1 /* Broadcom NetXtreme-C/E network driver. 2 * 3 * Copyright (c) 2017 Broadcom Limited 4 * 5 * This program is free software; you can redistribute it and/or modify 6 * it under the terms of the GNU General Public License as published by 7 * the Free Software Foundation. 8 */ 9 10 #include <linux/netdevice.h> 11 #include <linux/inetdevice.h> 12 #include <linux/if_vlan.h> 13 #include <net/flow_dissector.h> 14 #include <net/pkt_cls.h> 15 #include <net/tc_act/tc_gact.h> 16 #include <net/tc_act/tc_skbedit.h> 17 #include <net/tc_act/tc_mirred.h> 18 #include <net/tc_act/tc_vlan.h> 19 #include <net/tc_act/tc_tunnel_key.h> 20 21 #include "bnxt_hsi.h" 22 #include "bnxt.h" 23 #include "bnxt_sriov.h" 24 #include "bnxt_tc.h" 25 #include "bnxt_vfr.h" 26 27 #define BNXT_FID_INVALID 0xffff 28 #define VLAN_TCI(vid, prio) ((vid) | ((prio) << VLAN_PRIO_SHIFT)) 29 30 /* Return the dst fid of the func for flow forwarding 31 * For PFs: src_fid is the fid of the PF 32 * For VF-reps: src_fid the fid of the VF 33 */ 34 static u16 bnxt_flow_get_dst_fid(struct bnxt *pf_bp, struct net_device *dev) 35 { 36 struct bnxt *bp; 37 38 /* check if dev belongs to the same switch */ 39 if (!switchdev_port_same_parent_id(pf_bp->dev, dev)) { 40 netdev_info(pf_bp->dev, "dev(ifindex=%d) not on same switch", 41 dev->ifindex); 42 return BNXT_FID_INVALID; 43 } 44 45 /* Is dev a VF-rep? */ 46 if (dev != pf_bp->dev) 47 return bnxt_vf_rep_get_fid(dev); 48 49 bp = netdev_priv(dev); 50 return bp->pf.fw_fid; 51 } 52 53 static int bnxt_tc_parse_redir(struct bnxt *bp, 54 struct bnxt_tc_actions *actions, 55 const struct tc_action *tc_act) 56 { 57 int ifindex = tcf_mirred_ifindex(tc_act); 58 struct net_device *dev; 59 60 dev = __dev_get_by_index(dev_net(bp->dev), ifindex); 61 if (!dev) { 62 netdev_info(bp->dev, "no dev for ifindex=%d", ifindex); 63 return -EINVAL; 64 } 65 66 actions->flags |= BNXT_TC_ACTION_FLAG_FWD; 67 actions->dst_dev = dev; 68 return 0; 69 } 70 71 static void bnxt_tc_parse_vlan(struct bnxt *bp, 72 struct bnxt_tc_actions *actions, 73 const struct tc_action *tc_act) 74 { 75 if (tcf_vlan_action(tc_act) == TCA_VLAN_ACT_POP) { 76 actions->flags |= BNXT_TC_ACTION_FLAG_POP_VLAN; 77 } else if (tcf_vlan_action(tc_act) == TCA_VLAN_ACT_PUSH) { 78 actions->flags |= BNXT_TC_ACTION_FLAG_PUSH_VLAN; 79 actions->push_vlan_tci = htons(tcf_vlan_push_vid(tc_act)); 80 actions->push_vlan_tpid = tcf_vlan_push_proto(tc_act); 81 } 82 } 83 84 static int bnxt_tc_parse_tunnel_set(struct bnxt *bp, 85 struct bnxt_tc_actions *actions, 86 const struct tc_action *tc_act) 87 { 88 struct ip_tunnel_info *tun_info = tcf_tunnel_info(tc_act); 89 struct ip_tunnel_key *tun_key = &tun_info->key; 90 91 if (ip_tunnel_info_af(tun_info) != AF_INET) { 92 netdev_info(bp->dev, "only IPv4 tunnel-encap is supported"); 93 return -EOPNOTSUPP; 94 } 95 96 actions->tun_encap_key = *tun_key; 97 actions->flags |= BNXT_TC_ACTION_FLAG_TUNNEL_ENCAP; 98 return 0; 99 } 100 101 static int bnxt_tc_parse_actions(struct bnxt *bp, 102 struct bnxt_tc_actions *actions, 103 struct tcf_exts *tc_exts) 104 { 105 const struct tc_action *tc_act; 106 LIST_HEAD(tc_actions); 107 int rc; 108 109 if (!tcf_exts_has_actions(tc_exts)) { 110 netdev_info(bp->dev, "no actions"); 111 return -EINVAL; 112 } 113 114 tcf_exts_to_list(tc_exts, &tc_actions); 115 list_for_each_entry(tc_act, &tc_actions, list) { 116 /* Drop action */ 117 if (is_tcf_gact_shot(tc_act)) { 118 actions->flags |= BNXT_TC_ACTION_FLAG_DROP; 119 return 0; /* don't bother with other actions */ 120 } 121 122 /* Redirect action */ 123 if (is_tcf_mirred_egress_redirect(tc_act)) { 124 rc = bnxt_tc_parse_redir(bp, actions, tc_act); 125 if (rc) 126 return rc; 127 continue; 128 } 129 130 /* Push/pop VLAN */ 131 if (is_tcf_vlan(tc_act)) { 132 bnxt_tc_parse_vlan(bp, actions, tc_act); 133 continue; 134 } 135 136 /* Tunnel encap */ 137 if (is_tcf_tunnel_set(tc_act)) { 138 rc = bnxt_tc_parse_tunnel_set(bp, actions, tc_act); 139 if (rc) 140 return rc; 141 continue; 142 } 143 144 /* Tunnel decap */ 145 if (is_tcf_tunnel_release(tc_act)) { 146 actions->flags |= BNXT_TC_ACTION_FLAG_TUNNEL_DECAP; 147 continue; 148 } 149 } 150 151 if (rc) 152 return rc; 153 154 if (actions->flags & BNXT_TC_ACTION_FLAG_FWD) { 155 if (actions->flags & BNXT_TC_ACTION_FLAG_TUNNEL_ENCAP) { 156 /* dst_fid is PF's fid */ 157 actions->dst_fid = bp->pf.fw_fid; 158 } else { 159 /* find the FID from dst_dev */ 160 actions->dst_fid = 161 bnxt_flow_get_dst_fid(bp, actions->dst_dev); 162 if (actions->dst_fid == BNXT_FID_INVALID) 163 return -EINVAL; 164 } 165 } 166 167 return rc; 168 } 169 170 #define GET_KEY(flow_cmd, key_type) \ 171 skb_flow_dissector_target((flow_cmd)->dissector, key_type,\ 172 (flow_cmd)->key) 173 #define GET_MASK(flow_cmd, key_type) \ 174 skb_flow_dissector_target((flow_cmd)->dissector, key_type,\ 175 (flow_cmd)->mask) 176 177 static int bnxt_tc_parse_flow(struct bnxt *bp, 178 struct tc_cls_flower_offload *tc_flow_cmd, 179 struct bnxt_tc_flow *flow) 180 { 181 struct flow_dissector *dissector = tc_flow_cmd->dissector; 182 u16 addr_type = 0; 183 184 /* KEY_CONTROL and KEY_BASIC are needed for forming a meaningful key */ 185 if ((dissector->used_keys & BIT(FLOW_DISSECTOR_KEY_CONTROL)) == 0 || 186 (dissector->used_keys & BIT(FLOW_DISSECTOR_KEY_BASIC)) == 0) { 187 netdev_info(bp->dev, "cannot form TC key: used_keys = 0x%x", 188 dissector->used_keys); 189 return -EOPNOTSUPP; 190 } 191 192 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_CONTROL)) { 193 struct flow_dissector_key_control *key = 194 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_CONTROL); 195 196 addr_type = key->addr_type; 197 } 198 199 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_BASIC)) { 200 struct flow_dissector_key_basic *key = 201 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_BASIC); 202 struct flow_dissector_key_basic *mask = 203 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_BASIC); 204 205 flow->l2_key.ether_type = key->n_proto; 206 flow->l2_mask.ether_type = mask->n_proto; 207 208 if (key->n_proto == htons(ETH_P_IP) || 209 key->n_proto == htons(ETH_P_IPV6)) { 210 flow->l4_key.ip_proto = key->ip_proto; 211 flow->l4_mask.ip_proto = mask->ip_proto; 212 } 213 } 214 215 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_ETH_ADDRS)) { 216 struct flow_dissector_key_eth_addrs *key = 217 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_ETH_ADDRS); 218 struct flow_dissector_key_eth_addrs *mask = 219 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_ETH_ADDRS); 220 221 flow->flags |= BNXT_TC_FLOW_FLAGS_ETH_ADDRS; 222 ether_addr_copy(flow->l2_key.dmac, key->dst); 223 ether_addr_copy(flow->l2_mask.dmac, mask->dst); 224 ether_addr_copy(flow->l2_key.smac, key->src); 225 ether_addr_copy(flow->l2_mask.smac, mask->src); 226 } 227 228 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_VLAN)) { 229 struct flow_dissector_key_vlan *key = 230 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_VLAN); 231 struct flow_dissector_key_vlan *mask = 232 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_VLAN); 233 234 flow->l2_key.inner_vlan_tci = 235 cpu_to_be16(VLAN_TCI(key->vlan_id, key->vlan_priority)); 236 flow->l2_mask.inner_vlan_tci = 237 cpu_to_be16((VLAN_TCI(mask->vlan_id, mask->vlan_priority))); 238 flow->l2_key.inner_vlan_tpid = htons(ETH_P_8021Q); 239 flow->l2_mask.inner_vlan_tpid = htons(0xffff); 240 flow->l2_key.num_vlans = 1; 241 } 242 243 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_IPV4_ADDRS)) { 244 struct flow_dissector_key_ipv4_addrs *key = 245 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_IPV4_ADDRS); 246 struct flow_dissector_key_ipv4_addrs *mask = 247 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_IPV4_ADDRS); 248 249 flow->flags |= BNXT_TC_FLOW_FLAGS_IPV4_ADDRS; 250 flow->l3_key.ipv4.daddr.s_addr = key->dst; 251 flow->l3_mask.ipv4.daddr.s_addr = mask->dst; 252 flow->l3_key.ipv4.saddr.s_addr = key->src; 253 flow->l3_mask.ipv4.saddr.s_addr = mask->src; 254 } else if (dissector_uses_key(dissector, 255 FLOW_DISSECTOR_KEY_IPV6_ADDRS)) { 256 struct flow_dissector_key_ipv6_addrs *key = 257 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_IPV6_ADDRS); 258 struct flow_dissector_key_ipv6_addrs *mask = 259 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_IPV6_ADDRS); 260 261 flow->flags |= BNXT_TC_FLOW_FLAGS_IPV6_ADDRS; 262 flow->l3_key.ipv6.daddr = key->dst; 263 flow->l3_mask.ipv6.daddr = mask->dst; 264 flow->l3_key.ipv6.saddr = key->src; 265 flow->l3_mask.ipv6.saddr = mask->src; 266 } 267 268 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_PORTS)) { 269 struct flow_dissector_key_ports *key = 270 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_PORTS); 271 struct flow_dissector_key_ports *mask = 272 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_PORTS); 273 274 flow->flags |= BNXT_TC_FLOW_FLAGS_PORTS; 275 flow->l4_key.ports.dport = key->dst; 276 flow->l4_mask.ports.dport = mask->dst; 277 flow->l4_key.ports.sport = key->src; 278 flow->l4_mask.ports.sport = mask->src; 279 } 280 281 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_ICMP)) { 282 struct flow_dissector_key_icmp *key = 283 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_ICMP); 284 struct flow_dissector_key_icmp *mask = 285 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_ICMP); 286 287 flow->flags |= BNXT_TC_FLOW_FLAGS_ICMP; 288 flow->l4_key.icmp.type = key->type; 289 flow->l4_key.icmp.code = key->code; 290 flow->l4_mask.icmp.type = mask->type; 291 flow->l4_mask.icmp.code = mask->code; 292 } 293 294 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_ENC_CONTROL)) { 295 struct flow_dissector_key_control *key = 296 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_ENC_CONTROL); 297 298 addr_type = key->addr_type; 299 } 300 301 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS)) { 302 struct flow_dissector_key_ipv4_addrs *key = 303 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS); 304 struct flow_dissector_key_ipv4_addrs *mask = 305 GET_MASK(tc_flow_cmd, 306 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS); 307 308 flow->flags |= BNXT_TC_FLOW_FLAGS_TUNL_IPV4_ADDRS; 309 flow->tun_key.u.ipv4.dst = key->dst; 310 flow->tun_mask.u.ipv4.dst = mask->dst; 311 flow->tun_key.u.ipv4.src = key->src; 312 flow->tun_mask.u.ipv4.src = mask->src; 313 } else if (dissector_uses_key(dissector, 314 FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS)) { 315 return -EOPNOTSUPP; 316 } 317 318 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_ENC_KEYID)) { 319 struct flow_dissector_key_keyid *key = 320 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_ENC_KEYID); 321 struct flow_dissector_key_keyid *mask = 322 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_ENC_KEYID); 323 324 flow->flags |= BNXT_TC_FLOW_FLAGS_TUNL_ID; 325 flow->tun_key.tun_id = key32_to_tunnel_id(key->keyid); 326 flow->tun_mask.tun_id = key32_to_tunnel_id(mask->keyid); 327 } 328 329 if (dissector_uses_key(dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) { 330 struct flow_dissector_key_ports *key = 331 GET_KEY(tc_flow_cmd, FLOW_DISSECTOR_KEY_ENC_PORTS); 332 struct flow_dissector_key_ports *mask = 333 GET_MASK(tc_flow_cmd, FLOW_DISSECTOR_KEY_ENC_PORTS); 334 335 flow->flags |= BNXT_TC_FLOW_FLAGS_TUNL_PORTS; 336 flow->tun_key.tp_dst = key->dst; 337 flow->tun_mask.tp_dst = mask->dst; 338 flow->tun_key.tp_src = key->src; 339 flow->tun_mask.tp_src = mask->src; 340 } 341 342 return bnxt_tc_parse_actions(bp, &flow->actions, tc_flow_cmd->exts); 343 } 344 345 static int bnxt_hwrm_cfa_flow_free(struct bnxt *bp, __le16 flow_handle) 346 { 347 struct hwrm_cfa_flow_free_input req = { 0 }; 348 int rc; 349 350 bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_FLOW_FREE, -1, -1); 351 req.flow_handle = flow_handle; 352 353 rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT); 354 if (rc) 355 netdev_info(bp->dev, "Error: %s: flow_handle=0x%x rc=%d", 356 __func__, flow_handle, rc); 357 return rc; 358 } 359 360 static int ipv6_mask_len(struct in6_addr *mask) 361 { 362 int mask_len = 0, i; 363 364 for (i = 0; i < 4; i++) 365 mask_len += inet_mask_len(mask->s6_addr32[i]); 366 367 return mask_len; 368 } 369 370 static bool is_wildcard(void *mask, int len) 371 { 372 const u8 *p = mask; 373 int i; 374 375 for (i = 0; i < len; i++) { 376 if (p[i] != 0) 377 return false; 378 } 379 return true; 380 } 381 382 static int bnxt_hwrm_cfa_flow_alloc(struct bnxt *bp, struct bnxt_tc_flow *flow, 383 __le16 ref_flow_handle, 384 __le32 tunnel_handle, __le16 *flow_handle) 385 { 386 struct hwrm_cfa_flow_alloc_output *resp = bp->hwrm_cmd_resp_addr; 387 struct bnxt_tc_actions *actions = &flow->actions; 388 struct bnxt_tc_l3_key *l3_mask = &flow->l3_mask; 389 struct bnxt_tc_l3_key *l3_key = &flow->l3_key; 390 struct hwrm_cfa_flow_alloc_input req = { 0 }; 391 u16 flow_flags = 0, action_flags = 0; 392 