2 * This file is part of the Chelsio T4 Ethernet driver for Linux.
4 * Copyright (c) 2003-2010 Chelsio Communications, Inc. All rights reserved.
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:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
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.
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
35 #include <linux/skbuff.h>
36 #include <linux/netdevice.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>
50 #define VLAN_NONE 0xfff
52 /* identifies sync vs async L2T_WRITE_REQs */
53 #define F_SYNC_WR (1 << 12)
56 L2T_STATE_VALID, /* entry is up to date */
57 L2T_STATE_STALE, /* entry may be used but needs revalidation */
58 L2T_STATE_RESOLVING, /* entry needs address resolution */
59 L2T_STATE_SYNC_WRITE, /* synchronous write of entry underway */
61 /* when state is one of the below the entry is not hashed */
62 L2T_STATE_SWITCHING, /* entry is being used by a switching filter */
63 L2T_STATE_UNUSED /* entry not in use */
68 atomic_t nfree; /* number of free entries */
69 struct l2t_entry *rover; /* starting point for next allocation */
70 struct l2t_entry l2tab[L2T_SIZE];
73 static inline unsigned int vlan_prio(const struct l2t_entry *e)
78 static inline void l2t_hold(struct l2t_data *d, struct l2t_entry *e)
80 if (atomic_add_return(1, &e->refcnt) == 1) /* 0 -> 1 transition */
81 atomic_dec(&d->nfree);
85 * To avoid having to check address families we do not allow v4 and v6
86 * neighbors to be on the same hash chain. We keep v4 entries in the first
87 * half of available hash buckets and v6 in the second.
90 L2T_SZ_HALF = L2T_SIZE / 2,
91 L2T_HASH_MASK = L2T_SZ_HALF - 1
94 static inline unsigned int arp_hash(const u32 *key, int ifindex)
96 return jhash_2words(*key, ifindex, 0) & L2T_HASH_MASK;
99 static inline unsigned int ipv6_hash(const u32 *key, int ifindex)
101 u32 xor = key[0] ^ key[1] ^ key[2] ^ key[3];
103 return L2T_SZ_HALF + (jhash_2words(xor, ifindex, 0) & L2T_HASH_MASK);
106 static unsigned int addr_hash(const u32 *addr, int addr_len, int ifindex)
108 return addr_len == 4 ? arp_hash(addr, ifindex) :
109 ipv6_hash(addr, ifindex);
113 * Checks if an L2T entry is for the given IP/IPv6 address. It does not check
114 * whether the L2T entry and the address are of the same address family.
115 * Callers ensure an address is only checked against L2T entries of the same
116 * family, something made trivial by the separation of IP and IPv6 hash chains
117 * mentioned above. Returns 0 if there's a match,
119 static int addreq(const struct l2t_entry *e, const u32 *addr)
122 return (e->addr[0] ^ addr[0]) | (e->addr[1] ^ addr[1]) |
123 (e->addr[2] ^ addr[2]) | (e->addr[3] ^ addr[3]);
124 return e->addr[0] ^ addr[0];
127 static void neigh_replace(struct l2t_entry *e, struct neighbour *n)
131 neigh_release(e->neigh);
136 * Write an L2T entry. Must be called with the entry locked.
137 * The write may be synchronous or asynchronous.
139 static int write_l2e(struct adapter *adap, struct l2t_entry *e, int sync)
142 struct cpl_l2t_write_req *req;
144 skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
148 req = (struct cpl_l2t_write_req *)__skb_put(skb, sizeof(*req));
151 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ,
152 e->idx | (sync ? F_SYNC_WR : 0) |
153 TID_QID(adap->sge.fw_evtq.abs_id)));
154 req->params = htons(L2T_W_PORT(e->lport) | L2T_W_NOREPLY(!sync));
155 req->l2t_idx = htons(e->idx);
156 req->vlan = htons(e->vlan);
157 if (e->neigh && !(e->neigh->dev->flags & IFF_LOOPBACK))
158 memcpy(e->dmac, e->neigh->ha, sizeof(e->dmac));
159 memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
161 set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
162 t4_ofld_send(adap, skb);
164 if (sync && e->state != L2T_STATE_SWITCHING)
165 e->state = L2T_STATE_SYNC_WRITE;
170 * Send packets waiting in an L2T entry's ARP queue. Must be called with the
173 static void send_pending(struct adapter *adap, struct l2t_entry *e)
175 while (e->arpq_head) {
176 struct sk_buff *skb = e->arpq_head;
178 e->arpq_head = skb->next;
180 t4_ofld_send(adap, skb);
186 * Process a CPL_L2T_WRITE_RPL. Wake up the ARP queue if it completes a
187 * synchronous L2T_WRITE. Note that the TID in the reply is really the L2T
188 * index it refers to.
