linux/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
<<
>>
Prefs
   1// SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
   2/* Copyright (C) 2015-2018 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/bpf_trace.h>
  17#include <linux/module.h>
  18#include <linux/kernel.h>
  19#include <linux/init.h>
  20#include <linux/fs.h>
  21#include <linux/netdevice.h>
  22#include <linux/etherdevice.h>
  23#include <linux/interrupt.h>
  24#include <linux/ip.h>
  25#include <linux/ipv6.h>
  26#include <linux/mm.h>
  27#include <linux/overflow.h>
  28#include <linux/page_ref.h>
  29#include <linux/pci.h>
  30#include <linux/pci_regs.h>
  31#include <linux/msi.h>
  32#include <linux/ethtool.h>
  33#include <linux/log2.h>
  34#include <linux/if_vlan.h>
  35#include <linux/random.h>
  36#include <linux/vmalloc.h>
  37#include <linux/ktime.h>
  38
  39#include <net/tls.h>
  40#include <net/vxlan.h>
  41
  42#include "nfpcore/nfp_nsp.h"
  43#include "ccm.h"
  44#include "nfp_app.h"
  45#include "nfp_net_ctrl.h"
  46#include "nfp_net.h"
  47#include "nfp_net_sriov.h"
  48#include "nfp_port.h"
  49#include "crypto/crypto.h"
  50#include "crypto/fw.h"
  51
  52/**
  53 * nfp_net_get_fw_version() - Read and parse the FW version
  54 * @fw_ver:     Output fw_version structure to read to
  55 * @ctrl_bar:   Mapped address of the control BAR
  56 */
  57void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
  58                            void __iomem *ctrl_bar)
  59{
  60        u32 reg;
  61
  62        reg = readl(ctrl_bar + NFP_NET_CFG_VERSION);
  63        put_unaligned_le32(reg, fw_ver);
  64}
  65
  66static dma_addr_t nfp_net_dma_map_rx(struct nfp_net_dp *dp, void *frag)
  67{
  68        return dma_map_single_attrs(dp->dev, frag + NFP_NET_RX_BUF_HEADROOM,
  69                                    dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
  70                                    dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
  71}
  72
  73static void
  74nfp_net_dma_sync_dev_rx(const struct nfp_net_dp *dp, dma_addr_t dma_addr)
  75{
  76        dma_sync_single_for_device(dp->dev, dma_addr,
  77                                   dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
  78                                   dp->rx_dma_dir);
  79}
  80
  81static void nfp_net_dma_unmap_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr)
  82{
  83        dma_unmap_single_attrs(dp->dev, dma_addr,
  84                               dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
  85                               dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
  86}
  87
  88static void nfp_net_dma_sync_cpu_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr,
  89                                    unsigned int len)
  90{
  91        dma_sync_single_for_cpu(dp->dev, dma_addr - NFP_NET_RX_BUF_HEADROOM,
  92                                len, dp->rx_dma_dir);
  93}
  94
  95/* Firmware reconfig
  96 *
  97 * Firmware reconfig may take a while so we have two versions of it -
  98 * synchronous and asynchronous (posted).  All synchronous callers are holding
  99 * RTNL so we don't have to worry about serializing them.
 100 */
 101static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update)
 102{
 103        nn_writel(nn, NFP_NET_CFG_UPDATE, update);
 104        /* ensure update is written before pinging HW */
 105        nn_pci_flush(nn);
 106        nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1);
 107        nn->reconfig_in_progress_update = update;
 108}
 109
 110/* Pass 0 as update to run posted reconfigs. */
 111static void nfp_net_reconfig_start_async(struct nfp_net *nn, u32 update)
 112{
 113        update |= nn->reconfig_posted;
 114        nn->reconfig_posted = 0;
 115
 116        nfp_net_reconfig_start(nn, update);
 117
 118        nn->reconfig_timer_active = true;
 119        mod_timer(&nn->reconfig_timer, jiffies + NFP_NET_POLL_TIMEOUT * HZ);
 120}
 121
 122static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check)
 123{
 124        u32 reg;
 125
 126        reg = nn_readl(nn, NFP_NET_CFG_UPDATE);
 127        if (reg == 0)
 128                return true;
 129        if (reg & NFP_NET_CFG_UPDATE_ERR) {
 130                nn_err(nn, "Reconfig error (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n",
 131                       reg, nn->reconfig_in_progress_update,
 132                       nn_readl(nn, NFP_NET_CFG_CTRL));
 133                return true;
 134        } else if (last_check) {
 135                nn_err(nn, "Reconfig timeout (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n",
 136                       reg, nn->reconfig_in_progress_update,
 137                       nn_readl(nn, NFP_NET_CFG_CTRL));
 138                return true;
 139        }
 140
 141        return false;
 142}
 143
 144static bool __nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline)
 145{
 146        bool timed_out = false;
 147        int i;
 148
 149        /* Poll update field, waiting for NFP to ack the config.
 150         * Do an opportunistic wait-busy loop, afterward sleep.
 151         */
 152        for (i = 0; i < 50; i++) {
 153                if (nfp_net_reconfig_check_done(nn, false))
 154                        return false;
 155                udelay(4);
 156        }
 157
 158        while (!nfp_net_reconfig_check_done(nn, timed_out)) {
 159                usleep_range(250, 500);
 160                timed_out = time_is_before_eq_jiffies(deadline);
 161        }
 162
 163        return timed_out;
 164}
 165
 166static int nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline)
 167{
 168        if (__nfp_net_reconfig_wait(nn, deadline))
 169                return -EIO;
 170
 171        if (nn_readl(nn, NFP_NET_CFG_UPDATE) & NFP_NET_CFG_UPDATE_ERR)
 172                return -EIO;
 173
 174        return 0;
 175}
 176
 177static void nfp_net_reconfig_timer(struct timer_list *t)
 178{
 179        struct nfp_net *nn = from_timer(nn, t, reconfig_timer);
 180
 181        spin_lock_bh(&nn->reconfig_lock);
 182
 183        nn->reconfig_timer_active = false;
 184
 185        /* If sync caller is present it will take over from us */
 186        if (nn->reconfig_sync_present)
 187                goto done;
 188
 189        /* Read reconfig status and report errors */
 190        nfp_net_reconfig_check_done(nn, true);
 191
 192        if (nn->reconfig_posted)
 193                nfp_net_reconfig_start_async(nn, 0);
 194done:
 195        spin_unlock_bh(&nn->reconfig_lock);
 196}
 197
 198/**
 199 * nfp_net_reconfig_post() - Post async reconfig request
 200 * @nn:      NFP Net device to reconfigure
 201 * @update:  The value for the update field in the BAR config
 202 *
 203 * Record FW reconfiguration request.  Reconfiguration will be kicked off
 204 * whenever reconfiguration machinery is idle.  Multiple requests can be
 205 * merged together!
 206 */
 207static void nfp_net_reconfig_post(struct nfp_net *nn, u32 update)
 208{
 209        spin_lock_bh(&nn->reconfig_lock);
 210
 211        /* Sync caller will kick off async reconf when it's done, just post */
 212        if (nn->reconfig_sync_present) {
 213                nn->reconfig_posted |= update;
 214                goto done;
 215        }
 216
 217        /* Opportunistically check if the previous command is done */
 218        if (!nn->reconfig_timer_active ||
 219            nfp_net_reconfig_check_done(nn, false))
 220                nfp_net_reconfig_start_async(nn, update);
 221        else
 222                nn->reconfig_posted |= update;
 223done:
 224        spin_unlock_bh(&nn->reconfig_lock);
 225}
 226
 227static void nfp_net_reconfig_sync_enter(struct nfp_net *nn)
 228{
 229        bool cancelled_timer = false;
 230        u32 pre_posted_requests;
 231
 232        spin_lock_bh(&nn->reconfig_lock);
 233
 234        WARN_ON(nn->reconfig_sync_present);
 235        nn->reconfig_sync_present = true;
 236
 237        if (nn->reconfig_timer_active) {
 238                nn->reconfig_timer_active = false;
 239                cancelled_timer = true;
 240        }
 241        pre_posted_requests = nn->reconfig_posted;
 242        nn->reconfig_posted = 0;
 243
 244        spin_unlock_bh(&nn->reconfig_lock);
 245
 246        if (cancelled_timer) {
 247                del_timer_sync(&nn->reconfig_timer);
 248                nfp_net_reconfig_wait(nn, nn->reconfig_timer.expires);
 249        }
 250
 251        /* Run the posted reconfigs which were issued before we started */
 252        if (pre_posted_requests) {
 253                nfp_net_reconfig_start(nn, pre_posted_requests);
 254                nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
 255        }
 256}
 257
 258static void nfp_net_reconfig_wait_posted(struct nfp_net *nn)
 259{
 260        nfp_net_reconfig_sync_enter(nn);
 261
 262        spin_lock_bh(&nn->reconfig_lock);
 263        nn->reconfig_sync_present = false;
 264        spin_unlock_bh(&nn->reconfig_lock);
 265}
 266
 267/**
 268 * __nfp_net_reconfig() - Reconfigure the firmware
 269 * @nn:      NFP Net device to reconfigure
 270 * @update:  The value for the update field in the BAR config
 271 *
 272 * Write the update word to the BAR and ping the reconfig queue.  The
 273 * poll until the firmware has acknowledged the update by zeroing the
 274 * update word.
 275 *
 276 * Return: Negative errno on error, 0 on success
 277 */
 278int __nfp_net_reconfig(struct nfp_net *nn, u32 update)
 279{
 280        int ret;
 281
 282        nfp_net_reconfig_sync_enter(nn);
 283
 284        nfp_net_reconfig_start(nn, update);
 285        ret = nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
 286
 287        spin_lock_bh(&nn->reconfig_lock);
 288
 289        if (nn->reconfig_posted)
 290                nfp_net_reconfig_start_async(nn, 0);
 291
 292        nn->reconfig_sync_present = false;
 293
 294        spin_unlock_bh(&nn->reconfig_lock);
 295
 296        return ret;
 297}
 298
 299int nfp_net_reconfig(struct nfp_net *nn, u32 update)
 300{
 301        int ret;
 302
 303        nn_ctrl_bar_lock(nn);
 304        ret = __nfp_net_reconfig(nn, update);
 305        nn_ctrl_bar_unlock(nn);
 306
 307        return ret;
 308}
 309
 310int nfp_net_mbox_lock(struct nfp_net *nn, unsigned int data_size)
 311{
 312        if (nn->tlv_caps.mbox_len < NFP_NET_CFG_MBOX_SIMPLE_VAL + data_size) {
 313                nn_err(nn, "mailbox too small for %u of data (%u)\n",
 314                       data_size, nn->tlv_caps.mbox_len);
 315                return -EIO;
 316        }
 317
 318        nn_ctrl_bar_lock(nn);
 319        return 0;
 320}
 321
 322/**
 323 * nfp_net_mbox_reconfig() - Reconfigure the firmware via the mailbox
 324 * @nn:        NFP Net device to reconfigure
 325 * @mbox_cmd:  The value for the mailbox command
 326 *
 327 * Helper function for mailbox updates
 328 *
 329 * Return: Negative errno on error, 0 on success
 330 */
 331int nfp_net_mbox_reconfig(struct nfp_net *nn, u32 mbox_cmd)
 332{
 333        u32 mbox = nn->tlv_caps.mbox_off;
 334        int ret;
 335
 336        nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd);
 337
 338        ret = __nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MBOX);
 339        if (ret) {
 340                nn_err(nn, "Mailbox update error\n");
 341                return ret;
 342        }
 343
 344        return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET);
 345}
 346
 347void nfp_net_mbox_reconfig_post(struct nfp_net *nn, u32 mbox_cmd)
 348{
 349        u32 mbox = nn->tlv_caps.mbox_off;
 350
 351        nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd);
 352
 353        nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_MBOX);
 354}
 355
 356int nfp_net_mbox_reconfig_wait_posted(struct nfp_net *nn)
 357{
 358        u32 mbox = nn->tlv_caps.mbox_off;
 359
 360        nfp_net_reconfig_wait_posted(nn);
 361
 362        return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET);
 363}
 364
 365int nfp_net_mbox_reconfig_and_unlock(struct nfp_net *nn, u32 mbox_cmd)
 366{
 367        int ret;
 368
 369        ret = nfp_net_mbox_reconfig(nn, mbox_cmd);
 370        nn_ctrl_bar_unlock(nn);
 371        return ret;
 372}
 373
 374/* Interrupt configuration and handling
 375 */
 376
 377/**
 378 * nfp_net_irq_unmask() - Unmask automasked interrupt
 379 * @nn:       NFP Network structure
 380 * @entry_nr: MSI-X table entry
 381 *
 382 * Clear the ICR for the IRQ entry.
 383 */
 384static void nfp_net_irq_unmask(struct nfp_net *nn, unsigned int entry_nr)
 385{
 386        nn_writeb(nn, NFP_NET_CFG_ICR(entry_nr), NFP_NET_CFG_ICR_UNMASKED);
 387        nn_pci_flush(nn);
 388}
 389
 390/**
 391 * nfp_net_irqs_alloc() - allocates MSI-X irqs
 392 * @pdev:        PCI device structure
 393 * @irq_entries: Array to be initialized and used to hold the irq entries
 394 * @min_irqs:    Minimal acceptable number of interrupts
 395 * @wanted_irqs: Target number of interrupts to allocate
 396 *
 397 * Return: Number of irqs obtained or 0 on error.
 398 */
 399unsigned int
 400nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries,
 401                   unsigned int min_irqs, unsigned int wanted_irqs)
 402{
 403        unsigned int i;
 404        int got_irqs;
 405
 406        for (i = 0; i < wanted_irqs; i++)
 407                irq_entries[i].entry = i;
 408
 409        got_irqs = pci_enable_msix_range(pdev, irq_entries,
 410                                         min_irqs, wanted_irqs);
 411        if (got_irqs < 0) {
 412                dev_err(&pdev->dev, "Failed to enable %d-%d MSI-X (err=%d)\n",
 413                        min_irqs, wanted_irqs, got_irqs);
 414                return 0;
 415        }
 416
 417        if (got_irqs < wanted_irqs)
 418                dev_warn(&pdev->dev, "Unable to allocate %d IRQs got only %d\n",
 419                         wanted_irqs, got_irqs);
 420
 421        return got_irqs;
 422}
 423
 424/**
 425 * nfp_net_irqs_assign() - Assign interrupts allocated externally to netdev
 426 * @nn:          NFP Network structure
 427 * @irq_entries: Table of allocated interrupts
 428 * @n:           Size of @irq_entries (number of entries to grab)
 429 *
 430 * After interrupts are allocated with nfp_net_irqs_alloc() this function
 431 * should be called to assign them to a specific netdev (port).
 432 */
 433void
 434nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries,
 435                    unsigned int n)
 436{
 437        struct nfp_net_dp *dp = &nn->dp;
 438
 439        nn->max_r_vecs = n - NFP_NET_NON_Q_VECTORS;
 440        dp->num_r_vecs = nn->max_r_vecs;
 441
 442        memcpy(nn->irq_entries, irq_entries, sizeof(*irq_entries) * n);
 443
 444        if (dp->num_rx_rings > dp->num_r_vecs ||
 445            dp->num_tx_rings > dp->num_r_vecs)
 446                dev_warn(nn->dp.dev, "More rings (%d,%d) than vectors (%d).\n",
 447                         dp->num_rx_rings, dp->num_tx_rings,
 448                         dp->num_r_vecs);
 449
 450        dp->num_rx_rings = min(dp->num_r_vecs, dp->num_rx_rings);
 451        dp->num_tx_rings = min(dp->num_r_vecs, dp->num_tx_rings);
 452        dp->num_stack_tx_rings = dp->num_tx_rings;
 453}
 454
 455/**
 456 * nfp_net_irqs_disable() - Disable interrupts
 457 * @pdev:        PCI device structure
 458 *
 459 * Undoes what @nfp_net_irqs_alloc() does.
 460 */
 461void nfp_net_irqs_disable(struct pci_dev *pdev)
 462{
 463        pci_disable_msix(pdev);
 464}
 465
 466/**
 467 * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings.
 468 * @irq:      Interrupt
 469 * @data:     Opaque data structure
 470 *
 471 * Return: Indicate if the interrupt has been handled.
 472 */
 473static irqreturn_t nfp_net_irq_rxtx(int irq, void *data)
 474{
 475        struct nfp_net_r_vector *r_vec = data;
 476
 477        napi_schedule_irqoff(&r_vec->napi);
 478
 479        /* The FW auto-masks any interrupt, either via the MASK bit in
 480         * the MSI-X table or via the per entry ICR field.  So there
 481         * is no need to disable interrupts here.
