1 /* 2 * Copyright (C) 2015 Cavium, Inc. 3 * 4 * This program is free software; you can redistribute it and/or modify it 5 * under the terms of version 2 of the GNU General Public License 6 * as published by the Free Software Foundation. 7 */ 8 9 #include <linux/module.h> 10 #include <linux/interrupt.h> 11 #include <linux/pci.h> 12 #include <linux/netdevice.h> 13 #include <linux/if_vlan.h> 14 #include <linux/etherdevice.h> 15 #include <linux/ethtool.h> 16 #include <linux/log2.h> 17 #include <linux/prefetch.h> 18 #include <linux/irq.h> 19 20 #include "nic_reg.h" 21 #include "nic.h" 22 #include "nicvf_queues.h" 23 #include "thunder_bgx.h" 24 25 #define DRV_NAME "thunder-nicvf" 26 #define DRV_VERSION "1.0" 27 28 /* Supported devices */ 29 static const struct pci_device_id nicvf_id_table[] = { 30 { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM, 31 PCI_DEVICE_ID_THUNDER_NIC_VF, 32 PCI_VENDOR_ID_CAVIUM, 0xA134) }, 33 { PCI_DEVICE_SUB(PCI_VENDOR_ID_CAVIUM, 34 PCI_DEVICE_ID_THUNDER_PASS1_NIC_VF, 35 PCI_VENDOR_ID_CAVIUM, 0xA11E) }, 36 { 0, } /* end of table */ 37 }; 38 39 MODULE_AUTHOR("Sunil Goutham"); 40 MODULE_DESCRIPTION("Cavium Thunder NIC Virtual Function Driver"); 41 MODULE_LICENSE("GPL v2"); 42 MODULE_VERSION(DRV_VERSION); 43 MODULE_DEVICE_TABLE(pci, nicvf_id_table); 44 45 static int debug = 0x00; 46 module_param(debug, int, 0644); 47 MODULE_PARM_DESC(debug, "Debug message level bitmap"); 48 49 static int cpi_alg = CPI_ALG_NONE; 50 module_param(cpi_alg, int, S_IRUGO); 51 MODULE_PARM_DESC(cpi_alg, 52 "PFC algorithm (0=none, 1=VLAN, 2=VLAN16, 3=IP Diffserv)"); 53 54 static inline u8 nicvf_netdev_qidx(struct nicvf *nic, u8 qidx) 55 { 56 if (nic->sqs_mode) 57 return qidx + ((nic->sqs_id + 1) * MAX_CMP_QUEUES_PER_QS); 58 else 59 return qidx; 60 } 61 62 static inline void nicvf_set_rx_frame_cnt(struct nicvf *nic, 63 struct sk_buff *skb) 64 { 65 if (skb->len <= 64) 66 nic->drv_stats.rx_frames_64++; 67 else if (skb->len <= 127) 68 nic->drv_stats.rx_frames_127++; 69 else if (skb->len <= 255) 70 nic->drv_stats.rx_frames_255++; 71 else if (skb->len <= 511) 72 nic->drv_stats.rx_frames_511++; 73 else if (skb->len <= 1023) 74 nic->drv_stats.rx_frames_1023++; 75 else if (skb->len <= 1518) 76 nic->drv_stats.rx_frames_1518++; 77 else 78 nic->drv_stats.rx_frames_jumbo++; 79 } 80 81 /* The Cavium ThunderX network controller can *only* be found in SoCs 82 * containing the ThunderX ARM64 CPU implementation. All accesses to the device 83 * registers on this platform are implicitly strongly ordered with respect 84 * to memory accesses. So writeq_relaxed() and readq_relaxed() are safe to use 85 * with no memory barriers in this driver. The readq()/writeq() functions add 86 * explicit ordering operation which in this case are redundant, and only 87 * add overhead. 88 */ 89 90 /* Register read/write APIs */ 91 void nicvf_reg_write(struct nicvf *nic, u64 offset, u64 val) 92 { 93 writeq_relaxed(val, nic->reg_base + offset); 94 } 95 96 u64 nicvf_reg_read(struct nicvf *nic, u64 offset) 97 { 98 return readq_relaxed(nic->reg_base + offset); 99 } 100 101 void nicvf_queue_reg_write(struct nicvf *nic, u64 offset, 102 u64 qidx, u64 val) 103 { 104 void __iomem *addr = nic->reg_base + offset; 105 106 writeq_relaxed(val, addr + (qidx << NIC_Q_NUM_SHIFT)); 107 } 108 109 u64 nicvf_queue_reg_read(struct nicvf *nic, u64 offset, u64 qidx) 110 { 111 void __iomem *addr = nic->reg_base + offset; 112 113 return readq_relaxed(addr + (qidx << NIC_Q_NUM_SHIFT)); 114 } 115 116 /* VF -> PF mailbox communication */ 117 static void nicvf_write_to_mbx(struct nicvf *nic, union nic_mbx *mbx) 118 { 119 u64 *msg = (u64 *)mbx; 120 121 nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 0, msg[0]); 122 nicvf_reg_write(nic, NIC_VF_PF_MAILBOX_0_1 + 8, msg[1]); 123 } 124 125 int nicvf_send_msg_to_pf(struct nicvf *nic, union nic_mbx *mbx) 126 { 127 int timeout = NIC_MBOX_MSG_TIMEOUT; 128 int sleep = 10; 129 130 nic->pf_acked = false; 131 nic->pf_nacked = false; 132 133 nicvf_write_to_mbx(nic, mbx); 134 135 /* Wait for previous message to be acked, timeout 2sec */ 136 while (!nic->pf_acked) { 137 if (nic->pf_nacked) 138 return -EINVAL; 139 msleep(sleep); 140 if (nic->pf_acked) 141 break; 142 timeout -= sleep; 143 if (!timeout) { 144 netdev_err(nic->netdev, 145 "PF didn't ack to mbox msg %d from VF%d\n", 146 (mbx->msg.msg & 0xFF), nic->vf_id); 147 return -EBUSY; 148 } 149 } 150 return 0; 151 } 152 153 /* Checks if VF is able to comminicate with PF 154 * and also gets the VNIC number this VF is associated to. 155 */ 156 static int nicvf_check_pf_ready(struct nicvf *nic) 157 { 158 union nic_mbx mbx = {}; 159 160 mbx.msg.msg = NIC_MBOX_MSG_READY; 161 if (nicvf_send_msg_to_pf(nic, &mbx)) { 162 netdev_err(nic->netdev, 163 "PF didn't respond to READY msg\n"); 164 return 0; 165 } 166 167 return 1; 168 } 169 170 static void nicvf_read_bgx_stats(struct nicvf *nic, struct bgx_stats_msg *bgx) 171 { 172 if (bgx->rx) 173 nic->bgx_stats.rx_stats[bgx->idx] = bgx->stats; 174 else 175 nic->bgx_stats.tx_stats[bgx->idx] = bgx->stats; 176 } 177 178 static void nicvf_handle_mbx_intr(struct nicvf *nic) 179 { 180 union nic_mbx mbx = {}; 181 u64 *mbx_data; 182 u64 mbx_addr; 183 int i; 184 185 mbx_addr = NIC_VF_PF_MAILBOX_0_1; 186 mbx_data = (u64 *)&mbx; 187 188 for (i = 0; i < NIC_PF_VF_MAILBOX_SIZE; i++) { 189 *mbx_data = nicvf_reg_read(nic, mbx_addr); 190 mbx_data++; 191 mbx_addr += sizeof(u64); 192 } 193 194 netdev_dbg(nic->netdev, "Mbox message: msg: 0x%x\n", mbx.msg.msg); 195 switch (mbx.msg.msg) { 196 case NIC_MBOX_MSG_READY: 197 nic->pf_acked = true; 198 nic->vf_id = mbx.nic_cfg.vf_id & 0x7F; 199 nic->tns_mode = mbx.nic_cfg.tns_mode & 0x7F; 200 nic->node = mbx.nic_cfg.node_id; 201 if (!nic->set_mac_pending) 202 ether_addr_copy(nic->netdev->dev_addr, 203 mbx.nic_cfg.mac_addr); 204 nic->sqs_mode = mbx.nic_cfg.sqs_mode; 205 nic->loopback_supported = mbx.nic_cfg.loopback_supported; 206 nic->link_up = false; 207 nic->duplex = 0; 208 nic->speed = 0; 209 break; 210 case NIC_MBOX_MSG_ACK: 211 nic->pf_acked = true; 212 break; 213 case NIC_MBOX_MSG_NACK: 214 nic->pf_nacked = true; 215 break; 216 case NIC_MBOX_MSG_RSS_SIZE: 217 nic->rss_info.rss_size = mbx.rss_size.ind_tbl_size; 218 nic->pf_acked = true; 219 break; 220 case NIC_MBOX_MSG_BGX_STATS: 221 nicvf_read_bgx_stats(nic, &mbx.bgx_stats); 222 nic->pf_acked = true; 223 break; 224 case NIC_MBOX_MSG_BGX_LINK_CHANGE: 225 nic->pf_acked = true; 226 nic->link_up = mbx.link_status.link_up; 227 nic->duplex = mbx.link_status.duplex; 228 nic->speed = mbx.link_status.speed; 229 if (nic->link_up) { 230 netdev_info(nic->netdev, "%s: Link is Up %d Mbps %s\n", 231 nic->netdev->name, nic->speed, 232 nic->duplex == DUPLEX_FULL ? 