1 /* Broadcom NetXtreme-C/E network driver. 2 * 3 * Copyright (c) 2016-2017 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 #include <linux/pci.h> 10 #include <linux/netdevice.h> 11 #include <linux/etherdevice.h> 12 #include <linux/rtnetlink.h> 13 #include <linux/jhash.h> 14 #include <net/pkt_cls.h> 15 16 #include "bnxt_hsi.h" 17 #include "bnxt.h" 18 #include "bnxt_vfr.h" 19 #include "bnxt_devlink.h" 20 #include "bnxt_tc.h" 21 22 #ifdef CONFIG_BNXT_SRIOV 23 24 #define CFA_HANDLE_INVALID 0xffff 25 #define VF_IDX_INVALID 0xffff 26 27 static int hwrm_cfa_vfr_alloc(struct bnxt *bp, u16 vf_idx, 28 u16 *tx_cfa_action, u16 *rx_cfa_code) 29 { 30 struct hwrm_cfa_vfr_alloc_output *resp = bp->hwrm_cmd_resp_addr; 31 struct hwrm_cfa_vfr_alloc_input req = { 0 }; 32 int rc; 33 34 bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_VFR_ALLOC, -1, -1); 35 req.vf_id = cpu_to_le16(vf_idx); 36 sprintf(req.vfr_name, "vfr%d", vf_idx); 37 38 mutex_lock(&bp->hwrm_cmd_lock); 39 rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT); 40 if (!rc) { 41 *tx_cfa_action = le16_to_cpu(resp->tx_cfa_action); 42 *rx_cfa_code = le16_to_cpu(resp->rx_cfa_code); 43 netdev_dbg(bp->dev, "tx_cfa_action=0x%x, rx_cfa_code=0x%x", 44 *tx_cfa_action, *rx_cfa_code); 45 } else { 46 netdev_info(bp->dev, "%s error rc=%d", __func__, rc); 47 } 48 49 mutex_unlock(&bp->hwrm_cmd_lock); 50 return rc; 51 } 52 53 static int hwrm_cfa_vfr_free(struct bnxt *bp, u16 vf_idx) 54 { 55 struct hwrm_cfa_vfr_free_input req = { 0 }; 56 int rc; 57 58 bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_VFR_FREE, -1, -1); 59 sprintf(req.vfr_name, "vfr%d", vf_idx); 60 61 rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT); 62 if (rc) 63 netdev_info(bp->dev, "%s error rc=%d", __func__, rc); 64 return rc; 65 } 66 67 static int bnxt_hwrm_vfr_qcfg(struct bnxt *bp, struct bnxt_vf_rep *vf_rep, 68 u16 *max_mtu) 69 { 70 struct hwrm_func_qcfg_output *resp = bp->hwrm_cmd_resp_addr; 71 struct hwrm_func_qcfg_input req = {0}; 72 u16 mtu; 73 int rc; 74 75 bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_QCFG, -1, -1); 76 req.fid = cpu_to_le16(bp->pf.vf[vf_rep->vf_idx].fw_fid); 77 78 mutex_lock(&bp->hwrm_cmd_lock); 79 80 rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT); 81 if (!rc) { 82 mtu = le16_to_cpu(resp->max_mtu_configured); 83 if (!mtu) 84 *max_mtu = BNXT_MAX_MTU; 85 else 86 *max_mtu = mtu; 87 } 88 mutex_unlock(&bp->hwrm_cmd_lock); 89 return rc; 90 } 91 92 static int bnxt_vf_rep_open(struct net_device *dev) 93 { 94 struct bnxt_vf_rep *vf_rep = netdev_priv(dev); 95 struct bnxt *bp = vf_rep->bp; 96 97 /* Enable link and TX only if the parent PF is open. */ 98 if (netif_running(bp->dev)) { 99 netif_carrier_on(dev); 100 netif_tx_start_all_queues(dev); 101 } 102 return 0; 103 } 104 105 static int bnxt_vf_rep_close(struct net_device *dev) 106 { 107 netif_carrier_off(dev); 108 netif_tx_disable(dev); 109 110 return 0; 111 } 112 113 static netdev_tx_t bnxt_vf_rep_xmit(struct sk_buff *skb, 114 struct net_device *dev) 115 { 116 struct bnxt_vf_rep *vf_rep = netdev_priv(dev); 117 int