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 #include <net/xfrm.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 static int nfp_net_mc_unsync(struct net_device *netdev, const unsigned char *addr); 57 58 /** 59 * nfp_net_get_fw_version() - Read and parse the FW version 60 * @fw_ver: Output fw_version structure to read to 61 * @ctrl_bar: Mapped address of the control BAR 62 */ 63 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver, 64 void __iomem *ctrl_bar) 65 { 66 u32 reg; 67 68 reg = readl(ctrl_bar + NFP_NET_CFG_VERSION); 69 put_unaligned_le32(reg, fw_ver); 70 } 71 72 u32 nfp_qcp_queue_offset(const struct nfp_dev_info *dev_info, u16 queue) 73 { 74 queue &= dev_info->qc_idx_mask; 75 return dev_info->qc_addr_offset + NFP_QCP_QUEUE_ADDR_SZ * queue; 76 } 77 78 /* Firmware reconfig 79 * 80 * Firmware reconfig may take a while so we have two versions of it - 81 * synchronous and asynchronous (posted). All synchronous callers are holding 82 * RTNL so we don't have to worry about serializing them. 83 */ 84 static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update) 85 { 86 nn_writel(nn, NFP_NET_CFG_UPDATE, update); 87 /* ensure update is written before pinging HW */ 88 nn_pci_flush(nn); 89 nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1); 90 nn->reconfig_in_progress_update = update; 91 } 92 93 /* Pass 0 as update to run posted reconfigs. */ 94 static void nfp_net_reconfig_start_async(struct nfp_net *nn, u32 update) 95 { 96 update |= nn->reconfig_posted; 97 nn->reconfig_posted = 0; 98 99 nfp_net_reconfig_start(nn, update); 100 101 nn->reconfig_timer_active = true; 102 mod_timer(&nn->reconfig_timer, jiffies + NFP_NET_POLL_TIMEOUT * HZ); 103 } 104 105 static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check) 106 { 107 u32 reg; 108 109 reg = nn_readl(nn, NFP_NET_CFG_UPDATE); 110 if (reg == 0) 111 return true; 112 if (reg & NFP_NET_CFG_UPDATE_ERR) { 113 nn_err(nn, "Reconfig error (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n", 114 reg, nn->reconfig_in_progress_update, 115 nn_readl(nn, NFP_NET_CFG_CTRL)); 116 return true; 117 } else if (last_check) { 118 nn_err(nn, "Reconfig timeout (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n", 119 reg, nn->reconfig_in_progress_update, 120 nn_readl(nn, NFP_NET_CFG_CTRL)); 121 return true; 122 } 123 124 return false; 125 } 126 127 static bool __nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline) 128 { 129 bool timed_out = false; 130 int i; 131 132 /* Poll update field, waiting for NFP to ack the config. 133 * Do an opportunistic wait-busy loop, afterward sleep. 134 */ 135 for (i = 0; i < 50; i++) { 136 if (nfp_net_reconfig_check_done(nn, false)) 137 return false; 138 udelay(4); 139 } 140 141 while (!nfp_net_reconfig_check_done(nn, timed_out)) { 142 usleep_range(250, 500); 143 timed_out = time_is_before_eq_jiffies(deadline); 144 } 145 146 return timed_out; 147 } 148 149 static int nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline) 150 { 151 if (__nfp_net_reconfig_wait(nn, deadline)) 152 return -EIO; 153 154 if (nn_readl(nn, NFP_NET_CFG_UPDATE) & NFP_NET_CFG_UPDATE_ERR) 155 return -EIO; 156 157 return 0; 158 } 159 160 static void nfp_net_reconfig_timer(struct timer_list *t) 161 { 162 struct nfp_net *nn = from_timer(nn, t, reconfig_timer); 163 164 spin_lock_bh(&nn->reconfig_lock); 165 166 nn->reconfig_timer_active = false; 167 168 /* If sync caller is present it will take over from us */ 169 if (nn->reconfig_sync_present) 170 goto done; 171 172 /* Read reconfig status and report errors */ 173 nfp_net_reconfig_check_done(nn, true); 174 175 if (nn->reconfig_posted) 176 nfp_net_reconfig_start_async(nn, 0); 177 done: 178 spin_unlock_bh(&nn->reconfig_lock); 179 } 180 181 /** 182 * nfp_net_reconfig_post() - Post async reconfig request 183 * @nn: NFP Net device to reconfigure 184 * @update: The value for the update field in the BAR config 185 * 186 * Record FW reconfiguration request. Reconfiguration will be kicked off 187 * whenever reconfiguration machinery is idle. Multiple requests can be 188 * merged together! 189 */ 190 static void nfp_net_reconfig_post(struct nfp_net *nn, u32 update) 191 { 192 spin_lock_bh(&nn->reconfig_lock); 193 194 /* Sync caller will kick off async reconf when it's done, just post */ 195 if (nn->reconfig_sync_present) { 196 nn->reconfig_posted |= update; 197 goto done; 198 } 199 200 /* Opportunistically check if the previous command is done */ 201 if (!nn->reconfig_timer_active || 202 nfp_net_reconfig_check_done(nn, false)) 203 nfp_net_reconfig_start_async(nn, update); 204 else 205 nn->reconfig_posted |= update; 206 done: 207 spin_unlock_bh(&nn->reconfig_lock); 208 } 209 210 static void nfp_net_reconfig_sync_enter(struct nfp_net *nn) 211 { 212 bool cancelled_timer = false; 213 u32 pre_posted_requests; 214 215 spin_lock_bh(&nn->reconfig_lock); 216 217 WARN_ON(nn->reconfig_sync_present); 218 nn->reconfig_sync_present = true; 219 220 if (nn->reconfig_timer_active) { 221 nn->reconfig_timer_active = false; 222 cancelled_timer = true; 223 } 224 pre_posted_requests = nn->reconfig_posted; 225 nn->reconfig_posted = 0; 226 227 spin_unlock_bh(&nn->reconfig_lock); 228 229 if (cancelled_timer) { 230 del_timer_sync(&nn->reconfig_timer); 231 nfp_net_reconfig_wait(nn, nn->reconfig_timer.expires); 232 } 233 234 /* Run the posted reconfigs which were issued before we started */ 235 if (pre_posted_requests) { 236 nfp_net_reconfig_start(nn, pre_posted_requests); 237 nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT); 238 } 239 } 240 241 static void nfp_net_reconfig_wait_posted(struct nfp_net *nn) 242 { 243 nfp_net_reconfig_sync_enter(nn); 244 245 spin_lock_bh(&nn->reconfig_lock); 246 nn->reconfig_sync_present = false; 247 spin_unlock_bh(&nn->reconfig_lock); 248 } 249 250 /** 251 * __nfp_net_reconfig() - Reconfigure the firmware 252 * @nn: NFP Net device to reconfigure 253 * @update: The value for the update field in the BAR config 254 * 255 * Write the update word to the BAR and ping the reconfig queue. The 256 * poll until the firmware has acknowledged the update by zeroing the 257 * update word. 258 * 259 * Return: Negative errno on error, 0 on success 260 */ 261 int __nfp_net_reconfig(struct nfp_net *nn, u32 update) 262 { 263 int ret; 264 265 nfp_net_reconfig_sync_enter(nn); 266 267 nfp_net_reconfig_start(nn, update); 268 ret = nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT); 269 270 spin_lock_bh(&nn->reconfig_lock); 271 272 if (nn->reconfig_posted) 273 nfp_net_reconfig_start_async(nn, 0); 274 275 nn->reconfig_sync_present = false; 276 277 spin_unlock_bh(&nn->reconfig_lock); 278 279 return ret; 280 } 281 282 int nfp_net_reconfig(struct nfp_net *nn, u32 update) 283 { 284 int ret; 285 286 nn_ctrl_bar_lock(nn); 287 ret = __nfp_net_reconfig(nn, update); 288 nn_ctrl_bar_unlock(nn); 289 290 return ret; 291 } 292 293 int nfp_net_mbox_lock(struct nfp_net *nn, unsigned int data_size) 294 { 295 if (nn->tlv_caps.mbox_len < NFP_NET_CFG_MBOX_SIMPLE_VAL + data_size) { 296 nn_err(nn, "mailbox too small for %u of data (%u)\n", 297 data_size, nn->tlv_caps.mbox_len); 298 return -EIO; 299 } 300 301 nn_ctrl_bar_lock(nn); 302 return 0; 303 } 304 305 /** 306 * nfp_net_mbox_reconfig() - Reconfigure the firmware via the mailbox 307 * @nn: NFP Net device to reconfigure 308 * @mbox_cmd: The value for the mailbox command 309 * 310 * Helper function for mailbox updates 311 * 312 * Return: Negative errno on error, 0 on success 313 */ 314 int nfp_net_mbox_reconfig(struct nfp_net *nn, u32 mbox_cmd) 315 { 316 u32 mbox = nn->tlv_caps.mbox_off; 317 int ret; 318 319 nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd); 320 321 ret = __nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MBOX); 322 if (ret) { 323 nn_err(nn, "Mailbox update error\n"); 324 return ret; 325 } 326 327 return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET); 328 } 329 330 void nfp_net_mbox_reconfig_post(struct nfp_net *nn, u32 mbox_cmd) 331 { 332 u32 mbox = nn->tlv_caps.mbox_off; 333 334 nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd); 335 336 nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_MBOX); 337 } 338 339 int nfp_net_mbox_reconfig_wait_posted(struct nfp_net *nn) 340 { 341 u32 mbox = nn->tlv_caps.mbox_off; 342 343 nfp_net_reconfig_wait_posted(nn); 344 345 return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET); 346 } 347 348 int nfp_net_mbox_reconfig_and_unlock(struct nfp_net *nn, u32 mbox_cmd) 349 { 350 int ret; 351 352 ret = nfp_net_mbox_reconfig(nn, mbox_cmd); 353 nn_ctrl_bar_unlock(nn); 354 return ret; 355 } 356 357 /* Interrupt configuration and handling 358 */ 359 360 /** 361 * nfp_net_irqs_alloc() - allocates MSI-X irqs 362 * @pdev: PCI device structure 363 * @irq_entries: Array to be initialized and used to hold the irq entries 364 * @min_irqs: Minimal acceptable number of interrupts 365 * @wanted_irqs: Target number of interrupts to allocate 366 * 367 * Return: Number of irqs obtained or 0 on error. 368 */ 369 unsigned int 370 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries, 371 unsigned int min_irqs, unsigned int wanted_irqs) 372 { 373 unsigned int i; 374 int got_irqs; 375 376 for (i = 0; i < wanted_irqs; i++) 377 irq_entries[i].entry = i; 378 379 got_irqs = pci_enable_msix_range(pdev, irq_entries, 380 min_irqs, wanted_irqs); 381 if (got_irqs < 0) { 382 dev_err(&pdev->dev, "Failed to enable %d-%d MSI-X (err=%d)\n", 383 min_irqs, wanted_irqs, got_irqs); 384 return 0; 385 } 386 387 if (got_irqs < wanted_irqs) 388 dev_warn(&pdev->dev, "Unable to allocate %d IRQs got only %d\n", 389 wanted_irqs, got_irqs); 390 391 return got_irqs; 392 } 393 394 /** 395 * nfp_net_irqs_assign() - Assign interrupts allocated externally to netdev 396 * @nn: NFP Network structure 397 * @irq_entries: Table of allocated interrupts 398 * @n: Size of @irq_entries (number of entries to grab) 399 * 400 * After interrupts are allocated with nfp_net_irqs_alloc() this function 401 * should be called to assign them to a specific netdev (port). 