1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) 2 /* Copyright (C) 2015-2019 Netronome Systems, Inc. */ 3 4 /* 5 * nfp_net_common.c 6 * Netronome network device driver: Common functions between PF and VF 7 * Authors: Jakub Kicinski <jakub.kicinski@netronome.com> 8 * Jason McMullan <jason.mcmullan@netronome.com> 9 * Rolf Neugebauer <rolf.neugebauer@netronome.com> 10 * Brad Petrus <brad.petrus@netronome.com> 11 * Chris Telfer <chris.telfer@netronome.com> 12 */ 13 14 #include <linux/bitfield.h> 15 #include <linux/bpf.h> 16 #include <linux/module.h> 17 #include <linux/kernel.h> 18 #include <linux/init.h> 19 #include <linux/fs.h> 20 #include <linux/netdevice.h> 21 #include <linux/etherdevice.h> 22 #include <linux/interrupt.h> 23 #include <linux/ip.h> 24 #include <linux/ipv6.h> 25 #include <linux/mm.h> 26 #include <linux/overflow.h> 27 #include <linux/page_ref.h> 28 #include <linux/pci.h> 29 #include <linux/pci_regs.h> 30 #include <linux/msi.h> 31 #include <linux/ethtool.h> 32 #include <linux/log2.h> 33 #include <linux/if_vlan.h> 34 #include <linux/if_bridge.h> 35 #include <linux/random.h> 36 #include <linux/vmalloc.h> 37 #include <linux/ktime.h> 38 39 #include <net/tls.h> 40 #include <net/vxlan.h> 41 #include <net/xdp_sock_drv.h> 42 43 #include "nfpcore/nfp_dev.h" 44 #include "nfpcore/nfp_nsp.h" 45 #include "ccm.h" 46 #include "nfp_app.h" 47 #include "nfp_net_ctrl.h" 48 #include "nfp_net.h" 49 #include "nfp_net_dp.h" 50 #include "nfp_net_sriov.h" 51 #include "nfp_net_xsk.h" 52 #include "nfp_port.h" 53 #include "crypto/crypto.h" 54 #include "crypto/fw.h" 55 56 /** 57 * nfp_net_get_fw_version() - Read and parse the FW version 58 * @fw_ver: Output fw_version structure to read to 59 * @ctrl_bar: Mapped address of the control BAR 60 */ 61 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver, 62 void __iomem *ctrl_bar) 63 { 64 u32 reg; 65 66 reg = readl(ctrl_bar + NFP_NET_CFG_VERSION); 67 put_unaligned_le32(reg, fw_ver); 68 } 69 70 u32 nfp_qcp_queue_offset(const struct nfp_dev_info *dev_info, u16 queue) 71 { 72 queue &= dev_info->qc_idx_mask; 73 return dev_info->qc_addr_offset + NFP_QCP_QUEUE_ADDR_SZ * queue; 74 } 75 76 /* Firmware reconfig 77 * 78 * Firmware reconfig may take a while so we have two versions of it - 79 * synchronous and asynchronous (posted). All synchronous callers are holding 80 * RTNL so we don't have to worry about serializing them. 81 */ 82 static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update) 83 { 84 nn_writel(nn, NFP_NET_CFG_UPDATE, update); 85 /* ensure update is written before pinging HW */ 86 nn_pci_flush(nn); 87 nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1); 88 nn->reconfig_in_progress_update = update; 89 } 90 91 /* Pass 0 as update to run posted reconfigs. */ 92 static void nfp_net_reconfig_start_async(struct nfp_net *nn, u32 update) 93 { 94 update |= nn->reconfig_posted; 95 nn->reconfig_posted = 0; 96 97 nfp_net_reconfig_start(nn, update); 98 99 nn->reconfig_timer_active = true; 100 mod_timer(&nn->reconfig_timer, jiffies + NFP_NET_POLL_TIMEOUT * HZ); 101 } 102 103 static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check) 104 { 105 u32 reg; 106 107 reg = nn_readl(nn, NFP_NET_CFG_UPDATE); 108 if (reg == 0) 109 return true; 110 if (reg & NFP_NET_CFG_UPDATE_ERR) { 111 nn_err(nn, "Reconfig error (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n", 112 reg, nn->reconfig_in_progress_update, 113 nn_readl(nn, NFP_NET_CFG_CTRL)); 114 return true; 115 } else if (last_check) { 116 nn_err(nn, "Reconfig timeout (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n", 117 reg, nn->reconfig_in_progress_update, 118 nn_readl(nn, NFP_NET_CFG_CTRL)); 119 return true; 120 } 121 122 return false; 123 } 124 125 static bool __nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline) 126 { 127 bool timed_out = false; 128 int i; 129 130 /* Poll update field, waiting for NFP to ack the config. 131 * Do an opportunistic wait-busy loop, afterward sleep. 132 */ 133 for (i = 0; i < 50; i++) { 134 if (nfp_net_reconfig_check_done(nn, false)) 135 return false; 136 udelay(4); 137 } 138 139 while (!nfp_net_reconfig_check_done(nn, timed_out)) { 140 usleep_range(250, 500); 141 timed_out = time_is_before_eq_jiffies(deadline); 142 } 143 144 return timed_out; 145 } 146 147 static int nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline) 148 { 149 if (__nfp_net_reconfig_wait(nn, deadline)) 150 return -EIO; 151 152 if (nn_readl(nn, NFP_NET_CFG_UPDATE) & NFP_NET_CFG_UPDATE_ERR) 153 return -EIO; 154 155 return 0; 156 } 157 158 static void nfp_net_reconfig_timer(struct timer_list *t) 159 { 160 struct nfp_net *nn = from_timer(nn, t, reconfig_timer); 161 162 spin_lock_bh(&nn->reconfig_lock); 163 164 nn->reconfig_timer_active = false; 165 166 /* If sync caller is present it will take over from us */ 167 if (nn->reconfig_sync_present) 168 goto done; 169 170 /* Read reconfig status and report errors */ 171 nfp_net_reconfig_check_done(nn, true); 172 173 if (nn->reconfig_posted) 174 nfp_net_reconfig_start_async(nn, 0); 175 done: 176 spin_unlock_bh(&nn->reconfig_lock); 177 } 178 179 /** 180 * nfp_net_reconfig_post() - Post async reconfig request 181 * @nn: NFP Net device to reconfigure 182 * @update: The value for the update field in the BAR config 183 * 184 * Record FW reconfiguration request. Reconfiguration will be kicked off 185 * whenever reconfiguration machinery is idle. Multiple requests can be 186 * merged together! 187 */ 188 static void nfp_net_reconfig_post(struct nfp_net *nn, u32 update) 189 { 190 spin_lock_bh(&nn->reconfig_lock); 191 192 /* Sync caller will kick off async reconf when it's done, just post */ 193 if (nn->reconfig_sync_present) { 194 nn->reconfig_posted |= update; 195 goto done; 196 } 197 198 /* Opportunistically check if the previous command is done */ 199 if (!nn->reconfig_timer_active || 200 nfp_net_reconfig_check_done(nn, false)) 201 nfp_net_reconfig_start_async(nn, update); 202 else 203 nn->reconfig_posted |= update; 204 done: 205 spin_unlock_bh(&nn->reconfig_lock); 206 } 207 208 static void nfp_net_reconfig_sync_enter(struct nfp_net *nn) 209 { 210 bool cancelled_timer = false; 211 u32 pre_posted_requests; 212 213 spin_lock_bh(&nn->reconfig_lock); 214 215 WARN_ON(nn->reconfig_sync_present); 216 nn->reconfig_sync_present = true; 217 218 if (nn->reconfig_timer_active) { 219 nn->reconfig_timer_active = false; 220 cancelled_timer = true; 221 } 222 pre_posted_requests = nn->reconfig_posted; 223 nn->reconfig_posted = 0; 224 225 spin_unlock_bh(&nn->reconfig_lock); 226 227 if (cancelled_timer) { 228 del_timer_sync(&nn->reconfig_timer); 229 nfp_net_reconfig_wait(nn, nn->reconfig_timer.expires); 230 } 231 232 /* Run the posted reconfigs which were issued before we started */ 233 if (pre_posted_requests) { 234 nfp_net_reconfig_start(nn, pre_posted_requests); 235 nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT); 236 } 237 } 238 239 static void nfp_net_reconfig_wait_posted(struct nfp_net *nn) 240 { 241 nfp_net_reconfig_sync_enter(nn); 242 243 spin_lock_bh(&nn->reconfig_lock); 244 nn->reconfig_sync_present = false; 245 spin_unlock_bh(&nn->reconfig_lock); 246 } 247 248 /** 249 * __nfp_net_reconfig() - Reconfigure the firmware 250 * @nn: NFP Net device to reconfigure 251 * @update: The value for the update field in the BAR config 252 * 253 * Write the update word to the BAR and ping the reconfig queue. The 254 * poll until the firmware has acknowledged the update by zeroing the 255 * update word. 256 * 257 * Return: Negative errno on error, 0 on success 258 */ 259 int __nfp_net_reconfig(struct nfp_net *nn, u32 update) 260 { 261 int ret; 262 263 nfp_net_reconfig_sync_enter(nn); 264 265 nfp_net_reconfig_start(nn, update); 266 ret = nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT); 267 268 spin_lock_bh(&nn->reconfig_lock); 269 270 if (nn->reconfig_posted) 271 nfp_net_reconfig_start_async(nn, 0); 272 273 nn->reconfig_sync_present = false; 274 275 spin_unlock_bh(&nn->reconfig_lock); 276 277 return ret; 278 } 279 280 int nfp_net_reconfig(struct nfp_net *nn, u32 update) 281 { 282 int ret; 283 284 nn_ctrl_bar_lock(nn); 285 ret = __nfp_net_reconfig(nn, update); 286 nn_ctrl_bar_unlock(nn); 287 288 return ret; 289 } 290 291 int nfp_net_mbox_lock(struct nfp_net *nn, unsigned int data_size) 292 { 293 if (nn->tlv_caps.mbox_len < NFP_NET_CFG_MBOX_SIMPLE_VAL + data_size) { 294 nn_err(nn, "mailbox too small for %u of data (%u)\n", 295 data_size, nn->tlv_caps.mbox_len); 296 return -EIO; 297 } 298 299 nn_ctrl_bar_lock(nn); 300 return 0; 301 } 302 303 /** 304 * nfp_net_mbox_reconfig() - Reconfigure the firmware via the mailbox 305 * @nn: NFP Net device to reconfigure 306 * @mbox_cmd: The value for the mailbox command 307 * 308 * Helper function for mailbox updates 309 * 310 * Return: Negative errno on error, 0 on success 311 */ 312 int nfp_net_mbox_reconfig(struct nfp_net *nn, u32 mbox_cmd) 313 { 314 u32 mbox = nn->tlv_caps.mbox_off; 315 int ret; 316 317 nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd); 318 319 ret = __nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MBOX); 320 if (ret) { 321 nn_err(nn, "Mailbox update error\n"); 322 return ret; 323 } 324 325 return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET); 326 } 327 328 void nfp_net_mbox_reconfig_post(struct nfp_net *nn, u32 mbox_cmd) 329 { 330 u32 mbox = nn->tlv_caps.mbox_off; 331 332 nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd); 333 334 nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_MBOX); 335 } 336 337 int nfp_net_mbox_reconfig_wait_posted(struct nfp_net *nn) 338 { 339 u32 mbox = nn->tlv_caps.mbox_off; 340 341 nfp_net_reconfig_wait_posted(nn); 342 343 return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET); 344 } 345 346 int nfp_net_mbox_reconfig_and_unlock(struct nfp_net *nn, u32 mbox_cmd) 347 { 348 int ret; 349 350 ret = nfp_net_mbox_reconfig(nn, mbox_cmd); 351 nn_ctrl_bar_unlock(nn); 352 return ret; 353 } 354 355 /* Interrupt configuration and handling 356 */ 357 358 /** 359 * nfp_net_irqs_alloc() - allocates MSI-X irqs 360 * @pdev: PCI device structure 361 * @irq_entries: Array to be initialized and used to hold the irq entries 362 * @min_irqs: Minimal acceptable number of interrupts 363 * @wanted_irqs: Target number of interrupts to allocate 364 * 365 * Return: Number of irqs obtained or 0 on error. 