rc, len = skb->len; 118 119 skb_dst_drop(skb); 120 dst_hold((struct dst_entry *)vf_rep->dst); 121 skb_dst_set(skb, (struct dst_entry *)vf_rep->dst); 122 skb->dev = vf_rep->dst->u.port_info.lower_dev; 123 124 rc = dev_queue_xmit(skb); 125 if (!rc) { 126 vf_rep->tx_stats.packets++; 127 vf_rep->tx_stats.bytes += len; 128 } 129 return rc; 130 } 131 132 static void 133 bnxt_vf_rep_get_stats64(struct net_device *dev, 134 struct rtnl_link_stats64 *stats) 135 { 136 struct bnxt_vf_rep *vf_rep = netdev_priv(dev); 137 138 stats->rx_packets = vf_rep->rx_stats.packets; 139 stats->rx_bytes = vf_rep->rx_stats.bytes; 140 stats->tx_packets = vf_rep->tx_stats.packets; 141 stats->tx_bytes = vf_rep->tx_stats.bytes; 142 } 143 144 static int bnxt_vf_rep_setup_tc_block_cb(enum tc_setup_type type, 145 void *type_data, 146 void *cb_priv) 147 { 148 struct bnxt_vf_rep *vf_rep = cb_priv; 149 struct bnxt *bp = vf_rep->bp; 150 int vf_fid = bp->pf.vf[vf_rep->vf_idx].fw_fid; 151 152 if (!bnxt_tc_flower_enabled(vf_rep->bp) || 153 !tc_cls_can_offload_and_chain0(bp->dev, type_data)) 154 return -EOPNOTSUPP; 155 156 switch (type) { 157 case TC_SETUP_CLSFLOWER: 158 return bnxt_tc_setup_flower(bp, vf_fid, type_data); 159 default: 160 return -EOPNOTSUPP; 161 } 162 } 163 164 static int bnxt_vf_rep_setup_tc_block(struct net_device *dev, 165 struct tc_block_offload *f) 166 { 167 struct bnxt_vf_rep *vf_rep = netdev_priv(dev); 168 169 if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS) 170 return -EOPNOTSUPP; 171 172 switch (f->command) { 173 case TC_BLOCK_BIND: 174 return tcf_block_cb_register(f->block, 175 bnxt_vf_rep_setup_tc_block_cb, 176 vf_rep, vf_rep, f->extack); 177 case TC_BLOCK_UNBIND: 178 tcf_block_cb_unregister(f->block, 179 bnxt_vf_rep_setup_tc_block_cb, vf_rep); 180 return 0; 181 default: 182 return -EOPNOTSUPP; 183 } 184 } 185 186 static int bnxt_vf_rep_setup_tc(struct net_device *dev, enum tc_setup_type type, 187 void *type_data) 188 { 189 switch (type) { 190 case TC_SETUP_BLOCK: 191 return bnxt_vf_rep_setup_tc_block(dev, type_data); 192 default: 193 return -EOPNOTSUPP; 194 } 195 } 196 197 struct net_device *bnxt_get_vf_rep(struct bnxt *bp, u16 cfa_code) 198 { 199 u16 vf_idx; 200 201 if (cfa_code && bp->cfa_code_map && BNXT_PF(bp)) { 202 vf_idx = bp->cfa_code_map[cfa_code]; 203 if (vf_idx != VF_IDX_INVALID) 204 return bp->vf_reps[vf_idx]->dev; 205 } 206 return NULL; 207 } 208 209 void bnxt_vf_rep_rx(struct bnxt *bp, struct sk_buff *skb) 210 { 211 struct bnxt_vf_rep *vf_rep = netdev_priv(skb->dev); 212 213 vf_rep->rx_stats.bytes += skb->len; 214 vf_rep->rx_stats.packets++; 215 216 netif_receive_skb(skb); 217 } 218 219 static int bnxt_vf_rep_get_phys_port_name(struct net_device *dev, char *buf, 220 size_t len) 221 { 222 struct bnxt_vf_rep *vf_rep = netdev_priv(dev); 223 struct pci_dev *pf_pdev = vf_rep->bp->pdev; 224 int rc; 225 226 rc = snprintf(buf, len, "pf%dvf%d", PCI_FUNC(pf_pdev->devfn), 227 vf_rep->vf_idx); 228 if (rc >= len) 229 return -EOPNOTSUPP; 230 return 0; 231 } 232 233 static void bnxt_vf_rep_get_drvinfo(struct net_device *dev, 234 