402 */ 403 void 404 nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries, 405 unsigned int n) 406 { 407 struct nfp_net_dp *dp = &nn->dp; 408 409 nn->max_r_vecs = n - NFP_NET_NON_Q_VECTORS; 410 dp->num_r_vecs = nn->max_r_vecs; 411 412 memcpy(nn->irq_entries, irq_entries, sizeof(*irq_entries) * n); 413 414 if (dp->num_rx_rings > dp->num_r_vecs || 415 dp->num_tx_rings > dp->num_r_vecs) 416 dev_warn(nn->dp.dev, "More rings (%d,%d) than vectors (%d).\n", 417 dp->num_rx_rings, dp->num_tx_rings, 418 dp->num_r_vecs); 419 420 dp->num_rx_rings = min(dp->num_r_vecs, dp->num_rx_rings); 421 dp->num_tx_rings = min(dp->num_r_vecs, dp->num_tx_rings); 422 dp->num_stack_tx_rings = dp->num_tx_rings; 423 } 424 425 /** 426 * nfp_net_irqs_disable() - Disable interrupts 427 * @pdev: PCI device structure 428 * 429 * Undoes what @nfp_net_irqs_alloc() does. 430 */ 431 void nfp_net_irqs_disable(struct pci_dev *pdev) 432 { 433 pci_disable_msix(pdev); 434 } 435 436 /** 437 * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings. 438 * @irq: Interrupt 439 * @data: Opaque data structure 440 * 441 * Return: Indicate if the interrupt has been handled. 442 */ 443 static irqreturn_t nfp_net_irq_rxtx(int irq, void *data) 444 { 445 struct nfp_net_r_vector *r_vec = data; 446 447 /* Currently we cannot tell if it's a rx or tx interrupt, 448 * since dim does not need accurate event_ctr to calculate, 449 * we just use this counter for both rx and tx dim. 450 */ 451 r_vec->event_ctr++; 452 453 napi_schedule_irqoff(&r_vec->napi); 454 455 /* The FW auto-masks any interrupt, either via the MASK bit in 456 * the MSI-X table or via the per entry ICR field. So there 457 * is no need to disable interrupts here. 458 */ 459 return IRQ_HANDLED; 460 } 461 462 static irqreturn_t nfp_ctrl_irq_rxtx(int irq, void *data) 463 { 464 struct nfp_net_r_vector *r_vec = data; 465 466 tasklet_schedule(&r_vec->tasklet); 467 468 return IRQ_HANDLED; 469 } 470 471 /** 472 * nfp_net_read_link_status() - Reread link status from control BAR 473 * @nn: NFP Network structure 474 */ 475 static void nfp_net_read_link_status(struct nfp_net *nn) 476 { 477 unsigned long flags; 478 bool link_up; 479 u16 sts; 480 481 spin_lock_irqsave(&nn->link_status_lock, flags); 482 483 sts = nn_readw(nn, NFP_NET_CFG_STS); 484 link_up = !!(sts & NFP_NET_CFG_STS_LINK); 485 486 if (nn->link_up == link_up) 487 goto out; 488 489 nn->link_up = link_up; 490 if (nn->port) { 491 set_bit(NFP_PORT_CHANGED, &nn->port->flags); 492 if (nn->port->link_cb) 493 nn->port->link_cb(nn->port); 494 } 495 496 if (nn->link_up) { 497 netif_carrier_on(nn->dp.netdev); 498 netdev_info(nn->dp.netdev, "NIC Link is Up\n"); 499 } else { 500 netif_carrier_off(nn->dp.netdev); 501 netdev_info(nn->dp.netdev, "NIC Link is Down\n"); 502 } 503 out: 504 spin_unlock_irqrestore(&nn->link_status_lock, flags); 505 } 506 507 /** 508 * nfp_net_irq_lsc() - Interrupt service routine for link state changes 509 * @irq: Interrupt 510 * @data: Opaque data structure 511 * 512 * Return: Indicate if the interrupt has been handled. 513 */ 514 static irqreturn_t nfp_net_irq_lsc(int irq, void *data) 515 { 516 struct nfp_net *nn = data; 517 struct msix_entry *entry; 518 519 entry = &nn->irq_entries[NFP_NET_IRQ_LSC_IDX]; 520 521 nfp_net_read_link_status(nn); 522 523 nfp_net_irq_unmask(nn, entry->entry); 524 525 return IRQ_HANDLED; 526 } 527 528 /** 529 * nfp_net_irq_exn() - Interrupt service routine for exceptions 530 * @irq: Interrupt 531 * @data: Opaque data structure 532 * 533 * Return: Indicate if the interrupt has been handled. 534 */ 535 static irqreturn_t nfp_net_irq_exn(int irq, void *data) 536 { 537 struct nfp_net *nn = data; 538 539 nn_err(nn, "%s: UNIMPLEMENTED.\n", __func__); 540 /* XXX TO BE IMPLEMENTED */ 541 return IRQ_HANDLED; 542 } 543 544 /** 545 * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN) 546 * @nn: NFP Network structure 547 * @ctrl_offset: Control BAR offset where IRQ configuration should be written 548 * @format: printf-style format to construct the interrupt name 549 * @name: Pointer to allocated space for interrupt name 550 * @name_sz: Size of space for interrupt name 551 * @vector_idx: Index of MSI-X vector used for this interrupt 552 * @handler: IRQ handler to register for this interrupt 553 */ 554 static int 555 nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset, 556 const char *format, char *name, size_t name_sz, 557 unsigned int vector_idx, irq_handler_t handler) 558 { 559 struct msix_entry *entry; 560 int err; 561 562 entry = &nn->irq_entries[vector_idx]; 563 564 snprintf(name, name_sz, format, nfp_net_name(nn)); 565 err = request_irq(entry->vector, handler, 0, name, nn); 566 if (err) { 567 nn_err(nn, "Failed to request IRQ %d (err=%d).\n", 568 entry->vector, err); 569 return err; 570 } 571 nn_writeb(nn, ctrl_offset, entry->entry); 572 nfp_net_irq_unmask(nn, entry->entry); 573 574 return 0; 575 } 576 577 /** 578 * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN) 579 * @nn: NFP Network structure 580 * @ctrl_offset: Control BAR offset where IRQ configuration should be written 581 * @vector_idx: Index of MSI-X vector used for this interrupt 582 */ 583 static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset, 584 unsigned int vector_idx) 585 { 586 nn_writeb(nn, ctrl_offset, 0xff); 587 nn_pci_flush(nn); 588 free_irq(nn->irq_entries[vector_idx].vector, nn); 589 } 590 591 struct sk_buff * 592 nfp_net_tls_tx(struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec, 593 struct sk_buff *skb, u64 *tls_handle, int *nr_frags) 594 { 595 #ifdef CONFIG_TLS_DEVICE 596 struct nfp_net_tls_offload_ctx *ntls; 597 struct sk_buff *nskb; 598 bool resync_pending; 599 u32 datalen, seq; 600 601 if (likely(!dp->ktls_tx)) 602 return skb; 603 if (!tls_is_skb_tx_device_offloaded(skb)) 604 return skb; 605 606 datalen = skb->len - skb_tcp_all_headers(skb); 607 seq = ntohl(tcp_hdr(skb)->seq); 608 ntls = tls_driver_ctx(skb->sk, TLS_OFFLOAD_CTX_DIR_TX); 609 resync_pending = tls_offload_tx_resync_pending(skb->sk); 610 if (unlikely(resync_pending || ntls->next_seq != seq)) { 611 /* Pure ACK out of order already */ 612 if (!datalen) 613 return skb; 614 615 u64_stats_update_begin(&r_vec->tx_sync); 616 r_vec->tls_tx_fallback++; 617 u64_stats_update_end(&r_vec->tx_sync); 618 619 nskb = tls_encrypt_skb(skb); 620 if (!nskb) { 621 u64_stats_update_begin(&r_vec->tx_sync); 622 r_vec->tls_tx_no_fallback++; 623 u64_stats_update_end(&r_vec->tx_sync); 624 return NULL; 625 } 626 /* encryption wasn't necessary */ 627 if (nskb == skb) 628 return skb; 629 /* we don't re-check ring space */ 630 if (unlikely(skb_is_nonlinear(nskb))) { 631 nn_dp_warn(dp, "tls_encrypt_skb() produced fragmented frame\n"); 632 u64_stats_update_begin(&r_vec->tx_sync); 633 r_vec->tx_errors++; 634 u64_stats_update_end(&r_vec->tx_sync); 635 dev_kfree_skb_any(nskb); 636 return NULL; 637 } 638 639 /* jump forward, a TX may have gotten lost, need to sync TX */ 640 if (!resync_pending && seq - ntls->next_seq < U32_MAX / 4) 641 tls_offload_tx_resync_request(nskb->sk, seq, 642 ntls->next_seq); 643 644 *nr_frags = 0; 645 return nskb; 646 } 647 648 if (datalen) { 649 u64_stats_update_begin(&r_vec->tx_sync); 650 if (!skb_is_gso(skb)) 651 r_vec->hw_tls_tx++; 652 else 653 r_vec->hw_tls_tx += skb_shinfo(skb)->gso_segs; 654 u64_stats_update_end(&r_vec->tx_sync); 655 } 656 657 memcpy(tls_handle, ntls->fw_handle, sizeof(ntls->fw_handle)); 658 ntls->next_seq += datalen; 659 #endif 660 return skb; 661 } 662 663 void nfp_net_tls_tx_undo(struct sk_buff *skb, u64 tls_handle) 664 { 665 #ifdef CONFIG_TLS_DEVICE 666 struct nfp_net_tls_offload_ctx *ntls; 667 u32 datalen, seq; 668 669 if (!tls_handle) 670 return; 671 if (WARN_ON_ONCE(!tls_is_skb_tx_device_offloaded(skb))) 672 return; 673 674 datalen = skb->len - skb_tcp_all_headers(skb); 675 seq = ntohl(tcp_hdr(skb)->seq); 676 677 ntls = tls_driver_ctx(skb->sk, TLS_OFFLOAD_CTX_DIR_TX); 678 if (ntls->next_seq == seq + datalen) 679 ntls->next_seq = seq; 680 else 681 WARN_ON_ONCE(1); 682 #endif 683 } 684 685 static void nfp_net_tx_timeout(struct net_device *netdev, unsigned int txqueue) 686 { 687 struct nfp_net *nn = netdev_priv(netdev); 688 689 nn_warn(nn, "TX watchdog timeout on ring: %u\n", txqueue); 690 } 691 692 /* Receive processing */ 693 static unsigned int 694 nfp_net_calc_fl_bufsz_data(struct nfp_net_dp *dp) 695 { 696 unsigned int fl_bufsz = 0; 697 698 if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC) 699 fl_bufsz += NFP_NET_MAX_PREPEND; 700 else 701 fl_bufsz += dp->rx_offset; 702 fl_bufsz += ETH_HLEN + VLAN_HLEN * 2 + dp->mtu; 703 704 return fl_bufsz; 705 } 706 707 static unsigned int nfp_net_calc_fl_bufsz(struct nfp_net_dp *dp) 708 { 709 unsigned int fl_bufsz; 710 711 fl_bufsz = NFP_NET_RX_BUF_HEADROOM; 712 fl_bufsz += dp->rx_dma_off; 713 fl_bufsz += nfp_net_calc_fl_bufsz_data(dp); 714 715 fl_bufsz = SKB_DATA_ALIGN(fl_bufsz); 716 fl_bufsz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 717 718 return fl_bufsz; 719 } 720 721 static unsigned int nfp_net_calc_fl_bufsz_xsk(struct nfp_net_dp *dp) 722 { 723 unsigned int fl_bufsz; 724 725 fl_bufsz = XDP_PACKET_HEADROOM; 726 fl_bufsz += nfp_net_calc_fl_bufsz_data(dp); 727 728 return fl_bufsz; 729 } 730 731 /* Setup and Configuration 732 */ 733 734 /** 735 * nfp_net_vecs_init() - Assign IRQs and setup rvecs. 