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/ethtool.h> 31 #include <linux/log2.h> 32 #include <linux/if_vlan.h> 33 #include <linux/if_bridge.h> 34 #include <linux/random.h> 35 #include <linux/vmalloc.h> 36 #include <linux/ktime.h> 37 38 #include <net/tls.h> 39 #include <net/vxlan.h> 40 #include <net/xdp_sock_drv.h> 41 42 #include "nfpcore/nfp_dev.h" 43 #include "nfpcore/nfp_nsp.h" 44 #include "ccm.h" 45 #include "nfp_app.h" 46 #include "nfp_net_ctrl.h" 47 #include "nfp_net.h" 48 #include "nfp_net_dp.h" 49 #include "nfp_net_sriov.h" 50 #include "nfp_net_xsk.h" 51 #include "nfp_port.h" 52 #include "crypto/crypto.h" 53 #include "crypto/fw.h" 54 55 /** 56 * nfp_net_get_fw_version() - Read and parse the FW version 57 * @fw_ver: Output fw_version structure to read to 58 * @ctrl_bar: Mapped address of the control BAR 59 */ 60 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver, 61 void __iomem *ctrl_bar) 62 { 63 u32 reg; 64 65 reg = readl(ctrl_bar + NFP_NET_CFG_VERSION); 66 put_unaligned_le32(reg, fw_ver); 67 } 68 69 u32 nfp_qcp_queue_offset(const struct nfp_dev_info *dev_info, u16 queue) 70 { 71 queue &= dev_info->qc_idx_mask; 72 return dev_info->qc_addr_offset + NFP_QCP_QUEUE_ADDR_SZ * queue; 73 } 74 75 /* Firmware reconfig 76 * 77 * Firmware reconfig may take a while so we have two versions of it - 78 * synchronous and asynchronous (posted). All synchronous callers are holding 79 * RTNL so we don't have to worry about serializing them. 80 */ 81 static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update) 82 { 83 nn_writel(nn, NFP_NET_CFG_UPDATE, update); 84 /* ensure update is written before pinging HW */ 85 nn_pci_flush(nn); 86 nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1); 87 nn->reconfig_in_progress_update = update; 88 } 89 90 /* Pass 0 as update to run posted reconfigs. */ 91 static void nfp_net_reconfig_start_async(struct nfp_net *nn, u32 update) 92 { 93 update |= nn->reconfig_posted; 94 nn->reconfig_posted = 0; 95 96 nfp_net_reconfig_start(nn, update); 97 98 nn->reconfig_timer_active = true; 99 mod_timer(&nn->reconfig_timer, jiffies + NFP_NET_POLL_TIMEOUT * HZ); 100 } 101 102 static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check) 103 { 104 u32 reg; 105 106 reg = nn_readl(nn, NFP_NET_CFG_UPDATE); 107 if (reg == 0) 108 return true; 109 if (reg & NFP_NET_CFG_UPDATE_ERR) { 110 nn_err(nn, "Reconfig error (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n", 111 reg, nn->reconfig_in_progress_update, 112 nn_readl(nn, NFP_NET_CFG_CTRL)); 113 return true; 114 } else if (last_check) { 115 nn_err(nn, "Reconfig timeout (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n", 116 reg, nn->reconfig_in_progress_update, 117 nn_readl(nn, NFP_NET_CFG_CTRL)); 118 return true; 119 } 120 121 return false; 122 } 123 124 static bool __nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline) 125 { 126 bool timed_out = false; 127 int i; 128 129 /* Poll update field, waiting for NFP to ack the config. 130 * Do an opportunistic wait-busy loop, afterward sleep. 131 */ 132 for (i = 0; i < 50; i++) { 133 if (nfp_net_reconfig_check_done(nn, false)) 134 return false; 135 udelay(4); 136 } 137 138 while (!nfp_net_reconfig_check_done(nn, timed_out)) { 139 usleep_range(250, 500); 140 timed_out = time_is_before_eq_jiffies(deadline); 141 } 142 143 return timed_out; 144 } 145 146 static int nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline) 147 { 148 if (__nfp_net_reconfig_wait(nn, deadline)) 149 return -EIO; 150 151 if (nn_readl(nn, NFP_NET_CFG_UPDATE) & NFP_NET_CFG_UPDATE_ERR) 152 return -EIO; 153 154 return 0; 155 } 156 157 static void nfp_net_reconfig_timer(struct timer_list *t) 158 { 159 struct nfp_net *nn = from_timer(nn, t, reconfig_timer); 160 161 spin_lock_bh(&nn->reconfig_lock); 162 163 nn->reconfig_timer_active = false; 164 165 /* If sync caller is present it will take over from us */ 166 if (nn->reconfig_sync_present) 167 goto done; 168 169 /* Read reconfig status and report errors */ 170 nfp_net_reconfig_check_done(nn, true); 171 172 if (nn->reconfig_posted) 173 nfp_net_reconfig_start_async(nn, 0); 174 done: 175 spin_unlock_bh(&nn->reconfig_lock); 176 } 177 178 /** 179 * nfp_net_reconfig_post() - Post async reconfig request 180 * @nn: NFP Net device to reconfigure 181 * @update: The value for the update field in the BAR config 182 * 183 * Record FW reconfiguration request. Reconfiguration will be kicked off 184 * whenever reconfiguration machinery is idle. Multiple requests can be 185 * merged together! 186 */ 187 static void nfp_net_reconfig_post(struct nfp_net *nn, u32 update) 188 { 189 spin_lock_bh(&nn->reconfig_lock); 190 191 /* Sync caller will kick off async reconf when it's done, just post */ 192 if (nn->reconfig_sync_present) { 193 nn->reconfig_posted |= update; 194 goto done; 195 } 196 197 /* Opportunistically check if the previous command is done */ 198 if (!nn->reconfig_timer_active || 199 nfp_net_reconfig_check_done(nn, false)) 200 nfp_net_reconfig_start_async(nn, update); 201 else 202 nn->reconfig_posted |= update; 203 done: 204 spin_unlock_bh(&nn->reconfig_lock); 205 } 206 207 static void nfp_net_reconfig_sync_enter(struct nfp_net *nn) 208 { 209 bool cancelled_timer = false; 210 u32 pre_posted_requests; 211 212 spin_lock_bh(&nn->reconfig_lock); 213 214 WARN_ON(nn->reconfig_sync_present); 215 nn->reconfig_sync_present = true; 216 217 if (nn->reconfig_timer_active) { 218 nn->reconfig_timer_active = false; 219 cancelled_timer = true; 220 } 221 pre_posted_requests = nn->reconfig_posted; 222 nn->reconfig_posted = 0; 223 224 spin_unlock_bh(&nn->reconfig_lock); 225 226 if (cancelled_timer) { 227 del_timer_sync(&nn->reconfig_timer); 228 nfp_net_reconfig_wait(nn, nn->reconfig_timer.expires); 229 } 230 231 /* Run the posted reconfigs which were issued before we started */ 232 if (pre_posted_requests) { 233 nfp_net_reconfig_start(nn, pre_posted_requests); 234 nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT); 235 } 236 } 237 238 static void nfp_net_reconfig_wait_posted(struct nfp_net *nn) 239 { 240 nfp_net_reconfig_sync_enter(nn); 241 242 spin_lock_bh(&nn->reconfig_lock); 243 nn->reconfig_sync_present = false; 244 spin_unlock_bh(&nn->reconfig_lock); 245 } 246 247 /** 248 * __nfp_net_reconfig() - Reconfigure the firmware 249 * @nn: NFP Net device to reconfigure 250 * @update: The value for the update field in the BAR config 251 * 252 * Write the update word to the BAR and ping the reconfig queue. The 253 * poll until the firmware has acknowledged the update by zeroing the 254 * update word. 255 * 256 * Return: Negative errno on error, 0 on success 257 */ 258 int __nfp_net_reconfig(struct nfp_net *nn, u32 update) 259 { 260 int ret; 261 262 nfp_net_reconfig_sync_enter(nn); 263 264 nfp_net_reconfig_start(nn, update); 265 ret = nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT); 266 267 spin_lock_bh(&nn->reconfig_lock); 268 269 if (nn->reconfig_posted) 270 nfp_net_reconfig_start_async(nn, 0); 271 272 nn->reconfig_sync_present = false; 273 274 spin_unlock_bh(&nn->reconfig_lock); 275 276 return ret; 277 } 278 279 int nfp_net_reconfig(struct nfp_net *nn, u32 update) 280 { 281 int ret; 282 283 nn_ctrl_bar_lock(nn); 284 ret = __nfp_net_reconfig(nn, update); 285 nn_ctrl_bar_unlock(nn); 286 287 return ret; 288 } 289 290 int nfp_net_mbox_lock(struct nfp_net *nn, unsigned int data_size) 291 { 292 if (nn->tlv_caps.mbox_len < NFP_NET_CFG_MBOX_SIMPLE_VAL + data_size) { 293 nn_err(nn, "mailbox too small for %u of data (%u)\n", 294 data_size, nn->tlv_caps.mbox_len); 295 return -EIO; 296 } 297 298 nn_ctrl_bar_lock(nn); 299 return 0; 300 } 301 302 /** 303 * nfp_net_mbox_reconfig() - Reconfigure the firmware via the mailbox 304 * @nn: NFP Net device to reconfigure 305 * @mbox_cmd: The value for the mailbox command 306 * 307 * Helper function for mailbox updates 308 * 309 * Return: Negative errno on error, 0 on success 310 */ 311 int nfp_net_mbox_reconfig(struct nfp_net *nn, u32 mbox_cmd) 312 { 313 u32 mbox = nn->tlv_caps.mbox_off; 314 int ret; 315 316 nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd); 317 318 ret = __nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MBOX); 319 if (ret) { 320 nn_err(nn, "Mailbox update error\n"); 321 return ret; 322 } 323 324 return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET); 325 } 326 327 void nfp_net_mbox_reconfig_post(struct nfp_net *nn, u32 mbox_cmd) 328 { 329 u32 mbox = nn->tlv_caps.mbox_off; 330 331 nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd); 332 333 nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_MBOX); 334 } 335 336 int nfp_net_mbox_reconfig_wait_posted(struct nfp_net *nn) 337 { 338 u32 mbox = nn->tlv_caps.mbox_off; 339 340 nfp_net_reconfig_wait_posted(nn); 341 342 return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET); 343 } 344 345 int nfp_net_mbox_reconfig_and_unlock(struct nfp_net *nn, u32 mbox_cmd) 346 { 347 int ret; 348 349 ret = nfp_net_mbox_reconfig(nn, mbox_cmd); 350 nn_ctrl_bar_unlock(nn); 351 return ret; 352 } 353 354 /* Interrupt configuration and handling 355 */ 356 357 /** 358 * nfp_net_irqs_alloc() - allocates MSI-X irqs 359 * @pdev: PCI device structure 360 * @irq_entries: Array to be initialized and used to hold the irq entries 361 * @min_irqs: Minimal acceptable number of interrupts 362 * @wanted_irqs: Target number of interrupts to allocate 363 * 364 * Return: Number of irqs obtained or 0 on error. 