linux/drivers/net/ethernet/chelsio/cxgb4/l2t.c
<<
>>
Prefs
   1/*
   2 * This file is part of the Chelsio T4 Ethernet driver for Linux.
   3 *
   4 * Copyright (c) 2003-2014 Chelsio Communications, Inc. All rights reserved.
   5 *
   6 * This software is available to you under a choice of one of two
   7 * licenses.  You may choose to be licensed under the terms of the GNU
   8 * General Public License (GPL) Version 2, available from the file
   9 * COPYING in the main directory of this source tree, or the
  10 * OpenIB.org BSD license below:
  11 *
  12 *     Redistribution and use in source and binary forms, with or
  13 *     without modification, are permitted provided that the following
  14 *     conditions are met:
  15 *
  16 *      - Redistributions of source code must retain the above
  17 *        copyright notice, this list of conditions and the following
  18 *        disclaimer.
  19 *
  20 *      - Redistributions in binary form must reproduce the above
  21 *        copyright notice, this list of conditions and the following
  22 *        disclaimer in the documentation and/or other materials
  23 *        provided with the distribution.
  24 *
  25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  32 * SOFTWARE.
  33 */
  34
  35#include <linux/skbuff.h>
  36#include <linux/netdevice.h>
  37#include <linux/if.h>
  38#include <linux/if_vlan.h>
  39#include <linux/jhash.h>
  40#include <linux/module.h>
  41#include <linux/debugfs.h>
  42#include <linux/seq_file.h>
  43#include <net/neighbour.h>
  44#include "cxgb4.h"
  45#include "l2t.h"
  46#include "t4_msg.h"
  47#include "t4fw_api.h"
  48#include "t4_regs.h"
  49#include "t4_values.h"
  50
  51/* identifies sync vs async L2T_WRITE_REQs */
  52#define SYNC_WR_S    12
  53#define SYNC_WR_V(x) ((x) << SYNC_WR_S)
  54#define SYNC_WR_F    SYNC_WR_V(1)
  55
  56struct l2t_data {
  57        unsigned int l2t_start;     /* start index of our piece of the L2T */
  58        unsigned int l2t_size;      /* number of entries in l2tab */
  59        rwlock_t lock;
  60        atomic_t nfree;             /* number of free entries */
  61        struct l2t_entry *rover;    /* starting point for next allocation */
  62        struct l2t_entry l2tab[0];  /* MUST BE LAST */
  63};
  64
  65static inline unsigned int vlan_prio(const struct l2t_entry *e)
  66{
  67        return e->vlan >> VLAN_PRIO_SHIFT;
  68}
  69
  70static inline void l2t_hold(struct l2t_data *d, struct l2t_entry *e)
  71{
  72        if (atomic_add_return(1, &e->refcnt) == 1)  /* 0 -> 1 transition */
  73                atomic_dec(&d->nfree);
  74}
  75
  76/*
  77 * To avoid having to check address families we do not allow v4 and v6
  78 * neighbors to be on the same hash chain.  We keep v4 entries in the first
  79 * half of available hash buckets and v6 in the second.  We need at least two
  80 * entries in our L2T for this scheme to work.
  81 */
  82enum {
  83        L2T_MIN_HASH_BUCKETS = 2,
  84};
  85
  86static inline unsigned int arp_hash(struct l2t_data *d, const u32 *key,
  87                                    int ifindex)
  88{
  89        unsigned int l2t_size_half = d->l2t_size / 2;
  90
  91        return jhash_2words(*key, ifindex, 0) % l2t_size_half;
  92}
  93
  94static inline unsigned int ipv6_hash(struct l2t_data *d, const u32 *key,
  95                                     int ifindex)
  96{
  97        unsigned int l2t_size_half = d->l2t_size / 2;
  98        u32 xor = key[0] ^ key[1] ^ key[2] ^ key[3];
  99
 100        return (l2t_size_half +
 101                (jhash_2words(xor, ifindex, 0) % l2t_size_half));
 102}
 103
 104static unsigned int addr_hash(struct l2t_data *d, const u32 *addr,
 105                              int addr_len, int ifindex)
 106{
 107        return addr_len == 4 ? arp_hash(d, addr, ifindex) :
 108                               ipv6_hash(d, addr, ifindex);
 109}
 110
 111/*
 112 * Checks if an L2T entry is for the given IP/IPv6 address.  It does not check
 113 * whether the L2T entry and the address are of the same address family.