int rc; 393 394 bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_FLOW_ALLOC, -1, -1); 395 396 req.src_fid = cpu_to_le16(flow->src_fid); 397 req.ref_flow_handle = ref_flow_handle; 398 399 if (actions->flags & BNXT_TC_ACTION_FLAG_TUNNEL_DECAP || 400 actions->flags & BNXT_TC_ACTION_FLAG_TUNNEL_ENCAP) { 401 req.tunnel_handle = tunnel_handle; 402 flow_flags |= CFA_FLOW_ALLOC_REQ_FLAGS_TUNNEL; 403 action_flags |= CFA_FLOW_ALLOC_REQ_ACTION_FLAGS_TUNNEL; 404 } 405 406 req.ethertype = flow->l2_key.ether_type; 407 req.ip_proto = flow->l4_key.ip_proto; 408 409 if (flow->flags & BNXT_TC_FLOW_FLAGS_ETH_ADDRS) { 410 memcpy(req.dmac, flow->l2_key.dmac, ETH_ALEN); 411 memcpy(req.smac, flow->l2_key.smac, ETH_ALEN); 412 } 413 414 if (flow->l2_key.num_vlans > 0) { 415 flow_flags |= CFA_FLOW_ALLOC_REQ_FLAGS_NUM_VLAN_ONE; 416 /* FW expects the inner_vlan_tci value to be set 417 * in outer_vlan_tci when num_vlans is 1 (which is 418 * always the case in TC.) 419 */ 420 req.outer_vlan_tci = flow->l2_key.inner_vlan_tci; 421 } 422 423 /* If all IP and L4 fields are wildcarded then this is an L2 flow */ 424 if (is_wildcard(&l3_mask, sizeof(l3_mask)) && 425 is_wildcard(&flow->l4_mask, sizeof(flow->l4_mask))) { 426 flow_flags |= CFA_FLOW_ALLOC_REQ_FLAGS_FLOWTYPE_L2; 427 } else { 428 flow_flags |= flow->l2_key.ether_type == htons(ETH_P_IP) ? 429 CFA_FLOW_ALLOC_REQ_FLAGS_FLOWTYPE_IPV4 : 430 CFA_FLOW_ALLOC_REQ_FLAGS_FLOWTYPE_IPV6; 431 432 if (flow->flags & BNXT_TC_FLOW_FLAGS_IPV4_ADDRS) { 433 req.ip_dst[0] = l3_key->ipv4.daddr.s_addr; 434 req.ip_dst_mask_len = 435 inet_mask_len(l3_mask->ipv4.daddr.s_addr); 436 req.ip_src[0] = l3_key->ipv4.saddr.s_addr; 437 req.ip_src_mask_len = 438 inet_mask_len(l3_mask->ipv4.saddr.s_addr); 439 } else if (flow->flags & BNXT_TC_FLOW_FLAGS_IPV6_ADDRS) { 440 memcpy(req.ip_dst, l3_key->ipv6.daddr.s6_addr32, 441 sizeof(req.ip_dst)); 442 req.ip_dst_mask_len = 443 ipv6_mask_len(&l3_mask->ipv6.daddr); 444 memcpy(req.ip_src, l3_key->ipv6.saddr.s6_addr32, 445 sizeof(req.ip_src)); 446 req.ip_src_mask_len = 447 ipv6_mask_len(&l3_mask->ipv6.saddr); 448 } 449 } 450 451 if (flow->flags & BNXT_TC_FLOW_FLAGS_PORTS) { 452 req.l4_src_port = flow->l4_key.ports.sport; 453 req.l4_src_port_mask = flow->l4_mask.ports.sport; 454 req.l4_dst_port = flow->l4_key.ports.dport; 455 req.l4_dst_port_mask = flow->l4_mask.ports.dport; 456 } else if (flow->flags & BNXT_TC_FLOW_FLAGS_ICMP) { 457 /* l4 ports serve as type/code when ip_proto is ICMP */ 458 req.l4_src_port = htons(flow->l4_key.icmp.type); 459 req.l4_src_port_mask = htons(flow->l4_mask.icmp.type); 460 req.l4_dst_port = htons(flow->l4_key.icmp.code); 461 req.l4_dst_port_mask = htons(flow->l4_mask.icmp.code); 462 } 463 req.flags = cpu_to_le16(flow_flags); 464 465 if (actions->flags & BNXT_TC_ACTION_FLAG_DROP) { 466 action_flags |= CFA_FLOW_ALLOC_REQ_ACTION_FLAGS_DROP; 467 } else { 468 if (actions->flags & BNXT_TC_ACTION_FLAG_FWD) { 469 action_flags |= CFA_FLOW_ALLOC_REQ_ACTION_FLAGS_FWD; 470 req.dst_fid = cpu_to_le16(actions->dst_fid); 471 } 472 if (actions->flags & BNXT_TC_ACTION_FLAG_PUSH_VLAN) { 473 action_flags |= 474 CFA_FLOW_ALLOC_REQ_ACTION_FLAGS_L2_HEADER_REWRITE; 475 req.l2_rewrite_vlan_tpid = actions->push_vlan_tpid; 476 req.l2_rewrite_vlan_tci = actions->push_vlan_tci; 477 memcpy(&req.l2_rewrite_dmac, &req.dmac, ETH_ALEN); 478 memcpy(&req.l2_rewrite_smac, &req.smac, ETH_ALEN); 479 } 480 if (actions->flags & BNXT_TC_ACTION_FLAG_POP_VLAN) { 481 action_flags |= 482 CFA_FLOW_ALLOC_REQ_ACTION_FLAGS_L2_HEADER_REWRITE; 483 /* Rewrite config with tpid = 0 implies vlan pop */ 484 req.l2_rewrite_vlan_tpid = 0; 485 memcpy(&req.l2_rewrite_dmac, &req.dmac, ETH_ALEN); 486 memcpy(&req.l2_rewrite_smac, &req.smac, ETH_ALEN); 487 } 488 } 489 req.action_flags = cpu_to_le16(action_flags); 490 491 mutex_lock(&bp->hwrm_cmd_lock); 492 493 rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT); 494 if (!rc) 495 *flow_handle = resp->flow_handle; 496 497 mutex_unlock(&bp->hwrm_cmd_lock); 498 499 return rc; 500 } 501 502 static int hwrm_cfa_decap_filter_alloc(struct bnxt *bp, 503 struct bnxt_tc_flow *flow, 504 struct bnxt_tc_l2_key *l2_info, 505 __le32 ref_decap_handle, 506 __le32 *decap_filter_handle) 507 { 508 struct hwrm_cfa_decap_filter_alloc_output *resp = 509 bp->hwrm_cmd_resp_addr; 510 struct hwrm_cfa_decap_filter_alloc_input req = { 0 }; 511 struct ip_tunnel_key *tun_key = &flow->tun_key; 512 u32 enables = 0; 513 int rc; 514 515 bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_DECAP_FILTER_ALLOC, -1, -1); 516 517 req.flags = cpu_to_le32(CFA_DECAP_FILTER_ALLOC_REQ_FLAGS_OVS_TUNNEL); 518 enables |= CFA_DECAP_FILTER_ALLOC_REQ_ENABLES_TUNNEL_TYPE | 519 CFA_DECAP_FILTER_ALLOC_REQ_ENABLES_IP_PROTOCOL; 520 req.tunnel_type = CFA_DECAP_FILTER_ALLOC_REQ_TUNNEL_TYPE_VXLAN; 521 req.ip_protocol = CFA_DECAP_FILTER_ALLOC_REQ_IP_PROTOCOL_UDP; 522 523 if (flow->flags & BNXT_TC_FLOW_FLAGS_TUNL_ID) { 524 enables |= CFA_DECAP_FILTER_ALLOC_REQ_ENABLES_TUNNEL_ID; 525 /* tunnel_id is wrongly defined in hsi defn. as __le32 */ 526 req.tunnel_id = tunnel_id_to_key32(tun_key->tun_id); 527 } 528 529 if (flow->flags & BNXT_TC_FLOW_FLAGS_TUNL_ETH_ADDRS) { 530 enables |= CFA_DECAP_FILTER_ALLOC_REQ_ENABLES_DST_MACADDR; 531 ether_addr_copy(req.dst_macaddr, l2_info->dmac); 532 } 533 if (l2_info->num_vlans) { 534 enables |= CFA_DECAP_FILTER_ALLOC_REQ_ENABLES_T_IVLAN_VID; 535 req.t_ivlan_vid = l2_info->inner_vlan_tci; 536 } 537 538 enables |= CFA_DECAP_FILTER_ALLOC_REQ_ENABLES_ETHERTYPE; 539 req.ethertype = htons(ETH_P_IP); 540 541 if (flow->flags & BNXT_TC_FLOW_FLAGS_TUNL_IPV4_ADDRS) { 542 enables |= CFA_DECAP_FILTER_ALLOC_REQ_ENABLES_SRC_IPADDR | 543 CFA_DECAP_FILTER_ALLOC_REQ_ENABLES_DST_IPADDR | 544 CFA_DECAP_FILTER_ALLOC_REQ_ENABLES_IPADDR_TYPE; 545 req.ip_addr_type = CFA_DECAP_FILTER_ALLOC_REQ_IP_ADDR_TYPE_IPV4; 546 req.dst_ipaddr[0] = tun_key->u.ipv4.dst; 547 req.src_ipaddr[0] = tun_key->u.ipv4.src; 548 } 549 550 if (flow->flags & BNXT_TC_FLOW_FLAGS_TUNL_PORTS) { 551 enables |= CFA_DECAP_FILTER_ALLOC_REQ_ENABLES_DST_PORT; 552 req.dst_port = tun_key->tp_dst; 553 } 554 555 /* Eventhough the decap_handle returned by hwrm_cfa_decap_filter_alloc 556 * is defined as __le32, l2_ctxt_ref_id is defined in HSI as __le16. 