365 */ 366 unsigned int 367 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries, 368 unsigned int min_irqs, unsigned int wanted_irqs) 369 { 370 unsigned int i; 371 int got_irqs; 372 373 for (i = 0; i < wanted_irqs; i++) 374 irq_entries[i].entry = i; 375 376 got_irqs = pci_enable_msix_range(pdev, irq_entries, 377 min_irqs, wanted_irqs); 378 if (got_irqs < 0) { 379 dev_err(&pdev->dev, "Failed to enable %d-%d MSI-X (err=%d)\n", 380 min_irqs, wanted_irqs, got_irqs); 381 return 0; 382 } 383 384 if (got_irqs < wanted_irqs) 385 dev_warn(&pdev->dev, "Unable to allocate %d IRQs got only %d\n", 386 wanted_irqs, got_irqs); 387 388 return got_irqs; 389 } 390 391 /** 392 * nfp_net_irqs_assign() - Assign interrupts allocated externally to netdev 393 * @nn: NFP Network structure 394 * @irq_entries: Table of allocated interrupts 395 * @n: Size of @irq_entries (number of entries to grab) 396 * 397 * After interrupts are allocated with nfp_net_irqs_alloc() this function 398 * should be called to assign them to a specific netdev (port). 399 */ 400 void 401 nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries, 402 unsigned int n) 403 { 404 struct nfp_net_dp *dp = &nn->dp; 405 406 nn->max_r_vecs = n - NFP_NET_NON_Q_VECTORS; 407 dp->num_r_vecs = nn->max_r_vecs; 408 409 memcpy(nn->irq_entries, irq_entries, sizeof(*irq_entries) * n); 410 411 if (dp->num_rx_rings > dp->num_r_vecs || 412 dp->num_tx_rings > dp->num_r_vecs) 413 dev_warn(nn->dp.dev, "More rings (%d,%d) than vectors (%d).\n", 414 dp->num_rx_rings, dp->num_tx_rings, 415 dp->num_r_vecs); 416 417 dp->num_rx_rings = min(dp->num_r_vecs, dp->num_rx_rings); 418 dp->num_tx_rings = min(dp->num_r_vecs, dp->num_tx_rings); 419 dp->num_stack_tx_rings = dp->num_tx_rings; 420 } 421 422 /** 423 * nfp_net_irqs_disable() - Disable interrupts 424 * @pdev: PCI device structure 425 * 426 * Undoes what @nfp_net_irqs_alloc() does. 427 */ 428 void nfp_net_irqs_disable(struct pci_dev *pdev) 429 { 430 pci_disable_msix(pdev); 431 } 432 433 /** 434 * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings. 435 * @irq: Interrupt 436 * @data: Opaque data structure 437 * 438 * Return: Indicate if the interrupt has been handled. 439 */ 440 static irqreturn_t nfp_net_irq_rxtx(int irq, void *data) 441 { 442 struct nfp_net_r_vector *r_vec = data; 443 444 /* Currently we cannot tell if it's a rx or tx interrupt, 445 * since dim does not need accurate event_ctr to calculate, 446 * we just use this counter for both rx and tx dim. 447 */ 448 r_vec->event_ctr++; 449 450 napi_schedule_irqoff(&r_vec->napi); 451 452 /* The FW auto-masks any interrupt, either via the MASK bit in 453 * the MSI-X table or via the per entry ICR field. So there 454 * is no need to disable interrupts here. 455 */ 456 return IRQ_HANDLED; 457 } 458 459 static irqreturn_t nfp_ctrl_irq_rxtx(int irq, void *data) 460 { 461 struct nfp_net_r_vector *r_vec = data; 462 463 tasklet_schedule(&r_vec->tasklet); 464 465 return IRQ_HANDLED; 466 } 467 468 /** 469 * nfp_net_read_link_status() - Reread link status from control BAR 470 * @nn: NFP Network structure 471 */ 472 static void nfp_net_read_link_status(struct nfp_net *nn) 473 { 474 unsigned long flags; 475 bool link_up; 476 u16 sts; 477 478 spin_lock_irqsave(&nn->link_status_lock, flags); 479 480 sts = nn_readw(nn, NFP_NET_CFG_STS); 481 link_up = !!(sts & NFP_NET_CFG_STS_LINK); 482 483 if (nn->link_up == link_up) 484 goto out; 485 486 nn->link_up = link_up; 487 if (nn->port) { 488 set_bit(NFP_PORT_CHANGED, &nn->port->flags); 489 if (nn->port->link_cb) 490 nn->port->link_cb(nn->port); 491 } 492 493 if (nn->link_up) { 494 netif_carrier_on(nn->dp.netdev); 495 netdev_info(nn->dp.netdev, "NIC Link is Up\n"); 496 } else { 497 netif_carrier_off(nn->dp.netdev); 498 netdev_info(nn->dp.netdev, "NIC Link is Down\n"); 499 } 500 out: 501 spin_unlock_irqrestore(&nn->link_status_lock, flags); 502 } 503 504 /** 505 * nfp_net_irq_lsc() - Interrupt service routine for link state changes 506 * @irq: Interrupt 507 * @data: Opaque data structure 508 * 509 * Return: Indicate if the interrupt has been handled. 510 */ 511 static irqreturn_t nfp_net_irq_lsc(int irq, void *data) 512 { 513 struct nfp_net *nn = data; 514 struct msix_entry *entry; 515 516 entry = &nn->irq_entries[NFP_NET_IRQ_LSC_IDX]; 517 518 nfp_net_read_link_status(nn); 519 520 nfp_net_irq_unmask(nn, entry->entry); 521 522 return IRQ_HANDLED; 523 } 524 525 /** 526 * nfp_net_irq_exn() - Interrupt service routine for exceptions 527 * @irq: Interrupt 528 * @data: Opaque data structure 529 * 530 * Return: Indicate if the interrupt has been handled. 531 */ 532 static irqreturn_t nfp_net_irq_exn(int irq, void *data) 533 { 534 struct nfp_net *nn = data; 535 536 nn_err(nn, "%s: UNIMPLEMENTED.\n", __func__); 537 /* XXX TO BE IMPLEMENTED */ 538 return IRQ_HANDLED; 539 } 540 541 /** 542 * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN) 543 * @nn: NFP Network structure 544 * @ctrl_offset: Control BAR offset where IRQ configuration should be written 545 * @format: printf-style format to construct the interrupt name 546 * @name: Pointer to allocated space for interrupt name 547 * @name_sz: Size of space for interrupt name 548 * @vector_idx: Index of MSI-X vector used for this interrupt 549 * @handler: IRQ handler to register for this interrupt 550 */ 551 static int 552 nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset, 553 const char *format, char *name, size_t name_sz, 554 unsigned int vector_idx, irq_handler_t handler) 555 { 556 struct msix_entry *entry; 557 int err; 558 559 entry = &nn->irq_entries[vector_idx]; 560 561 snprintf(name, name_sz, format, nfp_net_name(nn)); 562 err = request_irq(entry->vector, handler, 0, name, nn); 563 if (err) { 564 nn_err(nn, "Failed to request IRQ %d (err=%d).\n", 565 entry->vector, err); 566 return err; 567 } 568 nn_writeb(nn, ctrl_offset, entry->entry); 569 nfp_net_irq_unmask(nn, entry->entry); 570 571 return 0; 572 } 573 574 /** 575 * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN) 576 * @nn: NFP Network structure 577 * @ctrl_offset: Control BAR offset where IRQ configuration should be written 578 * @vector_idx: Index of MSI-X vector used for this interrupt 579 */ 580 static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset, 581 unsigned int vector_idx) 582 { 583 nn_writeb(nn, ctrl_offset, 0xff); 584 nn_pci_flush(nn); 585 free_irq(nn->irq_entries[vector_idx].vector, nn); 586 } 587 588 struct sk_buff * 589 nfp_net_tls_tx(struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec, 590 struct sk_buff *skb, u64 *tls_handle, int *nr_frags) 591 { 592 #ifdef CONFIG_TLS_DEVICE 593 struct nfp_net_tls_offload_ctx *ntls; 594 struct sk_buff *nskb; 595 bool resync_pending; 596 u32 datalen, seq; 597 598 if (likely(!dp->ktls_tx)) 599 return skb; 600 if (!skb->sk || !tls_is_sk_tx_device_offloaded(skb->sk)) 601 return skb; 602 603 datalen = skb->len - skb_tcp_all_headers(skb); 604 seq = ntohl(tcp_hdr(skb)->seq); 605 ntls = tls_driver_ctx(skb->sk, TLS_OFFLOAD_CTX_DIR_TX); 606 resync_pending = tls_offload_tx_resync_pending(skb->sk); 607 if (unlikely(resync_pending || ntls->next_seq != seq)) { 608 /* Pure ACK out of order already */ 609 if (!datalen) 610 return skb; 611 612 u64_stats_update_begin(&r_vec->tx_sync); 613 r_vec->tls_tx_fallback++; 614 u64_stats_update_end(&r_vec->tx_sync); 615 616 nskb = tls_encrypt_skb(skb); 617 if (!nskb) { 618 u64_stats_update_begin(&r_vec->tx_sync); 619 r_vec->tls_tx_no_fallback++; 620 u64_stats_update_end(&r_vec->tx_sync); 621 return NULL; 622 } 623 /* encryption wasn't necessary */ 624 if (nskb == skb) 625 return skb; 626 /* we don't re-check ring space */ 627 if (unlikely(skb_is_nonlinear(nskb))) { 628 nn_dp_warn(dp, "tls_encrypt_skb() produced fragmented frame\n"); 629 u64_stats_update_begin(&r_vec->tx_sync); 630 r_vec->tx_errors++; 631 u64_stats_update_end(&r_vec->tx_sync); 632 dev_kfree_skb_any(nskb); 633 return NULL; 634 } 635 636 /* jump forward, a TX may have gotten lost, need to sync TX */ 637 if (!resync_pending && seq - ntls->next_seq < U32_MAX / 4) 638 tls_offload_tx_resync_request(nskb->sk, seq, 639 ntls->next_seq); 640 641 *nr_frags = 0; 642 return nskb; 643 } 644 645 if (datalen) { 646 u64_stats_update_begin(&r_vec->tx_sync); 647 if (!skb_is_gso(skb)) 648 r_vec->hw_tls_tx++; 649 else 650 r_vec->hw_tls_tx += skb_shinfo(skb)->gso_segs; 651 u64_stats_update_end(&r_vec->tx_sync); 652 } 653 654 memcpy(tls_handle, ntls->fw_handle, sizeof(ntls->fw_handle)); 655 ntls->next_seq += datalen; 656 #endif 657 return skb; 658 } 659 660 void nfp_net_tls_tx_undo(struct sk_buff *skb, u64 tls_handle) 661 { 662 #ifdef CONFIG_TLS_DEVICE 663 struct nfp_net_tls_offload_ctx *ntls; 664 u32 datalen, seq; 665 666 if (!tls_handle) 667 return; 668 if (WARN_ON_ONCE(!skb->sk || !tls_is_sk_tx_device_offloaded(skb->sk))) 669 return; 670 671 datalen = skb->len - skb_tcp_all_headers(skb); 672 seq = ntohl(tcp_hdr(skb)->seq); 673 674 ntls = tls_driver_ctx(skb->sk, TLS_OFFLOAD_CTX_DIR_TX); 675 if (ntls->next_seq == seq + datalen) 676 ntls->next_seq = seq; 677 else 678 WARN_ON_ONCE(1); 679 #endif 680 } 681 682 static void nfp_net_tx_timeout(struct net_device *netdev, unsigned int txqueue) 683 { 684 struct nfp_net *nn = netdev_priv(netdev); 685 686 nn_warn(nn, "TX watchdog timeout on ring: %u\n", txqueue); 687 } 688 689 /* Receive processing */ 690 static unsigned int 691 nfp_net_calc_fl_bufsz_data(struct nfp_net_dp *dp) 692 { 693 unsigned int fl_bufsz = 0; 694 695 if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC) 696 fl_bufsz += NFP_NET_MAX_PREPEND; 697 else 698 fl_bufsz += dp->rx_offset; 699 fl_bufsz += ETH_HLEN + VLAN_HLEN * 2 + dp->mtu; 700 701 return fl_bufsz; 702 } 703 704 static unsigned int nfp_net_calc_fl_bufsz(struct nfp_net_dp *dp) 705 { 706 unsigned int fl_bufsz; 707 708 fl_bufsz = NFP_NET_RX_BUF_HEADROOM; 709 fl_bufsz += dp->rx_dma_off; 710 fl_bufsz += nfp_net_calc_fl_bufsz_data(dp); 711 712 fl_bufsz = SKB_DATA_ALIGN(fl_bufsz); 713 fl_bufsz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 714 715 return fl_bufsz; 716 } 717 718 static unsigned int nfp_net_calc_fl_bufsz_xsk(struct nfp_net_dp *dp) 719 { 720 unsigned int fl_bufsz; 721 722 fl_bufsz = XDP_PACKET_HEADROOM; 723 fl_bufsz += nfp_net_calc_fl_bufsz_data(dp); 724 725 return fl_bufsz; 726 } 727 728 /* Setup and Configuration 729 */ 730 731 /** 732 * nfp_net_vecs_init() - Assign IRQs and setup rvecs. 