 482         */
 483        return IRQ_HANDLED;
 484}
 485
 486static irqreturn_t nfp_ctrl_irq_rxtx(int irq, void *data)
 487{
 488        struct nfp_net_r_vector *r_vec = data;
 489
 490        tasklet_schedule(&r_vec->tasklet);
 491
 492        return IRQ_HANDLED;
 493}
 494
 495/**
 496 * nfp_net_read_link_status() - Reread link status from control BAR
 497 * @nn:       NFP Network structure
 498 */
 499static void nfp_net_read_link_status(struct nfp_net *nn)
 500{
 501        unsigned long flags;
 502        bool link_up;
 503        u32 sts;
 504
 505        spin_lock_irqsave(&nn->link_status_lock, flags);
 506
 507        sts = nn_readl(nn, NFP_NET_CFG_STS);
 508        link_up = !!(sts & NFP_NET_CFG_STS_LINK);
 509
 510        if (nn->link_up == link_up)
 511                goto out;
 512
 513        nn->link_up = link_up;
 514        if (nn->port)
 515                set_bit(NFP_PORT_CHANGED, &nn->port->flags);
 516
 517        if (nn->link_up) {
 518                netif_carrier_on(nn->dp.netdev);
 519                netdev_info(nn->dp.netdev, "NIC Link is Up\n");
 520        } else {
 521                netif_carrier_off(nn->dp.netdev);
 522                netdev_info(nn->dp.netdev, "NIC Link is Down\n");
 523        }
 524out:
 525        spin_unlock_irqrestore(&nn->link_status_lock, flags);
 526}
 527
 528/**
 529 * nfp_net_irq_lsc() - Interrupt service routine for link state changes
 530 * @irq:      Interrupt
 531 * @data:     Opaque data structure
 532 *
 533 * Return: Indicate if the interrupt has been handled.
 534 */
 535static irqreturn_t nfp_net_irq_lsc(int irq, void *data)
 536{
 537        struct nfp_net *nn = data;
 538        struct msix_entry *entry;
 539
 540        entry = &nn->irq_entries[NFP_NET_IRQ_LSC_IDX];
 541
 542        nfp_net_read_link_status(nn);
 543
 544        nfp_net_irq_unmask(nn, entry->entry);
 545
 546        return IRQ_HANDLED;
 547}
 548
 549/**
 550 * nfp_net_irq_exn() - Interrupt service routine for exceptions
 551 * @irq:      Interrupt
 552 * @data:     Opaque data structure
 553 *
 554 * Return: Indicate if the interrupt has been handled.
 555 */
 556static irqreturn_t nfp_net_irq_exn(int irq, void *data)
 557{
 558        struct nfp_net *nn = data;
 559
 560        nn_err(nn, "%s: UNIMPLEMENTED.\n", __func__);
 561        /* XXX TO BE IMPLEMENTED */
 562        return IRQ_HANDLED;
 563}
 564
 565/**
 566 * nfp_net_tx_ring_init() - Fill in the boilerplate for a TX ring
 567 * @tx_ring:  TX ring structure
 568 * @r_vec:    IRQ vector servicing this ring
 569 * @idx:      Ring index
 570 * @is_xdp:   Is this an XDP TX ring?
 571 */
 572static void
 573nfp_net_tx_ring_init(struct nfp_net_tx_ring *tx_ring,
 574                     struct nfp_net_r_vector *r_vec, unsigned int idx,
 575                     bool is_xdp)
 576{
 577        struct nfp_net *nn = r_vec->nfp_net;
 578
 579        tx_ring->idx = idx;
 580        tx_ring->r_vec = r_vec;
 581        tx_ring->is_xdp = is_xdp;
 582        u64_stats_init(&tx_ring->r_vec->tx_sync);
 583
 584        tx_ring->qcidx = tx_ring->idx * nn->stride_tx;
 585        tx_ring->qcp_q = nn->tx_bar + NFP_QCP_QUEUE_OFF(tx_ring->qcidx);
 586}
 587
 588/**
 589 * nfp_net_rx_ring_init() - Fill in the boilerplate for a RX ring
 590 * @rx_ring:  RX ring structure
 591 * @r_vec:    IRQ vector servicing this ring
 592 * @idx:      Ring index
 593 */
 594static void
 595nfp_net_rx_ring_init(struct nfp_net_rx_ring *rx_ring,
 596                     struct nfp_net_r_vector *r_vec, unsigned int idx)
 597{
 598        struct nfp_net *nn = r_vec->nfp_net;
 599
 600        rx_ring->idx = idx;
 601        rx_ring->r_vec = r_vec;
 602        u64_stats_init(&rx_ring->r_vec->rx_sync);
 603
 604        rx_ring->fl_qcidx = rx_ring->idx * nn->stride_rx;
 605        rx_ring->qcp_fl = nn->rx_bar + NFP_QCP_QUEUE_OFF(rx_ring->fl_qcidx);
 606}
 607
 608/**
 609 * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN)
 610 * @nn:         NFP Network structure
 611 * @ctrl_offset: Control BAR offset where IRQ configuration should be written
 612 * @format:     printf-style format to construct the interrupt name
 613 * @name:       Pointer to allocated space for interrupt name
 614 * @name_sz:    Size of space for interrupt name
 615 * @vector_idx: Index of MSI-X vector used for this interrupt
 616 * @handler:    IRQ handler to register for this interrupt
 617 */
 618static int
 619nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset,
 620                        const char *format, char *name, size_t name_sz,
 621                        unsigned int vector_idx, irq_handler_t handler)
 622{
 623        struct msix_entry *entry;
 624        int err;
 625
 626        entry = &nn->irq_entries[vector_idx];
 627
 628        snprintf(name, name_sz, format, nfp_net_name(nn));
 629        err = request_irq(entry->vector, handler, 0, name, nn);
 630        if (err) {
 631                nn_err(nn, "Failed to request IRQ %d (err=%d).\n",
 632                       entry->vector, err);
 633                return err;
 634        }
 635        nn_writeb(nn, ctrl_offset, entry->entry);
 636        nfp_net_irq_unmask(nn, entry->entry);
 637
 638        return 0;
 639}
 640
 641/**
 642 * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN)
 643 * @nn:         NFP Network structure
 644 * @ctrl_offset: Control BAR offset where IRQ configuration should be written
 645 * @vector_idx: Index of MSI-X vector used for this interrupt
 646 */
 647static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset,
 648                                 unsigned int vector_idx)
 649{
 650        nn_writeb(nn, ctrl_offset, 0xff);
 651        nn_pci_flush(nn);
 652        free_irq(nn->irq_entries[vector_idx].vector, nn);
 653}
 654
 655/* Transmit
 656 *
 657 * One queue controller peripheral queue is used for transmit.  The
 658 * driver en-queues packets for transmit by advancing the write
 659 * pointer.  The device indicates that packets have transmitted by
 660 * advancing the read pointer.  The driver maintains a local copy of
 661 * the read and write pointer in @struct nfp_net_tx_ring.  The driver
 662 * keeps @wr_p in sync with the queue controller write pointer and can
 663 * determine how many packets have been transmitted by comparing its
 664 * copy of the read pointer @rd_p with the read pointer maintained by
 665 * the queue controller peripheral.
 666 */
 667
 668/**
 669 * nfp_net_tx_full() - Check if the TX ring is full
 670 * @tx_ring: TX ring to check
 671 * @dcnt:    Number of descriptors that need to be enqueued (must be >= 1)
 672 *
 673 * This function checks, based on the *host copy* of read/write
 674 * pointer if a given TX ring is full.  The real TX queue may have
 675 * some newly made available slots.
 676 *
 677 * Return: True if the ring is full.
 678 */
 679static int nfp_net_tx_full(struct nfp_net_tx_ring *tx_ring, int dcnt)
 680{
 681        return (tx_ring->wr_p - tx_ring->rd_p) >= (tx_ring->cnt - dcnt);
 682}
 683
 684/* Wrappers for deciding when to stop and restart TX queues */
 685static int nfp_net_tx_ring_should_wake(struct nfp_net_tx_ring *tx_ring)
 686{
 687        return !nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS * 4);
 688}
 689
 690static int nfp_net_tx_ring_should_stop(struct nfp_net_tx_ring *tx_ring)
 691{
 692        return nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS + 1);
 693}
 694
 695/**
 696 * nfp_net_tx_ring_stop() - stop tx ring
 697 * @nd_q:    netdev queue
 698 * @tx_ring: driver tx queue structure
 699 *
 700 * Safely stop TX ring.  Remember that while we are running .start_xmit()
 701 * someone else may be cleaning the TX ring completions so we need to be
 702 * extra careful here.
 703 */
 704static void nfp_net_tx_ring_stop(struct netdev_queue *nd_q,
 705                                 struct nfp_net_tx_ring *tx_ring)
 706{
 707        netif_tx_stop_queue(nd_q);
 708
 709        /* We can race with the TX completion out of NAPI so recheck */
 710        smp_mb();
 711        if (unlikely(nfp_net_tx_ring_should_wake(tx_ring)))
 712                netif_tx_start_queue(nd_q);
 713}
 714
 715/**
 716 * nfp_net_tx_tso() - Set up Tx descriptor for LSO
 717 * @r_vec: per-ring structure
 718 * @txbuf: Pointer to driver soft TX descriptor
 719 * @txd: Pointer to HW TX descriptor
 720 * @skb: Pointer to SKB
 721 * @md_bytes: Prepend length
 722 *
 723 * Set up Tx descriptor for LSO, do nothing for non-LSO skbs.
 724 * Return error on packet header greater than maximum supported LSO header size.
 725 */
 726static void nfp_net_tx_tso(struct nfp_net_r_vector *r_vec,
 727                           struct nfp_net_tx_buf *txbuf,
 728                           struct nfp_net_tx_desc *txd, struct sk_buff *skb,
 729                           u32 md_bytes)
 730{
 731        u32 l3_offset, l4_offset, hdrlen;
 732        u16 mss;
 733
 734        if (!skb_is_gso(skb))
 735                return;
 736
 737        if (!skb->encapsulation) {
 738                l3_offset = skb_network_offset(skb);
 739                l4_offset = skb_transport_offset(skb);
 740                hdrlen = skb_transport_offset(skb) + tcp_hdrlen(skb);
 741        } else {
 742                l3_offset = skb_inner_network_offset(skb);
 743                l4_offset = skb_inner_transport_offset(skb);
 744                hdrlen = skb_inner_transport_header(skb) - skb->data +
 745                        inner_tcp_hdrlen(skb);
 746        }
 747
 748        txbuf->pkt_cnt = skb_shinfo(skb)->gso_segs;
 749        txbuf->real_len += hdrlen * (txbuf->pkt_cnt - 1);
 750
 751        mss = skb_shinfo(skb)->gso_size & PCIE_DESC_TX_MSS_MASK;
 752        txd->l3_offset = l3_offset - md_bytes;
 753        txd->l4_offset = l4_offset - md_bytes;
 754        txd->lso_hdrlen = hdrlen - md_bytes;
 755        txd->mss = cpu_to_le16(mss);
 756        txd->flags |= PCIE_DESC_TX_LSO;
 757
 758        u64_stats_update_begin(&r_vec->tx_sync);
 759        r_vec->tx_lso++;
 760        u64_stats_update_end(&r_vec->tx_sync);
 761}
 762
 763/**
 764 * nfp_net_tx_csum() - Set TX CSUM offload flags in TX descriptor
 765 * @dp:  NFP Net data path struct
 766 * @r_vec: per-ring structure
 767 * @txbuf: Pointer to driver soft TX descriptor
 768 * @txd: Pointer to TX descriptor
 769 * @skb: Pointer to SKB
 770 *
 771 * This function sets the TX checksum flags in the TX descriptor based
 772 * on the configuration and the protocol of the packet to be transmitted.
 773 */
 774static void nfp_net_tx_csum(struct nfp_net_dp *dp,
 775                            struct nfp_net_r_vector *r_vec,
 776                            struct nfp_net_tx_buf *txbuf,
 777                            struct nfp_net_tx_desc *txd, struct sk_buff *skb)
 778{
 779        struct ipv6hdr *ipv6h;
 780        struct iphdr *iph;
 781        u8 l4_hdr;
 782
 783        if (!(dp->ctrl & NFP_NET_CFG_CTRL_TXCSUM))
 784                return;
 785
 786        if (skb->ip_summed != CHECKSUM_PARTIAL)
 787                return;
 788
 789        txd->flags |= PCIE_DESC_TX_CSUM;
 790        if (skb->encapsulation)
 791                txd->flags |= PCIE_DESC_TX_ENCAP;
 792
 793        iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
 794        ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
 795
 796        if (iph->version == 4) {
 797                txd->flags |= PCIE_DESC_TX_IP4_CSUM;
 798                l4_hdr = iph->protocol;
 799        } else if (ipv6h->version == 6) {
 800                l4_hdr = ipv6h->nexthdr;
 801        } else {
 802                nn_dp_warn(dp, "partial checksum but ipv=%x!\n", iph->version);
 803                return;
 804        }
 805
 806        switch (l4_hdr) {
 807        case IPPROTO_TCP:
 808                txd->flags |= PCIE_DESC_TX_TCP_CSUM;
 809                break;
 810        case IPPROTO_UDP:
 811                txd->flags |= PCIE_DESC_TX_UDP_CSUM;
 812                break;
 813        default:
 814                nn_dp_warn(dp, "partial checksum but l4 proto=%x!\n", l4_hdr);
 815                return;
 816        }
 817
 818        u64_stats_update_begin(&r_vec->tx_sync);
 819        if (skb->encapsulation)
 820                r_vec->hw_csum_tx_inner += txbuf->pkt_cnt;
 821        else
 822                r_vec->hw_csum_tx += txbuf->pkt_cnt;
 823        u64_stats_update_end(&r_vec->tx_sync);
 824}
 825
 826static struct sk_buff *
 827nfp_net_tls_tx(struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec,
 828               struct sk_buff *skb, u64 *tls_handle, int *nr_frags)
 829{
 830#ifdef CONFIG_TLS_DEVICE
 831        struct nfp_net_tls_offload_ctx *ntls;
 832        struct sk_buff *nskb;
 833        bool resync_pending;
 834        u32 datalen, seq;
 835
 836        if (likely(!dp->ktls_tx))
 837                return skb;
 838        if (!skb->sk || !tls_is_sk_tx_device_offloaded(skb->sk))
 839                return skb;
 840
 841        datalen = skb->len - (skb_transport_offset(skb) + tcp_hdrlen(skb));
 842        seq = ntohl(tcp_hdr(skb)->seq);
 843        ntls = tls_driver_ctx(skb->sk, TLS_OFFLOAD_CTX_DIR_TX);
 844        resync_pending = tls_offload_tx_resync_pending(skb->sk);
 845        if (unlikely(resync_pending || ntls->next_seq != seq)) {
 846                /* Pure ACK out of order already */
 847                if (!datalen)
 848                        return skb;
 849
 850                u64_stats_update_begin(&r_vec->tx_sync);
 851                r_vec->tls_tx_fallback++;
 852                u64_stats_update_end(&r_vec->tx_sync);
 853
 854                nskb = tls_encrypt_skb(skb);
 855                if (!nskb) {
 856                        u64_stats_update_begin(&r_vec->tx_sync);
 857                        r_vec->tls_tx_no_fallback++;
 858                        u64_stats_update_end(&r_vec->tx_sync);
 859                        return NULL;
 860                }
 861                /* encryption wasn't necessary */
 862                if (nskb == skb)
 863                        return skb;
 864                /* we don't re-check ring space */
 865                if (unlikely(skb_is_nonlinear(nskb))) {
 866                        nn_dp_warn(dp, "tls_encrypt_skb() produced fragmented frame\n");
 867                        u64_stats_update_begin(&r_vec->tx_sync);
 868                        r_vec->tx_errors++;
 869                        u64_stats_update_end(&r_vec->tx_sync);
 870                        dev_kfree_skb_any(nskb);
 871                        return NULL;
 872                }
 873
 874                /* jump forward, a TX may have gotten lost, need to sync TX */
 875                if (!resync_pending && seq - ntls->next_seq < U32_MAX / 4)
 876                        tls_offload_tx_resync_request(nskb->sk, seq,
 877                                                      ntls->next_seq);
 878
 879                *nr_frags = 0;
 880                return nskb;
 881        }
 882
 883        if (datalen) {
 884                u64_stats_update_begin(&r_vec->tx_sync);
 885                if (!skb_is_gso(skb))
 886                        r_vec->hw_tls_tx++;
 887                else
 888                        r_vec->hw_tls_tx += skb_shinfo(skb)->gso_segs;
 889                u64_stats_update_end(&r_vec->tx_sync);
 890        }
 891
 892        memcpy(tls_handle, ntls->fw_handle, sizeof(ntls->fw_handle));
 893        ntls->next_seq += datalen;
 894#endif
 895        return skb;
 896}
 897
 898static void nfp_net_tls_tx_undo(struct sk_buff *skb, u64 tls_handle)
 899{
 900#ifdef CONFIG_TLS_DEVICE
 901        struct nfp_net_tls_offload_ctx *ntls;
 902        u32 datalen, seq;
 903
 904        if (!tls_handle)
 905                return;
 906        if (WARN_ON_ONCE(!skb->sk || !tls_is_sk_tx_device_offloaded(skb->sk)))
 907                return;
 908
 909        datalen = skb->len - (skb_transport_offset(skb) + tcp_hdrlen(skb));
 910        seq = ntohl(tcp_hdr(skb)->seq);
 911
 912        ntls = tls_driver_ctx(skb->sk, TLS_OFFLOAD_CTX_DIR_TX);
 913        if (ntls->next_seq == seq + datalen)
 914                ntls->next_seq = seq;
 915        else
 916                WARN_ON_ONCE(1);
 917#endif
 918}
 919
 920static void nfp_net_tx_xmit_more_flush(struct nfp_net_tx_ring *tx_ring)
 921{
 922        wmb();
 923        nfp_qcp_wr_ptr_add(tx_ring->qcp_q, tx_ring->wr_ptr_add);
 924        tx_ring->wr_ptr_add = 0;
 925}
 926
 927static int nfp_net_prep_tx_meta(struct sk_buff *skb, u64 tls_handle)
 928{
 929        struct metadata_dst *md_dst = skb_metadata_dst(skb);
 930        unsigned char *data;
 931        u32 meta_id = 0;
 932        int md_bytes;
 933
 934        if (likely(!md_dst && !tls_handle))
 935                return 0;
 936        if (unlikely(md_dst && md_dst->type != METADATA_HW_PORT_MUX)) {
 937                if (!tls_handle)
 938                        return 0;
 939                md_dst = NULL;
 940        }
 941
 942        md_bytes = 4 + !!md_dst * 4 + !!tls_handle * 8;
 943
 944        if (unlikely(skb_cow_head(skb, md_bytes)))
 945                return -ENOMEM;
 946
 947        meta_id = 0;
 948        data = skb_push(skb, md_bytes) + md_bytes;
 949        if (md_dst) {
 950                data -= 4;
 951                put_unaligned_be32(md_dst->u.port_info.port_id, data);
 952                meta_id = NFP_NET_META_PORTID;
 953        }
 954        if (tls_handle) {
 955                /* conn handle is opaque, we just use u64 to be able to quickly
 956                 * compare it to zero
 957                 */
 958                data -= 8;
 959                memcpy(data, &tls_handle, sizeof(tls_handle));
 960                meta_id <<= NFP_NET_META_FIELD_SIZE;
 961                meta_id |= NFP_NET_META_CONN_HANDLE;
 962        }
 963
 964        data -= 4;
 965        put_unaligned_be32(meta_id, data);
 966
 967        return md_bytes;
 968}
 969
 970/**
 971 * nfp_net_tx() - Main transmit entry point
 972 * @skb:    SKB to transmit
 973 * @netdev: netdev structure
 974 *
 975 * Return: NETDEV_TX_OK on success.