 114 * Callers ensure an address is only checked against L2T entries of the same
 115 * family, something made trivial by the separation of IP and IPv6 hash chains
 116 * mentioned above.  Returns 0 if there's a match,
 117 */
 118static int addreq(const struct l2t_entry *e, const u32 *addr)
 119{
 120        if (e->v6)
 121                return (e->addr[0] ^ addr[0]) | (e->addr[1] ^ addr[1]) |
 122                       (e->addr[2] ^ addr[2]) | (e->addr[3] ^ addr[3]);
 123        return e->addr[0] ^ addr[0];
 124}
 125
 126static void neigh_replace(struct l2t_entry *e, struct neighbour *n)
 127{
 128        neigh_hold(n);
 129        if (e->neigh)
 130                neigh_release(e->neigh);
 131        e->neigh = n;
 132}
 133
 134/*
 135 * Write an L2T entry.  Must be called with the entry locked.
 136 * The write may be synchronous or asynchronous.
 137 */
 138static int write_l2e(struct adapter *adap, struct l2t_entry *e, int sync)
 139{
 140        struct l2t_data *d = adap->l2t;
 141        unsigned int l2t_idx = e->idx + d->l2t_start;
 142        struct sk_buff *skb;
 143        struct cpl_l2t_write_req *req;
 144
 145        skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
 146        if (!skb)
 147                return -ENOMEM;
 148
 149        req = __skb_put(skb, sizeof(*req));
 150        INIT_TP_WR(req, 0);
 151
 152        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ,
 153                                        l2t_idx | (sync ? SYNC_WR_F : 0) |
 154                                        TID_QID_V(adap->sge.fw_evtq.abs_id)));
 155        req->params = htons(L2T_W_PORT_V(e->lport) | L2T_W_NOREPLY_V(!sync));
 156        req->l2t_idx = htons(l2t_idx);
 157        req->vlan = htons(e->vlan);
 158        if (e->neigh && !(e->neigh->dev->flags & IFF_LOOPBACK))
 159                memcpy(e->dmac, e->neigh->ha, sizeof(e->dmac));
 160        memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
 161
 162        t4_mgmt_tx(adap, skb);
 163
 164        if (sync && e->state != L2T_STATE_SWITCHING)
 165                e->state = L2T_STATE_SYNC_WRITE;
 166        return 0;
 167}
 168
 169/*
 170 * Send packets waiting in an L2T entry's ARP queue.  Must be called with the
 171 * entry locked.
 172 */
 173static void send_pending(struct adapter *adap, struct l2t_entry *e)
 174{
 175        struct sk_buff *skb;
 176
 177        while ((skb = __skb_dequeue(&e->arpq)) != NULL)
 178                t4_ofld_send(adap, skb);
 179}
 180
 181/*
 182 * Process a CPL_L2T_WRITE_RPL.  Wake up the ARP queue if it completes a
 183 * synchronous L2T_WRITE.  Note that the TID in the reply is really the L2T
 184 * index it refers to.
 185 */
 186void do_l2t_write_rpl(struct adapter *adap, const struct cpl_l2t_write_rpl *rpl)
 187{
 188        struct l2t_data *d = adap->l2t;
 189        unsigned int tid = GET_TID(rpl);
 190        unsigned int l2t_idx = tid % L2T_SIZE;
 191
 192        if (unlikely(rpl->status != CPL_ERR_NONE)) {
 193                dev_err(adap->pdev_dev,
 194                        "Unexpected L2T_WRITE_RPL status %u for entry %u\n",
 195                        rpl->status, l2t_idx);
 196                return;
 197        }
 198
 199        if (tid & SYNC_WR_F) {
 200                struct l2t_entry *e = &d->l2tab[l2t_idx - d->l2t_start];
 201
 202                spin_lock(&e->lock);
 203                if (e->state != L2T_STATE_SWITCHING) {
 204                        send_pending(adap, e);
 205                        e->state = (e->neigh->nud_state & NUD_STALE) ?
 206                                        L2T_STATE_STALE : L2T_STATE_VALID;
 207                }
 208                spin_unlock(&e->lock);
 209        }
 210}
 211
 212/*
 213 * Add a packet to an L2T entry's queue of packets awaiting resolution.
