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_hwrm.h" 19 #include "bnxt_vfr.h" 20 #include "bnxt_devlink.h" 21 #include "bnxt_tc.h" 22 23 #ifdef CONFIG_BNXT_SRIOV 24 25 #define CFA_HANDLE_INVALID 0xffff 26 #define VF_IDX_INVALID 0xffff 27 28 static int hwrm_cfa_vfr_alloc(struct bnxt *bp, u16 vf_idx, 29 u16 *tx_cfa_action, u16 *rx_cfa_code) 30 { 31 struct hwrm_cfa_vfr_alloc_output *resp; 32 struct hwrm_cfa_vfr_alloc_input *req; 33 int rc; 34 35 rc = hwrm_req_init(bp, req, HWRM_CFA_VFR_ALLOC); 36 if (!rc) { 37 req->vf_id = cpu_to_le16(vf_idx); 38 sprintf(req->vfr_name, "vfr%d", vf_idx); 39 40 resp = hwrm_req_hold(bp, req); 41 rc = hwrm_req_send(bp, req); 42 if (!rc) { 43 *tx_cfa_action = le16_to_cpu(resp->tx_cfa_action); 44 *rx_cfa_code = le16_to_cpu(resp->rx_cfa_code); 45 netdev_dbg(bp->dev, "tx_cfa_action=0x%x, rx_cfa_code=0x%x", 46 *tx_cfa_action, *rx_cfa_code); 47 } 48 hwrm_req_drop(bp, req); 49 } 50 if (rc) 51 netdev_info(bp->dev, "%s error rc=%d\n", __func__, rc); 52 return rc; 53 } 54 55 static int hwrm_cfa_vfr_free(struct bnxt *bp, u16 vf_idx) 56 { 57 struct hwrm_cfa_vfr_free_input *req; 58 int rc; 59 60 rc = hwrm_req_init(bp, req, HWRM_CFA_VFR_FREE); 61 if (!rc) { 62 sprintf(req->vfr_name, "vfr%d", vf_idx); 63 rc = hwrm_req_send(bp, req); 64 } 65 if (rc) 66 netdev_info(bp->dev, "%s error rc=%d\n", __func__, rc); 67 return rc; 68 } 69 70 static int bnxt_hwrm_vfr_qcfg(struct bnxt *bp, struct bnxt_vf_rep *vf_rep, 71 u16 *max_mtu) 72 { 73 struct hwrm_func_qcfg_output *resp; 74 struct hwrm_func_qcfg_input *req; 75 u16 mtu; 76 int rc; 77 78 rc = hwrm_req_init(bp, req, HWRM_FUNC_QCFG); 79 if (rc) 80 return rc; 81 82 req->fid = cpu_to_le16(bp->pf.vf[vf_rep->vf_idx].fw_fid); 83 resp = hwrm_req_hold(bp, req); 84 rc = hwrm_req_send(bp, req); 85 if (!rc) { 86 mtu = le16_to_cpu(resp->max_mtu_configured); 87 if (!mtu) 88 *max_mtu = BNXT_MAX_MTU; 89 else 90 *max_mtu = mtu; 91 } 92 hwrm_req_drop(bp, req); 93 return rc; 94 } 95 96 static int bnxt_vf_rep_open(struct net_device *dev) 97 { 98 struct bnxt_vf_rep *vf_rep = netdev_priv(dev); 99 struct bnxt *bp = vf_rep->bp; 100 101 /* Enable link and TX only if the parent PF is open. */ 102 if (netif_running(bp->dev)) { 103 netif_carrier_on(dev); 104 netif_tx_start_all_queues(dev); 105 } 106 return 0; 107 } 108 109 static int bnxt_vf_rep_close(struct net_device *dev) 110 { 111 netif_carrier_off(dev); 112 netif_tx_disable(dev); 113 114 return 0; 115 } 116 117 static netdev_tx_t bnxt_vf_rep_xmit(struct sk_buff *skb, 118 struct net_device *dev) 119 { 120 struct bnxt_vf_rep *vf_rep = netdev_priv(dev); 121 int rc, len = skb->len; 122 123 skb_dst_drop(skb); 124 dst_hold((struct dst_entry *)vf_rep->dst); 125 skb_dst_set(skb, (struct dst_entry *)vf_rep->dst); 126 skb->dev = vf_rep->dst->u.port_info.lower_dev; 127 128 rc = dev_queue_xmit(skb); 129 if (!rc) { 130 vf_rep->tx_stats.packets++; 131 vf_rep->tx_stats.bytes += len; 132 } 133 return rc; 134 } 135 136 static void 137 bnxt_vf_rep_get_stats64(struct net_device *dev, 138 struct rtnl_link_stats64 *stats) 139 { 140 struct bnxt_vf_rep *vf_rep = netdev_priv(dev); 141 142 stats->rx_packets = vf_rep->rx_stats.packets; 143 stats->rx_bytes = vf_rep->rx_stats.bytes; 144 stats->tx_packets = vf_rep->tx_stats.packets; 145 stats->tx_bytes = vf_rep->tx_stats.bytes; 146 } 147 148 static int bnxt_vf_rep_setup_tc_block_cb(enum tc_setup_type type, 149 void *type_data, 150 void *cb_priv) 151 { 152 struct bnxt_vf_rep *vf_rep = cb_priv; 153 struct bnxt *bp = vf_rep->bp; 154 int vf_fid = bp->pf.vf[vf_rep->vf_idx].fw_fid; 155 156 if (!bnxt_tc_flower_enabled(vf_rep->bp) || 157 !tc_cls_can_offload_and_chain0(bp->dev, type_data)) 158 return -EOPNOTSUPP; 159 160 switch (type) { 161 case TC_SETUP_CLSFLOWER: 162 return bnxt_tc_setup_flower(bp, vf_fid, type_data); 163 default: 164 return -EOPNOTSUPP; 165 } 166 } 167 168 static LIST_HEAD(bnxt_vf_block_cb_list); 169 170 static int bnxt_vf_rep_setup_tc(struct net_device *dev, enum tc_setup_type type, 171 void *type_data) 172 { 173 struct bnxt_vf_rep *vf_rep = netdev_priv(dev); 174 175 switch (type) { 176 case TC_SETUP_BLOCK: 177 return flow_block_cb_setup_simple(type_data, 178 &bnxt_vf_block_cb_list, 179 bnxt_vf_rep_setup_tc_block_cb, 180 vf_rep, vf_rep, true); 181 default: 182 return -EOPNOTSUPP; 183 } 184 } 185 186 struct net_device *bnxt_get_vf_rep(struct bnxt *bp, u16 cfa_code) 187 { 188 u16 vf_idx; 189 190 if (cfa_code && bp->cfa_code_map && BNXT_PF(bp)) { 191 vf_idx = bp->cfa_code_map[cfa_code]; 192 if (vf_idx != VF_IDX_INVALID) 193 return bp->vf_reps[vf_idx]->dev; 194 } 195 return NULL; 196 } 197 198 void bnxt_vf_rep_rx(struct bnxt *bp, struct sk_buff *skb) 199 { 200 struct bnxt_vf_rep *vf_rep = netdev_priv(skb->dev); 201 202 vf_rep->rx_stats.bytes += skb->len; 203 vf_rep->rx_stats.packets++; 204 205 netif_receive_skb(skb); 206 } 207 208 static int bnxt_vf_rep_get_phys_port_name(struct net_device *dev, char *buf, 209 size_t len) 210 { 211 struct bnxt_vf_rep *vf_rep = netdev_priv(dev); 212 struct pci_dev *pf_pdev = vf_rep->bp->pdev; 213 int rc; 214 215 rc = snprintf(buf, len, "pf%dvf%d", PCI_FUNC(pf_pdev->devfn), 216 vf_rep->vf_idx); 217 if (rc >= len) 218 return -EOPNOTSUPP; 219 return 0; 220 } 221 222 static void bnxt_vf_rep_get_drvinfo(struct net_device *dev, 223 struct ethtool_drvinfo *info) 224 { 225 strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver)); 226 } 227 228 static int bnxt_vf_rep_get_port_parent_id(struct net_device *dev, 229 struct netdev_phys_item_id *ppid) 230 { 231 struct bnxt_vf_rep *vf_rep = netdev_priv(dev); 232 233 /* as only PORT_PARENT_ID is supported currently use common code 234 * between PF and VF-rep for now. 235 */ 236 return bnxt_get_port_parent_id(vf_rep->bp->dev, ppid); 237 } 238 239 static const struct ethtool_ops bnxt_vf_rep_ethtool_ops = { 240 .get_drvinfo = bnxt_vf_rep_get_drvinfo 241 }; 242 243 static const struct net_device_ops bnxt_vf_rep_netdev_ops = { 244 .ndo_open = bnxt_vf_rep_open, 245 .ndo_stop = bnxt_vf_rep_close, 246 .ndo_start_xmit = bnxt_vf_rep_xmit, 247 .ndo_get_stats64 = bnxt_vf_rep_get_stats64, 248 .ndo_setup_tc = bnxt_vf_rep_setup_tc, 249 .ndo_get_port_parent_id = bnxt_vf_rep_get_port_parent_id, 250 .ndo_get_phys_port_name = bnxt_vf_rep_get_phys_port_name 251 }; 252 253 bool bnxt_dev_is_vf_rep(struct net_device *dev) 254 { 255 return dev->netdev_ops == &bnxt_vf_rep_netdev_ops; 256 } 257 258 /* Called when the parent PF interface is closed: 259 * As the mode transition from SWITCHDEV to LEGACY 260 * happens under the rtnl_lock() this routine is safe 261 * under the rtnl_lock() 262 */ 263 void bnxt_vf_reps_close(struct bnxt *bp) 264 { 265 struct bnxt_vf_rep *vf_rep; 266 u16 num_vfs, i; 267 268 if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV) 269 return; 270 271 num_vfs = pci_num_vf(bp->pdev); 272 for (i = 0; i < num_vfs; i++) { 273 vf_rep = bp->vf_reps[i]; 274 if (netif_running(vf_rep->dev)) 275 bnxt_vf_rep_close(vf_rep->dev); 276 } 277 } 278 279 /* Called when the parent PF interface is opened (re-opened): 280 * As the mode transition from SWITCHDEV to LEGACY 281 * happen under the rtnl_lock() this routine is safe 282 * under the rtnl_lock() 283 */ 284 void bnxt_vf_reps_open(struct bnxt *bp) 285 { 286 int i; 287 288 if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV) 289 return; 290 291 for (i = 0; i < pci_num_vf(bp->pdev); i++) { 292 /* Open the VF-Rep only if it is allocated in the FW */ 293 if (bp->vf_reps[i]->tx_cfa_action != CFA_HANDLE_INVALID) 294 bnxt_vf_rep_open(bp->vf_reps[i]->dev); 295 } 296 } 297 298 static void __bnxt_free_one_vf_rep(struct bnxt *bp, struct bnxt_vf_rep *vf_rep) 299 { 300 if (!vf_rep) 301 return; 302 303 if (vf_rep->dst) { 304 dst_release((struct dst_entry *)vf_rep->dst); 305 vf_rep->dst = NULL; 306 } 307 if (vf_rep->tx_cfa_action != CFA_HANDLE_INVALID) { 308 hwrm_cfa_vfr_free(bp, vf_rep->vf_idx); 309 vf_rep->tx_cfa_action = CFA_HANDLE_INVALID; 310 } 311 } 312 313 static void __bnxt_vf_reps_destroy(struct bnxt *bp) 314 { 315 u16 num_vfs = pci_num_vf(bp->pdev); 316 struct bnxt_vf_rep *vf_rep; 317 int i; 318 319 for (i = 0; i < num_vfs; i++) { 320 vf_rep = bp->vf_reps[i]; 321 if (vf_rep) { 322 __bnxt_free_one_vf_rep(bp, vf_rep); 323 if (vf_rep->dev) { 324 /* if register_netdev failed, then netdev_ops 325 * would have been set to NULL 326 */ 327 if (vf_rep->dev->netdev_ops) 328 unregister_netdev(vf_rep->dev); 329 free_netdev(vf_rep->dev); 330 } 331 } 332 } 333 334 kfree(bp->vf_reps); 335 bp->vf_reps = NULL; 336 } 337 338 void bnxt_vf_reps_destroy(struct bnxt *bp) 339 { 340 bool closed = false; 341 342 if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV) 343 return; 344 345 if (!bp->vf_reps) 346 return; 347 348 /* Ensure that parent PF's and VF-reps' RX/TX has been quiesced 349 * before proceeding with VF-rep cleanup. 350 */ 351 rtnl_lock(); 352 if (netif_running(bp->dev)) { 353 bnxt_close_nic(bp, false, false); 354 closed = true; 355 } 356 /* un-publish cfa_code_map so that RX path can't see it anymore */ 357 kfree(bp->cfa_code_map); 358 bp->cfa_code_map = NULL; 359 bp->eswitch_mode = DEVLINK_ESWITCH_MODE_LEGACY; 360 361 if (closed) 362 bnxt_open_nic(bp, false, false); 363 rtnl_unlock(); 364 365 /* Need to call vf_reps_destroy() outside of rntl_lock 366 * as unregister_netdev takes rtnl_lock 367 */ 368 __bnxt_vf_reps_destroy(bp); 369 } 370 371 /* Free the VF-Reps in firmware, during firmware hot-reset processing. 