 976 */
 977static netdev_tx_t nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
 978{
 979        struct nfp_net *nn = netdev_priv(netdev);
 980        const skb_frag_t *frag;
 981        int f, nr_frags, wr_idx, md_bytes;
 982        struct nfp_net_tx_ring *tx_ring;
 983        struct nfp_net_r_vector *r_vec;
 984        struct nfp_net_tx_buf *txbuf;
 985        struct nfp_net_tx_desc *txd;
 986        struct netdev_queue *nd_q;
 987        struct nfp_net_dp *dp;
 988        dma_addr_t dma_addr;
 989        unsigned int fsize;
 990        u64 tls_handle = 0;
 991        u16 qidx;
 992
 993        dp = &nn->dp;
 994        qidx = skb_get_queue_mapping(skb);
 995        tx_ring = &dp->tx_rings[qidx];
 996        r_vec = tx_ring->r_vec;
 997
 998        nr_frags = skb_shinfo(skb)->nr_frags;
 999
1000        if (unlikely(nfp_net_tx_full(tx_ring, nr_frags + 1))) {
1001                nn_dp_warn(dp, "TX ring %d busy. wrp=%u rdp=%u\n",
1002                           qidx, tx_ring->wr_p, tx_ring->rd_p);
1003                nd_q = netdev_get_tx_queue(dp->netdev, qidx);
1004                netif_tx_stop_queue(nd_q);
1005                nfp_net_tx_xmit_more_flush(tx_ring);
1006                u64_stats_update_begin(&r_vec->tx_sync);
1007                r_vec->tx_busy++;
1008                u64_stats_update_end(&r_vec->tx_sync);
1009                return NETDEV_TX_BUSY;
1010        }
1011
1012        skb = nfp_net_tls_tx(dp, r_vec, skb, &tls_handle, &nr_frags);
1013        if (unlikely(!skb)) {
1014                nfp_net_tx_xmit_more_flush(tx_ring);
1015                return NETDEV_TX_OK;
1016        }
1017
1018        md_bytes = nfp_net_prep_tx_meta(skb, tls_handle);
1019        if (unlikely(md_bytes < 0))
1020                goto err_flush;
1021
1022        /* Start with the head skbuf */
1023        dma_addr = dma_map_single(dp->dev, skb->data, skb_headlen(skb),
1024                                  DMA_TO_DEVICE);
1025        if (dma_mapping_error(dp->dev, dma_addr))
1026                goto err_dma_err;
1027
1028        wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
1029
1030        /* Stash the soft descriptor of the head then initialize it */
1031        txbuf = &tx_ring->txbufs[wr_idx];
1032        txbuf->skb = skb;
1033        txbuf->dma_addr = dma_addr;
1034        txbuf->fidx = -1;
1035        txbuf->pkt_cnt = 1;
1036        txbuf->real_len = skb->len;
1037
1038        /* Build TX descriptor */
1039        txd = &tx_ring->txds[wr_idx];
1040        txd->offset_eop = (nr_frags ? 0 : PCIE_DESC_TX_EOP) | md_bytes;
1041        txd->dma_len = cpu_to_le16(skb_headlen(skb));
1042        nfp_desc_set_dma_addr(txd, dma_addr);
1043        txd->data_len = cpu_to_le16(skb->len);
1044
1045        txd->flags = 0;
1046        txd->mss = 0;
1047        txd->lso_hdrlen = 0;
1048
1049        /* Do not reorder - tso may adjust pkt cnt, vlan may override fields */
1050        nfp_net_tx_tso(r_vec, txbuf, txd, skb, md_bytes);
1051        nfp_net_tx_csum(dp, r_vec, txbuf, txd, skb);
1052        if (skb_vlan_tag_present(skb) && dp->ctrl & NFP_NET_CFG_CTRL_TXVLAN) {
1053                txd->flags |= PCIE_DESC_TX_VLAN;
1054                txd->vlan = cpu_to_le16(skb_vlan_tag_get(skb));
1055        }
1056
1057        /* Gather DMA */
1058        if (nr_frags > 0) {
1059                __le64 second_half;
1060
1061                /* all descs must match except for in addr, length and eop */
1062                second_half = txd->vals8[1];
1063
1064                for (f = 0; f < nr_frags; f++) {
1065                        frag = &skb_shinfo(skb)->frags[f];
1066                        fsize = skb_frag_size(frag);
1067
1068                        dma_addr = skb_frag_dma_map(dp->dev, frag, 0,
1069                                                    fsize, DMA_TO_DEVICE);
1070                        if (dma_mapping_error(dp->dev, dma_addr))
1071                                goto err_unmap;
1072
1073                        wr_idx = D_IDX(tx_ring, wr_idx + 1);
1074                        tx_ring->txbufs[wr_idx].skb = skb;
1075                        tx_ring->txbufs[wr_idx].dma_addr = dma_addr;
1076                        tx_ring->txbufs[wr_idx].fidx = f;
1077
1078                        txd = &tx_ring->txds[wr_idx];
1079                        txd->dma_len = cpu_to_le16(fsize);
1080                        nfp_desc_set_dma_addr(txd, dma_addr);
1081                        txd->offset_eop = md_bytes |
1082                                ((f == nr_frags - 1) ? PCIE_DESC_TX_EOP : 0);
1083                        txd->vals8[1] = second_half;
1084                }
1085
1086                u64_stats_update_begin(&r_vec->tx_sync);
1087                r_vec->tx_gather++;
1088                u64_stats_update_end(&r_vec->tx_sync);
1089        }
1090
1091        skb_tx_timestamp(skb);
1092
1093        nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1094
1095        tx_ring->wr_p += nr_frags + 1;
1096        if (nfp_net_tx_ring_should_stop(tx_ring))
1097                nfp_net_tx_ring_stop(nd_q, tx_ring);
1098
1099        tx_ring->wr_ptr_add += nr_frags + 1;
1100        if (__netdev_tx_sent_queue(nd_q, txbuf->real_len, netdev_xmit_more()))
1101                nfp_net_tx_xmit_more_flush(tx_ring);
1102
1103        return NETDEV_TX_OK;
1104
1105err_unmap:
1106        while (--f >= 0) {
1107                frag = &skb_shinfo(skb)->frags[f];
1108                dma_unmap_page(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
1109                               skb_frag_size(frag), DMA_TO_DEVICE);
1110                tx_ring->txbufs[wr_idx].skb = NULL;
1111                tx_ring->txbufs[wr_idx].dma_addr = 0;
1112                tx_ring->txbufs[wr_idx].fidx = -2;
1113                wr_idx = wr_idx - 1;
1114                if (wr_idx < 0)
1115                        wr_idx += tx_ring->cnt;
1116        }
1117        dma_unmap_single(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
1118                         skb_headlen(skb), DMA_TO_DEVICE);
1119        tx_ring->txbufs[wr_idx].skb = NULL;
1120        tx_ring->txbufs[wr_idx].dma_addr = 0;
1121        tx_ring->txbufs[wr_idx].fidx = -2;
1122err_dma_err:
1123        nn_dp_warn(dp, "Failed to map DMA TX buffer\n");
1124err_flush:
1125        nfp_net_tx_xmit_more_flush(tx_ring);
1126        u64_stats_update_begin(&r_vec->tx_sync);
1127        r_vec->tx_errors++;
1128        u64_stats_update_end(&r_vec->tx_sync);
1129        nfp_net_tls_tx_undo(skb, tls_handle);
1130        dev_kfree_skb_any(skb);
1131        return NETDEV_TX_OK;
1132}
1133
1134/**
1135 * nfp_net_tx_complete() - Handled completed TX packets
1136 * @tx_ring:    TX ring structure
1137 * @budget:     NAPI budget (only used as bool to determine if in NAPI context)
1138 */
1139static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring, int budget)
1140{
1141        struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
1142        struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
1143        struct netdev_queue *nd_q;
1144        u32 done_pkts = 0, done_bytes = 0;
1145        u32 qcp_rd_p;
1146        int todo;
1147
1148        if (tx_ring->wr_p == tx_ring->rd_p)
1149                return;
1150
1151        /* Work out how many descriptors have been transmitted */
1152        qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
1153
1154        if (qcp_rd_p == tx_ring->qcp_rd_p)
1155                return;
1156
1157        todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
1158
1159        while (todo--) {
1160                const skb_frag_t *frag;
1161                struct nfp_net_tx_buf *tx_buf;
1162                struct sk_buff *skb;
1163                int fidx, nr_frags;
1164                int idx;
1165
1166                idx = D_IDX(tx_ring, tx_ring->rd_p++);
1167                tx_buf = &tx_ring->txbufs[idx];
1168
1169                skb = tx_buf->skb;
1170                if (!skb)
1171                        continue;
1172
1173                nr_frags = skb_shinfo(skb)->nr_frags;
1174                fidx = tx_buf->fidx;
1175
1176                if (fidx == -1) {
1177                        /* unmap head */
1178                        dma_unmap_single(dp->dev, tx_buf->dma_addr,
1179                                         skb_headlen(skb), DMA_TO_DEVICE);
1180
1181                        done_pkts += tx_buf->pkt_cnt;
1182                        done_bytes += tx_buf->real_len;
1183                } else {
1184                        /* unmap fragment */
1185                        frag = &skb_shinfo(skb)->frags[fidx];
1186                        dma_unmap_page(dp->dev, tx_buf->dma_addr,
1187                                       skb_frag_size(frag), DMA_TO_DEVICE);
1188                }
1189
1190                /* check for last gather fragment */
1191                if (fidx == nr_frags - 1)
1192                        napi_consume_skb(skb, budget);
1193
1194                tx_buf->dma_addr = 0;
1195                tx_buf->skb = NULL;
1196                tx_buf->fidx = -2;
1197        }
1198
1199        tx_ring->qcp_rd_p = qcp_rd_p;
1200
1201        u64_stats_update_begin(&r_vec->tx_sync);
1202        r_vec->tx_bytes += done_bytes;
1203        r_vec->tx_pkts += done_pkts;
1204        u64_stats_update_end(&r_vec->tx_sync);
1205
1206        if (!dp->netdev)
1207                return;
1208
1209        nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1210        netdev_tx_completed_queue(nd_q, done_pkts, done_bytes);
1211        if (nfp_net_tx_ring_should_wake(tx_ring)) {
1212                /* Make sure TX thread will see updated tx_ring->rd_p */
1213                smp_mb();
1214
1215                if (unlikely(netif_tx_queue_stopped(nd_q)))
1216                        netif_tx_wake_queue(nd_q);
1217        }
1218
1219        WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1220                  "TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1221                  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1222}
1223
1224static bool nfp_net_xdp_complete(struct nfp_net_tx_ring *tx_ring)
1225{
1226        struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
1227        u32 done_pkts = 0, done_bytes = 0;
1228        bool done_all;
1229        int idx, todo;
1230        u32 qcp_rd_p;
1231
1232        /* Work out how many descriptors have been transmitted */
1233        qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
1234
1235        if (qcp_rd_p == tx_ring->qcp_rd_p)
1236                return true;
1237
1238        todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
1239
1240        done_all = todo <= NFP_NET_XDP_MAX_COMPLETE;
1241        todo = min(todo, NFP_NET_XDP_MAX_COMPLETE);
1242
1243        tx_ring->qcp_rd_p = D_IDX(tx_ring, tx_ring->qcp_rd_p + todo);
1244
1245        done_pkts = todo;
1246        while (todo--) {
1247                idx = D_IDX(tx_ring, tx_ring->rd_p);
1248                tx_ring->rd_p++;
1249
1250                done_bytes += tx_ring->txbufs[idx].real_len;
1251        }
1252
1253        u64_stats_update_begin(&r_vec->tx_sync);
1254        r_vec->tx_bytes += done_bytes;
1255        r_vec->tx_pkts += done_pkts;
1256        u64_stats_update_end(&r_vec->tx_sync);
1257
1258        WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1259                  "XDP TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1260                  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1261
1262        return done_all;
1263}
1264
1265/**
1266 * nfp_net_tx_ring_reset() - Free any untransmitted buffers and reset pointers
1267 * @dp:         NFP Net data path struct
1268 * @tx_ring:    TX ring structure
1269 *
1270 * Assumes that the device is stopped, must be idempotent.
1271 */
1272static void
1273nfp_net_tx_ring_reset(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
1274{
1275        const skb_frag_t *frag;
1276        struct netdev_queue *nd_q;
1277
1278        while (!tx_ring->is_xdp && tx_ring->rd_p != tx_ring->wr_p) {
1279                struct nfp_net_tx_buf *tx_buf;
1280                struct sk_buff *skb;
1281                int idx, nr_frags;
1282
1283                idx = D_IDX(tx_ring, tx_ring->rd_p);
1284                tx_buf = &tx_ring->txbufs[idx];
1285
1286                skb = tx_ring->txbufs[idx].skb;
1287                nr_frags = skb_shinfo(skb)->nr_frags;
1288
1289                if (tx_buf->fidx == -1) {
1290                        /* unmap head */
1291                        dma_unmap_single(dp->dev, tx_buf->dma_addr,
1292                                         skb_headlen(skb), DMA_TO_DEVICE);
1293                } else {
1294                        /* unmap fragment */
1295                        frag = &skb_shinfo(skb)->frags[tx_buf->fidx];
1296                        dma_unmap_page(dp->dev, tx_buf->dma_addr,
1297                                       skb_frag_size(frag), DMA_TO_DEVICE);
1298                }
1299
1300                /* check for last gather fragment */
1301                if (tx_buf->fidx == nr_frags - 1)
1302                        dev_kfree_skb_any(skb);
1303
1304                tx_buf->dma_addr = 0;
1305                tx_buf->skb = NULL;
1306                tx_buf->fidx = -2;
1307
1308                tx_ring->qcp_rd_p++;
1309                tx_ring->rd_p++;
1310        }
1311
1312        memset(tx_ring->txds, 0, tx_ring->size);
1313        tx_ring->wr_p = 0;
1314        tx_ring->rd_p = 0;
1315        tx_ring->qcp_rd_p = 0;
1316        tx_ring->wr_ptr_add = 0;
1317
1318        if (tx_ring->is_xdp || !dp->netdev)
1319                return;
1320
1321        nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1322        netdev_tx_reset_queue(nd_q);
1323}
1324
1325static void nfp_net_tx_timeout(struct net_device *netdev, unsigned int txqueue)
1326{
1327        struct nfp_net *nn = netdev_priv(netdev);
1328
1329        nn_warn(nn, "TX watchdog timeout on ring: %u\n", txqueue);
1330}
1331
1332/* Receive processing
1333 */
1334static unsigned int
1335nfp_net_calc_fl_bufsz(struct nfp_net_dp *dp)
1336{
1337        unsigned int fl_bufsz;
1338
1339        fl_bufsz = NFP_NET_RX_BUF_HEADROOM;
1340        fl_bufsz += dp->rx_dma_off;
1341        if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1342                fl_bufsz += NFP_NET_MAX_PREPEND;
1343        else
1344                fl_bufsz += dp->rx_offset;
1345        fl_bufsz += ETH_HLEN + VLAN_HLEN * 2 + dp->mtu;
1346
1347        fl_bufsz = SKB_DATA_ALIGN(fl_bufsz);
1348        fl_bufsz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1349
1350        return fl_bufsz;
1351}
1352
1353static void
1354nfp_net_free_frag(void *frag, bool xdp)
1355{
1356        if (!xdp)
1357                skb_free_frag(frag);
1358        else
1359                __free_page(virt_to_page(frag));
1360}
1361
1362/**
1363 * nfp_net_rx_alloc_one() - Allocate and map page frag for RX
1364 * @dp:         NFP Net data path struct
1365 * @dma_addr:   Pointer to storage for DMA address (output param)
1366 *
1367 * This function will allcate a new page frag, map it for DMA.
1368 *
1369 * Return: allocated page frag or NULL on failure.