 214 * Must be called with the entry's lock held.
 215 */
 216static inline void arpq_enqueue(struct l2t_entry *e, struct sk_buff *skb)
 217{
 218        __skb_queue_tail(&e->arpq, skb);
 219}
 220
 221int cxgb4_l2t_send(struct net_device *dev, struct sk_buff *skb,
 222                   struct l2t_entry *e)
 223{
 224        struct adapter *adap = netdev2adap(dev);
 225
 226again:
 227        switch (e->state) {
 228        case L2T_STATE_STALE:     /* entry is stale, kick off revalidation */
 229                neigh_event_send(e->neigh, NULL);
 230                spin_lock_bh(&e->lock);
 231                if (e->state == L2T_STATE_STALE)
 232                        e->state = L2T_STATE_VALID;
 233                spin_unlock_bh(&e->lock);
 234        case L2T_STATE_VALID:     /* fast-path, send the packet on */
 235                return t4_ofld_send(adap, skb);
 236        case L2T_STATE_RESOLVING:
 237        case L2T_STATE_SYNC_WRITE:
 238                spin_lock_bh(&e->lock);
 239                if (e->state != L2T_STATE_SYNC_WRITE &&
 240                    e->state != L2T_STATE_RESOLVING) {
 241                        spin_unlock_bh(&e->lock);
 242                        goto again;
 243                }
 244                arpq_enqueue(e, skb);
 245                spin_unlock_bh(&e->lock);
 246
 247                if (e->state == L2T_STATE_RESOLVING &&
 248                    !neigh_event_send(e->neigh, NULL)) {
 249                        spin_lock_bh(&e->lock);
 250                        if (e->state == L2T_STATE_RESOLVING &&
 251                            !skb_queue_empty(&e->arpq))
 252                                write_l2e(adap, e, 1);
 253                        spin_unlock_bh(&e->lock);
 254                }
 255        }
 256        return 0;
 257}
 258EXPORT_SYMBOL(cxgb4_l2t_send);
 259
 260/*
 261 * Allocate a free L2T entry.  Must be called with l2t_data.lock held.
 262 */
 263static struct l2t_entry *alloc_l2e(struct l2t_data *d)
 264{
 265        struct l2t_entry *end, *e, **p;
 266
 267        if (!atomic_read(&d->nfree))
 268                return NULL;
 269
 270        /* there's definitely a free entry */
 271        for (e = d->rover, end = &d->l2tab[d->l2t_size]; e != end; ++e)
 272                if (atomic_read(&e->refcnt) == 0)
 273                        goto found;
 274
 275        for (e = d->l2tab; atomic_read(&e->refcnt); ++e)
 276                ;
 277found:
 278        d->rover = e + 1;
 279        atomic_dec(&d->nfree);
 280
 281        /*
 282         * The entry we found may be an inactive entry that is
 283         * presently in the hash table.  We need to remove it.
 284         */
 285        if (e->state < L2T_STATE_SWITCHING)
 286                for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next)
 287                        if (*p == e) {
 288                                *p = e->next;
 289                                e->next = NULL;
 290                                break;
 291                        }
 292
 293        e->state = L2T_STATE_UNUSED;
 294        return e;
 295}
 296
 297static struct l2t_entry *find_or_alloc_l2e(struct l2t_data *d, u16 vlan,
 298                                           u8 port, u8 *dmac)
 299{
 300        struct l2t_entry *end, *e, **p;
 301        struct l2t_entry *first_free = NULL;
 302
 303        for (e = &d->l2tab[0], end = &d->l2tab[d->l2t_size]; e != end; ++e) {
 304                if (atomic_read(&e->refcnt) == 0) {
 305                        if (!first_free)
 306                                first_free = e;
 307                } else {
 308                        if (e->state == L2T_STATE_SWITCHING) {
 309                                if (ether_addr_equal(e->dmac, dmac) &&
 310                                    (e->vlan == vlan) && (e->lport == port))
 311                                        goto exists;
 312                        }
 313                }
 314        }
 315
 316        if (first_free) {
 317                e = first_free;
 318                goto found;
 319        }
 320
 321        return NULL;
 322
 323found:
 324        /* The entry we found may be an inactive entry that is
 325         * presently in the hash table.  We need to remove it.
