1/* 2 * This file is part of the Chelsio T4 Ethernet driver for Linux. 3 * 4 * Copyright (c) 2003-2010 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 <net/neighbour.h> 41#include "cxgb4.h" 42#include "l2t.h" 43#include "t4_msg.h" 44#include "t4fw_api.h" 45 46#define VLAN_NONE 0xfff 47 48/* identifies sync vs async L2T_WRITE_REQs */ 49#define F_SYNC_WR (1 << 12) 50 51enum { 52 L2T_STATE_VALID, /* entry is up to date */ 53 L2T_STATE_STALE, /* entry may be used but needs revalidation */ 54 L2T_STATE_RESOLVING, /* entry needs address resolution */ 55 L2T_STATE_SYNC_WRITE, /* synchronous write of entry underway */ 56 57 /* when state is one of the below the entry is not hashed */ 58 L2T_STATE_SWITCHING, /* entry is being used by a switching filter */ 59 L2T_STATE_UNUSED /* entry not in use */ 60}; 61 62struct l2t_data { 63 rwlock_t lock; 64 atomic_t nfree; /* number of free entries */ 65 struct l2t_entry *rover; /* starting point for next allocation */ 66 struct l2t_entry l2tab[L2T_SIZE]; 67}; 68 69static inline unsigned int vlan_prio(const struct l2t_entry *e) 70{ 71 return e->vlan >> 13; 72} 73 74static inline void l2t_hold(struct l2t_data *d, struct l2t_entry *e) 75{ 76 if (atomic_add_return(1, &e->refcnt) == 1) /* 0 -> 1 transition */ 77 atomic_dec(&d->nfree); 78} 79 80/* 81 * To avoid having to check address families we do not allow v4 and v6 82 * neighbors to be on the same hash chain. We keep v4 entries in the first 83 * half of available hash buckets and v6 in the second. 84 */ 85enum { 86 L2T_SZ_HALF = L2T_SIZE / 2, 87 L2T_HASH_MASK = L2T_SZ_HALF - 1 88}; 89 90static inline unsigned int arp_hash(const u32 *key, int ifindex) 91{ 92 return jhash_2words(*key, ifindex, 0) & L2T_HASH_MASK; 93} 94 95static inline unsigned int ipv6_hash(const u32 *key, int ifindex) 96{ 97 u32 xor = key[0] ^ key[1] ^ key[2] ^ key[3]; 98 99 return L2T_SZ_HALF + (jhash_2words(xor, ifindex, 0) & L2T_HASH_MASK); 100} 101 102static unsigned int addr_hash(const u32 *addr, int addr_len, int ifindex) 103{ 104 return addr_len == 4 ? arp_hash(addr, ifindex) : 105 ipv6_hash(addr, ifindex); 106} 107 108/* 109 * Checks if an L2T entry is for the given IP/IPv6 address. It does not check 110 * whether the L2T entry and the address are of the same address family. 111 * Callers ensure an address is only checked against L2T entries of the same 112 * family, something made trivial by the separation of IP and IPv6 hash chains 113 * mentioned above. Returns 0 if there's a match, 114 */ 115static int addreq(const struct l2t_entry *e, const u32 *addr) 116{ 117 if (e->v6) 118 return (e->addr[0] ^ addr[0]) | (e->addr[1] ^ addr[1]) | 119 (e->addr[2] ^ addr[2]) | (e->addr[3] ^ addr[3]); 120 return e->addr[0] ^ addr[0]; 121} 122 123static void neigh_replace(struct l2t_entry *e, struct neighbour *n) 124{ 125 neigh_hold(n); 126 if (e->neigh) 127 neigh_release(e->neigh); 128 e->neigh = n; 129} 130 131/* 132 * Write an L2T entry. Must be called with the entry locked. 133 * The write may be synchronous or asynchronous. 