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_get_port_parent_id(struct net_device *dev, 241 struct netdev_phys_item_id *ppid) 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_get_port_parent_id(vf_rep->bp->dev, ppid); 249 } 250 251 static const struct ethtool_ops bnxt_vf_rep_ethtool_ops = { 252 .get_drvinfo = bnxt_vf_rep_get_drvinfo 253 }; 254 255 static const struct net_device_ops bnxt_vf_rep_netdev_ops = { 256 .ndo_open = bnxt_vf_rep_open, 257 .ndo_stop = bnxt_vf_rep_close, 258 .ndo_start_xmit = bnxt_vf_rep_xmit, 259 .ndo_get_stats64 = bnxt_vf_rep_get_stats64, 260 .ndo_setup_tc = bnxt_vf_rep_setup_tc, 261 .ndo_get_port_parent_id = bnxt_vf_rep_get_port_parent_id, 262 .ndo_get_phys_port_name = bnxt_vf_rep_get_phys_port_name 263 }; 264 265 bool bnxt_dev_is_vf_rep(struct net_device *dev) 266 { 267 return dev->netdev_ops == &bnxt_vf_rep_netdev_ops; 268 } 269 270 /* Called when the parent PF interface is closed: 271 * As the mode transition from SWITCHDEV to LEGACY 272 * happens under the rtnl_lock() this routine is safe 273 * under the rtnl_lock() 274 */ 275 void bnxt_vf_reps_close(struct bnxt *bp) 276 { 277 struct bnxt_vf_rep *vf_rep; 278 u16 num_vfs, i; 279 280 if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV) 281 return; 282 283 num_vfs = pci_num_vf(bp->pdev); 284 for (i = 0; i < num_vfs; i++) { 285 vf_rep = bp->vf_reps[i]; 286 if (netif_running(vf_rep->dev)) 287 bnxt_vf_rep_close(vf_rep->dev); 288 } 289 } 290 291 /* Called when the parent PF interface is opened (re-opened): 292 * As the mode transition from SWITCHDEV to LEGACY 293 * happen under the rtnl_lock() this routine is safe 294 * under the rtnl_lock() 295 */ 296 void bnxt_vf_reps_open(struct bnxt *bp) 297 { 298 int i; 299 300 if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV) 301 return; 302 303 for (i = 0; i < pci_num_vf(bp->pdev); i++) 304 bnxt_vf_rep_open(bp->vf_reps[i]->dev); 305 } 306 307 static void __bnxt_vf_reps_destroy(struct bnxt *bp) 308 { 309 u16 num_vfs = pci_num_vf(bp->pdev); 310 struct bnxt_vf_rep *vf_rep; 311 int i; 312 313 for (i = 0; i < num_vfs; i++) { 314 vf_rep = bp->vf_reps[i]; 315 if (vf_rep) { 316 dst_release((struct dst_entry *)vf_rep->dst); 317 318 if (vf_rep->tx_cfa_action != CFA_HANDLE_INVALID) 319 hwrm_cfa_vfr_free(bp, vf_rep->vf_idx); 320 321 if (vf_rep->dev) { 322 /* if register_netdev failed, then netdev_ops 323 * would have been set to NULL 324 */ 325 if (vf_rep->dev->netdev_ops) 326 unregister_netdev(vf_rep->dev); 327 free_netdev(vf_rep->dev); 328 } 329 } 330 } 331 332 kfree(bp->vf_reps); 333 bp->vf_reps = NULL; 334 } 335 336 void bnxt_vf_reps_destroy(struct bnxt *bp) 337 { 338 bool closed = false; 339 340 if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV) 341 return; 342 343 if (!bp->vf_reps) 344 return; 345 346 /* Ensure that parent PF's and VF-reps' RX/TX has been quiesced 347 * before proceeding with VF-rep cleanup. 