1370 */
1371static void *nfp_net_rx_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
1372{
1373        void *frag;
1374
1375        if (!dp->xdp_prog) {
1376                frag = netdev_alloc_frag(dp->fl_bufsz);
1377        } else {
1378                struct page *page;
1379
1380                page = alloc_page(GFP_KERNEL);
1381                frag = page ? page_address(page) : NULL;
1382        }
1383        if (!frag) {
1384                nn_dp_warn(dp, "Failed to alloc receive page frag\n");
1385                return NULL;
1386        }
1387
1388        *dma_addr = nfp_net_dma_map_rx(dp, frag);
1389        if (dma_mapping_error(dp->dev, *dma_addr)) {
1390                nfp_net_free_frag(frag, dp->xdp_prog);
1391                nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
1392                return NULL;
1393        }
1394
1395        return frag;
1396}
1397
1398static void *nfp_net_napi_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
1399{
1400        void *frag;
1401
1402        if (!dp->xdp_prog) {
1403                frag = napi_alloc_frag(dp->fl_bufsz);
1404                if (unlikely(!frag))
1405                        return NULL;
1406        } else {
1407                struct page *page;
1408
1409                page = dev_alloc_page();
1410                if (unlikely(!page))
1411                        return NULL;
1412                frag = page_address(page);
1413        }
1414
1415        *dma_addr = nfp_net_dma_map_rx(dp, frag);
1416        if (dma_mapping_error(dp->dev, *dma_addr)) {
1417                nfp_net_free_frag(frag, dp->xdp_prog);
1418                nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
1419                return NULL;
1420        }
1421
1422        return frag;
1423}
1424
1425/**
1426 * nfp_net_rx_give_one() - Put mapped skb on the software and hardware rings
1427 * @dp:         NFP Net data path struct
1428 * @rx_ring:    RX ring structure
1429 * @frag:       page fragment buffer
1430 * @dma_addr:   DMA address of skb mapping
1431 */
1432static void nfp_net_rx_give_one(const struct nfp_net_dp *dp,
1433                                struct nfp_net_rx_ring *rx_ring,
1434                                void *frag, dma_addr_t dma_addr)
1435{
1436        unsigned int wr_idx;
1437
1438        wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
1439
1440        nfp_net_dma_sync_dev_rx(dp, dma_addr);
1441
1442        /* Stash SKB and DMA address away */
1443        rx_ring->rxbufs[wr_idx].frag = frag;
1444        rx_ring->rxbufs[wr_idx].dma_addr = dma_addr;
1445
1446        /* Fill freelist descriptor */
1447        rx_ring->rxds[wr_idx].fld.reserved = 0;
1448        rx_ring->rxds[wr_idx].fld.meta_len_dd = 0;
1449        nfp_desc_set_dma_addr(&rx_ring->rxds[wr_idx].fld,
1450                              dma_addr + dp->rx_dma_off);
1451
1452        rx_ring->wr_p++;
1453        if (!(rx_ring->wr_p % NFP_NET_FL_BATCH)) {
1454                /* Update write pointer of the freelist queue. Make
1455                 * sure all writes are flushed before telling the hardware.
1456                 */
1457                wmb();
1458                nfp_qcp_wr_ptr_add(rx_ring->qcp_fl, NFP_NET_FL_BATCH);
1459        }
1460}
1461
1462/**
1463 * nfp_net_rx_ring_reset() - Reflect in SW state of freelist after disable
1464 * @rx_ring:    RX ring structure
1465 *
1466 * Assumes that the device is stopped, must be idempotent.
1467 */
1468static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring)
1469{
1470        unsigned int wr_idx, last_idx;
1471
1472        /* wr_p == rd_p means ring was never fed FL bufs.  RX rings are always
1473         * kept at cnt - 1 FL bufs.
1474         */
1475        if (rx_ring->wr_p == 0 && rx_ring->rd_p == 0)
1476                return;
1477
1478        /* Move the empty entry to the end of the list */
1479        wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
1480        last_idx = rx_ring->cnt - 1;
1481        rx_ring->rxbufs[wr_idx].dma_addr = rx_ring->rxbufs[last_idx].dma_addr;
1482        rx_ring->rxbufs[wr_idx].frag = rx_ring->rxbufs[last_idx].frag;
1483        rx_ring->rxbufs[last_idx].dma_addr = 0;
1484        rx_ring->rxbufs[last_idx].frag = NULL;
1485
1486        memset(rx_ring->rxds, 0, rx_ring->size);
1487        rx_ring->wr_p = 0;
1488        rx_ring->rd_p = 0;
1489}
1490
1491/**
1492 * nfp_net_rx_ring_bufs_free() - Free any buffers currently on the RX ring
1493 * @dp:         NFP Net data path struct
1494 * @rx_ring:    RX ring to remove buffers from
1495 *
1496 * Assumes that the device is stopped and buffers are in [0, ring->cnt - 1)
1497 * entries.  After device is disabled nfp_net_rx_ring_reset() must be called
1498 * to restore required ring geometry.
1499 */
1500static void
1501nfp_net_rx_ring_bufs_free(struct nfp_net_dp *dp,
1502                          struct nfp_net_rx_ring *rx_ring)
1503{
1504        unsigned int i;
1505
1506        for (i = 0; i < rx_ring->cnt - 1; i++) {
1507                /* NULL skb can only happen when initial filling of the ring
1508                 * fails to allocate enough buffers and calls here to free
1509                 * already allocated ones.
1510                 */
1511                if (!rx_ring->rxbufs[i].frag)
1512                        continue;
1513
1514                nfp_net_dma_unmap_rx(dp, rx_ring->rxbufs[i].dma_addr);
1515                nfp_net_free_frag(rx_ring->rxbufs[i].frag, dp->xdp_prog);
1516                rx_ring->rxbufs[i].dma_addr = 0;
1517                rx_ring->rxbufs[i].frag = NULL;
1518        }
1519}
1520
1521/**
1522 * nfp_net_rx_ring_bufs_alloc() - Fill RX ring with buffers (don't give to FW)
1523 * @dp:         NFP Net data path struct
1524 * @rx_ring:    RX ring to remove buffers from
1525 */
1526static int
1527nfp_net_rx_ring_bufs_alloc(struct nfp_net_dp *dp,
1528                           struct nfp_net_rx_ring *rx_ring)
1529{
1530        struct nfp_net_rx_buf *rxbufs;
1531        unsigned int i;
1532
1533        rxbufs = rx_ring->rxbufs;
1534
1535        for (i = 0; i < rx_ring->cnt - 1; i++) {
1536                rxbufs[i].frag = nfp_net_rx_alloc_one(dp, &rxbufs[i].dma_addr);
1537                if (!rxbufs[i].frag) {
1538                        nfp_net_rx_ring_bufs_free(dp, rx_ring);
1539                        return -ENOMEM;
1540                }
1541        }
1542
1543        return 0;
1544}
1545
1546/**
1547 * nfp_net_rx_ring_fill_freelist() - Give buffers from the ring to FW
1548 * @dp:      NFP Net data path struct
1549 * @rx_ring: RX ring to fill
1550 */
1551static void
1552nfp_net_rx_ring_fill_freelist(struct nfp_net_dp *dp,
1553                              struct nfp_net_rx_ring *rx_ring)
1554{
1555        unsigned int i;
1556
1557        for (i = 0; i < rx_ring->cnt - 1; i++)
1558                nfp_net_rx_give_one(dp, rx_ring, rx_ring->rxbufs[i].frag,
1559                                    rx_ring->rxbufs[i].dma_addr);
1560}
1561
1562/**
1563 * nfp_net_rx_csum_has_errors() - group check if rxd has any csum errors
1564 * @flags: RX descriptor flags field in CPU byte order
1565 */
1566static int nfp_net_rx_csum_has_errors(u16 flags)
1567{
1568        u16 csum_all_checked, csum_all_ok;
1569
1570        csum_all_checked = flags & __PCIE_DESC_RX_CSUM_ALL;
1571        csum_all_ok = flags & __PCIE_DESC_RX_CSUM_ALL_OK;
1572
1573        return csum_all_checked != (csum_all_ok << PCIE_DESC_RX_CSUM_OK_SHIFT);
1574}
1575
1576/**
1577 * nfp_net_rx_csum() - set SKB checksum field based on RX descriptor flags
1578 * @dp:  NFP Net data path struct
1579 * @r_vec: per-ring structure
1580 * @rxd: Pointer to RX descriptor
1581 * @meta: Parsed metadata prepend
1582 * @skb: Pointer to SKB
1583 */
1584static void nfp_net_rx_csum(struct nfp_net_dp *dp,
1585                            struct nfp_net_r_vector *r_vec,
1586                            struct nfp_net_rx_desc *rxd,
1587                            struct nfp_meta_parsed *meta, struct sk_buff *skb)
1588{
1589        skb_checksum_none_assert(skb);
1590
1591        if (!(dp->netdev->features & NETIF_F_RXCSUM))
1592                return;
1593
1594        if (meta->csum_type) {
1595                skb->ip_summed = meta->csum_type;
1596                skb->csum = meta->csum;
1597                u64_stats_update_begin(&r_vec->rx_sync);
1598                r_vec->hw_csum_rx_complete++;
1599                u64_stats_update_end(&r_vec->rx_sync);
1600                return;
1601        }
1602
1603        if (nfp_net_rx_csum_has_errors(le16_to_cpu(rxd->rxd.flags))) {
1604                u64_stats_update_begin(&r_vec->rx_sync);
1605                r_vec->hw_csum_rx_error++;
1606                u64_stats_update_end(&r_vec->rx_sync);
1607                return;
1608        }
1609
1610        /* Assume that the firmware will never report inner CSUM_OK unless outer
1611         * L4 headers were successfully parsed. FW will always report zero UDP
1612         * checksum as CSUM_OK.
1613         */
1614        if (rxd->rxd.flags & PCIE_DESC_RX_TCP_CSUM_OK ||
1615            rxd->rxd.flags & PCIE_DESC_RX_UDP_CSUM_OK) {
1616                __skb_incr_checksum_unnecessary(skb);
1617                u64_stats_update_begin(&r_vec->rx_sync);
1618                r_vec->hw_csum_rx_ok++;
1619                u64_stats_update_end(&r_vec->rx_sync);
1620        }
1621
1622        if (rxd->rxd.flags & PCIE_DESC_RX_I_TCP_CSUM_OK ||
1623            rxd->rxd.flags & PCIE_DESC_RX_I_UDP_CSUM_OK) {
1624                __skb_incr_checksum_unnecessary(skb);
1625                u64_stats_update_begin(&r_vec->rx_sync);
1626                r_vec->hw_csum_rx_inner_ok++;
1627                u64_stats_update_end(&r_vec->rx_sync);
1628        }
1629}
1630
1631static void
1632nfp_net_set_hash(struct net_device *netdev, struct nfp_meta_parsed *meta,
1633                 unsigned int type, __be32 *hash)
1634{
1635        if (!(netdev->features & NETIF_F_RXHASH))
1636                return;
1637
1638        switch (type) {
1639        case NFP_NET_RSS_IPV4:
1640        case NFP_NET_RSS_IPV6:
1641        case NFP_NET_RSS_IPV6_EX:
1642                meta->hash_type = PKT_HASH_TYPE_L3;
1643                break;
1644        default:
1645                meta->hash_type = PKT_HASH_TYPE_L4;
1646                break;
1647        }
1648
1649        meta->hash = get_unaligned_be32(hash);
1650}
1651
1652static void
1653nfp_net_set_hash_desc(struct net_device *netdev, struct nfp_meta_parsed *meta,
1654                      void *data, struct nfp_net_rx_desc *rxd)
1655{
1656        struct nfp_net_rx_hash *rx_hash = data;
1657
1658        if (!(rxd->rxd.flags & PCIE_DESC_RX_RSS))
1659                return;
1660
1661        nfp_net_set_hash(netdev, meta, get_unaligned_be32(&rx_hash->hash_type),
1662                         &rx_hash->hash);
1663}
1664
1665static bool
1666nfp_net_parse_meta(struct net_device *netdev, struct nfp_meta_parsed *meta,
1667                   void *data, void *pkt, unsigned int pkt_len, int meta_len)
1668{
1669        u32 meta_info;
1670
1671        meta_info = get_unaligned_be32(data);
1672        data += 4;
1673
1674        while (meta_info) {
1675                switch (meta_info & NFP_NET_META_FIELD_MASK) {
1676                case NFP_NET_META_HASH:
1677                        meta_info >>= NFP_NET_META_FIELD_SIZE;
1678                        nfp_net_set_hash(netdev, meta,
1679                                         meta_info & NFP_NET_META_FIELD_MASK,
1680                                         (__be32 *)data);
1681                        data += 4;
1682                        break;
1683                case NFP_NET_META_MARK:
1684                        meta->mark = get_unaligned_be32(data);
1685                        data += 4;
1686                        break;
1687                case NFP_NET_META_PORTID:
1688                        meta->portid = get_unaligned_be32(data);
1689                        data += 4;
1690                        break;
1691                case NFP_NET_META_CSUM:
1692                        meta->csum_type = CHECKSUM_COMPLETE;
1693                        meta->csum =
1694                                (__force __wsum)__get_unaligned_cpu32(data);
1695                        data += 4;
1696                        break;
1697                case NFP_NET_META_RESYNC_INFO:
1698                        if (nfp_net_tls_rx_resync_req(netdev, data, pkt,
1699                                                      pkt_len))
1700                                return NULL;
1701                        data += sizeof(struct nfp_net_tls_resync_req);
1702                        break;
1703                default:
1704                        return true;
1705                }
1706
1707                meta_info >>= NFP_NET_META_FIELD_SIZE;
1708        }
1709
1710        return data != pkt;
1711}
1712
1713static void
1714nfp_net_rx_drop(const struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec,
1715                struct nfp_net_rx_ring *rx_ring, struct nfp_net_rx_buf *rxbuf,
1716                struct sk_buff *skb)
1717{
1718        u64_stats_update_begin(&r_vec->rx_sync);
1719        r_vec->rx_drops++;
1720        /* If we have both skb and rxbuf the replacement buffer allocation
1721         * must have failed, count this as an alloc failure.
1722         */
1723        if (skb && rxbuf)
1724                r_vec->rx_replace_buf_alloc_fail++;
1725        u64_stats_update_end(&r_vec->rx_sync);
1726
1727        /* skb is build based on the frag, free_skb() would free the frag
1728         * so to be able to reuse it we need an extra ref.
1729         */
1730        if (skb && rxbuf && skb->head == rxbuf->frag)
1731                page_ref_inc(virt_to_head_page(rxbuf->frag));
1732        if (rxbuf)
1733                nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag, rxbuf->dma_addr);
1734        if (skb)
1735                dev_kfree_skb_any(skb);
1736}
1737
1738static bool
1739nfp_net_tx_xdp_buf(struct nfp_net_dp *dp, struct nfp_net_rx_ring *rx_ring,
1740                   struct nfp_net_tx_ring *tx_ring,
1741                   struct nfp_net_rx_buf *rxbuf, unsigned int dma_off,
1742                   unsigned int pkt_len, bool *completed)
1743{
1744        unsigned int dma_map_sz = dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA;
1745        struct nfp_net_tx_buf *txbuf;
1746        struct nfp_net_tx_desc *txd;
1747        int wr_idx;
1748
1749        /* Reject if xdp_adjust_tail grow packet beyond DMA area */
1750        if (pkt_len + dma_off > dma_map_sz)
1751                return false;
1752
1753        if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1754                if (!*completed) {
1755                        nfp_net_xdp_complete(tx_ring);
1756                        *completed = true;
1757                }
1758
1759                if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1760                        nfp_net_rx_drop(dp, rx_ring->r_vec, rx_ring, rxbuf,
1761                                        NULL);
1762                        return false;
1763                }
1764        }
1765
1766        wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
1767
1768        /* Stash the soft descriptor of the head then initialize it */
1769        txbuf = &tx_ring->txbufs[wr_idx];
1770
1771        nfp_net_rx_give_one(dp, rx_ring, txbuf->frag, txbuf->dma_addr);
1772
1773        txbuf->frag = rxbuf->frag;
1774        txbuf->dma_addr = rxbuf->dma_addr;
1775        txbuf->fidx = -1;
1776        txbuf->pkt_cnt = 1;
1777        txbuf->real_len = pkt_len;
1778
1779        dma_sync_single_for_device(dp->dev, rxbuf->dma_addr + dma_off,
1780                                   pkt_len, DMA_BIDIRECTIONAL);
1781
1782        /* Build TX descriptor */
1783        txd = &tx_ring->txds[wr_idx];
1784        txd->offset_eop = PCIE_DESC_TX_EOP;
1785        txd->dma_len = cpu_to_le16(pkt_len);
1786        nfp_desc_set_dma_addr(txd, rxbuf->dma_addr + dma_off);
1787        txd->data_len = cpu_to_le16(pkt_len);
1788
1789        txd->flags = 0;
1790        txd->mss = 0;
1791        txd->lso_hdrlen = 0;
1792
1793        tx_ring->wr_p++;
1794        tx_ring->wr_ptr_add++;
1795        return true;
1796}
1797
1798/**
1799 * nfp_net_rx() - receive up to @budget packets on @rx_ring
1800 * @rx_ring:   RX ring to receive from
1801 * @budget:    NAPI budget
1802 *
1803 * Note, this function is separated out from the napi poll function to
1804 * more cleanly separate packet receive code from other bookkeeping
1805 * functions performed in the napi poll function.
1806 *
1807 * Return: Number of packets received.