 326         */
 327        if (e->state < L2T_STATE_SWITCHING)
 328                for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next)
 329                        if (*p == e) {
 330                                *p = e->next;
 331                                e->next = NULL;
 332                                break;
 333                        }
 334        e->state = L2T_STATE_UNUSED;
 335
 336exists:
 337        return e;
 338}
 339
 340/* Called when an L2T entry has no more users.  The entry is left in the hash
 341 * table since it is likely to be reused but we also bump nfree to indicate
 342 * that the entry can be reallocated for a different neighbor.  We also drop
 343 * the existing neighbor reference in case the neighbor is going away and is
 344 * waiting on our reference.
 345 *
 346 * Because entries can be reallocated to other neighbors once their ref count
 347 * drops to 0 we need to take the entry's lock to avoid races with a new
 348 * incarnation.
 349 */
 350static void _t4_l2e_free(struct l2t_entry *e)
 351{
 352        struct l2t_data *d;
 353        struct sk_buff *skb;
 354
 355        if (atomic_read(&e->refcnt) == 0) {  /* hasn't been recycled */
 356                if (e->neigh) {
 357                        neigh_release(e->neigh);
 358                        e->neigh = NULL;
 359                }
 360                while ((skb = __skb_dequeue(&e->arpq)) != NULL)
 361                        kfree_skb(skb);
 362        }
 363
 364        d = container_of(e, struct l2t_data, l2tab[e->idx]);
 365        atomic_inc(&d->nfree);
 366}
 367
 368/* Locked version of _t4_l2e_free */
 369static void t4_l2e_free(struct l2t_entry *e)
 370{
 371        struct l2t_data *d;
 372        struct sk_buff *skb;
 373
 374        spin_lock_bh(&e->lock);
 375        if (atomic_read(&e->refcnt) == 0) {  /* hasn't been recycled */
 376                if (e->neigh) {
 377                        neigh_release(e->neigh);
 378                        e->neigh = NULL;
 379                }
 380                while ((skb = __skb_dequeue(&e->arpq)) != NULL)
 381                        kfree_skb(skb);
 382        }
 383        spin_unlock_bh(&e->lock);
 384
 385        d = container_of(e, struct l2t_data, l2tab[e->idx]);
 386        atomic_inc(&d->nfree);
 387}
 388
 389void cxgb4_l2t_release(struct l2t_entry *e)
 390{
 391        if (atomic_dec_and_test(&e->refcnt))
 392                t4_l2e_free(e);
 393}
 394EXPORT_SYMBOL(cxgb4_l2t_release);
 395
 396/*
 397 * Update an L2T entry that was previously used for the same next hop as neigh.
 398 * Must be called with softirqs disabled.
 399 */
 400static void reuse_entry(struct l2t_entry *e, struct neighbour *neigh)
 401{
 402        unsigned int nud_state;
 403
 404        spin_lock(&e->lock);                /* avoid race with t4_l2t_free */
 405        if (neigh != e->neigh)
 406                neigh_replace(e, neigh);
 407        nud_state = neigh->nud_state;
 408        if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)) ||
 409            !(nud_state & NUD_VALID))
 410                e->state = L2T_STATE_RESOLVING;
 411        else if (nud_state & NUD_CONNECTED)
 412                e->state = L2T_STATE_VALID;
 413        else
 414                e->state = L2T_STATE_STALE;
 415        spin_unlock(&e->lock);
 416}
 417
 418struct l2t_entry *cxgb4_l2t_get(struct l2t_data *d, struct neighbour *neigh,
 419                                const struct net_device *physdev,
 420                                unsigned int priority)
 421{
 422        u8 lport;
 423        u16 vlan;
 424        struct l2t_entry *e;
 425        unsigned int addr_len = neigh->tbl->key_len;
 426        u32 *addr = (u32 *)neigh->primary_key;
 427        int ifidx = neigh->dev->ifindex;
 428        int hash = addr_hash(d, addr, addr_len, ifidx);
 429
 430        if (neigh->dev->flags & IFF_LOOPBACK)
 431                lport = netdev2pinfo(physdev)->tx_chan + 4;
 432        else