733 * @nn: NFP Network structure 734 */ 735 static void nfp_net_vecs_init(struct nfp_net *nn) 736 { 737 int numa_node = dev_to_node(&nn->pdev->dev); 738 struct nfp_net_r_vector *r_vec; 739 unsigned 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(cpumask_local_spread(r, numa_node), &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) ? dp->ops->xsk_poll : dp->ops->poll); 775 else 776 tasklet_enable(&r_vec->tasklet); 777 } 778 779 static void 780 nfp_net_napi_del(struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec) 781 { 782 if (dp->netdev) 783 netif_napi_del(&r_vec->napi); 784 else 785 tasklet_disable(&r_vec->tasklet); 786 } 787 788 static void 789 nfp_net_vector_assign_rings(struct nfp_net_dp *dp, 790 struct nfp_net_r_vector *r_vec, int idx) 791 { 792 r_vec->rx_ring = idx < dp->num_rx_rings ? &dp->rx_rings[idx] : NULL; 793 r_vec->tx_ring = 794 idx < dp->num_stack_tx_rings ? &dp->tx_rings[idx] : NULL; 795 796 r_vec->xdp_ring = idx < dp->num_tx_rings - dp->num_stack_tx_rings ? 797 &dp->tx_rings[dp->num_stack_tx_rings + idx] : NULL; 798 799 if (nfp_net_has_xsk_pool_slow(dp, idx) || r_vec->xsk_pool) { 800 r_vec->xsk_pool = dp->xdp_prog ? dp->xsk_pools[idx] : NULL; 801 802 if (r_vec->xsk_pool) 803 xsk_pool_set_rxq_info(r_vec->xsk_pool, 804 &r_vec->rx_ring->xdp_rxq); 805 806 nfp_net_napi_del(dp, r_vec); 807 nfp_net_napi_add(dp, r_vec, idx); 808 } 809 } 810 811 static int 812 nfp_net_prepare_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec, 813 int idx) 814 { 815 int err; 816 817 nfp_net_napi_add(&nn->dp, r_vec, idx); 818 819 snprintf(r_vec->name, sizeof(r_vec->name), 820 "%s-rxtx-%d", nfp_net_name(nn), idx); 821 err = request_irq(r_vec->irq_vector, r_vec->handler, 0, r_vec->name, 822 r_vec); 823 if (err) { 824 nfp_net_napi_del(&nn->dp, r_vec); 825 nn_err(nn, "Error requesting IRQ %d\n", r_vec->irq_vector); 826 return err; 827 } 828 disable_irq(r_vec->irq_vector); 829 830 irq_set_affinity_hint(r_vec->irq_vector, &r_vec->affinity_mask); 831 832 nn_dbg(nn, "RV%02d: irq=%03d/%03d\n", idx, r_vec->irq_vector, 833 r_vec->irq_entry); 834 835 return 0; 836 } 837 838 static void 839 nfp_net_cleanup_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec) 840 { 841 irq_set_affinity_hint(r_vec->irq_vector, NULL); 842 nfp_net_napi_del(&nn->dp, r_vec); 843 free_irq(r_vec->irq_vector, r_vec); 844 } 845 846 /** 847 * nfp_net_rss_write_itbl() - Write RSS indirection table to device 848 * @nn: NFP Net device to reconfigure 849 */ 850 void nfp_net_rss_write_itbl(struct nfp_net *nn) 851 { 852 int i; 853 854 for (i = 0; i < NFP_NET_CFG_RSS_ITBL_SZ; i += 4) 855 nn_writel(nn, NFP_NET_CFG_RSS_ITBL + i, 856 get_unaligned_le32(nn->rss_itbl + i)); 857 } 858 859 /** 860 * nfp_net_rss_write_key() - Write RSS hash key to device 861 * @nn: NFP Net device to reconfigure 862 */ 863 void nfp_net_rss_write_key(struct nfp_net *nn) 864 { 865 int i; 866 867 for (i = 0; i < nfp_net_rss_key_sz(nn); i += 4) 868 nn_writel(nn, NFP_NET_CFG_RSS_KEY + i, 869 get_unaligned_le32(nn->rss_key + i)); 870 } 871 872 /** 873 * nfp_net_coalesce_write_cfg() - Write irq coalescence configuration to HW 874 * @nn: NFP Net device to reconfigure 875 */ 876 void nfp_net_coalesce_write_cfg(struct nfp_net *nn) 877 { 878 u8 i; 879 u32 factor; 880 u32 value; 881 882 /* Compute factor used to convert coalesce '_usecs' parameters to 883 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp 884 * count. 885 */ 886 factor = nn->tlv_caps.me_freq_mhz / 16; 887 888 /* copy RX interrupt coalesce parameters */ 889 value = (nn->rx_coalesce_max_frames << 16) | 890 (factor * nn->rx_coalesce_usecs); 891 for (i = 0; i < nn->dp.num_rx_rings; i++) 892 nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(i), value); 893 894 /* copy TX interrupt coalesce parameters */ 895 value = (nn->tx_coalesce_max_frames << 16) | 896 (factor * nn->tx_coalesce_usecs); 897 for (i = 0; i < nn->dp.num_tx_rings; i++) 898 nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(i), value); 899 } 900 901 /** 902 * nfp_net_write_mac_addr() - Write mac address to the device control BAR 903 * @nn: NFP Net device to reconfigure 904 * @addr: MAC address to write 905 * 906 * Writes the MAC address from the netdev to the device control BAR. Does not 907 * perform the required reconfig. We do a bit of byte swapping dance because 908 * firmware is LE. 909 */ 910 static void nfp_net_write_mac_addr(struct nfp_net *nn, const u8 *addr) 911 { 912 nn_writel(nn, NFP_NET_CFG_MACADDR + 0, get_unaligned_be32(addr)); 913 nn_writew(nn, NFP_NET_CFG_MACADDR + 6, get_unaligned_be16(addr + 4)); 914 } 915 916 /** 917 * nfp_net_clear_config_and_disable() - Clear control BAR and disable NFP 918 * @nn: NFP Net device to reconfigure 919 * 920 * Warning: must be fully idempotent. 921 */ 922 static void nfp_net_clear_config_and_disable(struct nfp_net *nn) 923 { 924 u32 new_ctrl, update; 925 unsigned int r; 926 int err; 927 928 new_ctrl = nn->dp.ctrl; 929 new_ctrl &= ~NFP_NET_CFG_CTRL_ENABLE; 930 update = NFP_NET_CFG_UPDATE_GEN; 931 update |= NFP_NET_CFG_UPDATE_MSIX; 932 update |= NFP_NET_CFG_UPDATE_RING; 933 934 if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG) 935 new_ctrl &= ~NFP_NET_CFG_CTRL_RINGCFG; 936 937 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0); 938 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0); 939 940 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 941 err = nfp_net_reconfig(nn, update); 942 if (err) 943 nn_err(nn, "Could not disable device: %d\n", err); 944 945 for (r = 0; r < nn->dp.num_rx_rings; r++) { 946 nfp_net_rx_ring_reset(&nn->dp.rx_rings[r]); 947 if (nfp_net_has_xsk_pool_slow(&nn->dp, nn->dp.rx_rings[r].idx)) 948 nfp_net_xsk_rx_bufs_free(&nn->dp.rx_rings[r]); 949 } 950 for (r = 0; r < nn->dp.num_tx_rings; r++) 951 nfp_net_tx_ring_reset(&nn->dp, &nn->dp.tx_rings[r]); 952 for (r = 0; r < nn->dp.num_r_vecs; r++) 953 nfp_net_vec_clear_ring_data(nn, r); 954 955 nn->dp.ctrl = new_ctrl; 956 } 957 958 /** 959 * nfp_net_set_config_and_enable() - Write control BAR and enable NFP 960 * @nn: NFP Net device to reconfigure 961 */ 962 static int nfp_net_set_config_and_enable(struct nfp_net *nn) 963 { 964 u32 bufsz, new_ctrl, update = 0; 965 unsigned int r; 966 int err; 967 968 new_ctrl = nn->dp.ctrl; 969 970 if (nn->dp.ctrl & NFP_NET_CFG_CTRL_RSS_ANY) { 971 nfp_net_rss_write_key(nn); 972 nfp_net_rss_write_itbl(nn); 973 nn_writel(nn, NFP_NET_CFG_RSS_CTRL, nn->rss_cfg); 974 update |= NFP_NET_CFG_UPDATE_RSS; 975 } 976 977 if (nn->dp.ctrl & NFP_NET_CFG_CTRL_IRQMOD) { 978 nfp_net_coalesce_write_cfg(nn); 979 update |= NFP_NET_CFG_UPDATE_IRQMOD; 980 } 981 982 for (r = 0; r < nn->dp.num_tx_rings; r++) 983 nfp_net_tx_ring_hw_cfg_write(nn, &nn->dp.tx_rings[r], r); 984 for (r = 0; r < nn->dp.num_rx_rings; r++) 985 nfp_net_rx_ring_hw_cfg_write(nn, &nn->dp.rx_rings[r], r); 986 987 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 988 U64_MAX >> (64 - nn->dp.num_tx_rings)); 989 990 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 991 U64_MAX >> (64 - nn->dp.num_rx_rings)); 992 993 if (nn->dp.netdev) 994 nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr); 995 996 nn_writel(nn, NFP_NET_CFG_MTU, nn->dp.mtu); 997 998 bufsz = nn->dp.fl_bufsz - nn->dp.rx_dma_off - NFP_NET_RX_BUF_NON_DATA; 999 nn_writel(nn, NFP_NET_CFG_FLBUFSZ, bufsz); 1000 1001 /* Enable device */ 1002 new_ctrl |= NFP_NET_CFG_CTRL_ENABLE; 1003 update |= NFP_NET_CFG_UPDATE_GEN; 1004 update |= NFP_NET_CFG_UPDATE_MSIX; 1005 update |= NFP_NET_CFG_UPDATE_RING; 1006 if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG) 1007 new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG; 1008 1009 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1010 nn_writel(nn, NFP_NET_CFG_CTRL_WORD1, nn->dp.ctrl_w1); 1011 err = nfp_net_reconfig(nn, update); 1012 if (err) { 1013 nfp_net_clear_config_and_disable(nn); 1014 return err; 1015 } 1016 1017 nn->dp.ctrl = new_ctrl; 1018 1019 for (r = 0; r < nn->dp.num_rx_rings; r++) 1020 nfp_net_rx_ring_fill_freelist(&nn->dp, &nn->dp.rx_rings[r]); 1021 1022 return 0; 1023 } 1024 1025 /** 1026 * nfp_net_close_stack() - Quiesce the stack (part of close) 1027 * @nn: NFP Net device to reconfigure 1028 */ 1029 static void nfp_net_close_stack(struct nfp_net *nn) 1030 { 1031 struct nfp_net_r_vector *r_vec; 1032 unsigned int r; 1033 1034 disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector); 1035 netif_carrier_off(nn->dp.netdev); 1036 nn->link_up = false; 1037 1038 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1039 r_vec = &nn->r_vecs[r]; 1040 1041 disable_irq(r_vec->irq_vector); 1042 napi_disable(&r_vec->napi); 1043 1044 if (r_vec->rx_ring) 1045 cancel_work_sync(&r_vec->rx_dim.work); 1046 1047 if (r_vec->tx_ring) 1048 cancel_work_sync(&r_vec->tx_dim.work); 1049 } 1050 1051 netif_tx_disable(nn->dp.netdev); 1052 } 1053 1054 /** 1055 * nfp_net_close_free_all() - Free all runtime resources 1056 * @nn: NFP Net device to reconfigure 1057 */ 1058 static void nfp_net_close_free_all(struct nfp_net *nn) 1059 { 1060 unsigned int r; 1061 1062 nfp_net_tx_rings_free(&nn->dp); 1063 nfp_net_rx_rings_free(&nn->dp); 1064 1065 for (r = 0; r < nn->dp.num_r_vecs; r++) 1066 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1067 1068 nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX); 1069 nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX); 1070 } 1071 1072 /** 1073 * nfp_net_netdev_close() - Called when the device is downed 1074 * @netdev: netdev structure 1075 */ 1076 static int nfp_net_netdev_close(struct net_device *netdev) 1077 { 1078 struct nfp_net *nn = netdev_priv(netdev); 1079 1080 /* Step 1: Disable RX and TX rings from the Linux kernel perspective 1081 */ 1082 nfp_net_close_stack(nn); 1083 1084 /* Step 2: Tell NFP 1085 */ 1086 nfp_net_clear_config_and_disable(nn); 1087 nfp_port_configure(netdev, false); 1088 1089 /* Step 3: Free resources 1090 */ 1091 nfp_net_close_free_all(nn); 1092 1093 nn_dbg(nn, "%s down", netdev->name); 1094 return 0; 1095 } 1096 1097 void nfp_ctrl_close(struct nfp_net *nn) 1098 { 1099 int r; 1100 1101 rtnl_lock(); 1102 1103 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1104 disable_irq(nn->r_vecs[r].irq_vector); 1105 tasklet_disable(&nn->r_vecs[r].tasklet); 1106 } 1107 1108 nfp_net_clear_config_and_disable(nn); 1109 1110 nfp_net_close_free_all(nn); 1111 1112 rtnl_unlock(); 1113 } 1114 1115 static void nfp_net_rx_dim_work(struct work_struct *work) 1116 { 1117 struct nfp_net_r_vector *r_vec; 1118 unsigned int factor, value; 1119 struct dim_cq_moder moder; 1120 struct nfp_net *nn; 1121 struct dim *dim; 1122 1123 dim = container_of(work, struct dim, work); 1124 moder = net_dim_get_rx_moderation(dim->mode, dim->profile_ix); 1125 r_vec = container_of(dim, struct nfp_net_r_vector, rx_dim); 1126 nn = r_vec->nfp_net; 1127 1128 /* Compute factor used to convert coalesce '_usecs' parameters to 1129 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp 1130 * count. 