1808 */
1809static int nfp_net_rx(struct nfp_net_rx_ring *rx_ring, int budget)
1810{
1811        struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
1812        struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
1813        struct nfp_net_tx_ring *tx_ring;
1814        struct bpf_prog *xdp_prog;
1815        bool xdp_tx_cmpl = false;
1816        unsigned int true_bufsz;
1817        struct sk_buff *skb;
1818        int pkts_polled = 0;
1819        struct xdp_buff xdp;
1820        int idx;
1821
1822        xdp_prog = READ_ONCE(dp->xdp_prog);
1823        true_bufsz = xdp_prog ? PAGE_SIZE : dp->fl_bufsz;
1824        xdp_init_buff(&xdp, PAGE_SIZE - NFP_NET_RX_BUF_HEADROOM,
1825                      &rx_ring->xdp_rxq);
1826        tx_ring = r_vec->xdp_ring;
1827
1828        while (pkts_polled < budget) {
1829                unsigned int meta_len, data_len, meta_off, pkt_len, pkt_off;
1830                struct nfp_net_rx_buf *rxbuf;
1831                struct nfp_net_rx_desc *rxd;
1832                struct nfp_meta_parsed meta;
1833                bool redir_egress = false;
1834                struct net_device *netdev;
1835                dma_addr_t new_dma_addr;
1836                u32 meta_len_xdp = 0;
1837                void *new_frag;
1838
1839                idx = D_IDX(rx_ring, rx_ring->rd_p);
1840
1841                rxd = &rx_ring->rxds[idx];
1842                if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
1843                        break;
1844
1845                /* Memory barrier to ensure that we won't do other reads
1846                 * before the DD bit.
1847                 */
1848                dma_rmb();
1849
1850                memset(&meta, 0, sizeof(meta));
1851
1852                rx_ring->rd_p++;
1853                pkts_polled++;
1854
1855                rxbuf = &rx_ring->rxbufs[idx];
1856                /*         < meta_len >
1857                 *  <-- [rx_offset] -->
1858                 *  ---------------------------------------------------------
1859                 * | [XX] |  metadata  |             packet           | XXXX |
1860                 *  ---------------------------------------------------------
1861                 *         <---------------- data_len --------------->
1862                 *
1863                 * The rx_offset is fixed for all packets, the meta_len can vary
1864                 * on a packet by packet basis. If rx_offset is set to zero
1865                 * (_RX_OFFSET_DYNAMIC) metadata starts at the beginning of the
1866                 * buffer and is immediately followed by the packet (no [XX]).
1867                 */
1868                meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
1869                data_len = le16_to_cpu(rxd->rxd.data_len);
1870                pkt_len = data_len - meta_len;
1871
1872                pkt_off = NFP_NET_RX_BUF_HEADROOM + dp->rx_dma_off;
1873                if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1874                        pkt_off += meta_len;
1875                else
1876                        pkt_off += dp->rx_offset;
1877                meta_off = pkt_off - meta_len;
1878
1879                /* Stats update */
1880                u64_stats_update_begin(&r_vec->rx_sync);
1881                r_vec->rx_pkts++;
1882                r_vec->rx_bytes += pkt_len;
1883                u64_stats_update_end(&r_vec->rx_sync);
1884
1885                if (unlikely(meta_len > NFP_NET_MAX_PREPEND ||
1886                             (dp->rx_offset && meta_len > dp->rx_offset))) {
1887                        nn_dp_warn(dp, "oversized RX packet metadata %u\n",
1888                                   meta_len);
1889                        nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
1890                        continue;
1891                }
1892
1893                nfp_net_dma_sync_cpu_rx(dp, rxbuf->dma_addr + meta_off,
1894                                        data_len);
1895
1896                if (!dp->chained_metadata_format) {
1897                        nfp_net_set_hash_desc(dp->netdev, &meta,
1898                                              rxbuf->frag + meta_off, rxd);
1899                } else if (meta_len) {
1900                        if (unlikely(nfp_net_parse_meta(dp->netdev, &meta,
1901                                                        rxbuf->frag + meta_off,
1902                                                        rxbuf->frag + pkt_off,
1903                                                        pkt_len, meta_len))) {
1904                                nn_dp_warn(dp, "invalid RX packet metadata\n");
1905                                nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf,
1906                                                NULL);
1907                                continue;
1908                        }
1909                }
1910
1911                if (xdp_prog && !meta.portid) {
1912                        void *orig_data = rxbuf->frag + pkt_off;
1913                        unsigned int dma_off;
1914                        int act;
1915
1916                        xdp_prepare_buff(&xdp,
1917                                         rxbuf->frag + NFP_NET_RX_BUF_HEADROOM,
1918                                         pkt_off - NFP_NET_RX_BUF_HEADROOM,
1919                                         pkt_len, true);
1920
1921                        act = bpf_prog_run_xdp(xdp_prog, &xdp);
1922
1923                        pkt_len = xdp.data_end - xdp.data;
1924                        pkt_off += xdp.data - orig_data;
1925
1926                        switch (act) {
1927                        case XDP_PASS:
1928                                meta_len_xdp = xdp.data - xdp.data_meta;
1929                                break;
1930                        case XDP_TX:
1931                                dma_off = pkt_off - NFP_NET_RX_BUF_HEADROOM;
1932                                if (unlikely(!nfp_net_tx_xdp_buf(dp, rx_ring,
1933                                                                 tx_ring, rxbuf,
1934                                                                 dma_off,
1935                                                                 pkt_len,
1936                                                                 &xdp_tx_cmpl)))
1937                                        trace_xdp_exception(dp->netdev,
1938                                                            xdp_prog, act);
1939                                continue;
1940                        default:
1941                                bpf_warn_invalid_xdp_action(act);
1942                                fallthrough;
1943                        case XDP_ABORTED:
1944                                trace_xdp_exception(dp->netdev, xdp_prog, act);
1945                                fallthrough;
1946                        case XDP_DROP:
1947                                nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag,
1948                                                    rxbuf->dma_addr);
1949                                continue;
1950                        }
1951                }
1952
1953                if (likely(!meta.portid)) {
1954                        netdev = dp->netdev;
1955                } else if (meta.portid == NFP_META_PORT_ID_CTRL) {
1956                        struct nfp_net *nn = netdev_priv(dp->netdev);
1957
1958                        nfp_app_ctrl_rx_raw(nn->app, rxbuf->frag + pkt_off,
1959                                            pkt_len);
1960                        nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag,
1961                                            rxbuf->dma_addr);
1962                        continue;
1963                } else {
1964                        struct nfp_net *nn;
1965
1966                        nn = netdev_priv(dp->netdev);
1967                        netdev = nfp_app_dev_get(nn->app, meta.portid,
1968                                                 &redir_egress);
1969                        if (unlikely(!netdev)) {
1970                                nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf,
1971                                                NULL);
1972                                continue;
1973                        }
1974
1975                        if (nfp_netdev_is_nfp_repr(netdev))
1976                                nfp_repr_inc_rx_stats(netdev, pkt_len);
1977                }
1978
1979                skb = build_skb(rxbuf->frag, true_bufsz);
1980                if (unlikely(!skb)) {
1981                        nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
1982                        continue;
1983                }
1984                new_frag = nfp_net_napi_alloc_one(dp, &new_dma_addr);
1985                if (unlikely(!new_frag)) {
1986                        nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, skb);
1987                        continue;
1988                }
1989
1990                nfp_net_dma_unmap_rx(dp, rxbuf->dma_addr);
1991
1992                nfp_net_rx_give_one(dp, rx_ring, new_frag, new_dma_addr);
1993
1994                skb_reserve(skb, pkt_off);
1995                skb_put(skb, pkt_len);
1996
1997                skb->mark = meta.mark;
1998                skb_set_hash(skb, meta.hash, meta.hash_type);
1999
2000                skb_record_rx_queue(skb, rx_ring->idx);
2001                skb->protocol = eth_type_trans(skb, netdev);
2002
2003                nfp_net_rx_csum(dp, r_vec, rxd, &meta, skb);
2004
2005#ifdef CONFIG_TLS_DEVICE
2006                if (rxd->rxd.flags & PCIE_DESC_RX_DECRYPTED) {
2007                        skb->decrypted = true;
2008                        u64_stats_update_begin(&r_vec->rx_sync);
2009                        r_vec->hw_tls_rx++;
2010                        u64_stats_update_end(&r_vec->rx_sync);
2011                }
2012#endif
2013
2014                if (rxd->rxd.flags & PCIE_DESC_RX_VLAN)
2015                        __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2016                                               le16_to_cpu(rxd->rxd.vlan));
2017                if (meta_len_xdp)
2018                        skb_metadata_set(skb, meta_len_xdp);
2019
2020                if (likely(!redir_egress)) {
2021                        napi_gro_receive(&rx_ring->r_vec->napi, skb);
2022                } else {
2023                        skb->dev = netdev;
2024                        skb_reset_network_header(skb);
2025                        __skb_push(skb, ETH_HLEN);
2026                        dev_queue_xmit(skb);
2027                }
2028        }
2029
2030        if (xdp_prog) {
2031                if (tx_ring->wr_ptr_add)
2032                        nfp_net_tx_xmit_more_flush(tx_ring);
2033                else if (unlikely(tx_ring->wr_p != tx_ring->rd_p) &&
2034                         !xdp_tx_cmpl)
2035                        if (!nfp_net_xdp_complete(tx_ring))
2036                                pkts_polled = budget;
2037        }
2038
2039        return pkts_polled;
2040}
2041
2042/**
2043 * nfp_net_poll() - napi poll function
2044 * @napi:    NAPI structure
2045 * @budget:  NAPI budget
2046 *
2047 * Return: number of packets polled.
2048 */
2049static int nfp_net_poll(struct napi_struct *napi, int budget)
2050{
2051        struct nfp_net_r_vector *r_vec =
2052                container_of(napi, struct nfp_net_r_vector, napi);
2053        unsigned int pkts_polled = 0;
2054
2055        if (r_vec->tx_ring)
2056                nfp_net_tx_complete(r_vec->tx_ring, budget);
2057        if (r_vec->rx_ring)
2058                pkts_polled = nfp_net_rx(r_vec->rx_ring, budget);
2059
2060        if (pkts_polled < budget)
2061                if (napi_complete_done(napi, pkts_polled))
2062                        nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_entry);
2063
2064        return pkts_polled;
2065}
2066
2067/* Control device data path
2068 */
2069
2070static bool
2071nfp_ctrl_tx_one(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
2072                struct sk_buff *skb, bool old)
2073{
2074        unsigned int real_len = skb->len, meta_len = 0;
2075        struct nfp_net_tx_ring *tx_ring;
2076        struct nfp_net_tx_buf *txbuf;
2077        struct nfp_net_tx_desc *txd;
2078        struct nfp_net_dp *dp;
2079        dma_addr_t dma_addr;
2080        int wr_idx;
2081
2082        dp = &r_vec->nfp_net->dp;
2083        tx_ring = r_vec->tx_ring;
2084
2085        if (WARN_ON_ONCE(skb_shinfo(skb)->nr_frags)) {
2086                nn_dp_warn(dp, "Driver's CTRL TX does not implement gather\n");
2087                goto err_free;
2088        }
2089
2090        if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
2091                u64_stats_update_begin(&r_vec->tx_sync);
2092                r_vec->tx_busy++;
2093                u64_stats_update_end(&r_vec->tx_sync);
2094                if (!old)
2095                        __skb_queue_tail(&r_vec->queue, skb);
2096                else
2097                        __skb_queue_head(&r_vec->queue, skb);
2098                return true;
2099        }
2100
2101        if (nfp_app_ctrl_has_meta(nn->app)) {
2102                if (unlikely(skb_headroom(skb) < 8)) {
2103                        nn_dp_warn(dp, "CTRL TX on skb without headroom\n");
2104                        goto err_free;
2105                }
2106                meta_len = 8;
2107                put_unaligned_be32(NFP_META_PORT_ID_CTRL, skb_push(skb, 4));
2108                put_unaligned_be32(NFP_NET_META_PORTID, skb_push(skb, 4));
2109        }
2110
2111        /* Start with the head skbuf */
2112        dma_addr = dma_map_single(dp->dev, skb->data, skb_headlen(skb),
2113                                  DMA_TO_DEVICE);
2114        if (dma_mapping_error(dp->dev, dma_addr))
2115                goto err_dma_warn;
2116
2117        wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
2118
2119        /* Stash the soft descriptor of the head then initialize it */
2120        txbuf = &tx_ring->txbufs[wr_idx];
2121        txbuf->skb = skb;
2122        txbuf->dma_addr = dma_addr;
2123        txbuf->fidx = -1;
2124        txbuf->pkt_cnt = 1;
2125        txbuf->real_len = real_len;
2126
2127        /* Build TX descriptor */
2128        txd = &tx_ring->txds[wr_idx];
2129        txd->offset_eop = meta_len | PCIE_DESC_TX_EOP;
2130        txd->dma_len = cpu_to_le16(skb_headlen(skb));
2131        nfp_desc_set_dma_addr(txd, dma_addr);
2132        txd->data_len = cpu_to_le16(skb->len);
2133
2134        txd->flags = 0;
2135        txd->mss = 0;
2136        txd->lso_hdrlen = 0;
2137
2138        tx_ring->wr_p++;
2139        tx_ring->wr_ptr_add++;
2140        nfp_net_tx_xmit_more_flush(tx_ring);
2141
2142        return false;
2143
2144err_dma_warn:
2145        nn_dp_warn(dp, "Failed to DMA map TX CTRL buffer\n");
2146err_free:
2147        u64_stats_update_begin(&r_vec->tx_sync);
2148        r_vec->tx_errors++;
2149        u64_stats_update_end(&r_vec->tx_sync);
2150        dev_kfree_skb_any(skb);
2151        return false;
2152}
2153
2154bool __nfp_ctrl_tx(struct nfp_net *nn, struct sk_buff *skb)
2155{
2156        struct nfp_net_r_vector *r_vec = &nn->r_vecs[0];
2157
2158        return nfp_ctrl_tx_one(nn, r_vec, skb, false);
2159}
2160
2161bool nfp_ctrl_tx(struct nfp_net *nn, struct sk_buff *skb)
2162{
2163        struct nfp_net_r_vector *r_vec = &nn->r_vecs[0];
2164        bool ret;
2165
2166        spin_lock_bh(&r_vec->lock);
2167        ret = nfp_ctrl_tx_one(nn, r_vec, skb, false);
2168        spin_unlock_bh(&r_vec->lock);
2169
2170        return ret;
2171}
2172
2173static void __nfp_ctrl_tx_queued(struct nfp_net_r_vector *r_vec)
2174{
2175        struct sk_buff *skb;
2176
2177        while ((skb = __skb_dequeue(&r_vec->queue)))
2178                if (nfp_ctrl_tx_one(r_vec->nfp_net, r_vec, skb, true))
2179                        return;
2180}
2181
2182static bool
2183nfp_ctrl_meta_ok(struct nfp_net *nn, void *data, unsigned int meta_len)
2184{
2185        u32 meta_type, meta_tag;
2186
2187        if (!nfp_app_ctrl_has_meta(nn->app))
2188                return !meta_len;
2189
2190        if (meta_len != 8)
2191                return false;
2192
2193        meta_type = get_unaligned_be32(data);
2194        meta_tag = get_unaligned_be32(data + 4);
2195
2196        return (meta_type == NFP_NET_META_PORTID &&
2197                meta_tag == NFP_META_PORT_ID_CTRL);
2198}
2199
2200static bool
2201nfp_ctrl_rx_one(struct nfp_net *nn, struct nfp_net_dp *dp,
2202                struct nfp_net_r_vector *r_vec, struct nfp_net_rx_ring *rx_ring)
2203{
2204        unsigned int meta_len, data_len, meta_off, pkt_len, pkt_off;
2205        struct nfp_net_rx_buf *rxbuf;
2206        struct nfp_net_rx_desc *rxd;
2207        dma_addr_t new_dma_addr;
2208        struct sk_buff *skb;
2209        void *new_frag;
2210        int idx;
2211
2212        idx = D_IDX(rx_ring, rx_ring->rd_p);
2213
2214        rxd = &rx_ring->rxds[idx];
2215        if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
2216                return false;
2217
2218        /* Memory barrier to ensure that we won't do other reads
2219         * before the DD bit.
