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, 0, r_vec->name, 825 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 disable_irq(r_vec->irq_vector); 832 833 irq_set_affinity_hint(r_vec->irq_vector, &r_vec->affinity_mask); 834 835 nn_dbg(nn, "RV%02d: irq=%03d/%03d\n", idx, r_vec->irq_vector, 836 r_vec->irq_entry); 837 838 return 0; 839 } 840 841 static void 842 nfp_net_cleanup_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec) 843 { 844 irq_set_affinity_hint(r_vec->irq_vector, NULL); 845 nfp_net_napi_del(&nn->dp, r_vec); 846 free_irq(r_vec->irq_vector, r_vec); 847 } 848 849 /** 850 * nfp_net_rss_write_itbl() - Write RSS indirection table to device 851 * @nn: NFP Net device to reconfigure 852 */ 853 void nfp_net_rss_write_itbl(struct nfp_net *nn) 854 { 855 int i; 856 857 for (i = 0; i < NFP_NET_CFG_RSS_ITBL_SZ; i += 4) 858 nn_writel(nn, NFP_NET_CFG_RSS_ITBL + i, 859 get_unaligned_le32(nn->rss_itbl + i)); 860 } 861 862 /** 863 * nfp_net_rss_write_key() - Write RSS hash key to device 864 * @nn: NFP Net device to reconfigure 865 */ 866 void nfp_net_rss_write_key(struct nfp_net *nn) 867 { 868 int i; 869 870 for (i = 0; i < nfp_net_rss_key_sz(nn); i += 4) 871 nn_writel(nn, NFP_NET_CFG_RSS_KEY + i, 872 get_unaligned_le32(nn->rss_key + i)); 873 } 874 875 /** 876 * nfp_net_coalesce_write_cfg() - Write irq coalescence configuration to HW 877 * @nn: NFP Net device to reconfigure 878 */ 879 void nfp_net_coalesce_write_cfg(struct nfp_net *nn) 880 { 881 u8 i; 882 u32 factor; 883 u32 value; 884 885 /* Compute factor used to convert coalesce '_usecs' parameters to 886 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp 887 * count. 888 */ 889 factor = nn->tlv_caps.me_freq_mhz / 16; 890 891 /* copy RX interrupt coalesce parameters */ 892 value = (nn->rx_coalesce_max_frames << 16) | 893 (factor * nn->rx_coalesce_usecs); 894 for (i = 0; i < nn->dp.num_rx_rings; i++) 895 nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(i), value); 896 897 /* copy TX interrupt coalesce parameters */ 898 value = (nn->tx_coalesce_max_frames << 16) | 899 (factor * nn->tx_coalesce_usecs); 900 for (i = 0; i < nn->dp.num_tx_rings; i++) 901 nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(i), value); 902 } 903 904 /** 905 * nfp_net_write_mac_addr() - Write mac address to the device control BAR 906 * @nn: NFP Net device to reconfigure 907 * @addr: MAC address to write 908 * 909 * Writes the MAC address from the netdev to the device control BAR. Does not 910 * perform the required reconfig. We do a bit of byte swapping dance because 911 * firmware is LE. 912 */ 913 static void nfp_net_write_mac_addr(struct nfp_net *nn, const u8 *addr) 914 { 915 nn_writel(nn, NFP_NET_CFG_MACADDR + 0, get_unaligned_be32(addr)); 916 nn_writew(nn, NFP_NET_CFG_MACADDR + 6, get_unaligned_be16(addr + 4)); 917 } 918 919 /** 920 * nfp_net_clear_config_and_disable() - Clear control BAR and disable NFP 921 * @nn: NFP Net device to reconfigure 922 * 923 * Warning: must be fully idempotent. 924 */ 925 static void nfp_net_clear_config_and_disable(struct nfp_net *nn) 926 { 927 u32 new_ctrl, new_ctrl_w1, update; 928 unsigned int r; 929 int err; 930 931 new_ctrl = nn->dp.ctrl; 932 new_ctrl &= ~NFP_NET_CFG_CTRL_ENABLE; 933 update = NFP_NET_CFG_UPDATE_GEN; 934 update |= NFP_NET_CFG_UPDATE_MSIX; 935 update |= NFP_NET_CFG_UPDATE_RING; 936 937 if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG) 938 new_ctrl &= ~NFP_NET_CFG_CTRL_RINGCFG; 939 940 if (!(nn->cap_w1 & NFP_NET_CFG_CTRL_FREELIST_EN)) { 941 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0); 942 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0); 943 } 944 945 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 946 err = nfp_net_reconfig(nn, update); 947 if (err) 948 nn_err(nn, "Could not disable device: %d\n", err); 949 950 if (nn->cap_w1 & NFP_NET_CFG_CTRL_FREELIST_EN) { 951 new_ctrl_w1 = nn->dp.ctrl_w1; 952 new_ctrl_w1 &= ~NFP_NET_CFG_CTRL_FREELIST_EN; 953 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0); 954 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0); 955 956 nn_writel(nn, NFP_NET_CFG_CTRL_WORD1, new_ctrl_w1); 957 err = nfp_net_reconfig(nn, update); 958 if (err) 959 nn_err(nn, "Could not disable FREELIST_EN: %d\n", err); 960 nn->dp.ctrl_w1 = new_ctrl_w1; 961 } 962 963 for (r = 0; r < nn->dp.num_rx_rings; r++) { 964 nfp_net_rx_ring_reset(&nn->dp.rx_rings[r]); 965 if (nfp_net_has_xsk_pool_slow(&nn->dp, nn->dp.rx_rings[r].idx)) 966 nfp_net_xsk_rx_bufs_free(&nn->dp.rx_rings[r]); 967 } 968 for (r = 0; r < nn->dp.num_tx_rings; r++) 969 nfp_net_tx_ring_reset(&nn->dp, &nn->dp.tx_rings[r]); 970 for (r = 0; r < nn->dp.num_r_vecs; r++) 971 nfp_net_vec_clear_ring_data(nn, r); 972 973 nn->dp.ctrl = new_ctrl; 974 } 975 976 /** 977 * nfp_net_set_config_and_enable() - Write control BAR and enable NFP 978 * @nn: NFP Net device to reconfigure 979 */ 980 static int nfp_net_set_config_and_enable(struct nfp_net *nn) 981 { 982 u32 bufsz, new_ctrl, new_ctrl_w1, update = 0; 983 unsigned int r; 984 int err; 985 986 new_ctrl = nn->dp.ctrl; 987 new_ctrl_w1 = nn->dp.ctrl_w1; 988 989 if (nn->dp.ctrl & NFP_NET_CFG_CTRL_RSS_ANY) { 990 nfp_net_rss_write_key(nn); 991 nfp_net_rss_write_itbl(nn); 992 nn_writel(nn, NFP_NET_CFG_RSS_CTRL, nn->rss_cfg); 993 update |= NFP_NET_CFG_UPDATE_RSS; 994 } 995 996 if (nn->dp.ctrl & NFP_NET_CFG_CTRL_IRQMOD) { 997 nfp_net_coalesce_write_cfg(nn); 998 update |= NFP_NET_CFG_UPDATE_IRQMOD; 999 } 1000 1001 for (r = 0; r < nn->dp.num_tx_rings; r++) 1002 nfp_net_tx_ring_hw_cfg_write(nn, &nn->dp.tx_rings[r], r); 1003 for (r = 0; r < nn->dp.num_rx_rings; r++) 1004 nfp_net_rx_ring_hw_cfg_write(nn, &nn->dp.rx_rings[r], r); 1005 1006 nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 1007 U64_MAX >> (64 - nn->dp.num_tx_rings)); 1008 1009 nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 1010 U64_MAX >> (64 - nn->dp.num_rx_rings)); 1011 1012 if (nn->dp.netdev) 1013 nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr); 1014 1015 nn_writel(nn, NFP_NET_CFG_MTU, nn->dp.mtu); 1016 1017 bufsz = nn->dp.fl_bufsz - nn->dp.rx_dma_off - NFP_NET_RX_BUF_NON_DATA; 1018 nn_writel(nn, NFP_NET_CFG_FLBUFSZ, bufsz); 1019 1020 /* Enable device 1021 * Step 1: Replace the CTRL_ENABLE by NFP_NET_CFG_CTRL_FREELIST_EN if 1022 * FREELIST_EN exits. 1023 */ 1024 if (nn->cap_w1 & NFP_NET_CFG_CTRL_FREELIST_EN) 1025 new_ctrl_w1 |= NFP_NET_CFG_CTRL_FREELIST_EN; 1026 else 1027 new_ctrl |= NFP_NET_CFG_CTRL_ENABLE; 1028 update |= NFP_NET_CFG_UPDATE_GEN; 1029 update |= NFP_NET_CFG_UPDATE_MSIX; 1030 update |= NFP_NET_CFG_UPDATE_RING; 1031 if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG) 1032 new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG; 1033 1034 /* Step 2: Send the configuration and write the freelist. 