 433                lport = netdev2pinfo(physdev)->lport;
 434
 435        if (is_vlan_dev(neigh->dev)) {
 436                vlan = vlan_dev_vlan_id(neigh->dev);
 437                vlan |= vlan_dev_get_egress_qos_mask(neigh->dev, priority);
 438        } else {
 439                vlan = VLAN_NONE;
 440        }
 441
 442        write_lock_bh(&d->lock);
 443        for (e = d->l2tab[hash].first; e; e = e->next)
 444                if (!addreq(e, addr) && e->ifindex == ifidx &&
 445                    e->vlan == vlan && e->lport == lport) {
 446                        l2t_hold(d, e);
 447                        if (atomic_read(&e->refcnt) == 1)
 448                                reuse_entry(e, neigh);
 449                        goto done;
 450                }
 451
 452        /* Need to allocate a new entry */
 453        e = alloc_l2e(d);
 454        if (e) {
 455                spin_lock(&e->lock);          /* avoid race with t4_l2t_free */
 456                e->state = L2T_STATE_RESOLVING;
 457                if (neigh->dev->flags & IFF_LOOPBACK)
 458                        memcpy(e->dmac, physdev->dev_addr, sizeof(e->dmac));
 459                memcpy(e->addr, addr, addr_len);
 460                e->ifindex = ifidx;
 461                e->hash = hash;
 462                e->lport = lport;
 463                e->v6 = addr_len == 16;
 464                atomic_set(&e->refcnt, 1);
 465                neigh_replace(e, neigh);
 466                e->vlan = vlan;
 467                e->next = d->l2tab[hash].first;
 468                d->l2tab[hash].first = e;
 469                spin_unlock(&e->lock);
 470        }
 471done:
 472        write_unlock_bh(&d->lock);
 473        return e;
 474}
 475EXPORT_SYMBOL(cxgb4_l2t_get);
 476
 477u64 cxgb4_select_ntuple(struct net_device *dev,
 478                        const struct l2t_entry *l2t)
 479{
 480        struct adapter *adap = netdev2adap(dev);
 481        struct tp_params *tp = &adap->params.tp;
 482        u64 ntuple = 0;
 483
 484        /* Initialize each of the fields which we care about which are present
 485         * in the Compressed Filter Tuple.
 486         */
 487        if (tp->vlan_shift >= 0 && l2t->vlan != VLAN_NONE)
 488                ntuple |= (u64)(FT_VLAN_VLD_F | l2t->vlan) << tp->vlan_shift;
 489
 490        if (tp->port_shift >= 0)
 491                ntuple |= (u64)l2t->lport << tp->port_shift;
 492
 493        if (tp->protocol_shift >= 0)
 494                ntuple |= (u64)IPPROTO_TCP << tp->protocol_shift;
 495
 496        if (tp->vnic_shift >= 0 && (tp->ingress_config & VNIC_F)) {
 497                u32 viid = cxgb4_port_viid(dev);
 498                u32 vf = FW_VIID_VIN_G(viid);
 499                u32 pf = FW_VIID_PFN_G(viid);
 500                u32 vld = FW_VIID_VIVLD_G(viid);
 501
 502                ntuple |= (u64)(FT_VNID_ID_VF_V(vf) |
 503                                FT_VNID_ID_PF_V(pf) |
 504                                FT_VNID_ID_VLD_V(vld)) << tp->vnic_shift;
 505        }
 506
 507        return ntuple;
 508}
 509EXPORT_SYMBOL(cxgb4_select_ntuple);
 510
 511/*
 512 * Called when address resolution fails for an L2T entry to handle packets
 513 * on the arpq head.  If a packet specifies a failure handler it is invoked,
 514 * otherwise the packet is sent to the device.
 515 */
 516static void handle_failed_resolution(struct adapter *adap, struct l2t_entry *e)
 517{
 518        struct sk_buff *skb;
 519
 520        while ((skb = __skb_dequeue(&e->arpq)) != NULL) {
 521                const struct l2t_skb_cb *cb = L2T_SKB_CB(skb);
 522
 523                spin_unlock(&e->lock);
 524                if (cb->arp_err_handler)
 525                        cb->arp_err_handler(cb->handle, skb);
 526                else
 527                        t4_ofld_send(adap, skb);
 528                spin_lock(&e->lock);
 529        }
 530}
 531
 532/*
 533 * Called when the host's neighbor layer makes a change to some entry that is
 534 * loaded into the HW L2 table.