2220         */
2221        dma_rmb();
2222
2223        rx_ring->rd_p++;
2224
2225        rxbuf = &rx_ring->rxbufs[idx];
2226        meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
2227        data_len = le16_to_cpu(rxd->rxd.data_len);
2228        pkt_len = data_len - meta_len;
2229
2230        pkt_off = NFP_NET_RX_BUF_HEADROOM + dp->rx_dma_off;
2231        if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
2232                pkt_off += meta_len;
2233        else
2234                pkt_off += dp->rx_offset;
2235        meta_off = pkt_off - meta_len;
2236
2237        /* Stats update */
2238        u64_stats_update_begin(&r_vec->rx_sync);
2239        r_vec->rx_pkts++;
2240        r_vec->rx_bytes += pkt_len;
2241        u64_stats_update_end(&r_vec->rx_sync);
2242
2243        nfp_net_dma_sync_cpu_rx(dp, rxbuf->dma_addr + meta_off, data_len);
2244
2245        if (unlikely(!nfp_ctrl_meta_ok(nn, rxbuf->frag + meta_off, meta_len))) {
2246                nn_dp_warn(dp, "incorrect metadata for ctrl packet (%d)\n",
2247                           meta_len);
2248                nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
2249                return true;
2250        }
2251
2252        skb = build_skb(rxbuf->frag, dp->fl_bufsz);
2253        if (unlikely(!skb)) {
2254                nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
2255                return true;
2256        }
2257        new_frag = nfp_net_napi_alloc_one(dp, &new_dma_addr);
2258        if (unlikely(!new_frag)) {
2259                nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, skb);
2260                return true;
2261        }
2262
2263        nfp_net_dma_unmap_rx(dp, rxbuf->dma_addr);
2264
2265        nfp_net_rx_give_one(dp, rx_ring, new_frag, new_dma_addr);
2266
2267        skb_reserve(skb, pkt_off);
2268        skb_put(skb, pkt_len);
2269
2270        nfp_app_ctrl_rx(nn->app, skb);
2271
2272        return true;
2273}
2274
2275static bool nfp_ctrl_rx(struct nfp_net_r_vector *r_vec)
2276{
2277        struct nfp_net_rx_ring *rx_ring = r_vec->rx_ring;
2278        struct nfp_net *nn = r_vec->nfp_net;
2279        struct nfp_net_dp *dp = &nn->dp;
2280        unsigned int budget = 512;
2281
2282        while (nfp_ctrl_rx_one(nn, dp, r_vec, rx_ring) && budget--)
2283                continue;
2284
2285        return budget;
2286}
2287
2288static void nfp_ctrl_poll(struct tasklet_struct *t)
2289{
2290        struct nfp_net_r_vector *r_vec = from_tasklet(r_vec, t, tasklet);
2291
2292        spin_lock(&r_vec->lock);
2293        nfp_net_tx_complete(r_vec->tx_ring, 0);
2294        __nfp_ctrl_tx_queued(r_vec);
2295        spin_unlock(&r_vec->lock);
2296
2297        if (nfp_ctrl_rx(r_vec)) {
2298                nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_entry);
2299        } else {
2300                tasklet_schedule(&r_vec->tasklet);
2301                nn_dp_warn(&r_vec->nfp_net->dp,
2302                           "control message budget exceeded!\n");
2303        }
2304}
2305
2306/* Setup and Configuration
2307 */
2308
2309/**
2310 * nfp_net_vecs_init() - Assign IRQs and setup rvecs.
2311 * @nn:         NFP Network structure
2312 */
2313static void nfp_net_vecs_init(struct nfp_net *nn)
2314{
2315        struct nfp_net_r_vector *r_vec;
2316        int r;
2317
2318        nn->lsc_handler = nfp_net_irq_lsc;
2319        nn->exn_handler = nfp_net_irq_exn;
2320
2321        for (r = 0; r < nn->max_r_vecs; r++) {
2322                struct msix_entry *entry;
2323
2324                entry = &nn->irq_entries[NFP_NET_NON_Q_VECTORS + r];
2325
2326                r_vec = &nn->r_vecs[r];
2327                r_vec->nfp_net = nn;
2328                r_vec->irq_entry = entry->entry;
2329                r_vec->irq_vector = entry->vector;
2330
2331                if (nn->dp.netdev) {
2332                        r_vec->handler = nfp_net_irq_rxtx;
2333                } else {
2334                        r_vec->handler = nfp_ctrl_irq_rxtx;
2335
2336                        __skb_queue_head_init(&r_vec->queue);
2337                        spin_lock_init(&r_vec->lock);
2338                        tasklet_setup(&r_vec->tasklet, nfp_ctrl_poll);
2339                        tasklet_disable(&r_vec->tasklet);
2340                }
2341
2342                cpumask_set_cpu(r, &r_vec->affinity_mask);
2343        }
2344}
2345
2346/**
2347 * nfp_net_tx_ring_free() - Free resources allocated to a TX ring
2348 * @tx_ring:   TX ring to free
2349 */
2350static void nfp_net_tx_ring_free(struct nfp_net_tx_ring *tx_ring)
2351{
2352        struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
2353        struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
2354
2355        kvfree(tx_ring->txbufs);
2356
2357        if (tx_ring->txds)
2358                dma_free_coherent(dp->dev, tx_ring->size,
2359                                  tx_ring->txds, tx_ring->dma);
2360
2361        tx_ring->cnt = 0;
2362        tx_ring->txbufs = NULL;
2363        tx_ring->txds = NULL;
2364        tx_ring->dma = 0;
2365        tx_ring->size = 0;
2366}
2367
2368/**
2369 * nfp_net_tx_ring_alloc() - Allocate resource for a TX ring
2370 * @dp:        NFP Net data path struct
2371 * @tx_ring:   TX Ring structure to allocate
2372 *
2373 * Return: 0 on success, negative errno otherwise.
2374 */
2375static int
2376nfp_net_tx_ring_alloc(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
2377{
2378        struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
2379
2380        tx_ring->cnt = dp->txd_cnt;
2381
2382        tx_ring->size = array_size(tx_ring->cnt, sizeof(*tx_ring->txds));
2383        tx_ring->txds = dma_alloc_coherent(dp->dev, tx_ring->size,
2384                                           &tx_ring->dma,
2385                                           GFP_KERNEL | __GFP_NOWARN);
2386        if (!tx_ring->txds) {
2387                netdev_warn(dp->netdev, "failed to allocate TX descriptor ring memory, requested descriptor count: %d, consider lowering descriptor count\n",
2388                            tx_ring->cnt);
2389                goto err_alloc;
2390        }
2391
2392        tx_ring->txbufs = kvcalloc(tx_ring->cnt, sizeof(*tx_ring->txbufs),
2393                                   GFP_KERNEL);
2394        if (!tx_ring->txbufs)
2395                goto err_alloc;
2396
2397        if (!tx_ring->is_xdp && dp->netdev)
2398                netif_set_xps_queue(dp->netdev, &r_vec->affinity_mask,
2399                                    tx_ring->idx);
2400
2401        return 0;
2402
2403err_alloc:
2404        nfp_net_tx_ring_free(tx_ring);
2405        return -ENOMEM;
2406}
2407
2408static void
2409nfp_net_tx_ring_bufs_free(struct nfp_net_dp *dp,
2410                          struct nfp_net_tx_ring *tx_ring)
2411{
2412        unsigned int i;
2413
2414        if (!tx_ring->is_xdp)
2415                return;
2416
2417        for (i = 0; i < tx_ring->cnt; i++) {
2418                if (!tx_ring->txbufs[i].frag)
2419                        return;
2420
2421                nfp_net_dma_unmap_rx(dp, tx_ring->txbufs[i].dma_addr);
2422                __free_page(virt_to_page(tx_ring->txbufs[i].frag));
2423        }
2424}
2425
2426static int
2427nfp_net_tx_ring_bufs_alloc(struct nfp_net_dp *dp,
2428                           struct nfp_net_tx_ring *tx_ring)
2429{
2430        struct nfp_net_tx_buf *txbufs = tx_ring->txbufs;
2431        unsigned int i;
2432
2433        if (!tx_ring->is_xdp)
2434                return 0;
2435
2436        for (i = 0; i < tx_ring->cnt; i++) {
2437                txbufs[i].frag = nfp_net_rx_alloc_one(dp, &txbufs[i].dma_addr);
2438                if (!txbufs[i].frag) {
2439                        nfp_net_tx_ring_bufs_free(dp, tx_ring);
2440                        return -ENOMEM;
2441                }
2442        }
2443
2444        return 0;
2445}
2446
2447static int nfp_net_tx_rings_prepare(struct nfp_net *nn, struct nfp_net_dp *dp)
2448{
2449        unsigned int r;
2450
2451        dp->tx_rings = kcalloc(dp->num_tx_rings, sizeof(*dp->tx_rings),
2452                               GFP_KERNEL);
2453        if (!dp->tx_rings)
2454                return -ENOMEM;
2455
2456        for (r = 0; r < dp->num_tx_rings; r++) {
2457                int bias = 0;
2458
2459                if (r >= dp->num_stack_tx_rings)
2460                        bias = dp->num_stack_tx_rings;
2461
2462                nfp_net_tx_ring_init(&dp->tx_rings[r], &nn->r_vecs[r - bias],
2463                                     r, bias);
2464
2465                if (nfp_net_tx_ring_alloc(dp, &dp->tx_rings[r]))
2466                        goto err_free_prev;
2467
2468                if (nfp_net_tx_ring_bufs_alloc(dp, &dp->tx_rings[r]))
2469                        goto err_free_ring;
2470        }
2471
2472        return 0;
2473
2474err_free_prev:
2475        while (r--) {
2476                nfp_net_tx_ring_bufs_free(dp, &dp->tx_rings[r]);
2477err_free_ring:
2478                nfp_net_tx_ring_free(&dp->tx_rings[r]);
2479        }
2480        kfree(dp->tx_rings);
2481        return -ENOMEM;
2482}
2483
2484static void nfp_net_tx_rings_free(struct nfp_net_dp *dp)
2485{
2486        unsigned int r;
2487
2488        for (r = 0; r < dp->num_tx_rings; r++) {
2489                nfp_net_tx_ring_bufs_free(dp, &dp->tx_rings[r]);
2490                nfp_net_tx_ring_free(&dp->tx_rings[r]);
2491        }
2492
2493        kfree(dp->tx_rings);
2494}
2495
2496/**
2497 * nfp_net_rx_ring_free() - Free resources allocated to a RX ring
2498 * @rx_ring:  RX ring to free
2499 */
2500static void nfp_net_rx_ring_free(struct nfp_net_rx_ring *rx_ring)
2501{
2502        struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
2503        struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
2504
2505        if (dp->netdev)
2506                xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
2507        kvfree(rx_ring->rxbufs);
2508
2509        if (rx_ring->rxds)
2510                dma_free_coherent(dp->dev, rx_ring->size,
2511                                  rx_ring->rxds, rx_ring->dma);
2512
2513        rx_ring->cnt = 0;
2514        rx_ring->rxbufs = NULL;
2515        rx_ring->rxds = NULL;
2516        rx_ring->dma = 0;
2517        rx_ring->size = 0;
2518}
2519
2520/**
2521 * nfp_net_rx_ring_alloc() - Allocate resource for a RX ring
2522 * @dp:       NFP Net data path struct
2523 * @rx_ring:  RX ring to allocate
2524 *
2525 * Return: 0 on success, negative errno otherwise.
2526 */
2527static int
2528nfp_net_rx_ring_alloc(struct nfp_net_dp *dp, struct nfp_net_rx_ring *rx_ring)
2529{
2530        int err;
2531
2532        if (dp->netdev) {
2533                err = xdp_rxq_info_reg(&rx_ring->xdp_rxq, dp->netdev,
2534                                       rx_ring->idx, rx_ring->r_vec->napi.napi_id);
2535                if (err < 0)
2536                        return err;
2537        }
2538
2539        rx_ring->cnt = dp->rxd_cnt;
2540        rx_ring->size = array_size(rx_ring->cnt, sizeof(*rx_ring->rxds));
2541        rx_ring->rxds = dma_alloc_coherent(dp->dev, rx_ring->size,
2542                                           &rx_ring->dma,
2543                                           GFP_KERNEL | __GFP_NOWARN);
2544        if (!rx_ring->rxds) {
2545                netdev_warn(dp->netdev, "failed to allocate RX descriptor ring memory, requested descriptor count: %d, consider lowering descriptor count\n",
2546                            rx_ring->cnt);
2547                goto err_alloc;
2548        }
2549
2550        rx_ring->rxbufs = kvcalloc(rx_ring->cnt, sizeof(*rx_ring->rxbufs),
2551                                   GFP_KERNEL);
2552        if (!rx_ring->rxbufs)
2553                goto err_alloc;
2554
2555        return 0;
2556
2557err_alloc:
2558        nfp_net_rx_ring_free(rx_ring);
2559        return -ENOMEM;
2560}
2561
2562static int nfp_net_rx_rings_prepare(struct nfp_net *nn, struct nfp_net_dp *dp)
2563{
2564        unsigned int r;
2565
2566        dp->rx_rings = kcalloc(dp->num_rx_rings, sizeof(*dp->rx_rings),
2567                               GFP_KERNEL);
2568        if (!dp->rx_rings)
2569                return -ENOMEM;
2570
2571        for (r = 0; r < dp->num_rx_rings; r++) {
2572                nfp_net_rx_ring_init(&dp->rx_rings[r], &nn->r_vecs[r], r);
2573
2574                if (nfp_net_rx_ring_alloc(dp, &dp->rx_rings[r]))
2575                        goto err_free_prev;
2576
2577                if (nfp_net_rx_ring_bufs_alloc(dp, &dp->rx_rings[r]))
2578                        goto err_free_ring;
2579        }
2580
2581        return 0;
2582
2583err_free_prev:
2584        while (r--) {
2585                nfp_net_rx_ring_bufs_free(dp, &dp->rx_rings[r]);
2586err_free_ring:
2587                nfp_net_rx_ring_free(&dp->rx_rings[r]);
2588        }
2589        kfree(dp->rx_rings);
2590        return -ENOMEM;
2591}
2592
2593static void nfp_net_rx_rings_free(struct nfp_net_dp *dp)
2594{
2595        unsigned int r;
2596
2597        for (r = 0; r < dp->num_rx_rings; r++) {
2598                nfp_net_rx_ring_bufs_free(dp, &dp->rx_rings[r]);
2599                nfp_net_rx_ring_free(&dp->rx_rings[r]);
2600        }
2601
2602        kfree(dp->rx_rings);
2603}
2604
2605static void
2606nfp_net_vector_assign_rings(struct nfp_net_dp *dp,
2607                            struct nfp_net_r_vector *r_vec, int idx)
2608{
2609        r_vec->rx_ring = idx < dp->num_rx_rings ? &dp->rx_rings[idx] : NULL;
2610        r_vec->tx_ring =
2611                idx < dp->num_stack_tx_rings ? &dp->tx_rings[idx] : NULL;
2612
2613        r_vec->xdp_ring = idx < dp->num_tx_rings - dp->num_stack_tx_rings ?
2614                &dp->tx_rings[dp->num_stack_tx_rings + idx] : NULL;
2615}
2616
2617static int
2618nfp_net_prepare_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
2619                       int idx)
2620{
2621        int err;
2622
2623        /* Setup NAPI */
2624        if (nn->dp.netdev)
2625                netif_napi_add(nn->dp.netdev, &r_vec->napi,
2626                               nfp_net_poll, NAPI_POLL_WEIGHT);
2627        else
2628                tasklet_enable(&r_vec->tasklet);
2629
2630        snprintf(r_vec->name, sizeof(r_vec->name),
2631                 "%s-rxtx-%d", nfp_net_name(nn), idx);
2632        err = request_irq(r_vec->irq_vector, r_vec->handler, 0, r_vec->name,
2633                          r_vec);
2634        if (err) {
2635                if (nn->dp.netdev)
2636                        netif_napi_del(&r_vec->napi);
2637                else
2638                        tasklet_disable(&r_vec->tasklet);
2639
2640                nn_err(nn, "Error requesting IRQ %d\n", r_vec->irq_vector);
2641                return err;
2642        }
2643        disable_irq(r_vec->irq_vector);
2644
2645        irq_set_affinity_hint(r_vec->irq_vector, &r_vec->affinity_mask);
2646
2647        nn_dbg(nn, "RV%02d: irq=%03d/%03d\n", idx, r_vec->irq_vector,
2648               r_vec->irq_entry);
2649
2650        return 0;
2651}
2652
2653static void
2654nfp_net_cleanup_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec)
2655{
2656        irq_set_affinity_hint(r_vec->irq_vector, NULL);
2657        if (nn->dp.netdev)
2658                netif_napi_del(&r_vec->napi);
2659        else
2660                tasklet_disable(&r_vec->tasklet);
2661
2662        free_irq(r_vec->irq_vector, r_vec);
2663}
2664
2665/**
2666 * nfp_net_rss_write_itbl() - Write RSS indirection table to device
2667 * @nn:      NFP Net device to reconfigure
2668 */
2669void nfp_net_rss_write_itbl(struct nfp_net *nn)
2670{
2671        int i;
2672
2673        for (i = 0; i < NFP_NET_CFG_RSS_ITBL_SZ; i += 4)
2674                nn_writel(nn, NFP_NET_CFG_RSS_ITBL + i,
2675                          get_unaligned_le32(nn->rss_itbl + i));
2676}
2677
2678/**
2679 * nfp_net_rss_write_key() - Write RSS hash key to device
2680 * @nn:      NFP Net device to reconfigure
2681 */
2682void nfp_net_rss_write_key(struct nfp_net *nn)
2683{
2684        int i;
2685
2686        for (i = 0; i < nfp_net_rss_key_sz(nn); i += 4)
2687                nn_writel(nn, NFP_NET_CFG_RSS_KEY + i,
2688                          get_unaligned_le32(nn->rss_key + i));
2689}
2690
2691/**
2692 * nfp_net_coalesce_write_cfg() - Write irq coalescence configuration to HW
2693 * @nn:      NFP Net device to reconfigure
2694 */
2695void nfp_net_coalesce_write_cfg(struct nfp_net *nn)
2696{
2697        u8 i;
2698        u32 factor;
2699        u32 value;
2700
2701        /* Compute factor used to convert coalesce '_usecs' parameters to
2702         * ME timestamp ticks.  There are 16 ME clock cycles for each timestamp
2703         * count.