736 * @nn: NFP Network structure 737 */ 738 static void nfp_net_vecs_init(struct nfp_net *nn) 739 { 740 int numa_node = dev_to_node(&nn->pdev->dev); 741 struct nfp_net_r_vector *r_vec; 742 unsigned int r; 743 744 nn->lsc_handler = nfp_net_irq_lsc; 745 nn->exn_handler = nfp_net_irq_exn; 746 747 for (r = 0; r < nn->max_r_vecs; r++) { 748 struct msix_entry *entry; 749 750 entry = &nn->irq_entries[NFP_NET_NON_Q_VECTORS + r]; 751 752 r_vec = &nn->r_vecs[r]; 753 r_vec->nfp_net = nn; 754 r_vec->irq_entry = entry->entry; 755 r_vec->irq_vector = entry->vector; 756 757 if (nn->dp.netdev) { 758 r_vec->handler = nfp_net_irq_rxtx; 759 } else { 760 r_vec->handler = nfp_ctrl_irq_rxtx; 761 762 __skb_queue_head_init(&r_vec->queue); 763 spin_lock_init(&r_vec->lock); 764 tasklet_setup(&r_vec->tasklet, nn->dp.ops->ctrl_poll); 765 tasklet_disable(&r_vec->tasklet); 766 } 767 768 cpumask_set_cpu(cpumask_local_spread(r, numa_node), &r_vec->affinity_mask); 769 } 770 } 771 772 static void 773 nfp_net_napi_add(struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec, int idx) 774 { 775 if (dp->netdev) 776 netif_napi_add(dp->netdev, &r_vec->napi, 777 nfp_net_has_xsk_pool_slow(dp, idx) ? dp->ops->xsk_poll : dp->ops->poll); 778 else 779 tasklet_enable(&r_vec->tasklet); 780 } 781 782 static void 783 nfp_net_napi_del(struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec) 784 { 785 if (dp->netdev) 786 netif_napi_del(&r_vec->napi); 787 else 788 tasklet_disable(&r_vec->tasklet); 789 } 790 791 static void 792 nfp_net_vector_assign_rings(struct nfp_net_dp *dp, 793 struct nfp_net_r_vector *r_vec, int idx) 794 { 795 r_vec->rx_ring = idx < dp->num_rx_rings ? &dp->rx_rings[idx] : NULL; 796 r_vec->tx_ring = 797 idx < dp->num_stack_tx_rings ? &dp->tx_rings[idx] : NULL; 798 799 r_vec->xdp_ring = idx < dp->num_tx_rings - dp->num_stack_tx_rings ? 800 &dp->tx_rings[dp->num_stack_tx_rings + idx] : NULL; 801 802 if (nfp_net_has_xsk_pool_slow(dp, idx) || r_vec->xsk_pool) { 803 r_vec->xsk_pool = dp->xdp_prog ? dp->xsk_pools[idx] : NULL; 804 805 if (r_vec->xsk_pool) 806 xsk_pool_set_rxq_info(r_vec->xsk_pool, 807 &r_vec->rx_ring->xdp_rxq); 808 809 nfp_net_napi_del(dp, r_vec); 810 nfp_net_napi_add(dp, r_vec, idx); 811 } 812 } 813 814 static int 815 nfp_net_prepare_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec, 816 int idx) 817 { 818 int err; 819 820 nfp_net_napi_add(&nn->dp, r_vec, idx); 821 822 snprintf(r_vec->name, sizeof(r_vec->name), 823 "%s-rxtx-%d", nfp_net_name(nn), idx); 824 err = request_irq(r_vec->irq_vector, r_vec->handler, IRQF_NO_AUTOEN, 825 r_vec->name, r_vec); 826 if (err) { 827 nfp_net_napi_del(&nn->dp, r_vec); 828 nn_err(nn, "Error requesting IRQ %d\n", r_vec->irq_vector); 829 return err; 830 } 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, new_ctrl_w1, 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 if (!(nn->cap_w1 & NFP_NET_CFG_CTRL_FREELIST_EN)) { 940 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0); 941 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0); 942 } 943 944 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 945 err = nfp_net_reconfig(nn, update); 946 if (err) 947 nn_err(nn, "Could not disable device: %d\n", err); 948 949 if (nn->cap_w1 & NFP_NET_CFG_CTRL_FREELIST_EN) { 950 new_ctrl_w1 = nn->dp.ctrl_w1; 951 new_ctrl_w1 &= ~NFP_NET_CFG_CTRL_FREELIST_EN; 952 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0); 953 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0); 954 955 nn_writel(nn, NFP_NET_CFG_CTRL_WORD1, new_ctrl_w1); 956 err = nfp_net_reconfig(nn, update); 957 if (err) 958 nn_err(nn, "Could not disable FREELIST_EN: %d\n", err); 959 nn->dp.ctrl_w1 = new_ctrl_w1; 960 } 961 962 for (r = 0; r < nn->dp.num_rx_rings; r++) { 963 nfp_net_rx_ring_reset(&nn->dp.rx_rings[r]); 964 if (nfp_net_has_xsk_pool_slow(&nn->dp, nn->dp.rx_rings[r].idx)) 965 nfp_net_xsk_rx_bufs_free(&nn->dp.rx_rings[r]); 966 } 967 for (r = 0; r < nn->dp.num_tx_rings; r++) 968 nfp_net_tx_ring_reset(&nn->dp, &nn->dp.tx_rings[r]); 969 for (r = 0; r < nn->dp.num_r_vecs; r++) 970 nfp_net_vec_clear_ring_data(nn, r); 971 972 nn->dp.ctrl = new_ctrl; 973 } 974 975 /** 976 * nfp_net_set_config_and_enable() - Write control BAR and enable NFP 977 * @nn: NFP Net device to reconfigure 978 */ 979 static int nfp_net_set_config_and_enable(struct nfp_net *nn) 980 { 981 u32 bufsz, new_ctrl, new_ctrl_w1, update = 0; 982 unsigned int r; 983 int err; 984 985 new_ctrl = nn->dp.ctrl; 986 new_ctrl_w1 = nn->dp.ctrl_w1; 987 988 if (nn->dp.ctrl & NFP_NET_CFG_CTRL_RSS_ANY) { 989 nfp_net_rss_write_key(nn); 990 nfp_net_rss_write_itbl(nn); 991 nn_writel(nn, NFP_NET_CFG_RSS_CTRL, nn->rss_cfg); 992 update |= NFP_NET_CFG_UPDATE_RSS; 993 } 994 995 if (nn->dp.ctrl & NFP_NET_CFG_CTRL_IRQMOD) { 996 nfp_net_coalesce_write_cfg(nn); 997 update |= NFP_NET_CFG_UPDATE_IRQMOD; 998 } 999 1000 for (r = 0; r < nn->dp.num_tx_rings; r++) 1001 nfp_net_tx_ring_hw_cfg_write(nn, &nn->dp.tx_rings[r], r); 1002 for (r = 0; r < nn->dp.num_rx_rings; r++) 1003 nfp_net_rx_ring_hw_cfg_write(nn, &nn->dp.rx_rings[r], r); 1004 1005 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 1006 U64_MAX >> (64 - nn->dp.num_tx_rings)); 1007 1008 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 1009 U64_MAX >> (64 - nn->dp.num_rx_rings)); 1010 1011 if (nn->dp.netdev) 1012 nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr); 1013 1014 nn_writel(nn, NFP_NET_CFG_MTU, nn->dp.mtu); 1015 1016 bufsz = nn->dp.fl_bufsz - nn->dp.rx_dma_off - NFP_NET_RX_BUF_NON_DATA; 1017 nn_writel(nn, NFP_NET_CFG_FLBUFSZ, bufsz); 1018 1019 /* Enable device 1020 * Step 1: Replace the CTRL_ENABLE by NFP_NET_CFG_CTRL_FREELIST_EN if 1021 * FREELIST_EN exits. 1022 */ 1023 if (nn->cap_w1 & NFP_NET_CFG_CTRL_FREELIST_EN) 1024 new_ctrl_w1 |= NFP_NET_CFG_CTRL_FREELIST_EN; 1025 else 1026 new_ctrl |= NFP_NET_CFG_CTRL_ENABLE; 1027 update |= NFP_NET_CFG_UPDATE_GEN; 1028 update |= NFP_NET_CFG_UPDATE_MSIX; 1029 update |= NFP_NET_CFG_UPDATE_RING; 1030 if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG) 1031 new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG; 1032 1033 /* Step 2: Send the configuration and write the freelist. 1034 * - The freelist only need to be written once. 1035 */ 1036 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1037 nn_writel(nn, NFP_NET_CFG_CTRL_WORD1, new_ctrl_w1); 1038 err = nfp_net_reconfig(nn, update); 1039 if (err) { 1040 nfp_net_clear_config_and_disable(nn); 1041 return err; 1042 } 1043 1044 nn->dp.ctrl = new_ctrl; 1045 nn->dp.ctrl_w1 = new_ctrl_w1; 1046 1047 for (r = 0; r < nn->dp.num_rx_rings; r++) 1048 nfp_net_rx_ring_fill_freelist(&nn->dp, &nn->dp.rx_rings[r]); 1049 1050 /* Step 3: Do the NFP_NET_CFG_CTRL_ENABLE. Send the configuration. 1051 */ 1052 if (nn->cap_w1 & NFP_NET_CFG_CTRL_FREELIST_EN) { 1053 new_ctrl |= NFP_NET_CFG_CTRL_ENABLE; 1054 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1055 1056 err = nfp_net_reconfig(nn, update); 1057 if (err) { 1058 nfp_net_clear_config_and_disable(nn); 1059 return err; 1060 } 1061 nn->dp.ctrl = new_ctrl; 1062 } 1063 1064 return 0; 1065 } 1066 1067 /** 1068 * nfp_net_close_stack() - Quiesce the stack (part of close) 1069 * @nn: NFP Net device to reconfigure 1070 */ 1071 static void nfp_net_close_stack(struct nfp_net *nn) 1072 { 1073 struct nfp_net_r_vector *r_vec; 1074 unsigned int r; 1075 1076 disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector); 1077 netif_carrier_off(nn->dp.netdev); 1078 nn->link_up = false; 1079 1080 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1081 r_vec = &nn->r_vecs[r]; 1082 1083 disable_irq(r_vec->irq_vector); 1084 napi_disable(&r_vec->napi); 1085 1086 if (r_vec->rx_ring) 1087 cancel_work_sync(&r_vec->rx_dim.work); 1088 1089 if (r_vec->tx_ring) 1090 cancel_work_sync(&r_vec->tx_dim.work); 1091 } 1092 1093 netif_tx_disable(nn->dp.netdev); 1094 } 1095 1096 /** 1097 * nfp_net_close_free_all() - Free all runtime resources 1098 * @nn: NFP Net device to reconfigure 1099 */ 1100 static void nfp_net_close_free_all(struct nfp_net *nn) 1101 { 1102 unsigned int r; 1103 1104 nfp_net_tx_rings_free(&nn->dp); 1105 nfp_net_rx_rings_free(&nn->dp); 1106 1107 for (r = 0; r < nn->dp.num_r_vecs; r++) 1108 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1109 1110 nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX); 1111 nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX); 1112 } 1113 1114 /** 1115 * nfp_net_netdev_close() - Called when the device is downed 1116 * @netdev: netdev structure 1117 */ 1118 static int nfp_net_netdev_close(struct net_device *netdev) 1119 { 1120 struct nfp_net *nn = netdev_priv(netdev); 1121 1122 /* Step 1: Disable RX and TX rings from the Linux kernel perspective 1123 */ 1124 nfp_net_close_stack(nn); 1125 1126 /* Step 2: Tell NFP 1127 */ 1128 if (nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER) 1129 __dev_mc_unsync(netdev, nfp_net_mc_unsync); 1130 1131 nfp_net_clear_config_and_disable(nn); 1132 nfp_port_configure(netdev, false); 1133 1134 /* Step 3: Free resources 1135 */ 1136 nfp_net_close_free_all(nn); 1137 1138 nn_dbg(nn, "%s down", netdev->name); 1139 return 0; 1140 } 1141 1142 void nfp_ctrl_close(struct nfp_net *nn) 1143 { 1144 int r; 1145 1146 rtnl_lock(); 1147 1148 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1149 disable_irq(nn->r_vecs[r].irq_vector); 1150 tasklet_disable(&nn->r_vecs[r].tasklet); 1151 } 1152 1153 nfp_net_clear_config_and_disable(nn); 1154 1155 nfp_net_close_free_all(nn); 1156 1157 rtnl_unlock(); 1158 } 1159 1160 static void nfp_net_rx_dim_work(struct work_struct *work) 1161 { 1162 struct nfp_net_r_vector *r_vec; 1163 unsigned int factor, value; 1164 struct dim_cq_moder moder; 1165 struct nfp_net *nn; 1166 struct dim *dim; 1167 1168 dim = container_of(work, struct dim, work); 1169 moder = net_dim_get_rx_moderation(dim->mode, dim->profile_ix); 1170 r_vec = container_of(dim, struct nfp_net_r_vector, rx_dim); 1171 nn = r_vec->nfp_net; 1172 1173 /* Compute factor used to convert coalesce '_usecs' parameters to 1174 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp 1175 * count. 