struct ethtool_drvinfo *info) 235 { 236 strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver)); 237 strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version)); 238 } 239 240 static int bnxt_vf_rep_port_attr_get(struct net_device *dev, 241 struct switchdev_attr *attr) 242 { 243 struct bnxt_vf_rep *vf_rep = netdev_priv(dev); 244 245 /* as only PORT_PARENT_ID is supported currently use common code 246 * between PF and VF-rep for now. 247 */ 248 return bnxt_port_attr_get(vf_rep->bp, attr); 249 } 250 251 static const struct switchdev_ops bnxt_vf_rep_switchdev_ops = { 252 .switchdev_port_attr_get = bnxt_vf_rep_port_attr_get 253 }; 254 255 static const struct ethtool_ops bnxt_vf_rep_ethtool_ops = { 256 .get_drvinfo = bnxt_vf_rep_get_drvinfo 257 }; 258 259 static const struct net_device_ops bnxt_vf_rep_netdev_ops = { 260 .ndo_open = bnxt_vf_rep_open, 261 .ndo_stop = bnxt_vf_rep_close, 262 .ndo_start_xmit = bnxt_vf_rep_xmit, 263 .ndo_get_stats64 = bnxt_vf_rep_get_stats64, 264 .ndo_setup_tc = bnxt_vf_rep_setup_tc, 265 .ndo_get_phys_port_name = bnxt_vf_rep_get_phys_port_name 266 }; 267 268 bool bnxt_dev_is_vf_rep(struct net_device *dev) 269 { 270 return dev->netdev_ops == &bnxt_vf_rep_netdev_ops; 271 } 272 273 /* Called when the parent PF interface is closed: 274 * As the mode transition from SWITCHDEV to LEGACY 275 * happens under the rtnl_lock() this routine is safe 276 * under the rtnl_lock() 277 */ 278 void bnxt_vf_reps_close(struct bnxt *bp) 279 { 280 struct bnxt_vf_rep *vf_rep; 281 u16 num_vfs, i; 282 283 if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV) 284 return; 285 286 num_vfs = pci_num_vf(bp->pdev); 287 for (i = 0; i < num_vfs; i++) { 288 vf_rep = bp->vf_reps[i]; 289 if (netif_running(vf_rep->dev)) 290 bnxt_vf_rep_close(vf_rep->dev); 291 } 292 } 293 294 /* Called when the parent PF interface is opened (re-opened): 295 * As the mode transition from SWITCHDEV to LEGACY 296 * happen under the rtnl_lock() this routine is safe 297 * under the rtnl_lock() 298 */ 299 void bnxt_vf_reps_open(struct bnxt *bp) 300 { 301 int i; 302 303 if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV) 304 return; 305 306 for (i = 0; i < pci_num_vf(bp->pdev); i++) 307 bnxt_vf_rep_open(bp->vf_reps[i]->dev); 308 } 309 310 static void __bnxt_vf_reps_destroy(struct bnxt *bp) 311 { 312 u16 num_vfs = pci_num_vf(bp->pdev); 313 struct bnxt_vf_rep *vf_rep; 314 int i; 315 316 for (i = 0; i < num_vfs; i++) { 317 vf_rep = bp->vf_reps[i]; 318 if (vf_rep) { 319 dst_release((struct dst_entry *)vf_rep->dst); 320 321 if (vf_rep->tx_cfa_action != CFA_HANDLE_INVALID) 322 hwrm_cfa_vfr_free(bp, vf_rep->vf_idx); 323 324 if (vf_rep->dev) { 325 /* if register_netdev failed, then netdev_ops 326 * would have been set to NULL 327 */ 328 if (vf_rep->dev->netdev_ops) 329 unregister_netdev(vf_rep->dev); 330 free_netdev(vf_rep->dev); 331 } 332 } 333 } 334 335 kfree(bp->vf_reps); 336 bp->vf_reps = NULL; 337 } 338 339 void bnxt_vf_reps_destroy(struct bnxt *bp) 340 { 341 bool closed = false; 342 343 if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV) 344 return; 345 346 if (!bp->vf_reps) 347 return; 348 349 /* Ensure that parent PF's and VF-reps' RX/TX has been quiesced 350 * before proceeding with VF-rep cleanup. 