1035 * - The freelist only need to be written once. 1036 */ 1037 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1038 nn_writel(nn, NFP_NET_CFG_CTRL_WORD1, new_ctrl_w1); 1039 err = nfp_net_reconfig(nn, update); 1040 if (err) { 1041 nfp_net_clear_config_and_disable(nn); 1042 return err; 1043 } 1044 1045 nn->dp.ctrl = new_ctrl; 1046 nn->dp.ctrl_w1 = new_ctrl_w1; 1047 1048 for (r = 0; r < nn->dp.num_rx_rings; r++) 1049 nfp_net_rx_ring_fill_freelist(&nn->dp, &nn->dp.rx_rings[r]); 1050 1051 /* Step 3: Do the NFP_NET_CFG_CTRL_ENABLE. Send the configuration. 1052 */ 1053 if (nn->cap_w1 & NFP_NET_CFG_CTRL_FREELIST_EN) { 1054 new_ctrl |= NFP_NET_CFG_CTRL_ENABLE; 1055 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1056 1057 err = nfp_net_reconfig(nn, update); 1058 if (err) { 1059 nfp_net_clear_config_and_disable(nn); 1060 return err; 1061 } 1062 nn->dp.ctrl = new_ctrl; 1063 } 1064 1065 return 0; 1066 } 1067 1068 /** 1069 * nfp_net_close_stack() - Quiesce the stack (part of close) 1070 * @nn: NFP Net device to reconfigure 1071 */ 1072 static void nfp_net_close_stack(struct nfp_net *nn) 1073 { 1074 struct nfp_net_r_vector *r_vec; 1075 unsigned int r; 1076 1077 disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector); 1078 netif_carrier_off(nn->dp.netdev); 1079 nn->link_up = false; 1080 1081 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1082 r_vec = &nn->r_vecs[r]; 1083 1084 disable_irq(r_vec->irq_vector); 1085 napi_disable(&r_vec->napi); 1086 1087 if (r_vec->rx_ring) 1088 cancel_work_sync(&r_vec->rx_dim.work); 1089 1090 if (r_vec->tx_ring) 1091 cancel_work_sync(&r_vec->tx_dim.work); 1092 } 1093 1094 netif_tx_disable(nn->dp.netdev); 1095 } 1096 1097 /** 1098 * nfp_net_close_free_all() - Free all runtime resources 1099 * @nn: NFP Net device to reconfigure 1100 */ 1101 static void nfp_net_close_free_all(struct nfp_net *nn) 1102 { 1103 unsigned int r; 1104 1105 nfp_net_tx_rings_free(&nn->dp); 1106 nfp_net_rx_rings_free(&nn->dp); 1107 1108 for (r = 0; r < nn->dp.num_r_vecs; r++) 1109 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1110 1111 nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX); 1112 nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX); 1113 } 1114 1115 /** 1116 * nfp_net_netdev_close() - Called when the device is downed 1117 * @netdev: netdev structure 1118 */ 1119 static int nfp_net_netdev_close(struct net_device *netdev) 1120 { 1121 struct nfp_net *nn = netdev_priv(netdev); 1122 1123 /* Step 1: Disable RX and TX rings from the Linux kernel perspective 1124 */ 1125 nfp_net_close_stack(nn); 1126 1127 /* Step 2: Tell NFP 1128 */ 1129 if (nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER) 1130 __dev_mc_unsync(netdev, nfp_net_mc_unsync); 1131 1132 nfp_net_clear_config_and_disable(nn); 1133 nfp_port_configure(netdev, false); 1134 1135 /* Step 3: Free resources 1136 */ 1137 nfp_net_close_free_all(nn); 1138 1139 nn_dbg(nn, "%s down", netdev->name); 1140 return 0; 1141 } 1142 1143 void nfp_ctrl_close(struct nfp_net *nn) 1144 { 1145 int r; 1146 1147 rtnl_lock(); 1148 1149 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1150 disable_irq(nn->r_vecs[r].irq_vector); 1151 tasklet_disable(&nn->r_vecs[r].tasklet); 1152 } 1153 1154 nfp_net_clear_config_and_disable(nn); 1155 1156 nfp_net_close_free_all(nn); 1157 1158 rtnl_unlock(); 1159 } 1160 1161 static void nfp_net_rx_dim_work(struct work_struct *work) 1162 { 1163 struct nfp_net_r_vector *r_vec; 1164 unsigned int factor, value; 1165 struct dim_cq_moder moder; 1166 struct nfp_net *nn; 1167 struct dim *dim; 1168 1169 dim = container_of(work, struct dim, work); 1170 moder = net_dim_get_rx_moderation(dim->mode, dim->profile_ix); 1171 r_vec = container_of(dim, struct nfp_net_r_vector, rx_dim); 1172 nn = r_vec->nfp_net; 1173 1174 /* Compute factor used to convert coalesce '_usecs' parameters to 1175 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp 1176 * count. 1177 */ 1178 factor = nn->tlv_caps.me_freq_mhz / 16; 1179 if (nfp_net_coalesce_para_check(factor * moder.usec, moder.pkts)) 1180 return; 1181 1182 /* copy RX interrupt coalesce parameters */ 1183 value = (moder.pkts << 16) | (factor * moder.usec); 1184 nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(r_vec->rx_ring->idx), value); 1185 (void)nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_IRQMOD); 1186 1187 dim->state = DIM_START_MEASURE; 1188 } 1189 1190 static void nfp_net_tx_dim_work(struct work_struct *work) 1191 { 1192 struct nfp_net_r_vector *r_vec; 1193 unsigned int factor, value; 1194 struct dim_cq_moder moder; 1195 struct nfp_net *nn; 1196 struct dim *dim; 1197 1198 dim = container_of(work, struct dim, work); 1199 moder = net_dim_get_tx_moderation(dim->mode, dim->profile_ix); 1200 r_vec = container_of(dim, struct nfp_net_r_vector, tx_dim); 1201 nn = r_vec->nfp_net; 1202 1203 /* Compute factor used to convert coalesce '_usecs' parameters to 1204 * ME timestamp ticks. There are 16 ME clock cycles for each timestamp 1205 * count. 1206 */ 1207 factor = nn->tlv_caps.me_freq_mhz / 16; 1208 if (nfp_net_coalesce_para_check(factor * moder.usec, moder.pkts)) 1209 return; 1210 1211 /* copy TX interrupt coalesce parameters */ 1212 value = (moder.pkts << 16) | (factor * moder.usec); 1213 nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(r_vec->tx_ring->idx), value); 1214 (void)nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_IRQMOD); 1215 1216 dim->state = DIM_START_MEASURE; 1217 } 1218 1219 /** 1220 * nfp_net_open_stack() - Start the device from stack's perspective 1221 * @nn: NFP Net device to reconfigure 1222 */ 1223 static void nfp_net_open_stack(struct nfp_net *nn) 1224 { 1225 struct nfp_net_r_vector *r_vec; 1226 unsigned int r; 1227 1228 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1229 r_vec = &nn->r_vecs[r]; 1230 1231 if (r_vec->rx_ring) { 1232 INIT_WORK(&r_vec->rx_dim.work, nfp_net_rx_dim_work); 1233 r_vec->rx_dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE; 1234 } 1235 1236 if (r_vec->tx_ring) { 1237 INIT_WORK(&r_vec->tx_dim.work, nfp_net_tx_dim_work); 1238 r_vec->tx_dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE; 1239 } 1240 1241 napi_enable(&r_vec->napi); 1242 enable_irq(r_vec->irq_vector); 1243 } 1244 1245 netif_tx_wake_all_queues(nn->dp.netdev); 1246 1247 enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector); 1248 nfp_net_read_link_status(nn); 1249 } 1250 1251 static int nfp_net_open_alloc_all(struct nfp_net *nn) 1252 { 1253 int err, r; 1254 1255 err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn", 1256 nn->exn_name, sizeof(nn->exn_name), 1257 NFP_NET_IRQ_EXN_IDX, nn->exn_handler); 1258 if (err) 1259 return err; 1260 err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc", 1261 nn->lsc_name, sizeof(nn->lsc_name), 1262 NFP_NET_IRQ_LSC_IDX, nn->lsc_handler); 1263 if (err) 1264 goto err_free_exn; 1265 disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector); 1266 1267 for (r = 0; r < nn->dp.num_r_vecs; r++) { 1268 err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r); 1269 if (err) 1270 goto err_cleanup_vec_p; 1271 } 1272 1273 err = nfp_net_rx_rings_prepare(nn, &nn->dp); 1274 if (err) 1275 goto err_cleanup_vec; 1276 1277 err = nfp_net_tx_rings_prepare(nn, &nn->dp); 1278 if (err) 1279 goto err_free_rx_rings; 1280 1281 for (r = 0; r < nn->max_r_vecs; r++) 1282 nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r); 1283 1284 return 0; 1285 1286 err_free_rx_rings: 1287 nfp_net_rx_rings_free(&nn->dp); 1288 err_cleanup_vec: 1289 r = nn->dp.num_r_vecs; 1290 err_cleanup_vec_p: 1291 while (r--) 1292 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1293 nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX); 1294 err_free_exn: 1295 nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX); 1296 return err; 1297 } 1298 1299 static int nfp_net_netdev_open(struct net_device *netdev) 1300 { 1301 struct nfp_net *nn = netdev_priv(netdev); 1302 int err; 1303 1304 /* Step 1: Allocate resources for rings and the like 1305 * - Request interrupts 1306 * - Allocate RX and TX ring resources 1307 * - Setup initial RSS table 1308 */ 1309 err = nfp_net_open_alloc_all(nn); 1310 if (err) 1311 return err; 1312 1313 err = netif_set_real_num_tx_queues(netdev, nn->dp.num_stack_tx_rings); 1314 if (err) 1315 goto err_free_all; 1316 1317 err = netif_set_real_num_rx_queues(netdev, nn->dp.num_rx_rings); 1318 if (err) 1319 goto err_free_all; 1320 1321 /* Step 2: Configure the NFP 1322 * - Ifup the physical interface if it exists 1323 * - Enable rings from 0 to tx_rings/rx_rings - 1. 1324 * - Write MAC address (in case it changed) 1325 * - Set the MTU 1326 * - Set the Freelist buffer size 1327 * - Enable the FW 1328 */ 1329 err = nfp_port_configure(netdev, true); 1330 if (err) 1331 goto err_free_all; 1332 1333 err = nfp_net_set_config_and_enable(nn); 1334 if (err) 1335 goto err_port_disable; 1336 1337 /* Step 3: Enable for kernel 1338 * - put some freelist descriptors on each RX ring 1339 * - enable NAPI on each ring 1340 * - enable all TX queues 1341 * - set link state 1342 */ 1343 nfp_net_open_stack(nn); 1344 1345 return 0; 1346 1347 err_port_disable: 1348 nfp_port_configure(netdev, false); 1349 err_free_all: 1350 nfp_net_close_free_all(nn); 1351 return err; 1352 } 1353 1354 int nfp_ctrl_open(struct nfp_net *nn) 1355 { 1356 int err, r; 1357 1358 /* ring dumping depends on vNICs being opened/closed under rtnl */ 1359 rtnl_lock(); 1360 1361 err = nfp_net_open_alloc_all(nn); 1362 if (err) 1363 goto err_unlock; 1364 1365 err = nfp_net_set_config_and_enable(nn); 1366 if (err) 1367 goto err_free_all; 1368 1369 for (r = 0; r < nn->dp.num_r_vecs; r++) 1370 enable_irq(nn->r_vecs[r].irq_vector); 1371 1372 rtnl_unlock(); 1373 1374 return 0; 1375 1376 err_free_all: 1377 nfp_net_close_free_all(nn); 1378 err_unlock: 1379 rtnl_unlock(); 1380 return err; 1381 } 1382 1383 int nfp_net_sched_mbox_amsg_work(struct nfp_net *nn, u32 cmd, const void *data, size_t len, 1384 int (*cb)(struct nfp_net *, struct nfp_mbox_amsg_entry *)) 1385 { 1386 struct nfp_mbox_amsg_entry *entry; 1387 1388 entry = kmalloc(sizeof(*entry) + len, GFP_ATOMIC); 1389 if (!entry) 1390 return -ENOMEM; 1391 1392 memcpy(entry->msg, data, len); 1393 entry->cmd = cmd; 1394 entry->cfg = cb; 1395 1396 spin_lock_bh(&nn->mbox_amsg.lock); 1397 list_add_tail(&entry->list, &nn->mbox_amsg.list); 1398 spin_unlock_bh(&nn->mbox_amsg.lock); 1399 1400 schedule_work(&nn->mbox_amsg.work); 1401 1402 return 0; 1403 } 1404 1405 static void nfp_net_mbox_amsg_work(struct work_struct *work) 1406 { 1407 struct nfp_net *nn = container_of(work, struct nfp_net, mbox_amsg.work); 1408 struct nfp_mbox_amsg_entry *entry, *tmp; 1409 struct list_head tmp_list; 1410 1411 INIT_LIST_HEAD(&tmp_list); 1412 1413 spin_lock_bh(&nn->mbox_amsg.lock); 1414 list_splice_init(&nn->mbox_amsg.list, &tmp_list); 1415 spin_unlock_bh(&nn->mbox_amsg.lock); 1416 1417 list_for_each_entry_safe(entry, tmp, &tmp_list, list) { 1418 int err = entry->cfg(nn, entry); 1419 1420 if (err) 1421 nn_err(nn, "Config cmd %d to HW failed %d.\n", entry->cmd, err); 1422 1423 list_del(&entry->list); 1424 kfree(entry); 1425 } 1426 } 1427 1428 static int nfp_net_mc_cfg(struct nfp_net *nn, struct nfp_mbox_amsg_entry *entry) 1429 { 1430 unsigned char *addr = entry->msg; 1431 int ret; 1432 1433 ret = nfp_net_mbox_lock(nn, NFP_NET_CFG_MULTICAST_SZ); 1434 if (ret) 1435 return ret; 1436 1437 nn_writel(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_MULTICAST_MAC_HI, 1438 get_unaligned_be32(addr)); 1439 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_MULTICAST_MAC_LO, 1440 get_unaligned_be16(addr + 4)); 1441 1442 return nfp_net_mbox_reconfig_and_unlock(nn, entry->cmd); 1443 } 1444 1445 static int nfp_net_mc_sync(struct net_device *netdev, const unsigned char *addr) 1446 { 1447 struct nfp_net *nn = netdev_priv(netdev); 1448 1449 if (netdev_mc_count(netdev) > NFP_NET_CFG_MAC_MC_MAX) { 1450 nn_err(nn, "Requested number of MC addresses (%d) exceeds maximum (%d).\n", 1451 netdev_mc_count(netdev), NFP_NET_CFG_MAC_MC_MAX); 1452 return -EINVAL; 1453 } 1454 1455 return nfp_net_sched_mbox_amsg_work(nn, NFP_NET_CFG_MBOX_CMD_MULTICAST_ADD, addr, 1456 NFP_NET_CFG_MULTICAST_SZ, nfp_net_mc_cfg); 1457 } 1458 1459 static int nfp_net_mc_unsync(struct net_device *netdev, const unsigned char *addr) 1460 { 1461 struct nfp_net *nn = netdev_priv(netdev); 1462 1463 return nfp_net_sched_mbox_amsg_work(nn, NFP_NET_CFG_MBOX_CMD_MULTICAST_DEL, addr, 1464 NFP_NET_CFG_MULTICAST_SZ, nfp_net_mc_cfg); 1465 } 1466 1467 static void nfp_net_set_rx_mode(struct net_device *netdev) 1468 { 1469 struct nfp_net *nn = netdev_priv(netdev); 1470 u32 new_ctrl, new_ctrl_w1; 1471 1472 new_ctrl = nn->dp.ctrl; 1473 new_ctrl_w1 = nn->dp.ctrl_w1; 1474 1475 if (!netdev_mc_empty(netdev) || netdev->flags & IFF_ALLMULTI) 1476 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_L2MC; 1477 else 1478 new_ctrl &= ~NFP_NET_CFG_CTRL_L2MC; 1479 1480 if (netdev->flags & IFF_ALLMULTI) 1481 new_ctrl_w1 &= ~NFP_NET_CFG_CTRL_MCAST_FILTER; 1482 else 1483 new_ctrl_w1 |= nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER; 1484 1485 if (netdev->flags & IFF_PROMISC) { 1486 if (nn->cap & NFP_NET_CFG_CTRL_PROMISC) 1487 new_ctrl |= NFP_NET_CFG_CTRL_PROMISC; 1488 else 1489 nn_warn(nn, "FW does not support promiscuous mode\n"); 1490 } else { 1491 new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC; 1492 } 1493 1494 if ((nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER) && 1495 __dev_mc_sync(netdev, nfp_net_mc_sync, nfp_net_mc_unsync)) 1496 netdev_err(netdev, "Sync mc address failed\n"); 1497 1498 if (new_ctrl == nn->dp.ctrl && new_ctrl_w1 == nn->dp.ctrl_w1) 1499 return; 1500 1501 if (new_ctrl != nn->dp.ctrl) 1502 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1503 if (new_ctrl_w1 != nn->dp.ctrl_w1) 1504 nn_writel(nn, NFP_NET_CFG_CTRL_WORD1, new_ctrl_w1); 1505 nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN); 1506 1507 nn->dp.ctrl = new_ctrl; 1508 nn->dp.ctrl_w1 = new_ctrl_w1; 1509 } 1510 1511 static void nfp_net_rss_init_itbl(struct nfp_net *nn) 1512 { 1513 int i; 1514 1515 for (i = 0; i < sizeof(nn->rss_itbl); i++) 1516 nn->rss_itbl[i] = 1517 ethtool_rxfh_indir_default(i, nn->dp.num_rx_rings); 1518 } 1519 1520 static void nfp_net_dp_swap(struct nfp_net *nn, struct nfp_net_dp *dp) 1521 { 1522 struct nfp_net_dp new_dp = *dp; 1523 1524 *dp = nn->dp; 1525 nn->dp = new_dp; 1526 1527 nn->dp.netdev->mtu = new_dp.mtu; 1528 1529 if (!netif_is_rxfh_configured(nn->dp.netdev)) 1530 nfp_net_rss_init_itbl(nn); 1531 } 1532 1533 static int nfp_net_dp_swap_enable(struct nfp_net *nn, struct nfp_net_dp *dp) 1534 { 1535 unsigned int r; 1536 int err; 1537 1538 nfp_net_dp_swap(nn, dp); 1539 1540 for (r = 0; r < nn->max_r_vecs; r++) 1541 nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r); 1542 1543 err = netif_set_real_num_queues(nn->dp.netdev, 1544 nn->dp.num_stack_tx_rings, 1545 nn->dp.num_rx_rings); 1546 if (err) 1547 return err; 1548 1549 return nfp_net_set_config_and_enable(nn); 1550 } 1551 1552 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn) 1553 { 1554 struct nfp_net_dp *new; 1555 1556 new = kmalloc(sizeof(*new), GFP_KERNEL); 1557 if (!new) 1558 return NULL; 1559 1560 *new = nn->dp; 1561 1562 new->xsk_pools = kmemdup(new->xsk_pools, 1563 array_size(nn->max_r_vecs, 1564 sizeof(new->xsk_pools)), 1565 GFP_KERNEL); 1566 if (!new->xsk_pools) { 1567 kfree(new); 1568 return NULL; 1569 } 1570 1571 /* Clear things which need to be recomputed */ 1572 new->fl_bufsz = 0; 1573 new->tx_rings = NULL; 1574 new->rx_rings = NULL; 1575 new->num_r_vecs = 0; 1576 new->num_stack_tx_rings = 0; 1577 new->txrwb = NULL; 1578 new->txrwb_dma = 0; 1579 1580 return new; 1581 } 1582 1583 static void nfp_net_free_dp(struct nfp_net_dp *dp) 1584 { 1585 kfree(dp->xsk_pools); 1586 kfree(dp); 1587 } 1588 1589 static int 1590 nfp_net_check_config(struct nfp_net *nn, struct nfp_net_dp *dp, 1591 struct netlink_ext_ack *extack) 1592 { 1593 unsigned int r, xsk_min_fl_bufsz; 1594 1595 /* XDP-enabled tests */ 1596 if (!dp->xdp_prog) 1597 return 0; 1598 if (dp->fl_bufsz > PAGE_SIZE) { 1599 NL_SET_ERR_MSG_MOD(extack, "MTU too large w/ XDP enabled"); 1600 return -EINVAL; 1601 } 1602 if (dp->num_tx_rings > nn->max_tx_rings) { 1603 NL_SET_ERR_MSG_MOD(extack, "Insufficient number of TX rings w/ XDP enabled"); 1604 return -EINVAL; 1605 } 1606 1607 xsk_min_fl_bufsz = nfp_net_calc_fl_bufsz_xsk(dp); 1608 for (r = 0; r < nn->max_r_vecs; r++) { 1609 if (!dp->xsk_pools[r]) 1610 continue; 1611 1612 if (xsk_pool_get_rx_frame_size(dp->xsk_pools[r]) < xsk_min_fl_bufsz) { 1613 NL_SET_ERR_MSG_MOD(extack, 1614 "XSK buffer pool chunk size too small"); 1615 return -EINVAL; 1616 } 1617 } 1618 1619 return 0; 1620 } 1621 1622 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *dp, 1623 struct netlink_ext_ack *extack) 1624 { 1625 int r, err; 1626 1627 dp->fl_bufsz = nfp_net_calc_fl_bufsz(dp); 1628 1629 dp->num_stack_tx_rings = dp->num_tx_rings; 1630 if (dp->xdp_prog) 1631 dp->num_stack_tx_rings -= dp->num_rx_rings; 1632 1633 dp->num_r_vecs = max(dp->num_rx_rings, dp->num_stack_tx_rings); 1634 1635 err = nfp_net_check_config(nn, dp, extack); 1636 if (err) 1637 goto exit_free_dp; 1638 1639 if (!netif_running(dp->netdev)) { 1640 nfp_net_dp_swap(nn, dp); 1641 err = 0; 1642 goto exit_free_dp; 1643 } 1644 1645 /* Prepare new rings */ 1646 for (r = nn->dp.num_r_vecs; r < dp->num_r_vecs; r++) { 1647 err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r); 1648 if (err) { 1649 dp->num_r_vecs = r; 1650 goto err_cleanup_vecs; 1651 } 1652 } 1653 1654 err = nfp_net_rx_rings_prepare(nn, dp); 1655 if (err) 1656 goto err_cleanup_vecs; 1657 1658 err = nfp_net_tx_rings_prepare(nn, dp); 1659 if (err) 1660 goto err_free_rx; 1661 1662 /* Stop device, swap in new rings, try to start the firmware */ 1663 nfp_net_close_stack(nn); 1664 nfp_net_clear_config_and_disable(nn); 1665 1666 err = nfp_net_dp_swap_enable(nn, dp); 1667 if (err) { 1668 int err2; 1669 1670 nfp_net_clear_config_and_disable(nn); 1671 1672 /* Try with old configuration and old rings */ 1673 err2 = nfp_net_dp_swap_enable(nn, dp); 1674 if (err2) 1675 nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n", 1676 err, err2); 1677 } 1678 for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--) 1679 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1680 1681 nfp_net_rx_rings_free(dp); 1682 nfp_net_tx_rings_free(dp); 1683 1684 nfp_net_open_stack(nn); 1685 exit_free_dp: 1686 nfp_net_free_dp(dp); 1687 1688 return err; 1689 1690 err_free_rx: 1691 nfp_net_rx_rings_free(dp); 1692 err_cleanup_vecs: 1693 for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--) 1694 nfp_net_cleanup_vector(nn, &nn->r_vecs[r]); 1695 nfp_net_free_dp(dp); 1696 return err; 1697 } 1698 1699 static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu) 1700 { 1701 struct nfp_net *nn = netdev_priv(netdev); 1702 struct nfp_net_dp *dp; 1703 int err; 1704 1705 err = nfp_app_check_mtu(nn->app, netdev, new_mtu); 1706 if (err) 1707 return err; 1708 1709 dp = nfp_net_clone_dp(nn); 1710 if (!dp) 1711 return -ENOMEM; 1712 1713 dp->mtu = new_mtu; 1714 1715 return nfp_net_ring_reconfig(nn, dp, NULL); 1716 } 1717 1718 static int 1719 nfp_net_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid) 1720 { 1721 const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_ADD; 1722 struct nfp_net *nn = netdev_priv(netdev); 1723 int err; 1724 1725 /* Priority tagged packets with vlan id 0 are processed by the 1726 * NFP as untagged packets 1727 */ 1728 if (!vid) 1729 return 0; 1730 1731 err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ); 1732 if (err) 1733 return err; 1734 1735 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid); 1736 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO, 1737 ETH_P_8021Q); 1738 1739 return nfp_net_mbox_reconfig_and_unlock(nn, cmd); 1740 } 1741 1742 static int 1743 nfp_net_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid) 1744 { 1745 const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_KILL; 1746 struct nfp_net *nn = netdev_priv(netdev); 1747 int err; 1748 1749 /* Priority tagged packets with vlan id 0 are processed by the 1750 * NFP as untagged packets 1751 */ 1752 if (!vid) 1753 return 0; 1754 1755 err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ); 1756 if (err) 1757 return err; 1758 1759 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid); 1760 nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO, 1761 ETH_P_8021Q); 1762 1763 return nfp_net_mbox_reconfig_and_unlock(nn, cmd); 1764 } 1765 1766 static void nfp_net_stat64(struct net_device *netdev, 1767 struct rtnl_link_stats64 *stats) 