348 */ 349 rtnl_lock(); 350 if (netif_running(bp->dev)) { 351 bnxt_close_nic(bp, false, false); 352 closed = true; 353 } 354 /* un-publish cfa_code_map so that RX path can't see it anymore */ 355 kfree(bp->cfa_code_map); 356 bp->cfa_code_map = NULL; 357 bp->eswitch_mode = DEVLINK_ESWITCH_MODE_LEGACY; 358 359 if (closed) 360 bnxt_open_nic(bp, false, false); 361 rtnl_unlock(); 362 363 /* Need to call vf_reps_destroy() outside of rntl_lock 364 * as unregister_netdev takes rtnl_lock 365 */ 366 __bnxt_vf_reps_destroy(bp); 367 } 368 369 /* Use the OUI of the PF's perm addr and report the same mac addr 370 * for the same VF-rep each time 371 */ 372 static void bnxt_vf_rep_eth_addr_gen(u8 *src_mac, u16 vf_idx, u8 *mac) 373 { 374 u32 addr; 375 376 ether_addr_copy(mac, src_mac); 377 378 addr = jhash(src_mac, ETH_ALEN, 0) + vf_idx; 379 mac[3] = (u8)(addr & 0xFF); 380 mac[4] = (u8)((addr >> 8) & 0xFF); 381 mac[5] = (u8)((addr >> 16) & 0xFF); 382 } 383 384 static void bnxt_vf_rep_netdev_init(struct bnxt *bp, struct bnxt_vf_rep *vf_rep, 385 struct net_device *dev) 386 { 387 struct net_device *pf_dev = bp->dev; 388 u16 max_mtu; 389 390 dev->netdev_ops = &bnxt_vf_rep_netdev_ops; 391 dev->ethtool_ops = &bnxt_vf_rep_ethtool_ops; 392 /* Just inherit all the featues of the parent PF as the VF-R 393 * uses the RX/TX rings of the parent PF 394 */ 395 dev->hw_features = pf_dev->hw_features; 396 dev->gso_partial_features = pf_dev->gso_partial_features; 397 dev->vlan_features = pf_dev->vlan_features; 398 dev->hw_enc_features = pf_dev->hw_enc_features; 399 dev->features |= pf_dev->features; 400 bnxt_vf_rep_eth_addr_gen(bp->pf.mac_addr, vf_rep->vf_idx, 401 dev->perm_addr); 402 ether_addr_copy(dev->dev_addr, dev->perm_addr); 403 /* Set VF-Rep's max-mtu to the corresponding VF's max-mtu */ 404 if (!bnxt_hwrm_vfr_qcfg(bp, vf_rep, &max_mtu)) 405 dev->max_mtu = max_mtu; 406 dev->min_mtu = ETH_ZLEN; 407 } 408 409 static int bnxt_pcie_dsn_get(struct bnxt *bp, u8 dsn[]) 410 { 411 struct pci_dev *pdev = bp->pdev; 412 int pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_DSN); 413 u32 dw; 414 415 if (!pos) { 416 netdev_info(bp->dev, "Unable do read adapter's DSN"); 417 return -EOPNOTSUPP; 418 } 419 420 /* DSN (two dw) is at an offset of 4 from the cap pos */ 421 pos += 4; 422 pci_read_config_dword(pdev, pos, &dw); 423 put_unaligned_le32(dw, &dsn[0]); 424 pci_read_config_dword(pdev, pos + 4, &dw); 425 put_unaligned_le32(dw, &dsn[4]); 426 return 0; 427 } 428 429 static int bnxt_vf_reps_create(struct bnxt *bp) 430 { 431 u16 *cfa_code_map = NULL, num_vfs = pci_num_vf(bp->pdev); 432 struct bnxt_vf_rep *vf_rep; 433 struct net_device *dev; 434 int rc, i; 435 436 bp->vf_reps = kcalloc(num_vfs, sizeof(vf_rep), GFP_KERNEL); 437 if (!bp->vf_reps) 438 return -ENOMEM; 439 440 /* storage for cfa_code to vf-idx mapping */ 441 cfa_code_map = kmalloc_array(MAX_CFA_CODE, sizeof(*bp->cfa_code_map), 442 GFP_KERNEL); 443 if (!cfa_code_map) { 444 rc = -ENOMEM; 445 goto err; 446 } 447 for (i = 0; i < MAX_CFA_CODE; i++) 448 cfa_code_map[i] = VF_IDX_INVALID; 449 450 for (i = 0; i < num_vfs; i++) { 451 dev = alloc_etherdev(sizeof(*vf_rep)); 452 if (!dev) { 453 rc = -ENOMEM; 454 goto err; 455 } 456 457 vf_rep = netdev_priv(dev); 458 bp->vf_reps[i] = vf_rep; 459 vf_rep->dev = dev; 460 vf_rep->bp = bp; 461 vf_rep->vf_idx = i; 462 vf_rep->tx_cfa_action = CFA_HANDLE_INVALID; 463 464 /* get cfa handles from FW */ 465 rc = hwrm_cfa_vfr_alloc(bp, vf_rep->vf_idx, 466 &vf_rep->tx_cfa_action, 467 &vf_rep->rx_cfa_code); 468 if (rc) { 469 rc = -ENOLINK; 470 goto err; 471 } 472 cfa_code_map[vf_rep->rx_cfa_code] = vf_rep->vf_idx; 473 474 vf_rep->dst = metadata_dst_alloc(0, METADATA_HW_PORT_MUX, 475 GFP_KERNEL); 476 if (!vf_rep->dst) { 477 rc = -ENOMEM; 478 goto err; 479 } 480 /* only cfa_action is needed to mux a packet while TXing */ 481 vf_rep->dst->u.port_info.port_id = vf_rep->tx_cfa_action; 482 vf_rep->dst->u.port_info.lower_dev = bp->dev; 483 484 bnxt_vf_rep_netdev_init(bp, vf_rep, dev); 485 rc = register_netdev(dev); 486 if (rc) { 487 /* no need for unregister_netdev in cleanup */ 488 dev->netdev_ops = NULL; 489 goto err; 490 } 491 } 492 493 /* Read the adapter's DSN to use as the eswitch switch_id */ 494 rc = bnxt_pcie_dsn_get(bp, bp->switch_id); 495 if (rc) 496 goto err; 497 498 /* publish cfa_code_map only after all VF-reps have been initialized */ 499 bp->cfa_code_map = cfa_code_map; 500 bp->eswitch_mode = DEVLINK_ESWITCH_MODE_SWITCHDEV; 501 netif_keep_dst(bp->dev); 502 return 0; 503 504 err: 505 netdev_info(bp->dev, "%s error=%d", __func__, rc); 506 kfree(cfa_code_map); 507 __bnxt_vf_reps_destroy(bp); 508 return rc; 509 } 510 511 /* Devlink related routines */ 512 int bnxt_dl_eswitch_mode_get(struct devlink *devlink, u16 *mode) 513 { 514 struct bnxt *bp = bnxt_get_bp_from_dl(devlink); 515 516 *mode = bp->eswitch_mode; 517 return 0; 518 } 519 520 int bnxt_dl_eswitch_mode_set(struct devlink *devlink, u16 mode, 521 struct netlink_ext_ack *extack) 522 { 523 struct bnxt *bp = bnxt_get_bp_from_dl(devlink); 524 int rc = 0; 525 526 mutex_lock(&bp->sriov_lock); 527 if (bp->eswitch_mode == mode) { 528 netdev_info(bp->dev, "already in %s eswitch mode", 529 mode == DEVLINK_ESWITCH_MODE_LEGACY ? 530 "legacy" : "switchdev"); 531 rc = -EINVAL; 532 goto done; 533 } 534 535 switch (mode) { 536 case DEVLINK_ESWITCH_MODE_LEGACY: 537 bnxt_vf_reps_destroy(bp); 538 break; 539 540 case DEVLINK_ESWITCH_MODE_SWITCHDEV: 541 if (bp->hwrm_spec_code < 0x10803) { 542 netdev_warn(bp->dev, "FW does not support SRIOV E-Switch SWITCHDEV mode\n"); 543 rc = -ENOTSUPP; 544 goto done; 545 } 546 547 if (pci_num_vf(bp->pdev) == 0) { 548 netdev_info(bp->dev, "Enable VFs before setting switchdev mode"); 549 rc = -EPERM; 550 goto done; 551 } 552 rc = bnxt_vf_reps_create(bp); 553 break; 554 555 default: 556 rc = -EINVAL; 557 goto done; 558 } 559 done: 560 mutex_unlock(&bp->sriov_lock); 561 return rc; 562 } 563 564 #endif 565