134 */ 135static int write_l2e(struct adapter *adap, struct l2t_entry *e, int sync) 136{ 137 struct sk_buff *skb; 138 struct cpl_l2t_write_req *req; 139 140 skb = alloc_skb(sizeof(*req), GFP_ATOMIC); 141 if (!skb) 142 return -ENOMEM; 143 144 req = (struct cpl_l2t_write_req *)__skb_put(skb, sizeof(*req)); 145 INIT_TP_WR(req, 0); 146 147 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, 148 e->idx | (sync ? F_SYNC_WR : 0) | 149 TID_QID(adap->sge.fw_evtq.abs_id))); 150 req->params = htons(L2T_W_PORT(e->lport) | L2T_W_NOREPLY(!sync)); 151 req->l2t_idx = htons(e->idx); 152 req->vlan = htons(e->vlan); 153 if (e->neigh) 154 memcpy(e->dmac, e->neigh->ha, sizeof(e->dmac)); 155 memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac)); 156 157 set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0); 158 t4_ofld_send(adap, skb); 159 160 if (sync && e->state != L2T_STATE_SWITCHING) 161 e->state = L2T_STATE_SYNC_WRITE; 162 return 0; 163} 164 165/* 166 * Send packets waiting in an L2T entry's ARP queue. Must be called with the 167 * entry locked. 168 */ 169static void send_pending(struct adapter *adap, struct l2t_entry *e) 170{ 171 while (e->arpq_head) { 172 struct sk_buff *skb = e->arpq_head; 173 174 e->arpq_head = skb->next; 175 skb->next = NULL; 176 t4_ofld_send(adap, skb); 177 } 178 e->arpq_tail = NULL; 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 unsigned int tid = GET_TID(rpl); 189 unsigned int idx = tid & (L2T_SIZE - 1); 190 191 if (unlikely(rpl->status != CPL_ERR_NONE)) { 192 dev_err(adap->pdev_dev, 193 "Unexpected L2T_WRITE_RPL status %u for entry %u\n", 194 rpl->status, idx); 195 return; 196 } 197 198 if (tid & F_SYNC_WR) { 199 struct l2t_entry *e = &adap->l2t->l2tab[idx]; 200 201 spin_lock(&e->lock); 202 if (e->state != L2T_STATE_SWITCHING) { 203 send_pending(adap, e); 204 e->state = (e->neigh->nud_state & NUD_STALE) ? 205 L2T_STATE_STALE : L2T_STATE_VALID; 206 } 207 spin_unlock(&e->lock); 208 } 209} 210 211/* 212 * Add a packet to an L2T entry's queue of packets awaiting resolution. 213 * Must be called with the entry's lock held. 214 */ 215static inline void arpq_enqueue(struct l2t_entry *e, struct sk_buff *skb) 216{ 217 skb->next = NULL; 218 if (e->arpq_head) 219 e->arpq_tail->next = skb; 220 else 221 e->arpq_head = skb; 222 e->arpq_tail = skb; 223} 224 225int cxgb4_l2t_send(struct net_device *dev, struct sk_buff *skb, 226 struct l2t_entry *e) 227{ 228 struct adapter *adap = netdev2adap(dev); 229 230again: 231 switch (e->state) { 232 case L2T_STATE_STALE: /* entry is stale, kick off revalidation */ 233 neigh_event_send(e->neigh, NULL); 234 spin_lock_bh(&e->lock); 235 if (e->state == L2T_STATE_STALE) 236 e->state = L2T_STATE_VALID; 237 spin_unlock_bh(&e->lock); 238 case L2T_STATE_VALID: /* fast-path, send the packet on */ 239 return t4_ofld_send(adap, skb); 240 case L2T_STATE_RESOLVING: 241 case L2T_STATE_SYNC_WRITE: 242 spin_lock_bh(&e->lock); 243 if (e->state != L2T_STATE_SYNC_WRITE && 244 e->state != L2T_STATE_RESOLVING) { 245 spin_unlock_bh(&e->lock); 246 goto again; 247 } 248 arpq_enqueue(e, skb); 249 spin_unlock_bh(&e->lock); 250 251 if (e->state == L2T_STATE_RESOLVING && 252 !neigh_event_send(e->neigh, NULL)) { 253 spin_lock_bh(&e->lock); 254 if (e->state == L2T_STATE_RESOLVING && e->arpq_head) 255 write_l2e(adap, e, 1); 256 spin_unlock_bh(&e->lock); 257 } 258 } 259 return 0; 260} 261EXPORT_SYMBOL(cxgb4_l2t_send); 