233 "Full duplex" : "Half duplex"); 234 netif_carrier_on(nic->netdev); 235 netif_tx_start_all_queues(nic->netdev); 236 } else { 237 netdev_info(nic->netdev, "%s: Link is Down\n", 238 nic->netdev->name); 239 netif_carrier_off(nic->netdev); 240 netif_tx_stop_all_queues(nic->netdev); 241 } 242 break; 243 case NIC_MBOX_MSG_ALLOC_SQS: 244 nic->sqs_count = mbx.sqs_alloc.qs_count; 245 nic->pf_acked = true; 246 break; 247 case NIC_MBOX_MSG_SNICVF_PTR: 248 /* Primary VF: make note of secondary VF's pointer 249 * to be used while packet transmission. 250 */ 251 nic->snicvf[mbx.nicvf.sqs_id] = 252 (struct nicvf *)mbx.nicvf.nicvf; 253 nic->pf_acked = true; 254 break; 255 case NIC_MBOX_MSG_PNICVF_PTR: 256 /* Secondary VF/Qset: make note of primary VF's pointer 257 * to be used while packet reception, to handover packet 258 * to primary VF's netdev. 259 */ 260 nic->pnicvf = (struct nicvf *)mbx.nicvf.nicvf; 261 nic->pf_acked = true; 262 break; 263 default: 264 netdev_err(nic->netdev, 265 "Invalid message from PF, msg 0x%x\n", mbx.msg.msg); 266 break; 267 } 268 nicvf_clear_intr(nic, NICVF_INTR_MBOX, 0); 269 } 270 271 static int nicvf_hw_set_mac_addr(struct nicvf *nic, struct net_device *netdev) 272 { 273 union nic_mbx mbx = {}; 274 275 mbx.mac.msg = NIC_MBOX_MSG_SET_MAC; 276 mbx.mac.vf_id = nic->vf_id; 277 ether_addr_copy(mbx.mac.mac_addr, netdev->dev_addr); 278 279 return nicvf_send_msg_to_pf(nic, &mbx); 280 } 281 282 static void nicvf_config_cpi(struct nicvf *nic) 283 { 284 union nic_mbx mbx = {}; 285 286 mbx.cpi_cfg.msg = NIC_MBOX_MSG_CPI_CFG; 287 mbx.cpi_cfg.vf_id = nic->vf_id; 288 mbx.cpi_cfg.cpi_alg = nic->cpi_alg; 289 mbx.cpi_cfg.rq_cnt = nic->qs->rq_cnt; 290 291 nicvf_send_msg_to_pf(nic, &mbx); 292 } 293 294 static void nicvf_get_rss_size(struct nicvf *nic) 295 { 296 union nic_mbx mbx = {}; 297 298 mbx.rss_size.msg = NIC_MBOX_MSG_RSS_SIZE; 299 mbx.rss_size.vf_id = nic->vf_id; 300 nicvf_send_msg_to_pf(nic, &mbx); 301 } 302 303 void nicvf_config_rss(struct nicvf *nic) 304 { 305 union nic_mbx mbx = {}; 306 struct nicvf_rss_info *rss = &nic->rss_info; 307 int ind_tbl_len = rss->rss_size; 308 int i, nextq = 0; 309 310 mbx.rss_cfg.vf_id = nic->vf_id; 311 mbx.rss_cfg.hash_bits = rss->hash_bits; 312 while (ind_tbl_len) { 313 mbx.rss_cfg.tbl_offset = nextq; 314 mbx.rss_cfg.tbl_len = min(ind_tbl_len, 315 RSS_IND_TBL_LEN_PER_MBX_MSG); 316 mbx.rss_cfg.msg = mbx.rss_cfg.tbl_offset ? 317 NIC_MBOX_MSG_RSS_CFG_CONT : NIC_MBOX_MSG_RSS_CFG; 318 319 for (i = 0; i < mbx.rss_cfg.tbl_len; i++) 320 mbx.rss_cfg.ind_tbl[i] = rss->ind_tbl[nextq++]; 321 322 nicvf_send_msg_to_pf(nic, &mbx); 323 324 ind_tbl_len -= mbx.rss_cfg.tbl_len; 325 } 326 } 327 328 void nicvf_set_rss_key(struct nicvf *nic) 329 { 330 struct nicvf_rss_info *rss = &nic->rss_info; 331 u64 key_addr = NIC_VNIC_RSS_KEY_0_4; 332 int idx; 333 334 for (idx = 0; idx < RSS_HASH_KEY_SIZE; idx++) { 335 nicvf_reg_write(nic, key_addr, rss->key[idx]); 336 key_addr += sizeof(u64); 337 } 338 } 339 340 static int nicvf_rss_init(struct nicvf *nic) 341 { 342 struct nicvf_rss_info *rss = &nic->rss_info; 343 int idx; 344 345 nicvf_get_rss_size(nic); 346 347 if (cpi_alg != CPI_ALG_NONE) { 348 rss->enable = false; 349 rss->hash_bits = 0; 350 return 0; 351 } 352 353 rss->enable = true; 354 355 /* Using the HW reset value for now */ 356 rss->key[0] = 0xFEED0BADFEED0BADULL; 357 rss->key[1] = 0xFEED0BADFEED0BADULL; 358 rss->key[2] = 0xFEED0BADFEED0BADULL; 359 rss->key[3] = 0xFEED0BADFEED0BADULL; 360 rss->key[4] = 0xFEED0BADFEED0BADULL; 361 362 nicvf_set_rss_key(nic); 363 364 rss->cfg = RSS_IP_HASH_ENA | RSS_TCP_HASH_ENA | RSS_UDP_HASH_ENA; 365 nicvf_reg_write(nic, NIC_VNIC_RSS_CFG, rss->cfg); 366 367 rss->hash_bits = ilog2(rounddown_pow_of_two(rss->rss_size)); 368 369 for (idx = 0; idx < rss->rss_size; idx++) 370 rss->ind_tbl[idx] = ethtool_rxfh_indir_default(idx, 371 nic->rx_queues); 372 nicvf_config_rss(nic); 373 return 1; 374 } 375 376 /* Request PF to allocate additional Qsets */ 377 static void nicvf_request_sqs(struct nicvf *nic) 378 { 379 union nic_mbx mbx = {}; 380 int sqs; 381 int sqs_count = nic->sqs_count; 382 int rx_queues = 0, tx_queues = 0; 383 384 /* Only primary VF should request */ 385 if (nic->sqs_mode || !nic->sqs_count) 386 return; 387 388 mbx.sqs_alloc.msg = NIC_MBOX_MSG_ALLOC_SQS; 389 mbx.sqs_alloc.vf_id = nic->vf_id; 390 mbx.sqs_alloc.qs_count = nic->sqs_count; 391 if (nicvf_send_msg_to_pf(nic, &mbx)) { 392 /* No response from PF */ 393 nic->sqs_count = 0; 394 return; 395 } 396 397 /* Return if no Secondary Qsets available */ 398 if (!nic->sqs_count) 399 return; 400 401 if (nic->rx_queues > MAX_RCV_QUEUES_PER_QS) 402 rx_queues = nic->rx_queues - MAX_RCV_QUEUES_PER_QS; 403 if (nic->tx_queues > MAX_SND_QUEUES_PER_QS) 404 tx_queues = nic->tx_queues - MAX_SND_QUEUES_PER_QS; 405 406 /* Set no of Rx/Tx queues in each of the SQsets */ 407 for (sqs = 0; sqs < nic->sqs_count; sqs++) { 408 mbx.nicvf.msg = NIC_MBOX_MSG_SNICVF_PTR; 409 mbx.nicvf.vf_id = nic->vf_id; 410 mbx.nicvf.sqs_id = sqs; 411 nicvf_send_msg_to_pf(nic, &mbx); 412 413 nic->snicvf[sqs]->sqs_id = sqs; 414 if (rx_queues > MAX_RCV_QUEUES_PER_QS) { 415 