 535 */
 536void t4_l2t_update(struct adapter *adap, struct neighbour *neigh)
 537{
 538        struct l2t_entry *e;
 539        struct sk_buff_head *arpq = NULL;
 540        struct l2t_data *d = adap->l2t;
 541        unsigned int addr_len = neigh->tbl->key_len;
 542        u32 *addr = (u32 *) neigh->primary_key;
 543        int ifidx = neigh->dev->ifindex;
 544        int hash = addr_hash(d, addr, addr_len, ifidx);
 545
 546        read_lock_bh(&d->lock);
 547        for (e = d->l2tab[hash].first; e; e = e->next)
 548                if (!addreq(e, addr) && e->ifindex == ifidx) {
 549                        spin_lock(&e->lock);
 550                        if (atomic_read(&e->refcnt))
 551                                goto found;
 552                        spin_unlock(&e->lock);
 553                        break;
 554                }
 555        read_unlock_bh(&d->lock);
 556        return;
 557
 558 found:
 559        read_unlock(&d->lock);
 560
 561        if (neigh != e->neigh)
 562                neigh_replace(e, neigh);
 563
 564        if (e->state == L2T_STATE_RESOLVING) {
 565                if (neigh->nud_state & NUD_FAILED) {
 566                        arpq = &e->arpq;
 567                } else if ((neigh->nud_state & (NUD_CONNECTED | NUD_STALE)) &&
 568                           !skb_queue_empty(&e->arpq)) {
 569                        write_l2e(adap, e, 1);
 570                }
 571        } else {
 572                e->state = neigh->nud_state & NUD_CONNECTED ?
 573                        L2T_STATE_VALID : L2T_STATE_STALE;
 574                if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)))
 575                        write_l2e(adap, e, 0);
 576        }
 577
 578        if (arpq)
 579                handle_failed_resolution(adap, e);
 580        spin_unlock_bh(&e->lock);
 581}
 582
 583/* Allocate an L2T entry for use by a switching rule.  Such need to be
 584 * explicitly freed and while busy they are not on any hash chain, so normal
 585 * address resolution updates do not see them.
 586 */
 587struct l2t_entry *t4_l2t_alloc_switching(struct adapter *adap, u16 vlan,
 588                                         u8 port, u8 *eth_addr)
 589{
 590        struct l2t_data *d = adap->l2t;
 591        struct l2t_entry *e;
 592        int ret;
 593
 594        write_lock_bh(&d->lock);
 595        e = find_or_alloc_l2e(d, vlan, port, eth_addr);
 596        if (e) {
 597                spin_lock(&e->lock);          /* avoid race with t4_l2t_free */
 598                if (!atomic_read(&e->refcnt)) {
 599                        e->state = L2T_STATE_SWITCHING;
 600                        e->vlan = vlan;
 601                        e->lport = port;
 602                        ether_addr_copy(e->dmac, eth_addr);
 603                        atomic_set(&e->refcnt, 1);
 604                        ret = write_l2e(adap, e, 0);
 605                        if (ret < 0) {
 606                                _t4_l2e_free(e);
 607                                spin_unlock(&e->lock);
 608                                write_unlock_bh(&d->lock);
 609                                return NULL;
 610                        }
 611                } else {
 612                        atomic_inc(&e->refcnt);
 613                }
 614
 615                spin_unlock(&e->lock);
 616        }
 617        write_unlock_bh(&d->lock);
 618        return e;
 619}
 620
 621/**
 622 * @dev: net_device pointer
 623 * @vlan: VLAN Id
 624 * @port: Associated port
 625 * @dmac: Destination MAC address to add to L2T
 626 * Returns pointer to the allocated l2t entry
 627 *
 628 * Allocates an L2T entry for use by switching rule of a filter
 629 */
 630struct l2t_entry *cxgb4_l2t_alloc_switching(struct net_device *dev, u16 vlan,
 631                                            u8 port, u8 *dmac)
 632{
 633        struct adapter *adap = netdev2adap(dev);
 634
 635        return t4_l2t_alloc_switching(adap, vlan, port, dmac);
 636}
 637EXPORT_SYMBOL(cxgb4_l2t_alloc_switching);
 638
 639struct l2t_data *t4_init_l2t(unsigned int l2t_start, unsigned int l2t_end)