1176 */ 1177 factor = nn->tlv_caps.me_freq_mhz / 16; 1178 if (nfp_net_coalesce_para_check(factor * moder.usec, moder.pkts)) 1179 return; 1180 1181 /* copy RX interrupt coalesce parameters */ 1182 value = (moder.pkts << 16) | (factor * moder.usec); 1183 nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(r_vec->rx_ring->idx), value); 1184 (void)nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_IRQMOD); 1185 1186 dim->state = DIM_START_MEASURE; 1187 } 1188 1189 static void nfp_net_tx_dim_work(struct work_struct *work) 1190 { 1191 struct nfp_net_r_vector *r_vec; 1192 unsigned int factor, value; 1193 struct dim_cq_moder moder; 1194 struct nfp_net *nn; 1195 struct dim *dim; 1196 1197 dim = container_of(work, struct dim, work); 1198 moder = net_dim_get_tx_moderation(dim->mode, dim->profile_ix); 1199 r_vec = container_of(dim, struct nfp_net_r_vector, tx_dim); 1200 nn = r_vec->nfp_net; 1201 1202 /* Compute factor used to convert coalesce '_usecs' parameters to 1203 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp 1204 * count. 1205 */ 1206 factor = nn->tlv_caps.me_freq_mhz / 16; 1207 if (nfp_net_coalesce_para_check(factor * moder.usec, moder.pkts)) 1208 return; 1209 1210 /* copy TX interrupt coalesce parameters */ 1211 value = (moder.pkts << 16) | (factor * moder.usec); 1212 nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(r_vec->tx_ring->idx), value); 1213 (void)nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_IRQMOD); 1214 1215 dim->state = DIM_START_MEASURE; 1216 } 1217 1218 /** 1219 * nfp_net_open_stack() - Start the device from stack's perspective 1220 * @nn: NFP Net device to reconfigure 1221 */ 1222 static void nfp_net_open_stack(struct nfp_net *nn) 1223 { 1224 struct nfp_net_r_vector *r_vec; 1225 unsigned int r; 1226 1227 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1228 r_vec = &nn->r_vecs[r]; 1229 1230 if (r_vec->rx_ring) { 1231 INIT_WORK(&r_vec->rx_dim.work, nfp_net_rx_dim_work); 1232 r_vec->rx_dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE; 1233 } 1234 1235 if (r_vec->tx_ring) { 1236 INIT_WORK(&r_vec->tx_dim.work, nfp_net_tx_dim_work); 1237 r_vec->tx_dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE; 1238 } 1239 1240 napi_enable(&r_vec->napi); 1241 enable_irq(r_vec->irq_vector); 1242 } 1243 1244 netif_tx_wake_all_queues(nn->dp.netdev); 1245 1246 enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector); 1247 nfp_net_read_link_status(nn); 1248 } 1249 1250 static int nfp_net_open_alloc_all(struct nfp_net *nn) 1251 { 1252 int err, r; 1253 1254 err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn", 1255 nn->exn_name, sizeof(nn->exn_name), 1256 NFP_NET_IRQ_EXN_IDX, nn->exn_handler); 1257 if (err) 1258 return err; 1259 err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc", 1260 nn->lsc_name, sizeof(nn->lsc_name), 1261 NFP_NET_IRQ_LSC_IDX, nn->lsc_handler); 1262 if (err) 1263 goto err_free_exn; 1264 disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector); 1265 1266 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1267 err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r); 1268 if (err) 1269 goto err_cleanup_vec_p; 1270 } 1271 1272 err = nfp_net_rx_rings_prepare(nn, &nn->dp); 1273 if (err) 1274 goto err_cleanup_vec; 1275 1276 err = nfp_net_tx_rings_prepare(nn, &nn->dp); 1277 if (err) 1278 goto err_free_rx_rings; 1279 1280 for (r = 0; r < nn->max_r_vecs; r++) 1281 nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r); 1282 1283 return 0; 1284 1285 err_free_rx_rings: 1286 nfp_net_rx_rings_free(&nn->dp); 1287 err_cleanup_vec: 1288 r = nn->dp.num_r_vecs; 1289 err_cleanup_vec_p: 1290 while (r--) 1291 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1292 nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX); 1293 err_free_exn: 1294 nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX); 1295 return err; 1296 } 1297 1298 static int nfp_net_netdev_open(struct net_device *netdev) 1299 { 1300 struct nfp_net *nn = netdev_priv(netdev); 1301 int err; 1302 1303 /* Step 1: Allocate resources for rings and the like 1304 * - Request interrupts 1305 * - Allocate RX and TX ring resources 1306 * - Setup initial RSS table 1307 */ 1308 err = nfp_net_open_alloc_all(nn); 1309 if (err) 1310 return err; 1311 1312 err = netif_set_real_num_tx_queues(netdev, nn->dp.num_stack_tx_rings); 1313 if (err) 1314 goto err_free_all; 1315 1316 err = netif_set_real_num_rx_queues(netdev, nn->dp.num_rx_rings); 1317 if (err) 1318 goto err_free_all; 1319 1320 /* Step 2: Configure the NFP 1321 * - Ifup the physical interface if it exists 1322 * - Enable rings from 0 to tx_rings/rx_rings - 1. 1323 * - Write MAC address (in case it changed) 1324 * - Set the MTU 1325 * - Set the Freelist buffer size 1326 * - Enable the FW 1327 */ 1328 err = nfp_port_configure(netdev, true); 1329 if (err) 1330 goto err_free_all; 1331 1332 err = nfp_net_set_config_and_enable(nn); 1333 if (err) 1334 goto err_port_disable; 1335 1336 /* Step 3: Enable for kernel 1337 * - put some freelist descriptors on each RX ring 1338 * - enable NAPI on each ring 1339 * - enable all TX queues 1340 * - set link state 1341 */ 1342 nfp_net_open_stack(nn); 1343 1344 return 0; 1345 1346 err_port_disable: 1347 nfp_port_configure(netdev, false); 1348 err_free_all: 1349 nfp_net_close_free_all(nn); 1350 return err; 1351 } 1352 1353 int nfp_ctrl_open(struct nfp_net *nn) 1354 { 1355 int err, r; 1356 1357 /* ring dumping depends on vNICs being opened/closed under rtnl */ 1358 rtnl_lock(); 1359 1360 err = nfp_net_open_alloc_all(nn); 1361 if (err) 1362 goto err_unlock; 1363 1364 err = nfp_net_set_config_and_enable(nn); 1365 if (err) 1366 goto err_free_all; 1367 1368 for (r = 0; r < nn->dp.num_r_vecs; r++) 1369 enable_irq(nn->r_vecs[r].irq_vector); 1370 1371 rtnl_unlock(); 1372 1373 return 0; 1374 1375 err_free_all: 1376 nfp_net_close_free_all(nn); 1377 err_unlock: 1378 rtnl_unlock(); 1379 return err; 1380 } 1381 1382 int nfp_net_sched_mbox_amsg_work(struct nfp_net *nn, u32 cmd, const void *data, size_t len, 1383 int (*cb)(struct nfp_net *, struct nfp_mbox_amsg_entry *)) 1384 { 1385 struct nfp_mbox_amsg_entry *entry; 1386 1387 entry = kmalloc(sizeof(*entry) + len, GFP_ATOMIC); 1388 if (!entry) 1389 return -ENOMEM; 1390 1391 memcpy(entry->msg, data, len); 1392 entry->cmd = cmd; 1393 entry->cfg = cb; 1394 1395 spin_lock_bh(&nn->mbox_amsg.lock); 1396 list_add_tail(&entry->list, &nn->mbox_amsg.list); 1397 spin_unlock_bh(&nn->mbox_amsg.lock); 1398 1399 schedule_work(&nn->mbox_amsg.work); 1400 1401 return 0; 1402 } 1403 1404 static void nfp_net_mbox_amsg_work(struct work_struct *work) 1405 { 1406 struct nfp_net *nn = container_of(work, struct nfp_net, mbox_amsg.work); 1407 struct nfp_mbox_amsg_entry *entry, *tmp; 1408 struct list_head tmp_list; 1409 1410 INIT_LIST_HEAD(&tmp_list); 1411 1412 spin_lock_bh(&nn->mbox_amsg.lock); 1413 list_splice_init(&nn->mbox_amsg.list, &tmp_list); 1414 spin_unlock_bh(&nn->mbox_amsg.lock); 1415 1416 list_for_each_entry_safe(entry, tmp, &tmp_list, list) { 1417 int err = entry->cfg(nn, entry); 1418 1419 if (err) 1420 nn_err(nn, "Config cmd %d to HW failed %d.\n", entry->cmd, err); 1421 1422 list_del(&entry->list); 1423 kfree(entry); 1424 } 1425 } 1426 1427 static int nfp_net_mc_cfg(struct nfp_net *nn, struct nfp_mbox_amsg_entry *entry) 1428 { 1429 unsigned char *addr = entry->msg; 1430 int ret; 1431 1432 ret = nfp_net_mbox_lock(nn, NFP_NET_CFG_MULTICAST_SZ); 1433 if (ret) 1434 return ret; 1435 1436 nn_writel(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_MULTICAST_MAC_HI, 1437 get_unaligned_be32(addr)); 1438 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_MULTICAST_MAC_LO, 1439 get_unaligned_be16(addr + 4)); 1440 1441 return nfp_net_mbox_reconfig_and_unlock(nn, entry->cmd); 1442 } 1443 1444 static int nfp_net_mc_sync(struct net_device *netdev, const unsigned char *addr) 1445 { 1446 struct nfp_net *nn = netdev_priv(netdev); 1447 1448 if (netdev_mc_count(netdev) > NFP_NET_CFG_MAC_MC_MAX) { 1449 nn_err(nn, "Requested number of MC addresses (%d) exceeds maximum (%d).\n", 1450 netdev_mc_count(netdev), NFP_NET_CFG_MAC_MC_MAX); 1451 return -EINVAL; 1452 } 1453 1454 return nfp_net_sched_mbox_amsg_work(nn, NFP_NET_CFG_MBOX_CMD_MULTICAST_ADD, addr, 1455 NFP_NET_CFG_MULTICAST_SZ, nfp_net_mc_cfg); 1456 } 1457 1458 static int nfp_net_mc_unsync(struct net_device *netdev, const unsigned char *addr) 1459 { 1460 struct nfp_net *nn = netdev_priv(netdev); 1461 1462 return nfp_net_sched_mbox_amsg_work(nn, NFP_NET_CFG_MBOX_CMD_MULTICAST_DEL, addr, 1463 NFP_NET_CFG_MULTICAST_SZ, nfp_net_mc_cfg); 1464 } 1465 1466 static void nfp_net_set_rx_mode(struct net_device *netdev) 1467 { 1468 struct nfp_net *nn = netdev_priv(netdev); 1469 u32 new_ctrl, new_ctrl_w1; 1470 1471 new_ctrl = nn->dp.ctrl; 1472 new_ctrl_w1 = nn->dp.ctrl_w1; 1473 1474 if (!netdev_mc_empty(netdev) || netdev->flags & IFF_ALLMULTI) 1475 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_L2MC; 1476 else 1477 new_ctrl &= ~NFP_NET_CFG_CTRL_L2MC; 1478 1479 if (netdev->flags & IFF_ALLMULTI) 1480 new_ctrl_w1 &= ~NFP_NET_CFG_CTRL_MCAST_FILTER; 1481 else 1482 new_ctrl_w1 |= nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER; 1483 1484 if (netdev->flags & IFF_PROMISC) { 1485 if (nn->cap & NFP_NET_CFG_CTRL_PROMISC) 1486 new_ctrl |= NFP_NET_CFG_CTRL_PROMISC; 1487 else 1488 nn_warn(nn, "FW does not support promiscuous mode\n"); 1489 } else { 1490 new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC; 1491 } 1492 1493 if ((nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER) && 1494 __dev_mc_sync(netdev, nfp_net_mc_sync, nfp_net_mc_unsync)) 1495 netdev_err(netdev, "Sync mc address failed\n"); 1496 1497 if (new_ctrl == nn->dp.ctrl && new_ctrl_w1 == nn->dp.ctrl_w1) 1498 return; 1499 1500 if (new_ctrl != nn->dp.ctrl) 1501 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1502 if (new_ctrl_w1 != nn->dp.ctrl_w1) 1503 nn_writel(nn, NFP_NET_CFG_CTRL_WORD1, new_ctrl_w1); 1504 nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN); 1505 1506 nn->dp.ctrl = new_ctrl; 1507 nn->dp.ctrl_w1 = new_ctrl_w1; 1508 } 1509 1510 static void nfp_net_rss_init_itbl(struct nfp_net *nn) 1511 { 1512 int i; 1513 1514 for (i = 0; i < sizeof(nn->rss_itbl); i++) 1515 nn->rss_itbl[i] = 1516 ethtool_rxfh_indir_default(i, nn->dp.num_rx_rings); 1517 } 1518 1519 static void nfp_net_dp_swap(struct nfp_net *nn, struct nfp_net_dp *dp) 1520 { 1521 struct nfp_net_dp new_dp = *dp; 1522 1523 *dp = nn->dp; 1524 nn->dp = new_dp; 1525 1526 nn->dp.netdev->mtu = new_dp.mtu; 1527 1528 if (!netif_is_rxfh_configured(nn->dp.netdev)) 1529 nfp_net_rss_init_itbl(nn); 1530 } 1531 1532 static int nfp_net_dp_swap_enable(struct nfp_net *nn, struct nfp_net_dp *dp) 1533 { 1534 unsigned int r; 1535 int err; 1536 1537 nfp_net_dp_swap(nn, dp); 1538 1539 for (r = 0; r < nn->max_r_vecs; r++) 1540 nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r); 1541 1542 err = netif_set_real_num_queues(nn->dp.netdev, 1543 nn->dp.num_stack_tx_rings, 1544 nn->dp.num_rx_rings); 1545 if (err) 1546 return err; 1547 1548 return nfp_net_set_config_and_enable(nn); 1549 } 1550 1551 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn) 1552 { 1553 struct nfp_net_dp *new; 1554 1555 new = kmalloc(sizeof(*new), GFP_KERNEL); 1556 if (!new) 1557 return NULL; 1558 1559 *new = nn->dp; 1560 1561 new->xsk_pools = kmemdup(new->xsk_pools, 1562 array_size(nn->max_r_vecs, 1563 sizeof(new->xsk_pools)), 1564 GFP_KERNEL); 1565 if (!new->xsk_pools) { 1566 kfree(new); 1567 return NULL; 1568 } 1569 1570 /* Clear things which need to be recomputed */ 1571 new->fl_bufsz = 0; 1572 new->tx_rings = NULL; 1573 new->rx_rings = NULL; 1574 new->num_r_vecs = 0; 1575 new->num_stack_tx_rings = 0; 1576 new->txrwb = NULL; 1577 new->txrwb_dma = 0; 1578 1579 return new; 1580 } 1581 1582 static void nfp_net_free_dp(struct nfp_net_dp *dp) 1583 { 1584 kfree(dp->xsk_pools); 1585 kfree(dp); 1586 } 1587 1588 static int 1589 nfp_net_check_config(struct nfp_net *nn, struct nfp_net_dp *dp, 1590 struct netlink_ext_ack *extack) 1591 { 1592 unsigned int r, xsk_min_fl_bufsz; 1593 1594 /* XDP-enabled tests */ 1595 if (!dp->xdp_prog) 1596 return 0; 1597 if (dp->fl_bufsz > PAGE_SIZE) { 1598 NL_SET_ERR_MSG_MOD(extack, "MTU too large w/ XDP enabled"); 1599 return -EINVAL; 1600 } 1601 if (dp->num_tx_rings > nn->max_tx_rings) { 1602 NL_SET_ERR_MSG_MOD(extack, "Insufficient number of TX rings w/ XDP enabled"); 1603 return -EINVAL; 1604 } 1605 1606 xsk_min_fl_bufsz = nfp_net_calc_fl_bufsz_xsk(dp); 1607 for (r = 0; r < nn->max_r_vecs; r++) { 1608 if (!dp->xsk_pools[r]) 1609 continue; 1610 1611 if (xsk_pool_get_rx_frame_size(dp->xsk_pools[r]) < xsk_min_fl_bufsz) { 1612 NL_SET_ERR_MSG_MOD(extack, 1613 "XSK buffer pool chunk size too small"); 1614 return -EINVAL; 1615 } 1616 } 1617 1618 return 0; 1619 } 1620 1621 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *dp, 1622 struct netlink_ext_ack *extack) 1623 { 1624 int r, err; 1625 1626 dp->fl_bufsz = nfp_net_calc_fl_bufsz(dp); 1627 1628 dp->num_stack_tx_rings = dp->num_tx_rings; 1629 if (dp->xdp_prog) 1630 dp->num_stack_tx_rings -= dp->num_rx_rings; 1631 1632 dp->num_r_vecs = max(dp->num_rx_rings, dp->num_stack_tx_rings); 1633 1634 err = nfp_net_check_config(nn, dp, extack); 1635 if (err) 1636 goto exit_free_dp; 1637 1638 if (!netif_running(dp->netdev)) { 1639 nfp_net_dp_swap(nn, dp); 1640 err = 0; 1641 goto exit_free_dp; 1642 } 1643 1644 /* Prepare new rings */ 1645 for (r = nn->dp.num_r_vecs; r < dp->num_r_vecs; r++) { 1646 err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r); 1647 if (err) { 1648 dp->num_r_vecs = r; 1649 goto err_cleanup_vecs; 1650 } 1651 } 1652 1653 err = nfp_net_rx_rings_prepare(nn, dp); 1654 if (err) 1655 goto err_cleanup_vecs; 1656 1657 err = nfp_net_tx_rings_prepare(nn, dp); 1658 if (err) 1659 goto err_free_rx; 1660 1661 /* Stop device, swap in new rings, try to start the firmware */ 1662 nfp_net_close_stack(nn); 1663 nfp_net_clear_config_and_disable(nn); 1664 1665 err = nfp_net_dp_swap_enable(nn, dp); 1666 if (err) { 1667 int err2; 1668 1669 nfp_net_clear_config_and_disable(nn); 1670 1671 /* Try with old configuration and old rings */ 1672 err2 = nfp_net_dp_swap_enable(nn, dp); 1673 if (err2) 1674 nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n", 1675 err, err2); 1676 } 1677 for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--) 1678 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1679 1680 nfp_net_rx_rings_free(dp); 1681 nfp_net_tx_rings_free(dp); 1682 1683 nfp_net_open_stack(nn); 1684 exit_free_dp: 1685 nfp_net_free_dp(dp); 1686 1687 return err; 1688 1689 err_free_rx: 1690 nfp_net_rx_rings_free(dp); 1691 err_cleanup_vecs: 1692 for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--) 1693 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1694 nfp_net_free_dp(dp); 1695 return err; 1696 } 1697 1698 static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu) 1699 { 1700 struct nfp_net *nn = netdev_priv(netdev); 1701 struct nfp_net_dp *dp; 1702 int err; 1703 1704 err = nfp_app_check_mtu(nn->app, netdev, new_mtu); 1705 if (err) 1706 return err; 1707 1708 dp = nfp_net_clone_dp(nn); 1709 if (!dp) 1710 return -ENOMEM; 1711 1712 dp->mtu = new_mtu; 1713 1714 return nfp_net_ring_reconfig(nn, dp, NULL); 1715 } 1716 1717 static int 1718 nfp_net_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid) 1719 { 1720 const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_ADD; 1721 struct nfp_net *nn = netdev_priv(netdev); 1722 int err; 1723 1724 /* Priority tagged packets with vlan id 0 are processed by the 1725 * NFP as untagged packets 1726 */ 1727 if (!vid) 1728 return 0; 1729 1730 err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ); 1731 if (err) 1732 return err; 1733 1734 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid); 1735 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO, 1736 ETH_P_8021Q); 1737 1738 return nfp_net_mbox_reconfig_and_unlock(nn, cmd); 1739 } 1740 1741 static int 1742 nfp_net_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid) 1743 { 1744 const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_KILL; 1745 struct nfp_net *nn = netdev_priv(netdev); 1746 int err; 1747 1748 /* Priority tagged packets with vlan id 0 are processed by the 1749 * NFP as untagged packets 1750 */ 1751 if (!vid) 1752 return 0; 1753 1754 err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ); 1755 if (err) 1756 return err; 1757 1758 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid); 1759 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO, 1760 ETH_P_8021Q); 1761 1762 return nfp_net_mbox_reconfig_and_unlock(nn, cmd); 1763 } 1764 1765 static void nfp_net_stat64(struct net_device *netdev, 1766 struct rtnl_link_stats64 *stats) 1767 { 1768 struct nfp_net *nn = netdev_priv(netdev); 1769 int r; 1770 1771 /* Collect software stats */ 1772 for (r = 0; r < nn->max_r_vecs; r++) { 1773 struct nfp_net_r_vector *r_vec = &nn->r_vecs[r]; 1774 u64 data[3]; 1775 unsigned int start; 1776 1777 do { 1778 start = u64_stats_fetch_begin(&r_vec->rx_sync); 1779 data[0] = r_vec->rx_pkts; 1780 data[1] = r_vec->rx_bytes; 1781 data[2] = r_vec->rx_drops; 1782 } while (u64_stats_fetch_retry(&r_vec->rx_sync, start)); 1783 stats->rx_packets += data[0]; 1784 stats->rx_bytes += data[1]; 1785 stats->rx_dropped += data[2]; 1786 1787 do { 1788 start = u64_stats_fetch_begin(&r_vec->tx_sync); 1789 data[0] = r_vec->tx_pkts; 1790 data[1] = r_vec->tx_bytes; 1791 data[2] = r_vec->tx_errors; 1792 } while (u64_stats_fetch_retry(&r_vec->tx_sync, start)); 1793 stats->tx_packets += data[0]; 1794 stats->tx_bytes += data[1]; 1795 stats->tx_errors += data[2]; 1796 } 1797 1798 /* Add in device stats */ 1799 stats->multicast += nn_readq(nn, NFP_NET_CFG_STATS_RX_MC_FRAMES); 1800 stats->rx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_RX_DISCARDS); 1801 stats->rx_errors += nn_readq(nn, NFP_NET_CFG_STATS_RX_ERRORS); 1802 1803 stats->tx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_TX_DISCARDS); 1804 stats->tx_errors += nn_readq(nn, NFP_NET_CFG_STATS_TX_ERRORS); 1805 } 1806 1807 static int nfp_net_set_features(struct net_device *netdev, 1808 netdev_features_t features) 1809 { 1810 netdev_features_t changed = netdev->features ^ features; 1811 struct nfp_net *nn = netdev_priv(netdev); 1812 u32 new_ctrl; 1813 int err; 1814 1815 /* Assume this is not called with features we have not advertised */ 1816 1817 new_ctrl = nn->dp.ctrl; 1818 1819 if (changed & NETIF_F_RXCSUM) { 1820 if (features & NETIF_F_RXCSUM) 1821 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY; 1822 else 1823 new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM_ANY; 1824 } 1825 1826 if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 1827 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) 1828 new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM; 1829 else 1830 new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM; 1831 } 1832 1833 if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) { 1834 if (features & (NETIF_F_TSO | NETIF_F_TSO6)) 1835 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?: 1836 NFP_NET_CFG_CTRL_LSO; 1837 else 1838 new_ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY; 1839 } 1840 1841 if (changed & NETIF_F_HW_VLAN_CTAG_RX) { 1842 if (features & NETIF_F_HW_VLAN_CTAG_RX) 1843 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ?: 1844 NFP_NET_CFG_CTRL_RXVLAN; 1845 else 1846 new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN_ANY; 1847 } 1848 1849 if (changed & NETIF_F_HW_VLAN_CTAG_TX) { 1850 if (features & NETIF_F_HW_VLAN_CTAG_TX) 1851 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ?: 1852 NFP_NET_CFG_CTRL_TXVLAN; 1853 else 1854 new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN_ANY; 1855 } 1856 1857 if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) { 