xref: /openbmc/linux/drivers/net/ethernet/chelsio/cxgb4/l2t.c (revision 943126417891372d56aa3fe46295cbf53db31370)
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 
56 struct 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 
65 static inline unsigned int vlan_prio(const struct l2t_entry *e)
66 {
67 	return e->vlan >> VLAN_PRIO_SHIFT;
68 }
69 
70 static 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  */
82 enum {
83 	L2T_MIN_HASH_BUCKETS = 2,
84 };
85 
86 static 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 
94 static 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 
104 static 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  */
118 static 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 
126 static 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  */
138 static 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  */
173 static 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  */
186 void 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  */
216 static inline void arpq_enqueue(struct l2t_entry *e, struct sk_buff *skb)
217 {
218 	__skb_queue_tail(&e->arpq, skb);
219 }
220 
221 int cxgb4_l2t_send(struct net_device *dev, struct sk_buff *skb,
222 		   struct l2t_entry *e)
223 {
224 	struct adapter *adap = netdev2adap(dev);
225 
226 again:
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 		/* fall through */
235 	case L2T_STATE_VALID:     /* fast-path, send the packet on */
236 		return t4_ofld_send(adap, skb);
237 	case L2T_STATE_RESOLVING:
238 	case L2T_STATE_SYNC_WRITE:
239 		spin_lock_bh(&e->lock);
240 		if (e->state != L2T_STATE_SYNC_WRITE &&
241 		    e->state != L2T_STATE_RESOLVING) {
242 			spin_unlock_bh(&e->lock);
243 			goto again;
244 		}
245 		arpq_enqueue(e, skb);
246 		spin_unlock_bh(&e->lock);
247 
248 		if (e->state == L2T_STATE_RESOLVING &&
249 		    !neigh_event_send(e->neigh, NULL)) {
250 			spin_lock_bh(&e->lock);
251 			if (e->state == L2T_STATE_RESOLVING &&
252 			    !skb_queue_empty(&e->arpq))
253 				write_l2e(adap, e, 1);
254 			spin_unlock_bh(&e->lock);
255 		}
256 	}
257 	return 0;
258 }
259 EXPORT_SYMBOL(cxgb4_l2t_send);
260 
261 /*
262  * Allocate a free L2T entry.  Must be called with l2t_data.lock held.
263  */
264 static struct l2t_entry *alloc_l2e(struct l2t_data *d)
265 {
266 	struct l2t_entry *end, *e, **p;
267 
268 	if (!atomic_read(&d->nfree))
269 		return NULL;
270 
271 	/* there's definitely a free entry */
272 	for (e = d->rover, end = &d->l2tab[d->l2t_size]; e != end; ++e)
273 		if (atomic_read(&e->refcnt) == 0)
274 			goto found;
275 
276 	for (e = d->l2tab; atomic_read(&e->refcnt); ++e)
277 		;
278 found:
279 	d->rover = e + 1;
280 	atomic_dec(&d->nfree);
281 
282 	/*
283 	 * The entry we found may be an inactive entry that is
284 	 * presently in the hash table.  We need to remove it.
285 	 */
286 	if (e->state < L2T_STATE_SWITCHING)
287 		for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next)
288 			if (*p == e) {
289 				*p = e->next;
290 				e->next = NULL;
291 				break;
292 			}
293 
294 	e->state = L2T_STATE_UNUSED;
295 	return e;
296 }
297 
298 static struct l2t_entry *find_or_alloc_l2e(struct l2t_data *d, u16 vlan,
299 					   u8 port, u8 *dmac)
300 {
301 	struct l2t_entry *end, *e, **p;
302 	struct l2t_entry *first_free = NULL;
303 
304 	for (e = &d->l2tab[0], end = &d->l2tab[d->l2t_size]; e != end; ++e) {
305 		if (atomic_read(&e->refcnt) == 0) {
306 			if (!first_free)
307 				first_free = e;
308 		} else {
309 			if (e->state == L2T_STATE_SWITCHING) {
310 				if (ether_addr_equal(e->dmac, dmac) &&
311 				    (e->vlan == vlan) && (e->lport == port))
312 					goto exists;
313 			}
314 		}
315 	}
316 
317 	if (first_free) {
318 		e = first_free;
319 		goto found;
320 	}
321 
322 	return NULL;
323 
324 found:
325 	/* The entry we found may be an inactive entry that is
326 	 * presently in the hash table.  We need to remove it.