1131 */ 1132 factor = nn->tlv_caps.me_freq_mhz / 16; 1133 if (nfp_net_coalesce_para_check(factor * moder.usec, moder.pkts)) 1134 return; 1135 1136 /* copy RX interrupt coalesce parameters */ 1137 value = (moder.pkts << 16) | (factor * moder.usec); 1138 nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(r_vec->rx_ring->idx), value); 1139 (void)nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_IRQMOD); 1140 1141 dim->state = DIM_START_MEASURE; 1142 } 1143 1144 static void nfp_net_tx_dim_work(struct work_struct *work) 1145 { 1146 struct nfp_net_r_vector *r_vec; 1147 unsigned int factor, value; 1148 struct dim_cq_moder moder; 1149 struct nfp_net *nn; 1150 struct dim *dim; 1151 1152 dim = container_of(work, struct dim, work); 1153 moder = net_dim_get_tx_moderation(dim->mode, dim->profile_ix); 1154 r_vec = container_of(dim, struct nfp_net_r_vector, tx_dim); 1155 nn = r_vec->nfp_net; 1156 1157 /* Compute factor used to convert coalesce '_usecs' parameters to 1158 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp 1159 * count. 1160 */ 1161 factor = nn->tlv_caps.me_freq_mhz / 16; 1162 if (nfp_net_coalesce_para_check(factor * moder.usec, moder.pkts)) 1163 return; 1164 1165 /* copy TX interrupt coalesce parameters */ 1166 value = (moder.pkts << 16) | (factor * moder.usec); 1167 nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(r_vec->tx_ring->idx), value); 1168 (void)nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_IRQMOD); 1169 1170 dim->state = DIM_START_MEASURE; 1171 } 1172 1173 /** 1174 * nfp_net_open_stack() - Start the device from stack's perspective 1175 * @nn: NFP Net device to reconfigure 1176 */ 1177 static void nfp_net_open_stack(struct nfp_net *nn) 1178 { 1179 struct nfp_net_r_vector *r_vec; 1180 unsigned int r; 1181 1182 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1183 r_vec = &nn->r_vecs[r]; 1184 1185 if (r_vec->rx_ring) { 1186 INIT_WORK(&r_vec->rx_dim.work, nfp_net_rx_dim_work); 1187 r_vec->rx_dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE; 1188 } 1189 1190 if (r_vec->tx_ring) { 1191 INIT_WORK(&r_vec->tx_dim.work, nfp_net_tx_dim_work); 1192 r_vec->tx_dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE; 1193 } 1194 1195 napi_enable(&r_vec->napi); 1196 enable_irq(r_vec->irq_vector); 1197 } 1198 1199 netif_tx_wake_all_queues(nn->dp.netdev); 1200 1201 enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector); 1202 nfp_net_read_link_status(nn); 1203 } 1204 1205 static int nfp_net_open_alloc_all(struct nfp_net *nn) 1206 { 1207 int err, r; 1208 1209 err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn", 1210 nn->exn_name, sizeof(nn->exn_name), 1211 NFP_NET_IRQ_EXN_IDX, nn->exn_handler); 1212 if (err) 1213 return err; 1214 err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc", 1215 nn->lsc_name, sizeof(nn->lsc_name), 1216 NFP_NET_IRQ_LSC_IDX, nn->lsc_handler); 1217 if (err) 1218 goto err_free_exn; 1219 disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector); 1220 1221 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1222 err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r); 1223 if (err) 1224 goto err_cleanup_vec_p; 1225 } 1226 1227 err = nfp_net_rx_rings_prepare(nn, &nn->dp); 1228 if (err) 1229 goto err_cleanup_vec; 1230 1231 err = nfp_net_tx_rings_prepare(nn, &nn->dp); 1232 if (err) 1233 goto err_free_rx_rings; 1234 1235 for (r = 0; r < nn->max_r_vecs; r++) 1236 nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r); 1237 1238 return 0; 1239 1240 err_free_rx_rings: 1241 nfp_net_rx_rings_free(&nn->dp); 1242 err_cleanup_vec: 1243 r = nn->dp.num_r_vecs; 1244 err_cleanup_vec_p: 1245 while (r--) 1246 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1247 nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX); 1248 err_free_exn: 1249 nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX); 1250 return err; 1251 } 1252 1253 static int nfp_net_netdev_open(struct net_device *netdev) 1254 { 1255 struct nfp_net *nn = netdev_priv(netdev); 1256 int err; 1257 1258 /* Step 1: Allocate resources for rings and the like 1259 * - Request interrupts 1260 * - Allocate RX and TX ring resources 1261 * - Setup initial RSS table 1262 */ 1263 err = nfp_net_open_alloc_all(nn); 1264 if (err) 1265 return err; 1266 1267 err = netif_set_real_num_tx_queues(netdev, nn->dp.num_stack_tx_rings); 1268 if (err) 1269 goto err_free_all; 1270 1271 err = netif_set_real_num_rx_queues(netdev, nn->dp.num_rx_rings); 1272 if (err) 1273 goto err_free_all; 1274 1275 /* Step 2: Configure the NFP 1276 * - Ifup the physical interface if it exists 1277 * - Enable rings from 0 to tx_rings/rx_rings - 1. 1278 * - Write MAC address (in case it changed) 1279 * - Set the MTU 1280 * - Set the Freelist buffer size 1281 * - Enable the FW 1282 */ 1283 err = nfp_port_configure(netdev, true); 1284 if (err) 1285 goto err_free_all; 1286 1287 err = nfp_net_set_config_and_enable(nn); 1288 if (err) 1289 goto err_port_disable; 1290 1291 /* Step 3: Enable for kernel 1292 * - put some freelist descriptors on each RX ring 1293 * - enable NAPI on each ring 1294 * - enable all TX queues 1295 * - set link state 1296 */ 1297 nfp_net_open_stack(nn); 1298 1299 return 0; 1300 1301 err_port_disable: 1302 nfp_port_configure(netdev, false); 1303 err_free_all: 1304 nfp_net_close_free_all(nn); 1305 return err; 1306 } 1307 1308 int nfp_ctrl_open(struct nfp_net *nn) 1309 { 1310 int err, r; 1311 1312 /* ring dumping depends on vNICs being opened/closed under rtnl */ 1313 rtnl_lock(); 1314 1315 err = nfp_net_open_alloc_all(nn); 1316 if (err) 1317 goto err_unlock; 1318 1319 err = nfp_net_set_config_and_enable(nn); 1320 if (err) 1321 goto err_free_all; 1322 1323 for (r = 0; r < nn->dp.num_r_vecs; r++) 1324 enable_irq(nn->r_vecs[r].irq_vector); 1325 1326 rtnl_unlock(); 1327 1328 return 0; 1329 1330 err_free_all: 1331 nfp_net_close_free_all(nn); 1332 err_unlock: 1333 rtnl_unlock(); 1334 return err; 1335 } 1336 1337 struct nfp_mc_addr_entry { 1338 u8 addr[ETH_ALEN]; 1339 u32 cmd; 1340 struct list_head list; 1341 }; 1342 1343 static int nfp_net_mc_cfg(struct nfp_net *nn, const unsigned char *addr, const u32 cmd) 1344 { 1345 int ret; 1346 1347 ret = nfp_net_mbox_lock(nn, NFP_NET_CFG_MULTICAST_SZ); 1348 if (ret) 1349 return ret; 1350 1351 nn_writel(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_MULTICAST_MAC_HI, 1352 get_unaligned_be32(addr)); 1353 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_MULTICAST_MAC_LO, 1354 get_unaligned_be16(addr + 4)); 1355 1356 return nfp_net_mbox_reconfig_and_unlock(nn, cmd); 1357 } 1358 1359 static int nfp_net_mc_prep(struct nfp_net *nn, const unsigned char *addr, const u32 cmd) 1360 { 1361 struct nfp_mc_addr_entry *entry; 1362 1363 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1364 if (!entry) 1365 return -ENOMEM; 1366 1367 ether_addr_copy(entry->addr, addr); 1368 entry->cmd = cmd; 1369 spin_lock_bh(&nn->mc_lock); 1370 list_add_tail(&entry->list, &nn->mc_addrs); 1371 spin_unlock_bh(&nn->mc_lock); 1372 1373 schedule_work(&nn->mc_work); 1374 1375 return 0; 1376 } 1377 1378 static int nfp_net_mc_sync(struct net_device *netdev, const unsigned char *addr) 1379 { 1380 struct nfp_net *nn = netdev_priv(netdev); 1381 1382 if (netdev_mc_count(netdev) > NFP_NET_CFG_MAC_MC_MAX) { 1383 nn_err(nn, "Requested number of MC addresses (%d) exceeds maximum (%d).\n", 1384 netdev_mc_count(netdev), NFP_NET_CFG_MAC_MC_MAX); 1385 return -EINVAL; 1386 } 1387 1388 return nfp_net_mc_prep(nn, addr, NFP_NET_CFG_MBOX_CMD_MULTICAST_ADD); 1389 } 1390 1391 static int nfp_net_mc_unsync(struct net_device *netdev, const unsigned char *addr) 1392 { 1393 struct nfp_net *nn = netdev_priv(netdev); 1394 1395 return nfp_net_mc_prep(nn, addr, NFP_NET_CFG_MBOX_CMD_MULTICAST_DEL); 1396 } 1397 1398 static void nfp_net_mc_addr_config(struct work_struct *work) 1399 { 1400 struct nfp_net *nn = container_of(work, struct nfp_net, mc_work); 1401 struct nfp_mc_addr_entry *entry, *tmp; 1402 struct list_head tmp_list; 1403 1404 INIT_LIST_HEAD(&tmp_list); 1405 1406 spin_lock_bh(&nn->mc_lock); 1407 list_splice_init(&nn->mc_addrs, &tmp_list); 1408 spin_unlock_bh(&nn->mc_lock); 1409 1410 list_for_each_entry_safe(entry, tmp, &tmp_list, list) { 1411 if (nfp_net_mc_cfg(nn, entry->addr, entry->cmd)) 1412 nn_err(nn, "Config mc address to HW failed.