366 */ 367 unsigned int 368 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries, 369 unsigned int min_irqs, unsigned int wanted_irqs) 370 { 371 unsigned int i; 372 int got_irqs; 373 374 for (i = 0; i < wanted_irqs; i++) 375 irq_entries[i].entry = i; 376 377 got_irqs = pci_enable_msix_range(pdev, irq_entries, 378 min_irqs, wanted_irqs); 379 if (got_irqs < 0) { 380 dev_err(&pdev->dev, "Failed to enable %d-%d MSI-X (err=%d)\n", 381 min_irqs, wanted_irqs, got_irqs); 382 return 0; 383 } 384 385 if (got_irqs < wanted_irqs) 386 dev_warn(&pdev->dev, "Unable to allocate %d IRQs got only %d\n", 387 wanted_irqs, got_irqs); 388 389 return got_irqs; 390 } 391 392 /** 393 * nfp_net_irqs_assign() - Assign interrupts allocated externally to netdev 394 * @nn: NFP Network structure 395 * @irq_entries: Table of allocated interrupts 396 * @n: Size of @irq_entries (number of entries to grab) 397 * 398 * After interrupts are allocated with nfp_net_irqs_alloc() this function 399 * should be called to assign them to a specific netdev (port). 400 */ 401 void 402 nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries, 403 unsigned int n) 404 { 405 struct nfp_net_dp *dp = &nn->dp; 406 407 nn->max_r_vecs = n - NFP_NET_NON_Q_VECTORS; 408 dp->num_r_vecs = nn->max_r_vecs; 409 410 memcpy(nn->irq_entries, irq_entries, sizeof(*irq_entries) * n); 411 412 if (dp->num_rx_rings > dp->num_r_vecs || 413 dp->num_tx_rings > dp->num_r_vecs) 414 dev_warn(nn->dp.dev, "More rings (%d,%d) than vectors (%d).\n", 415 dp->num_rx_rings, dp->num_tx_rings, 416 dp->num_r_vecs); 417 418 dp->num_rx_rings = min(dp->num_r_vecs, dp->num_rx_rings); 419 dp->num_tx_rings = min(dp->num_r_vecs, dp->num_tx_rings); 420 dp->num_stack_tx_rings = dp->num_tx_rings; 421 } 422 423 /** 424 * nfp_net_irqs_disable() - Disable interrupts 425 * @pdev: PCI device structure 426 * 427 * Undoes what @nfp_net_irqs_alloc() does. 428 */ 429 void nfp_net_irqs_disable(struct pci_dev *pdev) 430 { 431 pci_disable_msix(pdev); 432 } 433 434 /** 435 * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings. 436 * @irq: Interrupt 437 * @data: Opaque data structure 438 * 439 * Return: Indicate if the interrupt has been handled. 440 */ 441 static irqreturn_t nfp_net_irq_rxtx(int irq, void *data) 442 { 443 struct nfp_net_r_vector *r_vec = data; 444 445 /* Currently we cannot tell if it's a rx or tx interrupt, 446 * since dim does not need accurate event_ctr to calculate, 447 * we just use this counter for both rx and tx dim. 448 */ 449 r_vec->event_ctr++; 450 451 napi_schedule_irqoff(&r_vec->napi); 452 453 /* The FW auto-masks any interrupt, either via the MASK bit in 454 * the MSI-X table or via the per entry ICR field. So there 455 * is no need to disable interrupts here. 456 */ 457 return IRQ_HANDLED; 458 } 459 460 static irqreturn_t nfp_ctrl_irq_rxtx(int irq, void *data) 461 { 462 struct nfp_net_r_vector *r_vec = data; 463 464 tasklet_schedule(&r_vec->tasklet); 465 466 return IRQ_HANDLED; 467 } 468 469 /** 470 * nfp_net_read_link_status() - Reread link status from control BAR 471 * @nn: NFP Network structure 472 */ 473 static void nfp_net_read_link_status(struct nfp_net *nn) 474 { 475 unsigned long flags; 476 bool link_up; 477 u16 sts; 478 479 spin_lock_irqsave(&nn->link_status_lock, flags); 480 481 sts = nn_readw(nn, NFP_NET_CFG_STS); 482 link_up = !!(sts & NFP_NET_CFG_STS_LINK); 483 484 if (nn->link_up == link_up) 485 goto out; 486 487 nn->link_up = link_up; 488 if (nn->port) { 489 set_bit(NFP_PORT_CHANGED, &nn->port->flags); 490 if (nn->port->link_cb) 491 nn->port->link_cb(nn->port); 492 } 493 494 if (nn->link_up) { 495 netif_carrier_on(nn->dp.netdev); 496 netdev_info(nn->dp.netdev, "NIC Link is Up\n"); 497 } else { 498 netif_carrier_off(nn->dp.netdev); 499 netdev_info(nn->dp.netdev, "NIC Link is Down\n"); 500 } 501 out: 502 spin_unlock_irqrestore(&nn->link_status_lock, flags); 503 } 504 505 /** 506 * nfp_net_irq_lsc() - Interrupt service routine for link state changes 507 * @irq: Interrupt 508 * @data: Opaque data structure 509 * 510 * Return: Indicate if the interrupt has been handled. 511 */ 512 static irqreturn_t nfp_net_irq_lsc(int irq, void *data) 513 { 514 struct nfp_net *nn = data; 515 struct msix_entry *entry; 516 517 entry = &nn->irq_entries[NFP_NET_IRQ_LSC_IDX]; 518 519 nfp_net_read_link_status(nn); 520 521 nfp_net_irq_unmask(nn, entry->entry); 522 523 return IRQ_HANDLED; 524 } 525 526 /** 527 * nfp_net_irq_exn() - Interrupt service routine for exceptions 528 * @irq: Interrupt 529 * @data: Opaque data structure 530 * 531 * Return: Indicate if the interrupt has been handled. 532 */ 533 static irqreturn_t nfp_net_irq_exn(int irq, void *data) 534 { 535 struct nfp_net *nn = data; 536 537 nn_err(nn, "%s: UNIMPLEMENTED.\n", __func__); 538 /* XXX TO BE IMPLEMENTED */ 539 return IRQ_HANDLED; 540 } 541 542 /** 543 * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN) 544 * @nn: NFP Network structure 545 * @ctrl_offset: Control BAR offset where IRQ configuration should be written 546 * @format: printf-style format to construct the interrupt name 547 * @name: Pointer to allocated space for interrupt name 548 * @name_sz: Size of space for interrupt name 549 * @vector_idx: Index of MSI-X vector used for this interrupt 550 * @handler: IRQ handler to register for this interrupt 551 */ 552 static int 553 nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset, 554 const char *format, char *name, size_t name_sz, 555 unsigned int vector_idx, irq_handler_t handler) 556 { 557 struct msix_entry *entry; 558 int err; 559 560 entry = &nn->irq_entries[vector_idx]; 561 562 snprintf(name, name_sz, format, nfp_net_name(nn)); 563 err = request_irq(entry->vector, handler, 0, name, nn); 564 if (err) { 565 nn_err(nn, "Failed to request IRQ %d (err=%d).\n", 566 entry->vector, err); 567 return err; 568 } 569 nn_writeb(nn, ctrl_offset, entry->entry); 570 nfp_net_irq_unmask(nn, entry->entry); 571 572 return 0; 573 } 574 575 /** 576 * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN) 577 * @nn: NFP Network structure 578 * @ctrl_offset: Control BAR offset where IRQ configuration should be written 579 * @vector_idx: Index of MSI-X vector used for this interrupt 580 */ 581 static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset, 582 unsigned int vector_idx) 583 { 584 nn_writeb(nn, ctrl_offset, 0xff); 585 nn_pci_flush(nn); 586 free_irq(nn->irq_entries[vector_idx].vector, nn); 587 } 588 589 struct sk_buff * 590 nfp_net_tls_tx(struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec, 591 struct sk_buff *skb, u64 *tls_handle, int *nr_frags) 592 { 593 #ifdef CONFIG_TLS_DEVICE 594 struct nfp_net_tls_offload_ctx *ntls; 595 struct sk_buff *nskb; 596 bool resync_pending; 597 u32 datalen, seq; 598 599 if (likely(!dp->ktls_tx)) 600 return skb; 601 if (!skb->sk || !tls_is_sk_tx_device_offloaded(skb->sk)) 602 return skb; 603 604 datalen = skb->len - skb_tcp_all_headers(skb); 605 seq = ntohl(tcp_hdr(skb)->seq); 606 ntls = tls_driver_ctx(skb->sk, TLS_OFFLOAD_CTX_DIR_TX); 607 resync_pending = tls_offload_tx_resync_pending(skb->sk); 608 if (unlikely(resync_pending || ntls->next_seq != seq)) { 609 /* Pure ACK out of order already */ 610 if (!datalen) 611 return skb; 612 613 u64_stats_update_begin(&r_vec->tx_sync); 614 r_vec->tls_tx_fallback++; 615 u64_stats_update_end(&r_vec->tx_sync); 616 617 nskb = tls_encrypt_skb(skb); 618 if (!nskb) { 619 u64_stats_update_begin(&r_vec->tx_sync); 620 r_vec->tls_tx_no_fallback++; 621 u64_stats_update_end(&r_vec->tx_sync); 622 return NULL; 623 } 624 /* encryption wasn't necessary */ 625 if (nskb == skb) 626 return skb; 627 /* we don't re-check ring space */ 628 if (unlikely(skb_is_nonlinear(nskb))) { 629 nn_dp_warn(dp, "tls_encrypt_skb() produced fragmented frame\n"); 630 u64_stats_update_begin(&r_vec->tx_sync); 631 r_vec->tx_errors++; 632 u64_stats_update_end(&r_vec->tx_sync); 633 dev_kfree_skb_any(nskb); 634 return NULL; 635 } 636 637 /* jump forward, a TX may have gotten lost, need to sync TX */ 638 if (!resync_pending && seq - ntls->next_seq < U32_MAX / 4) 639 tls_offload_tx_resync_request(nskb->sk, seq, 640 ntls->next_seq); 641 642 *nr_frags = 0; 643 return nskb; 644 } 645 646 if (datalen) { 647 u64_stats_update_begin(&r_vec->tx_sync); 648 if (!skb_is_gso(skb)) 649 r_vec->hw_tls_tx++; 650 else 651 r_vec->hw_tls_tx += skb_shinfo(skb)->gso_segs; 652 u64_stats_update_end(&r_vec->tx_sync); 653 } 654 655 memcpy(tls_handle, ntls->fw_handle, sizeof(ntls->fw_handle)); 656 ntls->next_seq += datalen; 657 #endif 658 return skb; 659 } 660 661 void nfp_net_tls_tx_undo(struct sk_buff *skb, u64 tls_handle) 662 { 663 #ifdef CONFIG_TLS_DEVICE 664 struct nfp_net_tls_offload_ctx *ntls; 665 u32 datalen, seq; 666 667 if (!tls_handle) 668 return; 669 if (WARN_ON_ONCE(!skb->sk || !tls_is_sk_tx_device_offloaded(skb->sk))) 670 return; 671 672 datalen = skb->len - skb_tcp_all_headers(skb); 673 seq = ntohl(tcp_hdr(skb)->seq); 674 675 ntls = tls_driver_ctx(skb->sk, TLS_OFFLOAD_CTX_DIR_TX); 676 if (ntls->next_seq == seq + datalen) 677 ntls->next_seq = seq; 678 else 679 WARN_ON_ONCE(1); 680 #endif 681 } 682 683 static void nfp_net_tx_timeout(struct net_device *netdev, unsigned int txqueue) 684 { 685 struct nfp_net *nn = netdev_priv(netdev); 686 687 nn_warn(nn, "TX watchdog timeout on ring: %u\n", txqueue); 688 } 689 690 /* Receive processing */ 691 static unsigned int 692 nfp_net_calc_fl_bufsz_data(struct nfp_net_dp *dp) 693 { 694 unsigned int fl_bufsz = 0; 695 696 if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC) 697 fl_bufsz += NFP_NET_MAX_PREPEND; 698 else 699 fl_bufsz += dp->rx_offset; 700 fl_bufsz += ETH_HLEN + VLAN_HLEN * 2 + dp->mtu; 701 702 return fl_bufsz; 703 } 704 705 static unsigned int nfp_net_calc_fl_bufsz(struct nfp_net_dp *dp) 706 { 707 unsigned int fl_bufsz; 708 709 fl_bufsz = NFP_NET_RX_BUF_HEADROOM; 710 fl_bufsz += dp->rx_dma_off; 711 fl_bufsz += nfp_net_calc_fl_bufsz_data(dp); 712 713 fl_bufsz = SKB_DATA_ALIGN(fl_bufsz); 714 fl_bufsz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 715 716 return fl_bufsz; 717 } 718 719 static unsigned int nfp_net_calc_fl_bufsz_xsk(struct nfp_net_dp *dp) 720 { 721 unsigned int fl_bufsz; 722 723 fl_bufsz = XDP_PACKET_HEADROOM; 724 fl_bufsz += nfp_net_calc_fl_bufsz_data(dp); 725 726 return fl_bufsz; 727 } 728 729 /* Setup and Configuration 730 */ 731 732 /** 733 * nfp_net_vecs_init() - Assign IRQs and setup rvecs. 