327 	 */
328 	if (e->state < L2T_STATE_SWITCHING)
329 		for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next)
330 			if (*p == e) {
331 				*p = e->next;
332 				e->next = NULL;
333 				break;
334 			}
335 	e->state = L2T_STATE_UNUSED;
336 
337 exists:
338 	return e;
339 }
340 
341 /* Called when an L2T entry has no more users.  The entry is left in the hash
342  * table since it is likely to be reused but we also bump nfree to indicate
343  * that the entry can be reallocated for a different neighbor.  We also drop
344  * the existing neighbor reference in case the neighbor is going away and is
345  * waiting on our reference.
346  *
347  * Because entries can be reallocated to other neighbors once their ref count
348  * drops to 0 we need to take the entry's lock to avoid races with a new
349  * incarnation.
350  */
351 static void _t4_l2e_free(struct l2t_entry *e)
352 {
353 	struct l2t_data *d;
354 	struct sk_buff *skb;
355 
356 	if (atomic_read(&e->refcnt) == 0) {  /* hasn't been recycled */
357 		if (e->neigh) {
358 			neigh_release(e->neigh);
359 			e->neigh = NULL;
360 		}
361 		while ((skb = __skb_dequeue(&e->arpq)) != NULL)
362 			kfree_skb(skb);
363 	}
364 
365 	d = container_of(e, struct l2t_data, l2tab[e->idx]);
366 	atomic_inc(&d->nfree);
367 }
368 
369 /* Locked version of _t4_l2e_free */
370 static void t4_l2e_free(struct l2t_entry *e)
371 {
372 	struct l2t_data *d;
373 	struct sk_buff *skb;
374 
375 	spin_lock_bh(&e->lock);
376 	if (atomic_read(&e->refcnt) == 0) {  /* hasn't been recycled */
377 		if (e->neigh) {
378 			neigh_release(e->neigh);
379 			e->neigh = NULL;
380 		}
381 		while ((skb = __skb_dequeue(&e->arpq)) != NULL)
382 			kfree_skb(skb);
383 	}
384 	spin_unlock_bh(&e->lock);
385 
386 	d = container_of(e, struct l2t_data, l2tab[e->idx]);
387 	atomic_inc(&d->nfree);
388 }
389 
390 void cxgb4_l2t_release(struct l2t_entry *e)
391 {
392 	if (atomic_dec_and_test(&e->refcnt))
393 		t4_l2e_free(e);
394 }
395 EXPORT_SYMBOL(cxgb4_l2t_release);
396 
397 /*
398  * Update an L2T entry that was previously used for the same next hop as neigh.
399  * Must be called with softirqs disabled.
400  */
401 static void reuse_entry(struct l2t_entry *e, struct neighbour *neigh)
402 {
403 	unsigned int nud_state;
404 
405 	spin_lock(&e->lock);                /* avoid race with t4_l2t_free */
406 	if (neigh != e->neigh)
407 		neigh_replace(e, neigh);
408 	nud_state = neigh->nud_state;
409 	if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)) ||
410 	    !(nud_state & NUD_VALID))
411 		e->state = L2T_STATE_RESOLVING;
412 	else if (nud_state & NUD_CONNECTED)
413 		e->state = L2T_STATE_VALID;
414 	else
415 		e->state = L2T_STATE_STALE;
416 	spin_unlock(&e->lock);
417 }
418 
419 struct l2t_entry *cxgb4_l2t_get(struct l2t_data *d, struct neighbour *neigh,
420 				const struct net_device *physdev,
421 				unsigned int priority)
422 {
423 	u8 lport;
424 	u16 vlan;
425 	struct l2t_entry *e;
426 	unsigned int addr_len = neigh->tbl->key_len;
427 	u32 *addr = (u32 *)neigh->primary_key;
428 	int ifidx = neigh->dev->ifindex;
429 	int hash = addr_hash(d, addr, addr_len, ifidx);
430 
431 	if (neigh->dev->flags & IFF_LOOPBACK)