\n"); 1413 1414 list_del(&entry->list); 1415 kfree(entry); 1416 } 1417 } 1418 1419 static void nfp_net_set_rx_mode(struct net_device *netdev) 1420 { 1421 struct nfp_net *nn = netdev_priv(netdev); 1422 u32 new_ctrl, new_ctrl_w1; 1423 1424 new_ctrl = nn->dp.ctrl; 1425 new_ctrl_w1 = nn->dp.ctrl_w1; 1426 1427 if (!netdev_mc_empty(netdev) || netdev->flags & IFF_ALLMULTI) 1428 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_L2MC; 1429 else 1430 new_ctrl &= ~NFP_NET_CFG_CTRL_L2MC; 1431 1432 if (netdev->flags & IFF_ALLMULTI) 1433 new_ctrl_w1 &= ~NFP_NET_CFG_CTRL_MCAST_FILTER; 1434 else 1435 new_ctrl_w1 |= nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER; 1436 1437 if (netdev->flags & IFF_PROMISC) { 1438 if (nn->cap & NFP_NET_CFG_CTRL_PROMISC) 1439 new_ctrl |= NFP_NET_CFG_CTRL_PROMISC; 1440 else 1441 nn_warn(nn, "FW does not support promiscuous mode\n"); 1442 } else { 1443 new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC; 1444 } 1445 1446 if ((nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER) && 1447 __dev_mc_sync(netdev, nfp_net_mc_sync, nfp_net_mc_unsync)) 1448 netdev_err(netdev, "Sync mc address failed\n"); 1449 1450 if (new_ctrl == nn->dp.ctrl && new_ctrl_w1 == nn->dp.ctrl_w1) 1451 return; 1452 1453 if (new_ctrl != nn->dp.ctrl) 1454 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1455 if (new_ctrl_w1 != nn->dp.ctrl_w1) 1456 nn_writel(nn, NFP_NET_CFG_CTRL_WORD1, new_ctrl_w1); 1457 nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN); 1458 1459 nn->dp.ctrl = new_ctrl; 1460 nn->dp.ctrl_w1 = new_ctrl_w1; 1461 } 1462 1463 static void nfp_net_rss_init_itbl(struct nfp_net *nn) 1464 { 1465 int i; 1466 1467 for (i = 0; i < sizeof(nn->rss_itbl); i++) 1468 nn->rss_itbl[i] = 1469 ethtool_rxfh_indir_default(i, nn->dp.num_rx_rings); 1470 } 1471 1472 static void nfp_net_dp_swap(struct nfp_net *nn, struct nfp_net_dp *dp) 1473 { 1474 struct nfp_net_dp new_dp = *dp; 1475 1476 *dp = nn->dp; 1477 nn->dp = new_dp; 1478 1479 nn->dp.netdev->mtu = new_dp.mtu; 1480 1481 if (!netif_is_rxfh_configured(nn->dp.netdev)) 1482 nfp_net_rss_init_itbl(nn); 1483 } 1484 1485 static int nfp_net_dp_swap_enable(struct nfp_net *nn, struct nfp_net_dp *dp) 1486 { 1487 unsigned int r; 1488 int err; 1489 1490 nfp_net_dp_swap(nn, dp); 1491 1492 for (r = 0; r < nn->max_r_vecs; r++) 1493 nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r); 1494 1495 err = netif_set_real_num_queues(nn->dp.netdev, 1496 nn->dp.num_stack_tx_rings, 1497 nn->dp.num_rx_rings); 1498 if (err) 1499 return err; 1500 1501 return nfp_net_set_config_and_enable(nn); 1502 } 1503 1504 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn) 1505 { 1506 struct nfp_net_dp *new; 1507 1508 new = kmalloc(sizeof(*new), GFP_KERNEL); 1509 if (!new) 1510 return NULL; 1511 1512 *new = nn->dp; 1513 1514 new->xsk_pools = kmemdup(new->xsk_pools, 1515 array_size(nn->max_r_vecs, 1516 sizeof(new->xsk_pools)), 1517 GFP_KERNEL); 1518 if (!new->xsk_pools) { 1519 kfree(new); 1520 return NULL; 1521 } 1522 1523 /* Clear things which need to be recomputed */ 1524 new->fl_bufsz = 0; 1525 new->tx_rings = NULL; 1526 new->rx_rings = NULL; 1527 new->num_r_vecs = 0; 1528 new->num_stack_tx_rings = 0; 1529 new->txrwb = NULL; 1530 new->txrwb_dma = 0; 1531 1532 return new; 1533 } 1534 1535 static void nfp_net_free_dp(struct nfp_net_dp *dp) 1536 { 1537 kfree(dp->xsk_pools); 1538 kfree(dp); 1539 } 1540 1541 static int 1542 nfp_net_check_config(struct nfp_net *nn, struct nfp_net_dp *dp, 1543 struct netlink_ext_ack *extack) 1544 { 1545 unsigned int r, xsk_min_fl_bufsz; 1546 1547 /* XDP-enabled tests */ 1548 if (!dp->xdp_prog) 1549 return 0; 1550 if (dp->fl_bufsz > PAGE_SIZE) { 1551 NL_SET_ERR_MSG_MOD(extack, "MTU too large w/ XDP enabled"); 1552 return -EINVAL; 1553 } 1554 if (dp->num_tx_rings > nn->max_tx_rings) { 1555 NL_SET_ERR_MSG_MOD(extack, "Insufficient number of TX rings w/ XDP enabled"); 1556 return -EINVAL; 1557 } 1558 1559 xsk_min_fl_bufsz = nfp_net_calc_fl_bufsz_xsk(dp); 1560 for (r = 0; r < nn->max_r_vecs; r++) { 1561 if (!dp->xsk_pools[r]) 1562 continue; 1563 1564 if (xsk_pool_get_rx_frame_size(dp->xsk_pools[r]) < xsk_min_fl_bufsz) { 1565 NL_SET_ERR_MSG_MOD(extack, 1566 "XSK buffer pool chunk size too small"); 1567 return -EINVAL; 1568 } 1569 } 1570 1571 return 0; 1572 } 1573 1574 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *dp, 1575 struct netlink_ext_ack *extack) 1576 { 1577 int r, err; 1578 1579 dp->fl_bufsz = nfp_net_calc_fl_bufsz(dp); 1580 1581 dp->num_stack_tx_rings = dp->num_tx_rings; 1582 if (dp->xdp_prog) 1583 dp->num_stack_tx_rings -= dp->num_rx_rings; 1584 1585 dp->num_r_vecs = max(dp->num_rx_rings, dp->num_stack_tx_rings); 1586 1587 err = nfp_net_check_config(nn, dp, extack); 1588 if (err) 1589 goto exit_free_dp; 1590 1591 if (!netif_running(dp->netdev)) { 1592 nfp_net_dp_swap(nn, dp); 1593 err = 0; 1594 goto exit_free_dp; 1595 } 1596 1597 /* Prepare new rings */ 1598 for (r = nn->dp.num_r_vecs; r < dp->num_r_vecs; r++) { 1599 err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r); 1600 if (err) { 1601 dp->num_r_vecs = r; 1602 goto err_cleanup_vecs; 1603 } 1604 } 1605 1606 err = nfp_net_rx_rings_prepare(nn, dp); 1607 if (err) 1608 goto err_cleanup_vecs; 1609 1610 err = nfp_net_tx_rings_prepare(nn, dp); 1611 if (err) 1612 goto err_free_rx; 1613 1614 /* Stop device, swap in new rings, try to start the firmware */ 1615 nfp_net_close_stack(nn); 1616 nfp_net_clear_config_and_disable(nn); 1617 1618 err = nfp_net_dp_swap_enable(nn, dp); 1619 if (err) { 1620 int err2; 1621 1622 nfp_net_clear_config_and_disable(nn); 1623 1624 /* Try with old configuration and old rings */ 1625 err2 = nfp_net_dp_swap_enable(nn, dp); 1626 if (err2) 1627 nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n", 1628 err, err2); 1629 } 1630 for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--) 1631 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1632 1633 nfp_net_rx_rings_free(dp); 1634 nfp_net_tx_rings_free(dp); 1635 1636 nfp_net_open_stack(nn); 1637 exit_free_dp: 1638 nfp_net_free_dp(dp); 1639 1640 return err; 1641 1642 err_free_rx: 1643 nfp_net_rx_rings_free(dp); 1644 err_cleanup_vecs: 1645 for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--) 1646 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1647 nfp_net_free_dp(dp); 1648 return err; 1649 } 1650 1651 static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu) 1652 { 1653 struct nfp_net *nn = netdev_priv(netdev); 1654 struct nfp_net_dp *dp; 1655 int err; 1656 1657 err = nfp_app_check_mtu(nn->app, netdev, new_mtu); 1658 if (err) 1659 return err; 1660 1661 dp = nfp_net_clone_dp(nn); 1662 if (!dp) 1663 return -ENOMEM; 1664 1665 dp->mtu = new_mtu; 1666 1667 return nfp_net_ring_reconfig(nn, dp, NULL); 1668 } 1669 1670 static int 1671 nfp_net_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid) 1672 { 1673 const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_ADD; 1674 struct nfp_net *nn = netdev_priv(netdev); 1675 int err; 1676 1677 /* Priority tagged packets with vlan id 0 are processed by the 1678 * NFP as untagged packets 1679 */ 1680 if (!vid) 1681 return 0; 1682 1683 err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ); 1684 if (err) 1685 return err; 1686 1687 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid); 1688 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO, 1689 ETH_P_8021Q); 1690 1691 return nfp_net_mbox_reconfig_and_unlock(nn, cmd); 1692 } 1693 1694 static int 1695 nfp_net_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid) 1696 { 1697 const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_KILL; 1698 struct nfp_net *nn = netdev_priv(netdev); 1699 int err; 1700 1701 /* Priority tagged packets with vlan id 0 are processed by the 1702 * NFP as untagged packets 1703 */ 1704 if (!vid) 1705 return 0; 1706 1707 err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ); 1708 if (err) 1709 return err; 1710 1711 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid); 1712 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO, 1713 ETH_P_8021Q); 1714 1715 return nfp_net_mbox_reconfig_and_unlock(nn, cmd); 1716 } 1717 1718 static void nfp_net_stat64(struct net_device *netdev, 1719 struct rtnl_link_stats64 *stats) 1720 { 1721 struct nfp_net *nn = netdev_priv(netdev); 1722 int r; 1723 1724 /* Collect software stats */ 1725 for (r = 0; r < nn->max_r_vecs; r++) { 1726 struct nfp_net_r_vector *r_vec = &nn->r_vecs[r]; 1727 u64 data[3]; 1728 unsigned int start; 1729 1730 do { 1731 start = u64_stats_fetch_begin(&r_vec->rx_sync); 1732 data[0] = r_vec->rx_pkts; 1733 data[1] = r_vec->rx_bytes; 1734 data[2] = r_vec->rx_drops; 1735 } while (u64_stats_fetch_retry(&r_vec->rx_sync, start)); 1736 stats->rx_packets += data[0]; 1737 stats->rx_bytes += data[1]; 1738 stats->rx_dropped += data[2]; 1739 1740 do { 1741 start = u64_stats_fetch_begin(&r_vec->tx_sync); 1742 data[0] = r_vec->tx_pkts; 1743 data[1] = r_vec->tx_bytes; 1744 data[2] = r_vec->tx_errors; 1745 } while (u64_stats_fetch_retry(&r_vec->tx_sync, start)); 1746 stats->tx_packets += data[0]; 1747 stats->tx_bytes += data[1]; 1748 stats->tx_errors += data[2]; 1749 } 1750 1751 /* Add in device stats */ 1752 stats->multicast += nn_readq(nn, NFP_NET_CFG_STATS_RX_MC_FRAMES); 1753 stats->rx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_RX_DISCARDS); 1754 stats->rx_errors += nn_readq(nn, NFP_NET_CFG_STATS_RX_ERRORS); 1755 1756 stats->tx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_TX_DISCARDS); 1757 stats->tx_errors += nn_readq(nn, NFP_NET_CFG_STATS_TX_ERRORS); 1758 } 1759 1760 static int nfp_net_set_features(struct net_device *netdev, 1761 netdev_features_t features) 1762 { 1763 netdev_features_t changed = netdev->features ^ features; 1764 struct nfp_net *nn = netdev_priv(netdev); 1765 u32 new_ctrl; 1766 int err; 1767 1768 /* Assume this is not called with features we have not advertised */ 1769 1770 new_ctrl = nn->dp.ctrl; 1771 1772 if (changed & NETIF_F_RXCSUM) { 1773 if (features & NETIF_F_RXCSUM) 1774 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY; 1775 else 1776 new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM_ANY; 1777 } 1778 1779 if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 1780 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) 1781 new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM; 1782 else 1783 new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM; 1784 } 1785 1786 if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) { 1787 if (features & (NETIF_F_TSO | NETIF_F_TSO6)) 1788 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?: 1789 NFP_NET_CFG_CTRL_LSO; 1790 else 1791 