734 * @nn: NFP Network structure 735 */ 736 static void nfp_net_vecs_init(struct nfp_net *nn) 737 { 738 struct nfp_net_r_vector *r_vec; 739 int r; 740 741 nn->lsc_handler = nfp_net_irq_lsc; 742 nn->exn_handler = nfp_net_irq_exn; 743 744 for (r = 0; r < nn->max_r_vecs; r++) { 745 struct msix_entry *entry; 746 747 entry = &nn->irq_entries[NFP_NET_NON_Q_VECTORS + r]; 748 749 r_vec = &nn->r_vecs[r]; 750 r_vec->nfp_net = nn; 751 r_vec->irq_entry = entry->entry; 752 r_vec->irq_vector = entry->vector; 753 754 if (nn->dp.netdev) { 755 r_vec->handler = nfp_net_irq_rxtx; 756 } else { 757 r_vec->handler = nfp_ctrl_irq_rxtx; 758 759 __skb_queue_head_init(&r_vec->queue); 760 spin_lock_init(&r_vec->lock); 761 tasklet_setup(&r_vec->tasklet, nn->dp.ops->ctrl_poll); 762 tasklet_disable(&r_vec->tasklet); 763 } 764 765 cpumask_set_cpu(r, &r_vec->affinity_mask); 766 } 767 } 768 769 static void 770 nfp_net_napi_add(struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec, int idx) 771 { 772 if (dp->netdev) 773 netif_napi_add(dp->netdev, &r_vec->napi, 774 nfp_net_has_xsk_pool_slow(dp, idx) ? 775 dp->ops->xsk_poll : dp->ops->poll, 776 NAPI_POLL_WEIGHT); 777 else 778 tasklet_enable(&r_vec->tasklet); 779 } 780 781 static void 782 nfp_net_napi_del(struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec) 783 { 784 if (dp->netdev) 785 netif_napi_del(&r_vec->napi); 786 else 787 tasklet_disable(&r_vec->tasklet); 788 } 789 790 static void 791 nfp_net_vector_assign_rings(struct nfp_net_dp *dp, 792 struct nfp_net_r_vector *r_vec, int idx) 793 { 794 r_vec->rx_ring = idx < dp->num_rx_rings ? &dp->rx_rings[idx] : NULL; 795 r_vec->tx_ring = 796 idx < dp->num_stack_tx_rings ? &dp->tx_rings[idx] : NULL; 797 798 r_vec->xdp_ring = idx < dp->num_tx_rings - dp->num_stack_tx_rings ? 799 &dp->tx_rings[dp->num_stack_tx_rings + idx] : NULL; 800 801 if (nfp_net_has_xsk_pool_slow(dp, idx) || r_vec->xsk_pool) { 802 r_vec->xsk_pool = dp->xdp_prog ? dp->xsk_pools[idx] : NULL; 803 804 if (r_vec->xsk_pool) 805 xsk_pool_set_rxq_info(r_vec->xsk_pool, 806 &r_vec->rx_ring->xdp_rxq); 807 808 nfp_net_napi_del(dp, r_vec); 809 nfp_net_napi_add(dp, r_vec, idx); 810 } 811 } 812 813 static int 814 nfp_net_prepare_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec, 815 int idx) 816 { 817 int err; 818 819 nfp_net_napi_add(&nn->dp, r_vec, idx); 820 821 snprintf(r_vec->name, sizeof(r_vec->name), 822 "%s-rxtx-%d", nfp_net_name(nn), idx); 823 err = request_irq(r_vec->irq_vector, r_vec->handler, 0, r_vec->name, 824 r_vec); 825 if (err) { 826 nfp_net_napi_del(&nn->dp, r_vec); 827 nn_err(nn, "Error requesting IRQ %d\n", r_vec->irq_vector); 828 return err; 829 } 830 disable_irq(r_vec->irq_vector); 831 832 irq_set_affinity_hint(r_vec->irq_vector, &r_vec->affinity_mask); 833 834 nn_dbg(nn, "RV%02d: irq=%03d/%03d\n", idx, r_vec->irq_vector, 835 r_vec->irq_entry); 836 837 return 0; 838 } 839 840 static void 841 nfp_net_cleanup_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec) 842 { 843 irq_set_affinity_hint(r_vec->irq_vector, NULL); 844 nfp_net_napi_del(&nn->dp, r_vec); 845 free_irq(r_vec->irq_vector, r_vec); 846 } 847 848 /** 849 * nfp_net_rss_write_itbl() - Write RSS indirection table to device 850 * @nn: NFP Net device to reconfigure 851 */ 852 void nfp_net_rss_write_itbl(struct nfp_net *nn) 853 { 854 int i; 855 856 for (i = 0; i < NFP_NET_CFG_RSS_ITBL_SZ; i += 4) 857 nn_writel(nn, NFP_NET_CFG_RSS_ITBL + i, 858 get_unaligned_le32(nn->rss_itbl + i)); 859 } 860 861 /** 862 * nfp_net_rss_write_key() - Write RSS hash key to device 863 * @nn: NFP Net device to reconfigure 864 */ 865 void nfp_net_rss_write_key(struct nfp_net *nn) 866 { 867 int i; 868 869 for (i = 0; i < nfp_net_rss_key_sz(nn); i += 4) 870 nn_writel(nn, NFP_NET_CFG_RSS_KEY + i, 871 get_unaligned_le32(nn->rss_key + i)); 872 } 873 874 /** 875 * nfp_net_coalesce_write_cfg() - Write irq coalescence configuration to HW 876 * @nn: NFP Net device to reconfigure 877 */ 878 void nfp_net_coalesce_write_cfg(struct nfp_net *nn) 879 { 880 u8 i; 881 u32 factor; 882 u32 value; 883 884 /* Compute factor used to convert coalesce '_usecs' parameters to 885 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp 886 * count. 887 */ 888 factor = nn->tlv_caps.me_freq_mhz / 16; 889 890 /* copy RX interrupt coalesce parameters */ 891 value = (nn->rx_coalesce_max_frames << 16) | 892 (factor * nn->rx_coalesce_usecs); 893 for (i = 0; i < nn->dp.num_rx_rings; i++) 894 nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(i), value); 895 896 /* copy TX interrupt coalesce parameters */ 897 value = (nn->tx_coalesce_max_frames << 16) | 898 (factor * nn->tx_coalesce_usecs); 899 for (i = 0; i < nn->dp.num_tx_rings; i++) 900 nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(i), value); 901 } 902 903 /** 904 * nfp_net_write_mac_addr() - Write mac address to the device control BAR 905 * @nn: NFP Net device to reconfigure 906 * @addr: MAC address to write 907 * 908 * Writes the MAC address from the netdev to the device control BAR. Does not 909 * perform the required reconfig. We do a bit of byte swapping dance because 910 * firmware is LE. 911 */ 912 static void nfp_net_write_mac_addr(struct nfp_net *nn, const u8 *addr) 913 { 914 nn_writel(nn, NFP_NET_CFG_MACADDR + 0, get_unaligned_be32(addr)); 915 nn_writew(nn, NFP_NET_CFG_MACADDR + 6, get_unaligned_be16(addr + 4)); 916 } 917 918 /** 919 * nfp_net_clear_config_and_disable() - Clear control BAR and disable NFP 920 * @nn: NFP Net device to reconfigure 921 * 922 * Warning: must be fully idempotent. 923 */ 924 static void nfp_net_clear_config_and_disable(struct nfp_net *nn) 925 { 926 u32 new_ctrl, update; 927 unsigned int r; 928 int err; 929 930 new_ctrl = nn->dp.ctrl; 931 new_ctrl &= ~NFP_NET_CFG_CTRL_ENABLE; 932 update = NFP_NET_CFG_UPDATE_GEN; 933 update |= NFP_NET_CFG_UPDATE_MSIX; 934 update |= NFP_NET_CFG_UPDATE_RING; 935 936 if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG) 937 new_ctrl &= ~NFP_NET_CFG_CTRL_RINGCFG; 938 939 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0); 940 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0); 941 942 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 943 err = nfp_net_reconfig(nn, update); 944 if (err) 945 nn_err(nn, "Could not disable device: %d\n", err); 946 947 for (r = 0; r < nn->dp.num_rx_rings; r++) { 948 nfp_net_rx_ring_reset(&nn->dp.rx_rings[r]); 949 if (nfp_net_has_xsk_pool_slow(&nn->dp, nn->dp.rx_rings[r].idx)) 950 nfp_net_xsk_rx_bufs_free(&nn->dp.rx_rings[r]); 951 } 952 for (r = 0; r < nn->dp.num_tx_rings; r++) 953 nfp_net_tx_ring_reset(&nn->dp, &nn->dp.tx_rings[r]); 954 for (r = 0; r < nn->dp.num_r_vecs; r++) 955 nfp_net_vec_clear_ring_data(nn, r); 956 957 nn->dp.ctrl = new_ctrl; 958 } 959 960 /** 961 * nfp_net_set_config_and_enable() - Write control BAR and enable NFP 962 * @nn: NFP Net device to reconfigure 963 */ 964 static int nfp_net_set_config_and_enable(struct nfp_net *nn) 965 { 966 u32 bufsz, new_ctrl, update = 0; 967 unsigned int r; 968 int err; 969 970 new_ctrl = nn->dp.ctrl; 971 972 if (nn->dp.ctrl & NFP_NET_CFG_CTRL_RSS_ANY) { 973 nfp_net_rss_write_key(nn); 974 nfp_net_rss_write_itbl(nn); 975 nn_writel(nn, NFP_NET_CFG_RSS_CTRL, nn->rss_cfg); 976 update |= NFP_NET_CFG_UPDATE_RSS; 977 } 978 979 if (nn->dp.ctrl & NFP_NET_CFG_CTRL_IRQMOD) { 980 nfp_net_coalesce_write_cfg(nn); 981 update |= NFP_NET_CFG_UPDATE_IRQMOD; 982 } 983 984 for (r = 0; r < nn->dp.num_tx_rings; r++) 985 nfp_net_tx_ring_hw_cfg_write(nn, &nn->dp.tx_rings[r], r); 986 for (r = 0; r < nn->dp.num_rx_rings; r++) 987 nfp_net_rx_ring_hw_cfg_write(nn, &nn->dp.rx_rings[r], r); 988 989 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 990 U64_MAX >> (64 - nn->dp.num_tx_rings)); 991 992 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 993 U64_MAX >> (64 - nn->dp.num_rx_rings)); 994 995 if (nn->dp.netdev) 996 nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr); 997 998 nn_writel(nn, NFP_NET_CFG_MTU, nn->dp.mtu); 999 1000 bufsz = nn->dp.fl_bufsz - nn->dp.rx_dma_off - NFP_NET_RX_BUF_NON_DATA; 1001 nn_writel(nn, NFP_NET_CFG_FLBUFSZ, bufsz); 1002 1003 /* Enable device */ 1004 new_ctrl |= NFP_NET_CFG_CTRL_ENABLE; 1005 update |= NFP_NET_CFG_UPDATE_GEN; 1006 update |= NFP_NET_CFG_UPDATE_MSIX; 1007 update |= NFP_NET_CFG_UPDATE_RING; 1008 if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG) 