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 28static 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 55static 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 70static 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 96static 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 109static 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 117static 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 136static void 137bnxt_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 148static 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 168static LIST_HEAD(bnxt_vf_block_cb_list); 169 170static 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 186struct 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 198void 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 208static 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 222static void bnxt_vf_rep_get_drvinfo(struct net_device *dev, 223 struct ethtool_drvinfo *info) 224{ 225 strscpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver)); 226} 227 228static 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 239static const struct ethtool_ops bnxt_vf_rep_ethtool_ops = { 240 .get_drvinfo = bnxt_vf_rep_get_drvinfo 241}; 242 243static 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 253bool 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 */ 263void 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 */ 284void 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 298static 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 313static 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 338void 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; 359 360 if (closed) { 361 /* Temporarily set legacy mode to avoid re-opening 362 * representors and restore switchdev mode after that. 363 */ 364 bp->eswitch_mode = DEVLINK_ESWITCH_MODE_LEGACY; 365 bnxt_open_nic(bp, false, false); 366 bp->eswitch_mode = DEVLINK_ESWITCH_MODE_SWITCHDEV; 367 } 368 rtnl_unlock(); 369 370 /* Need to call vf_reps_destroy() outside of rntl_lock 371 * as unregister_netdev takes rtnl_lock 372 */ 373 __bnxt_vf_reps_destroy(bp); 374} 375 376/* Free the VF-Reps in firmware, during firmware hot-reset processing. 377 * Note that the VF-Rep netdevs are still active (not unregistered) during 378 * this process. As the mode transition from SWITCHDEV to LEGACY happens 379 * under the rtnl_lock() this routine is safe under the rtnl_lock(). 380 */ 381void bnxt_vf_reps_free(struct bnxt *bp) 382{ 383 u16 num_vfs = pci_num_vf(bp->pdev); 384 int i; 385 386 if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV) 387 return; 388 389 for (i = 0; i < num_vfs; i++) 390 __bnxt_free_one_vf_rep(bp, bp->vf_reps[i]); 391} 392 393static int bnxt_alloc_vf_rep(struct bnxt *bp, struct bnxt_vf_rep *vf_rep, 394 u16 *cfa_code_map) 395{ 396 /* get cfa handles from FW */ 397 if (hwrm_cfa_vfr_alloc(bp, vf_rep->vf_idx, &vf_rep->tx_cfa_action, 398 &vf_rep->rx_cfa_code)) 399 return -ENOLINK; 400 401 cfa_code_map[vf_rep->rx_cfa_code] = vf_rep->vf_idx; 402 vf_rep->dst = metadata_dst_alloc(0, METADATA_HW_PORT_MUX, GFP_KERNEL); 403 if (!vf_rep->dst) 404 return -ENOMEM; 405 406 /* only cfa_action is needed to mux a packet while TXing */ 407 vf_rep->dst->u.port_info.port_id = vf_rep->tx_cfa_action; 408 vf_rep->dst->u.port_info.lower_dev = bp->dev; 409 410 return 0; 411} 412 413/* Allocate the VF-Reps in firmware, during firmware hot-reset processing. 414 * Note that the VF-Rep netdevs are still active (not unregistered) during 415 * this process. As the mode transition from SWITCHDEV to LEGACY happens 416 * under the rtnl_lock() this routine is safe under the rtnl_lock(). 