nic->snicvf[sqs]->qs->rq_cnt = MAX_RCV_QUEUES_PER_QS; 416 rx_queues -= MAX_RCV_QUEUES_PER_QS; 417 } else { 418 nic->snicvf[sqs]->qs->rq_cnt = rx_queues; 419 rx_queues = 0; 420 } 421 422 if (tx_queues > MAX_SND_QUEUES_PER_QS) { 423 nic->snicvf[sqs]->qs->sq_cnt = MAX_SND_QUEUES_PER_QS; 424 tx_queues -= MAX_SND_QUEUES_PER_QS; 425 } else { 426 nic->snicvf[sqs]->qs->sq_cnt = tx_queues; 427 tx_queues = 0; 428 } 429 430 nic->snicvf[sqs]->qs->cq_cnt = 431 max(nic->snicvf[sqs]->qs->rq_cnt, nic->snicvf[sqs]->qs->sq_cnt); 432 433 /* Initialize secondary Qset's queues and its interrupts */ 434 nicvf_open(nic->snicvf[sqs]->netdev); 435 } 436 437 /* Update stack with actual Rx/Tx queue count allocated */ 438 if (sqs_count != nic->sqs_count) 439 nicvf_set_real_num_queues(nic->netdev, 440 nic->tx_queues, nic->rx_queues); 441 } 442 443 /* Send this Qset's nicvf pointer to PF. 444 * PF inturn sends primary VF's nicvf struct to secondary Qsets/VFs 445 * so that packets received by these Qsets can use primary VF's netdev 446 */ 447 static void nicvf_send_vf_struct(struct nicvf *nic) 448 { 449 union nic_mbx mbx = {}; 450 451 mbx.nicvf.msg = NIC_MBOX_MSG_NICVF_PTR; 452 mbx.nicvf.sqs_mode = nic->sqs_mode; 453 mbx.nicvf.nicvf = (u64)nic; 454 nicvf_send_msg_to_pf(nic, &mbx); 455 } 456 457 static void nicvf_get_primary_vf_struct(struct nicvf *nic) 458 { 459 union nic_mbx mbx = {}; 460 461 mbx.nicvf.msg = NIC_MBOX_MSG_PNICVF_PTR; 462 nicvf_send_msg_to_pf(nic, &mbx); 463 } 464 465 int nicvf_set_real_num_queues(struct net_device *netdev, 466 int tx_queues, int rx_queues) 467 { 468 int err = 0; 469 470 err = netif_set_real_num_tx_queues(netdev, tx_queues); 471 if (err) { 472 netdev_err(netdev, 473 "Failed to set no of Tx queues: %d\n", tx_queues); 474 return err; 475 } 476 477 err = netif_set_real_num_rx_queues(netdev, rx_queues); 478 if (err) 479 netdev_err(netdev, 480 "Failed to set no of Rx queues: %d\n", rx_queues); 481 return err; 482 } 483 484 static int nicvf_init_resources(struct nicvf *nic) 485 { 486 int err; 487 union nic_mbx mbx = {}; 488 489 mbx.msg.msg = NIC_MBOX_MSG_CFG_DONE; 490 491 /* Enable Qset */ 492 nicvf_qset_config(nic, true); 493 494 /* Initialize queues and HW for data transfer */ 495 err = nicvf_config_data_transfer(nic, true); 496 if (err) { 497 netdev_err(nic->netdev, 498 "Failed to alloc/config VF's QSet resources\n"); 499 return err; 500 } 501 502 /* Send VF config done msg to PF */ 503 nicvf_write_to_mbx(nic, &mbx); 504 505 return 0; 506 } 507 508 static void nicvf_snd_pkt_handler(struct net_device *netdev, 509 struct cmp_queue *cq, 510 struct cqe_send_t *cqe_tx, int cqe_type) 511 { 512 struct sk_buff *skb = NULL; 513 struct nicvf *nic = netdev_priv(netdev); 514 struct snd_queue *sq; 515 struct sq_hdr_subdesc *hdr; 516 517 sq = &nic->qs->sq[cqe_tx->sq_idx]; 518 519 hdr = (struct sq_hdr_subdesc *)GET_SQ_DESC(sq, cqe_tx->sqe_ptr); 520 if (hdr->subdesc_type != SQ_DESC_TYPE_HEADER) 521 return; 522 523 netdev_dbg(nic->netdev, 524 "%s Qset #%d SQ #%d SQ ptr #%d subdesc count %d\n", 525 __func__, cqe_tx->sq_qs, cqe_tx->sq_idx, 526 cqe_tx->sqe_ptr, hdr->subdesc_cnt); 527 528 nicvf_check_cqe_tx_errs(nic, cq, cqe_tx); 529 skb = (struct sk_buff *)sq->skbuff[cqe_tx->sqe_ptr]; 530 /* For TSO offloaded packets only one SQE will have a valid SKB */ 531 if (skb) { 532 nicvf_put_sq_desc(sq, hdr->subdesc_cnt + 1); 533 prefetch(skb); 534 dev_consume_skb_any(skb); 535 sq->skbuff[cqe_tx->sqe_ptr] = (u64)NULL; 536 } else { 537 /* In case of HW TSO, HW sends a CQE for each segment of a TSO 538 * packet instead of a single CQE for the whole TSO packet 539 * transmitted. Each of this CQE points to the same SQE, so 540 * avoid freeing same SQE multiple times. 541 */ 542 if (!nic->hw_tso) 543 nicvf_put_sq_desc(sq, hdr->subdesc_cnt + 1); 544 } 545 } 546 547 static inline void nicvf_set_rxhash(struct net_device *netdev, 548 struct cqe_rx_t *cqe_rx, 549 struct sk_buff *skb) 550 { 551 u8 hash_type; 552 u32 hash; 553 554 if (!(netdev->features & NETIF_F_RXHASH)) 555 return; 556 557 switch (cqe_rx->rss_alg) { 558 case RSS_ALG_TCP_IP: 559 case RSS_ALG_UDP_IP: 560 hash_type = PKT_HASH_TYPE_L4; 561 hash = cqe_rx->rss_tag; 562 break; 563 case RSS_ALG_IP: 564 hash_type = PKT_HASH_TYPE_L3; 565 hash = cqe_rx->rss_tag; 566 break; 567 default: 568 hash_type = PKT_HASH_TYPE_NONE; 569 hash = 0; 570 } 571 572 skb_set_hash(skb, hash, hash_type); 573 } 574 575 static void nicvf_rcv_pkt_handler(struct net_device *netdev, 576 struct napi_struct *napi, 577 struct cqe_rx_t *cqe_rx) 578 { 579 struct sk_buff *skb; 580 struct nicvf *nic = netdev_priv(netdev); 581 int err = 0; 582 int rq_idx; 583 584 rq_idx = nicvf_netdev_qidx(nic, cqe_rx->rq_idx); 585 586 if (nic->sqs_mode) { 587 /* Use primary VF's 'nicvf' struct */ 588 nic = nic->pnicvf; 589 netdev = nic->netdev; 590 } 591 592 /* Check for errors */ 593 err = nicvf_check_cqe_rx_errs(nic, cqe_rx); 594 if (err && !cqe_rx->rb_cnt) 595 return; 596 597 skb = nicvf_get_rcv_skb(nic, cqe_rx); 598 if (!skb) { 599 netdev_dbg(nic->netdev, "Packet not received\n"); 600 return; 601 } 602 603 if (netif_msg_pktdata(nic)) { 604 netdev_info(nic->netdev, "%s: skb 0x%p, len=%d\n", netdev->name, 605 skb, skb->len); 606 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1, 607 skb->data, skb->len, true); 608 } 609 610 /* If error packet, drop it here */ 611 if (err) { 612 dev_kfree_skb_any(skb); 613 return; 614 } 615 616 nicvf_set_rx_frame_cnt(nic, skb); 617 618 nicvf_set_rxhash(netdev, cqe_rx, skb); 619 620 skb_record_rx_queue(skb, rq_idx); 621 if (netdev->hw_features & NETIF_F_RXCSUM) { 622 /* HW by default verifies TCP/UDP/SCTP checksums */ 623 skb->ip_summed = CHECKSUM_UNNECESSARY; 624 } else { 625 skb_checksum_none_assert(skb); 626 } 627 628 skb->protocol = eth_type_trans(skb, netdev); 629 630 /* Check for stripped VLAN */ 631 if (cqe_rx->vlan_found && cqe_rx->vlan_stripped) 632 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), 