2704         */
2705        factor = nn->tlv_caps.me_freq_mhz / 16;
2706
2707        /* copy RX interrupt coalesce parameters */
2708        value = (nn->rx_coalesce_max_frames << 16) |
2709                (factor * nn->rx_coalesce_usecs);
2710        for (i = 0; i < nn->dp.num_rx_rings; i++)
2711                nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(i), value);
2712
2713        /* copy TX interrupt coalesce parameters */
2714        value = (nn->tx_coalesce_max_frames << 16) |
2715                (factor * nn->tx_coalesce_usecs);
2716        for (i = 0; i < nn->dp.num_tx_rings; i++)
2717                nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(i), value);
2718}
2719
2720/**
2721 * nfp_net_write_mac_addr() - Write mac address to the device control BAR
2722 * @nn:      NFP Net device to reconfigure
2723 * @addr:    MAC address to write
2724 *
2725 * Writes the MAC address from the netdev to the device control BAR.  Does not
2726 * perform the required reconfig.  We do a bit of byte swapping dance because
2727 * firmware is LE.
2728 */
2729static void nfp_net_write_mac_addr(struct nfp_net *nn, const u8 *addr)
2730{
2731        nn_writel(nn, NFP_NET_CFG_MACADDR + 0, get_unaligned_be32(addr));
2732        nn_writew(nn, NFP_NET_CFG_MACADDR + 6, get_unaligned_be16(addr + 4));
2733}
2734
2735static void nfp_net_vec_clear_ring_data(struct nfp_net *nn, unsigned int idx)
2736{
2737        nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), 0);
2738        nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), 0);
2739        nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), 0);
2740
2741        nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), 0);
2742        nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), 0);
2743        nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), 0);
2744}
2745
2746/**
2747 * nfp_net_clear_config_and_disable() - Clear control BAR and disable NFP
2748 * @nn:      NFP Net device to reconfigure
2749 *
2750 * Warning: must be fully idempotent.
2751 */
2752static void nfp_net_clear_config_and_disable(struct nfp_net *nn)
2753{
2754        u32 new_ctrl, update;
2755        unsigned int r;
2756        int err;
2757
2758        new_ctrl = nn->dp.ctrl;
2759        new_ctrl &= ~NFP_NET_CFG_CTRL_ENABLE;
2760        update = NFP_NET_CFG_UPDATE_GEN;
2761        update |= NFP_NET_CFG_UPDATE_MSIX;
2762        update |= NFP_NET_CFG_UPDATE_RING;
2763
2764        if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2765                new_ctrl &= ~NFP_NET_CFG_CTRL_RINGCFG;
2766
2767        nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
2768        nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
2769
2770        nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2771        err = nfp_net_reconfig(nn, update);
2772        if (err)
2773                nn_err(nn, "Could not disable device: %d\n", err);
2774
2775        for (r = 0; r < nn->dp.num_rx_rings; r++)
2776                nfp_net_rx_ring_reset(&nn->dp.rx_rings[r]);
2777        for (r = 0; r < nn->dp.num_tx_rings; r++)
2778                nfp_net_tx_ring_reset(&nn->dp, &nn->dp.tx_rings[r]);
2779        for (r = 0; r < nn->dp.num_r_vecs; r++)
2780                nfp_net_vec_clear_ring_data(nn, r);
2781
2782        nn->dp.ctrl = new_ctrl;
2783}
2784
2785static void
2786nfp_net_rx_ring_hw_cfg_write(struct nfp_net *nn,
2787                             struct nfp_net_rx_ring *rx_ring, unsigned int idx)
2788{
2789        /* Write the DMA address, size and MSI-X info to the device */
2790        nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), rx_ring->dma);
2791        nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), ilog2(rx_ring->cnt));
2792        nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), rx_ring->r_vec->irq_entry);
2793}
2794
2795static void
2796nfp_net_tx_ring_hw_cfg_write(struct nfp_net *nn,
2797                             struct nfp_net_tx_ring *tx_ring, unsigned int idx)
2798{
2799        nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), tx_ring->dma);
2800        nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), ilog2(tx_ring->cnt));
2801        nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), tx_ring->r_vec->irq_entry);
2802}
2803
2804/**
2805 * nfp_net_set_config_and_enable() - Write control BAR and enable NFP
2806 * @nn:      NFP Net device to reconfigure
2807 */
2808static int nfp_net_set_config_and_enable(struct nfp_net *nn)
2809{
2810        u32 bufsz, new_ctrl, update = 0;
2811        unsigned int r;
2812        int err;
2813
2814        new_ctrl = nn->dp.ctrl;
2815
2816        if (nn->dp.ctrl & NFP_NET_CFG_CTRL_RSS_ANY) {
2817                nfp_net_rss_write_key(nn);
2818                nfp_net_rss_write_itbl(nn);
2819                nn_writel(nn, NFP_NET_CFG_RSS_CTRL, nn->rss_cfg);
2820                update |= NFP_NET_CFG_UPDATE_RSS;
2821        }
2822
2823        if (nn->dp.ctrl & NFP_NET_CFG_CTRL_IRQMOD) {
2824                nfp_net_coalesce_write_cfg(nn);
2825                update |= NFP_NET_CFG_UPDATE_IRQMOD;
2826        }
2827
2828        for (r = 0; r < nn->dp.num_tx_rings; r++)
2829                nfp_net_tx_ring_hw_cfg_write(nn, &nn->dp.tx_rings[r], r);
2830        for (r = 0; r < nn->dp.num_rx_rings; r++)
2831                nfp_net_rx_ring_hw_cfg_write(nn, &nn->dp.rx_rings[r], r);
2832
2833        nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, nn->dp.num_tx_rings == 64 ?
2834                  0xffffffffffffffffULL : ((u64)1 << nn->dp.num_tx_rings) - 1);
2835
2836        nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, nn->dp.num_rx_rings == 64 ?
2837                  0xffffffffffffffffULL : ((u64)1 << nn->dp.num_rx_rings) - 1);
2838
2839        if (nn->dp.netdev)
2840                nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
2841
2842        nn_writel(nn, NFP_NET_CFG_MTU, nn->dp.mtu);
2843
2844        bufsz = nn->dp.fl_bufsz - nn->dp.rx_dma_off - NFP_NET_RX_BUF_NON_DATA;
2845        nn_writel(nn, NFP_NET_CFG_FLBUFSZ, bufsz);
2846
2847        /* Enable device */
2848        new_ctrl |= NFP_NET_CFG_CTRL_ENABLE;
2849        update |= NFP_NET_CFG_UPDATE_GEN;
2850        update |= NFP_NET_CFG_UPDATE_MSIX;
2851        update |= NFP_NET_CFG_UPDATE_RING;
2852        if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2853                new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG;
2854
2855        nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2856        err = nfp_net_reconfig(nn, update);
2857        if (err) {
2858                nfp_net_clear_config_and_disable(nn);
2859                return err;
2860        }
2861
2862        nn->dp.ctrl = new_ctrl;
2863
2864        for (r = 0; r < nn->dp.num_rx_rings; r++)
2865                nfp_net_rx_ring_fill_freelist(&nn->dp, &nn->dp.rx_rings[r]);
2866
2867        return 0;
2868}
2869
2870/**
2871 * nfp_net_close_stack() - Quiesce the stack (part of close)
2872 * @nn:      NFP Net device to reconfigure
2873 */
2874static void nfp_net_close_stack(struct nfp_net *nn)
2875{
2876        unsigned int r;
2877
2878        disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2879        netif_carrier_off(nn->dp.netdev);
2880        nn->link_up = false;
2881
2882        for (r = 0; r < nn->dp.num_r_vecs; r++) {
2883                disable_irq(nn->r_vecs[r].irq_vector);
2884                napi_disable(&nn->r_vecs[r].napi);
2885        }
2886
2887        netif_tx_disable(nn->dp.netdev);
2888}
2889
2890/**
2891 * nfp_net_close_free_all() - Free all runtime resources
2892 * @nn:      NFP Net device to reconfigure
2893 */
2894static void nfp_net_close_free_all(struct nfp_net *nn)
2895{
2896        unsigned int r;
2897
2898        nfp_net_tx_rings_free(&nn->dp);
2899        nfp_net_rx_rings_free(&nn->dp);
2900
2901        for (r = 0; r < nn->dp.num_r_vecs; r++)
2902                nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2903
2904        nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2905        nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2906}
2907
2908/**
2909 * nfp_net_netdev_close() - Called when the device is downed
2910 * @netdev:      netdev structure
2911 */
2912static int nfp_net_netdev_close(struct net_device *netdev)
2913{
2914        struct nfp_net *nn = netdev_priv(netdev);
2915
2916        /* Step 1: Disable RX and TX rings from the Linux kernel perspective
2917         */
2918        nfp_net_close_stack(nn);
2919
2920        /* Step 2: Tell NFP
2921         */
2922        nfp_net_clear_config_and_disable(nn);
2923        nfp_port_configure(netdev, false);
2924
2925        /* Step 3: Free resources
2926         */
2927        nfp_net_close_free_all(nn);
2928
2929        nn_dbg(nn, "%s down", netdev->name);
2930        return 0;
2931}
2932
2933void nfp_ctrl_close(struct nfp_net *nn)
2934{
2935        int r;
2936
2937        rtnl_lock();
2938
2939        for (r = 0; r < nn->dp.num_r_vecs; r++) {
2940                disable_irq(nn->r_vecs[r].irq_vector);
2941                tasklet_disable(&nn->r_vecs[r].tasklet);
2942        }
2943
2944        nfp_net_clear_config_and_disable(nn);
2945
2946        nfp_net_close_free_all(nn);
2947
2948        rtnl_unlock();
2949}
2950
2951/**
2952 * nfp_net_open_stack() - Start the device from stack's perspective
2953 * @nn:      NFP Net device to reconfigure
2954 */
2955static void nfp_net_open_stack(struct nfp_net *nn)
2956{
2957        unsigned int r;
2958
2959        for (r = 0; r < nn->dp.num_r_vecs; r++) {
2960                napi_enable(&nn->r_vecs[r].napi);
2961                enable_irq(nn->r_vecs[r].irq_vector);
2962        }
2963
2964        netif_tx_wake_all_queues(nn->dp.netdev);
2965
2966        enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2967        nfp_net_read_link_status(nn);
2968}
2969
2970static int nfp_net_open_alloc_all(struct nfp_net *nn)
2971{
2972        int err, r;
2973
2974        err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn",
2975                                      nn->exn_name, sizeof(nn->exn_name),
2976                                      NFP_NET_IRQ_EXN_IDX, nn->exn_handler);
2977        if (err)
2978                return err;
2979        err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc",
2980                                      nn->lsc_name, sizeof(nn->lsc_name),
2981                                      NFP_NET_IRQ_LSC_IDX, nn->lsc_handler);
2982        if (err)
2983                goto err_free_exn;
2984        disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2985
2986        for (r = 0; r < nn->dp.num_r_vecs; r++) {
2987                err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
2988                if (err)
2989                        goto err_cleanup_vec_p;
2990        }
2991
2992        err = nfp_net_rx_rings_prepare(nn, &nn->dp);
2993        if (err)
2994                goto err_cleanup_vec;
2995
2996        err = nfp_net_tx_rings_prepare(nn, &nn->dp);
2997        if (err)
2998                goto err_free_rx_rings;
2999
3000        for (r = 0; r < nn->max_r_vecs; r++)
3001                nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
3002
3003        return 0;
3004
3005err_free_rx_rings:
3006        nfp_net_rx_rings_free(&nn->dp);
3007err_cleanup_vec:
3008        r = nn->dp.num_r_vecs;
3009err_cleanup_vec_p:
3010        while (r--)
3011                nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3012        nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
3013err_free_exn:
3014        nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
3015        return err;
3016}
3017
3018static int nfp_net_netdev_open(struct net_device *netdev)
3019{
3020        struct nfp_net *nn = netdev_priv(netdev);
3021        int err;
3022
3023        /* Step 1: Allocate resources for rings and the like
3024         * - Request interrupts
3025         * - Allocate RX and TX ring resources
3026         * - Setup initial RSS table
3027         */
3028        err = nfp_net_open_alloc_all(nn);
3029        if (err)
3030                return err;
3031
3032        err = netif_set_real_num_tx_queues(netdev, nn->dp.num_stack_tx_rings);
3033        if (err)
3034                goto err_free_all;
3035
3036        err = netif_set_real_num_rx_queues(netdev, nn->dp.num_rx_rings);
3037        if (err)
3038                goto err_free_all;
3039
3040        /* Step 2: Configure the NFP
3041         * - Ifup the physical interface if it exists
3042         * - Enable rings from 0 to tx_rings/rx_rings - 1.
3043         * - Write MAC address (in case it changed)
3044         * - Set the MTU
3045         * - Set the Freelist buffer size
3046         * - Enable the FW
3047         */
3048        err = nfp_port_configure(netdev, true);
3049        if (err)
3050                goto err_free_all;
3051
3052        err = nfp_net_set_config_and_enable(nn);
3053        if (err)
3054                goto err_port_disable;
3055
3056        /* Step 3: Enable for kernel
3057         * - put some freelist descriptors on each RX ring
3058         * - enable NAPI on each ring
3059         * - enable all TX queues
3060         * - set link state
3061         */
3062        nfp_net_open_stack(nn);
3063
3064        return 0;
3065
3066err_port_disable:
3067        nfp_port_configure(netdev, false);
3068err_free_all:
3069        nfp_net_close_free_all(nn);
3070        return err;
3071}
3072
3073int nfp_ctrl_open(struct nfp_net *nn)
3074{
3075        int err, r;
3076
3077        /* ring dumping depends on vNICs being opened/closed under rtnl */
3078        rtnl_lock();
3079
3080        err = nfp_net_open_alloc_all(nn);
3081        if (err)
3082                goto err_unlock;
3083
3084        err = nfp_net_set_config_and_enable(nn);
3085        if (err)
3086                goto err_free_all;
3087
3088        for (r = 0; r < nn->dp.num_r_vecs; r++)
3089                enable_irq(nn->r_vecs[r].irq_vector);
3090
3091        rtnl_unlock();
3092
3093        return 0;
3094
3095err_free_all:
3096        nfp_net_close_free_all(nn);
3097err_unlock:
3098        rtnl_unlock();
3099        return err;
3100}
3101
3102static void nfp_net_set_rx_mode(struct net_device *netdev)
3103{
3104        struct nfp_net *nn = netdev_priv(netdev);
3105        u32 new_ctrl;
3106
3107        new_ctrl = nn->dp.ctrl;
3108
3109        if (!netdev_mc_empty(netdev) || netdev->flags & IFF_ALLMULTI)
3110                new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_L2MC;
3111        else
3112                new_ctrl &= ~NFP_NET_CFG_CTRL_L2MC;
3113
3114        if (netdev->flags & IFF_PROMISC) {
3115                if (nn->cap & NFP_NET_CFG_CTRL_PROMISC)
3116                        new_ctrl |= NFP_NET_CFG_CTRL_PROMISC;
3117                else
3118                        nn_warn(nn, "FW does not support promiscuous mode\n");
3119        } else {
3120                new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC;
3121        }
3122
3123        if (new_ctrl == nn->dp.ctrl)
3124                return;
3125
3126        nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
3127        nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN);
3128
3129        nn->dp.ctrl = new_ctrl;
3130}
3131
3132static void nfp_net_rss_init_itbl(struct nfp_net *nn)
3133{
3134        int i;
3135
3136        for (i = 0; i < sizeof(nn->rss_itbl); i++)
3137                nn->rss_itbl[i] =
3138                        ethtool_rxfh_indir_default(i, nn->dp.num_rx_rings);
3139}
3140
3141static void nfp_net_dp_swap(struct nfp_net *nn, struct nfp_net_dp *dp)
3142{
3143        struct nfp_net_dp new_dp = *dp;
3144
3145        *dp = nn->dp;
3146        nn->dp = new_dp;
3147
3148        nn->dp.netdev->mtu = new_dp.mtu;
3149
3150        if (!netif_is_rxfh_configured(nn->dp.netdev))
3151                nfp_net_rss_init_itbl(nn);
3152}
3153
3154static int nfp_net_dp_swap_enable(struct nfp_net *nn, struct nfp_net_dp *dp)
3155{
3156        unsigned int r;
3157        int err;
3158
3159        nfp_net_dp_swap(nn, dp);
3160
3161        for (r = 0; r < nn->max_r_vecs; r++)
3162                nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
3163
3164        err = netif_set_real_num_rx_queues(nn->dp.netdev, nn->dp.num_rx_rings);
3165        if (err)
3166                return err;
3167
3168        if (nn->dp.netdev->real_num_tx_queues != nn->dp.num_stack_tx_rings) {
3169                err = netif_set_real_num_tx_queues(nn->dp.netdev,
3170                                                   nn->dp.num_stack_tx_rings);
3171                if (err)
3172                        return err;
3173        }
3174
3175        return nfp_net_set_config_and_enable(nn);
3176}
3177
3178struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn)
3179{
3180        struct nfp_net_dp *new;
3181
3182        new = kmalloc(sizeof(*new), GFP_KERNEL);
3183        if (!new)
3184                return NULL;
3185
3186        *new = nn->dp;
3187
3188        /* Clear things which need to be recomputed */
3189        new->fl_bufsz = 0;
3190        new->tx_rings = NULL;
3191        new->rx_rings = NULL;
3192        new->num_r_vecs = 0;
3193        new->num_stack_tx_rings = 0;
3194
3195        return new;
3196}
3197
3198static int
3199nfp_net_check_config(struct nfp_net *nn, struct nfp_net_dp *dp,
3200                     struct netlink_ext_ack *extack)
3201{
3202        /* XDP-enabled tests */
3203        if (!dp->xdp_prog)
3204                return 0;
3205        if (dp->fl_bufsz > PAGE_SIZE) {
3206                NL_SET_ERR_MSG_MOD(extack, "MTU too large w/ XDP enabled");
3207                return -EINVAL;
3208        }
3209        if (dp->num_tx_rings > nn->max_tx_rings) {
3210                NL_SET_ERR_MSG_MOD(extack, "Insufficient number of TX rings w/ XDP enabled");
3211                return -EINVAL;
3212        }
3213
3214        return 0;
3215}
3216
3217int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *dp,
3218                          struct netlink_ext_ack *extack)
3219{
3220        int r, err;
3221
3222        dp->fl_bufsz = nfp_net_calc_fl_bufsz(dp);
3223
3224        dp->num_stack_tx_rings = dp->num_tx_rings;
3225        if (dp->xdp_prog)
3226                dp->num_stack_tx_rings -= dp->num_rx_rings;
3227
3228        dp->num_r_vecs = max(dp->num_rx_rings, dp->num_stack_tx_rings);
3229