432 		lport = netdev2pinfo(physdev)->tx_chan + 4;
433 	else
434 		lport = netdev2pinfo(physdev)->lport;
435 
436 	if (is_vlan_dev(neigh->dev)) {
437 		vlan = vlan_dev_vlan_id(neigh->dev);
438 		vlan |= vlan_dev_get_egress_qos_mask(neigh->dev, priority);
439 	} else {
440 		vlan = VLAN_NONE;
441 	}
442 
443 	write_lock_bh(&d->lock);
444 	for (e = d->l2tab[hash].first; e; e = e->next)
445 		if (!addreq(e, addr) && e->ifindex == ifidx &&
446 		    e->vlan == vlan && e->lport == lport) {
447 			l2t_hold(d, e);
448 			if (atomic_read(&e->refcnt) == 1)
449 				reuse_entry(e, neigh);
450 			goto done;
451 		}
452 
453 	/* Need to allocate a new entry */
454 	e = alloc_l2e(d);
455 	if (e) {
456 		spin_lock(&e->lock);          /* avoid race with t4_l2t_free */
457 		e->state = L2T_STATE_RESOLVING;
458 		if (neigh->dev->flags & IFF_LOOPBACK)
459 			memcpy(e->dmac, physdev->dev_addr, sizeof(e->dmac));
460 		memcpy(e->addr, addr, addr_len);
461 		e->ifindex = ifidx;
462 		e->hash = hash;
463 		e->lport = lport;
464 		e->v6 = addr_len == 16;
465 		atomic_set(&e->refcnt, 1);
466 		neigh_replace(e, neigh);
467 		e->vlan = vlan;
468 		e->next = d->l2tab[hash].first;
469 		d->l2tab[hash].first = e;
470 		spin_unlock(&e->lock);
471 	}
472 done:
473 	write_unlock_bh(&d->lock);
474 	return e;
475 }
476 EXPORT_SYMBOL(cxgb4_l2t_get);
477 
478 u64 cxgb4_select_ntuple(struct net_device *dev,
479 			const struct l2t_entry *l2t)
480 {
481 	struct adapter *adap = netdev2adap(dev);
482 	struct tp_params *tp = &adap->params.tp;
483 	u64 ntuple = 0;
484 
485 	/* Initialize each of the fields which we care about which are present
486 	 * in the Compressed Filter Tuple.
487 	 */
488 	if (tp->vlan_shift >= 0 && l2t->vlan != VLAN_NONE)
489 		ntuple |= (u64)(FT_VLAN_VLD_F | l2t->vlan) << tp->vlan_shift;
490 
491 	if (tp->port_shift >= 0)
492 		ntuple |= (u64)l2t->lport << tp->port_shift;
493 
494 	if (tp->protocol_shift >= 0)
495 		ntuple |= (u64)IPPROTO_TCP << tp->protocol_shift;
496 
497 	if (tp->vnic_shift >= 0 && (tp->ingress_config & VNIC_F)) {
498 		u32 viid = cxgb4_port_viid(dev);
499 		u32 vf = FW_VIID_VIN_G(viid);
500 		u32 pf = FW_VIID_PFN_G(viid);
501 		u32 vld = FW_VIID_VIVLD_G(viid);
502 
503 		ntuple |= (u64)(FT_VNID_ID_VF_V(vf) |
504 				FT_VNID_ID_PF_V(pf) |
505 				FT_VNID_ID_VLD_V(vld)) << tp->vnic_shift;
506 	}
507 
508 	return ntuple;
509 }
510 EXPORT_SYMBOL(cxgb4_select_ntuple);
511 
512 /*
513  * Called when address resolution fails for an L2T entry to handle packets
514  * on the arpq head.  If a packet specifies a failure handler it is invoked,
515  * otherwise the packet is sent to the device.
516  */
517 static void handle_failed_resolution(struct adapter *adap, struct l2t_entry *e)
518 {
519 	struct sk_buff *skb;
520 
521 	while ((skb = __skb_dequeue(&e->arpq)) != NULL) {
522 		const struct l2t_skb_cb *cb = L2T_SKB_CB(skb);
523 
524 		spin_unlock(&e->lock);
525 		if (cb->arp_err_handler)
526 			cb->arp_err_handler(cb->handle, skb);
527 		else
528 			t4_ofld_send(adap, skb);
529 		spin_lock(&e->lock);
530 	}
531 }
532 
533 /*
534  * Called when the host's neighbor layer makes a change to some entry that is
535  * loaded into the HW L2 table.