262 263/* 264 * Allocate a free L2T entry. Must be called with l2t_data.lock held. 265 */ 266static struct l2t_entry *alloc_l2e(struct l2t_data *d) 267{ 268 struct l2t_entry *end, *e, **p; 269 270 if (!atomic_read(&d->nfree)) 271 return NULL; 272 273 /* there's definitely a free entry */ 274 for (e = d->rover, end = &d->l2tab[L2T_SIZE]; e != end; ++e) 275 if (atomic_read(&e->refcnt) == 0) 276 goto found; 277 278 for (e = d->l2tab; atomic_read(&e->refcnt); ++e) 279 ; 280found: 281 d->rover = e + 1; 282 atomic_dec(&d->nfree); 283 284 /* 285 * The entry we found may be an inactive entry that is 286 * presently in the hash table. We need to remove it. 287 */ 288 if (e->state < L2T_STATE_SWITCHING) 289 for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next) 290 if (*p == e) { 291 *p = e->next; 292 e->next = NULL; 293 break; 294 } 295 296 e->state = L2T_STATE_UNUSED; 297 return e; 298} 299 300/* 301 * Called when an L2T entry has no more users. 302 */ 303static void t4_l2e_free(struct l2t_entry *e) 304{ 305 struct l2t_data *d; 306 307 spin_lock_bh(&e->lock); 308 if (atomic_read(&e->refcnt) == 0) { /* hasn't been recycled */ 309 if (e->neigh) { 310 neigh_release(e->neigh); 311 e->neigh = NULL; 312 } 313 while (e->arpq_head) { 314 struct sk_buff *skb = e->arpq_head; 315 316 e->arpq_head = skb->next; 317 kfree_skb(skb); 318 } 319 e->arpq_tail = NULL; 320 } 321 spin_unlock_bh(&e->lock); 322 323 d = container_of(e, struct l2t_data, l2tab[e->idx]); 324 atomic_inc(&d->nfree); 325} 326 327void cxgb4_l2t_release(struct l2t_entry *e) 328{ 329 if (atomic_dec_and_test(&e->refcnt)) 330 t4_l2e_free(e); 331} 332EXPORT_SYMBOL(cxgb4_l2t_release); 333 334/* 335 * Update an L2T entry that was previously used for the same next hop as neigh. 336 * Must be called with softirqs disabled. 337 */ 338static void reuse_entry(struct l2t_entry *e, struct neighbour *neigh) 339{ 340 unsigned int nud_state; 341 342 spin_lock(&e->lock); /* avoid race with t4_l2t_free */ 343 if (neigh != e->neigh) 344 neigh_replace(e, neigh); 345 nud_state = neigh->nud_state; 346 if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)) || 347 !(nud_state & NUD_VALID)) 348 e->state = L2T_STATE_RESOLVING; 349 else if (nud_state & NUD_CONNECTED) 350 e->state = L2T_STATE_VALID; 351 else 352 e->state = L2T_STATE_STALE; 353 spin_unlock(&e->lock); 354} 355 356struct l2t_entry *cxgb4_l2t_get(struct l2t_data *d, struct neighbour *neigh, 357 const struct net_device *physdev, 358 unsigned int priority) 359{ 360 u8 lport; 361 u16 vlan; 362 struct l2t_entry *e; 363 int addr_len = neigh->tbl->key_len; 364 u32 *addr = (u32 *)neigh->primary_key; 365 int ifidx = neigh->dev->ifindex; 366 int hash = addr_hash(addr, addr_len, ifidx); 367 368 if (neigh->dev->flags & IFF_LOOPBACK) 369 lport = netdev2pinfo(physdev)->tx_chan + 4; 370 else 371 lport = netdev2pinfo(physdev)->lport; 372 373 if (neigh->dev->priv_flags & IFF_802_1Q_VLAN) 374 vlan = vlan_dev_vlan_id(neigh->dev); 375 else 376 vlan = VLAN_NONE; 377 378 write_lock_bh(&d->lock); 379 for (e = d->l2tab[hash].first; e; e = e->next) 380 if (!addreq(e, addr) && e->ifindex == ifidx && 381 e->vlan == vlan && e->lport == lport) { 382 l2t_hold(d, e); 383 if (atomic_read(&e->refcnt) == 