557 */ 558 req.l2_ctxt_ref_id = (__force __le16)ref_decap_handle; 559 req.enables = cpu_to_le32(enables); 560 561 mutex_lock(&bp->hwrm_cmd_lock); 562 rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT); 563 if (!rc) 564 *decap_filter_handle = resp->decap_filter_id; 565 else 566 netdev_info(bp->dev, "%s: Error rc=%d", __func__, rc); 567 mutex_unlock(&bp->hwrm_cmd_lock); 568 569 return rc; 570 } 571 572 static int hwrm_cfa_decap_filter_free(struct bnxt *bp, 573 __le32 decap_filter_handle) 574 { 575 struct hwrm_cfa_decap_filter_free_input req = { 0 }; 576 int rc; 577 578 bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_DECAP_FILTER_FREE, -1, -1); 579 req.decap_filter_id = decap_filter_handle; 580 581 rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT); 582 if (rc) 583 netdev_info(bp->dev, "%s: Error rc=%d", __func__, rc); 584 return rc; 585 } 586 587 static int hwrm_cfa_encap_record_alloc(struct bnxt *bp, 588 struct ip_tunnel_key *encap_key, 589 struct bnxt_tc_l2_key *l2_info, 590 __le32 *encap_record_handle) 591 { 592 struct hwrm_cfa_encap_record_alloc_output *resp = 593 bp->hwrm_cmd_resp_addr; 594 struct hwrm_cfa_encap_record_alloc_input req = { 0 }; 595 struct hwrm_cfa_encap_data_vxlan *encap = 596 (struct hwrm_cfa_encap_data_vxlan *)&req.encap_data; 597 struct hwrm_vxlan_ipv4_hdr *encap_ipv4 = 598 (struct hwrm_vxlan_ipv4_hdr *)encap->l3; 599 int rc; 600 601 bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_ENCAP_RECORD_ALLOC, -1, -1); 602 603 req.encap_type = CFA_ENCAP_RECORD_ALLOC_REQ_ENCAP_TYPE_VXLAN; 604 605 ether_addr_copy(encap->dst_mac_addr, l2_info->dmac); 606 ether_addr_copy(encap->src_mac_addr, l2_info->smac); 607 if (l2_info->num_vlans) { 608 encap->num_vlan_tags = l2_info->num_vlans; 609 encap->ovlan_tci = l2_info->inner_vlan_tci; 610 encap->ovlan_tpid = l2_info->inner_vlan_tpid; 611 } 612 613 encap_ipv4->ver_hlen = 4 << VXLAN_IPV4_HDR_VER_HLEN_VERSION_SFT; 614 encap_ipv4->ver_hlen |= 5 << VXLAN_IPV4_HDR_VER_HLEN_HEADER_LENGTH_SFT; 615 encap_ipv4->ttl = encap_key->ttl; 616 617 encap_ipv4->dest_ip_addr = encap_key->u.ipv4.dst; 618 encap_ipv4->src_ip_addr = encap_key->u.ipv4.src; 619 encap_ipv4->protocol = IPPROTO_UDP; 620 621 encap->dst_port = encap_key->tp_dst; 622 encap->vni = tunnel_id_to_key32(encap_key->tun_id); 623 624 mutex_lock(&bp->hwrm_cmd_lock); 625 rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT); 626 if (!rc) 627 *encap_record_handle = resp->encap_record_id; 628 else 629 netdev_info(bp->dev, "%s: Error rc=%d", __func__, rc); 630 mutex_unlock(&bp->hwrm_cmd_lock); 631 632 return rc; 633 } 634 635 static int hwrm_cfa_encap_record_free(struct bnxt *bp, 636 __le32 encap_record_handle) 637 { 638 struct hwrm_cfa_encap_record_free_input req = { 0 }; 639 int rc; 640 641 bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_ENCAP_RECORD_FREE, -1, -1); 642 req.encap_record_id = encap_record_handle; 643 644 rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT); 645 if (rc) 646 netdev_info(bp->dev, "%s: Error rc=%d", __func__, rc); 647 return rc; 648 } 649 650 static int bnxt_tc_put_l2_node(struct bnxt *bp, 651 struct bnxt_tc_flow_node *flow_node) 652 { 653 struct bnxt_tc_l2_node *l2_node = flow_node->l2_node; 654 struct bnxt_tc_info *tc_info = bp->tc_info; 655 int rc; 656 657 /* remove flow_node from the L2 shared flow list */ 658 list_del(&flow_node->l2_list_node); 659 if (--l2_node->refcount == 0) { 660 rc = rhashtable_remove_fast(&tc_info->l2_table, &l2_node->node, 661 tc_info->l2_ht_params); 662 if (rc) 663 netdev_err(bp->dev, 664 "Error: %s: rhashtable_remove_fast: %d", 665 __func__, rc); 666 kfree_rcu(l2_node, rcu); 667 } 668 return 0; 669 } 670 671 static struct bnxt_tc_l2_node * 672 bnxt_tc_get_l2_node(struct bnxt *bp, struct rhashtable *l2_table, 673 struct rhashtable_params ht_params, 674 struct bnxt_tc_l2_key *l2_key) 675 { 676 struct bnxt_tc_l2_node *l2_node; 677 int rc; 678 679 l2_node = rhashtable_lookup_fast(l2_table, l2_key, ht_params); 680 if (!l2_node) { 681 l2_node = kzalloc(sizeof(*l2_node), GFP_KERNEL); 682 if (!l2_node) { 683 rc = -ENOMEM; 684 return NULL; 685 } 686 687 l2_node->key = *l2_key; 688 rc = rhashtable_insert_fast(l2_table, &l2_node->node, 689 ht_params); 690 if (rc) { 691 kfree_rcu(l2_node, rcu); 692 netdev_err(bp->dev, 693 "Error: %s: rhashtable_insert_fast: %d", 694 __func__, rc); 695 return NULL; 696 } 697 INIT_LIST_HEAD(&l2_node->common_l2_flows); 698 } 699 return l2_node; 700 } 701 702 /* Get the ref_flow_handle for a flow by checking if there are any other 703 * flows that share the same L2 key as this flow. 704 */ 705 static int 706 bnxt_tc_get_ref_flow_handle(struct bnxt *bp, struct bnxt_tc_flow *flow, 707 struct bnxt_tc_flow_node *flow_node, 708 __le16 *ref_flow_handle) 709 { 710 struct bnxt_tc_info *tc_info = bp->tc_info; 711 struct bnxt_tc_flow_node *ref_flow_node; 712 struct bnxt_tc_l2_node *l2_node; 713 714 l2_node = bnxt_tc_get_l2_node(bp, &tc_info->l2_table, 715 tc_info->l2_ht_params, 716 &flow->l2_key); 717 if (!l2_node) 718 return -1; 719 720 /* If any other flow is using this l2_node, use it's flow_handle 721 * as the ref_flow_handle 722 */ 723 if (l2_node->refcount > 0) { 724 ref_flow_node = list_first_entry(&l2_node->common_l2_flows, 725 struct bnxt_tc_flow_node, 726 l2_list_node); 727 *ref_flow_handle = ref_flow_node->flow_handle; 728 } else { 729 *ref_flow_handle = cpu_to_le16(0xffff); 730 } 731 732 /* Insert the l2_node into the flow_node so that subsequent flows 733 * with a matching l2 key can use the flow_handle of this flow 734 * as their ref_flow_handle 735 */ 736 flow_node->l2_node = l2_node; 737 list_add(&flow_node->l2_list_node, &l2_node->common_l2_flows); 738 l2_node->refcount++; 739 return 0; 740 } 741 742 /* After the flow parsing is done, this routine is used for checking 743 * if there are any aspects of the flow that prevent it from being 744 * offloaded. 745 */ 746 static bool bnxt_tc_can_offload(struct bnxt *bp, struct bnxt_tc_flow *flow) 747 { 748 /* If L4 ports are specified then ip_proto must be TCP or UDP */ 749 if ((flow->flags & BNXT_TC_FLOW_FLAGS_PORTS) && 750 (flow->l4_key.ip_proto != IPPROTO_TCP && 751 flow->l4_key.ip_proto != IPPROTO_UDP)) { 752 netdev_info(bp->dev, "Cannot offload non-TCP/UDP (%d) ports", 753 flow->l4_key.ip_proto); 754 return false; 755 } 756 757 return true; 758 } 759 760 /* Returns the final refcount of the node on success 761 * or a -ve error code on failure 762 */ 763 static int bnxt_tc_put_tunnel_node(struct bnxt *bp, 764 struct rhashtable *tunnel_table, 765 struct rhashtable_params *ht_params, 766 struct bnxt_tc_tunnel_node *tunnel_node) 767 { 768 int rc; 769 770 if (--tunnel_node->refcount == 0) { 771 rc = rhashtable_remove_fast(tunnel_table, &tunnel_node->node, 772 *ht_params); 773 if (rc) { 774 netdev_err(bp->dev, "rhashtable_remove_fast rc=%d", rc); 775 rc = -1; 776 } 777 kfree_rcu(tunnel_node, rcu); 778 return rc; 779 } else { 780 return tunnel_node->refcount; 781 } 782 } 783 784 /* Get (or add) either encap or decap tunnel node from/to the supplied 785 * hash table. 