1768 { 1769 struct nfp_net *nn = netdev_priv(netdev); 1770 int r; 1771 1772 /* Collect software stats */ 1773 for (r = 0; r < nn->max_r_vecs; r++) { 1774 struct nfp_net_r_vector *r_vec = &nn->r_vecs[r]; 1775 u64 data[3]; 1776 unsigned int start; 1777 1778 do { 1779 start = u64_stats_fetch_begin(&r_vec->rx_sync); 1780 data[0] = r_vec->rx_pkts; 1781 data[1] = r_vec->rx_bytes; 1782 data[2] = r_vec->rx_drops; 1783 } while (u64_stats_fetch_retry(&r_vec->rx_sync, start)); 1784 stats->rx_packets += data[0]; 1785 stats->rx_bytes += data[1]; 1786 stats->rx_dropped += data[2]; 1787 1788 do { 1789 start = u64_stats_fetch_begin(&r_vec->tx_sync); 1790 data[0] = r_vec->tx_pkts; 1791 data[1] = r_vec->tx_bytes; 1792 data[2] = r_vec->tx_errors; 1793 } while (u64_stats_fetch_retry(&r_vec->tx_sync, start)); 1794 stats->tx_packets += data[0]; 1795 stats->tx_bytes += data[1]; 1796 stats->tx_errors += data[2]; 1797 } 1798 1799 /* Add in device stats */ 1800 stats->multicast += nn_readq(nn, NFP_NET_CFG_STATS_RX_MC_FRAMES); 1801 stats->rx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_RX_DISCARDS); 1802 stats->rx_errors += nn_readq(nn, NFP_NET_CFG_STATS_RX_ERRORS); 1803 1804 stats->tx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_TX_DISCARDS); 1805 stats->tx_errors += nn_readq(nn, NFP_NET_CFG_STATS_TX_ERRORS); 1806 } 1807 1808 static int nfp_net_set_features(struct net_device *netdev, 1809 netdev_features_t features) 1810 { 1811 netdev_features_t changed = netdev->features ^ features; 1812 struct nfp_net *nn = netdev_priv(netdev); 1813 u32 new_ctrl; 1814 int err; 1815 1816 /* Assume this is not called with features we have not advertised */ 1817 1818 new_ctrl = nn->dp.ctrl; 1819 1820 if (changed & NETIF_F_RXCSUM) { 1821 if (features & NETIF_F_RXCSUM) 1822 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY; 1823 else 1824 new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM_ANY; 1825 } 1826 1827 if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) { 1828 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) 1829 new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM; 1830 else 1831 new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM; 1832 } 1833 1834 if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) { 1835 if (features & (NETIF_F_TSO | NETIF_F_TSO6)) 1836 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?: 1837 NFP_NET_CFG_CTRL_LSO; 1838 else 1839 new_ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY; 1840 } 1841 1842 if (changed & NETIF_F_HW_VLAN_CTAG_RX) { 1843 if (features & NETIF_F_HW_VLAN_CTAG_RX) 1844 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ?: 1845 NFP_NET_CFG_CTRL_RXVLAN; 1846 else 1847 new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN_ANY; 1848 } 1849 1850 if (changed & NETIF_F_HW_VLAN_CTAG_TX) { 1851 if (features & NETIF_F_HW_VLAN_CTAG_TX) 1852 new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ?: 1853 NFP_NET_CFG_CTRL_TXVLAN; 1854 else 1855 new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN_ANY; 1856 } 1857 1858 if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) { 1859 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) 1860 new_ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER; 1861 else 1862 new_ctrl &= ~NFP_NET_CFG_CTRL_CTAG_FILTER; 1863 } 1864 1865 if (changed & NETIF_F_HW_VLAN_STAG_RX) { 1866 if (features & NETIF_F_HW_VLAN_STAG_RX) 1867 new_ctrl |= NFP_NET_CFG_CTRL_RXQINQ; 1868 else 1869 new_ctrl &= ~NFP_NET_CFG_CTRL_RXQINQ; 1870 } 1871 1872 if (changed & NETIF_F_SG) { 1873 if (features & NETIF_F_SG) 1874 new_ctrl |= NFP_NET_CFG_CTRL_GATHER; 1875 else 1876 new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER; 1877 } 1878 1879 err = nfp_port_set_features(netdev, features); 1880 if (err) 1881 return err; 1882 1883 nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n", 1884 netdev->features, features, changed); 1885 1886 if (new_ctrl == nn->dp.ctrl) 1887 return 0; 1888 1889 nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->dp.ctrl, new_ctrl); 1890 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 1891 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN); 1892 if (err) 1893 return err; 1894 1895 nn->dp.ctrl = new_ctrl; 1896 1897 return 0; 1898 } 1899 1900 static netdev_features_t 1901 nfp_net_fix_features(struct net_device *netdev, 1902 netdev_features_t features) 1903 { 1904 if ((features & NETIF_F_HW_VLAN_CTAG_RX) && 1905 (features & NETIF_F_HW_VLAN_STAG_RX)) { 1906 if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) { 1907 features &= ~NETIF_F_HW_VLAN_CTAG_RX; 1908 netdev->wanted_features &= ~NETIF_F_HW_VLAN_CTAG_RX; 1909 netdev_warn(netdev, 1910 "S-tag and C-tag stripping can't be enabled at the same time. Enabling S-tag stripping and disabling C-tag stripping\n"); 1911 } else if (netdev->features & NETIF_F_HW_VLAN_STAG_RX) { 1912 features &= ~NETIF_F_HW_VLAN_STAG_RX; 1913 netdev->wanted_features &= ~NETIF_F_HW_VLAN_STAG_RX; 1914 netdev_warn(netdev, 1915 "S-tag and C-tag stripping can't be enabled at the same time. Enabling C-tag stripping and disabling S-tag stripping\n"); 1916 } 1917 } 1918 return features; 1919 } 1920 1921 static netdev_features_t 1922 nfp_net_features_check(struct sk_buff *skb, struct net_device *dev, 1923 netdev_features_t features) 1924 { 1925 u8 l4_hdr; 1926 1927 /* We can't do TSO over double tagged packets (802.1AD) */ 1928 features &= vlan_features_check(skb, features); 1929 1930 if (!skb->encapsulation) 1931 return features; 1932 1933 /* Ensure that inner L4 header offset fits into TX descriptor field */ 1934 if (skb_is_gso(skb)) { 1935 u32 hdrlen; 1936 1937 hdrlen = skb_inner_tcp_all_headers(skb); 1938 1939 /* Assume worst case scenario of having longest possible 1940 * metadata prepend - 8B 1941 */ 1942 if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ - 8)) 1943 features &= ~NETIF_F_GSO_MASK; 1944 } 1945 1946 if (xfrm_offload(skb)) 1947 return features; 1948 1949 /* VXLAN/GRE check */ 1950 switch (vlan_get_protocol(skb)) { 1951 case htons(ETH_P_IP): 1952 l4_hdr = ip_hdr(skb)->protocol; 1953 break; 1954 case htons(ETH_P_IPV6): 1955 l4_hdr = ipv6_hdr(skb)->nexthdr; 1956 break; 1957 default: 1958 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 1959 } 1960 1961 if (skb->inner_protocol_type != ENCAP_TYPE_ETHER || 1962 skb->inner_protocol != htons(ETH_P_TEB) || 1963 (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) || 1964 (l4_hdr == IPPROTO_UDP && 1965 (skb_inner_mac_header(skb) - skb_transport_header(skb) != 1966 sizeof(struct udphdr) + sizeof(struct vxlanhdr)))) 1967 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); 1968 1969 return features; 1970 } 1971 1972 static int 1973 nfp_net_get_phys_port_name(struct net_device *netdev, char *name, size_t len) 1974 { 1975 struct nfp_net *nn = netdev_priv(netdev); 1976 int n; 1977 1978 /* If port is defined, devlink_port is registered and devlink core 1979 * is taking care of name formatting. 