190 void do_l2t_write_rpl(struct adapter *adap, const struct cpl_l2t_write_rpl *rpl)
192 unsigned int tid = GET_TID(rpl);
193 unsigned int idx = tid & (L2T_SIZE - 1);
195 if (unlikely(rpl->status != CPL_ERR_NONE)) {
196 dev_err(adap->pdev_dev,
197 "Unexpected L2T_WRITE_RPL status %u for entry %u\n",
202 if (tid & F_SYNC_WR) {
203 struct l2t_entry *e = &adap->l2t->l2tab[idx];
206 if (e->state != L2T_STATE_SWITCHING) {
207 send_pending(adap, e);
208 e->state = (e->neigh->nud_state & NUD_STALE) ?
209 L2T_STATE_STALE : L2T_STATE_VALID;
211 spin_unlock(&e->lock);
216 * Add a packet to an L2T entry's queue of packets awaiting resolution.
217 * Must be called with the entry's lock held.
219 static inline void arpq_enqueue(struct l2t_entry *e, struct sk_buff *skb)
223 e->arpq_tail->next = skb;
229 int cxgb4_l2t_send(struct net_device *dev, struct sk_buff *skb,
232 struct adapter *adap = netdev2adap(dev);
236 case L2T_STATE_STALE: /* entry is stale, kick off revalidation */
237 neigh_event_send(e->neigh, NULL);
238 spin_lock_bh(&e->lock);
239 if (e->state == L2T_STATE_STALE)
240 e->state = L2T_STATE_VALID;
241 spin_unlock_bh(&e->lock);
242 case L2T_STATE_VALID: /* fast-path, send the packet on */
243 return t4_ofld_send(adap, skb);
244 case L2T_STATE_RESOLVING:
245 case L2T_STATE_SYNC_WRITE:
246 spin_lock_bh(&e->lock);
247 if (e->state != L2T_STATE_SYNC_WRITE &&
248 e->state != L2T_STATE_RESOLVING) {
249 spin_unlock_bh(&e->lock);
252 arpq_enqueue(e, skb);
253 spin_unlock_bh(&e->lock);
255 if (e->state == L2T_STATE_RESOLVING &&
256 !neigh_event_send(e->neigh, NULL)) {
257 spin_lock_bh(&e->lock);
258 if (e->state == L2T_STATE_RESOLVING && e->arpq_head)
259 write_l2e(adap, e, 1);
260 spin_unlock_bh(&e->lock);
265 EXPORT_SYMBOL(cxgb4_l2t_send);
268 * Allocate a free L2T entry. Must be called with l2t_data.lock held.
270 static struct l2t_entry *alloc_l2e(struct l2t_data *d)
272 struct l2t_entry *end, *e, **p;
274 if (!atomic_read(&d->nfree))
277 /* there's definitely a free entry */
278 for (e = d->rover, end = &d->l2tab[L2T_SIZE]; e != end; ++e)
279 if (atomic_read(&e->refcnt) == 0)
282 for (e = d->l2tab; atomic_read(&e->refcnt); ++e)
286 atomic_dec(&d->nfree);
289 * The entry we found may be an inactive entry that is
290 * presently in the hash table. We need to remove it.
292 if (e->state < L2T_STATE_SWITCHING)
293 for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next)
300 e->state = L2T_STATE_UNUSED;
305 * Called when an L2T entry has no more users.
307 static void t4_l2e_free(struct l2t_entry *e)
311 spin_lock_bh(&e->lock);
312 if (atomic_read(&e->refcnt) == 0) { /* hasn't been recycled */
314 neigh_release(e->neigh);
317 while (e->arpq_head) {
318 struct sk_buff *skb = e->arpq_head;
320 e->arpq_head = skb->next;
325 spin_unlock_bh(&e->lock);
327 d = container_of(e, struct l2t_data, l2tab[e->idx]);
328 atomic_inc(&d->nfree);
331 void cxgb4_l2t_release(struct l2t_entry *e)
333 if (atomic_dec_and_test(&e->refcnt))
336 EXPORT_SYMBOL(cxgb4_l2t_release);
339 * Update an L2T entry that was previously used for the same next hop as neigh.
340 * Must be called with softirqs disabled.