new_ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY; 1792 } 1793 1794 if (changed & NETIF_F_HW_VLAN_CTAG_RX) { 1795 if (features & NETIF_F_HW_VLAN_CTAG_RX) 1796 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ?: 1797 NFP_NET_CFG_CTRL_RXVLAN; 1798 else 1799 new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN_ANY; 1800 } 1801 1802 if (changed & NETIF_F_HW_VLAN_CTAG_TX) { 1803 if (features & NETIF_F_HW_VLAN_CTAG_TX) 1804 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ?: 1805 NFP_NET_CFG_CTRL_TXVLAN; 1806 else 1807 new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN_ANY; 1808 } 1809 1810 if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) { 1811 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) 1812 new_ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER; 1813 else 1814 new_ctrl &= ~NFP_NET_CFG_CTRL_CTAG_FILTER; 1815 } 1816 1817 if (changed & NETIF_F_HW_VLAN_STAG_RX) { 1818 if (features & NETIF_F_HW_VLAN_STAG_RX) 1819 new_ctrl |= NFP_NET_CFG_CTRL_RXQINQ; 1820 else 1821 new_ctrl &= ~NFP_NET_CFG_CTRL_RXQINQ; 1822 } 1823 1824 if (changed & NETIF_F_SG) { 1825 if (features & NETIF_F_SG) 1826 new_ctrl |= NFP_NET_CFG_CTRL_GATHER; 1827 else 1828 new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER; 1829 } 1830 1831 err = nfp_port_set_features(netdev, features); 1832 if (err) 1833 return err; 1834 1835 nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n", 1836 netdev->features, features, changed); 1837 1838 if (new_ctrl == nn->dp.ctrl) 1839 return 0; 1840 1841 nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->dp.ctrl, new_ctrl); 1842 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1843 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN); 1844 if (err) 1845 return err; 1846 1847 nn->dp.ctrl = new_ctrl; 1848 1849 return 0; 1850 } 1851 1852 static netdev_features_t 1853 nfp_net_fix_features(struct net_device *netdev, 1854 netdev_features_t features) 1855 { 1856 if ((features & NETIF_F_HW_VLAN_CTAG_RX) && 1857 (features & NETIF_F_HW_VLAN_STAG_RX)) { 1858 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) { 1859 features &= ~NETIF_F_HW_VLAN_CTAG_RX; 1860 netdev->wanted_features &= ~NETIF_F_HW_VLAN_CTAG_RX; 1861 netdev_warn(netdev, 1862 "S-tag and C-tag stripping can't be enabled at the same time. Enabling S-tag stripping and disabling C-tag stripping\n"); 1863 } else if (netdev->features & NETIF_F_HW_VLAN_STAG_RX) { 1864 features &= ~NETIF_F_HW_VLAN_STAG_RX; 1865 netdev->wanted_features &= ~NETIF_F_HW_VLAN_STAG_RX; 1866 netdev_warn(netdev, 1867 "S-tag and C-tag stripping can't be enabled at the same time. Enabling C-tag stripping and disabling S-tag stripping\n"); 1868 } 1869 } 1870 return features; 1871 } 1872 1873 static netdev_features_t 1874 nfp_net_features_check(struct sk_buff *skb, struct net_device *dev, 1875 netdev_features_t features) 1876 { 1877 u8 l4_hdr; 1878 1879 /* We can't do TSO over double tagged packets (802.1AD) */ 1880 features &= vlan_features_check(skb, features); 1881 1882 if (!skb->encapsulation) 1883 return features; 1884 1885 /* Ensure that inner L4 header offset fits into TX descriptor field */ 1886 if (skb_is_gso(skb)) { 1887 u32 hdrlen; 1888 1889 hdrlen = skb_inner_tcp_all_headers(skb); 1890 1891 /* Assume worst case scenario of having longest possible 1892 * metadata prepend - 8B 1893 */ 1894 if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ - 8)) 1895 features &= ~NETIF_F_GSO_MASK; 1896 } 1897 1898 /* VXLAN/GRE check */ 1899 switch (vlan_get_protocol(skb)) { 1900 case htons(ETH_P_IP): 1901 l4_hdr = ip_hdr(skb)->protocol; 1902 break; 1903 case htons(ETH_P_IPV6): 1904 l4_hdr = ipv6_hdr(skb)->nexthdr; 1905 break; 1906 default: 1907 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 1908 } 1909 1910 if (skb->inner_protocol_type != ENCAP_TYPE_ETHER || 1911 skb->inner_protocol != htons(ETH_P_TEB) || 1912 (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) || 1913 (l4_hdr == IPPROTO_UDP && 1914 (skb_inner_mac_header(skb) - skb_transport_header(skb) != 1915 sizeof(struct udphdr) + sizeof(struct vxlanhdr)))) 1916 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 1917 1918 return features; 1919 } 1920 1921 static int 1922 nfp_net_get_phys_port_name(struct net_device *netdev, char *name, size_t len) 1923 { 1924 struct nfp_net *nn = netdev_priv(netdev); 1925 int n; 1926 1927 /* If port is defined, devlink_port is registered and devlink core 1928 * is taking care of name formatting. 1929 */ 1930 if (nn->port) 1931 return -EOPNOTSUPP; 1932 1933 if (nn->dp.is_vf || nn->vnic_no_name) 1934 return -EOPNOTSUPP; 1935 1936 n = snprintf(name, len, "n%d", nn->id); 1937 if (n >= len) 1938 return -EINVAL; 1939 1940 return 0; 1941 } 1942 1943 static int nfp_net_xdp_setup_drv(struct nfp_net *nn, struct netdev_bpf *bpf) 1944 { 1945 struct bpf_prog *prog = bpf->prog; 1946 struct nfp_net_dp *dp; 1947 int err; 1948 1949 if (!prog == !nn->dp.xdp_prog) { 1950 WRITE_ONCE(nn->dp.xdp_prog, prog); 1951 xdp_attachment_setup(&nn->xdp, bpf); 1952 return 0; 1953 } 1954 1955 dp = nfp_net_clone_dp(nn); 1956 if (!dp) 1957 return -ENOMEM; 1958 1959 dp->xdp_prog = prog; 1960 dp->num_tx_rings += prog ? nn->dp.num_rx_rings : -nn->dp.num_rx_rings; 1961 dp->rx_dma_dir = prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE; 1962 dp->rx_dma_off = prog ? XDP_PACKET_HEADROOM - nn->dp.rx_offset : 0; 1963 1964 /* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */ 1965 err = nfp_net_ring_reconfig(nn, dp, bpf->extack); 1966 if (err) 1967 return err; 1968 1969 xdp_attachment_setup(&nn->xdp, bpf); 1970 return 0; 1971 } 1972 1973 static int nfp_net_xdp_setup_hw(struct nfp_net *nn, struct netdev_bpf *bpf) 1974 { 1975 int err; 1976 1977 err = nfp_app_xdp_offload(nn->app, nn, bpf->prog, bpf->extack); 1978 if (err) 1979 return err; 1980 1981 xdp_attachment_setup(&nn->xdp_hw, bpf); 1982 return 0; 1983 } 1984 1985 static int nfp_net_xdp(struct net_device *netdev, struct netdev_bpf *xdp) 1986 { 1987 struct nfp_net *nn = netdev_priv(netdev); 1988 1989 switch (xdp->command) { 1990 case XDP_SETUP_PROG: 1991 return nfp_net_xdp_setup_drv(nn, xdp); 1992 case XDP_SETUP_PROG_HW: 1993 return nfp_net_xdp_setup_hw(nn, xdp); 1994 case XDP_SETUP_XSK_POOL: 1995 return nfp_net_xsk_setup_pool(netdev, xdp->xsk.pool, 1996 xdp->xsk.queue_id); 1997 default: 1998 return nfp_app_bpf(nn->app, nn, xdp); 1999 } 2000 } 2001 2002 static int nfp_net_set_mac_address(struct net_device *netdev, void *addr) 2003 { 2004 struct nfp_net *nn = netdev_priv(netdev); 2005 struct sockaddr *saddr = addr; 2006 int err; 2007 2008 err = eth_prepare_mac_addr_change(netdev, addr); 2009 if (err) 2010 return err; 2011 2012 nfp_net_write_mac_addr(nn, saddr->sa_data); 2013 2014 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MACADDR); 2015 if (err) 2016 return err; 2017 2018 eth_commit_mac_addr_change(netdev, addr); 2019 2020 return 0; 2021 } 2022 2023 static int nfp_net_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq, 2024 struct net_device *dev, u32 filter_mask, 2025 int nlflags) 2026 { 2027 struct nfp_net *nn = netdev_priv(dev); 2028 u16 mode; 2029 2030 if (!(nn->cap & NFP_NET_CFG_CTRL_VEPA)) 2031 return -EOPNOTSUPP; 2032 2033 mode = (nn->dp.ctrl & NFP_NET_CFG_CTRL_VEPA) ? 2034 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB; 2035 2036 return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode, 0, 0, 2037 nlflags, filter_mask, NULL); 2038 } 2039 2040 static int nfp_net_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh, 2041 u16 flags, struct netlink_ext_ack *extack) 2042 { 2043 struct nfp_net *nn = netdev_priv(dev); 2044 struct nlattr *attr, *br_spec; 2045 int rem, err; 2046 u32 new_ctrl; 2047 u16 mode; 2048 2049 if (!(nn->cap & NFP_NET_CFG_CTRL_VEPA)) 2050 return -EOPNOTSUPP; 2051 2052 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC); 2053 if (!br_spec) 2054 return -EINVAL; 2055 2056 nla_for_each_nested(attr, br_spec, rem) { 2057 if (nla_type(attr) != IFLA_BRIDGE_MODE) 2058 continue; 2059 2060 if (nla_len(attr) < sizeof(mode)) 2061 return -EINVAL; 2062 2063 new_ctrl = nn->dp.ctrl; 2064 mode = nla_get_u16(attr); 2065 if (mode == BRIDGE_MODE_VEPA) 2066 new_ctrl |= NFP_NET_CFG_CTRL_VEPA; 2067 else if (mode == BRIDGE_MODE_VEB) 2068 new_ctrl &= ~NFP_NET_CFG_CTRL_VEPA; 2069 else 2070 return -EOPNOTSUPP; 2071 2072 if (new_ctrl == nn->dp.ctrl) 2073 return 0; 2074 2075 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 2076 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN); 2077 if (!err) 2078 nn->dp.ctrl = new_ctrl; 2079 2080 return err; 2081 } 2082 2083 return -EINVAL; 2084 } 2085 2086 const struct net_device_ops nfp_nfd3_netdev_ops = { 2087 .ndo_init = nfp_app_ndo_init, 2088 .ndo_uninit = nfp_app_ndo_uninit, 2089 .ndo_open = nfp_net_netdev_open, 2090 .ndo_stop = nfp_net_netdev_close, 2091 .ndo_start_xmit = nfp_net_tx, 2092 .ndo_get_stats64 = nfp_net_stat64, 2093 .ndo_vlan_rx_add_vid = nfp_net_vlan_rx_add_vid, 2094 .ndo_vlan_rx_kill_vid = nfp_net_vlan_rx_kill_vid, 2095 .ndo_set_vf_mac = nfp_app_set_vf_mac, 2096 .ndo_set_vf_vlan = nfp_app_set_vf_vlan, 2097 .ndo_set_vf_rate = nfp_app_set_vf_rate, 2098 .ndo_set_vf_spoofchk = nfp_app_set_vf_spoofchk, 2099 .ndo_set_vf_trust = nfp_app_set_vf_trust, 2100 .ndo_get_vf_config = nfp_app_get_vf_config, 2101 .ndo_set_vf_link_state = nfp_app_set_vf_link_state, 2102 .ndo_setup_tc = nfp_port_setup_tc, 2103 .ndo_tx_timeout = nfp_net_tx_timeout, 2104 .ndo_set_rx_mode = nfp_net_set_rx_mode, 2105 .ndo_change_mtu = nfp_net_change_mtu, 2106 .ndo_set_mac_address = nfp_net_set_mac_address, 2107 .ndo_set_features = nfp_net_set_features, 2108 .ndo_fix_features = nfp_net_fix_features, 2109 .ndo_features_check = nfp_net_features_check, 2110 .ndo_get_phys_port_name = nfp_net_get_phys_port_name, 2111 .ndo_bpf = nfp_net_xdp, 2112 .ndo_xsk_wakeup = nfp_net_xsk_wakeup, 2113 .ndo_bridge_getlink = nfp_net_bridge_getlink, 2114 .ndo_bridge_setlink = nfp_net_bridge_setlink, 2115 }; 2116 2117 const struct net_device_ops