372 * Note that the VF-Rep netdevs are still active (not unregistered) during 373 * this process. As the mode transition from SWITCHDEV to LEGACY happens 374 * under the rtnl_lock() this routine is safe under the rtnl_lock(). 375 */ 376 void bnxt_vf_reps_free(struct bnxt *bp) 377 { 378 u16 num_vfs = pci_num_vf(bp->pdev); 379 int i; 380 381 if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV) 382 return; 383 384 for (i = 0; i < num_vfs; i++) 385 __bnxt_free_one_vf_rep(bp, bp->vf_reps[i]); 386 } 387 388 static int bnxt_alloc_vf_rep(struct bnxt *bp, struct bnxt_vf_rep *vf_rep, 389 u16 *cfa_code_map) 390 { 391 /* get cfa handles from FW */ 392 if (hwrm_cfa_vfr_alloc(bp, vf_rep->vf_idx, &vf_rep->tx_cfa_action, 393 &vf_rep->rx_cfa_code)) 394 return -ENOLINK; 395 396 cfa_code_map[vf_rep->rx_cfa_code] = vf_rep->vf_idx; 397 vf_rep->dst = metadata_dst_alloc(0, METADATA_HW_PORT_MUX, GFP_KERNEL); 398 if (!vf_rep->dst) 399 return -ENOMEM; 400 401 /* only cfa_action is needed to mux a packet while TXing */ 402 vf_rep->dst->u.port_info.port_id = vf_rep->tx_cfa_action; 403 vf_rep->dst->u.port_info.lower_dev = bp->dev; 404 405 return 0; 406 } 407 408 /* Allocate the VF-Reps in firmware, during firmware hot-reset processing. 409 * Note that the VF-Rep netdevs are still active (not unregistered) during 410 * this process. As the mode transition from SWITCHDEV to LEGACY happens 411 * under the rtnl_lock() this routine is safe under the rtnl_lock(). 412 */ 413 int bnxt_vf_reps_alloc(struct bnxt *bp) 414 { 415 u16 *cfa_code_map = bp->cfa_code_map, num_vfs = pci_num_vf(bp->pdev); 416 struct bnxt_vf_rep *vf_rep; 417 int rc, i; 418 419 if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV) 420 return 0; 421 422 if (!cfa_code_map) 423 return -EINVAL; 424 425 for (i = 0; i < MAX_CFA_CODE; i++) 426 cfa_code_map[i] = VF_IDX_INVALID; 427 428 for (i = 0; i < num_vfs; i++) { 429 vf_rep = bp->vf_reps[i]; 430 vf_rep->vf_idx = i; 431 432 rc = bnxt_alloc_vf_rep(bp, vf_rep, cfa_code_map); 433 if (rc) 434 goto err; 435 } 436 437 return 0; 438 439 err: 440 netdev_info(bp->dev, "%s error=%d\n", __func__, rc); 441 bnxt_vf_reps_free(bp); 442 return rc; 443 } 444 445 /* Use the OUI of the PF's perm addr and report the same mac addr 446 * for the same VF-rep each time 447 */ 448 static void bnxt_vf_rep_eth_addr_gen(u8 *src_mac, u16 vf_idx, u8 *mac) 449 { 450 u32 addr; 451 452 ether_addr_copy(mac, src_mac); 453 454 addr = jhash(src_mac, ETH_ALEN, 0) + vf_idx; 455 mac[3] = (u8)(addr & 0xFF); 456 mac[4] = (u8)((addr >> 8) & 0xFF); 457 mac[5] = (u8)((addr >> 16) & 0xFF); 458 } 459 460 static void bnxt_vf_rep_netdev_init(struct bnxt *bp, struct bnxt_vf_rep *vf_rep, 461 struct net_device *dev) 462 { 463 struct net_device *pf_dev = bp->dev; 464 u16 max_mtu; 465 466 dev->netdev_ops = &bnxt_vf_rep_netdev_ops; 467 dev->ethtool_ops = &bnxt_vf_rep_ethtool_ops; 468 /* Just inherit all the featues of the parent PF as the VF-R 469 * uses the RX/TX rings of the parent PF 470 */ 471 dev->hw_features = pf_dev->hw_features; 472 dev->gso_partial_features = pf_dev->gso_partial_features; 473 dev->vlan_features = pf_dev->vlan_features; 474 dev->hw_enc_features = pf_dev->hw_enc_features; 475 dev->features |= pf_dev->features; 476 bnxt_vf_rep_eth_addr_gen(bp->pf.mac_addr, vf_rep->vf_idx, 477 dev->perm_addr); 478 ether_addr_copy(dev->dev_addr, dev->perm_addr); 479 /* Set VF-Rep's max-mtu to the corresponding VF's max-mtu */ 480 if (!bnxt_hwrm_vfr_qcfg(bp, vf_rep, &max_mtu)) 481 dev->max_mtu = max_mtu; 482 dev->min_mtu = ETH_ZLEN; 483 } 484 485 static int bnxt_vf_reps_create(struct bnxt *bp) 486 { 487 u16 *cfa_code_map = NULL, num_vfs = pci_num_vf(bp->pdev); 488 struct bnxt_vf_rep *vf_rep; 489 struct net_device *dev; 490 int rc, i; 491 492 if (!