1009 new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG; 1010 1011 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1012 err = nfp_net_reconfig(nn, update); 1013 if (err) { 1014 nfp_net_clear_config_and_disable(nn); 1015 return err; 1016 } 1017 1018 nn->dp.ctrl = new_ctrl; 1019 1020 for (r = 0; r < nn->dp.num_rx_rings; r++) 1021 nfp_net_rx_ring_fill_freelist(&nn->dp, &nn->dp.rx_rings[r]); 1022 1023 return 0; 1024 } 1025 1026 /** 1027 * nfp_net_close_stack() - Quiesce the stack (part of close) 1028 * @nn: NFP Net device to reconfigure 1029 */ 1030 static void nfp_net_close_stack(struct nfp_net *nn) 1031 { 1032 struct nfp_net_r_vector *r_vec; 1033 unsigned int r; 1034 1035 disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector); 1036 netif_carrier_off(nn->dp.netdev); 1037 nn->link_up = false; 1038 1039 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1040 r_vec = &nn->r_vecs[r]; 1041 1042 disable_irq(r_vec->irq_vector); 1043 napi_disable(&r_vec->napi); 1044 1045 if (r_vec->rx_ring) 1046 cancel_work_sync(&r_vec->rx_dim.work); 1047 1048 if (r_vec->tx_ring) 1049 cancel_work_sync(&r_vec->tx_dim.work); 1050 } 1051 1052 netif_tx_disable(nn->dp.netdev); 1053 } 1054 1055 /** 1056 * nfp_net_close_free_all() - Free all runtime resources 1057 * @nn: NFP Net device to reconfigure 1058 */ 1059 static void nfp_net_close_free_all(struct nfp_net *nn) 1060 { 1061 unsigned int r; 1062 1063 nfp_net_tx_rings_free(&nn->dp); 1064 nfp_net_rx_rings_free(&nn->dp); 1065 1066 for (r = 0; r < nn->dp.num_r_vecs; r++) 1067 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1068 1069 nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX); 1070 nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX); 1071 } 1072 1073 /** 1074 * nfp_net_netdev_close() - Called when the device is downed 1075 * @netdev: netdev structure 1076 */ 1077 static int nfp_net_netdev_close(struct net_device *netdev) 1078 { 1079 struct nfp_net *nn = netdev_priv(netdev); 1080 1081 /* Step 1: Disable RX and TX rings from the Linux kernel perspective 1082 */ 1083 nfp_net_close_stack(nn); 1084 1085 /* Step 2: Tell NFP 1086 */ 1087 nfp_net_clear_config_and_disable(nn); 1088 nfp_port_configure(netdev, false); 1089 1090 /* Step 3: Free resources 1091 */ 1092 nfp_net_close_free_all(nn); 1093 1094 nn_dbg(nn, "%s down", netdev->name); 1095 return 0; 1096 } 1097 1098 void nfp_ctrl_close(struct nfp_net *nn) 1099 { 1100 int r; 1101 1102 rtnl_lock(); 1103 1104 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1105 disable_irq(nn->r_vecs[r].irq_vector); 1106 tasklet_disable(&nn->r_vecs[r].tasklet); 1107 } 1108 1109 nfp_net_clear_config_and_disable(nn); 1110 1111 nfp_net_close_free_all(nn); 1112 1113 rtnl_unlock(); 1114 } 1115 1116 static void nfp_net_rx_dim_work(struct work_struct *work) 1117 { 1118 struct nfp_net_r_vector *r_vec; 1119 unsigned int factor, value; 1120 struct dim_cq_moder moder; 1121 struct nfp_net *nn; 1122 struct dim *dim; 1123 1124 dim = container_of(work, struct dim, work); 1125 moder = net_dim_get_rx_moderation(dim->mode, dim->profile_ix); 1126 r_vec = container_of(dim, struct nfp_net_r_vector, rx_dim); 1127 nn = r_vec->nfp_net; 1128 1129 /* Compute factor used to convert coalesce '_usecs' parameters to 1130 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp 1131 * count. 1132 */ 1133 factor = nn->tlv_caps.me_freq_mhz / 16; 1134 if (nfp_net_coalesce_para_check(factor * moder.usec, moder.pkts)) 1135 return; 1136 1137 /* copy RX interrupt coalesce parameters */ 1138 value = (moder.pkts << 16) | (factor * moder.usec); 1139 nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(r_vec->rx_ring->idx), value); 1140 (void)nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_IRQMOD); 1141 1142 dim->state = DIM_START_MEASURE; 1143 } 1144 1145 static void nfp_net_tx_dim_work(struct work_struct *work) 1146 { 1147 struct nfp_net_r_vector *r_vec; 1148 unsigned int factor, value; 1149 struct dim_cq_moder moder; 1150 struct nfp_net *nn; 1151 struct dim *dim; 1152 1153 dim = container_of(work, struct dim, work); 1154 moder = net_dim_get_tx_moderation(dim->mode, dim->profile_ix); 1155 r_vec = container_of(dim, struct nfp_net_r_vector, tx_dim); 1156 nn = r_vec->nfp_net; 1157 1158 /* Compute factor used to convert coalesce '_usecs' parameters to 1159 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp 1160 * count. 1161 */ 1162 factor = nn->tlv_caps.me_freq_mhz / 16; 1163 if (nfp_net_coalesce_para_check(factor * moder.usec, moder.pkts)) 1164 return; 1165 1166 /* copy TX interrupt coalesce parameters */ 1167 value = (moder.pkts << 16) | (factor * moder.usec); 1168 nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(r_vec->tx_ring->idx), value); 1169 (void)nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_IRQMOD); 1170 1171 dim->state = DIM_START_MEASURE; 1172 } 1173 1174 /** 1175 * nfp_net_open_stack() - Start the device from stack's perspective 1176 * @nn: NFP Net device to reconfigure 1177 */ 1178 static void nfp_net_open_stack(struct nfp_net *nn) 1179 { 1180 struct nfp_net_r_vector *r_vec; 1181 unsigned int r; 1182 1183 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1184 r_vec = &nn->r_vecs[r]; 1185 1186 if (r_vec->rx_ring) { 1187 INIT_WORK(&r_vec->rx_dim.work, nfp_net_rx_dim_work); 1188 r_vec->rx_dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE; 1189 } 1190 1191 if (r_vec->tx_ring) { 1192 INIT_WORK(&r_vec->tx_dim.work, nfp_net_tx_dim_work); 1193 r_vec->tx_dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE; 1194 } 1195 1196 napi_enable(&r_vec->napi); 1197 enable_irq(r_vec->irq_vector); 1198 } 1199 1200 netif_tx_wake_all_queues(nn->dp.netdev); 1201 1202 enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector); 1203 nfp_net_read_link_status(nn); 1204 } 1205 1206 static int nfp_net_open_alloc_all(struct nfp_net *nn) 1207 { 1208 int err, r; 1209 1210 err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn", 1211 nn->exn_name, sizeof(nn->exn_name), 1212 NFP_NET_IRQ_EXN_IDX, nn->exn_handler); 1213 if (err) 1214 return err; 1215 err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc", 1216 nn->lsc_name, sizeof(nn->lsc_name), 1217 NFP_NET_IRQ_LSC_IDX, nn->lsc_handler); 1218 if (err) 1219 goto err_free_exn; 1220 disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector); 1221 1222 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1223 err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r); 1224 if (err) 1225 goto err_cleanup_vec_p; 1226 } 1227 1228 err = nfp_net_rx_rings_prepare(nn, &nn->dp); 1229 if (err) 1230 goto err_cleanup_vec; 1231 1232 err = nfp_net_tx_rings_prepare(nn, &nn->dp); 1233 if (err) 1234 goto err_free_rx_rings; 1235 1236 for (r = 0; r < nn->max_r_vecs; r++) 1237 nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r); 1238 1239 return 0; 1240 1241 err_free_rx_rings: 1242 nfp_net_rx_rings_free(&nn->dp); 1243 err_cleanup_vec: 1244 r = nn->dp.num_r_vecs; 1245 err_cleanup_vec_p: 1246 while (r--) 1247 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1248 nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX); 1249 err_free_exn: 1250 nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX); 1251 return err; 1252 } 1253 1254 static int nfp_net_netdev_open(struct net_device *netdev) 1255 { 1256 struct nfp_net *nn = netdev_priv(netdev); 1257 int err; 1258 1259 /* Step 1: Allocate resources for rings and the like 1260 * - Request interrupts 1261 * - Allocate RX and TX ring resources 1262 * - Setup initial RSS table 1263 */ 1264 err = nfp_net_open_alloc_all(nn); 1265 if (err) 1266 return err; 1267 1268 err = netif_set_real_num_tx_queues(netdev, nn->dp.num_stack_tx_rings); 1269 if (err) 1270 goto err_free_all; 1271 1272 err = netif_set_real_num_rx_queues(netdev, nn->dp.num_rx_rings); 1273 if (err) 1274 goto err_free_all; 1275 1276 /* Step 2: Configure the NFP 1277 * - Ifup the physical interface if it exists 1278 * - Enable rings from 0 to tx_rings/rx_rings - 1. 1279 * - Write MAC address (in case it changed) 1280 * - Set the MTU 1281 * - Set the Freelist buffer size 1282 * - Enable the FW 1283 */ 1284 err = nfp_port_configure(netdev, true); 1285 if (err) 1286 goto err_free_all; 1287 1288 err = nfp_net_set_config_and_enable(nn); 1289 if (err) 1290 goto err_port_disable; 1291 1292 /* Step 3: Enable for kernel 1293 * - put some freelist descriptors on each RX ring 1294 * - enable NAPI on each ring 1295 * - enable all TX queues 1296 * - set link state 1297 */ 1298 nfp_net_open_stack(nn); 1299 1300 return 0; 1301 1302 err_port_disable: 1303 nfp_port_configure(netdev, false); 1304 err_free_all: 1305 nfp_net_close_free_all(nn); 1306 return err; 1307 } 1308 1309 int nfp_ctrl_open(struct nfp_net *nn) 1310 { 1311 int err, r; 1312 1313 /* ring dumping depends on vNICs being opened/closed under rtnl */ 1314 rtnl_lock(); 1315 1316 err = nfp_net_open_alloc_all(nn); 1317 if (err) 1318 goto err_unlock; 1319 1320 err = nfp_net_set_config_and_enable(nn); 1321 if (err) 1322 goto err_free_all; 1323 1324 for (r = 0; r < nn->dp.num_r_vecs; r++) 1325 enable_irq(nn->r_vecs[r].irq_vector); 1326 1327 rtnl_unlock(); 1328 1329 return 0; 1330 1331 err_free_all: 1332 nfp_net_close_free_all(nn); 1333 err_unlock: 1334 rtnl_unlock(); 1335 return err; 1336 } 1337 1338 static void nfp_net_set_rx_mode(struct net_device *netdev) 1339 { 1340 struct nfp_net *nn = netdev_priv(netdev); 1341 u32 new_ctrl; 1342 1343 new_ctrl = nn->dp.ctrl; 1344 1345 if (!netdev_mc_empty(netdev) || netdev->flags & IFF_ALLMULTI) 1346 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_L2MC; 1347 else 1348 new_ctrl &= ~NFP_NET_CFG_CTRL_L2MC; 1349 1350 if (netdev->flags & IFF_PROMISC) { 1351 if (nn->cap & NFP_NET_CFG_CTRL_PROMISC) 