1858 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) 1859 new_ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER; 1860 else 1861 new_ctrl &= ~NFP_NET_CFG_CTRL_CTAG_FILTER; 1862 } 1863 1864 if (changed & NETIF_F_HW_VLAN_STAG_RX) { 1865 if (features & NETIF_F_HW_VLAN_STAG_RX) 1866 new_ctrl |= NFP_NET_CFG_CTRL_RXQINQ; 1867 else 1868 new_ctrl &= ~NFP_NET_CFG_CTRL_RXQINQ; 1869 } 1870 1871 if (changed & NETIF_F_SG) { 1872 if (features & NETIF_F_SG) 1873 new_ctrl |= NFP_NET_CFG_CTRL_GATHER; 1874 else 1875 new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER; 1876 } 1877 1878 err = nfp_port_set_features(netdev, features); 1879 if (err) 1880 return err; 1881 1882 nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n", 1883 netdev->features, features, changed); 1884 1885 if (new_ctrl == nn->dp.ctrl) 1886 return 0; 1887 1888 nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->dp.ctrl, new_ctrl); 1889 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1890 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN); 1891 if (err) 1892 return err; 1893 1894 nn->dp.ctrl = new_ctrl; 1895 1896 return 0; 1897 } 1898 1899 static netdev_features_t 1900 nfp_net_fix_features(struct net_device *netdev, 1901 netdev_features_t features) 1902 { 1903 if ((features & NETIF_F_HW_VLAN_CTAG_RX) && 1904 (features & NETIF_F_HW_VLAN_STAG_RX)) { 1905 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) { 1906 features &= ~NETIF_F_HW_VLAN_CTAG_RX; 1907 netdev->wanted_features &= ~NETIF_F_HW_VLAN_CTAG_RX; 1908 netdev_warn(netdev, 1909 "S-tag and C-tag stripping can't be enabled at the same time. Enabling S-tag stripping and disabling C-tag stripping\n"); 1910 } else if (netdev->features & NETIF_F_HW_VLAN_STAG_RX) { 1911 features &= ~NETIF_F_HW_VLAN_STAG_RX; 1912 netdev->wanted_features &= ~NETIF_F_HW_VLAN_STAG_RX; 1913 netdev_warn(netdev, 1914 "S-tag and C-tag stripping can't be enabled at the same time. Enabling C-tag stripping and disabling S-tag stripping\n"); 1915 } 1916 } 1917 return features; 1918 } 1919 1920 static netdev_features_t 1921 nfp_net_features_check(struct sk_buff *skb, struct net_device *dev, 1922 netdev_features_t features) 1923 { 1924 u8 l4_hdr; 1925 1926 /* We can't do TSO over double tagged packets (802.1AD) */ 1927 features &= vlan_features_check(skb, features); 1928 1929 if (!skb->encapsulation) 1930 return features; 1931 1932 /* Ensure that inner L4 header offset fits into TX descriptor field */ 1933 if (skb_is_gso(skb)) { 1934 u32 hdrlen; 1935 1936 hdrlen = skb_inner_tcp_all_headers(skb); 1937 1938 /* Assume worst case scenario of having longest possible 1939 * metadata prepend - 8B 1940 */ 1941 if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ - 8)) 1942 features &= ~NETIF_F_GSO_MASK; 1943 } 1944 1945 if (xfrm_offload(skb)) 1946 return features; 1947 1948 /* VXLAN/GRE check */ 1949 switch (vlan_get_protocol(skb)) { 1950 case htons(ETH_P_IP): 1951 l4_hdr = ip_hdr(skb)->protocol; 1952 break; 1953 case htons(ETH_P_IPV6): 1954 l4_hdr = ipv6_hdr(skb)->nexthdr; 1955 break; 1956 default: 1957 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 1958 } 1959 1960 if (skb->inner_protocol_type != ENCAP_TYPE_ETHER || 1961 skb->inner_protocol != htons(ETH_P_TEB) || 1962 (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) || 1963 (l4_hdr == IPPROTO_UDP && 1964 (skb_inner_mac_header(skb) - skb_transport_header(skb) != 1965 sizeof(struct udphdr) + sizeof(struct vxlanhdr)))) 1966 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 1967 1968 return features; 1969 } 1970 1971 static int 1972 nfp_net_get_phys_port_name(struct net_device *netdev, char *name, size_t len) 1973 { 1974 struct nfp_net *nn = netdev_priv(netdev); 1975 int n; 1976 1977 /* If port is defined, devlink_port is registered and devlink core 1978 * is taking care of name formatting. 1979 */ 1980 if (nn->port) 1981 return -EOPNOTSUPP; 1982 1983 if (nn->dp.is_vf || nn->vnic_no_name) 1984 return -EOPNOTSUPP; 1985 1986 n = snprintf(name, len, "n%d", nn->id); 1987 if (n >= len) 1988 return -EINVAL; 1989 1990 return 0; 1991 } 1992 1993 static int nfp_net_xdp_setup_drv(struct nfp_net *nn, struct netdev_bpf *bpf) 1994 { 1995 struct bpf_prog *prog = bpf->prog; 1996 struct nfp_net_dp *dp; 1997 int err; 1998 1999 if (!prog == !nn->dp.xdp_prog) { 2000 WRITE_ONCE(nn->dp.xdp_prog, prog); 2001 xdp_attachment_setup(&nn->xdp, bpf); 2002 return 0; 2003 } 2004 2005 dp = nfp_net_clone_dp(nn); 2006 if (!dp) 2007 return -ENOMEM; 2008 2009 dp->xdp_prog = prog; 2010 dp->num_tx_rings += prog ? nn->dp.num_rx_rings : -nn->dp.num_rx_rings; 2011 dp->rx_dma_dir = prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE; 2012 dp->rx_dma_off = prog ? XDP_PACKET_HEADROOM - nn->dp.rx_offset : 0; 2013 2014 /* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */ 2015 err = nfp_net_ring_reconfig(nn, dp, bpf->extack); 2016 if (err) 2017 return err; 2018 2019 xdp_attachment_setup(&nn->xdp, bpf); 2020 return 0; 2021 } 2022 2023 static int nfp_net_xdp_setup_hw(struct nfp_net *nn, struct netdev_bpf *bpf) 2024 { 2025 int err; 2026 2027 err = nfp_app_xdp_offload(nn->app, nn, bpf->prog, bpf->extack); 2028 if (err) 2029 return err; 2030 2031 xdp_attachment_setup(&nn->xdp_hw, bpf); 2032 return 0; 2033 } 2034 2035 static int nfp_net_xdp(struct net_device *netdev, struct netdev_bpf *xdp) 2036 { 2037 struct nfp_net *nn = netdev_priv(netdev); 2038 2039 switch (xdp->command) { 2040 case XDP_SETUP_PROG: 2041 return nfp_net_xdp_setup_drv(nn, xdp); 2042 case XDP_SETUP_PROG_HW: 2043 return nfp_net_xdp_setup_hw(nn, xdp); 2044 case XDP_SETUP_XSK_POOL: 2045 return nfp_net_xsk_setup_pool(netdev, xdp->xsk.pool, 2046 xdp->xsk.queue_id); 2047 default: 2048 return nfp_app_bpf(nn->app, nn, xdp); 2049 } 2050 } 2051 2052 static int nfp_net_set_mac_address(struct net_device *netdev, void *addr) 2053 { 2054 struct nfp_net *nn = netdev_priv(netdev); 2055 struct sockaddr *saddr = addr; 2056 int err; 2057 2058 err = eth_prepare_mac_addr_change(netdev, addr); 2059 if (err) 2060 return err; 2061 2062 nfp_net_write_mac_addr(nn, saddr->sa_data); 2063 2064 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MACADDR); 2065 if (err) 2066 return err; 2067 2068 eth_commit_mac_addr_change(netdev, addr); 2069 2070 return 0; 2071 } 2072 2073 static int nfp_net_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq, 2074 struct net_device *dev, u32 filter_mask, 2075 int nlflags) 2076 { 2077 struct nfp_net *nn = netdev_priv(dev); 2078 u16 mode; 2079 2080 if (!(nn->cap & NFP_NET_CFG_CTRL_VEPA)) 2081 return -EOPNOTSUPP; 2082 2083 mode = (nn->dp.ctrl & NFP_NET_CFG_CTRL_VEPA) ? 2084 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB; 2085 2086 return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode, 0, 0, 2087 nlflags, filter_mask, NULL); 2088 } 2089 2090 static int nfp_net_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh, 2091 u16 flags, struct netlink_ext_ack *extack) 2092 { 2093 struct nfp_net *nn = netdev_priv(dev); 2094 struct nlattr *attr, *br_spec; 2095 int rem, err; 2096 u32 new_ctrl; 2097 u16 mode; 2098 2099 if (!(nn->cap & NFP_NET_CFG_CTRL_VEPA)) 2100 return -EOPNOTSUPP; 2101 2102 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC); 2103 if (!br_spec) 2104 return -EINVAL; 2105 2106 nla_for_each_nested(attr, br_spec, rem) { 2107 if (nla_type(attr) != IFLA_BRIDGE_MODE) 2108 continue; 2109 2110 new_ctrl = nn->dp.ctrl; 2111 mode = nla_get_u16(attr); 2112 if (mode == BRIDGE_MODE_VEPA) 2113 new_ctrl |= NFP_NET_CFG_CTRL_VEPA; 2114 else if (mode == BRIDGE_MODE_VEB) 2115 new_ctrl &= ~NFP_NET_CFG_CTRL_VEPA; 2116 else 2117 return -EOPNOTSUPP; 2118 2119 if (new_ctrl == nn->dp.ctrl) 2120 return 0; 2121 2122 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 2123 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN); 2124 if (!err) 2125 nn->dp.ctrl = new_ctrl; 2126 2127 return err; 2128 } 2129 2130 return -EINVAL; 2131 } 2132 2133 const struct net_device_ops nfp_nfd3_netdev_ops = { 2134 .ndo_init = nfp_app_ndo_init, 2135 .ndo_uninit = nfp_app_ndo_uninit, 2136 .ndo_open = nfp_net_netdev_open, 2137 .ndo_stop = nfp_net_netdev_close, 2138 .ndo_start_xmit = nfp_net_tx, 2139 .ndo_get_stats64 = nfp_net_stat64, 2140 .ndo_vlan_rx_add_vid = nfp_net_vlan_rx_add_vid, 2141 .ndo_vlan_rx_kill_vid = nfp_net_vlan_rx_kill_vid, 2142 .ndo_set_vf_mac = nfp_app_set_vf_mac, 2143 .ndo_set_vf_vlan = nfp_app_set_vf_vlan, 2144 .ndo_set_vf_rate = nfp_app_set_vf_rate, 2145 .ndo_set_vf_spoofchk = nfp_app_set_vf_spoofchk, 2146 .ndo_set_vf_trust = nfp_app_set_vf_trust, 2147 .ndo_get_vf_config = nfp_app_get_vf_config, 2148 .ndo_set_vf_link_state = nfp_app_set_vf_link_state, 2149 .ndo_setup_tc = nfp_port_setup_tc, 2150 .ndo_tx_timeout = nfp_net_tx_timeout, 2151 .ndo_set_rx_mode = nfp_net_set_rx_mode, 2152 .ndo_change_mtu = nfp_net_change_mtu, 2153 .ndo_set_mac_address = nfp_net_set_mac_address, 2154 .ndo_set_features = nfp_net_set_features, 2155 .ndo_fix_features = nfp_net_fix_features, 2156 .ndo_features_check = nfp_net_features_check, 2157 .ndo_get_phys_port_name = nfp_net_get_phys_port_name, 2158 .ndo_bpf = nfp_net_xdp, 2159 .ndo_xsk_wakeup = nfp_net_xsk_wakeup, 2160 .ndo_bridge_getlink = nfp_net_bridge_getlink, 2161 .ndo_bridge_setlink = nfp_net_bridge_setlink, 2162 }; 2163 2164 const struct net_device_ops nfp_nfdk_netdev_ops = { 2165 .ndo_init = nfp_app_ndo_init, 2166 .ndo_uninit = nfp_app_ndo_uninit, 2167 .ndo_open = nfp_net_netdev_open, 2168 .ndo_stop = nfp_net_netdev_close, 2169 .ndo_start_xmit = nfp_net_tx, 2170 .ndo_get_stats64 = nfp_net_stat64, 2171 .ndo_vlan_rx_add_vid = nfp_net_vlan_rx_add_vid, 2172 .ndo_vlan_rx_kill_vid = nfp_net_vlan_rx_kill_vid, 2173 .ndo_set_vf_mac = nfp_app_set_vf_mac, 2174 .ndo_set_vf_vlan = nfp_app_set_vf_vlan, 2175 .ndo_set_vf_rate = nfp_app_set_vf_rate, 2176 .ndo_set_vf_spoofchk = nfp_app_set_vf_spoofchk, 2177 .ndo_set_vf_trust = nfp_app_set_vf_trust, 2178 .ndo_get_vf_config = nfp_app_get_vf_config, 2179 .ndo_set_vf_link_state = nfp_app_set_vf_link_state, 2180 .ndo_setup_tc = nfp_port_setup_tc, 2181 .ndo_tx_timeout = nfp_net_tx_timeout, 2182 .ndo_set_rx_mode = nfp_net_set_rx_mode, 2183 .ndo_change_mtu = nfp_net_change_mtu, 2184 .ndo_set_mac_address = nfp_net_set_mac_address, 2185 .ndo_set_features = nfp_net_set_features, 2186 .ndo_fix_features = nfp_net_fix_features, 2187 .ndo_features_check = nfp_net_features_check, 2188 .ndo_get_phys_port_name = nfp_net_get_phys_port_name, 2189 .ndo_bpf = nfp_net_xdp, 2190 .ndo_bridge_getlink = nfp_net_bridge_getlink, 2191 .ndo_bridge_setlink = nfp_net_bridge_setlink, 2192 }; 2193 2194 static int nfp_udp_tunnel_sync(struct net_device *netdev, unsigned int table) 2195 { 2196 struct nfp_net *nn = netdev_priv(netdev); 2197 int i; 2198 2199 BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1); 2200 for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2) { 2201 struct udp_tunnel_info ti0, ti1; 2202 2203 udp_tunnel_nic_get_port(netdev, table, i, &ti0); 2204 udp_tunnel_nic_get_port(netdev, table, i + 1, &ti1); 2205 2206 nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(ti0.port), 2207 be16_to_cpu(ti1.port) << 16 | be16_to_cpu(ti0.port)); 2208 } 2209 2210 return nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_VXLAN); 2211 } 2212 2213 static const struct udp_tunnel_nic_info nfp_udp_tunnels = { 2214 .sync_table = nfp_udp_tunnel_sync, 2215 .flags = UDP_TUNNEL_NIC_INFO_MAY_SLEEP | 2216 UDP_TUNNEL_NIC_INFO_OPEN_ONLY, 2217 .tables = { 2218 { 2219 .n_entries = NFP_NET_N_VXLAN_PORTS, 2220 .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, 2221 }, 2222 }, 2223 }; 2224 2225 /** 2226 * nfp_net_info() - Print general info about the NIC 2227 * @nn: NFP Net device to reconfigure 2228 */ 2229 void nfp_net_info(struct nfp_net *nn) 2230 { 2231 nn_info(nn, "NFP-6xxx %sNetdev: TxQs=%d/%d RxQs=%d/%d\n", 2232 nn->dp.is_vf ? "VF " : "", 2233 nn->dp.num_tx_rings, nn->max_tx_rings, 2234 nn->dp.num_rx_rings, nn->max_rx_rings); 2235 nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n", 2236 nn->fw_ver.extend, nn->fw_ver.class, 2237 nn->fw_ver.major, nn->fw_ver.minor, 2238 nn->max_mtu); 2239 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", 2240 nn->cap, 2241 nn->cap & NFP_NET_CFG_CTRL_PROMISC ? "PROMISC " : "", 2242 nn->cap & NFP_NET_CFG_CTRL_L2BC ? "L2BCFILT " : "", 2243 nn->cap & NFP_NET_CFG_CTRL_L2MC ? "L2MCFILT " : "", 2244 nn->cap & NFP_NET_CFG_CTRL_RXCSUM ? "RXCSUM " : "", 2245 nn->cap & NFP_NET_CFG_CTRL_TXCSUM ? "TXCSUM " : "", 2246 nn->cap & NFP_NET_CFG_CTRL_RXVLAN ? "RXVLAN " : "", 2247 nn->cap & NFP_NET_CFG_CTRL_TXVLAN ? "TXVLAN " : "", 2248 nn->cap & NFP_NET_CFG_CTRL_RXQINQ ? "RXQINQ " : "", 2249 nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ? "RXVLANv2 " : "", 2250 nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ? "TXVLANv2 " : "", 2251 nn->cap & NFP_NET_CFG_CTRL_SCATTER ? "SCATTER " : "", 2252 nn->cap & NFP_NET_CFG_CTRL_GATHER ? "GATHER " : "", 2253 nn->cap & NFP_NET_CFG_CTRL_LSO ? "TSO1 " : "", 2254 nn->cap & NFP_NET_CFG_CTRL_LSO2 ? "TSO2 " : "", 2255 nn->cap & NFP_NET_CFG_CTRL_RSS ? "RSS1 " : "", 2256 nn->cap & NFP_NET_CFG_CTRL_RSS2 ? "RSS2 " : "", 2257 nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER ? "CTAG_FILTER " : "", 2258 nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "", 2259 nn->cap & NFP_NET_CFG_CTRL_IRQMOD ? "IRQMOD " : "", 2260 nn->cap & NFP_NET_CFG_CTRL_TXRWB ? "TXRWB " : "", 2261 nn->cap & NFP_NET_CFG_CTRL_VEPA ? "VEPA " : "", 2262 nn->cap & NFP_NET_CFG_CTRL_VXLAN ? "VXLAN " : "", 2263 nn->cap & NFP_NET_CFG_CTRL_NVGRE ? "NVGRE " : "", 2264 nn->cap & NFP_NET_CFG_CTRL_CSUM_COMPLETE ? 2265 "RXCSUM_COMPLETE " : "", 2266 nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR ? "LIVE_ADDR " : "", 2267 nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER ? "MULTICAST_FILTER " : "", 2268 nfp_app_extra_cap(nn->app, nn)); 2269 } 2270 2271 /** 2272 * nfp_net_alloc() - Allocate netdev and related structure 2273 * @pdev: PCI device 2274 * @dev_info: NFP ASIC params 2275 * @ctrl_bar: PCI IOMEM with vNIC config memory 2276 * @needs_netdev: Whether to allocate a netdev for this vNIC 2277 * @max_tx_rings: Maximum number of TX rings supported by device 2278 * @max_rx_rings: Maximum number of RX rings supported by device 2279 * 2280 * This function allocates a netdev device and fills in the initial 2281 * part of the @struct nfp_net structure. In case of control device 2282 * nfp_net structure is allocated without the netdev. 2283 * 2284 * Return: NFP Net device structure, or ERR_PTR on error. 2285 */ 2286 struct nfp_net * 2287 nfp_net_alloc(struct pci_dev *pdev, const struct nfp_dev_info *dev_info, 2288 void __iomem *ctrl_bar, bool needs_netdev, 2289 unsigned int max_tx_rings, unsigned int max_rx_rings) 2290 { 2291 u64 dma_mask = dma_get_mask(&pdev->dev); 2292 struct nfp_net *nn; 2293 int err; 2294 2295 if (needs_netdev) { 2296 struct net_device *netdev; 2297 2298 netdev = alloc_etherdev_mqs(sizeof(struct nfp_net), 2299 max_tx_rings, max_rx_rings); 2300 if (!netdev) 2301 return ERR_PTR(-ENOMEM); 2302 2303 SET_NETDEV_DEV(netdev, &pdev->dev); 2304 nn = netdev_priv(netdev); 2305 nn->dp.netdev = netdev; 2306 } else { 2307 nn = vzalloc(sizeof(*nn)); 2308 if (!nn) 2309 return ERR_PTR(-ENOMEM); 2310 } 2311 2312 nn->dp.dev = &pdev->dev; 2313 nn->dp.ctrl_bar = ctrl_bar; 2314 nn->dev_info = dev_info; 2315 nn->pdev = pdev; 2316 nfp_net_get_fw_version(&nn->fw_ver, ctrl_bar); 2317 2318 switch (FIELD_GET(NFP_NET_CFG_VERSION_DP_MASK, nn->fw_ver.extend)) { 2319 case NFP_NET_CFG_VERSION_DP_NFD3: 2320 nn->dp.ops = &nfp_nfd3_ops; 2321 break; 2322 case NFP_NET_CFG_VERSION_DP_NFDK: 2323 if (nn->fw_ver.major < 5) { 2324 dev_err(&pdev->dev, 2325 "NFDK must use ABI 5 or newer, found: %d\n", 2326 nn->fw_ver.major); 2327 err = -EINVAL; 2328 goto err_free_nn; 2329 } 2330 nn->dp.ops = &nfp_nfdk_ops; 2331 break; 2332 default: 2333 err = -EINVAL; 2334 goto err_free_nn; 2335 } 2336 2337 if ((dma_mask & nn->dp.ops->dma_mask) != dma_mask) { 2338 dev_err(&pdev->dev, 2339 "DMA mask of loaded firmware: %llx, required DMA mask: %llx\n", 2340 nn->dp.ops->dma_mask, dma_mask); 2341 err = -EINVAL; 2342 goto err_free_nn; 2343 } 2344 2345 nn->max_tx_rings = max_tx_rings; 2346 nn->max_rx_rings = max_rx_rings; 2347 2348 nn->dp.num_tx_rings = min_t(unsigned int, 2349 max_tx_rings, num_online_cpus()); 2350 nn->dp.num_rx_rings = min_t(unsigned int, max_rx_rings, 2351 netif_get_num_default_rss_queues()); 2352 2353 nn->dp.num_r_vecs = max(nn->dp.num_tx_rings, nn->dp.num_rx_rings); 2354 nn->dp.num_r_vecs = min_t(unsigned int, 2355 nn->dp.num_r_vecs, num_online_cpus()); 2356 nn->max_r_vecs = nn->dp.num_r_vecs; 2357 2358 nn->dp.xsk_pools = kcalloc(nn->max_r_vecs, sizeof(nn->dp.xsk_pools), 2359 GFP_KERNEL); 2360 if (!nn->dp.xsk_pools) { 2361 err = -ENOMEM; 2362 goto err_free_nn; 2363 } 2364 2365 nn->dp.txd_cnt = NFP_NET_TX_DESCS_DEFAULT; 2366 nn->dp.rxd_cnt = NFP_NET_RX_DESCS_DEFAULT; 2367 2368 sema_init(&nn->bar_lock, 1); 2369 2370 spin_lock_init(&nn->reconfig_lock); 2371 spin_lock_init(&nn->link_status_lock); 2372 2373 timer_setup(&nn->reconfig_timer, nfp_net_reconfig_timer, 0); 2374 2375 err = nfp_net_tlv_caps_parse(&nn->pdev->dev, nn->dp.ctrl_bar, 2376 &nn->tlv_caps); 2377 if (err) 2378 goto err_free_nn; 2379 2380 err = nfp_ccm_mbox_alloc(nn); 2381 if (err) 2382 goto err_free_nn; 2383 2384 return nn; 2385 2386 err_free_nn: 2387 if (nn->dp.netdev) 2388 free_netdev(nn->dp.netdev); 2389 else 2390 vfree(nn); 2391 return ERR_PTR(err); 2392 } 2393 2394 /** 2395 * nfp_net_free() - Undo what @nfp_net_alloc() did 2396 * @nn: NFP Net device to reconfigure 2397 */ 2398 void nfp_net_free(struct nfp_net *nn) 2399 { 2400 WARN_ON(timer_pending(&nn->reconfig_timer) || nn->reconfig_posted); 2401 nfp_ccm_mbox_free(nn); 2402 2403 kfree(nn->dp.xsk_pools); 2404 if (nn->dp.netdev) 2405 free_netdev(nn->dp.netdev); 2406 else 2407 vfree(nn); 2408 } 2409 2410 /** 2411 * nfp_net_rss_key_sz() - Get current size of the RSS key 2412 * @nn: NFP Net device instance 2413 * 2414 * Return: size of the RSS key for currently selected hash function. 2415 */ 2416 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn) 2417 { 2418 switch (nn->rss_hfunc) { 2419 case ETH_RSS_HASH_TOP: 2420 return NFP_NET_CFG_RSS_KEY_SZ; 2421 case ETH_RSS_HASH_XOR: 2422 return 0; 2423 case ETH_RSS_HASH_CRC32: 2424 return 4; 2425 } 2426 2427 nn_warn(nn, "Unknown hash function: %u\n", nn->rss_hfunc); 2428 return 0; 2429 } 2430 2431 /** 2432 * nfp_net_rss_init() - Set the initial RSS parameters 2433 * @nn: NFP Net device to reconfigure 2434 */ 2435 static void nfp_net_rss_init(struct nfp_net *nn) 2436 { 2437 unsigned long func_bit, rss_cap_hfunc; 2438 u32 reg; 2439 2440 /* Read the RSS function capability and select first supported func */ 2441 reg = nn_readl(nn, NFP_NET_CFG_RSS_CAP); 2442 rss_cap_hfunc = FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, reg); 2443 if (!rss_cap_hfunc) 2444 rss_cap_hfunc = FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, 2445 NFP_NET_CFG_RSS_TOEPLITZ); 2446 2447 func_bit = find_first_bit(&rss_cap_hfunc, NFP_NET_CFG_RSS_HFUNCS); 2448 if (func_bit == NFP_NET_CFG_RSS_HFUNCS) { 2449 dev_warn(nn->dp.dev, 2450 "Bad RSS config, defaulting to Toeplitz hash\n"); 2451 func_bit = ETH_RSS_HASH_TOP_BIT; 2452 } 2453 nn->rss_hfunc = 1 << func_bit; 2454 2455 netdev_rss_key_fill(nn->rss_key, nfp_net_rss_key_sz(nn)); 2456 2457 nfp_net_rss_init_itbl(nn); 2458 2459 /* Enable IPv4/IPv6 TCP by default */ 2460 nn->rss_cfg = NFP_NET_CFG_RSS_IPV4_TCP | 2461 NFP_NET_CFG_RSS_IPV6_TCP | 2462 NFP_NET_CFG_RSS_IPV4_UDP | 2463 NFP_NET_CFG_RSS_IPV6_UDP | 2464 FIELD_PREP(NFP_NET_CFG_RSS_HFUNC, nn->rss_hfunc) | 2465 NFP_NET_CFG_RSS_MASK; 2466 } 2467 2468 /** 2469 * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters 2470 * @nn: NFP Net device to reconfigure 2471 */ 2472 static void nfp_net_irqmod_init(struct nfp_net *nn) 2473 { 2474 nn->rx_coalesce_usecs = 50; 2475 nn->rx_coalesce_max_frames = 64; 2476 nn->tx_coalesce_usecs = 50; 2477 nn->tx_coalesce_max_frames = 64; 2478 2479 nn->rx_coalesce_adapt_on = true; 2480 nn->tx_coalesce_adapt_on = true; 2481 } 2482 2483 static void nfp_net_netdev_init(struct nfp_net *nn) 2484 { 2485 struct net_device *netdev = nn->dp.netdev; 2486 2487 nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr); 2488 2489 netdev->mtu = nn->dp.mtu; 2490 2491 /* Advertise/enable offloads based on capabilities 2492 * 2493 * Note: netdev->features show the currently enabled features 2494 * and netdev->hw_features advertises which features are 2495 * supported. By default we enable most features. 