633 ntohs((__force __be16)cqe_rx->vlan_tci)); 634 635 if (napi && (netdev->features & NETIF_F_GRO)) 636 napi_gro_receive(napi, skb); 637 else 638 netif_receive_skb(skb); 639 } 640 641 static int nicvf_cq_intr_handler(struct net_device *netdev, u8 cq_idx, 642 struct napi_struct *napi, int budget) 643 { 644 int processed_cqe, work_done = 0, tx_done = 0; 645 int cqe_count, cqe_head; 646 struct nicvf *nic = netdev_priv(netdev); 647 struct queue_set *qs = nic->qs; 648 struct cmp_queue *cq = &qs->cq[cq_idx]; 649 struct cqe_rx_t *cq_desc; 650 struct netdev_queue *txq; 651 652 spin_lock_bh(&cq->lock); 653 loop: 654 processed_cqe = 0; 655 /* Get no of valid CQ entries to process */ 656 cqe_count = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS, cq_idx); 657 cqe_count &= CQ_CQE_COUNT; 658 if (!cqe_count) 659 goto done; 660 661 /* Get head of the valid CQ entries */ 662 cqe_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD, cq_idx) >> 9; 663 cqe_head &= 0xFFFF; 664 665 netdev_dbg(nic->netdev, "%s CQ%d cqe_count %d cqe_head %d\n", 666 __func__, cq_idx, cqe_count, cqe_head); 667 while (processed_cqe < cqe_count) { 668 /* Get the CQ descriptor */ 669 cq_desc = (struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head); 670 cqe_head++; 671 cqe_head &= (cq->dmem.q_len - 1); 672 /* Initiate prefetch for next descriptor */ 673 prefetch((struct cqe_rx_t *)GET_CQ_DESC(cq, cqe_head)); 674 675 if ((work_done >= budget) && napi && 676 (cq_desc->cqe_type != CQE_TYPE_SEND)) { 677 break; 678 } 679 680 netdev_dbg(nic->netdev, "CQ%d cq_desc->cqe_type %d\n", 681 cq_idx, cq_desc->cqe_type); 682 switch (cq_desc->cqe_type) { 683 case CQE_TYPE_RX: 684 nicvf_rcv_pkt_handler(netdev, napi, cq_desc); 685 work_done++; 686 break; 687 case CQE_TYPE_SEND: 688 nicvf_snd_pkt_handler(netdev, cq, 689 (void *)cq_desc, CQE_TYPE_SEND); 690 tx_done++; 691 break; 692 case CQE_TYPE_INVALID: 693 case CQE_TYPE_RX_SPLIT: 694 case CQE_TYPE_RX_TCP: 695 case CQE_TYPE_SEND_PTP: 696 /* Ignore for now */ 697 break; 698 } 699 processed_cqe++; 700 } 701 netdev_dbg(nic->netdev, 702 "%s CQ%d processed_cqe %d work_done %d budget %d\n", 703 __func__, cq_idx, processed_cqe, work_done, budget); 704 705 /* Ring doorbell to inform H/W to reuse processed CQEs */ 706 nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_DOOR, 707 cq_idx, processed_cqe); 708 709 if ((work_done < budget) && napi) 710 goto loop; 711 712 done: 713 /* Wakeup TXQ if its stopped earlier due to SQ full */ 714 if (tx_done) { 715 netdev = nic->pnicvf->netdev; 716 txq = netdev_get_tx_queue(netdev, 717 nicvf_netdev_qidx(nic, cq_idx)); 718 nic = nic->pnicvf; 719 if (netif_tx_queue_stopped(txq) && netif_carrier_ok(netdev)) { 720 netif_tx_start_queue(txq); 721 nic->drv_stats.txq_wake++; 722 if (netif_msg_tx_err(nic)) 723 netdev_warn(netdev, 724 "%s: Transmit queue wakeup SQ%d\n", 725 netdev->name, cq_idx); 726 } 727 } 728 729 spin_unlock_bh(&cq->lock); 730 return work_done; 731 } 732 733 static int nicvf_poll(struct napi_struct *napi, int budget) 734 { 735 u64 cq_head; 736 int work_done = 0; 737 struct net_device *netdev = napi->dev; 738 struct nicvf *nic = netdev_priv(netdev); 739 struct nicvf_cq_poll *cq; 740 741 cq = container_of(napi, struct nicvf_cq_poll, napi); 742 work_done = nicvf_cq_intr_handler(netdev, cq->cq_idx, napi, budget); 743 744 if (work_done < budget) { 745 /* Slow packet rate, exit polling */ 746 napi_complete(napi); 747 /* Re-enable interrupts */ 748 cq_head = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_HEAD, 749 cq->cq_idx); 750 nicvf_clear_intr(nic, NICVF_INTR_CQ, cq->cq_idx); 751 nicvf_queue_reg_write(nic, NIC_QSET_CQ_0_7_HEAD, 752 cq->cq_idx, cq_head); 753 nicvf_enable_intr(nic, NICVF_INTR_CQ, cq->cq_idx); 754 } 755 return work_done; 756 } 757 758 /* Qset error interrupt handler 759 * 760 * As of now only CQ errors are handled 761 */ 762 static void nicvf_handle_qs_err(unsigned long data) 763 { 764 struct nicvf *nic = (struct nicvf *)data; 765 struct queue_set *qs = nic->qs; 766 int qidx; 767 u64 status; 768 769 netif_tx_disable(nic->netdev); 770 771 /* Check if it is CQ err */ 772 for (qidx = 0; qidx < qs->cq_cnt; qidx++) { 773 status = nicvf_queue_reg_read(nic, NIC_QSET_CQ_0_7_STATUS, 774 qidx); 775 if (!(status & CQ_ERR_MASK)) 776 continue; 777 /* Process already queued CQEs and reconfig CQ */ 778 nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx); 779 nicvf_sq_disable(nic, qidx); 780 nicvf_cq_intr_handler(nic->netdev, qidx, NULL, 0); 781 nicvf_cmp_queue_config(nic, qs, qidx, true); 782 nicvf_sq_free_used_descs(nic->netdev, &qs->sq[qidx], qidx); 783 nicvf_sq_enable(nic, &qs->sq[qidx], qidx); 784 785 nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx); 786 } 787 788 netif_tx_start_all_queues(nic->netdev); 789 /* Re-enable Qset error interrupt */ 790 nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0); 791 } 792 793 static void nicvf_dump_intr_status(struct nicvf *nic) 794 { 795 if (netif_msg_intr(nic)) 796 netdev_info(nic->netdev, "%s: interrupt status 0x%llx\n", 797 nic->netdev->name, nicvf_reg_read(nic, NIC_VF_INT)); 798 } 799 800 static irqreturn_t nicvf_misc_intr_handler(int irq, void *nicvf_irq) 801 { 802 struct nicvf *nic = (struct nicvf *)nicvf_irq; 803 u64 intr; 804 805 nicvf_dump_intr_status(nic); 806 807 intr = nicvf_reg_read(nic, NIC_VF_INT); 808 /* Check for spurious interrupt */ 809 if (!