417 */ 418int bnxt_vf_reps_alloc(struct bnxt *bp) 419{ 420 u16 *cfa_code_map = bp->cfa_code_map, num_vfs = pci_num_vf(bp->pdev); 421 struct bnxt_vf_rep *vf_rep; 422 int rc, i; 423 424 if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV) 425 return 0; 426 427 if (!cfa_code_map) 428 return -EINVAL; 429 430 for (i = 0; i < MAX_CFA_CODE; i++) 431 cfa_code_map[i] = VF_IDX_INVALID; 432 433 for (i = 0; i < num_vfs; i++) { 434 vf_rep = bp->vf_reps[i]; 435 vf_rep->vf_idx = i; 436 437 rc = bnxt_alloc_vf_rep(bp, vf_rep, cfa_code_map); 438 if (rc) 439 goto err; 440 } 441 442 return 0; 443 444err: 445 netdev_info(bp->dev, "%s error=%d\n", __func__, rc); 446 bnxt_vf_reps_free(bp); 447 return rc; 448} 449 450/* Use the OUI of the PF's perm addr and report the same mac addr 451 * for the same VF-rep each time 452 */ 453static void bnxt_vf_rep_eth_addr_gen(u8 *src_mac, u16 vf_idx, u8 *mac) 454{ 455 u32 addr; 456 457 ether_addr_copy(mac, src_mac); 458 459 addr = jhash(src_mac, ETH_ALEN, 0) + vf_idx; 460 mac[3] = (u8)(addr & 0xFF); 461 mac[4] = (u8)((addr >> 8) & 0xFF); 462 mac[5] = (u8)((addr >> 16) & 0xFF); 463} 464 465static void bnxt_vf_rep_netdev_init(struct bnxt *bp, struct bnxt_vf_rep *vf_rep, 466 struct net_device *dev) 467{ 468 struct net_device *pf_dev = bp->dev; 469 u16 max_mtu; 470 471 SET_NETDEV_DEV(dev, &bp->pdev->dev); 472 dev->netdev_ops = &bnxt_vf_rep_netdev_ops; 473 dev->ethtool_ops = &bnxt_vf_rep_ethtool_ops; 474 /* Just inherit all the featues of the parent PF as the VF-R 475 * uses the RX/TX rings of the parent PF 476 */ 477 dev->hw_features = pf_dev->hw_features; 478 dev->gso_partial_features = pf_dev->gso_partial_features; 479 dev->vlan_features = pf_dev->vlan_features; 480 dev->hw_enc_features = pf_dev->hw_enc_features; 481 dev->features |= pf_dev->features; 482 bnxt_vf_rep_eth_addr_gen(bp->pf.mac_addr, vf_rep->vf_idx, 483 dev->perm_addr); 484 eth_hw_addr_set(dev, dev->perm_addr); 485 /* Set VF-Rep's max-mtu to the corresponding VF's max-mtu */ 486 if (!bnxt_hwrm_vfr_qcfg(bp, vf_rep, &max_mtu)) 487 dev->max_mtu = max_mtu; 488 dev->min_mtu = ETH_ZLEN; 489} 490 491int bnxt_vf_reps_create(struct bnxt *bp) 492{ 493 u16 *cfa_code_map = NULL, num_vfs = pci_num_vf(bp->pdev); 494 struct bnxt_vf_rep *vf_rep; 495 struct net_device *dev; 496 int rc, i; 497 498 if (!(bp->flags & BNXT_FLAG_DSN_VALID)) 499 return -ENODEV; 500 501 bp->vf_reps = kcalloc(num_vfs, sizeof(vf_rep), GFP_KERNEL); 502 if (!bp->vf_reps) 503 return -ENOMEM; 504 505 /* storage for cfa_code to vf-idx mapping */ 506 cfa_code_map = kmalloc_array(MAX_CFA_CODE, sizeof(*bp->cfa_code_map), 507 GFP_KERNEL); 508 if (!cfa_code_map) { 509 rc = -ENOMEM; 510 goto err; 511 } 512 for (i = 0; i < MAX_CFA_CODE; i++) 513 cfa_code_map[i] = VF_IDX_INVALID; 514 515 for (i = 0; i < num_vfs; i++) { 516 dev = alloc_etherdev(sizeof(*vf_rep)); 517 if (!dev) { 518 rc = -ENOMEM; 519 goto err; 520 } 521 522 vf_rep = netdev_priv(dev); 523 bp->vf_reps[i] = vf_rep; 524 vf_rep->dev = dev; 525 vf_rep->bp = bp; 526 vf_rep->vf_idx = i; 527 vf_rep->tx_cfa_action = CFA_HANDLE_INVALID; 528 529 rc = bnxt_alloc_vf_rep(bp, vf_rep, cfa_code_map); 530 if (rc) 531 goto err; 532 533 bnxt_vf_rep_netdev_init(bp, vf_rep, dev); 534 rc = register_netdev(dev); 535 if (rc) { 536 /* no need for unregister_netdev in cleanup */ 537 dev->netdev_ops = NULL; 538 goto err; 539 } 540 } 541 542 /* publish cfa_code_map only after all VF-reps have been initialized */ 543 bp->cfa_code_map = cfa_code_map; 544 netif_keep_dst(bp->dev); 545 return 0; 546 547err: 548 netdev_info(bp->dev, "%s error=%d\n", __func__, rc); 549 kfree(cfa_code_map); 550 __bnxt_vf_reps_destroy(bp); 551 return rc; 552} 553 554/* Devlink related routines */ 555int bnxt_dl_eswitch_mode_get(struct devlink *devlink, u16 *mode) 556{ 557 struct bnxt *bp = bnxt_get_bp_from_dl(devlink); 558 559 *mode = bp->eswitch_mode; 560 return 0; 561} 562 563int bnxt_dl_eswitch_mode_set(struct devlink *devlink, u16 mode, 564 struct netlink_ext_ack *extack) 565{ 566 struct bnxt *bp = bnxt_get_bp_from_dl(devlink); 567 int ret = 0; 568 569 if (bp->eswitch_mode == mode) { 570 netdev_info(bp->dev, "already in %s eswitch mode\n", 571 mode == DEVLINK_ESWITCH_MODE_LEGACY ? 572 "legacy" : "switchdev"); 573 return -EINVAL; 574 } 575 576 switch (mode) { 577 case DEVLINK_ESWITCH_MODE_LEGACY: 578 bnxt_vf_reps_destroy(bp); 579 break; 580 581 case DEVLINK_ESWITCH_MODE_SWITCHDEV: 582 if (bp->hwrm_spec_code < 0x10803) { 583 netdev_warn(bp->dev, "FW does not support SRIOV E-Switch SWITCHDEV mode\n"); 584 return -ENOTSUPP; 585 } 586 587 /* Create representors for existing VFs */ 588 if (pci_num_vf(bp->pdev) > 0) 589 ret = bnxt_vf_reps_create(bp); 590 break; 591 592 default: 593 return -EINVAL; 594 } 595 596 if (!ret) 597 bp->eswitch_mode = mode; 598 599 return ret; 600} 601 602#endif 603