1) 384 reuse_entry(e, neigh); 385 goto done; 386 } 387 388 /* Need to allocate a new entry */ 389 e = alloc_l2e(d); 390 if (e) { 391 spin_lock(&e->lock); /* avoid race with t4_l2t_free */ 392 e->state = L2T_STATE_RESOLVING; 393 memcpy(e->addr, addr, addr_len); 394 e->ifindex = ifidx; 395 e->hash = hash; 396 e->lport = lport; 397 e->v6 = addr_len == 16; 398 atomic_set(&e->refcnt, 1); 399 neigh_replace(e, neigh); 400 e->vlan = vlan; 401 e->next = d->l2tab[hash].first; 402 d->l2tab[hash].first = e; 403 spin_unlock(&e->lock); 404 } 405done: 406 write_unlock_bh(&d->lock); 407 return e; 408} 409EXPORT_SYMBOL(cxgb4_l2t_get); 410 411/* 412 * Called when address resolution fails for an L2T entry to handle packets 413 * on the arpq head. If a packet specifies a failure handler it is invoked, 414 * otherwise the packet is sent to the device. 415 */ 416static void handle_failed_resolution(struct adapter *adap, struct sk_buff *arpq) 417{ 418 while (arpq) { 419 struct sk_buff *skb = arpq; 420 const struct l2t_skb_cb *cb = L2T_SKB_CB(skb); 421 422 arpq = skb->next; 423 skb->next = NULL; 424 if (cb->arp_err_handler) 425 cb->arp_err_handler(cb->handle, skb); 426 else 427 t4_ofld_send(adap, skb); 428 } 429} 430 431/* 432 * Called when the host's neighbor layer makes a change to some entry that is 433 * loaded into the HW L2 table. 434 */ 435void t4_l2t_update(struct adapter *adap, struct neighbour *neigh) 436{ 437 struct l2t_entry *e; 438 struct sk_buff *arpq = NULL; 439 struct l2t_data *d = adap->l2t; 440 int addr_len = neigh->tbl->key_len; 441 u32 *addr = (u32 *) neigh->primary_key; 442 int ifidx = neigh->dev->ifindex; 443 int hash = addr_hash(addr, addr_len, ifidx); 444 445 read_lock_bh(&d->lock); 446 for (e = d->l2tab[hash].first; e; e = e->next) 447 if (!addreq(e, addr) && e->ifindex == ifidx) { 448 spin_lock(&e->lock); 449 if (atomic_read(&e->refcnt)) 450 goto found; 451 spin_unlock(&e->lock); 452 break; 453 } 454 read_unlock_bh(&d->lock); 455 return; 456 457 found: 458 read_unlock(&d->lock); 459 460 if (neigh != e->neigh) 461 neigh_replace(e, neigh); 462 463 if (e->state == L2T_STATE_RESOLVING) { 464 if (neigh->nud_state & NUD_FAILED) { 465 arpq = e->arpq_head; 466 e->arpq_head = e->arpq_tail = NULL; 467 } else if ((neigh->nud_state & (NUD_CONNECTED | NUD_STALE)) && 468 e->arpq_head) { 469 write_l2e(adap, e, 1); 470 } 471 } else { 472 e->state = neigh->nud_state & NUD_CONNECTED ? 473 L2T_STATE_VALID : L2T_STATE_STALE; 474 if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac))) 475 write_l2e(adap, e, 0); 476 } 477 478 spin_unlock_bh(&e->lock); 479 480 if (arpq) 481 handle_failed_resolution(adap, arpq); 482} 483 484/* 485 * Allocate an L2T entry for use by a switching rule. Such entries need to be 486 * explicitly freed and while busy they are not on any hash chain, so normal 487 * address resolution updates do not see them. 488 */ 489struct l2t_entry *t4_l2t_alloc_switching(struct l2t_data *d) 490{ 491 struct l2t_entry *e; 492 493 write_lock_bh(&d->lock); 494 e = alloc_l2e(d); 495 if (e) { 496 spin_lock(&e->lock); /* avoid race with t4_l2t_free */ 497 e->state = L2T_STATE_SWITCHING; 498 atomic_set(&e->refcnt, 1); 499 spin_unlock(&e->lock); 500 } 501 write_unlock_bh(&d->lock); 502 return e; 503} 504 505/* 506 * Sets/updates the contents of a switching L2T entry that has been allocated 507 * with an earlier call to @t4_l2t_alloc_switching. 