(intr & NICVF_INTR_MBOX_MASK)) 810 return IRQ_HANDLED; 811 812 nicvf_handle_mbx_intr(nic); 813 814 return IRQ_HANDLED; 815 } 816 817 static irqreturn_t nicvf_intr_handler(int irq, void *cq_irq) 818 { 819 struct nicvf_cq_poll *cq_poll = (struct nicvf_cq_poll *)cq_irq; 820 struct nicvf *nic = cq_poll->nicvf; 821 int qidx = cq_poll->cq_idx; 822 823 nicvf_dump_intr_status(nic); 824 825 /* Disable interrupts */ 826 nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx); 827 828 /* Schedule NAPI */ 829 napi_schedule_irqoff(&cq_poll->napi); 830 831 /* Clear interrupt */ 832 nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx); 833 834 return IRQ_HANDLED; 835 } 836 837 static irqreturn_t nicvf_rbdr_intr_handler(int irq, void *nicvf_irq) 838 { 839 struct nicvf *nic = (struct nicvf *)nicvf_irq; 840 u8 qidx; 841 842 843 nicvf_dump_intr_status(nic); 844 845 /* Disable RBDR interrupt and schedule softirq */ 846 for (qidx = 0; qidx < nic->qs->rbdr_cnt; qidx++) { 847 if (!nicvf_is_intr_enabled(nic, NICVF_INTR_RBDR, qidx)) 848 continue; 849 nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx); 850 tasklet_hi_schedule(&nic->rbdr_task); 851 /* Clear interrupt */ 852 nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx); 853 } 854 855 return IRQ_HANDLED; 856 } 857 858 static irqreturn_t nicvf_qs_err_intr_handler(int irq, void *nicvf_irq) 859 { 860 struct nicvf *nic = (struct nicvf *)nicvf_irq; 861 862 nicvf_dump_intr_status(nic); 863 864 /* Disable Qset err interrupt and schedule softirq */ 865 nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0); 866 tasklet_hi_schedule(&nic->qs_err_task); 867 nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0); 868 869 return IRQ_HANDLED; 870 } 871 872 static int nicvf_enable_msix(struct nicvf *nic) 873 { 874 int ret, vec; 875 876 nic->num_vec = NIC_VF_MSIX_VECTORS; 877 878 for (vec = 0; vec < nic->num_vec; vec++) 879 nic->msix_entries[vec].entry = vec; 880 881 ret = pci_enable_msix(nic->pdev, nic->msix_entries, nic->num_vec); 882 if (ret) { 883 netdev_err(nic->netdev, 884 "Req for #%d msix vectors failed\n", nic->num_vec); 885 return 0; 886 } 887 nic->msix_enabled = 1; 888 return 1; 889 } 890 891 static void nicvf_disable_msix(struct nicvf *nic) 892 { 893 if (nic->msix_enabled) { 894 pci_disable_msix(nic->pdev); 895 nic->msix_enabled = 0; 896 nic->num_vec = 0; 897 } 898 } 899 900 static void nicvf_set_irq_affinity(struct nicvf *nic) 901 { 902 int vec, cpu; 903 int irqnum; 904 905 for (vec = 0; vec < nic->num_vec; vec++) { 906 if (!nic->irq_allocated[vec]) 907 continue; 908 909 if (!zalloc_cpumask_var(&nic->affinity_mask[vec], GFP_KERNEL)) 910 return; 911 /* CQ interrupts */ 912 if (vec < NICVF_INTR_ID_SQ) 913 /* Leave CPU0 for RBDR and other interrupts */ 914 cpu = nicvf_netdev_qidx(nic, vec) + 1; 915 else 916 cpu = 0; 917 918 cpumask_set_cpu(cpumask_local_spread(cpu, nic->node), 919 nic->affinity_mask[vec]); 920 irqnum = nic->msix_entries[vec].vector; 921 irq_set_affinity_hint(irqnum, nic->affinity_mask[vec]); 922 } 923 } 924 925 static int nicvf_register_interrupts(struct nicvf *nic) 926 { 927 int irq, ret = 0; 928 int vector; 929 930 for_each_cq_irq(irq) 931 sprintf(nic->irq_name[irq], "NICVF%d CQ%d", 932 nic->vf_id, irq); 933 934 for_each_sq_irq(irq) 935 sprintf(nic->irq_name[irq], "NICVF%d SQ%d", 936 nic->vf_id, irq - NICVF_INTR_ID_SQ); 937 938 for_each_rbdr_irq(irq) 939 sprintf(nic->irq_name[irq], "NICVF%d RBDR%d", 940 nic->vf_id, irq - NICVF_INTR_ID_RBDR); 941 942 /* Register CQ interrupts */ 943 for (irq = 0; irq < nic->qs->cq_cnt; irq++) { 944 vector = nic->msix_entries[irq].vector; 945 ret = request_irq(vector, nicvf_intr_handler, 946 0, nic->irq_name[irq], nic->napi[irq]); 947 if (ret) 948 goto err; 949 nic->irq_allocated[irq] = true; 950 } 951 952 /* Register RBDR interrupt */ 953 for (irq = NICVF_INTR_ID_RBDR; 954 irq < (NICVF_INTR_ID_RBDR + nic->qs->rbdr_cnt); irq++) { 955 vector = nic->msix_entries[irq].vector; 956 ret = request_irq(vector, nicvf_rbdr_intr_handler, 957 0, nic->irq_name[irq], nic); 958 if (ret) 959 goto err; 960 nic->irq_allocated[irq] = true; 961 } 962 963 /* Register QS error interrupt */ 964 sprintf(nic->irq_name[NICVF_INTR_ID_QS_ERR], 965 "NICVF%d Qset error", nic->vf_id); 966 irq = NICVF_INTR_ID_QS_ERR; 967 ret = request_irq(nic->msix_entries[irq].vector, 968 nicvf_qs_err_intr_handler, 969 0, nic->irq_name[irq], nic); 970 if (ret) 971 goto err; 972 973 nic->irq_allocated[irq] = true; 974 975 /* Set IRQ affinities */ 976 nicvf_set_irq_affinity(nic); 977 978 err: 979 if (ret) 980 netdev_err(nic->netdev, "request_irq failed, vector %d\n", irq); 981 982 return ret; 983 } 984 985 static void nicvf_unregister_interrupts(struct nicvf *nic) 986 { 987 int irq; 988 989 /* Free registered interrupts */ 990 for (irq = 0; irq < nic->num_vec; irq++) { 991 if (!nic->irq_allocated[irq]) 992 continue; 993 994 irq_set_affinity_hint(nic->msix_entries[irq].vector, NULL); 995 free_cpumask_var(nic->affinity_mask[irq]); 996 997 if (irq < NICVF_INTR_ID_SQ) 998 free_irq(nic->msix_entries[irq].vector, nic->napi[irq]); 999 else 1000 free_irq(nic->msix_entries[irq].vector, nic); 1001 1002 nic->irq_allocated[irq] = false; 1003 } 1004 1005 /* Disable MSI-X */ 1006 nicvf_disable_msix(nic); 1007 } 1008 1009 /* Initialize MSIX vectors and register MISC interrupt. 1010 * Send READY message to PF to check if its alive 1011 */ 1012 static int nicvf_register_misc_interrupt(struct nicvf *nic) 1013 { 1014 int ret = 0; 1015 int irq = NICVF_INTR_ID_MISC; 1016 1017 /* Return if mailbox interrupt is already registered */ 1018 if (nic->msix_enabled) 1019 return 0; 1020 1021 /* Enable MSI-X */ 1022 if (!nicvf_enable_msix(nic)) 1023 return 1; 1024 1025 sprintf(nic->irq_name[irq], "%s Mbox", "NICVF"); 1026 /* Register Misc interrupt */ 1027 ret = request_irq(nic->msix_entries[irq].vector, 1028 nicvf_misc_intr_handler, 0, nic->irq_name[irq], nic); 1029 1030 if (ret) 1031 return ret; 1032 nic->irq_allocated[irq] = true; 1033 1034 /* Enable mailbox interrupt */ 1035 nicvf_enable_intr(nic, NICVF_INTR_MBOX, 0); 1036 1037 /* Check if VF is able to communicate with PF */ 1038 if (!nicvf_check_pf_ready(nic)) { 1039 nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0); 1040 nicvf_unregister_interrupts(nic); 1041 return 1; 1042 } 1043 1044 return 0; 1045 } 1046 1047 static netdev_tx_t nicvf_xmit(struct sk_buff *skb, struct net_device *netdev) 1048 { 1049 struct nicvf *nic = netdev_priv(netdev); 1050 int qid = skb_get_queue_mapping(skb); 1051 struct netdev_queue *txq = netdev_get_tx_queue(netdev, qid); 1052 1053 /* Check for minimum packet length */ 1054 if (skb->len <= ETH_HLEN) { 1055 dev_kfree_skb(skb); 1056 return NETDEV_TX_OK; 1057 } 1058 1059 if (!netif_tx_queue_stopped(txq) && !nicvf_sq_append_skb(nic, skb)) { 1060 netif_tx_stop_queue(txq); 1061 nic->drv_stats.txq_stop++; 1062 if (netif_msg_tx_err(nic)) 1063 netdev_warn(netdev, 1064 "%s: Transmit ring full, stopping SQ%d\n", 1065 netdev->name, qid); 1066 return NETDEV_TX_BUSY; 1067 } 1068 1069 return NETDEV_TX_OK; 1070 } 1071 1072 static inline void nicvf_free_cq_poll(struct nicvf *nic) 1073 { 1074 struct nicvf_cq_poll *cq_poll; 1075 int qidx; 1076 1077 for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) { 1078 cq_poll = nic->napi[qidx]; 1079 if (!cq_poll) 1080 continue; 1081 nic->napi[qidx] = NULL; 1082 kfree(cq_poll); 1083 } 1084 } 1085 1086 int nicvf_stop(struct net_device *netdev) 1087 { 1088 int irq, qidx; 1089 struct nicvf *nic = netdev_priv(netdev); 1090 struct queue_set *qs = nic->qs; 1091 struct nicvf_cq_poll *cq_poll = NULL; 1092 union nic_mbx mbx = {}; 1093 1094 mbx.msg.msg = NIC_MBOX_MSG_SHUTDOWN; 1095 nicvf_send_msg_to_pf(nic, &mbx); 1096 1097 netif_carrier_off(netdev); 1098 netif_tx_stop_all_queues(nic->netdev); 1099 nic->link_up = false; 1100 1101 /* Teardown secondary qsets first */ 1102 if (!nic->sqs_mode) { 1103 for (qidx = 0; qidx < nic->sqs_count; qidx++) { 1104 if (!nic->snicvf[qidx]) 1105 continue; 1106 nicvf_stop(nic->snicvf[qidx]->netdev); 1107 nic->snicvf[qidx] = NULL; 1108 } 1109 } 1110 1111 /* Disable RBDR & QS error interrupts */ 1112 for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) { 1113 nicvf_disable_intr(nic, NICVF_INTR_RBDR, qidx); 1114 nicvf_clear_intr(nic, NICVF_INTR_RBDR, qidx); 1115 } 1116 nicvf_disable_intr(nic, NICVF_INTR_QS_ERR, 0); 1117 nicvf_clear_intr(nic, NICVF_INTR_QS_ERR, 0); 1118 1119 /* Wait for pending IRQ handlers to finish */ 1120 for (irq = 0; irq < nic->num_vec; irq++) 1121 synchronize_irq(nic->msix_entries[irq].vector); 1122 1123 tasklet_kill(&nic->rbdr_task); 1124 tasklet_kill(&nic->qs_err_task); 1125 if (nic->rb_work_scheduled) 1126 cancel_delayed_work_sync(&nic->rbdr_work); 1127 1128 for (qidx = 0; qidx < nic->qs->cq_cnt; qidx++) { 1129 cq_poll = nic->napi[qidx]; 1130 if (!cq_poll) 1131 continue; 1132 napi_synchronize(&cq_poll->napi); 1133 /* CQ intr is enabled while napi_complete, 1134 * so disable it now 1135 */ 1136 nicvf_disable_intr(nic, NICVF_INTR_CQ, qidx); 1137 nicvf_clear_intr(nic, NICVF_INTR_CQ, qidx); 1138 napi_disable(&cq_poll->napi); 1139 netif_napi_del(&cq_poll->napi); 1140 } 1141 1142 netif_tx_disable(netdev); 1143 1144 /* Free resources */ 1145 nicvf_config_data_transfer(nic, false); 1146 1147 /* Disable HW Qset */ 1148 nicvf_qset_config(nic, false); 1149 1150 /* disable mailbox interrupt */ 1151 nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0); 1152 1153 nicvf_unregister_interrupts(nic); 1154 1155 nicvf_free_cq_poll(nic); 1156 1157 /* Clear multiqset info */ 1158 nic->pnicvf = nic; 1159 1160 return 0; 1161 } 1162 1163 int nicvf_open(struct net_device *netdev) 1164 { 1165 int err, qidx; 1166 struct nicvf *nic = netdev_priv(netdev); 1167 struct queue_set *qs = nic->qs; 1168 struct nicvf_cq_poll *cq_poll = NULL; 1169 1170 nic->mtu = netdev->mtu; 1171 1172 netif_carrier_off(netdev); 1173 1174 err = nicvf_register_misc_interrupt(nic); 1175 if (err) 1176 return err; 1177 1178 /* Register NAPI handler for processing CQEs */ 1179 for (qidx = 0; qidx < qs->cq_cnt; qidx++) { 1180 cq_poll = kzalloc(sizeof(*cq_poll), GFP_KERNEL); 1181 if (!cq_poll) { 1182 err = -ENOMEM; 1183 goto napi_del; 1184 } 1185 cq_poll->cq_idx = qidx; 1186 cq_poll->nicvf = nic; 1187 netif_napi_add(netdev, &cq_poll->napi, nicvf_poll, 1188 NAPI_POLL_WEIGHT); 1189 napi_enable(&cq_poll->napi); 1190 nic->napi[qidx] = cq_poll; 1191 } 1192 1193 /* Check if we got MAC address from PF or else generate a radom MAC */ 1194 if (is_zero_ether_addr(netdev->dev_addr)) { 1195 eth_hw_addr_random(netdev); 1196 nicvf_hw_set_mac_addr(nic, netdev); 1197 } 1198 1199 if (nic->set_mac_pending) { 1200 nic->set_mac_pending = false; 1201 nicvf_hw_set_mac_addr(nic, netdev); 1202 } 1203 1204 /* Init tasklet for handling Qset err interrupt */ 1205 tasklet_init(&nic->qs_err_task, nicvf_handle_qs_err, 1206 (unsigned long)nic); 1207 1208 /* Init RBDR tasklet which will refill RBDR */ 1209 tasklet_init(&nic->rbdr_task, nicvf_rbdr_task, 1210 (unsigned long)nic); 1211 INIT_DELAYED_WORK(&nic->rbdr_work, nicvf_rbdr_work); 1212 1213 /* Configure CPI alorithm */ 1214 nic->cpi_alg = cpi_alg; 1215 if (!nic->sqs_mode) 1216 nicvf_config_cpi(nic); 1217 1218 nicvf_request_sqs(nic); 1219 if (nic->sqs_mode) 1220 nicvf_get_primary_vf_struct(nic); 1221 1222 /* Configure receive side scaling */ 1223 if (!nic->sqs_mode) 1224 nicvf_rss_init(nic); 1225 1226 err = nicvf_register_interrupts(nic); 1227 if (err) 1228 goto cleanup; 1229 1230 /* Initialize the queues */ 1231 err = nicvf_init_resources(nic); 1232 if (err) 1233 goto cleanup; 1234 1235 /* Make sure queue initialization is written */ 1236 wmb(); 1237 1238 nicvf_reg_write(nic, NIC_VF_INT, -1); 1239 /* Enable Qset err interrupt */ 1240 nicvf_enable_intr(nic, NICVF_INTR_QS_ERR, 0); 1241 1242 /* Enable completion queue interrupt */ 1243 for (qidx = 0; qidx < qs->cq_cnt; qidx++) 1244 nicvf_enable_intr(nic, NICVF_INTR_CQ, qidx); 1245 1246 /* Enable RBDR threshold interrupt */ 1247 for (qidx = 0; qidx < qs->rbdr_cnt; qidx++) 1248 nicvf_enable_intr(nic, NICVF_INTR_RBDR, qidx); 1249 1250 nic->drv_stats.txq_stop = 0; 1251 nic->drv_stats.txq_wake = 0; 1252 1253 return 0; 1254 cleanup: 1255 nicvf_disable_intr(nic, NICVF_INTR_MBOX, 0); 1256 nicvf_unregister_interrupts(nic); 1257 tasklet_kill(&nic->qs_err_task); 1258 tasklet_kill(&nic->rbdr_task); 1259 napi_del: 1260 for (qidx = 0; qidx < qs->cq_cnt; qidx++) { 1261 