1352 new_ctrl |= NFP_NET_CFG_CTRL_PROMISC; 1353 else 1354 nn_warn(nn, "FW does not support promiscuous mode\n"); 1355 } else { 1356 new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC; 1357 } 1358 1359 if (new_ctrl == nn->dp.ctrl) 1360 return; 1361 1362 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1363 nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN); 1364 1365 nn->dp.ctrl = new_ctrl; 1366 } 1367 1368 static void nfp_net_rss_init_itbl(struct nfp_net *nn) 1369 { 1370 int i; 1371 1372 for (i = 0; i < sizeof(nn->rss_itbl); i++) 1373 nn->rss_itbl[i] = 1374 ethtool_rxfh_indir_default(i, nn->dp.num_rx_rings); 1375 } 1376 1377 static void nfp_net_dp_swap(struct nfp_net *nn, struct nfp_net_dp *dp) 1378 { 1379 struct nfp_net_dp new_dp = *dp; 1380 1381 *dp = nn->dp; 1382 nn->dp = new_dp; 1383 1384 nn->dp.netdev->mtu = new_dp.mtu; 1385 1386 if (!netif_is_rxfh_configured(nn->dp.netdev)) 1387 nfp_net_rss_init_itbl(nn); 1388 } 1389 1390 static int nfp_net_dp_swap_enable(struct nfp_net *nn, struct nfp_net_dp *dp) 1391 { 1392 unsigned int r; 1393 int err; 1394 1395 nfp_net_dp_swap(nn, dp); 1396 1397 for (r = 0; r < nn->max_r_vecs; r++) 1398 nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r); 1399 1400 err = netif_set_real_num_queues(nn->dp.netdev, 1401 nn->dp.num_stack_tx_rings, 1402 nn->dp.num_rx_rings); 1403 if (err) 1404 return err; 1405 1406 return nfp_net_set_config_and_enable(nn); 1407 } 1408 1409 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn) 1410 { 1411 struct nfp_net_dp *new; 1412 1413 new = kmalloc(sizeof(*new), GFP_KERNEL); 1414 if (!new) 1415 return NULL; 1416 1417 *new = nn->dp; 1418 1419 new->xsk_pools = kmemdup(new->xsk_pools, 1420 array_size(nn->max_r_vecs, 1421 sizeof(new->xsk_pools)), 1422 GFP_KERNEL); 1423 if (!new->xsk_pools) { 1424 kfree(new); 1425 return NULL; 1426 } 1427 1428 /* Clear things which need to be recomputed */ 1429 new->fl_bufsz = 0; 1430 new->tx_rings = NULL; 1431 new->rx_rings = NULL; 1432 new->num_r_vecs = 0; 1433 new->num_stack_tx_rings = 0; 1434 new->txrwb = NULL; 1435 new->txrwb_dma = 0; 1436 1437 return new; 1438 } 1439 1440 static void nfp_net_free_dp(struct nfp_net_dp *dp) 1441 { 1442 kfree(dp->xsk_pools); 1443 kfree(dp); 1444 } 1445 1446 static int 1447 nfp_net_check_config(struct nfp_net *nn, struct nfp_net_dp *dp, 1448 struct netlink_ext_ack *extack) 1449 { 1450 unsigned int r, xsk_min_fl_bufsz; 1451 1452 /* XDP-enabled tests */ 1453 if (!dp->xdp_prog) 1454 return 0; 1455 if (dp->fl_bufsz > PAGE_SIZE) { 1456 NL_SET_ERR_MSG_MOD(extack, "MTU too large w/ XDP enabled"); 1457 return -EINVAL; 1458 } 1459 if (dp->num_tx_rings > nn->max_tx_rings) { 1460 NL_SET_ERR_MSG_MOD(extack, "Insufficient number of TX rings w/ XDP enabled"); 1461 return -EINVAL; 1462 } 1463 1464 xsk_min_fl_bufsz = nfp_net_calc_fl_bufsz_xsk(dp); 1465 for (r = 0; r < nn->max_r_vecs; r++) { 1466 if (!dp->xsk_pools[r]) 1467 continue; 1468 1469 if (xsk_pool_get_rx_frame_size(dp->xsk_pools[r]) < xsk_min_fl_bufsz) { 1470 NL_SET_ERR_MSG_MOD(extack, 1471 "XSK buffer pool chunk size too small"); 1472 return -EINVAL; 1473 } 1474 } 1475 1476 return 0; 1477 } 1478 1479 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *dp, 1480 struct netlink_ext_ack *extack) 1481 { 1482 int r, err; 1483 1484 dp->fl_bufsz = nfp_net_calc_fl_bufsz(dp); 1485 1486 dp->num_stack_tx_rings = dp->num_tx_rings; 1487 if (dp->xdp_prog) 1488 dp->num_stack_tx_rings -= dp->num_rx_rings; 1489 1490 dp->num_r_vecs = max(dp->num_rx_rings, dp->num_stack_tx_rings); 1491 1492 err = nfp_net_check_config(nn, dp, extack); 1493 if (err) 1494 goto exit_free_dp; 1495 1496 if (!netif_running(dp->netdev)) { 1497 nfp_net_dp_swap(nn, dp); 1498 err = 0; 1499 goto exit_free_dp; 1500 } 1501 1502 /* Prepare new rings */ 1503 for (r = nn->dp.num_r_vecs; r < dp->num_r_vecs; r++) { 1504 err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r); 1505 if (err) { 1506 dp->num_r_vecs = r; 1507 goto err_cleanup_vecs; 1508 } 1509 } 1510 1511 err = nfp_net_rx_rings_prepare(nn, dp); 1512 if (err) 1513 goto err_cleanup_vecs; 1514 1515 err = nfp_net_tx_rings_prepare(nn, dp); 1516 if (err) 1517 goto err_free_rx; 1518 1519 /* Stop device, swap in new rings, try to start the firmware */ 1520 nfp_net_close_stack(nn); 1521 nfp_net_clear_config_and_disable(nn); 1522 1523 err = nfp_net_dp_swap_enable(nn, dp); 1524 if (err) { 1525 int err2; 1526 1527 nfp_net_clear_config_and_disable(nn); 1528 1529 /* Try with old configuration and old rings */ 1530 err2 = nfp_net_dp_swap_enable(nn, dp); 1531 if (err2) 1532 nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n", 1533 err, err2); 1534 } 1535 for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--) 1536 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1537 1538 nfp_net_rx_rings_free(dp); 1539 nfp_net_tx_rings_free(dp); 1540 1541 nfp_net_open_stack(nn); 1542 exit_free_dp: 1543 nfp_net_free_dp(dp); 1544 1545 return err; 1546 1547 err_free_rx: 1548 nfp_net_rx_rings_free(dp); 1549 err_cleanup_vecs: 1550 for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--) 1551 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1552 nfp_net_free_dp(dp); 1553 return err; 1554 } 1555 1556 static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu) 1557 { 1558 struct nfp_net *nn = netdev_priv(netdev); 1559 struct nfp_net_dp *dp; 1560 int err; 1561 1562 err = nfp_app_check_mtu(nn->app, netdev, new_mtu); 1563 if (err) 1564 return err; 1565 1566 dp = nfp_net_clone_dp(nn); 1567 if (!dp) 1568 return -ENOMEM; 1569 1570 dp->mtu = new_mtu; 1571 1572 return nfp_net_ring_reconfig(nn, dp, NULL); 1573 } 1574 1575 static int 1576 nfp_net_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid) 1577 { 1578 const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_ADD; 1579 struct nfp_net *nn = netdev_priv(netdev); 1580 int err; 1581 1582 /* Priority tagged packets with vlan id 0 are processed by the 1583 * NFP as untagged packets 1584 */ 1585 if (!vid) 1586 return 0; 1587 1588 err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ); 1589 if (err) 1590 return err; 1591 1592 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid); 1593 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO, 1594 ETH_P_8021Q); 1595 1596 return nfp_net_mbox_reconfig_and_unlock(nn, cmd); 1597 } 1598 1599 static int 1600 nfp_net_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid) 1601 { 1602 const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_KILL; 1603 struct nfp_net *nn = netdev_priv(netdev); 1604 int err; 1605 1606 /* Priority tagged packets with vlan id 0 are processed by the 1607 * NFP as untagged packets 1608 */ 1609 if (!vid) 1610 return 0; 1611 1612 err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ); 1613 if (err) 1614 return err; 1615 1616 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid); 1617 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO, 1618 ETH_P_8021Q); 1619 1620 return nfp_net_mbox_reconfig_and_unlock(nn, cmd); 1621 } 1622 1623 static void nfp_net_stat64(struct net_device *netdev, 1624 struct rtnl_link_stats64 *stats) 1625 { 1626 struct nfp_net *nn = netdev_priv(netdev); 1627 int r; 1628 1629 /* Collect software stats */ 1630 for (r = 0; r < nn->max_r_vecs; r++) { 1631 struct nfp_net_r_vector *r_vec = &nn->r_vecs[r]; 1632 u64 data[3]; 1633 unsigned int start; 1634 1635 do { 1636 start = u64_stats_fetch_begin_irq(&r_vec->rx_sync); 1637 data[0] = r_vec->rx_pkts; 1638 data[1] = r_vec->rx_bytes; 1639 data[2] = r_vec->rx_drops; 1640 } while (u64_stats_fetch_retry_irq(&r_vec->rx_sync, start)); 1641 stats->rx_packets += data[0]; 1642 stats->rx_bytes += data[1]; 1643 stats->rx_dropped += data[2]; 1644 1645 do { 1646 start = u64_stats_fetch_begin_irq(&r_vec->tx_sync); 1647 data[0] = r_vec->tx_pkts; 1648 data[1] = r_vec->tx_bytes; 1649 data[2] = r_vec->tx_errors; 1650 } while (u64_stats_fetch_retry_irq(&r_vec->tx_sync, start)); 1651 stats->tx_packets += data[0]; 1652 stats->tx_bytes += data[1]; 1653 stats->tx_errors += data[2]; 1654 } 1655 1656 /* Add in device stats */ 1657 stats->multicast += nn_readq(nn, NFP_NET_CFG_STATS_RX_MC_FRAMES); 1658 stats->rx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_RX_DISCARDS); 1659 stats->rx_errors += nn_readq(nn, NFP_NET_CFG_STATS_RX_ERRORS); 1660 1661 stats->tx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_TX_DISCARDS); 1662 stats->tx_errors += nn_readq(nn, NFP_NET_CFG_STATS_TX_ERRORS); 1663 } 1664 1665 static int nfp_net_set_features(struct net_device *netdev, 1666 netdev_features_t features) 1667 { 1668 netdev_features_t changed = netdev->features ^ features; 1669 struct nfp_net *nn = netdev_priv(netdev); 1670 u32 new_ctrl; 1671 int err; 1672 1673 /* Assume this is not called with features we have not advertised */ 1674 1675 new_ctrl = nn->dp.ctrl; 1676 1677 if (changed & NETIF_F_RXCSUM) { 1678 if (features & NETIF_F_RXCSUM) 1679 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY; 1680 else 1681 new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM_ANY; 1682 } 1683 1684 if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 1685 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) 1686 new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM; 1687 else 1688 new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM; 1689 } 1690 1691 if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) { 1692 if (features & (NETIF_F_TSO | NETIF_F_TSO6)) 1693 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?: 1694 NFP_NET_CFG_CTRL_LSO; 1695 else 1696 new_ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY; 1697 } 1698 1699 if (changed & NETIF_F_HW_VLAN_CTAG_RX) { 1700 if (features & NETIF_F_HW_VLAN_CTAG_RX) 1701 