 640{
 641        unsigned int l2t_size;
 642        int i;
 643        struct l2t_data *d;
 644
 645        if (l2t_start >= l2t_end || l2t_end >= L2T_SIZE)
 646                return NULL;
 647        l2t_size = l2t_end - l2t_start + 1;
 648        if (l2t_size < L2T_MIN_HASH_BUCKETS)
 649                return NULL;
 650
 651        d = kvzalloc(sizeof(*d) + l2t_size * sizeof(struct l2t_entry), GFP_KERNEL);
 652        if (!d)
 653                return NULL;
 654
 655        d->l2t_start = l2t_start;
 656        d->l2t_size = l2t_size;
 657
 658        d->rover = d->l2tab;
 659        atomic_set(&d->nfree, l2t_size);
 660        rwlock_init(&d->lock);
 661
 662        for (i = 0; i < d->l2t_size; ++i) {
 663                d->l2tab[i].idx = i;
 664                d->l2tab[i].state = L2T_STATE_UNUSED;
 665                spin_lock_init(&d->l2tab[i].lock);
 666                atomic_set(&d->l2tab[i].refcnt, 0);
 667                skb_queue_head_init(&d->l2tab[i].arpq);
 668        }
 669        return d;
 670}
 671
 672static inline void *l2t_get_idx(struct seq_file *seq, loff_t pos)
 673{
 674        struct l2t_data *d = seq->private;
 675
 676        return pos >= d->l2t_size ? NULL : &d->l2tab[pos];
 677}
 678
 679static void *l2t_seq_start(struct seq_file *seq, loff_t *pos)
 680{
 681        return *pos ? l2t_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 682}
 683
 684static void *l2t_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 685{
 686        v = l2t_get_idx(seq, *pos);
 687        if (v)
 688                ++*pos;
 689        return v;
 690}
 691
 692static void l2t_seq_stop(struct seq_file *seq, void *v)
 693{
 694}
 695
 696static char l2e_state(const struct l2t_entry *e)
 697{
 698        switch (e->state) {
 699        case L2T_STATE_VALID: return 'V';
 700        case L2T_STATE_STALE: return 'S';
 701        case L2T_STATE_SYNC_WRITE: return 'W';
 702        case L2T_STATE_RESOLVING:
 703                return skb_queue_empty(&e->arpq) ? 'R' : 'A';
 704        case L2T_STATE_SWITCHING: return 'X';
 705        default:
 706                return 'U';
 707        }
 708}
 709
 710static int l2t_seq_show(struct seq_file *seq, void *v)
 711{
 712        if (v == SEQ_START_TOKEN)
 713                seq_puts(seq, " Idx IP address                "
 714                         "Ethernet address  VLAN/P LP State Users Port\n");
 715        else {
 716                char ip[60];
 717                struct l2t_data *d = seq->private;
 718                struct l2t_entry *e = v;
 719
 720                spin_lock_bh(&e->lock);
 721                if (e->state == L2T_STATE_SWITCHING)
 722                        ip[0] = '\0';
 723                else
 724                        sprintf(ip, e->v6 ? "%pI6c" : "%pI4", e->addr);
 725                seq_printf(seq, "%4u %-25s %17pM %4d %u %2u   %c   %5u %s\n",
 726                           e->idx + d->l2t_start, ip, e->dmac,
 727                           e->vlan & VLAN_VID_MASK, vlan_prio(e), e->lport,
 728                           l2e_state(e), atomic_read(&e->refcnt),
 729                           e->neigh ? e->neigh->dev->name : "");
 730                spin_unlock_bh(&e->lock);
 731        }
 732        return 0;
 733}
 734
 735static const struct seq_operations l2t_seq_ops = {
 736        .start = l2t_seq_start,
 737        .next = l2t_seq_next,
 738        .stop = l2t_seq_stop,
 739        .show = l2t_seq_show
 740};
 741
 742static int l2t_seq_open(struct inode *inode, struct file *file)
 743{
 744        int rc = seq_open(file, &l2t_seq_ops);
 745
 746        if (!rc) {
 747                struct adapter *adap = inode->i_private;
 748                struct seq_file *seq = file->private_data;
 749
 750                seq->private = adap->l2t;
 751        }
 752        return rc;
 753}
 754
 755const struct file_operations t4_l2t_fops = {
 756        .owner = THIS_MODULE,
 757        .open = l2t_seq_open,
 758        .read = seq_read,
 759        .llseek = seq_lseek,
 760        .release = seq_release,
 761};
 762