cq_poll = nic->napi[qidx]; 1262 if (!cq_poll) 1263 continue; 1264 napi_disable(&cq_poll->napi); 1265 netif_napi_del(&cq_poll->napi); 1266 } 1267 nicvf_free_cq_poll(nic); 1268 return err; 1269 } 1270 1271 static int nicvf_update_hw_max_frs(struct nicvf *nic, int mtu) 1272 { 1273 union nic_mbx mbx = {}; 1274 1275 mbx.frs.msg = NIC_MBOX_MSG_SET_MAX_FRS; 1276 mbx.frs.max_frs = mtu; 1277 mbx.frs.vf_id = nic->vf_id; 1278 1279 return nicvf_send_msg_to_pf(nic, &mbx); 1280 } 1281 1282 static int nicvf_change_mtu(struct net_device *netdev, int new_mtu) 1283 { 1284 struct nicvf *nic = netdev_priv(netdev); 1285 1286 if (new_mtu > NIC_HW_MAX_FRS) 1287 return -EINVAL; 1288 1289 if (new_mtu < NIC_HW_MIN_FRS) 1290 return -EINVAL; 1291 1292 if (nicvf_update_hw_max_frs(nic, new_mtu)) 1293 return -EINVAL; 1294 netdev->mtu = new_mtu; 1295 nic->mtu = new_mtu; 1296 1297 return 0; 1298 } 1299 1300 static int nicvf_set_mac_address(struct net_device *netdev, void *p) 1301 { 1302 struct sockaddr *addr = p; 1303 struct nicvf *nic = netdev_priv(netdev); 1304 1305 if (!is_valid_ether_addr(addr->sa_data)) 1306 return -EADDRNOTAVAIL; 1307 1308 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len); 1309 1310 if (nic->msix_enabled) { 1311 if (nicvf_hw_set_mac_addr(nic, netdev)) 1312 return -EBUSY; 1313 } else { 1314 nic->set_mac_pending = true; 1315 } 1316 1317 return 0; 1318 } 1319 1320 void nicvf_update_lmac_stats(struct nicvf *nic) 1321 { 1322 int stat = 0; 1323 union nic_mbx mbx = {}; 1324 1325 if (!netif_running(nic->netdev)) 1326 return; 1327 1328 mbx.bgx_stats.msg = NIC_MBOX_MSG_BGX_STATS; 1329 mbx.bgx_stats.vf_id = nic->vf_id; 1330 /* Rx stats */ 1331 mbx.bgx_stats.rx = 1; 1332 while (stat < BGX_RX_STATS_COUNT) { 1333 mbx.bgx_stats.idx = stat; 1334 if (nicvf_send_msg_to_pf(nic, &mbx)) 1335 return; 1336 stat++; 1337 } 1338 1339 stat = 0; 1340 1341 /* Tx stats */ 1342 mbx.bgx_stats.rx = 0; 1343 while (stat < BGX_TX_STATS_COUNT) { 1344 mbx.bgx_stats.idx = stat; 1345 if (nicvf_send_msg_to_pf(nic, &mbx)) 1346 return; 1347 stat++; 1348 } 1349 } 1350 1351 void nicvf_update_stats(struct nicvf *nic) 1352 { 1353 int qidx; 1354 struct nicvf_hw_stats *stats = &nic->hw_stats; 1355 struct nicvf_drv_stats *drv_stats = &nic->drv_stats; 1356 struct queue_set *qs = nic->qs; 1357 1358 #define GET_RX_STATS(reg) \ 1359 nicvf_reg_read(nic, NIC_VNIC_RX_STAT_0_13 | (reg << 3)) 1360 #define GET_TX_STATS(reg) \ 1361 nicvf_reg_read(nic, NIC_VNIC_TX_STAT_0_4 | (reg << 3)) 1362 1363 stats->rx_bytes = GET_RX_STATS(RX_OCTS); 1364 stats->rx_ucast_frames = GET_RX_STATS(RX_UCAST); 1365 stats->rx_bcast_frames = GET_RX_STATS(RX_BCAST); 1366 stats->rx_mcast_frames = GET_RX_STATS(RX_MCAST); 1367 stats->rx_fcs_errors = GET_RX_STATS(RX_FCS); 1368 stats->rx_l2_errors = GET_RX_STATS(RX_L2ERR); 1369 stats->rx_drop_red = GET_RX_STATS(RX_RED); 1370 stats->rx_drop_red_bytes = GET_RX_STATS(RX_RED_OCTS); 1371 stats->rx_drop_overrun = GET_RX_STATS(RX_ORUN); 1372 stats->rx_drop_overrun_bytes = GET_RX_STATS(RX_ORUN_OCTS); 1373 stats->rx_drop_bcast = GET_RX_STATS(RX_DRP_BCAST); 1374 stats->rx_drop_mcast = GET_RX_STATS(RX_DRP_MCAST); 1375 stats->rx_drop_l3_bcast = GET_RX_STATS(RX_DRP_L3BCAST); 1376 stats->rx_drop_l3_mcast = GET_RX_STATS(RX_DRP_L3MCAST); 1377 1378 stats->tx_bytes_ok = GET_TX_STATS(TX_OCTS); 1379 stats->tx_ucast_frames_ok = GET_TX_STATS(TX_UCAST); 1380 stats->tx_bcast_frames_ok = GET_TX_STATS(TX_BCAST); 1381 stats->tx_mcast_frames_ok = GET_TX_STATS(TX_MCAST); 1382 stats->tx_drops = GET_TX_STATS(TX_DROP); 1383 1384 drv_stats->tx_frames_ok = stats->tx_ucast_frames_ok + 1385 stats->tx_bcast_frames_ok + 1386 stats->tx_mcast_frames_ok; 1387 drv_stats->rx_frames_ok = stats->rx_ucast_frames + 1388 stats->rx_bcast_frames + 1389 stats->rx_mcast_frames; 1390 drv_stats->rx_drops = stats->rx_drop_red + 1391 stats->rx_drop_overrun; 1392 drv_stats->tx_drops = stats->tx_drops; 1393 1394 /* Update RQ and SQ stats */ 1395 for (qidx = 0; qidx < qs->rq_cnt; qidx++) 1396 nicvf_update_rq_stats(nic, qidx); 1397 for (qidx = 0; qidx < qs->sq_cnt; qidx++) 1398 nicvf_update_sq_stats(nic, qidx); 1399 } 1400 1401 static struct rtnl_link_stats64 *nicvf_get_stats64(struct net_device *netdev, 1402 struct rtnl_link_stats64 *stats) 1403 { 1404 struct nicvf *nic = netdev_priv(netdev); 1405 struct nicvf_hw_stats *hw_stats = &nic->hw_stats; 1406 struct nicvf_drv_stats *drv_stats = &nic->drv_stats; 1407 1408 nicvf_update_stats(nic); 1409 1410 stats->rx_bytes = hw_stats->rx_bytes; 1411 stats->rx_packets = drv_stats->rx_frames_ok; 1412 stats->rx_dropped = drv_stats->rx_drops; 1413 stats->multicast = hw_stats->rx_mcast_frames; 1414 1415 stats->tx_bytes = hw_stats->tx_bytes_ok; 1416 stats->tx_packets = drv_stats->tx_frames_ok; 1417 stats->tx_dropped = drv_stats->tx_drops; 1418 1419 return stats; 1420 } 1421 1422 static void nicvf_tx_timeout(struct net_device *dev) 1423 { 1424 struct nicvf *nic = netdev_priv(dev); 1425 1426 if (netif_msg_tx_err(nic)) 1427 netdev_warn(dev, "%s: Transmit timed out, resetting\n", 1428 dev->name); 1429 1430 nic->drv_stats.tx_timeout++; 1431 schedule_work(&nic->reset_task); 1432 } 1433 1434 static void nicvf_reset_task(struct work_struct *work) 1435 { 1436 struct nicvf *nic; 1437 1438 nic = container_of(work, struct nicvf, reset_task); 1439 1440 if (!netif_running(nic->netdev)) 1441 return; 1442 1443 nicvf_stop(nic->netdev); 1444 nicvf_open(nic->netdev); 1445 netif_trans_update(nic->netdev); 1446 } 1447 1448 static int nicvf_config_loopback(struct nicvf *nic, 1449 netdev_features_t features) 1450 { 1451 union nic_mbx mbx = {}; 1452 1453 mbx.lbk.msg = NIC_MBOX_MSG_LOOPBACK; 1454 mbx.lbk.vf_id = nic->vf_id; 1455 mbx.lbk.enable = (features & NETIF_F_LOOPBACK) != 0; 1456 1457 return nicvf_send_msg_to_pf(nic, &mbx); 1458 } 1459 1460 static netdev_features_t nicvf_fix_features(struct net_device *netdev, 1461 netdev_features_t features) 1462 { 1463 struct nicvf *nic = netdev_priv(netdev); 