786 */ 787 static struct bnxt_tc_tunnel_node * 788 bnxt_tc_get_tunnel_node(struct bnxt *bp, struct rhashtable *tunnel_table, 789 struct rhashtable_params *ht_params, 790 struct ip_tunnel_key *tun_key) 791 { 792 struct bnxt_tc_tunnel_node *tunnel_node; 793 int rc; 794 795 tunnel_node = rhashtable_lookup_fast(tunnel_table, tun_key, *ht_params); 796 if (!tunnel_node) { 797 tunnel_node = kzalloc(sizeof(*tunnel_node), GFP_KERNEL); 798 if (!tunnel_node) { 799 rc = -ENOMEM; 800 goto err; 801 } 802 803 tunnel_node->key = *tun_key; 804 tunnel_node->tunnel_handle = INVALID_TUNNEL_HANDLE; 805 rc = rhashtable_insert_fast(tunnel_table, &tunnel_node->node, 806 *ht_params); 807 if (rc) { 808 kfree_rcu(tunnel_node, rcu); 809 goto err; 810 } 811 } 812 tunnel_node->refcount++; 813 return tunnel_node; 814 err: 815 netdev_info(bp->dev, "error rc=%d", rc); 816 return NULL; 817 } 818 819 static int bnxt_tc_get_ref_decap_handle(struct bnxt *bp, 820 struct bnxt_tc_flow *flow, 821 struct bnxt_tc_l2_key *l2_key, 822 struct bnxt_tc_flow_node *flow_node, 823 __le32 *ref_decap_handle) 824 { 825 struct bnxt_tc_info *tc_info = bp->tc_info; 826 struct bnxt_tc_flow_node *ref_flow_node; 827 struct bnxt_tc_l2_node *decap_l2_node; 828 829 decap_l2_node = bnxt_tc_get_l2_node(bp, &tc_info->decap_l2_table, 830 tc_info->decap_l2_ht_params, 831 l2_key); 832 if (!decap_l2_node) 833 return -1; 834 835 /* If any other flow is using this decap_l2_node, use it's decap_handle 836 * as the ref_decap_handle 837 */ 838 if (decap_l2_node->refcount > 0) { 839 ref_flow_node = 840 list_first_entry(&decap_l2_node->common_l2_flows, 841 struct bnxt_tc_flow_node, 842 decap_l2_list_node); 843 *ref_decap_handle = ref_flow_node->decap_node->tunnel_handle; 844 } else { 845 *ref_decap_handle = INVALID_TUNNEL_HANDLE; 846 } 847 848 /* Insert the l2_node into the flow_node so that subsequent flows 849 * with a matching decap l2 key can use the decap_filter_handle of 850 * this flow as their ref_decap_handle 851 */ 852 flow_node->decap_l2_node = decap_l2_node; 853 list_add(&flow_node->decap_l2_list_node, 854 &decap_l2_node->common_l2_flows); 855 decap_l2_node->refcount++; 856 return 0; 857 } 858 859 static void bnxt_tc_put_decap_l2_node(struct bnxt *bp, 860 struct bnxt_tc_flow_node *flow_node) 861 { 862 struct bnxt_tc_l2_node *decap_l2_node = flow_node->decap_l2_node; 863 struct bnxt_tc_info *tc_info = bp->tc_info; 864 int rc; 865 866 /* remove flow_node from the decap L2 sharing flow list */ 867 list_del(&flow_node->decap_l2_list_node); 868 if (--decap_l2_node->refcount == 0) { 869 rc = rhashtable_remove_fast(&tc_info->decap_l2_table, 870 &decap_l2_node->node, 871 tc_info->decap_l2_ht_params); 872 if (rc) 873 netdev_err(bp->dev, "rhashtable_remove_fast rc=%d", rc); 874 kfree_rcu(decap_l2_node, rcu); 875 } 876 } 877 878 static void bnxt_tc_put_decap_handle(struct bnxt *bp, 879 struct bnxt_tc_flow_node *flow_node) 880 { 881 __le32 decap_handle = flow_node->decap_node->tunnel_handle; 882 struct bnxt_tc_info *tc_info = bp->tc_info; 883 int rc; 884 885 if (flow_node->decap_l2_node) 886 bnxt_tc_put_decap_l2_node(bp, flow_node); 887 888 rc = bnxt_tc_put_tunnel_node(bp, &tc_info->decap_table, 889 &tc_info->decap_ht_params, 890 flow_node->decap_node); 891 if (!rc && decap_handle != INVALID_TUNNEL_HANDLE) 892 hwrm_cfa_decap_filter_free(bp, decap_handle); 893 } 894 895 static int bnxt_tc_resolve_tunnel_hdrs(struct bnxt *bp, 896 struct ip_tunnel_key *tun_key, 897 struct bnxt_tc_l2_key *l2_info) 898 { 899 #ifdef CONFIG_INET 900 struct net_device *real_dst_dev = bp->dev; 901 struct flowi4 flow = { {0} }; 902 struct net_device *dst_dev; 903 struct neighbour *nbr; 904 struct rtable *rt; 905 int rc; 906 907 flow.flowi4_proto = IPPROTO_UDP; 908 flow.fl4_dport = tun_key->tp_dst; 909 flow.daddr = tun_key->u.ipv4.dst; 910 911 rt = ip_route_output_key(dev_net(real_dst_dev), &flow); 912 if (IS_ERR(rt)) { 913 netdev_info(bp->dev, "no route to %pI4b", &flow.daddr); 914 return -EOPNOTSUPP; 915 } 916 917 /* The route must either point to the real_dst_dev or a dst_dev that 918 * uses the real_dst_dev. 919 */ 920 dst_dev = rt->dst.dev; 921 if (is_vlan_dev(dst_dev)) { 922 #if IS_ENABLED(CONFIG_VLAN_8021Q) 923 struct vlan_dev_priv *vlan = vlan_dev_priv(dst_dev); 924 925 if (vlan->real_dev != real_dst_dev) { 926 netdev_info(bp->dev, 927 "dst_dev(%s) doesn't use PF-if(%s)", 928 netdev_name(dst_dev), 929 netdev_name(real_dst_dev)); 930 rc = -EOPNOTSUPP; 931 goto put_rt; 932 } 933 l2_info->inner_vlan_tci = htons(vlan->vlan_id); 934 l2_info->inner_vlan_tpid = vlan->vlan_proto; 935 l2_info->num_vlans = 1; 936 #endif 937 } else if (dst_dev != real_dst_dev) { 938 netdev_info(bp->dev, 939 "dst_dev(%s) for %pI4b is not PF-if(%s)", 940 netdev_name(dst_dev), &flow.daddr, 941 netdev_name(real_dst_dev)); 942 rc = -EOPNOTSUPP; 943 goto put_rt; 944 } 945 946 nbr = dst_neigh_lookup(&rt->dst, &flow.daddr); 947 if (!nbr) { 948 netdev_info(bp->dev, "can't lookup neighbor for %pI4b", 949 &flow.daddr); 950 rc = -EOPNOTSUPP; 951 goto put_rt; 952 } 953 954 tun_key->u.ipv4.src = flow.saddr; 955 tun_key->ttl = ip4_dst_hoplimit(&rt->dst); 956 neigh_ha_snapshot(l2_info->dmac, nbr, dst_dev); 957 ether_addr_copy(l2_info->smac, dst_dev->dev_addr); 958 neigh_release(nbr); 959 ip_rt_put(rt); 960 961 return 0; 962 put_rt: 963 ip_rt_put(rt); 964 return rc; 965 #else 966 return -EOPNOTSUPP; 967 #endif 968 } 969 970 static int bnxt_tc_get_decap_handle(struct bnxt *bp, struct bnxt_tc_flow *flow, 971 struct bnxt_tc_flow_node *flow_node, 972 __le32 *decap_filter_handle) 973 { 974 struct ip_tunnel_key *decap_key = &flow->tun_key; 975 struct bnxt_tc_info *tc_info = bp->tc_info; 976 struct bnxt_tc_l2_key l2_info = { {0} }; 977 struct bnxt_tc_tunnel_node *decap_node; 978 struct ip_tunnel_key tun_key = { 0 }; 979 struct bnxt_tc_l2_key *decap_l2_info; 980 __le32 ref_decap_handle; 981 int rc; 982 983 /* Check if there's another flow using the same tunnel decap. 