nfp_nfdk_netdev_ops = { 2118 .ndo_init = nfp_app_ndo_init, 2119 .ndo_uninit = nfp_app_ndo_uninit, 2120 .ndo_open = nfp_net_netdev_open, 2121 .ndo_stop = nfp_net_netdev_close, 2122 .ndo_start_xmit = nfp_net_tx, 2123 .ndo_get_stats64 = nfp_net_stat64, 2124 .ndo_vlan_rx_add_vid = nfp_net_vlan_rx_add_vid, 2125 .ndo_vlan_rx_kill_vid = nfp_net_vlan_rx_kill_vid, 2126 .ndo_set_vf_mac = nfp_app_set_vf_mac, 2127 .ndo_set_vf_vlan = nfp_app_set_vf_vlan, 2128 .ndo_set_vf_rate = nfp_app_set_vf_rate, 2129 .ndo_set_vf_spoofchk = nfp_app_set_vf_spoofchk, 2130 .ndo_set_vf_trust = nfp_app_set_vf_trust, 2131 .ndo_get_vf_config = nfp_app_get_vf_config, 2132 .ndo_set_vf_link_state = nfp_app_set_vf_link_state, 2133 .ndo_setup_tc = nfp_port_setup_tc, 2134 .ndo_tx_timeout = nfp_net_tx_timeout, 2135 .ndo_set_rx_mode = nfp_net_set_rx_mode, 2136 .ndo_change_mtu = nfp_net_change_mtu, 2137 .ndo_set_mac_address = nfp_net_set_mac_address, 2138 .ndo_set_features = nfp_net_set_features, 2139 .ndo_fix_features = nfp_net_fix_features, 2140 .ndo_features_check = nfp_net_features_check, 2141 .ndo_get_phys_port_name = nfp_net_get_phys_port_name, 2142 .ndo_bpf = nfp_net_xdp, 2143 .ndo_bridge_getlink = nfp_net_bridge_getlink, 2144 .ndo_bridge_setlink = nfp_net_bridge_setlink, 2145 }; 2146 2147 static int nfp_udp_tunnel_sync(struct net_device *netdev, unsigned int table) 2148 { 2149 struct nfp_net *nn = netdev_priv(netdev); 2150 int i; 2151 2152 BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1); 2153 for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2) { 2154 struct udp_tunnel_info ti0, ti1; 2155 2156 udp_tunnel_nic_get_port(netdev, table, i, &ti0); 2157 udp_tunnel_nic_get_port(netdev, table, i + 1, &ti1); 2158 2159 nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(ti0.port), 2160 be16_to_cpu(ti1.port) << 16 | be16_to_cpu(ti0.port)); 2161 } 2162 2163 return nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_VXLAN); 2164 } 2165 2166 static const struct udp_tunnel_nic_info nfp_udp_tunnels = { 2167 .sync_table = nfp_udp_tunnel_sync, 2168 .flags = UDP_TUNNEL_NIC_INFO_MAY_SLEEP | 2169 UDP_TUNNEL_NIC_INFO_OPEN_ONLY, 2170 .tables = { 2171 { 2172 .n_entries = NFP_NET_N_VXLAN_PORTS, 2173 .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, 2174 }, 2175 }, 2176 }; 2177 2178 /** 2179 * nfp_net_info() - Print general info about the NIC 2180 * @nn: NFP Net device to reconfigure 2181 */ 2182 void nfp_net_info(struct nfp_net *nn) 2183 { 2184 nn_info(nn, "NFP-6xxx %sNetdev: TxQs=%d/%d RxQs=%d/%d\n", 2185 nn->dp.is_vf ? "VF " : "", 2186 nn->dp.num_tx_rings, nn->max_tx_rings, 2187 nn->dp.num_rx_rings, nn->max_rx_rings); 2188 nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n", 2189 nn->fw_ver.extend, nn->fw_ver.class, 2190 nn->fw_ver.major, nn->fw_ver.minor, 2191 nn->max_mtu); 2192 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%s\n", 2193 nn->cap, 2194 nn->cap & NFP_NET_CFG_CTRL_PROMISC ? "PROMISC " : "", 2195 nn->cap & NFP_NET_CFG_CTRL_L2BC ? "L2BCFILT " : "", 2196 nn->cap & NFP_NET_CFG_CTRL_L2MC ? "L2MCFILT " : "", 2197 nn->cap & NFP_NET_CFG_CTRL_RXCSUM ? "RXCSUM " : "", 2198 nn->cap & NFP_NET_CFG_CTRL_TXCSUM ? "TXCSUM " : "", 2199 nn->cap & NFP_NET_CFG_CTRL_RXVLAN ? "RXVLAN " : "", 2200 nn->cap & NFP_NET_CFG_CTRL_TXVLAN ? "TXVLAN " : "", 2201 nn->cap & NFP_NET_CFG_CTRL_RXQINQ ? "RXQINQ " : "", 2202 nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ? "RXVLANv2 " : "", 2203 nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ? "TXVLANv2 " : "", 2204 nn->cap & NFP_NET_CFG_CTRL_SCATTER ? "SCATTER " : "", 2205 nn->cap & NFP_NET_CFG_CTRL_GATHER ? "GATHER " : "", 2206 nn->cap & NFP_NET_CFG_CTRL_LSO ? "TSO1 " : "", 2207 nn->cap & NFP_NET_CFG_CTRL_LSO2 ? "TSO2 " : "", 2208 nn->cap & NFP_NET_CFG_CTRL_RSS ? "RSS1 " : "", 2209 nn->cap & NFP_NET_CFG_CTRL_RSS2 ? "RSS2 " : "", 2210 nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER ? "CTAG_FILTER " : "", 2211 nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "", 2212 nn->cap & NFP_NET_CFG_CTRL_IRQMOD ? "IRQMOD " : "", 2213 nn->cap & NFP_NET_CFG_CTRL_TXRWB ? "TXRWB " : "", 2214 nn->cap & NFP_NET_CFG_CTRL_VEPA ? "VEPA " : "", 2215 nn->cap & NFP_NET_CFG_CTRL_VXLAN ? "VXLAN " : "", 2216 nn->cap & NFP_NET_CFG_CTRL_NVGRE ? "NVGRE " : "", 2217 nn->cap & NFP_NET_CFG_CTRL_CSUM_COMPLETE ? 2218 "RXCSUM_COMPLETE " : "", 2219 nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR ? "LIVE_ADDR " : "", 2220 nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER ? "MULTICAST_FILTER " : "", 2221 nfp_app_extra_cap(nn->app, nn)); 2222 } 2223 2224 /** 2225 * nfp_net_alloc() - Allocate netdev and related structure 2226 * @pdev: PCI device 2227 * @dev_info: NFP ASIC params 2228 * @ctrl_bar: PCI IOMEM with vNIC config memory 2229 * @needs_netdev: Whether to allocate a netdev for this vNIC 2230 * @max_tx_rings: Maximum number of TX rings supported by device 2231 * @max_rx_rings: Maximum number of RX rings supported by device 2232 * 2233 * This function allocates a netdev device and fills in the initial 2234 * part of the @struct nfp_net structure. In case of control device 2235 * nfp_net structure is allocated without the netdev. 2236 * 2237 * Return: NFP Net device structure, or ERR_PTR on error. 2238 */ 2239 struct nfp_net * 2240 nfp_net_alloc(struct pci_dev *pdev, const struct nfp_dev_info *dev_info, 2241 void __iomem *ctrl_bar, bool needs_netdev, 2242 unsigned int max_tx_rings, unsigned int max_rx_rings) 2243 { 2244 u64 dma_mask = dma_get_mask(&pdev->dev); 2245 struct nfp_net *nn; 2246 int err; 2247 2248 if (needs_netdev) { 2249 struct net_device *netdev; 2250 2251 netdev = alloc_etherdev_mqs(sizeof(struct nfp_net), 2252 max_tx_rings, max_rx_rings); 2253 if (!netdev) 2254 return ERR_PTR(-ENOMEM); 2255 2256 SET_NETDEV_DEV(netdev, &pdev->dev); 2257 nn = netdev_priv(netdev); 2258 nn->dp.netdev = netdev; 2259 } else { 2260 nn = vzalloc(sizeof(*nn)); 2261 if (!nn) 2262 return ERR_PTR(-ENOMEM); 2263 } 2264 2265 nn->dp.dev = &pdev->dev; 2266 nn->dp.ctrl_bar = ctrl_bar; 2267 nn->dev_info = dev_info; 2268 nn->pdev = pdev; 2269 nfp_net_get_fw_version(&nn->fw_ver, ctrl_bar); 2270 2271 switch (FIELD_GET(NFP_NET_CFG_VERSION_DP_MASK, nn->fw_ver.extend)) { 2272 case NFP_NET_CFG_VERSION_DP_NFD3: 2273 nn->dp.ops = &nfp_nfd3_ops; 2274 break; 2275 case NFP_NET_CFG_VERSION_DP_NFDK: 2276 if (nn->fw_ver.major < 5) { 2277 dev_err(&pdev->dev, 2278 "NFDK must use ABI 5 or newer, found: %d\n", 2279 nn->fw_ver.major); 2280 err = -EINVAL; 2281 goto err_free_nn; 2282 } 2283 nn->dp.ops = &nfp_nfdk_ops; 2284 break; 2285 default: 2286 err = -EINVAL; 2287 goto err_free_nn; 2288 } 2289 2290 if ((dma_mask & nn->dp.ops->dma_mask) != dma_mask) { 2291 dev_err(&pdev->dev, 2292 "DMA mask of loaded firmware: %llx, required DMA mask: %llx\n", 2293 nn->dp.ops->dma_mask, dma_mask); 2294 err = -EINVAL; 2295 goto err_free_nn; 2296 } 2297 2298 nn->max_tx_rings = max_tx_rings; 2299 nn->max_rx_rings = max_rx_rings; 2300 2301 nn->dp.num_tx_rings = min_t(unsigned int, 2302 max_tx_rings, num_online_cpus()); 2303 nn->dp.num_rx_rings = min_t(unsigned int, max_rx_rings, 2304 netif_get_num_default_rss_queues()); 2305 2306 nn->dp.num_r_vecs = max(nn->dp.num_tx_rings, nn->dp.num_rx_rings); 2307 nn->dp.num_r_vecs = min_t(unsigned int, 2308 nn->dp.num_r_vecs, num_online_cpus()); 2309 nn->max_r_vecs = nn->dp.num_r_vecs; 2310 2311 nn->dp.xsk_pools = kcalloc(nn->max_r_vecs, sizeof(nn->dp.xsk_pools), 2312 GFP_KERNEL); 2313 if (!nn->dp.xsk_pools) { 2314 err = -ENOMEM; 2315 goto err_free_nn; 2316 } 2317 2318 nn->dp.txd_cnt = NFP_NET_TX_DESCS_DEFAULT; 2319 nn->dp.rxd_cnt = NFP_NET_RX_DESCS_DEFAULT; 2320 2321 sema_init(&nn->bar_lock, 1); 2322 2323 spin_lock_init(&nn->reconfig_lock); 2324 spin_lock_init(&nn->link_status_lock); 2325 2326 timer_setup(&nn->reconfig_timer, nfp_net_reconfig_timer, 0); 2327 2328 err = nfp_net_tlv_caps_parse(&nn->pdev->dev, nn->dp.ctrl_bar, 2329 &nn->tlv_caps); 2330 if (err) 2331 goto err_free_nn; 2332 2333 err = nfp_ccm_mbox_alloc(nn); 2334 if (err) 2335 goto err_free_nn; 2336 2337 return nn; 2338 2339 err_free_nn: 2340 if (nn->dp.netdev) 2341 free_netdev(nn->dp.netdev); 2342 else 2343 vfree(nn); 2344 return ERR_PTR(err); 2345 } 2346 2347 /** 2348 * nfp_net_free() - Undo what @nfp_net_alloc() did 2349 * @nn: NFP Net device to reconfigure 2350 */ 2351 void nfp_net_free(struct nfp_net *nn) 2352 { 2353 WARN_ON(timer_pending(&nn->reconfig_timer) || nn->reconfig_posted); 2354 nfp_ccm_mbox_free(nn); 2355 2356 kfree(nn->dp.xsk_pools); 2357 if (nn->dp.netdev) 2358 free_netdev(nn->dp.netdev); 2359 else 2360 vfree(nn); 2361 } 2362 2363 /** 2364 * nfp_net_rss_key_sz() - Get current size of the RSS key 2365 * @nn: NFP Net device instance 2366 * 2367 * Return: size of the RSS key for currently selected hash function. 