351 */ 352 rtnl_lock(); 353 if (netif_running(bp->dev)) { 354 bnxt_close_nic(bp, false, false); 355 closed = true; 356 } 357 /* un-publish cfa_code_map so that RX path can't see it anymore */ 358 kfree(bp->cfa_code_map); 359 bp->cfa_code_map = NULL; 360 bp->eswitch_mode = DEVLINK_ESWITCH_MODE_LEGACY; 361 362 if (closed) 363 bnxt_open_nic(bp, false, false); 364 rtnl_unlock(); 365 366 /* Need to call vf_reps_destroy() outside of rntl_lock 367 * as unregister_netdev takes rtnl_lock 368 */ 369 __bnxt_vf_reps_destroy(bp); 370 } 371 372 /* Use the OUI of the PF's perm addr and report the same mac addr 373 * for the same VF-rep each time 374 */ 375 static void bnxt_vf_rep_eth_addr_gen(u8 *src_mac, u16 vf_idx, u8 *mac) 376 { 377 u32 addr; 378 379 ether_addr_copy(mac, src_mac); 380 381 addr = jhash(src_mac, ETH_ALEN, 0) + vf_idx; 382 mac[3] = (u8)(addr & 0xFF); 383 mac[4] = (u8)((addr >> 8) & 0xFF); 384 mac[5] = (u8)((addr >> 16) & 0xFF); 385 } 386 387 static void bnxt_vf_rep_netdev_init(struct bnxt *bp, struct bnxt_vf_rep *vf_rep, 388 struct net_device *dev) 389 { 390 struct net_device *pf_dev = bp->dev; 391 u16 max_mtu; 392 393 dev->netdev_ops = &bnxt_vf_rep_netdev_ops; 394 dev->ethtool_ops = &bnxt_vf_rep_ethtool_ops; 395 SWITCHDEV_SET_OPS(dev, &bnxt_vf_rep_switchdev_ops); 396 /* Just inherit all the featues of the parent PF as the VF-R 397 * uses the RX/TX rings of the parent PF 398 */ 399 dev->hw_features = pf_dev->hw_features; 400 dev->gso_partial_features = pf_dev->gso_partial_features; 401 dev->vlan_features = pf_dev->vlan_features; 402 dev->hw_enc_features = pf_dev->hw_enc_features; 403 dev->features |= pf_dev->features; 404 bnxt_vf_rep_eth_addr_gen(bp->pf.mac_addr, vf_rep->vf_idx, 405 dev->perm_addr); 406 ether_addr_copy(dev->dev_addr, dev->perm_addr); 407 /* Set VF-Rep's max-mtu to the corresponding VF's max-mtu */ 408 if (!bnxt_hwrm_vfr_qcfg(bp, vf_rep, &max_mtu)) 409 dev->max_mtu = max_mtu; 410 dev->min_mtu = ETH_ZLEN; 411 } 412 413 static int bnxt_pcie_dsn_get(struct bnxt *bp, u8 dsn[]) 414 { 415 struct pci_dev *pdev = bp->pdev; 416 int pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_DSN); 417 u32 dw; 418 419 if (!pos) { 420 netdev_info(bp->dev, "Unable do read adapter's DSN"); 421 return -EOPNOTSUPP; 422 } 423 424 /* DSN (two dw) is at an offset of 4 from the cap pos */ 425 pos += 4; 426 pci_read_config_dword(pdev, pos, &dw); 427 put_unaligned_le32(dw, &dsn[0]); 428 pci_read_config_dword(pdev, pos + 4, &dw); 429 put_unaligned_le32(dw, &dsn[4]); 430 return 0; 431 } 432 433 static int bnxt_vf_reps_create(struct bnxt *bp) 434 { 435 u16 *cfa_code_map = NULL, num_vfs = pci_num_vf(bp->pdev); 436 struct bnxt_vf_rep *vf_rep; 437 struct net_device *dev; 438 int rc, i; 439 440 bp->vf_reps = kcalloc(num_vfs, sizeof(vf_rep), GFP_KERNEL); 441 if (!bp->vf_reps) 442 return -ENOMEM; 443 444 /* storage for cfa_code to vf-idx mapping */ 445 cfa_code_map = kmalloc_array(MAX_CFA_CODE, sizeof(*bp->cfa_code_map), 446 GFP_KERNEL); 447 if (!cfa_code_map) { 448 rc = -ENOMEM; 449 goto err; 450 } 451 for (i = 0; i < MAX_CFA_CODE; i++) 452 cfa_code_map[i] = VF_IDX_INVALID; 453 454 for (i = 0; i < num_vfs; i++) { 455 dev = alloc_etherdev(sizeof(*vf_rep)); 456 if (!dev) { 457 rc = -ENOMEM; 458 goto err; 459 } 460 461 vf_rep = netdev_priv(dev); 462 bp->vf_reps[i] = vf_rep; 463 vf_rep->dev = dev; 464 vf_rep->bp = bp; 465 vf_rep->vf_idx = i; 466 vf_rep->tx_cfa_action = CFA_HANDLE_INVALID; 467 468 /* get cfa handles from FW */ 469 rc = hwrm_cfa_vfr_alloc(bp, vf_rep->vf_idx, 470 &vf_rep->tx_cfa_action, 471 &vf_rep->rx_cfa_code); 472 if (rc) { 473 rc = -ENOLINK; 474 goto err; 475 } 476 cfa_code_map[vf_rep->rx_cfa_code] = vf_rep->vf_idx; 477 478 vf_rep->dst = metadata_dst_alloc(0, METADATA_HW_PORT_MUX, 479 GFP_KERNEL); 480 if (!vf_rep->dst) { 481 rc = -ENOMEM; 482 goto err; 483 } 484 /* only cfa_action is needed to mux a packet while TXing */ 485 vf_rep->dst->u.port_info.port_id = vf_rep->tx_cfa_action; 486 vf_rep->dst->u.port_info.lower_dev = bp->dev; 487 488 bnxt_vf_rep_netdev_init(bp, vf_rep, dev); 489 rc = register_netdev(dev); 490 if (rc) { 491 /* no need for unregister_netdev in cleanup */ 492 dev->netdev_ops = NULL; 493 goto err; 494 } 495 } 496 497 /* Read the adapter's DSN to use as the eswitch switch_id */ 498 rc = bnxt_pcie_dsn_get(bp, bp->switch_id); 499 if (rc) 500 goto err; 501 502 /* publish cfa_code_map only after all VF-reps have been initialized */ 503 bp->cfa_code_map = cfa_code_map; 504 bp->eswitch_mode = DEVLINK_ESWITCH_MODE_SWITCHDEV; 505 netif_keep_dst(bp->dev); 506 return 0; 507 508 err: 509 netdev_info(bp->dev, "%s error=%d", __func__, rc); 510 kfree(cfa_code_map); 511 __bnxt_vf_reps_destroy(bp); 512 return rc; 513 } 514 515 /* Devlink related routines */ 516 int bnxt_dl_eswitch_mode_get(struct devlink *devlink, u16 *mode) 517 { 518 struct bnxt *bp = bnxt_get_bp_from_dl(devlink); 519 520 *mode = bp->eswitch_mode; 521 return 0; 522 } 523 524 int bnxt_dl_eswitch_mode_set(struct devlink *devlink, u16 mode, 525 struct netlink_ext_ack *extack) 526 { 527 struct bnxt *bp = bnxt_get_bp_from_dl(devlink); 528 int rc = 0; 529 530 mutex_lock(&bp->sriov_lock); 531 if (bp->eswitch_mode == mode) { 532 netdev_info(bp->dev, "already in %s eswitch mode", 533 mode == DEVLINK_ESWITCH_MODE_LEGACY ? 534 "legacy" : "switchdev"); 535 rc = -EINVAL; 536 goto done; 537 } 538 539 switch (mode) { 540 case DEVLINK_ESWITCH_MODE_LEGACY: 541 bnxt_vf_reps_destroy(bp); 542 break; 543 544 case DEVLINK_ESWITCH_MODE_SWITCHDEV: 545 if (bp->hwrm_spec_code < 0x10803) { 546 netdev_warn(bp->dev, "FW does not support SRIOV E-Switch SWITCHDEV mode\n"); 547 rc = -ENOTSUPP; 548 goto done; 549 } 550 551 if (pci_num_vf(bp->pdev) == 0) { 552 netdev_info(bp->dev, "Enable VFs before setting switchdev mode"); 553 rc = -EPERM; 554 goto done; 555 } 556 rc = bnxt_vf_reps_create(bp); 557 break; 558 559 default: 560 rc = -EINVAL; 561 goto done; 562 } 563 done: 564 mutex_unlock(&bp->sriov_lock); 565 return rc; 566 } 567 568 #endif 569