342 static void reuse_entry(struct l2t_entry *e, struct neighbour *neigh)
344 unsigned int nud_state;
346 spin_lock(&e->lock); /* avoid race with t4_l2t_free */
347 if (neigh != e->neigh)
348 neigh_replace(e, neigh);
349 nud_state = neigh->nud_state;
350 if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)) ||
351 !(nud_state & NUD_VALID))
352 e->state = L2T_STATE_RESOLVING;
353 else if (nud_state & NUD_CONNECTED)
354 e->state = L2T_STATE_VALID;
356 e->state = L2T_STATE_STALE;
357 spin_unlock(&e->lock);
360 struct l2t_entry *cxgb4_l2t_get(struct l2t_data *d, struct neighbour *neigh,
361 const struct net_device *physdev,
362 unsigned int priority)
367 int addr_len = neigh->tbl->key_len;
368 u32 *addr = (u32 *)neigh->primary_key;
369 int ifidx = neigh->dev->ifindex;
370 int hash = addr_hash(addr, addr_len, ifidx);
372 if (neigh->dev->flags & IFF_LOOPBACK)
373 lport = netdev2pinfo(physdev)->tx_chan + 4;
375 lport = netdev2pinfo(physdev)->lport;
377 if (neigh->dev->priv_flags & IFF_802_1Q_VLAN)
378 vlan = vlan_dev_vlan_id(neigh->dev);
382 write_lock_bh(&d->lock);
383 for (e = d->l2tab[hash].first; e; e = e->next)
384 if (!addreq(e, addr) && e->ifindex == ifidx &&
385 e->vlan == vlan && e->lport == lport) {
387 if (atomic_read(&e->refcnt) == 1)
388 reuse_entry(e, neigh);
392 /* Need to allocate a new entry */
395 spin_lock(&e->lock); /* avoid race with t4_l2t_free */
396 e->state = L2T_STATE_RESOLVING;
397 if (neigh->dev->flags & IFF_LOOPBACK)
398 memcpy(e->dmac, physdev->dev_addr, sizeof(e->dmac));
399 memcpy(e->addr, addr, addr_len);
403 e->v6 = addr_len == 16;
404 atomic_set(&e->refcnt, 1);
405 neigh_replace(e, neigh);
407 e->next = d->l2tab[hash].first;
408 d->l2tab[hash].first = e;
409 spin_unlock(&e->lock);
412 write_unlock_bh(&d->lock);
415 EXPORT_SYMBOL(cxgb4_l2t_get);
417 u64 cxgb4_select_ntuple(struct net_device *dev,
418 const struct l2t_entry *l2t)
420 struct adapter *adap = netdev2adap(dev);
421 struct tp_params *tp = &adap->params.tp;
424 /* Initialize each of the fields which we care about which are present
425 * in the Compressed Filter Tuple.
427 if (tp->vlan_shift >= 0 && l2t->vlan != VLAN_NONE)
428 ntuple |= (u64)(F_FT_VLAN_VLD | l2t->vlan) << tp->vlan_shift;
430 if (tp->port_shift >= 0)
431 ntuple |= (u64)l2t->lport << tp->port_shift;
433 if (tp->protocol_shift >= 0)
434 ntuple |= (u64)IPPROTO_TCP << tp->protocol_shift;
436 if (tp->vnic_shift >= 0) {
437 u32 viid = cxgb4_port_viid(dev);
438 u32 vf = FW_VIID_VIN_GET(viid);
439 u32 pf = FW_VIID_PFN_GET(viid);
440 u32 vld = FW_VIID_VIVLD_GET(viid);
442 ntuple |= (u64)(V_FT_VNID_ID_VF(vf) |
443 V_FT_VNID_ID_PF(pf) |
444 V_FT_VNID_ID_VLD(vld)) << tp->vnic_shift;
449 EXPORT_SYMBOL(cxgb4_select_ntuple);
452 * Called when address resolution fails for an L2T entry to handle packets
453 * on the arpq head. If a packet specifies a failure handler it is invoked,
454 * otherwise the packet is sent to the device.
456 static void handle_failed_resolution(struct adapter *adap, struct sk_buff *arpq)
459 struct sk_buff *skb = arpq;
460 const struct l2t_skb_cb *cb = L2T_SKB_CB(skb);
464 if (cb->arp_err_handler)
465 cb->arp_err_handler(cb->handle, skb);
467 t4_ofld_send(adap, skb);
472 * Called when the host's neighbor layer makes a change to some entry that is
473 * loaded into the HW L2 table.
475 void t4_l2t_update(struct adapter *adap, struct neighbour *neigh)
478 struct sk_buff *arpq = NULL;
479 struct l2t_data *d = adap->l2t;
480 int addr_len = neigh->tbl->key_len;
481 u32 *addr = (u32 *) neigh->primary_key;
482 int ifidx = neigh->dev->ifindex;
483 int hash = addr_hash(addr, addr_len, ifidx);
485 read_lock_bh(&d->lock);
486 for (e = d->l2tab[hash].first; e; e = e->next)
487 if (!addreq(e, addr) && e->ifindex == ifidx) {
489 if (atomic_read(&e->refcnt))
491 spin_unlock(&e->lock);
494 read_unlock_bh(&d->lock);
498 read_unlock(&d->lock);
500 if (neigh != e->neigh)
501 neigh_replace(e, neigh);
503 if (e->state == L2T_STATE_RESOLVING) {
504 if (neigh->nud_state & NUD_FAILED) {
506 e->arpq_head = e->arpq_tail = NULL;
507 } else if ((neigh->nud_state & (NUD_CONNECTED | NUD_STALE)) &&
509 write_l2e(adap, e, 1);
512 e->state = neigh->nud_state & NUD_CONNECTED ?