536  */
537 void t4_l2t_update(struct adapter *adap, struct neighbour *neigh)
538 {
539 	struct l2t_entry *e;
540 	struct sk_buff_head *arpq = NULL;
541 	struct l2t_data *d = adap->l2t;
542 	unsigned int addr_len = neigh->tbl->key_len;
543 	u32 *addr = (u32 *) neigh->primary_key;
544 	int ifidx = neigh->dev->ifindex;
545 	int hash = addr_hash(d, addr, addr_len, ifidx);
546 
547 	read_lock_bh(&d->lock);
548 	for (e = d->l2tab[hash].first; e; e = e->next)
549 		if (!addreq(e, addr) && e->ifindex == ifidx) {
550 			spin_lock(&e->lock);
551 			if (atomic_read(&e->refcnt))
552 				goto found;
553 			spin_unlock(&e->lock);
554 			break;
555 		}
556 	read_unlock_bh(&d->lock);
557 	return;
558 
559  found:
560 	read_unlock(&d->lock);
561 
562 	if (neigh != e->neigh)
563 		neigh_replace(e, neigh);
564 
565 	if (e->state == L2T_STATE_RESOLVING) {
566 		if (neigh->nud_state & NUD_FAILED) {
567 			arpq = &e->arpq;
568 		} else if ((neigh->nud_state & (NUD_CONNECTED | NUD_STALE)) &&
569 			   !skb_queue_empty(&e->arpq)) {
570 			write_l2e(adap, e, 1);
571 		}
572 	} else {
573 		e->state = neigh->nud_state & NUD_CONNECTED ?
574 			L2T_STATE_VALID : L2T_STATE_STALE;
575 		if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)))
576 			write_l2e(adap, e, 0);
577 	}
578 
579 	if (arpq)
580 		handle_failed_resolution(adap, e);
581 	spin_unlock_bh(&e->lock);
582 }
583 
584 /* Allocate an L2T entry for use by a switching rule.  Such need to be
585  * explicitly freed and while busy they are not on any hash chain, so normal
586  * address resolution updates do not see them.
587  */
588 struct l2t_entry *t4_l2t_alloc_switching(struct adapter *adap, u16 vlan,
589 					 u8 port, u8 *eth_addr)
590 {
591 	struct l2t_data *d = adap->l2t;
592 	struct l2t_entry *e;
593 	int ret;
594 
595 	write_lock_bh(&d->lock);
596 	e = find_or_alloc_l2e(d, vlan, port, eth_addr);
597 	if (e) {
598 		spin_lock(&e->lock);          /* avoid race with t4_l2t_free */
599 		if (!atomic_read(&e->refcnt)) {
600 			e->state = L2T_STATE_SWITCHING;
601 			e->vlan = vlan;
602 			e->lport = port;
603 			ether_addr_copy(e->dmac, eth_addr);
604 			atomic_set(&e->refcnt, 1);
605 			ret = write_l2e(adap, e, 0);
606 			if (ret < 0) {
607 				_t4_l2e_free(e);
608 				spin_unlock(&e->lock);
609 				write_unlock_bh(&d->lock);
610 				return NULL;
611 			}
612 		} else {
613 			atomic_inc(&e->refcnt);
614 		}
615 
616 		spin_unlock(&e->lock);
617 	}
618 	write_unlock_bh(&d->lock);
619 	return e;
620 }
621 
622 /**
623  * @dev: net_device pointer
624  * @vlan: VLAN Id
625  * @port: Associated port
626  * @dmac: Destination MAC address to add to L2T
627  * Returns pointer to the allocated l2t entry
628  *
629  * Allocates an L2T entry for use by switching rule of a filter
630  */
631 struct l2t_entry *cxgb4_l2t_alloc_switching(struct net_device *dev, u16 vlan,
632 					    u8 port, u8 *dmac)
633 {
634 	struct adapter *adap = netdev2adap(dev);
635 
636 	return t4_l2t_alloc_switching(adap, vlan, port, dmac);
637 }
638 EXPORT_SYMBOL(cxgb4_l2t_alloc_switching);
639 
640 struct l2t_data *t4_init_l2t(unsigned int l2t_start, unsigned int l2t_end)
641 {
642 	unsigned int l2t_size;
643 	int i;
644 	struct l2t_data *d;