984 * If not, add this tunnel to the table and resolve the other 985 * tunnel header fileds 986 */ 987 decap_node = bnxt_tc_get_tunnel_node(bp, &tc_info->decap_table, 988 &tc_info->decap_ht_params, 989 decap_key); 990 if (!decap_node) 991 return -ENOMEM; 992 993 flow_node->decap_node = decap_node; 994 995 if (decap_node->tunnel_handle != INVALID_TUNNEL_HANDLE) 996 goto done; 997 998 /* Resolve the L2 fields for tunnel decap 999 * Resolve the route for remote vtep (saddr) of the decap key 1000 * Find it's next-hop mac addrs 1001 */ 1002 tun_key.u.ipv4.dst = flow->tun_key.u.ipv4.src; 1003 tun_key.tp_dst = flow->tun_key.tp_dst; 1004 rc = bnxt_tc_resolve_tunnel_hdrs(bp, &tun_key, &l2_info); 1005 if (rc) 1006 goto put_decap; 1007 1008 decap_l2_info = &decap_node->l2_info; 1009 /* decap smac is wildcarded */ 1010 ether_addr_copy(decap_l2_info->dmac, l2_info.smac); 1011 if (l2_info.num_vlans) { 1012 decap_l2_info->num_vlans = l2_info.num_vlans; 1013 decap_l2_info->inner_vlan_tpid = l2_info.inner_vlan_tpid; 1014 decap_l2_info->inner_vlan_tci = l2_info.inner_vlan_tci; 1015 } 1016 flow->flags |= BNXT_TC_FLOW_FLAGS_TUNL_ETH_ADDRS; 1017 1018 /* For getting a decap_filter_handle we first need to check if 1019 * there are any other decap flows that share the same tunnel L2 1020 * key and if so, pass that flow's decap_filter_handle as the 1021 * ref_decap_handle for this flow. 1022 */ 1023 rc = bnxt_tc_get_ref_decap_handle(bp, flow, decap_l2_info, flow_node, 1024 &ref_decap_handle); 1025 if (rc) 1026 goto put_decap; 1027 1028 /* Issue the hwrm cmd to allocate a decap filter handle */ 1029 rc = hwrm_cfa_decap_filter_alloc(bp, flow, decap_l2_info, 1030 ref_decap_handle, 1031 &decap_node->tunnel_handle); 1032 if (rc) 1033 goto put_decap_l2; 1034 1035 done: 1036 *decap_filter_handle = decap_node->tunnel_handle; 1037 return 0; 1038 1039 put_decap_l2: 1040 bnxt_tc_put_decap_l2_node(bp, flow_node); 1041 put_decap: 1042 bnxt_tc_put_tunnel_node(bp, &tc_info->decap_table, 1043 &tc_info->decap_ht_params, 1044 flow_node->decap_node); 1045 return rc; 1046 } 1047 1048 static void bnxt_tc_put_encap_handle(struct bnxt *bp, 1049 struct bnxt_tc_tunnel_node *encap_node) 1050 { 1051 __le32 encap_handle = encap_node->tunnel_handle; 1052 struct bnxt_tc_info *tc_info = bp->tc_info; 1053 int rc; 1054 1055 rc = bnxt_tc_put_tunnel_node(bp, &tc_info->encap_table, 1056 &tc_info->encap_ht_params, encap_node); 1057 if (!rc && encap_handle != INVALID_TUNNEL_HANDLE) 1058 hwrm_cfa_encap_record_free(bp, encap_handle); 1059 } 1060 1061 /* Lookup the tunnel encap table and check if there's an encap_handle 1062 * alloc'd already. 1063 * If not, query L2 info via a route lookup and issue an encap_record_alloc 1064 * cmd to FW. 1065 */ 1066 static int bnxt_tc_get_encap_handle(struct bnxt *bp, struct bnxt_tc_flow *flow, 1067 struct bnxt_tc_flow_node *flow_node, 1068 __le32 *encap_handle) 1069 { 1070 struct ip_tunnel_key *encap_key = &flow->actions.tun_encap_key; 1071 struct bnxt_tc_info *tc_info = bp->tc_info; 1072 struct bnxt_tc_tunnel_node *encap_node; 1073 int rc; 1074 1075 /* Check if there's another flow using the same tunnel encap. 1076 * If not, add this tunnel to the table and resolve the other 1077 * tunnel header fileds 1078 */ 1079 encap_node = bnxt_tc_get_tunnel_node(bp, &tc_info->encap_table, 1080 &tc_info->encap_ht_params, 1081 encap_key); 1082 if (!encap_node) 1083 return -ENOMEM; 1084 1085 flow_node->encap_node = encap_node; 1086 1087 if (encap_node->tunnel_handle != INVALID_TUNNEL_HANDLE) 1088 goto done; 1089 1090 rc = bnxt_tc_resolve_tunnel_hdrs(bp, encap_key, &encap_node->l2_info); 1091 if (rc) 1092 goto put_encap; 1093 1094 /* Allocate a new tunnel encap record */ 1095 rc = hwrm_cfa_encap_record_alloc(bp, encap_key, &encap_node->l2_info, 1096 &encap_node->tunnel_handle); 1097 if (rc) 1098 goto put_encap; 1099 1100 done: 1101 *encap_handle = encap_node->tunnel_handle; 1102 return 0; 1103 1104 put_encap: 1105 bnxt_tc_put_tunnel_node(bp, &tc_info->encap_table, 1106 &tc_info->encap_ht_params, encap_node); 1107 return rc; 1108 } 1109 1110 static void bnxt_tc_put_tunnel_handle(struct bnxt *bp, 1111 struct bnxt_tc_flow *flow, 1112 struct bnxt_tc_flow_node *flow_node) 1113 { 1114 if (flow->actions.flags & BNXT_TC_ACTION_FLAG_TUNNEL_DECAP) 1115 bnxt_tc_put_decap_handle(bp, flow_node); 1116 else if (flow->actions.flags & BNXT_TC_ACTION_FLAG_TUNNEL_ENCAP) 1117 bnxt_tc_put_encap_handle(bp, flow_node->encap_node); 1118 } 1119 1120 static int bnxt_tc_get_tunnel_handle(struct bnxt *bp, 1121 struct bnxt_tc_flow *flow, 1122 struct bnxt_tc_flow_node *flow_node, 1123 __le32 *tunnel_handle) 1124 { 1125 if (flow->actions.flags & BNXT_TC_ACTION_FLAG_TUNNEL_DECAP) 1126 return bnxt_tc_get_decap_handle(bp, flow, flow_node, 1127 tunnel_handle); 1128 else if (flow->actions.flags & BNXT_TC_ACTION_FLAG_TUNNEL_ENCAP) 1129 return bnxt_tc_get_encap_handle(bp, flow, flow_node, 1130 tunnel_handle); 1131 else 1132 return 0; 1133 } 1134 static int __bnxt_tc_del_flow(struct bnxt *bp, 1135 struct bnxt_tc_flow_node *flow_node) 1136 { 1137 struct bnxt_tc_info *tc_info = bp->tc_info; 1138 int rc; 1139 1140 /* send HWRM cmd to free the flow-id */ 1141 bnxt_hwrm_cfa_flow_free(bp, flow_node->flow_handle); 1142 1143 mutex_lock(&tc_info->lock); 1144 1145 /* release references to any tunnel encap/decap nodes */ 1146 bnxt_tc_put_tunnel_handle(bp, &flow_node->flow, flow_node); 1147 1148 /* release reference to l2 node */ 1149 bnxt_tc_put_l2_node(bp, flow_node); 1150 1151 mutex_unlock(&tc_info->lock); 1152 1153 rc = rhashtable_remove_fast(&tc_info->flow_table, &flow_node->node, 1154 tc_info->flow_ht_params); 1155 if (rc) 1156 netdev_err(bp->dev, "Error: %s: rhashtable_remove_fast rc=%d", 1157 __func__, rc); 1158 1159 kfree_rcu(flow_node, rcu); 1160 return 0; 1161 } 1162 1163 static void bnxt_tc_set_src_fid(struct bnxt *bp, struct bnxt_tc_flow *flow, 1164 u16 src_fid) 1165 { 1166 if (flow->actions.flags & BNXT_TC_ACTION_FLAG_TUNNEL_DECAP) 1167 flow->src_fid = bp->pf.fw_fid; 1168 else 1169 flow->src_fid = src_fid; 1170 } 1171 1172 /* Add a new flow or replace an existing flow. 1173 * Notes on locking: 1174 * There are essentially two critical sections here. 1175 * 1. while adding a new flow 1176 * a) lookup l2-key 1177 * b) issue HWRM cmd and get flow_handle 1178 * c) link l2-key with flow 1179 * 2. while deleting a flow 1180 * a) unlinking l2-key from flow 1181 * A lock is needed to protect these two critical sections. 