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ?: 1702 NFP_NET_CFG_CTRL_RXVLAN; 1703 else 1704 new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN_ANY; 1705 } 1706 1707 if (changed & NETIF_F_HW_VLAN_CTAG_TX) { 1708 if (features & NETIF_F_HW_VLAN_CTAG_TX) 1709 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ?: 1710 NFP_NET_CFG_CTRL_TXVLAN; 1711 else 1712 new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN_ANY; 1713 } 1714 1715 if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) { 1716 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) 1717 new_ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER; 1718 else 1719 new_ctrl &= ~NFP_NET_CFG_CTRL_CTAG_FILTER; 1720 } 1721 1722 if (changed & NETIF_F_HW_VLAN_STAG_RX) { 1723 if (features & NETIF_F_HW_VLAN_STAG_RX) 1724 new_ctrl |= NFP_NET_CFG_CTRL_RXQINQ; 1725 else 1726 new_ctrl &= ~NFP_NET_CFG_CTRL_RXQINQ; 1727 } 1728 1729 if (changed & NETIF_F_SG) { 1730 if (features & NETIF_F_SG) 1731 new_ctrl |= NFP_NET_CFG_CTRL_GATHER; 1732 else 1733 new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER; 1734 } 1735 1736 err = nfp_port_set_features(netdev, features); 1737 if (err) 1738 return err; 1739 1740 nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n", 1741 netdev->features, features, changed); 1742 1743 if (new_ctrl == nn->dp.ctrl) 1744 return 0; 1745 1746 nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->dp.ctrl, new_ctrl); 1747 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1748 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN); 1749 if (err) 1750 return err; 1751 1752 nn->dp.ctrl = new_ctrl; 1753 1754 return 0; 1755 } 1756 1757 static netdev_features_t 1758 nfp_net_fix_features(struct net_device *netdev, 1759 netdev_features_t features) 1760 { 1761 if ((features & NETIF_F_HW_VLAN_CTAG_RX) && 1762 (features & NETIF_F_HW_VLAN_STAG_RX)) { 1763 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) { 1764 features &= ~NETIF_F_HW_VLAN_CTAG_RX; 1765 netdev->wanted_features &= ~NETIF_F_HW_VLAN_CTAG_RX; 1766 netdev_warn(netdev, 1767 "S-tag and C-tag stripping can't be enabled at the same time. Enabling S-tag stripping and disabling C-tag stripping\n"); 1768 } else if (netdev->features & NETIF_F_HW_VLAN_STAG_RX) { 1769 features &= ~NETIF_F_HW_VLAN_STAG_RX; 1770 netdev->wanted_features &= ~NETIF_F_HW_VLAN_STAG_RX; 1771 netdev_warn(netdev, 1772 "S-tag and C-tag stripping can't be enabled at the same time. Enabling C-tag stripping and disabling S-tag stripping\n"); 1773 } 1774 } 1775 return features; 1776 } 1777 1778 static netdev_features_t 1779 nfp_net_features_check(struct sk_buff *skb, struct net_device *dev, 1780 netdev_features_t features) 1781 { 1782 u8 l4_hdr; 1783 1784 /* We can't do TSO over double tagged packets (802.1AD) */ 1785 features &= vlan_features_check(skb, features); 1786 1787 if (!skb->encapsulation) 1788 return features; 1789 1790 /* Ensure that inner L4 header offset fits into TX descriptor field */ 1791 if (skb_is_gso(skb)) { 1792 u32 hdrlen; 1793 1794 hdrlen = skb_inner_tcp_all_headers(skb); 1795 1796 /* Assume worst case scenario of having longest possible 1797 * metadata prepend - 8B 1798 */ 1799 if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ - 8)) 1800 features &= ~NETIF_F_GSO_MASK; 1801 } 1802 1803 /* VXLAN/GRE check */ 1804 switch (vlan_get_protocol(skb)) { 1805 case htons(ETH_P_IP): 1806 l4_hdr = ip_hdr(skb)->protocol; 1807 break; 1808 case htons(ETH_P_IPV6): 1809 l4_hdr = ipv6_hdr(skb)->nexthdr; 1810 break; 1811 default: 1812 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 1813 } 1814 1815 if (skb->inner_protocol_type != ENCAP_TYPE_ETHER || 1816 skb->inner_protocol != htons(ETH_P_TEB) || 1817 (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) || 1818 (l4_hdr == IPPROTO_UDP && 1819 (skb_inner_mac_header(skb) - skb_transport_header(skb) != 1820 sizeof(struct udphdr) + sizeof(struct vxlanhdr)))) 1821 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 1822 1823 return features; 1824 } 1825 1826 static int 1827 nfp_net_get_phys_port_name(struct net_device *netdev, char *name, size_t len) 1828 { 1829 struct nfp_net *nn = netdev_priv(netdev); 1830 int n; 1831 1832 /* If port is defined, devlink_port is registered and devlink core 1833 * is taking care of name formatting. 1834 */ 1835 if (nn->port) 1836 return -EOPNOTSUPP; 1837 1838 if (nn->dp.is_vf || nn->vnic_no_name) 1839 return -EOPNOTSUPP; 1840 1841 n = snprintf(name, len, "n%d", nn->id); 1842 if (n >= len) 1843 return -EINVAL; 1844 1845 return 0; 1846 } 1847 1848 static int nfp_net_xdp_setup_drv(struct nfp_net *nn, struct netdev_bpf *bpf) 1849 { 1850 struct bpf_prog *prog = bpf->prog; 1851 struct nfp_net_dp *dp; 1852 int err; 1853 1854 if (!prog == !nn->dp.xdp_prog) { 1855 WRITE_ONCE(nn->dp.xdp_prog, prog); 1856 xdp_attachment_setup(&nn->xdp, bpf); 1857 return 0; 1858 } 1859 1860 dp = nfp_net_clone_dp(nn); 1861 if (!dp) 1862 return -ENOMEM; 1863 1864 dp->xdp_prog = prog; 1865 dp->num_tx_rings += prog ? nn->dp.num_rx_rings : -nn->dp.num_rx_rings; 1866 dp->rx_dma_dir = prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE; 1867 dp->rx_dma_off = prog ? XDP_PACKET_HEADROOM - nn->dp.rx_offset : 0; 1868 1869 /* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */ 1870 err = nfp_net_ring_reconfig(nn, dp, bpf->extack); 1871 if (err) 1872 return err; 1873 1874 xdp_attachment_setup(&nn->xdp, bpf); 1875 return 0; 1876 } 1877 1878 static int nfp_net_xdp_setup_hw(struct nfp_net *nn, struct netdev_bpf *bpf) 1879 { 1880 int err; 1881 1882 err = nfp_app_xdp_offload(nn->app, nn, bpf->prog, bpf->extack); 1883 if (err) 1884 return err; 1885 1886 xdp_attachment_setup(&nn->xdp_hw, bpf); 1887 return 0; 1888 } 1889 1890 static int nfp_net_xdp(struct net_device *netdev, struct netdev_bpf *xdp) 1891 { 1892 struct nfp_net *nn = netdev_priv(netdev); 1893 1894 switch (xdp->command) { 1895 case XDP_SETUP_PROG: 1896 return nfp_net_xdp_setup_drv(nn, xdp); 1897 case XDP_SETUP_PROG_HW: 1898 return nfp_net_xdp_setup_hw(nn, xdp); 1899 case XDP_SETUP_XSK_POOL: 1900 return nfp_net_xsk_setup_pool(netdev, xdp->xsk.pool, 1901 xdp->xsk.queue_id); 1902 default: 1903 return nfp_app_bpf(nn->app, nn, xdp); 1904 } 1905 } 1906 1907 static int nfp_net_set_mac_address(struct net_device *netdev, void *addr) 1908 { 1909 struct nfp_net *nn = netdev_priv(netdev); 1910 struct sockaddr *saddr = addr; 1911 int err; 1912 1913 err = eth_prepare_mac_addr_change(netdev, addr); 1914 if (err) 1915 return err; 1916 1917 nfp_net_write_mac_addr(nn, saddr->sa_data); 1918 1919 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MACADDR); 1920 if (err) 1921 return err; 1922 1923 eth_commit_mac_addr_change(netdev, addr); 1924 1925 return 0; 1926 } 1927 1928 static int nfp_net_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq, 1929 struct net_device *dev, u32 filter_mask, 1930 int nlflags) 1931 { 1932 struct nfp_net *nn = netdev_priv(dev); 1933 u16 mode; 1934 1935 if (!(nn->cap & NFP_NET_CFG_CTRL_VEPA)) 1936 return -EOPNOTSUPP; 1937 1938 mode = (nn->dp.ctrl & NFP_NET_CFG_CTRL_VEPA) ? 1939 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB; 1940 1941 return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode, 0, 0, 1942 nlflags, filter_mask, NULL); 1943 } 1944 1945 static int nfp_net_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh, 1946 u16 flags, struct netlink_ext_ack *extack) 1947 { 1948 struct nfp_net *nn = netdev_priv(dev); 1949 struct nlattr *attr, *br_spec; 1950 int rem, err; 1951 u32 new_ctrl; 1952 u16 mode; 1953 1954 if (!(nn->cap & NFP_NET_CFG_CTRL_VEPA)) 1955 return -EOPNOTSUPP; 1956 1957 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC); 1958 if (!br_spec) 1959 return -EINVAL; 1960 1961 nla_for_each_nested(attr, br_spec, rem) { 1962 if (nla_type(attr) != IFLA_BRIDGE_MODE) 1963 continue; 1964 1965 if (nla_len(attr) < sizeof(mode)) 1966 return -EINVAL; 1967 1968 new_ctrl = nn->dp.ctrl; 1969 mode = nla_get_u16(attr); 1970 if (mode == BRIDGE_MODE_VEPA) 1971 new_ctrl |= NFP_NET_CFG_CTRL_VEPA; 1972 else if (mode == BRIDGE_MODE_VEB) 1973 new_ctrl &= ~NFP_NET_CFG_CTRL_VEPA; 1974 else 1975 return -EOPNOTSUPP; 1976 1977 if (new_ctrl == nn->dp.ctrl) 1978 return 0; 1979 1980 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1981 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN); 1982 if (!err) 1983 nn->dp.ctrl = new_ctrl; 1984 1985 return err; 1986 } 1987 1988 return -EINVAL; 1989 } 1990 1991 const struct net_device_ops nfp_nfd3_netdev_ops = { 1992 .ndo_init = nfp_app_ndo_init, 1993 .ndo_uninit = nfp_app_ndo_uninit, 1994 .ndo_open = nfp_net_netdev_open, 1995 .ndo_stop = nfp_net_netdev_close, 1996 .ndo_start_xmit = nfp_net_tx, 1997 .ndo_get_stats64 = nfp_net_stat64, 1998 .ndo_vlan_rx_add_vid = nfp_net_vlan_rx_add_vid, 1999 .ndo_vlan_rx_kill_vid = nfp_net_vlan_rx_kill_vid, 2000 .ndo_set_vf_mac = nfp_app_set_vf_mac, 2001 .ndo_set_vf_vlan = nfp_app_set_vf_vlan, 2002 .ndo_set_vf_rate = nfp_app_set_vf_rate, 2003 .ndo_set_vf_spoofchk = nfp_app_set_vf_spoofchk, 2004 .ndo_set_vf_trust = nfp_app_set_vf_trust, 2005 .ndo_get_vf_config = nfp_app_get_vf_config, 2006 .ndo_set_vf_link_state = nfp_app_set_vf_link_state, 2007 .ndo_setup_tc = nfp_port_setup_tc, 2008 .ndo_tx_timeout = nfp_net_tx_timeout, 2009 .ndo_set_rx_mode = nfp_net_set_rx_mode, 2010 .ndo_change_mtu = nfp_net_change_mtu, 2011 .ndo_set_mac_address = nfp_net_set_mac_address, 2012 .ndo_set_features = nfp_net_set_features, 2013 .ndo_fix_features = nfp_net_fix_features, 2014 .ndo_features_check = nfp_net_features_check, 2015 .ndo_get_phys_port_name = nfp_net_get_phys_port_name, 2016 .ndo_bpf = nfp_net_xdp, 2017 .ndo_xsk_wakeup = nfp_net_xsk_wakeup, 2018 .ndo_get_devlink_port = nfp_devlink_get_devlink_port, 