1980 */ 1981 if (nn->port) 1982 return -EOPNOTSUPP; 1983 1984 if (nn->dp.is_vf || nn->vnic_no_name) 1985 return -EOPNOTSUPP; 1986 1987 n = snprintf(name, len, "n%d", nn->id); 1988 if (n >= len) 1989 return -EINVAL; 1990 1991 return 0; 1992 } 1993 1994 static int nfp_net_xdp_setup_drv(struct nfp_net *nn, struct netdev_bpf *bpf) 1995 { 1996 struct bpf_prog *prog = bpf->prog; 1997 struct nfp_net_dp *dp; 1998 int err; 1999 2000 if (!prog == !nn->dp.xdp_prog) { 2001 WRITE_ONCE(nn->dp.xdp_prog, prog); 2002 xdp_attachment_setup(&nn->xdp, bpf); 2003 return 0; 2004 } 2005 2006 dp = nfp_net_clone_dp(nn); 2007 if (!dp) 2008 return -ENOMEM; 2009 2010 dp->xdp_prog = prog; 2011 dp->num_tx_rings += prog ? nn->dp.num_rx_rings : -nn->dp.num_rx_rings; 2012 dp->rx_dma_dir = prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE; 2013 dp->rx_dma_off = prog ? XDP_PACKET_HEADROOM - nn->dp.rx_offset : 0; 2014 2015 /* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */ 2016 err = nfp_net_ring_reconfig(nn, dp, bpf->extack); 2017 if (err) 2018 return err; 2019 2020 xdp_attachment_setup(&nn->xdp, bpf); 2021 return 0; 2022 } 2023 2024 static int nfp_net_xdp_setup_hw(struct nfp_net *nn, struct netdev_bpf *bpf) 2025 { 2026 int err; 2027 2028 err = nfp_app_xdp_offload(nn->app, nn, bpf->prog, bpf->extack); 2029 if (err) 2030 return err; 2031 2032 xdp_attachment_setup(&nn->xdp_hw, bpf); 2033 return 0; 2034 } 2035 2036 static int nfp_net_xdp(struct net_device *netdev, struct netdev_bpf *xdp) 2037 { 2038 struct nfp_net *nn = netdev_priv(netdev); 2039 2040 switch (xdp->command) { 2041 case XDP_SETUP_PROG: 2042 return nfp_net_xdp_setup_drv(nn, xdp); 2043 case XDP_SETUP_PROG_HW: 2044 return nfp_net_xdp_setup_hw(nn, xdp); 2045 case XDP_SETUP_XSK_POOL: 2046 return nfp_net_xsk_setup_pool(netdev, xdp->xsk.pool, 2047 xdp->xsk.queue_id); 2048 default: 2049 return nfp_app_bpf(nn->app, nn, xdp); 2050 } 2051 } 2052 2053 static int nfp_net_set_mac_address(struct net_device *netdev, void *addr) 2054 { 2055 struct nfp_net *nn = netdev_priv(netdev); 2056 struct sockaddr *saddr = addr; 2057 int err; 2058 2059 err = eth_prepare_mac_addr_change(netdev, addr); 2060 if (err) 2061 return err; 2062 2063 nfp_net_write_mac_addr(nn, saddr->sa_data); 2064 2065 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MACADDR); 2066 if (err) 2067 return err; 2068 2069 eth_commit_mac_addr_change(netdev, addr); 2070 2071 return 0; 2072 } 2073 2074 static int nfp_net_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq, 2075 struct net_device *dev, u32 filter_mask, 2076 int nlflags) 2077 { 2078 struct nfp_net *nn = netdev_priv(dev); 2079 u16 mode; 2080 2081 if (!(nn->cap & NFP_NET_CFG_CTRL_VEPA)) 2082 return -EOPNOTSUPP; 2083 2084 mode = (nn->dp.ctrl & NFP_NET_CFG_CTRL_VEPA) ? 2085 BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB; 2086 2087 return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode, 0, 0, 2088 nlflags, filter_mask, NULL); 2089 } 2090 2091 static int nfp_net_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh, 2092 u16 flags, struct netlink_ext_ack *extack) 2093 { 2094 struct nfp_net *nn = netdev_priv(dev); 2095 struct nlattr *attr, *br_spec; 2096 int rem, err; 2097 u32 new_ctrl; 2098 u16 mode; 2099 2100 if (!(nn->cap & NFP_NET_CFG_CTRL_VEPA)) 2101 return -EOPNOTSUPP; 2102 2103 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC); 2104 if (!br_spec) 2105 return -EINVAL; 2106 2107 nla_for_each_nested(attr, br_spec, rem) { 2108 if (nla_type(attr) != IFLA_BRIDGE_MODE) 2109 continue; 2110 2111 new_ctrl = nn->dp.ctrl; 2112 mode = nla_get_u16(attr); 2113 if (mode == BRIDGE_MODE_VEPA) 2114 new_ctrl |= NFP_NET_CFG_CTRL_VEPA; 2115 else if (mode == BRIDGE_MODE_VEB) 2116 new_ctrl &= ~NFP_NET_CFG_CTRL_VEPA; 2117 else 2118 return -EOPNOTSUPP; 2119 2120 if (new_ctrl == nn->dp.ctrl) 2121 return 0; 2122 2123 nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl); 2124 err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN); 2125 if (!err) 2126 nn->dp.ctrl = new_ctrl; 2127 2128 return err; 2129 } 2130 2131 return -EINVAL; 2132 } 2133 2134 const struct net_device_ops nfp_nfd3_netdev_ops = { 2135 .ndo_init = nfp_app_ndo_init, 2136 .ndo_uninit = nfp_app_ndo_uninit, 2137 .ndo_open = nfp_net_netdev_open, 2138 .ndo_stop = nfp_net_netdev_close, 2139 .ndo_start_xmit = nfp_net_tx, 2140 .ndo_get_stats64 = nfp_net_stat64, 2141 .ndo_vlan_rx_add_vid = nfp_net_vlan_rx_add_vid, 2142 .ndo_vlan_rx_kill_vid = nfp_net_vlan_rx_kill_vid, 2143 .ndo_set_vf_mac = nfp_app_set_vf_mac, 2144 .ndo_set_vf_vlan = nfp_app_set_vf_vlan, 2145 .ndo_set_vf_rate = nfp_app_set_vf_rate, 2146 .ndo_set_vf_spoofchk = nfp_app_set_vf_spoofchk, 2147 .ndo_set_vf_trust = nfp_app_set_vf_trust, 2148 .ndo_get_vf_config = nfp_app_get_vf_config, 2149 .ndo_set_vf_link_state = nfp_app_set_vf_link_state, 2150 .ndo_setup_tc = nfp_port_setup_tc, 2151 .ndo_tx_timeout = nfp_net_tx_timeout, 2152 .ndo_set_rx_mode = nfp_net_set_rx_mode, 2153 .ndo_change_mtu = nfp_net_change_mtu, 2154 .ndo_set_mac_address = nfp_net_set_mac_address, 2155 .ndo_set_features = nfp_net_set_features, 2156 .ndo_fix_features = nfp_net_fix_features, 2157 .ndo_features_check = nfp_net_features_check, 2158 .ndo_get_phys_port_name = nfp_net_get_phys_port_name, 2159 .ndo_bpf = nfp_net_xdp, 2160 .ndo_xsk_wakeup = nfp_net_xsk_wakeup, 2161 .ndo_bridge_getlink = nfp_net_bridge_getlink, 2162 .ndo_bridge_setlink = nfp_net_bridge_setlink, 2163 }; 2164 2165 const struct net_device_ops nfp_nfdk_netdev_ops = { 2166 .ndo_init = nfp_app_ndo_init, 2167 .ndo_uninit = nfp_app_ndo_uninit, 2168 .ndo_open = nfp_net_netdev_open, 2169 .ndo_stop = nfp_net_netdev_close, 2170 .ndo_start_xmit = nfp_net_tx, 2171 .ndo_get_stats64 = nfp_net_stat64, 2172 .ndo_vlan_rx_add_vid = nfp_net_vlan_rx_add_vid, 2173 .ndo_vlan_rx_kill_vid = nfp_net_vlan_rx_kill_vid, 2174 .ndo_set_vf_mac = nfp_app_set_vf_mac, 2175 .ndo_set_vf_vlan = nfp_app_set_vf_vlan, 2176 .ndo_set_vf_rate = nfp_app_set_vf_rate, 2177 .ndo_set_vf_spoofchk = nfp_app_set_vf_spoofchk, 2178 .ndo_set_vf_trust = nfp_app_set_vf_trust, 2179 .ndo_get_vf_config = nfp_app_get_vf_config, 2180 .ndo_set_vf_link_state = nfp_app_set_vf_link_state, 2181 .ndo_setup_tc = nfp_port_setup_tc, 2182 .ndo_tx_timeout = nfp_net_tx_timeout, 2183 .ndo_set_rx_mode = nfp_net_set_rx_mode, 2184 .ndo_change_mtu = nfp_net_change_mtu, 2185 .ndo_set_mac_address = nfp_net_set_mac_address, 2186 .ndo_set_features = nfp_net_set_features, 2187 .ndo_fix_features = nfp_net_fix_features, 2188 .ndo_features_check = nfp_net_features_check, 2189 .ndo_get_phys_port_name = nfp_net_get_phys_port_name, 2190 .ndo_bpf = nfp_net_xdp, 2191 .ndo_bridge_getlink = nfp_net_bridge_getlink, 2192 .ndo_bridge_setlink = nfp_net_bridge_setlink, 2193 }; 2194 2195 static int nfp_udp_tunnel_sync(struct net_device *netdev, unsigned int table) 2196 { 2197 struct nfp_net *nn = netdev_priv(netdev); 2198 int i; 2199 2200 BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1); 2201 for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2) { 2202 struct udp_tunnel_info ti0, ti1; 2203 2204 udp_tunnel_nic_get_port(netdev, table, i, &ti0); 2205 udp_tunnel_nic_get_port(netdev, table, i + 1, &ti1); 2206 2207 nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(ti0.port), 2208 be16_to_cpu(ti1.port) << 16 | be16_to_cpu(ti0.port)); 2209 } 2210 2211 return nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_VXLAN); 2212 } 2213 2214 static const struct udp_tunnel_nic_info nfp_udp_tunnels = { 2215 .sync_table = nfp_udp_tunnel_sync, 2216 .flags = UDP_TUNNEL_NIC_INFO_MAY_SLEEP | 2217 UDP_TUNNEL_NIC_INFO_OPEN_ONLY, 2218 .tables = { 2219 { 2220 .n_entries = NFP_NET_N_VXLAN_PORTS, 2221 .