2368 */ 2369 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn) 2370 { 2371 switch (nn->rss_hfunc) { 2372 case ETH_RSS_HASH_TOP: 2373 return NFP_NET_CFG_RSS_KEY_SZ; 2374 case ETH_RSS_HASH_XOR: 2375 return 0; 2376 case ETH_RSS_HASH_CRC32: 2377 return 4; 2378 } 2379 2380 nn_warn(nn, "Unknown hash function: %u\n", nn->rss_hfunc); 2381 return 0; 2382 } 2383 2384 /** 2385 * nfp_net_rss_init() - Set the initial RSS parameters 2386 * @nn: NFP Net device to reconfigure 2387 */ 2388 static void nfp_net_rss_init(struct nfp_net *nn) 2389 { 2390 unsigned long func_bit, rss_cap_hfunc; 2391 u32 reg; 2392 2393 /* Read the RSS function capability and select first supported func */ 2394 reg = nn_readl(nn, NFP_NET_CFG_RSS_CAP); 2395 rss_cap_hfunc = FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, reg); 2396 if (!rss_cap_hfunc) 2397 rss_cap_hfunc = FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, 2398 NFP_NET_CFG_RSS_TOEPLITZ); 2399 2400 func_bit = find_first_bit(&rss_cap_hfunc, NFP_NET_CFG_RSS_HFUNCS); 2401 if (func_bit == NFP_NET_CFG_RSS_HFUNCS) { 2402 dev_warn(nn->dp.dev, 2403 "Bad RSS config, defaulting to Toeplitz hash\n"); 2404 func_bit = ETH_RSS_HASH_TOP_BIT; 2405 } 2406 nn->rss_hfunc = 1 << func_bit; 2407 2408 netdev_rss_key_fill(nn->rss_key, nfp_net_rss_key_sz(nn)); 2409 2410 nfp_net_rss_init_itbl(nn); 2411 2412 /* Enable IPv4/IPv6 TCP by default */ 2413 nn->rss_cfg = NFP_NET_CFG_RSS_IPV4_TCP | 2414 NFP_NET_CFG_RSS_IPV6_TCP | 2415 FIELD_PREP(NFP_NET_CFG_RSS_HFUNC, nn->rss_hfunc) | 2416 NFP_NET_CFG_RSS_MASK; 2417 } 2418 2419 /** 2420 * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters 2421 * @nn: NFP Net device to reconfigure 2422 */ 2423 static void nfp_net_irqmod_init(struct nfp_net *nn) 2424 { 2425 nn->rx_coalesce_usecs = 50; 2426 nn->rx_coalesce_max_frames = 64; 2427 nn->tx_coalesce_usecs = 50; 2428 nn->tx_coalesce_max_frames = 64; 2429 2430 nn->rx_coalesce_adapt_on = true; 2431 nn->tx_coalesce_adapt_on = true; 2432 } 2433 2434 static void nfp_net_netdev_init(struct nfp_net *nn) 2435 { 2436 struct net_device *netdev = nn->dp.netdev; 2437 2438 nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr); 2439 2440 netdev->mtu = nn->dp.mtu; 2441 2442 /* Advertise/enable offloads based on capabilities 2443 * 2444 * Note: netdev->features show the currently enabled features 2445 * and netdev->hw_features advertises which features are 2446 * supported. By default we enable most features. 2447 */ 2448 if (nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR) 2449 netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 2450 2451 netdev->hw_features = NETIF_F_HIGHDMA; 2452 if (nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY) { 2453 netdev->hw_features |= NETIF_F_RXCSUM; 2454 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY; 2455 } 2456 if (nn->cap & NFP_NET_CFG_CTRL_TXCSUM) { 2457 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; 2458 nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXCSUM; 2459 } 2460 if (nn->cap & NFP_NET_CFG_CTRL_GATHER) { 2461 netdev->hw_features |= NETIF_F_SG; 2462 nn->dp.ctrl |= NFP_NET_CFG_CTRL_GATHER; 2463 } 2464 if ((nn->cap & NFP_NET_CFG_CTRL_LSO && nn->fw_ver.major > 2) || 2465 nn->cap & NFP_NET_CFG_CTRL_LSO2) { 2466 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6; 2467 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?: 2468 NFP_NET_CFG_CTRL_LSO; 2469 } 2470 if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) 2471 netdev->hw_features |= NETIF_F_RXHASH; 2472 2473 #ifdef CONFIG_NFP_NET_IPSEC 2474 if (nn->cap_w1 & NFP_NET_CFG_CTRL_IPSEC) 2475 netdev->hw_features |= NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM; 2476 #endif 2477 2478 if (nn->cap & NFP_NET_CFG_CTRL_VXLAN) { 2479 if (nn->cap & NFP_NET_CFG_CTRL_LSO) { 2480 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL | 2481 NETIF_F_GSO_UDP_TUNNEL_CSUM | 2482 NETIF_F_GSO_PARTIAL; 2483 netdev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM; 2484 } 2485 netdev->udp_tunnel_nic_info = &nfp_udp_tunnels; 2486 nn->dp.ctrl |= NFP_NET_CFG_CTRL_VXLAN; 2487 } 2488 if (nn->cap & NFP_NET_CFG_CTRL_NVGRE) { 2489 if (nn->cap & NFP_NET_CFG_CTRL_LSO) 2490 netdev->hw_features |= NETIF_F_GSO_GRE; 2491 nn->dp.ctrl |= NFP_NET_CFG_CTRL_NVGRE; 2492 } 2493 if (nn->cap & (NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE)) 2494 netdev->hw_enc_features = netdev->hw_features; 2495 2496 netdev->vlan_features = netdev->hw_features; 2497 2498 if (nn->cap & NFP_NET_CFG_CTRL_RXVLAN_ANY) { 2499 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX; 2500 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ?: 2501 NFP_NET_CFG_CTRL_RXVLAN; 2502 } 2503 if (nn->cap & NFP_NET_CFG_CTRL_TXVLAN_ANY) { 2504 if (nn->cap & NFP_NET_CFG_CTRL_LSO2) { 2505 nn_warn(nn, "Device advertises both TSO2 and TXVLAN. Refusing to enable TXVLAN.\n"); 2506 } else { 2507 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX; 2508 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ?: 2509 NFP_NET_CFG_CTRL_TXVLAN; 2510 } 2511 } 2512 if (nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER) { 2513 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER; 2514 nn->dp.ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER; 2515 } 2516 if (nn->cap & NFP_NET_CFG_CTRL_RXQINQ) { 2517 netdev->hw_features |= NETIF_F_HW_VLAN_STAG_RX; 2518 nn->dp.ctrl |= NFP_NET_CFG_CTRL_RXQINQ; 2519 } 2520 2521 netdev->features = netdev->hw_features; 2522 2523 if (nfp_app_has_tc(nn->app) && nn->port) 2524 netdev->hw_features |= NETIF_F_HW_TC; 2525 2526 /* C-Tag strip and S-Tag strip can't be supported simultaneously, 2527 * so enable C-Tag strip and disable S-Tag strip by default. 2528 */ 2529 netdev->features &= ~NETIF_F_HW_VLAN_STAG_RX; 2530 nn->dp.ctrl &= ~NFP_NET_CFG_CTRL_RXQINQ; 2531 2532 /* Finalise the netdev setup */ 2533 switch (nn->dp.ops->version) { 2534 case NFP_NFD_VER_NFD3: 2535 netdev->netdev_ops = &nfp_nfd3_netdev_ops; 2536 break; 2537 case NFP_NFD_VER_NFDK: 2538 netdev->netdev_ops = &nfp_nfdk_netdev_ops; 2539 break; 2540 } 2541 2542 netdev->watchdog_timeo = msecs_to_jiffies(5 * 1000); 2543 2544 /* MTU range: 68 - hw-specific max */ 2545 netdev->min_mtu = ETH_MIN_MTU; 2546 netdev->max_mtu = nn->max_mtu; 2547 2548 netif_set_tso_max_segs(netdev, NFP_NET_LSO_MAX_SEGS); 2549 2550 netif_carrier_off(netdev); 2551 2552 nfp_net_set_ethtool_ops(netdev); 2553 } 2554 2555 static int nfp_net_read_caps(struct nfp_net *nn) 2556 { 2557 /* Get some of the read-only fields from the BAR */ 2558 nn->cap = nn_readl(nn, NFP_NET_CFG_CAP); 2559 nn->cap_w1 = nn_readl(nn, NFP_NET_CFG_CAP_WORD1); 2560 nn->max_mtu = nn_readl(nn, NFP_NET_CFG_MAX_MTU); 2561 2562 /* ABI 4.x and ctrl vNIC always use chained metadata, in other cases 2563 * we allow use of non-chained metadata if RSS(v1) is the only 2564 * advertised capability requiring metadata. 2565 */ 2566 nn->dp.chained_metadata_format = nn->fw_ver.major == 4 || 2567 !nn->dp.netdev || 2568 !(nn->cap & NFP_NET_CFG_CTRL_RSS) || 2569 nn->cap & NFP_NET_CFG_CTRL_CHAIN_META; 2570 /* RSS(v1) uses non-chained metadata format, except in ABI 4.x where 2571 * it has the same meaning as RSSv2. 2572 */ 2573 if (nn->dp.chained_metadata_format && nn->fw_ver.major != 4) 2574 nn->cap &= ~NFP_NET_CFG_CTRL_RSS; 2575 2576 /* Determine RX packet/metadata boundary offset */ 2577 if (nn->fw_ver.major >= 2) { 2578 u32 reg; 2579 2580 reg = nn_readl(nn, NFP_NET_CFG_RX_OFFSET); 2581 if (reg > NFP_NET_MAX_PREPEND) { 2582 nn_err(nn, "Invalid rx offset: %d\n", reg); 2583 return -EINVAL; 2584 } 2585 nn->dp.rx_offset = reg; 2586 } else { 2587 nn->dp.rx_offset = NFP_NET_RX_OFFSET; 2588 } 2589 2590 /* Mask out NFD-version-specific features */ 2591 nn->cap &= nn->dp.ops->cap_mask; 2592 2593 /* For control vNICs mask out the capabilities app doesn't want. */ 2594 if (!nn->dp.netdev) 2595 nn->cap &= nn->app->type->ctrl_cap_mask; 2596 2597 return 0; 2598 } 2599 2600 /** 2601 * nfp_net_init() - Initialise/finalise the nfp_net structure 2602 * @nn: NFP Net device structure 2603 * 2604 * Return: 0 on success or negative errno on error. 2605 */ 2606 int nfp_net_init(struct nfp_net *nn) 2607 { 2608 int err; 2609 2610 nn->dp.rx_dma_dir = DMA_FROM_DEVICE; 2611 2612 err = nfp_net_read_caps(nn); 2613 if (err) 2614 return err; 2615 2616 /* Set default MTU and Freelist buffer size */ 2617 if (!nfp_net_is_data_vnic(nn) && nn->app->ctrl_mtu) { 2618 nn->dp.mtu = min(nn->app->ctrl_mtu, nn->max_mtu); 2619 } else if (nn->max_mtu < NFP_NET_DEFAULT_MTU) { 2620 nn->dp.mtu = nn->max_mtu; 2621 } else { 2622 nn->dp.mtu = NFP_NET_DEFAULT_MTU; 2623 } 2624 nn->dp.fl_bufsz = nfp_net_calc_fl_bufsz(&nn->dp); 2625 2626 if (nfp_app_ctrl_uses_data_vnics(nn->app)) 2627 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_CMSG_DATA; 2628 2629 if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) { 2630 nfp_net_rss_init(nn); 2631 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RSS2 ?: 2632 NFP_NET_CFG_CTRL_RSS; 2633 } 2634 2635 /* Allow L2 Broadcast and Multicast through by default, if supported */ 2636 if (nn->cap & NFP_NET_CFG_CTRL_L2BC) 2637 nn->dp.ctrl |= NFP_NET_CFG_CTRL_L2BC; 2638 2639 /* Allow IRQ moderation, if supported */ 2640 if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) { 2641 nfp_net_irqmod_init(nn); 2642 nn->dp.ctrl |= NFP_NET_CFG_CTRL_IRQMOD; 2643 } 2644 2645 /* Enable TX pointer writeback, if supported */ 2646 if (nn->cap & NFP_NET_CFG_CTRL_TXRWB) 2647 nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXRWB; 2648 2649 if (nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER) 2650 nn->dp.ctrl_w1 |= NFP_NET_CFG_CTRL_MCAST_FILTER; 2651 2652 /* Stash the re-configuration queue away. First odd queue in TX Bar */ 2653 nn->qcp_cfg = nn->tx_bar + NFP_QCP_QUEUE_ADDR_SZ; 2654 2655 /* Make sure the FW knows the netdev is supposed to be disabled here */ 2656 nn_writel(nn, NFP_NET_CFG_CTRL, 0); 2657 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0); 2658 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0); 2659 nn_writel(nn, NFP_NET_CFG_CTRL_WORD1, 0); 2660 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_RING | 2661 NFP_NET_CFG_UPDATE_GEN); 2662 if (err) 2663 return err; 2664 2665 if (nn->dp.netdev) { 2666 nfp_net_netdev_init(nn); 2667 2668 err = nfp_ccm_mbox_init(nn); 2669 if (err) 2670 return err; 2671 2672 err = nfp_net_tls_init(nn); 2673 if (err) 2674 goto err_clean_mbox; 2675 2676 nfp_net_ipsec_init(nn); 2677 } 2678 2679 nfp_net_vecs_init(nn); 2680 2681 if (!nn->dp.netdev) 2682 return 0; 2683 2684 spin_lock_init(&nn->mc_lock); 2685 INIT_LIST_HEAD(&nn->mc_addrs); 2686 INIT_WORK(&nn->mc_work, nfp_net_mc_addr_config); 2687 2688 return register_netdev(nn->dp.netdev); 2689 2690 err_clean_mbox: 2691 nfp_ccm_mbox_clean(nn); 2692 return err; 2693 } 2694 2695 /** 2696 * nfp_net_clean() - Undo what nfp_net_init() did. 2697 * @nn: NFP Net device structure 2698 */ 2699 void nfp_net_clean(struct nfp_net *nn) 2700 { 2701 if (!nn->dp.netdev) 2702 return; 2703 2704 unregister_netdev(nn->dp.netdev); 2705 nfp_net_ipsec_clean(nn); 2706 nfp_ccm_mbox_clean(nn); 2707 flush_work(&nn->mc_work); 2708 nfp_net_reconfig_wait_posted(nn); 2709 } 2710