2019 .ndo_bridge_getlink = nfp_net_bridge_getlink, 2020 .ndo_bridge_setlink = nfp_net_bridge_setlink, 2021 }; 2022 2023 const struct net_device_ops nfp_nfdk_netdev_ops = { 2024 .ndo_init = nfp_app_ndo_init, 2025 .ndo_uninit = nfp_app_ndo_uninit, 2026 .ndo_open = nfp_net_netdev_open, 2027 .ndo_stop = nfp_net_netdev_close, 2028 .ndo_start_xmit = nfp_net_tx, 2029 .ndo_get_stats64 = nfp_net_stat64, 2030 .ndo_vlan_rx_add_vid = nfp_net_vlan_rx_add_vid, 2031 .ndo_vlan_rx_kill_vid = nfp_net_vlan_rx_kill_vid, 2032 .ndo_set_vf_mac = nfp_app_set_vf_mac, 2033 .ndo_set_vf_vlan = nfp_app_set_vf_vlan, 2034 .ndo_set_vf_rate = nfp_app_set_vf_rate, 2035 .ndo_set_vf_spoofchk = nfp_app_set_vf_spoofchk, 2036 .ndo_set_vf_trust = nfp_app_set_vf_trust, 2037 .ndo_get_vf_config = nfp_app_get_vf_config, 2038 .ndo_set_vf_link_state = nfp_app_set_vf_link_state, 2039 .ndo_setup_tc = nfp_port_setup_tc, 2040 .ndo_tx_timeout = nfp_net_tx_timeout, 2041 .ndo_set_rx_mode = nfp_net_set_rx_mode, 2042 .ndo_change_mtu = nfp_net_change_mtu, 2043 .ndo_set_mac_address = nfp_net_set_mac_address, 2044 .ndo_set_features = nfp_net_set_features, 2045 .ndo_fix_features = nfp_net_fix_features, 2046 .ndo_features_check = nfp_net_features_check, 2047 .ndo_get_phys_port_name = nfp_net_get_phys_port_name, 2048 .ndo_bpf = nfp_net_xdp, 2049 .ndo_get_devlink_port = nfp_devlink_get_devlink_port, 2050 .ndo_bridge_getlink = nfp_net_bridge_getlink, 2051 .ndo_bridge_setlink = nfp_net_bridge_setlink, 2052 }; 2053 2054 static int nfp_udp_tunnel_sync(struct net_device *netdev, unsigned int table) 2055 { 2056 struct nfp_net *nn = netdev_priv(netdev); 2057 int i; 2058 2059 BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1); 2060 for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2) { 2061 struct udp_tunnel_info ti0, ti1; 2062 2063 udp_tunnel_nic_get_port(netdev, table, i, &ti0); 2064 udp_tunnel_nic_get_port(netdev, table, i + 1, &ti1); 2065 2066 nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(ti0.port), 2067 be16_to_cpu(ti1.port) << 16 | be16_to_cpu(ti0.port)); 2068 } 2069 2070 return nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_VXLAN); 2071 } 2072 2073 static const struct udp_tunnel_nic_info nfp_udp_tunnels = { 2074 .sync_table = nfp_udp_tunnel_sync, 2075 .flags = UDP_TUNNEL_NIC_INFO_MAY_SLEEP | 2076 UDP_TUNNEL_NIC_INFO_OPEN_ONLY, 2077 .tables = { 2078 { 2079 .n_entries = NFP_NET_N_VXLAN_PORTS, 2080 .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, 2081 }, 2082 }, 2083 }; 2084 2085 /** 2086 * nfp_net_info() - Print general info about the NIC 2087 * @nn: NFP Net device to reconfigure 2088 */ 2089 void nfp_net_info(struct nfp_net *nn) 2090 { 2091 nn_info(nn, "NFP-6xxx %sNetdev: TxQs=%d/%d RxQs=%d/%d\n", 2092 nn->dp.is_vf ? "VF " : "", 2093 nn->dp.num_tx_rings, nn->max_tx_rings, 2094 nn->dp.num_rx_rings, nn->max_rx_rings); 2095 nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n", 2096 nn->fw_ver.extend, nn->fw_ver.class, 2097 nn->fw_ver.major, nn->fw_ver.minor, 2098 nn->max_mtu); 2099 nn_info(nn, "CAP: %#x %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n", 2100 nn->cap, 2101 nn->cap & NFP_NET_CFG_CTRL_PROMISC ? "PROMISC " : "", 2102 nn->cap & NFP_NET_CFG_CTRL_L2BC ? "L2BCFILT " : "", 2103 nn->cap & NFP_NET_CFG_CTRL_L2MC ? "L2MCFILT " : "", 2104 nn->cap & NFP_NET_CFG_CTRL_RXCSUM ? "RXCSUM " : "", 2105 nn->cap & NFP_NET_CFG_CTRL_TXCSUM ? "TXCSUM " : "", 2106 nn->cap & NFP_NET_CFG_CTRL_RXVLAN ? "RXVLAN " : "", 2107 nn->cap & NFP_NET_CFG_CTRL_TXVLAN ? "TXVLAN " : "", 2108 nn->cap & NFP_NET_CFG_CTRL_RXQINQ ? "RXQINQ " : "", 2109 nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ? "RXVLANv2 " : "", 2110 nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ? "TXVLANv2 " : "", 2111 nn->cap & NFP_NET_CFG_CTRL_SCATTER ? "SCATTER " : "", 2112 nn->cap & NFP_NET_CFG_CTRL_GATHER ? "GATHER " : "", 2113 nn->cap & NFP_NET_CFG_CTRL_LSO ? "TSO1 " : "", 2114 nn->cap & NFP_NET_CFG_CTRL_LSO2 ? "TSO2 " : "", 2115 nn->cap & NFP_NET_CFG_CTRL_RSS ? "RSS1 " : "", 2116 nn->cap & NFP_NET_CFG_CTRL_RSS2 ? "RSS2 " : "", 2117 nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER ? "CTAG_FILTER " : "", 2118 nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "", 2119 nn->cap & NFP_NET_CFG_CTRL_IRQMOD ? "IRQMOD " : "", 2120 nn->cap & NFP_NET_CFG_CTRL_TXRWB ? "TXRWB " : "", 2121 nn->cap & NFP_NET_CFG_CTRL_VEPA ? "VEPA " : "", 2122 nn->cap & NFP_NET_CFG_CTRL_VXLAN ? "VXLAN " : "", 2123 nn->cap & NFP_NET_CFG_CTRL_NVGRE ? "NVGRE " : "", 2124 nn->cap & NFP_NET_CFG_CTRL_CSUM_COMPLETE ? 2125 "RXCSUM_COMPLETE " : "", 2126 nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR ? "LIVE_ADDR " : "", 2127 nfp_app_extra_cap(nn->app, nn)); 2128 } 2129 2130 /** 2131 * nfp_net_alloc() - Allocate netdev and related structure 2132 * @pdev: PCI device 2133 * @dev_info: NFP ASIC params 2134 * @ctrl_bar: PCI IOMEM with vNIC config memory 2135 * @needs_netdev: Whether to allocate a netdev for this vNIC 2136 * @max_tx_rings: Maximum number of TX rings supported by device 2137 * @max_rx_rings: Maximum number of RX rings supported by device 2138 * 2139 * This function allocates a netdev device and fills in the initial 2140 * part of the @struct nfp_net structure. In case of control device 2141 * nfp_net structure is allocated without the netdev. 2142 * 2143 * Return: NFP Net device structure, or ERR_PTR on error. 2144 */ 2145 struct nfp_net * 2146 nfp_net_alloc(struct pci_dev *pdev, const struct nfp_dev_info *dev_info, 2147 void __iomem *ctrl_bar, bool needs_netdev, 2148 unsigned int max_tx_rings, unsigned int max_rx_rings) 2149 { 2150 u64 dma_mask = dma_get_mask(&pdev->dev); 2151 struct nfp_net *nn; 2152 int err; 2153 2154 if (needs_netdev) { 2155 struct net_device *netdev; 2156 2157 netdev = alloc_etherdev_mqs(sizeof(struct nfp_net), 2158 max_tx_rings, max_rx_rings); 2159 if (!netdev) 2160 return ERR_PTR(-ENOMEM); 2161 2162 SET_NETDEV_DEV(netdev, &pdev->dev); 2163 nn = netdev_priv(netdev); 2164 nn->dp.netdev = netdev; 2165 } else { 2166 nn = vzalloc(sizeof(*nn)); 2167 if (!nn) 2168 return ERR_PTR(-ENOMEM); 2169 } 2170 2171 nn->dp.dev = &pdev->dev; 2172 nn->dp.ctrl_bar = ctrl_bar; 2173 nn->dev_info = dev_info; 2174 nn->pdev = pdev; 2175 nfp_net_get_fw_version(&nn->fw_ver, ctrl_bar); 2176 2177 switch (FIELD_GET(NFP_NET_CFG_VERSION_DP_MASK, nn->fw_ver.extend)) { 2178 case NFP_NET_CFG_VERSION_DP_NFD3: 2179 nn->dp.ops = &nfp_nfd3_ops; 2180 break; 2181 case NFP_NET_CFG_VERSION_DP_NFDK: 2182 if (nn->fw_ver.major < 5) { 2183 dev_err(&pdev->dev, 2184 "NFDK must use ABI 5 or newer, found: %d\n", 2185 nn->fw_ver.major); 2186 err = -EINVAL; 2187 goto err_free_nn; 2188 } 2189 nn->dp.ops = &nfp_nfdk_ops; 2190 break; 2191 default: 2192 err = -EINVAL; 2193 goto err_free_nn; 2194 } 2195 2196 if ((dma_mask & nn->dp.ops->dma_mask) != dma_mask) { 2197 dev_err(&pdev->dev, 2198 "DMA mask of loaded firmware: %llx, required DMA mask: %llx\n", 2199 nn->dp.ops->dma_mask, dma_mask); 2200 err = -EINVAL; 2201 goto err_free_nn; 2202 } 2203 2204 nn->max_tx_rings = max_tx_rings; 2205 nn->max_rx_rings = max_rx_rings; 2206 2207 nn->dp.num_tx_rings = min_t(unsigned int, 2208 max_tx_rings, num_online_cpus()); 2209 nn->dp.num_rx_rings = min_t(unsigned int, max_rx_rings, 2210 netif_get_num_default_rss_queues()); 2211 2212 nn->dp.num_r_vecs = max(nn->dp.num_tx_rings, nn->dp.num_rx_rings); 2213 nn->dp.num_r_vecs = min_t(unsigned int, 2214 nn->dp.num_r_vecs, num_online_cpus()); 2215 nn->max_r_vecs = nn->dp.num_r_vecs; 2216 2217 nn->dp.xsk_pools = kcalloc(nn->max_r_vecs, sizeof(nn->dp.xsk_pools), 2218 GFP_KERNEL); 2219 if (!nn->dp.xsk_pools) { 2220 err = -ENOMEM; 2221 goto err_free_nn; 2222 } 2223 2224 nn->dp.txd_cnt = NFP_NET_TX_DESCS_DEFAULT; 2225 nn->dp.rxd_cnt = NFP_NET_RX_DESCS_DEFAULT; 2226 2227 sema_init(&nn->bar_lock, 1); 2228 2229 spin_lock_init(&nn->reconfig_lock); 2230 spin_lock_init(&nn->link_status_lock); 2231 2232 timer_setup(&nn->reconfig_timer, nfp_net_reconfig_timer, 0); 2233 2234 err = nfp_net_tlv_caps_parse(&nn->pdev->dev, nn->dp.ctrl_bar, 2235 &nn->tlv_caps); 2236 if (err) 2237 goto err_free_nn; 2238 2239 err = nfp_ccm_mbox_alloc(nn); 2240 if (err) 2241 goto err_free_nn; 2242 2243 return nn; 2244 2245 err_free_nn: 2246 if (nn->dp.netdev) 2247 free_netdev(nn->dp.netdev); 2248 else 2249 vfree(nn); 2250 return ERR_PTR(err); 2251 } 2252 2253 /** 2254 * nfp_net_free() - Undo what @nfp_net_alloc() did 2255 * @nn: NFP Net device to reconfigure 2256 */ 2257 void nfp_net_free(struct nfp_net *nn) 2258 { 2259 WARN_ON(timer_pending(&nn->reconfig_timer) || nn->reconfig_posted); 2260 nfp_ccm_mbox_free(nn); 2261 2262 kfree(nn->dp.xsk_pools); 2263 if (nn->dp.netdev) 2264 free_netdev(nn->dp.netdev); 2265 else 2266 vfree(nn); 2267 } 2268 2269 /** 2270 * nfp_net_rss_key_sz() - Get current size of the RSS key 2271 * @nn: NFP Net device instance 2272 * 2273 * Return: size of the RSS key for currently selected hash function. 