(bp->flags & BNXT_FLAG_DSN_VALID)) 493 return -ENODEV; 494 495 bp->vf_reps = kcalloc(num_vfs, sizeof(vf_rep), GFP_KERNEL); 496 if (!bp->vf_reps) 497 return -ENOMEM; 498 499 /* storage for cfa_code to vf-idx mapping */ 500 cfa_code_map = kmalloc_array(MAX_CFA_CODE, sizeof(*bp->cfa_code_map), 501 GFP_KERNEL); 502 if (!cfa_code_map) { 503 rc = -ENOMEM; 504 goto err; 505 } 506 for (i = 0; i < MAX_CFA_CODE; i++) 507 cfa_code_map[i] = VF_IDX_INVALID; 508 509 for (i = 0; i < num_vfs; i++) { 510 dev = alloc_etherdev(sizeof(*vf_rep)); 511 if (!dev) { 512 rc = -ENOMEM; 513 goto err; 514 } 515 516 vf_rep = netdev_priv(dev); 517 bp->vf_reps[i] = vf_rep; 518 vf_rep->dev = dev; 519 vf_rep->bp = bp; 520 vf_rep->vf_idx = i; 521 vf_rep->tx_cfa_action = CFA_HANDLE_INVALID; 522 523 rc = bnxt_alloc_vf_rep(bp, vf_rep, cfa_code_map); 524 if (rc) 525 goto err; 526 527 bnxt_vf_rep_netdev_init(bp, vf_rep, dev); 528 rc = register_netdev(dev); 529 if (rc) { 530 /* no need for unregister_netdev in cleanup */ 531 dev->netdev_ops = NULL; 532 goto err; 533 } 534 } 535 536 /* publish cfa_code_map only after all VF-reps have been initialized */ 537 bp->cfa_code_map = cfa_code_map; 538 bp->eswitch_mode = DEVLINK_ESWITCH_MODE_SWITCHDEV; 539 netif_keep_dst(bp->dev); 540 return 0; 541 542 err: 543 netdev_info(bp->dev, "%s error=%d\n", __func__, rc); 544 kfree(cfa_code_map); 545 __bnxt_vf_reps_destroy(bp); 546 return rc; 547 } 548 549 /* Devlink related routines */ 550 int bnxt_dl_eswitch_mode_get(struct devlink *devlink, u16 *mode) 551 { 552 struct bnxt *bp = bnxt_get_bp_from_dl(devlink); 553 554 *mode = bp->eswitch_mode; 555 return 0; 556 } 557 558 int bnxt_dl_eswitch_mode_set(struct devlink *devlink, u16 mode, 559 struct netlink_ext_ack *extack) 560 { 561 struct bnxt *bp = bnxt_get_bp_from_dl(devlink); 562 int rc = 0; 563 564 mutex_lock(&bp->sriov_lock); 565 if (bp->eswitch_mode == mode) { 566 netdev_info(bp->dev, "already in %s eswitch mode\n", 567 mode == DEVLINK_ESWITCH_MODE_LEGACY ? 568 "legacy" : "switchdev"); 569 rc = -EINVAL; 570 goto done; 571 } 572 573 switch (mode) { 574 case DEVLINK_ESWITCH_MODE_LEGACY: 575 bnxt_vf_reps_destroy(bp); 576 break; 577 578 case DEVLINK_ESWITCH_MODE_SWITCHDEV: 579 if (bp->hwrm_spec_code < 0x10803) { 580 netdev_warn(bp->dev, "FW does not support SRIOV E-Switch SWITCHDEV mode\n"); 581 rc = -ENOTSUPP; 582 goto done; 583 } 584 585 if (pci_num_vf(bp->pdev) == 0) { 586 netdev_info(bp->dev, "Enable VFs before setting switchdev mode\n"); 587 rc = -EPERM; 588 goto done; 589 } 590 rc = bnxt_vf_reps_create(bp); 591 break; 592 593 default: 594 rc = -EINVAL; 595 goto done; 596 } 597 done: 598 mutex_unlock(&bp->sriov_lock); 599 return rc; 600 } 601 602 #endif 603