645 
646 	if (l2t_start >= l2t_end || l2t_end >= L2T_SIZE)
647 		return NULL;
648 	l2t_size = l2t_end - l2t_start + 1;
649 	if (l2t_size < L2T_MIN_HASH_BUCKETS)
650 		return NULL;
651 
652 	d = kvzalloc(sizeof(*d) + l2t_size * sizeof(struct l2t_entry), GFP_KERNEL);
653 	if (!d)
654 		return NULL;
655 
656 	d->l2t_start = l2t_start;
657 	d->l2t_size = l2t_size;
658 
659 	d->rover = d->l2tab;
660 	atomic_set(&d->nfree, l2t_size);
661 	rwlock_init(&d->lock);
662 
663 	for (i = 0; i < d->l2t_size; ++i) {
664 		d->l2tab[i].idx = i;
665 		d->l2tab[i].state = L2T_STATE_UNUSED;
666 		spin_lock_init(&d->l2tab[i].lock);
667 		atomic_set(&d->l2tab[i].refcnt, 0);
668 		skb_queue_head_init(&d->l2tab[i].arpq);
669 	}
670 	return d;
671 }
672 
673 static inline void *l2t_get_idx(struct seq_file *seq, loff_t pos)
674 {
675 	struct l2t_data *d = seq->private;
676 
677 	return pos >= d->l2t_size ? NULL : &d->l2tab[pos];
678 }
679 
680 static void *l2t_seq_start(struct seq_file *seq, loff_t *pos)
681 {
682 	return *pos ? l2t_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
683 }
684 
685 static void *l2t_seq_next(struct seq_file *seq, void *v, loff_t *pos)
686 {
687 	v = l2t_get_idx(seq, *pos);
688 	if (v)
689 		++*pos;
690 	return v;
691 }
692 
693 static void l2t_seq_stop(struct seq_file *seq, void *v)
694 {
695 }
696 
697 static char l2e_state(const struct l2t_entry *e)
698 {
699 	switch (e->state) {
700 	case L2T_STATE_VALID: return 'V';
701 	case L2T_STATE_STALE: return 'S';
702 	case L2T_STATE_SYNC_WRITE: return 'W';
703 	case L2T_STATE_RESOLVING:
704 		return skb_queue_empty(&e->arpq) ? 'R' : 'A';
705 	case L2T_STATE_SWITCHING: return 'X';
706 	default:
707 		return 'U';
708 	}
709 }
710 
711 static int l2t_seq_show(struct seq_file *seq, void *v)
712 {
713 	if (v == SEQ_START_TOKEN)
714 		seq_puts(seq, " Idx IP address                "
715 			 "Ethernet address  VLAN/P LP State Users Port\n");
716 	else {
717 		char ip[60];
718 		struct l2t_data *d = seq->private;
719 		struct l2t_entry *e = v;
720 
721 		spin_lock_bh(&e->lock);
722 		if (e->state == L2T_STATE_SWITCHING)
723 			ip[0] = '\0';
724 		else
725 			sprintf(ip, e->v6 ? "%pI6c" : "%pI4", e->addr);
726 		seq_printf(seq, "%4u %-25s %17pM %4d %u %2u   %c   %5u %s\n",
727 			   e->idx + d->l2t_start, ip, e->dmac,
728 			   e->vlan & VLAN_VID_MASK, vlan_prio(e), e->lport,
729 			   l2e_state(e), atomic_read(&e->refcnt),
730 			   e->neigh ? e->neigh->dev->name : "");
731 		spin_unlock_bh(&e->lock);
732 	}
733 	return 0;
734 }
735 
736 static const struct seq_operations l2t_seq_ops = {
737 	.start = l2t_seq_start,
738 	.next = l2t_seq_next,
739 	.stop = l2t_seq_stop,
740 	.show = l2t_seq_show
741 };
742 
743 static int l2t_seq_open(struct inode *inode, struct file *file)
744 {
745 	int rc = seq_open(file, &l2t_seq_ops);
746 
747 	if (!rc) {
748 		struct adapter *adap = inode->i_private;
749 		struct seq_file *seq = file->private_data;
750 
751 		seq->private = adap->l2t;
752 	}
753 	return rc;
754 }
755 
756 const struct file_operations t4_l2t_fops = {
757 	.owner = THIS_MODULE,
758 	.open = l2t_seq_open,
759 	.read = seq_read,
760 	.llseek = seq_lseek,
761 	.release = seq_release,
762 };
763