2274 */ 2275 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn) 2276 { 2277 switch (nn->rss_hfunc) { 2278 case ETH_RSS_HASH_TOP: 2279 return NFP_NET_CFG_RSS_KEY_SZ; 2280 case ETH_RSS_HASH_XOR: 2281 return 0; 2282 case ETH_RSS_HASH_CRC32: 2283 return 4; 2284 } 2285 2286 nn_warn(nn, "Unknown hash function: %u\n", nn->rss_hfunc); 2287 return 0; 2288 } 2289 2290 /** 2291 * nfp_net_rss_init() - Set the initial RSS parameters 2292 * @nn: NFP Net device to reconfigure 2293 */ 2294 static void nfp_net_rss_init(struct nfp_net *nn) 2295 { 2296 unsigned long func_bit, rss_cap_hfunc; 2297 u32 reg; 2298 2299 /* Read the RSS function capability and select first supported func */ 2300 reg = nn_readl(nn, NFP_NET_CFG_RSS_CAP); 2301 rss_cap_hfunc = FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, reg); 2302 if (!rss_cap_hfunc) 2303 rss_cap_hfunc = FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, 2304 NFP_NET_CFG_RSS_TOEPLITZ); 2305 2306 func_bit = find_first_bit(&rss_cap_hfunc, NFP_NET_CFG_RSS_HFUNCS); 2307 if (func_bit == NFP_NET_CFG_RSS_HFUNCS) { 2308 dev_warn(nn->dp.dev, 2309 "Bad RSS config, defaulting to Toeplitz hash\n"); 2310 func_bit = ETH_RSS_HASH_TOP_BIT; 2311 } 2312 nn->rss_hfunc = 1 << func_bit; 2313 2314 netdev_rss_key_fill(nn->rss_key, nfp_net_rss_key_sz(nn)); 2315 2316 nfp_net_rss_init_itbl(nn); 2317 2318 /* Enable IPv4/IPv6 TCP by default */ 2319 nn->rss_cfg = NFP_NET_CFG_RSS_IPV4_TCP | 2320 NFP_NET_CFG_RSS_IPV6_TCP | 2321 FIELD_PREP(NFP_NET_CFG_RSS_HFUNC, nn->rss_hfunc) | 2322 NFP_NET_CFG_RSS_MASK; 2323 } 2324 2325 /** 2326 * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters 2327 * @nn: NFP Net device to reconfigure 2328 */ 2329 static void nfp_net_irqmod_init(struct nfp_net *nn) 2330 { 2331 nn->rx_coalesce_usecs = 50; 2332 nn->rx_coalesce_max_frames = 64; 2333 nn->tx_coalesce_usecs = 50; 2334 nn->tx_coalesce_max_frames = 64; 2335 2336 nn->rx_coalesce_adapt_on = true; 2337 nn->tx_coalesce_adapt_on = true; 2338 } 2339 2340 static void nfp_net_netdev_init(struct nfp_net *nn) 2341 { 2342 struct net_device *netdev = nn->dp.netdev; 2343 2344 nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr); 2345 2346 netdev->mtu = nn->dp.mtu; 2347 2348 /* Advertise/enable offloads based on capabilities 2349 * 2350 * Note: netdev->features show the currently enabled features 2351 * and netdev->hw_features advertises which features are 2352 * supported. By default we enable most features. 2353 */ 2354 if (nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR) 2355 netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 2356 2357 netdev->hw_features = NETIF_F_HIGHDMA; 2358 if (nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY) { 2359 netdev->hw_features |= NETIF_F_RXCSUM; 2360 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY; 2361 } 2362 if (nn->cap & NFP_NET_CFG_CTRL_TXCSUM) { 2363 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; 2364 nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXCSUM; 2365 } 2366 if (nn->cap & NFP_NET_CFG_CTRL_GATHER) { 2367 netdev->hw_features |= NETIF_F_SG; 2368 nn->dp.ctrl |= NFP_NET_CFG_CTRL_GATHER; 2369 } 2370 if ((nn->cap & NFP_NET_CFG_CTRL_LSO && nn->fw_ver.major > 2) || 2371 nn->cap & NFP_NET_CFG_CTRL_LSO2) { 2372 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6; 2373 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?: 2374 NFP_NET_CFG_CTRL_LSO; 2375 } 2376 if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) 2377 netdev->hw_features |= NETIF_F_RXHASH; 2378 if (nn->cap & NFP_NET_CFG_CTRL_VXLAN) { 2379 if (nn->cap & NFP_NET_CFG_CTRL_LSO) { 2380 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL | 2381 NETIF_F_GSO_UDP_TUNNEL_CSUM | 2382 NETIF_F_GSO_PARTIAL; 2383 netdev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM; 2384 } 2385 netdev->udp_tunnel_nic_info = &nfp_udp_tunnels; 2386 nn->dp.ctrl |= NFP_NET_CFG_CTRL_VXLAN; 2387 } 2388 if (nn->cap & NFP_NET_CFG_CTRL_NVGRE) { 2389 if (nn->cap & NFP_NET_CFG_CTRL_LSO) 2390 netdev->hw_features |= NETIF_F_GSO_GRE; 2391 nn->dp.ctrl |= NFP_NET_CFG_CTRL_NVGRE; 2392 } 2393 if (nn->cap & (NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE)) 2394 netdev->hw_enc_features = netdev->hw_features; 2395 2396 netdev->vlan_features = netdev->hw_features; 2397 2398 if (nn->cap & NFP_NET_CFG_CTRL_RXVLAN_ANY) { 2399 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX; 2400 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ?: 2401 NFP_NET_CFG_CTRL_RXVLAN; 2402 } 2403 if (nn->cap & NFP_NET_CFG_CTRL_TXVLAN_ANY) { 2404 if (nn->cap & NFP_NET_CFG_CTRL_LSO2) { 2405 nn_warn(nn, "Device advertises both TSO2 and TXVLAN. Refusing to enable TXVLAN.\n"); 2406 } else { 2407 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX; 2408 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ?: 2409 NFP_NET_CFG_CTRL_TXVLAN; 2410 } 2411 } 2412 if (nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER) { 2413 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER; 2414 nn->dp.ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER; 2415 } 2416 if (nn->cap & NFP_NET_CFG_CTRL_RXQINQ) { 2417 netdev->hw_features |= NETIF_F_HW_VLAN_STAG_RX; 2418 nn->dp.ctrl |= NFP_NET_CFG_CTRL_RXQINQ; 2419 } 2420 2421 netdev->features = netdev->hw_features; 2422 2423 if (nfp_app_has_tc(nn->app) && nn->port) 2424 netdev->hw_features |= NETIF_F_HW_TC; 2425 2426 /* C-Tag strip and S-Tag strip can't be supported simultaneously, 2427 * so enable C-Tag strip and disable S-Tag strip by default. 2428 */ 2429 netdev->features &= ~NETIF_F_HW_VLAN_STAG_RX; 2430 nn->dp.ctrl &= ~NFP_NET_CFG_CTRL_RXQINQ; 2431 2432 /* Finalise the netdev setup */ 2433 switch (nn->dp.ops->version) { 2434 case NFP_NFD_VER_NFD3: 2435 netdev->netdev_ops = &nfp_nfd3_netdev_ops; 2436 break; 2437 case NFP_NFD_VER_NFDK: 2438 netdev->netdev_ops = &nfp_nfdk_netdev_ops; 2439 break; 2440 } 2441 2442 netdev->watchdog_timeo = msecs_to_jiffies(5 * 1000); 2443 2444 /* MTU range: 68 - hw-specific max */ 2445 netdev->min_mtu = ETH_MIN_MTU; 2446 netdev->max_mtu = nn->max_mtu; 2447 2448 netif_set_tso_max_segs(netdev, NFP_NET_LSO_MAX_SEGS); 2449 2450 netif_carrier_off(netdev); 2451 2452 nfp_net_set_ethtool_ops(netdev); 2453 } 2454 2455 static int nfp_net_read_caps(struct nfp_net *nn) 2456 { 2457 /* Get some of the read-only fields from the BAR */ 2458 nn->cap = nn_readl(nn, NFP_NET_CFG_CAP); 2459 nn->max_mtu = nn_readl(nn, NFP_NET_CFG_MAX_MTU); 2460 2461 /* ABI 4.x and ctrl vNIC always use chained metadata, in other cases 2462 * we allow use of non-chained metadata if RSS(v1) is the only 2463 * advertised capability requiring metadata. 2464 */ 2465 nn->dp.chained_metadata_format = nn->fw_ver.major == 4 || 2466 !nn->dp.netdev || 2467 !(nn->cap & NFP_NET_CFG_CTRL_RSS) || 2468 nn->cap & NFP_NET_CFG_CTRL_CHAIN_META; 2469 /* RSS(v1) uses non-chained metadata format, except in ABI 4.x where 2470 * it has the same meaning as RSSv2. 2471 */ 2472 if (nn->dp.chained_metadata_format && nn->fw_ver.major != 4) 2473 nn->cap &= ~NFP_NET_CFG_CTRL_RSS; 2474 2475 /* Determine RX packet/metadata boundary offset */ 2476 if (nn->fw_ver.major >= 2) { 2477 u32 reg; 2478 2479 reg = nn_readl(nn, NFP_NET_CFG_RX_OFFSET); 2480 if (reg > NFP_NET_MAX_PREPEND) { 2481 nn_err(nn, "Invalid rx offset: %d\n", reg); 2482 return -EINVAL; 2483 } 2484 nn->dp.rx_offset = reg; 2485 } else { 2486 nn->dp.rx_offset = NFP_NET_RX_OFFSET; 2487 } 2488 2489 /* Mask out NFD-version-specific features */ 2490 nn->cap &= nn->dp.ops->cap_mask; 2491 2492 /* For control vNICs mask out the capabilities app doesn't want. */ 2493 if (!nn->dp.netdev) 2494 nn->cap &= nn->app->type->ctrl_cap_mask; 2495 2496 return 0; 2497 } 2498 2499 /** 2500 * nfp_net_init() - Initialise/finalise the nfp_net structure 2501 * @nn: NFP Net device structure 2502 * 2503 * Return: 0 on success or negative errno on error. 2504 */ 2505 int nfp_net_init(struct nfp_net *nn) 2506 { 2507 int err; 2508 2509 nn->dp.rx_dma_dir = DMA_FROM_DEVICE; 2510 2511 err = nfp_net_read_caps(nn); 2512 if (err) 2513 return err; 2514 2515 /* Set default MTU and Freelist buffer size */ 2516 if (!nfp_net_is_data_vnic(nn) && nn->app->ctrl_mtu) { 2517 nn->dp.mtu = min(nn->app->ctrl_mtu, nn->max_mtu); 2518 } else if (nn->max_mtu < NFP_NET_DEFAULT_MTU) { 2519 nn->dp.mtu = nn->max_mtu; 2520 } else { 2521 nn->dp.mtu = NFP_NET_DEFAULT_MTU; 2522 } 2523 nn->dp.fl_bufsz = nfp_net_calc_fl_bufsz(&nn->dp); 2524 2525 if (nfp_app_ctrl_uses_data_vnics(nn->app)) 2526 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_CMSG_DATA; 2527 2528 if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) { 2529 nfp_net_rss_init(nn); 2530 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RSS2 ?: 2531 NFP_NET_CFG_CTRL_RSS; 2532 } 2533 2534 /* Allow L2 Broadcast and Multicast through by default, if supported */ 2535 if (nn->cap & NFP_NET_CFG_CTRL_L2BC) 2536 nn->dp.ctrl |= NFP_NET_CFG_CTRL_L2BC; 2537 2538 /* Allow IRQ moderation, if supported */ 2539 if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) { 2540 nfp_net_irqmod_init(nn); 2541 nn->dp.ctrl |= NFP_NET_CFG_CTRL_IRQMOD; 2542 } 2543 2544 /* Enable TX pointer writeback, if supported */ 2545 if (nn->cap & NFP_NET_CFG_CTRL_TXRWB) 2546 nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXRWB; 2547 2548 /* Stash the re-configuration queue away. First odd queue in TX Bar */ 2549 nn->qcp_cfg = nn->tx_bar + NFP_QCP_QUEUE_ADDR_SZ; 2550 2551 /* Make sure the FW knows the netdev is supposed to be disabled here */ 2552 nn_writel(nn, NFP_NET_CFG_CTRL, 0); 2553 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0); 2554 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0); 2555 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_RING | 2556 NFP_NET_CFG_UPDATE_GEN); 2557 if (err) 2558 return err; 2559 2560 if (nn->dp.netdev) { 2561 nfp_net_netdev_init(nn); 2562 2563 err = nfp_ccm_mbox_init(nn); 2564 if (err) 2565 return err; 2566 2567 err = nfp_net_tls_init(nn); 2568 if (err) 2569 goto err_clean_mbox; 2570 } 2571 2572 nfp_net_vecs_init(nn); 2573 2574 if (!nn->dp.netdev) 2575 return 0; 2576 return register_netdev(nn->dp.netdev); 2577 2578 err_clean_mbox: 2579 nfp_ccm_mbox_clean(nn); 2580 return err; 2581 } 2582 2583 /** 2584 * nfp_net_clean() - Undo what nfp_net_init() did. 2585 * @nn: NFP Net device structure 2586 */ 2587 void nfp_net_clean(struct nfp_net *nn) 2588 { 2589 if (!nn->dp.netdev) 2590 return; 2591 2592 unregister_netdev(nn->dp.netdev); 2593 nfp_ccm_mbox_clean(nn); 2594 nfp_net_reconfig_wait_posted(nn); 2595 } 2596