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, 2222 }, 2223 }, 2224 }; 2225 2226 /** 2227 * nfp_net_info() - Print general info about the NIC 2228 * @nn: NFP Net device to reconfigure 2229 */ 2230 void nfp_net_info(struct nfp_net *nn) 2231 { 2232 nn_info(nn, "NFP-6xxx %sNetdev: TxQs=%d/%d RxQs=%d/%d\n", 2233 nn->dp.is_vf ? "VF " : "", 2234 nn->dp.num_tx_rings, nn->max_tx_rings, 2235 nn->dp.num_rx_rings, nn->max_rx_rings); 2236 nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n", 2237 nn->fw_ver.extend, nn->fw_ver.class, 2238 nn->fw_ver.major, nn->fw_ver.minor, 2239 nn->max_mtu); 2240 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", 2241 nn->cap, 2242 nn->cap & NFP_NET_CFG_CTRL_PROMISC ? "PROMISC " : "", 2243 nn->cap & NFP_NET_CFG_CTRL_L2BC ? "L2BCFILT " : "", 2244 nn->cap & NFP_NET_CFG_CTRL_L2MC ? "L2MCFILT " : "", 2245 nn->cap & NFP_NET_CFG_CTRL_RXCSUM ? "RXCSUM " : "", 2246 nn->cap & NFP_NET_CFG_CTRL_TXCSUM ? "TXCSUM " : "", 2247 nn->cap & NFP_NET_CFG_CTRL_RXVLAN ? "RXVLAN " : "", 2248 nn->cap & NFP_NET_CFG_CTRL_TXVLAN ? "TXVLAN " : "", 2249 nn->cap & NFP_NET_CFG_CTRL_RXQINQ ? "RXQINQ " : "", 2250 nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ? "RXVLANv2 " : "", 2251 nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ? "TXVLANv2 " : "", 2252 nn->cap & NFP_NET_CFG_CTRL_SCATTER ? "SCATTER " : "", 2253 nn->cap & NFP_NET_CFG_CTRL_GATHER ? "GATHER " : "", 2254 nn->cap & NFP_NET_CFG_CTRL_LSO ? "TSO1 " : "", 2255 nn->cap & NFP_NET_CFG_CTRL_LSO2 ? "TSO2 " : "", 2256 nn->cap & NFP_NET_CFG_CTRL_RSS ? "RSS1 " : "", 2257 nn->cap & NFP_NET_CFG_CTRL_RSS2 ? "RSS2 " : "", 2258 nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER ? "CTAG_FILTER " : "", 2259 nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "", 2260 nn->cap & NFP_NET_CFG_CTRL_IRQMOD ? "IRQMOD " : "", 2261 nn->cap & NFP_NET_CFG_CTRL_TXRWB ? "TXRWB " : "", 2262 nn->cap & NFP_NET_CFG_CTRL_VEPA ? "VEPA " : "", 2263 nn->cap & NFP_NET_CFG_CTRL_VXLAN ? "VXLAN " : "", 2264 nn->cap & NFP_NET_CFG_CTRL_NVGRE ? "NVGRE " : "", 2265 nn->cap & NFP_NET_CFG_CTRL_CSUM_COMPLETE ? 2266 "RXCSUM_COMPLETE " : "", 2267 nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR ? "LIVE_ADDR " : "", 2268 nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER ? "MULTICAST_FILTER " : "", 2269 nfp_app_extra_cap(nn->app, nn)); 2270 } 2271 2272 /** 2273 * nfp_net_alloc() - Allocate netdev and related structure 2274 * @pdev: PCI device 2275 * @dev_info: NFP ASIC params 2276 * @ctrl_bar: PCI IOMEM with vNIC config memory 2277 * @needs_netdev: Whether to allocate a netdev for this vNIC 2278 * @max_tx_rings: Maximum number of TX rings supported by device 2279 * @max_rx_rings: Maximum number of RX rings supported by device 2280 * 2281 * This function allocates a netdev device and fills in the initial 2282 * part of the @struct nfp_net structure. In case of control device 2283 * nfp_net structure is allocated without the netdev. 2284 * 2285 * Return: NFP Net device structure, or ERR_PTR on error. 2286 */ 2287 struct nfp_net * 2288 nfp_net_alloc(struct pci_dev *pdev, const struct nfp_dev_info *dev_info, 2289 void __iomem *ctrl_bar, bool needs_netdev, 2290 unsigned int max_tx_rings, unsigned int max_rx_rings) 2291 { 2292 u64 dma_mask = dma_get_mask(&pdev->dev); 2293 struct nfp_net *nn; 2294 int err; 2295 2296 if (needs_netdev) { 2297 struct net_device *netdev; 2298 2299 netdev = alloc_etherdev_mqs(sizeof(struct nfp_net), 2300 max_tx_rings, max_rx_rings); 2301 if (!netdev) 2302 return ERR_PTR(-ENOMEM); 2303 2304 SET_NETDEV_DEV(netdev, &pdev->dev); 2305 nn = netdev_priv(netdev); 2306 nn->dp.netdev = netdev; 2307 } else { 2308 nn = vzalloc(sizeof(*nn)); 2309 if (!nn) 2310 return ERR_PTR(-ENOMEM); 2311 } 2312 2313 nn->dp.dev = &pdev->dev; 2314 nn->dp.ctrl_bar = ctrl_bar; 2315 nn->dev_info = dev_info; 2316 nn->pdev = pdev; 2317 nfp_net_get_fw_version(&nn->fw_ver, ctrl_bar); 2318 2319 switch (FIELD_GET(NFP_NET_CFG_VERSION_DP_MASK, nn->fw_ver.extend)) { 2320 case NFP_NET_CFG_VERSION_DP_NFD3: 2321 nn->dp.ops = &nfp_nfd3_ops; 2322 break; 2323 case NFP_NET_CFG_VERSION_DP_NFDK: 2324 if (nn->fw_ver.major < 5) { 2325 dev_err(&pdev->dev, 2326 "NFDK must use ABI 5 or newer, found: %d\n", 2327 nn->fw_ver.major); 2328 err = -EINVAL; 2329 goto err_free_nn; 2330 } 2331 nn->dp.ops = &nfp_nfdk_ops; 2332 break; 2333 default: 2334 err = -EINVAL; 2335 goto err_free_nn; 2336 } 2337 2338 if ((dma_mask & nn->dp.ops->dma_mask) != dma_mask) { 2339 dev_err(&pdev->dev, 2340 "DMA mask of loaded firmware: %llx, required DMA mask: %llx\n", 2341 nn->dp.ops->dma_mask, dma_mask); 2342 err = -EINVAL; 2343 goto err_free_nn; 2344 } 2345 2346 nn->max_tx_rings = max_tx_rings; 2347 nn->max_rx_rings = max_rx_rings; 2348 2349 nn->dp.num_tx_rings = min_t(unsigned int, 2350 max_tx_rings, num_online_cpus()); 2351 nn->dp.num_rx_rings = min_t(unsigned int, max_rx_rings, 2352 netif_get_num_default_rss_queues()); 2353 2354 nn->dp.num_r_vecs = max(nn->dp.num_tx_rings, nn->dp.num_rx_rings); 2355 nn->dp.num_r_vecs = min_t(unsigned int, 2356 nn->dp.num_r_vecs, num_online_cpus()); 2357 nn->max_r_vecs = nn->dp.num_r_vecs; 2358 2359 nn->dp.xsk_pools = kcalloc(nn->max_r_vecs, sizeof(nn->dp.xsk_pools), 2360 GFP_KERNEL); 2361 if (!nn->dp.xsk_pools) { 2362 err = -ENOMEM; 2363 goto err_free_nn; 2364 } 2365 2366 nn->dp.txd_cnt = NFP_NET_TX_DESCS_DEFAULT; 2367 nn->dp.rxd_cnt = NFP_NET_RX_DESCS_DEFAULT; 2368 2369 sema_init(&nn->bar_lock, 1); 2370 2371 spin_lock_init(&nn->reconfig_lock); 2372 spin_lock_init(&nn->link_status_lock); 2373 2374 timer_setup(&nn->reconfig_timer, nfp_net_reconfig_timer, 0); 2375 2376 err = nfp_net_tlv_caps_parse(&nn->pdev->dev, nn->dp.ctrl_bar, 2377 &nn->tlv_caps); 2378 if (err) 2379 goto err_free_nn; 2380 2381 err = nfp_ccm_mbox_alloc(nn); 2382 if (err) 2383 goto err_free_nn; 2384 2385 return nn; 2386 2387 err_free_nn: 2388 if (nn->dp.netdev) 2389 free_netdev(nn->dp.netdev); 2390 else 2391 vfree(nn); 2392 return ERR_PTR(err); 2393 } 2394 2395 /** 2396 * nfp_net_free() - Undo what @nfp_net_alloc() did 2397 * @nn: NFP Net device to reconfigure 2398 */ 2399 void nfp_net_free(struct nfp_net *nn) 2400 { 2401 WARN_ON(timer_pending(&nn->reconfig_timer) || nn->reconfig_posted); 2402 nfp_ccm_mbox_free(nn); 2403 2404 kfree(nn->dp.xsk_pools); 2405 if (nn->dp.netdev) 2406 free_netdev(nn->dp.netdev); 2407 else 2408 vfree(nn); 2409 } 2410 2411 /** 2412 * nfp_net_rss_key_sz() - Get current size of the RSS key 2413 * @nn: NFP Net device instance 2414 * 2415 * Return: size of the RSS key for currently selected hash function. 