3230        err = nfp_net_check_config(nn, dp, extack);
3231        if (err)
3232                goto exit_free_dp;
3233
3234        if (!netif_running(dp->netdev)) {
3235                nfp_net_dp_swap(nn, dp);
3236                err = 0;
3237                goto exit_free_dp;
3238        }
3239
3240        /* Prepare new rings */
3241        for (r = nn->dp.num_r_vecs; r < dp->num_r_vecs; r++) {
3242                err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
3243                if (err) {
3244                        dp->num_r_vecs = r;
3245                        goto err_cleanup_vecs;
3246                }
3247        }
3248
3249        err = nfp_net_rx_rings_prepare(nn, dp);
3250        if (err)
3251                goto err_cleanup_vecs;
3252
3253        err = nfp_net_tx_rings_prepare(nn, dp);
3254        if (err)
3255                goto err_free_rx;
3256
3257        /* Stop device, swap in new rings, try to start the firmware */
3258        nfp_net_close_stack(nn);
3259        nfp_net_clear_config_and_disable(nn);
3260
3261        err = nfp_net_dp_swap_enable(nn, dp);
3262        if (err) {
3263                int err2;
3264
3265                nfp_net_clear_config_and_disable(nn);
3266
3267                /* Try with old configuration and old rings */
3268                err2 = nfp_net_dp_swap_enable(nn, dp);
3269                if (err2)
3270                        nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n",
3271                               err, err2);
3272        }
3273        for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
3274                nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3275
3276        nfp_net_rx_rings_free(dp);
3277        nfp_net_tx_rings_free(dp);
3278
3279        nfp_net_open_stack(nn);
3280exit_free_dp:
3281        kfree(dp);
3282
3283        return err;
3284
3285err_free_rx:
3286        nfp_net_rx_rings_free(dp);
3287err_cleanup_vecs:
3288        for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
3289                nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3290        kfree(dp);
3291        return err;
3292}
3293
3294static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu)
3295{
3296        struct nfp_net *nn = netdev_priv(netdev);
3297        struct nfp_net_dp *dp;
3298        int err;
3299
3300        err = nfp_app_check_mtu(nn->app, netdev, new_mtu);
3301        if (err)
3302                return err;
3303
3304        dp = nfp_net_clone_dp(nn);
3305        if (!dp)
3306                return -ENOMEM;
3307
3308        dp->mtu = new_mtu;
3309
3310        return nfp_net_ring_reconfig(nn, dp, NULL);
3311}
3312
3313static int
3314nfp_net_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
3315{
3316        const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_ADD;
3317        struct nfp_net *nn = netdev_priv(netdev);
3318        int err;
3319
3320        /* Priority tagged packets with vlan id 0 are processed by the
3321         * NFP as untagged packets
3322         */
3323        if (!vid)
3324                return 0;
3325
3326        err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ);
3327        if (err)
3328                return err;
3329
3330        nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid);
3331        nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO,
3332                  ETH_P_8021Q);
3333
3334        return nfp_net_mbox_reconfig_and_unlock(nn, cmd);
3335}
3336
3337static int
3338nfp_net_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
3339{
3340        const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_KILL;
3341        struct nfp_net *nn = netdev_priv(netdev);
3342        int err;
3343
3344        /* Priority tagged packets with vlan id 0 are processed by the
3345         * NFP as untagged packets
3346         */
3347        if (!vid)
3348                return 0;
3349
3350        err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ);
3351        if (err)
3352                return err;
3353
3354        nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid);
3355        nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO,
3356                  ETH_P_8021Q);
3357
3358        return nfp_net_mbox_reconfig_and_unlock(nn, cmd);
3359}
3360
3361static void nfp_net_stat64(struct net_device *netdev,
3362                           struct rtnl_link_stats64 *stats)
3363{
3364        struct nfp_net *nn = netdev_priv(netdev);
3365        int r;
3366
3367        /* Collect software stats */
3368        for (r = 0; r < nn->max_r_vecs; r++) {
3369                struct nfp_net_r_vector *r_vec = &nn->r_vecs[r];
3370                u64 data[3];
3371                unsigned int start;
3372
3373                do {
3374                        start = u64_stats_fetch_begin(&r_vec->rx_sync);
3375                        data[0] = r_vec->rx_pkts;
3376                        data[1] = r_vec->rx_bytes;
3377                        data[2] = r_vec->rx_drops;
3378                } while (u64_stats_fetch_retry(&r_vec->rx_sync, start));
3379                stats->rx_packets += data[0];
3380                stats->rx_bytes += data[1];
3381                stats->rx_dropped += data[2];
3382
3383                do {
3384                        start = u64_stats_fetch_begin(&r_vec->tx_sync);
3385                        data[0] = r_vec->tx_pkts;
3386                        data[1] = r_vec->tx_bytes;
3387                        data[2] = r_vec->tx_errors;
3388                } while (u64_stats_fetch_retry(&r_vec->tx_sync, start));
3389                stats->tx_packets += data[0];
3390                stats->tx_bytes += data[1];
3391                stats->tx_errors += data[2];
3392        }
3393
3394        /* Add in device stats */
3395        stats->multicast += nn_readq(nn, NFP_NET_CFG_STATS_RX_MC_FRAMES);
3396        stats->rx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_RX_DISCARDS);
3397        stats->rx_errors += nn_readq(nn, NFP_NET_CFG_STATS_RX_ERRORS);
3398
3399        stats->tx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_TX_DISCARDS);
3400        stats->tx_errors += nn_readq(nn, NFP_NET_CFG_STATS_TX_ERRORS);
3401}
3402
3403static int nfp_net_set_features(struct net_device *netdev,
3404                                netdev_features_t features)
3405{
3406        netdev_features_t changed = netdev->features ^ features;
3407        struct nfp_net *nn = netdev_priv(netdev);
3408        u32 new_ctrl;
3409        int err;
3410
3411        /* Assume this is not called with features we have not advertised */
3412
3413        new_ctrl = nn->dp.ctrl;
3414
3415        if (changed & NETIF_F_RXCSUM) {
3416                if (features & NETIF_F_RXCSUM)
3417                        new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
3418                else
3419                        new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM_ANY;
3420        }
3421
3422        if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3423                if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))
3424                        new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
3425                else
3426                        new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM;
3427        }
3428
3429        if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
3430                if (features & (NETIF_F_TSO | NETIF_F_TSO6))
3431                        new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
3432                                              NFP_NET_CFG_CTRL_LSO;
3433                else
3434                        new_ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
3435        }
3436
3437        if (changed & NETIF_F_HW_VLAN_CTAG_RX) {
3438                if (features & NETIF_F_HW_VLAN_CTAG_RX)
3439                        new_ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
3440                else
3441                        new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN;
3442        }
3443
3444        if (changed & NETIF_F_HW_VLAN_CTAG_TX) {
3445                if (features & NETIF_F_HW_VLAN_CTAG_TX)
3446                        new_ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3447                else
3448                        new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN;
3449        }
3450
3451        if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) {
3452                if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
3453                        new_ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
3454                else
3455                        new_ctrl &= ~NFP_NET_CFG_CTRL_CTAG_FILTER;
3456        }
3457
3458        if (changed & NETIF_F_SG) {
3459                if (features & NETIF_F_SG)
3460                        new_ctrl |= NFP_NET_CFG_CTRL_GATHER;
3461                else
3462                        new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER;
3463        }
3464
3465        err = nfp_port_set_features(netdev, features);
3466        if (err)
3467                return err;
3468
3469        nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n",
3470               netdev->features, features, changed);
3471
3472        if (new_ctrl == nn->dp.ctrl)
3473                return 0;
3474
3475        nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->dp.ctrl, new_ctrl);
3476        nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
3477        err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
3478        if (err)
3479                return err;
3480
3481        nn->dp.ctrl = new_ctrl;
3482
3483        return 0;
3484}
3485
3486static netdev_features_t
3487nfp_net_features_check(struct sk_buff *skb, struct net_device *dev,
3488                       netdev_features_t features)
3489{
3490        u8 l4_hdr;
3491
3492        /* We can't do TSO over double tagged packets (802.1AD) */
3493        features &= vlan_features_check(skb, features);
3494
3495        if (!skb->encapsulation)
3496                return features;
3497
3498        /* Ensure that inner L4 header offset fits into TX descriptor field */
3499        if (skb_is_gso(skb)) {
3500                u32 hdrlen;
3501
3502                hdrlen = skb_inner_transport_header(skb) - skb->data +
3503                        inner_tcp_hdrlen(skb);
3504
3505                /* Assume worst case scenario of having longest possible
3506                 * metadata prepend - 8B
3507                 */
3508                if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ - 8))
3509                        features &= ~NETIF_F_GSO_MASK;
3510        }
3511
3512        /* VXLAN/GRE check */
3513        switch (vlan_get_protocol(skb)) {
3514        case htons(ETH_P_IP):
3515                l4_hdr = ip_hdr(skb)->protocol;
3516                break;
3517        case htons(ETH_P_IPV6):
3518                l4_hdr = ipv6_hdr(skb)->nexthdr;
3519                break;
3520        default:
3521                return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3522        }
3523
3524        if (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
3525            skb->inner_protocol != htons(ETH_P_TEB) ||
3526            (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) ||
3527            (l4_hdr == IPPROTO_UDP &&
3528             (skb_inner_mac_header(skb) - skb_transport_header(skb) !=
3529              sizeof(struct udphdr) + sizeof(struct vxlanhdr))))
3530                return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3531
3532        return features;
3533}
3534
3535static int
3536nfp_net_get_phys_port_name(struct net_device *netdev, char *name, size_t len)
3537{
3538        struct nfp_net *nn = netdev_priv(netdev);
3539        int n;
3540
3541        /* If port is defined, devlink_port is registered and devlink core
3542         * is taking care of name formatting.
3543         */
3544        if (nn->port)
3545                return -EOPNOTSUPP;
3546
3547        if (nn->dp.is_vf || nn->vnic_no_name)
3548                return -EOPNOTSUPP;
3549
3550        n = snprintf(name, len, "n%d", nn->id);
3551        if (n >= len)
3552                return -EINVAL;
3553
3554        return 0;
3555}
3556
3557static int nfp_net_xdp_setup_drv(struct nfp_net *nn, struct netdev_bpf *bpf)
3558{
3559        struct bpf_prog *prog = bpf->prog;
3560        struct nfp_net_dp *dp;
3561        int err;
3562
3563        if (!prog == !nn->dp.xdp_prog) {
3564                WRITE_ONCE(nn->dp.xdp_prog, prog);
3565                xdp_attachment_setup(&nn->xdp, bpf);
3566                return 0;
3567        }
3568
3569        dp = nfp_net_clone_dp(nn);
3570        if (!dp)
3571                return -ENOMEM;
3572
3573        dp->xdp_prog = prog;
3574        dp->num_tx_rings += prog ? nn->dp.num_rx_rings : -nn->dp.num_rx_rings;
3575        dp->rx_dma_dir = prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
3576        dp->rx_dma_off = prog ? XDP_PACKET_HEADROOM - nn->dp.rx_offset : 0;
3577
3578        /* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */
3579        err = nfp_net_ring_reconfig(nn, dp, bpf->extack);
3580        if (err)
3581                return err;
3582
3583        xdp_attachment_setup(&nn->xdp, bpf);
3584        return 0;
3585}
3586
3587static int nfp_net_xdp_setup_hw(struct nfp_net *nn, struct netdev_bpf *bpf)
3588{
3589        int err;
3590
3591        err = nfp_app_xdp_offload(nn->app, nn, bpf->prog, bpf->extack);
3592        if (err)
3593                return err;
3594
3595        xdp_attachment_setup(&nn->xdp_hw, bpf);
3596        return 0;
3597}
3598
3599static int nfp_net_xdp(struct net_device *netdev, struct netdev_bpf *xdp)
3600{
3601        struct nfp_net *nn = netdev_priv(netdev);
3602
3603        switch (xdp->command) {
3604        case XDP_SETUP_PROG:
3605                return nfp_net_xdp_setup_drv(nn, xdp);
3606        case XDP_SETUP_PROG_HW:
3607                return nfp_net_xdp_setup_hw(nn, xdp);
3608        default:
3609                return nfp_app_bpf(nn->app, nn, xdp);
3610        }
3611}
3612
3613static int nfp_net_set_mac_address(struct net_device *netdev, void *addr)
3614{
3615        struct nfp_net *nn = netdev_priv(netdev);
3616        struct sockaddr *saddr = addr;
3617        int err;
3618
3619        err = eth_prepare_mac_addr_change(netdev, addr);
3620        if (err)
3621                return err;
3622
3623        nfp_net_write_mac_addr(nn, saddr->sa_data);
3624
3625        err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MACADDR);
3626        if (err)
3627                return err;
3628
3629        eth_commit_mac_addr_change(netdev, addr);
3630
3631        return 0;
3632}
3633
3634const struct net_device_ops nfp_net_netdev_ops = {
3635        .ndo_init               = nfp_app_ndo_init,
3636        .ndo_uninit             = nfp_app_ndo_uninit,
3637        .ndo_open               = nfp_net_netdev_open,
3638        .ndo_stop               = nfp_net_netdev_close,
3639        .ndo_start_xmit         = nfp_net_tx,
3640        .ndo_get_stats64        = nfp_net_stat64,
3641        .ndo_vlan_rx_add_vid    = nfp_net_vlan_rx_add_vid,
3642        .ndo_vlan_rx_kill_vid   = nfp_net_vlan_rx_kill_vid,
3643        .ndo_set_vf_mac         = nfp_app_set_vf_mac,
3644        .ndo_set_vf_vlan        = nfp_app_set_vf_vlan,
3645        .ndo_set_vf_spoofchk    = nfp_app_set_vf_spoofchk,
3646        .ndo_set_vf_trust       = nfp_app_set_vf_trust,
3647        .ndo_get_vf_config      = nfp_app_get_vf_config,
3648        .ndo_set_vf_link_state  = nfp_app_set_vf_link_state,
3649        .ndo_setup_tc           = nfp_port_setup_tc,
3650        .ndo_tx_timeout         = nfp_net_tx_timeout,
3651        .ndo_set_rx_mode        = nfp_net_set_rx_mode,
3652        .ndo_change_mtu         = nfp_net_change_mtu,
3653        .ndo_set_mac_address    = nfp_net_set_mac_address,
3654        .ndo_set_features       = nfp_net_set_features,
3655        .ndo_features_check     = nfp_net_features_check,
3656        .ndo_get_phys_port_name = nfp_net_get_phys_port_name,
3657        .ndo_bpf                = nfp_net_xdp,
3658        .ndo_get_devlink_port   = nfp_devlink_get_devlink_port,
3659};
3660
3661static int nfp_udp_tunnel_sync(struct net_device *netdev, unsigned int table)
3662{
3663        struct nfp_net *nn = netdev_priv(netdev);
3664        int i;
3665
3666        BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1);
3667        for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2) {
3668                struct udp_tunnel_info ti0, ti1;
3669
3670                udp_tunnel_nic_get_port(netdev, table, i, &ti0);
3671                udp_tunnel_nic_get_port(netdev, table, i + 1, &ti1);
3672
3673                nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(ti0.port),
3674                          be16_to_cpu(ti1.port) << 16 | be16_to_cpu(ti0.port));
3675        }
3676
3677        return nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_VXLAN);
3678}
3679
3680static const struct udp_tunnel_nic_info nfp_udp_tunnels = {
3681        .sync_table     = nfp_udp_tunnel_sync,
3682        .flags          = UDP_TUNNEL_NIC_INFO_MAY_SLEEP |
3683                          UDP_TUNNEL_NIC_INFO_OPEN_ONLY,
3684        .tables         = {
3685                {
3686                        .n_entries      = NFP_NET_N_VXLAN_PORTS,
3687                        .tunnel_types   = UDP_TUNNEL_TYPE_VXLAN,
3688                },
3689        },
3690};
3691
3692/**
3693 * nfp_net_info() - Print general info about the NIC
3694 * @nn:      NFP Net device to reconfigure
3695 */
3696void nfp_net_info(struct nfp_net *nn)
3697{
3698        nn_info(nn, "Netronome NFP-6xxx %sNetdev: TxQs=%d/%d RxQs=%d/%d\n",
3699                nn->dp.is_vf ? "VF " : "",
3700                nn->dp.num_tx_rings, nn->max_tx_rings,
3701                nn->dp.num_rx_rings, nn->max_rx_rings);
3702        nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n",
3703                nn->fw_ver.resv, nn->fw_ver.class,
3704                nn->fw_ver.major, nn->fw_ver.minor,
3705                nn->max_mtu);
3706        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\n",
3707                nn->cap,
3708                nn->cap & NFP_NET_CFG_CTRL_PROMISC  ? "PROMISC "  : "",
3709                nn->cap & NFP_NET_CFG_CTRL_L2BC     ? "L2BCFILT " : "",