2496 */ 2497 if (nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR) 2498 netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 2499 2500 netdev->hw_features = NETIF_F_HIGHDMA; 2501 if (nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY) { 2502 netdev->hw_features |= NETIF_F_RXCSUM; 2503 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY; 2504 } 2505 if (nn->cap & NFP_NET_CFG_CTRL_TXCSUM) { 2506 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; 2507 nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXCSUM; 2508 } 2509 if (nn->cap & NFP_NET_CFG_CTRL_GATHER) { 2510 netdev->hw_features |= NETIF_F_SG; 2511 nn->dp.ctrl |= NFP_NET_CFG_CTRL_GATHER; 2512 } 2513 if ((nn->cap & NFP_NET_CFG_CTRL_LSO && nn->fw_ver.major > 2) || 2514 nn->cap & NFP_NET_CFG_CTRL_LSO2) { 2515 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6; 2516 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?: 2517 NFP_NET_CFG_CTRL_LSO; 2518 } 2519 if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) 2520 netdev->hw_features |= NETIF_F_RXHASH; 2521 2522 #ifdef CONFIG_NFP_NET_IPSEC 2523 if (nn->cap_w1 & NFP_NET_CFG_CTRL_IPSEC) 2524 netdev->hw_features |= NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM; 2525 #endif 2526 2527 if (nn->cap & NFP_NET_CFG_CTRL_VXLAN) { 2528 if (nn->cap & NFP_NET_CFG_CTRL_LSO) { 2529 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL | 2530 NETIF_F_GSO_UDP_TUNNEL_CSUM | 2531 NETIF_F_GSO_PARTIAL; 2532 netdev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM; 2533 } 2534 netdev->udp_tunnel_nic_info = &nfp_udp_tunnels; 2535 nn->dp.ctrl |= NFP_NET_CFG_CTRL_VXLAN; 2536 } 2537 if (nn->cap & NFP_NET_CFG_CTRL_NVGRE) { 2538 if (nn->cap & NFP_NET_CFG_CTRL_LSO) 2539 netdev->hw_features |= NETIF_F_GSO_GRE; 2540 nn->dp.ctrl |= NFP_NET_CFG_CTRL_NVGRE; 2541 } 2542 if (nn->cap & (NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE)) 2543 netdev->hw_enc_features = netdev->hw_features; 2544 2545 netdev->vlan_features = netdev->hw_features; 2546 2547 if (nn->cap & NFP_NET_CFG_CTRL_RXVLAN_ANY) { 2548 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX; 2549 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ?: 2550 NFP_NET_CFG_CTRL_RXVLAN; 2551 } 2552 if (nn->cap & NFP_NET_CFG_CTRL_TXVLAN_ANY) { 2553 if (nn->cap & NFP_NET_CFG_CTRL_LSO2) { 2554 nn_warn(nn, "Device advertises both TSO2 and TXVLAN. Refusing to enable TXVLAN.\n"); 2555 } else { 2556 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX; 2557 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ?: 2558 NFP_NET_CFG_CTRL_TXVLAN; 2559 } 2560 } 2561 if (nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER) { 2562 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER; 2563 nn->dp.ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER; 2564 } 2565 if (nn->cap & NFP_NET_CFG_CTRL_RXQINQ) { 2566 netdev->hw_features |= NETIF_F_HW_VLAN_STAG_RX; 2567 nn->dp.ctrl |= NFP_NET_CFG_CTRL_RXQINQ; 2568 } 2569 2570 netdev->features = netdev->hw_features; 2571 2572 if (nfp_app_has_tc(nn->app) && nn->port) 2573 netdev->hw_features |= NETIF_F_HW_TC; 2574 2575 /* C-Tag strip and S-Tag strip can't be supported simultaneously, 2576 * so enable C-Tag strip and disable S-Tag strip by default. 2577 */ 2578 netdev->features &= ~NETIF_F_HW_VLAN_STAG_RX; 2579 nn->dp.ctrl &= ~NFP_NET_CFG_CTRL_RXQINQ; 2580 2581 netdev->xdp_features = NETDEV_XDP_ACT_BASIC; 2582 if (nn->app && nn->app->type->id == NFP_APP_BPF_NIC) 2583 netdev->xdp_features |= NETDEV_XDP_ACT_HW_OFFLOAD; 2584 2585 /* Finalise the netdev setup */ 2586 switch (nn->dp.ops->version) { 2587 case NFP_NFD_VER_NFD3: 2588 netdev->netdev_ops = &nfp_nfd3_netdev_ops; 2589 netdev->xdp_features |= NETDEV_XDP_ACT_XSK_ZEROCOPY; 2590 netdev->xdp_features |= NETDEV_XDP_ACT_REDIRECT; 2591 break; 2592 case NFP_NFD_VER_NFDK: 2593 netdev->netdev_ops = &nfp_nfdk_netdev_ops; 2594 break; 2595 } 2596 2597 netdev->watchdog_timeo = msecs_to_jiffies(5 * 1000); 2598 2599 /* MTU range: 68 - hw-specific max */ 2600 netdev->min_mtu = ETH_MIN_MTU; 2601 netdev->max_mtu = nn->max_mtu; 2602 2603 netif_set_tso_max_segs(netdev, NFP_NET_LSO_MAX_SEGS); 2604 2605 netif_carrier_off(netdev); 2606 2607 nfp_net_set_ethtool_ops(netdev); 2608 } 2609 2610 static int nfp_net_read_caps(struct nfp_net *nn) 2611 { 2612 /* Get some of the read-only fields from the BAR */ 2613 nn->cap = nn_readl(nn, NFP_NET_CFG_CAP); 2614 nn->cap_w1 = nn_readl(nn, NFP_NET_CFG_CAP_WORD1); 2615 nn->max_mtu = nn_readl(nn, NFP_NET_CFG_MAX_MTU); 2616 2617 /* ABI 4.x and ctrl vNIC always use chained metadata, in other cases 2618 * we allow use of non-chained metadata if RSS(v1) is the only 2619 * advertised capability requiring metadata. 2620 */ 2621 nn->dp.chained_metadata_format = nn->fw_ver.major == 4 || 2622 !nn->dp.netdev || 2623 !(nn->cap & NFP_NET_CFG_CTRL_RSS) || 2624 nn->cap & NFP_NET_CFG_CTRL_CHAIN_META; 2625 /* RSS(v1) uses non-chained metadata format, except in ABI 4.x where 2626 * it has the same meaning as RSSv2. 2627 */ 2628 if (nn->dp.chained_metadata_format && nn->fw_ver.major != 4) 2629 nn->cap &= ~NFP_NET_CFG_CTRL_RSS; 2630 2631 /* Determine RX packet/metadata boundary offset */ 2632 if (nn->fw_ver.major >= 2) { 2633 u32 reg; 2634 2635 reg = nn_readl(nn, NFP_NET_CFG_RX_OFFSET); 2636 if (reg > NFP_NET_MAX_PREPEND) { 2637 nn_err(nn, "Invalid rx offset: %d\n", reg); 2638 return -EINVAL; 2639 } 2640 nn->dp.rx_offset = reg; 2641 } else { 2642 nn->dp.rx_offset = NFP_NET_RX_OFFSET; 2643 } 2644 2645 /* Mask out NFD-version-specific features */ 2646 nn->cap &= nn->dp.ops->cap_mask; 2647 2648 /* For control vNICs mask out the capabilities app doesn't want. */ 2649 if (!nn->dp.netdev) 2650 nn->cap &= nn->app->type->ctrl_cap_mask; 2651 2652 return 0; 2653 } 2654 2655 /** 2656 * nfp_net_init() - Initialise/finalise the nfp_net structure 2657 * @nn: NFP Net device structure 2658 * 2659 * Return: 0 on success or negative errno on error. 2660 */ 2661 int nfp_net_init(struct nfp_net *nn) 2662 { 2663 int err; 2664 2665 nn->dp.rx_dma_dir = DMA_FROM_DEVICE; 2666 2667 err = nfp_net_read_caps(nn); 2668 if (err) 2669 return err; 2670 2671 /* Set default MTU and Freelist buffer size */ 2672 if (!nfp_net_is_data_vnic(nn) && nn->app->ctrl_mtu) { 2673 nn->dp.mtu = min(nn->app->ctrl_mtu, nn->max_mtu); 2674 } else if (nn->max_mtu < NFP_NET_DEFAULT_MTU) { 2675 nn->dp.mtu = nn->max_mtu; 2676 } else { 2677 nn->dp.mtu = NFP_NET_DEFAULT_MTU; 2678 } 2679 nn->dp.fl_bufsz = nfp_net_calc_fl_bufsz(&nn->dp); 2680 2681 if (nfp_app_ctrl_uses_data_vnics(nn->app)) 2682 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_CMSG_DATA; 2683 2684 if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) { 2685 nfp_net_rss_init(nn); 2686 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RSS2 ?: 2687 NFP_NET_CFG_CTRL_RSS; 2688 } 2689 2690 /* Allow L2 Broadcast and Multicast through by default, if supported */ 2691 if (nn->cap & NFP_NET_CFG_CTRL_L2BC) 2692 nn->dp.ctrl |= NFP_NET_CFG_CTRL_L2BC; 2693 2694 /* Allow IRQ moderation, if supported */ 2695 if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) { 2696 nfp_net_irqmod_init(nn); 2697 nn->dp.ctrl |= NFP_NET_CFG_CTRL_IRQMOD; 2698 } 2699 2700 /* Enable TX pointer writeback, if supported */ 2701 if (nn->cap & NFP_NET_CFG_CTRL_TXRWB) 2702 nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXRWB; 2703 2704 if (nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER) 2705 nn->dp.ctrl_w1 |= NFP_NET_CFG_CTRL_MCAST_FILTER; 2706 2707 /* Stash the re-configuration queue away. First odd queue in TX Bar */ 2708 nn->qcp_cfg = nn->tx_bar + NFP_QCP_QUEUE_ADDR_SZ; 2709 2710 /* Make sure the FW knows the netdev is supposed to be disabled here */ 2711 nn_writel(nn, NFP_NET_CFG_CTRL, 0); 2712 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0); 2713 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0); 2714 nn_writel(nn, NFP_NET_CFG_CTRL_WORD1, 0); 2715 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_RING | 2716 NFP_NET_CFG_UPDATE_GEN); 2717 if (err) 2718 return err; 2719 2720 if (nn->dp.netdev) { 2721 nfp_net_netdev_init(nn); 2722 2723 err = nfp_ccm_mbox_init(nn); 2724 if (err) 2725 return err; 2726 2727 err = nfp_net_tls_init(nn); 2728 if (err) 2729 goto err_clean_mbox; 2730 2731 nfp_net_ipsec_init(nn); 2732 } 2733 2734 nfp_net_vecs_init(nn); 2735 2736 if (!nn->dp.netdev) 2737 return 0; 2738 2739 spin_lock_init(&nn->mbox_amsg.lock); 2740 INIT_LIST_HEAD(&nn->mbox_amsg.list); 2741 INIT_WORK(&nn->mbox_amsg.work, nfp_net_mbox_amsg_work); 2742 2743 return register_netdev(nn->dp.netdev); 2744 2745 err_clean_mbox: 2746 nfp_ccm_mbox_clean(nn); 2747 return err; 2748 } 2749 2750 /** 2751 * nfp_net_clean() - Undo what nfp_net_init() did. 2752 * @nn: NFP Net device structure 2753 */ 2754 void nfp_net_clean(struct nfp_net *nn) 2755 { 2756 if (!nn->dp.netdev) 2757 return; 2758 2759 unregister_netdev(nn->dp.netdev); 2760 nfp_net_ipsec_clean(nn); 2761 nfp_ccm_mbox_clean(nn); 2762 flush_work(&nn->mbox_amsg.work); 2763 nfp_net_reconfig_wait_posted(nn); 2764 } 2765