1464 1465 if ((features & NETIF_F_LOOPBACK) && 1466 netif_running(netdev) && !nic->loopback_supported) 1467 features &= ~NETIF_F_LOOPBACK; 1468 1469 return features; 1470 } 1471 1472 static int nicvf_set_features(struct net_device *netdev, 1473 netdev_features_t features) 1474 { 1475 struct nicvf *nic = netdev_priv(netdev); 1476 netdev_features_t changed = features ^ netdev->features; 1477 1478 if (changed & NETIF_F_HW_VLAN_CTAG_RX) 1479 nicvf_config_vlan_stripping(nic, features); 1480 1481 if ((changed & NETIF_F_LOOPBACK) && netif_running(netdev)) 1482 return nicvf_config_loopback(nic, features); 1483 1484 return 0; 1485 } 1486 1487 static const struct net_device_ops nicvf_netdev_ops = { 1488 .ndo_open = nicvf_open, 1489 .ndo_stop = nicvf_stop, 1490 .ndo_start_xmit = nicvf_xmit, 1491 .ndo_change_mtu = nicvf_change_mtu, 1492 .ndo_set_mac_address = nicvf_set_mac_address, 1493 .ndo_get_stats64 = nicvf_get_stats64, 1494 .ndo_tx_timeout = nicvf_tx_timeout, 1495 .ndo_fix_features = nicvf_fix_features, 1496 .ndo_set_features = nicvf_set_features, 1497 }; 1498 1499 static int nicvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 1500 { 1501 struct device *dev = &pdev->dev; 1502 struct net_device *netdev; 1503 struct nicvf *nic; 1504 int err, qcount; 1505 1506 err = pci_enable_device(pdev); 1507 if (err) { 1508 dev_err(dev, "Failed to enable PCI device\n"); 1509 return err; 1510 } 1511 1512 err = pci_request_regions(pdev, DRV_NAME); 1513 if (err) { 1514 dev_err(dev, "PCI request regions failed 0x%x\n", err); 1515 goto err_disable_device; 1516 } 1517 1518 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(48)); 1519 if (err) { 1520 dev_err(dev, "Unable to get usable DMA configuration\n"); 1521 goto err_release_regions; 1522 } 1523 1524 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48)); 1525 if (err) { 1526 dev_err(dev, "unable to get 48-bit DMA for consistent allocations\n"); 1527 goto err_release_regions; 1528 } 1529 1530 qcount = MAX_CMP_QUEUES_PER_QS; 1531 1532 /* Restrict multiqset support only for host bound VFs */ 1533 if (pdev->is_virtfn) { 1534 /* Set max number of queues per VF */ 1535 qcount = roundup(num_online_cpus(), MAX_CMP_QUEUES_PER_QS); 1536 qcount = min(qcount, 1537 (MAX_SQS_PER_VF + 1) * MAX_CMP_QUEUES_PER_QS); 1538 } 1539 1540 netdev = alloc_etherdev_mqs(sizeof(struct nicvf), qcount, qcount); 1541 if (!netdev) { 1542 err = -ENOMEM; 1543 goto err_release_regions; 1544 } 1545 1546 pci_set_drvdata(pdev, netdev); 1547 1548 SET_NETDEV_DEV(netdev, &pdev->dev); 1549 1550 nic = netdev_priv(netdev); 1551 nic->netdev = netdev; 1552 nic->pdev = pdev; 1553 nic->pnicvf = nic; 1554 nic->max_queues = qcount; 1555 1556 /* MAP VF's configuration registers */ 1557 nic->reg_base = pcim_iomap(pdev, PCI_CFG_REG_BAR_NUM, 0); 1558 if (!nic->reg_base) { 1559 dev_err(dev, "Cannot map config register space, aborting\n"); 1560 err = -ENOMEM; 1561 goto err_free_netdev; 1562 } 1563 1564 err = nicvf_set_qset_resources(nic); 1565 if (err) 1566 goto err_free_netdev; 1567 1568 /* Check if PF is alive and get MAC address for this VF */ 1569 err = nicvf_register_misc_interrupt(nic); 1570 if (err) 1571 goto err_free_netdev; 1572 1573 nicvf_send_vf_struct(nic); 1574 1575 if (!pass1_silicon(nic->pdev)) 1576 nic->hw_tso = true; 1577 1578 /* Check if this VF is in QS only mode */ 1579 if (nic->sqs_mode) 1580 return 0; 1581 1582 err = nicvf_set_real_num_queues(netdev, nic->tx_queues, nic->rx_queues); 1583 if (err) 1584 goto err_unregister_interrupts; 1585 1586 netdev->hw_features = (NETIF_F_RXCSUM | NETIF_F_IP_CSUM | NETIF_F_SG | 1587 NETIF_F_TSO | NETIF_F_GRO | 1588 NETIF_F_HW_VLAN_CTAG_RX); 1589 1590 netdev->hw_features |= NETIF_F_RXHASH; 1591 1592 netdev->features |= netdev->hw_features; 1593 netdev->hw_features |= NETIF_F_LOOPBACK; 1594 1595 netdev->vlan_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO; 1596 1597 netdev->netdev_ops = &nicvf_netdev_ops; 1598 netdev->watchdog_timeo = NICVF_TX_TIMEOUT; 1599 1600 INIT_WORK(&nic->reset_task, nicvf_reset_task); 1601 1602 err = register_netdev(netdev); 1603 if (err) { 1604 dev_err(dev, "Failed to register netdevice\n"); 1605 goto err_unregister_interrupts; 1606 } 1607 1608 nic->msg_enable = debug; 1609 1610 nicvf_set_ethtool_ops(netdev); 1611 1612 return 0; 1613 1614 err_unregister_interrupts: 1615 nicvf_unregister_interrupts(nic); 1616 err_free_netdev: 1617 pci_set_drvdata(pdev, NULL); 1618 free_netdev(netdev); 1619 err_release_regions: 1620 pci_release_regions(pdev); 1621 err_disable_device: 1622 pci_disable_device(pdev); 1623 return err; 1624 } 1625 1626 static void nicvf_remove(struct pci_dev *pdev) 1627 { 1628 struct net_device *netdev = pci_get_drvdata(pdev); 1629 struct nicvf *nic; 1630 struct net_device *pnetdev; 1631 1632 if (!netdev) 1633 return; 1634 1635 nic = netdev_priv(netdev); 1636 pnetdev = nic->pnicvf->netdev; 1637 1638 /* Check if this Qset is assigned to different VF. 1639 * If yes, clean primary and all secondary Qsets. 1640 */ 1641 if (pnetdev && (pnetdev->reg_state == NETREG_REGISTERED)) 1642 unregister_netdev(pnetdev); 1643 nicvf_unregister_interrupts(nic); 1644 pci_set_drvdata(pdev, NULL); 1645 free_netdev(netdev); 1646 pci_release_regions(pdev); 1647 pci_disable_device(pdev); 1648 } 1649 1650 static void nicvf_shutdown(struct pci_dev *pdev) 1651 { 1652 nicvf_remove(pdev); 1653 } 1654 1655 static struct pci_driver nicvf_driver = { 1656 .name = DRV_NAME, 1657 .id_table = nicvf_id_table, 1658 .probe = nicvf_probe, 1659 .remove = nicvf_remove, 1660 .shutdown = nicvf_shutdown, 1661 }; 1662 1663 static int __init nicvf_init_module(void) 1664 { 1665 pr_info("%s, ver %s\n", DRV_NAME, DRV_VERSION); 1666 1667 return pci_register_driver(&nicvf_driver); 1668 } 1669 1670 static void __exit nicvf_cleanup_module(void) 1671 { 1672 pci_unregister_driver(&nicvf_driver); 1673 } 1674 1675 module_init(nicvf_init_module); 1676 module_exit(nicvf_cleanup_module); 1677