1182 * 1183 * The hash-tables are already protected by the rhashtable API. 1184 */ 1185 static int bnxt_tc_add_flow(struct bnxt *bp, u16 src_fid, 1186 struct tc_cls_flower_offload *tc_flow_cmd) 1187 { 1188 struct bnxt_tc_flow_node *new_node, *old_node; 1189 struct bnxt_tc_info *tc_info = bp->tc_info; 1190 struct bnxt_tc_flow *flow; 1191 __le32 tunnel_handle = 0; 1192 __le16 ref_flow_handle; 1193 int rc; 1194 1195 /* allocate memory for the new flow and it's node */ 1196 new_node = kzalloc(sizeof(*new_node), GFP_KERNEL); 1197 if (!new_node) { 1198 rc = -ENOMEM; 1199 goto done; 1200 } 1201 new_node->cookie = tc_flow_cmd->cookie; 1202 flow = &new_node->flow; 1203 1204 rc = bnxt_tc_parse_flow(bp, tc_flow_cmd, flow); 1205 if (rc) 1206 goto free_node; 1207 1208 bnxt_tc_set_src_fid(bp, flow, src_fid); 1209 1210 if (!bnxt_tc_can_offload(bp, flow)) { 1211 rc = -ENOSPC; 1212 goto free_node; 1213 } 1214 1215 /* If a flow exists with the same cookie, delete it */ 1216 old_node = rhashtable_lookup_fast(&tc_info->flow_table, 1217 &tc_flow_cmd->cookie, 1218 tc_info->flow_ht_params); 1219 if (old_node) 1220 __bnxt_tc_del_flow(bp, old_node); 1221 1222 /* Check if the L2 part of the flow has been offloaded already. 1223 * If so, bump up it's refcnt and get it's reference handle. 1224 */ 1225 mutex_lock(&tc_info->lock); 1226 rc = bnxt_tc_get_ref_flow_handle(bp, flow, new_node, &ref_flow_handle); 1227 if (rc) 1228 goto unlock; 1229 1230 /* If the flow involves tunnel encap/decap, get tunnel_handle */ 1231 rc = bnxt_tc_get_tunnel_handle(bp, flow, new_node, &tunnel_handle); 1232 if (rc) 1233 goto put_l2; 1234 1235 /* send HWRM cmd to alloc the flow */ 1236 rc = bnxt_hwrm_cfa_flow_alloc(bp, flow, ref_flow_handle, 1237 tunnel_handle, &new_node->flow_handle); 1238 if (rc) 1239 goto put_tunnel; 1240 1241 flow->lastused = jiffies; 1242 spin_lock_init(&flow->stats_lock); 1243 /* add new flow to flow-table */ 1244 rc = rhashtable_insert_fast(&tc_info->flow_table, &new_node->node, 1245 tc_info->flow_ht_params); 1246 if (rc) 1247 goto hwrm_flow_free; 1248 1249 mutex_unlock(&tc_info->lock); 1250 return 0; 1251 1252 hwrm_flow_free: 1253 bnxt_hwrm_cfa_flow_free(bp, new_node->flow_handle); 1254 put_tunnel: 1255 bnxt_tc_put_tunnel_handle(bp, flow, new_node); 1256 put_l2: 1257 bnxt_tc_put_l2_node(bp, new_node); 1258 unlock: 1259 mutex_unlock(&tc_info->lock); 1260 free_node: 1261 kfree_rcu(new_node, rcu); 1262 done: 1263 netdev_err(bp->dev, "Error: %s: cookie=0x%lx error=%d", 1264 __func__, tc_flow_cmd->cookie, rc); 1265 return rc; 1266 } 1267 1268 static int bnxt_tc_del_flow(struct bnxt *bp, 1269 struct tc_cls_flower_offload *tc_flow_cmd) 1270 { 1271 struct bnxt_tc_info *tc_info = bp->tc_info; 1272 struct bnxt_tc_flow_node *flow_node; 1273 1274 flow_node = rhashtable_lookup_fast(&tc_info->flow_table, 1275 &tc_flow_cmd->cookie, 1276 tc_info->flow_ht_params); 1277 if (!flow_node) { 1278 netdev_info(bp->dev, "ERROR: no flow_node for cookie %lx", 1279 tc_flow_cmd->cookie); 1280 return -EINVAL; 1281 } 1282 1283 return __bnxt_tc_del_flow(bp, flow_node); 1284 } 1285 1286 static int bnxt_tc_get_flow_stats(struct bnxt *bp, 1287 struct tc_cls_flower_offload *tc_flow_cmd) 1288 { 1289 struct bnxt_tc_flow_stats stats, *curr_stats, *prev_stats; 1290 struct bnxt_tc_info *tc_info = bp->tc_info; 1291 struct bnxt_tc_flow_node *flow_node; 1292 struct bnxt_tc_flow *flow; 1293 unsigned long lastused; 1294 1295 flow_node = rhashtable_lookup_fast(&tc_info->flow_table, 1296 &tc_flow_cmd->cookie, 1297 tc_info->flow_ht_params); 1298 if (!flow_node) { 1299 netdev_info(bp->dev, "Error: no flow_node for cookie %lx", 1300 tc_flow_cmd->cookie); 1301 return -1; 1302 } 1303 1304 flow = &flow_node->flow; 1305 curr_stats = &flow->stats; 1306 prev_stats = &flow->prev_stats; 1307 1308 spin_lock(&flow->stats_lock); 1309 stats.packets = curr_stats->packets - prev_stats->packets; 1310 stats.bytes = curr_stats->bytes - prev_stats->bytes; 1311 *prev_stats = *curr_stats; 1312 lastused = flow->lastused; 1313 spin_unlock(&flow->stats_lock); 1314 1315 tcf_exts_stats_update(tc_flow_cmd->exts, stats.bytes, stats.packets, 1316 lastused); 1317 return 0; 1318 } 1319 1320 static int 1321 bnxt_hwrm_cfa_flow_stats_get(struct bnxt *bp, int num_flows, 1322 struct bnxt_tc_stats_batch stats_batch[]) 1323 { 1324 struct hwrm_cfa_flow_stats_output *resp = bp->hwrm_cmd_resp_addr; 1325 struct hwrm_cfa_flow_stats_input req = { 0 }; 1326 __le16 *req_flow_handles = &req.flow_handle_0; 1327 int rc, i; 1328 1329 bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_FLOW_STATS, -1, -1); 1330 req.num_flows = cpu_to_le16(num_flows); 1331 for (i = 0; i < num_flows; i++) { 1332 struct bnxt_tc_flow_node *flow_node = stats_batch[i].flow_node; 1333 1334 req_flow_handles[i] = flow_node->flow_handle; 1335 } 1336 1337 mutex_lock(&bp->hwrm_cmd_lock); 1338 rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT); 1339 if (!rc) { 1340 __le64 *resp_packets = &resp->packet_0; 1341 __le64 *resp_bytes = &resp->byte_0; 1342 1343 for (i = 0; i < num_flows; i++) { 1344 stats_batch[i].hw_stats.packets = 1345 le64_to_cpu(resp_packets[i]); 1346 stats_batch[i].hw_stats.bytes = 1347 le64_to_cpu(resp_bytes[i]); 1348 } 1349 } else { 1350 netdev_info(bp->dev, "error rc=%d", rc); 1351 } 1352 1353 mutex_unlock(&bp->hwrm_cmd_lock); 1354 return rc; 1355 } 1356 1357 /* Add val to accum while handling a possible wraparound 1358 * of val. Eventhough val is of type u64, its actual width 1359 * is denoted by mask and will wrap-around beyond that width. 