2416 */ 2417 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn) 2418 { 2419 switch (nn->rss_hfunc) { 2420 case ETH_RSS_HASH_TOP: 2421 return NFP_NET_CFG_RSS_KEY_SZ; 2422 case ETH_RSS_HASH_XOR: 2423 return 0; 2424 case ETH_RSS_HASH_CRC32: 2425 return 4; 2426 } 2427 2428 nn_warn(nn, "Unknown hash function: %u\n", nn->rss_hfunc); 2429 return 0; 2430 } 2431 2432 /** 2433 * nfp_net_rss_init() - Set the initial RSS parameters 2434 * @nn: NFP Net device to reconfigure 2435 */ 2436 static void nfp_net_rss_init(struct nfp_net *nn) 2437 { 2438 unsigned long func_bit, rss_cap_hfunc; 2439 u32 reg; 2440 2441 /* Read the RSS function capability and select first supported func */ 2442 reg = nn_readl(nn, NFP_NET_CFG_RSS_CAP); 2443 rss_cap_hfunc = FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, reg); 2444 if (!rss_cap_hfunc) 2445 rss_cap_hfunc = FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, 2446 NFP_NET_CFG_RSS_TOEPLITZ); 2447 2448 func_bit = find_first_bit(&rss_cap_hfunc, NFP_NET_CFG_RSS_HFUNCS); 2449 if (func_bit == NFP_NET_CFG_RSS_HFUNCS) { 2450 dev_warn(nn->dp.dev, 2451 "Bad RSS config, defaulting to Toeplitz hash\n"); 2452 func_bit = ETH_RSS_HASH_TOP_BIT; 2453 } 2454 nn->rss_hfunc = 1 << func_bit; 2455 2456 netdev_rss_key_fill(nn->rss_key, nfp_net_rss_key_sz(nn)); 2457 2458 nfp_net_rss_init_itbl(nn); 2459 2460 /* Enable IPv4/IPv6 TCP by default */ 2461 nn->rss_cfg = NFP_NET_CFG_RSS_IPV4_TCP | 2462 NFP_NET_CFG_RSS_IPV6_TCP | 2463 NFP_NET_CFG_RSS_IPV4_UDP | 2464 NFP_NET_CFG_RSS_IPV6_UDP | 2465 FIELD_PREP(NFP_NET_CFG_RSS_HFUNC, nn->rss_hfunc) | 2466 NFP_NET_CFG_RSS_MASK; 2467 } 2468 2469 /** 2470 * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters 2471 * @nn: NFP Net device to reconfigure 2472 */ 2473 static void nfp_net_irqmod_init(struct nfp_net *nn) 2474 { 2475 nn->rx_coalesce_usecs = 50; 2476 nn->rx_coalesce_max_frames = 64; 2477 nn->tx_coalesce_usecs = 50; 2478 nn->tx_coalesce_max_frames = 64; 2479 2480 nn->rx_coalesce_adapt_on = true; 2481 nn->tx_coalesce_adapt_on = true; 2482 } 2483 2484 static void nfp_net_netdev_init(struct nfp_net *nn) 2485 { 2486 struct net_device *netdev = nn->dp.netdev; 2487 2488 nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr); 2489 2490 netdev->mtu = nn->dp.mtu; 2491 2492 /* Advertise/enable offloads based on capabilities 2493 * 2494 * Note: netdev->features show the currently enabled features 2495 * and netdev->hw_features advertises which features are 2496 * supported. By default we enable most features. 2497 */ 2498 if (nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR) 2499 netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 2500 2501 netdev->hw_features = NETIF_F_HIGHDMA; 2502 if (nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY) { 2503 netdev->hw_features |= NETIF_F_RXCSUM; 2504 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY; 2505 } 2506 if (nn->cap & NFP_NET_CFG_CTRL_TXCSUM) { 2507 netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM; 2508 nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXCSUM; 2509 } 2510 if (nn->cap & NFP_NET_CFG_CTRL_GATHER) { 2511 netdev->hw_features |= NETIF_F_SG; 2512 nn->dp.ctrl |= NFP_NET_CFG_CTRL_GATHER; 2513 } 2514 if ((nn->cap & NFP_NET_CFG_CTRL_LSO && nn->fw_ver.major > 2) || 2515 nn->cap & NFP_NET_CFG_CTRL_LSO2) { 2516 netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6; 2517 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?: 2518 NFP_NET_CFG_CTRL_LSO; 2519 } 2520 if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) 2521 netdev->hw_features |= NETIF_F_RXHASH; 2522 2523 #ifdef CONFIG_NFP_NET_IPSEC 2524 if (nn->cap_w1 & NFP_NET_CFG_CTRL_IPSEC) 2525 netdev->hw_features |= NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM; 2526 #endif 2527 2528 if (nn->cap & NFP_NET_CFG_CTRL_VXLAN) { 2529 if (nn->cap & NFP_NET_CFG_CTRL_LSO) { 2530 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL | 2531 NETIF_F_GSO_UDP_TUNNEL_CSUM | 2532 NETIF_F_GSO_PARTIAL; 2533 netdev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM; 2534 } 2535 netdev->udp_tunnel_nic_info = &nfp_udp_tunnels; 2536 nn->dp.ctrl |= NFP_NET_CFG_CTRL_VXLAN; 2537 } 2538 if (nn->cap & NFP_NET_CFG_CTRL_NVGRE) { 2539 if (nn->cap & NFP_NET_CFG_CTRL_LSO) 2540 netdev->hw_features |= NETIF_F_GSO_GRE; 2541 nn->dp.ctrl |= NFP_NET_CFG_CTRL_NVGRE; 2542 } 2543 if (nn->cap & (NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE)) 2544 netdev->hw_enc_features = netdev->hw_features; 2545 2546 netdev->vlan_features = netdev->hw_features; 2547 2548 if (nn->cap & NFP_NET_CFG_CTRL_RXVLAN_ANY) { 2549 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX; 2550 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ?: 2551 NFP_NET_CFG_CTRL_RXVLAN; 2552 } 2553 if (nn->cap & NFP_NET_CFG_CTRL_TXVLAN_ANY) { 2554 if (nn->cap & NFP_NET_CFG_CTRL_LSO2) { 2555 nn_warn(nn, "Device advertises both TSO2 and TXVLAN. Refusing to enable TXVLAN.\n"); 2556 } else { 2557 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX; 2558 nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ?: 2559 NFP_NET_CFG_CTRL_TXVLAN; 2560 } 2561 } 2562 if (nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER) { 2563 netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER; 2564 nn->dp.ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER; 2565 } 2566 if (nn->cap & NFP_NET_CFG_CTRL_RXQINQ) { 2567 netdev->hw_features |= NETIF_F_HW_VLAN_STAG_RX; 2568 nn->dp.ctrl |= NFP_NET_CFG_CTRL_RXQINQ; 2569 } 2570 2571 netdev->features = netdev->hw_features; 2572 2573 if (nfp_app_has_tc(nn->app) && nn->port) 2574 netdev->hw_features |= NETIF_F_HW_TC; 2575 2576 /* C-Tag strip and S-Tag strip can't be supported simultaneously, 2577 * so enable C-Tag strip and disable S-Tag strip by default. 2578 */ 2579 netdev->features &= ~NETIF_F_HW_VLAN_STAG_RX; 2580 nn->dp.ctrl &= ~NFP_NET_CFG_CTRL_RXQINQ; 2581 2582 netdev->xdp_features = NETDEV_XDP_ACT_BASIC; 2583 if (nn->app && nn->app->type->id == NFP_APP_BPF_NIC) 2584 netdev->xdp_features |= NETDEV_XDP_ACT_HW_OFFLOAD; 2585 2586 /* Finalise the netdev setup */ 2587 switch (nn->dp.ops->version) { 2588 case NFP_NFD_VER_NFD3: 2589 netdev->netdev_ops = &nfp_nfd3_netdev_ops; 2590 netdev->xdp_features |= NETDEV_XDP_ACT_XSK_ZEROCOPY; 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