508 */ 509int t4_l2t_set_switching(struct adapter *adap, struct l2t_entry *e, u16 vlan, 510 u8 port, u8 *eth_addr) 511{ 512 e->vlan = vlan; 513 e->lport = port; 514 memcpy(e->dmac, eth_addr, ETH_ALEN); 515 return write_l2e(adap, e, 0); 516} 517 518struct l2t_data *t4_init_l2t(void) 519{ 520 int i; 521 struct l2t_data *d; 522 523 d = t4_alloc_mem(sizeof(*d)); 524 if (!d) 525 return NULL; 526 527 d->rover = d->l2tab; 528 atomic_set(&d->nfree, L2T_SIZE); 529 rwlock_init(&d->lock); 530 531 for (i = 0; i < L2T_SIZE; ++i) { 532 d->l2tab[i].idx = i; 533 d->l2tab[i].state = L2T_STATE_UNUSED; 534 spin_lock_init(&d->l2tab[i].lock); 535 atomic_set(&d->l2tab[i].refcnt, 0); 536 } 537 return d; 538} 539 540#include <linux/module.h> 541#include <linux/debugfs.h> 542#include <linux/seq_file.h> 543 544static inline void *l2t_get_idx(struct seq_file *seq, loff_t pos) 545{ 546 struct l2t_entry *l2tab = seq->private; 547 548 return pos >= L2T_SIZE ? NULL : &l2tab[pos]; 549} 550 551static void *l2t_seq_start(struct seq_file *seq, loff_t *pos) 552{ 553 return *pos ? l2t_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; 554} 555 556static void *l2t_seq_next(struct seq_file *seq, void *v, loff_t *pos) 557{ 558 v = l2t_get_idx(seq, *pos); 559 if (v) 560 ++*pos; 561 return v; 562} 563 564static void l2t_seq_stop(struct seq_file *seq, void *v) 565{ 566} 567 568static char l2e_state(const struct l2t_entry *e) 569{ 570 switch (e->state) { 571 case L2T_STATE_VALID: return 'V'; 572 case L2T_STATE_STALE: return 'S'; 573 case L2T_STATE_SYNC_WRITE: return 'W'; 574 case L2T_STATE_RESOLVING: return e->arpq_head ? 'A' : 'R'; 575 case L2T_STATE_SWITCHING: return 'X'; 576 default: 577 return 'U'; 578 } 579} 580 581static int l2t_seq_show(struct seq_file *seq, void *v) 582{ 583 if (v == SEQ_START_TOKEN) 584 seq_puts(seq, " Idx IP address " 585 "Ethernet address VLAN/P LP State Users Port\n"); 586 else { 587 char ip[60]; 588 struct l2t_entry *e = v; 589 590 spin_lock_bh(&e->lock); 591 if (e->state == L2T_STATE_SWITCHING) 592 ip[0] = '\0'; 593 else 594 sprintf(ip, e->v6 ? "%pI6c" : "%pI4", e->addr); 595 seq_printf(seq, "%4u %-25s %17pM %4d %u %2u %c %5u %s\n", 596 e->idx, ip, e->dmac, 597 e->vlan & VLAN_VID_MASK, vlan_prio(e), e->lport, 598 l2e_state(e), atomic_read(&e->refcnt), 599 e->neigh ? e->neigh->dev->name : ""); 600 spin_unlock_bh(&e->lock); 601 } 602 return 0; 603} 604 605static const struct seq_operations l2t_seq_ops = { 606 .start = l2t_seq_start, 607 .next = l2t_seq_next, 608 .stop = l2t_seq_stop, 609 .show = l2t_seq_show 610}; 611 612static int l2t_seq_open(struct inode *inode, struct file *file) 613{ 614 int rc = seq_open(file, &l2t_seq_ops); 615 616 if (!rc) { 617 struct adapter *adap = inode->i_private; 618 struct seq_file *seq = file->private_data; 619 620 seq->private = adap->l2t->l2tab; 621 } 622 return rc; 623} 624 625const struct file_operations t4_l2t_fops = { 626 .owner = THIS_MODULE, 627 .open = l2t_seq_open, 628 .read = seq_read, 629 .llseek = seq_lseek, 630 .release = seq_release, 631}; 632