1360 */ 1361 static void accumulate_val(u64 *accum, u64 val, u64 mask) 1362 { 1363 #define low_bits(x, mask) ((x) & (mask)) 1364 #define high_bits(x, mask) ((x) & ~(mask)) 1365 bool wrapped = val < low_bits(*accum, mask); 1366 1367 *accum = high_bits(*accum, mask) + val; 1368 if (wrapped) 1369 *accum += (mask + 1); 1370 } 1371 1372 /* The HW counters' width is much less than 64bits. 1373 * Handle possible wrap-around while updating the stat counters 1374 */ 1375 static void bnxt_flow_stats_accum(struct bnxt_tc_info *tc_info, 1376 struct bnxt_tc_flow_stats *acc_stats, 1377 struct bnxt_tc_flow_stats *hw_stats) 1378 { 1379 accumulate_val(&acc_stats->bytes, hw_stats->bytes, tc_info->bytes_mask); 1380 accumulate_val(&acc_stats->packets, hw_stats->packets, 1381 tc_info->packets_mask); 1382 } 1383 1384 static int 1385 bnxt_tc_flow_stats_batch_update(struct bnxt *bp, int num_flows, 1386 struct bnxt_tc_stats_batch stats_batch[]) 1387 { 1388 struct bnxt_tc_info *tc_info = bp->tc_info; 1389 int rc, i; 1390 1391 rc = bnxt_hwrm_cfa_flow_stats_get(bp, num_flows, stats_batch); 1392 if (rc) 1393 return rc; 1394 1395 for (i = 0; i < num_flows; i++) { 1396 struct bnxt_tc_flow_node *flow_node = stats_batch[i].flow_node; 1397 struct bnxt_tc_flow *flow = &flow_node->flow; 1398 1399 spin_lock(&flow->stats_lock); 1400 bnxt_flow_stats_accum(tc_info, &flow->stats, 1401 &stats_batch[i].hw_stats); 1402 if (flow->stats.packets != flow->prev_stats.packets) 1403 flow->lastused = jiffies; 1404 spin_unlock(&flow->stats_lock); 1405 } 1406 1407 return 0; 1408 } 1409 1410 static int 1411 bnxt_tc_flow_stats_batch_prep(struct bnxt *bp, 1412 struct bnxt_tc_stats_batch stats_batch[], 1413 int *num_flows) 1414 { 1415 struct bnxt_tc_info *tc_info = bp->tc_info; 1416 struct rhashtable_iter *iter = &tc_info->iter; 1417 void *flow_node; 1418 int rc, i; 1419 1420 rc = rhashtable_walk_start(iter); 1421 if (rc && rc != -EAGAIN) { 1422 i = 0; 1423 goto done; 1424 } 1425 1426 rc = 0; 1427 for (i = 0; i < BNXT_FLOW_STATS_BATCH_MAX; i++) { 1428 flow_node = rhashtable_walk_next(iter); 1429 if (IS_ERR(flow_node)) { 1430 i = 0; 1431 if (PTR_ERR(flow_node) == -EAGAIN) { 1432 continue; 1433 } else { 1434 rc = PTR_ERR(flow_node); 1435 goto done; 1436 } 1437 } 1438 1439 /* No more flows */ 1440 if (!flow_node) 1441 goto done; 1442 1443 stats_batch[i].flow_node = flow_node; 1444 } 1445 done: 1446 rhashtable_walk_stop(iter); 1447 *num_flows = i; 1448 return rc; 1449 } 1450 1451 void bnxt_tc_flow_stats_work(struct bnxt *bp) 1452 { 1453 struct bnxt_tc_info *tc_info = bp->tc_info; 1454 int num_flows, rc; 1455 1456 num_flows = atomic_read(&tc_info->flow_table.nelems); 1457 if (!num_flows) 1458 return; 1459 1460 rhashtable_walk_enter(&tc_info->flow_table, &tc_info->iter); 1461 1462 for (;;) { 1463 rc = bnxt_tc_flow_stats_batch_prep(bp, tc_info->stats_batch, 1464 &num_flows); 1465 if (rc) { 1466 if (rc == -EAGAIN) 1467 continue; 1468 break; 1469 } 1470 1471 if (!num_flows) 1472 break; 1473 1474 bnxt_tc_flow_stats_batch_update(bp, num_flows, 1475 tc_info->stats_batch); 1476 } 1477 1478 rhashtable_walk_exit(&tc_info->iter); 1479 } 1480 1481 int bnxt_tc_setup_flower(struct bnxt *bp, u16 src_fid, 1482 struct tc_cls_flower_offload *cls_flower) 1483 { 1484 int rc = 0; 1485 1486 if (cls_flower->common.chain_index) 1487 return -EOPNOTSUPP; 1488 1489 switch (cls_flower->command) { 1490 case TC_CLSFLOWER_REPLACE: 1491 rc = bnxt_tc_add_flow(bp, src_fid, cls_flower); 1492 break; 1493 1494 case TC_CLSFLOWER_DESTROY: 1495 rc = bnxt_tc_del_flow(bp, cls_flower); 1496 break; 1497 1498 case TC_CLSFLOWER_STATS: 1499 rc = bnxt_tc_get_flow_stats(bp, cls_flower); 1500 break; 1501 } 1502 return rc; 1503 } 1504 1505 static const struct rhashtable_params bnxt_tc_flow_ht_params = { 1506 .head_offset = offsetof(struct bnxt_tc_flow_node, node), 1507 .key_offset = offsetof(struct bnxt_tc_flow_node, cookie), 1508 .key_len = sizeof(((struct bnxt_tc_flow_node *)0)->cookie), 1509 .automatic_shrinking = true 1510 }; 1511 1512 static const struct rhashtable_params bnxt_tc_l2_ht_params = { 1513 .head_offset = offsetof(struct bnxt_tc_l2_node, node), 1514 .key_offset = offsetof(struct bnxt_tc_l2_node, key), 1515 .key_len = BNXT_TC_L2_KEY_LEN, 1516 .automatic_shrinking = true 1517 }; 1518 1519 static const struct rhashtable_params bnxt_tc_decap_l2_ht_params = { 1520 .head_offset = offsetof(struct bnxt_tc_l2_node, node), 1521 .key_offset = offsetof(struct bnxt_tc_l2_node, key), 1522 .key_len = BNXT_TC_L2_KEY_LEN, 1523 .automatic_shrinking = true 1524 }; 1525 1526 static const struct rhashtable_params bnxt_tc_tunnel_ht_params = { 1527 .head_offset = offsetof(struct bnxt_tc_tunnel_node, node), 1528 .key_offset = offsetof(struct bnxt_tc_tunnel_node, key), 1529 .key_len = sizeof(struct ip_tunnel_key), 1530 .automatic_shrinking = true 1531 }; 1532 1533 /* convert counter width in bits to a mask */ 1534 #define mask(width) ((u64)~0 >> (64 - (width))) 1535 1536 int bnxt_init_tc(struct bnxt *bp) 1537 { 1538 struct bnxt_tc_info *tc_info; 1539 int rc; 1540 1541 if (bp->hwrm_spec_code < 0x10803) { 1542 netdev_warn(bp->dev, 1543 "Firmware does not support TC flower offload.\n"); 1544 return -ENOTSUPP; 1545 } 1546 1547 tc_info = kzalloc(sizeof(*tc_info), GFP_KERNEL); 1548 if (!tc_info) 1549 return -ENOMEM; 1550 mutex_init(&tc_info->lock); 1551 1552 /* Counter widths are programmed by FW */ 1553 tc_info->bytes_mask = mask(36); 1554 tc_info->packets_mask = mask(28); 1555 1556 tc_info->flow_ht_params = bnxt_tc_flow_ht_params; 1557 rc = rhashtable_init(&tc_info->flow_table, &tc_info->flow_ht_params); 1558 if (rc) 1559 goto free_tc_info; 1560 1561 tc_info->l2_ht_params = bnxt_tc_l2_ht_params; 1562 rc = rhashtable_init(&tc_info->l2_table, &tc_info->l2_ht_params); 1563 if (rc) 1564 goto destroy_flow_table; 1565 1566 tc_info->decap_l2_ht_params = bnxt_tc_decap_l2_ht_params; 1567 rc = rhashtable_init(&tc_info->decap_l2_table, 1568 &tc_info->decap_l2_ht_params); 1569 if (rc) 1570 goto destroy_l2_table; 1571 1572 tc_info->decap_ht_params = bnxt_tc_tunnel_ht_params; 1573 rc = rhashtable_init(&tc_info->decap_table, 1574 &tc_info->decap_ht_params); 1575 if (rc) 1576 goto destroy_decap_l2_table; 1577 1578 tc_info->encap_ht_params = bnxt_tc_tunnel_ht_params; 1579 rc = rhashtable_init(&tc_info->encap_table, 1580 &tc_info->encap_ht_params); 1581 if (rc) 1582 goto destroy_decap_table; 1583 1584 tc_info->enabled = true; 1585 bp->dev->hw_features |= NETIF_F_HW_TC; 1586 bp->dev->features |= NETIF_F_HW_TC; 1587 bp->tc_info = tc_info; 1588 return 0; 1589 1590 destroy_decap_table: 1591 rhashtable_destroy(&tc_info->decap_table); 1592 destroy_decap_l2_table: 1593 rhashtable_destroy(&tc_info->decap_l2_table); 1594 destroy_l2_table: 1595 rhashtable_destroy(&tc_info->l2_table); 1596 destroy_flow_table: 1597 rhashtable_destroy(&tc_info->flow_table); 1598 free_tc_info: 1599 kfree(tc_info); 1600 return rc; 1601 } 1602 1603 void bnxt_shutdown_tc(struct bnxt *bp) 1604 { 1605 struct bnxt_tc_info *tc_info = bp->tc_info; 1606 1607 if (!bnxt_tc_flower_enabled(bp)) 1608 return; 1609 1610 rhashtable_destroy(&tc_info->flow_table); 1611 rhashtable_destroy(&tc_info->l2_table); 1612 rhashtable_destroy(&tc_info->decap_l2_table); 1613 rhashtable_destroy(&tc_info->decap_table); 1614 rhashtable_destroy(&tc_info->encap_table); 1615 kfree(tc_info); 1616 bp->tc_info = NULL; 1617 } 1618