513 L2T_STATE_VALID : L2T_STATE_STALE;
514 if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)))
515 write_l2e(adap, e, 0);
518 spin_unlock_bh(&e->lock);
521 handle_failed_resolution(adap, arpq);
524 /* Allocate an L2T entry for use by a switching rule. Such need to be
525 * explicitly freed and while busy they are not on any hash chain, so normal
526 * address resolution updates do not see them.
528 struct l2t_entry *t4_l2t_alloc_switching(struct l2t_data *d)
532 write_lock_bh(&d->lock);
535 spin_lock(&e->lock); /* avoid race with t4_l2t_free */
536 e->state = L2T_STATE_SWITCHING;
537 atomic_set(&e->refcnt, 1);
538 spin_unlock(&e->lock);
540 write_unlock_bh(&d->lock);
544 /* Sets/updates the contents of a switching L2T entry that has been allocated
545 * with an earlier call to @t4_l2t_alloc_switching.
547 int t4_l2t_set_switching(struct adapter *adap, struct l2t_entry *e, u16 vlan,
548 u8 port, u8 *eth_addr)
552 memcpy(e->dmac, eth_addr, ETH_ALEN);
553 return write_l2e(adap, e, 0);
556 struct l2t_data *t4_init_l2t(void)
561 d = t4_alloc_mem(sizeof(*d));
566 atomic_set(&d->nfree, L2T_SIZE);
567 rwlock_init(&d->lock);
569 for (i = 0; i < L2T_SIZE; ++i) {
571 d->l2tab[i].state = L2T_STATE_UNUSED;
572 spin_lock_init(&d->l2tab[i].lock);
573 atomic_set(&d->l2tab[i].refcnt, 0);
578 static inline void *l2t_get_idx(struct seq_file *seq, loff_t pos)
580 struct l2t_entry *l2tab = seq->private;
582 return pos >= L2T_SIZE ? NULL : &l2tab[pos];
585 static void *l2t_seq_start(struct seq_file *seq, loff_t *pos)
587 return *pos ? l2t_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
590 static void *l2t_seq_next(struct seq_file *seq, void *v, loff_t *pos)
592 v = l2t_get_idx(seq, *pos);
598 static void l2t_seq_stop(struct seq_file *seq, void *v)
602 static char l2e_state(const struct l2t_entry *e)
605 case L2T_STATE_VALID: return 'V';
606 case L2T_STATE_STALE: return 'S';
607 case L2T_STATE_SYNC_WRITE: return 'W';
608 case L2T_STATE_RESOLVING: return e->arpq_head ? 'A' : 'R';
609 case L2T_STATE_SWITCHING: return 'X';
615 static int l2t_seq_show(struct seq_file *seq, void *v)
617 if (v == SEQ_START_TOKEN)
618 seq_puts(seq, " Idx IP address "
619 "Ethernet address VLAN/P LP State Users Port\n");
622 struct l2t_entry *e = v;
624 spin_lock_bh(&e->lock);
625 if (e->state == L2T_STATE_SWITCHING)
628 sprintf(ip, e->v6 ? "%pI6c" : "%pI4", e->addr);
629 seq_printf(seq, "%4u %-25s %17pM %4d %u %2u %c %5u %s\n",
631 e->vlan & VLAN_VID_MASK, vlan_prio(e), e->lport,
632 l2e_state(e), atomic_read(&e->refcnt),
633 e->neigh ? e->neigh->dev->name : "");
634 spin_unlock_bh(&e->lock);
639 static const struct seq_operations l2t_seq_ops = {
640 .start = l2t_seq_start,
641 .next = l2t_seq_next,
642 .stop = l2t_seq_stop,
646 static int l2t_seq_open(struct inode *inode, struct file *file)
648 int rc = seq_open(file, &l2t_seq_ops);
651 struct adapter *adap = inode->i_private;
652 struct seq_file *seq = file->private_data;
654 seq->private = adap->l2t->l2tab;
659 const struct file_operations t4_l2t_fops = {
660 .owner = THIS_MODULE,
661 .open = l2t_seq_open,
664 .release = seq_release,