2 * This file is part of the Chelsio T4 Ethernet driver for Linux.
4 * Copyright (c) 2003-2014 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>
49 #include "t4_values.h"
51 #define VLAN_NONE 0xfff
53 /* identifies sync vs async L2T_WRITE_REQs */
55 #define SYNC_WR_V(x) ((x) << SYNC_WR_S)
56 #define SYNC_WR_F SYNC_WR_V(1)
59 unsigned int l2t_start; /* start index of our piece of the L2T */
60 unsigned int l2t_size; /* number of entries in l2tab */
62 atomic_t nfree; /* number of free entries */
63 struct l2t_entry *rover; /* starting point for next allocation */
64 struct l2t_entry l2tab[0]; /* MUST BE LAST */
67 static inline unsigned int vlan_prio(const struct l2t_entry *e)
69 return e->vlan >> VLAN_PRIO_SHIFT;
72 static inline void l2t_hold(struct l2t_data *d, struct l2t_entry *e)
74 if (atomic_add_return(1, &e->refcnt) == 1) /* 0 -> 1 transition */
75 atomic_dec(&d->nfree);
79 * To avoid having to check address families we do not allow v4 and v6
80 * neighbors to be on the same hash chain. We keep v4 entries in the first
81 * half of available hash buckets and v6 in the second. We need at least two
82 * entries in our L2T for this scheme to work.
85 L2T_MIN_HASH_BUCKETS = 2,
88 static inline unsigned int arp_hash(struct l2t_data *d, const u32 *key,
91 unsigned int l2t_size_half = d->l2t_size / 2;
93 return jhash_2words(*key, ifindex, 0) % l2t_size_half;
96 static inline unsigned int ipv6_hash(struct l2t_data *d, const u32 *key,
99 unsigned int l2t_size_half = d->l2t_size / 2;
100 u32 xor = key[0] ^ key[1] ^ key[2] ^ key[3];
102 return (l2t_size_half +
103 (jhash_2words(xor, ifindex, 0) % l2t_size_half));
106 static unsigned int addr_hash(struct l2t_data *d, const u32 *addr,
107 int addr_len, int ifindex)
109 return addr_len == 4 ? arp_hash(d, addr, ifindex) :
110 ipv6_hash(d, addr, ifindex);
114 * Checks if an L2T entry is for the given IP/IPv6 address. It does not check
115 * whether the L2T entry and the address are of the same address family.
116 * Callers ensure an address is only checked against L2T entries of the same
117 * family, something made trivial by the separation of IP and IPv6 hash chains
118 * mentioned above. Returns 0 if there's a match,
120 static int addreq(const struct l2t_entry *e, const u32 *addr)
123 return (e->addr[0] ^ addr[0]) | (e->addr[1] ^ addr[1]) |
124 (e->addr[2] ^ addr[2]) | (e->addr[3] ^ addr[3]);
125 return e->addr[0] ^ addr[0];
128 static void neigh_replace(struct l2t_entry *e, struct neighbour *n)
132 neigh_release(e->neigh);
137 * Write an L2T entry. Must be called with the entry locked.
138 * The write may be synchronous or asynchronous.
140 static int write_l2e(struct adapter *adap, struct l2t_entry *e, int sync)
142 struct l2t_data *d = adap->l2t;
143 unsigned int l2t_idx = e->idx + d->l2t_start;
145 struct cpl_l2t_write_req *req;
147 skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
151 req = (struct cpl_l2t_write_req *)__skb_put(skb, sizeof(*req));
154 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ,
155 l2t_idx | (sync ? SYNC_WR_F : 0) |
156 TID_QID_V(adap->sge.fw_evtq.abs_id)));
157 req->params = htons(L2T_W_PORT_V(e->lport) | L2T_W_NOREPLY_V(!sync));
158 req->l2t_idx = htons(l2t_idx);
159 req->vlan = htons(e->vlan);
160 if (e->neigh && !(e->neigh->dev->flags & IFF_LOOPBACK))
161 memcpy(e->dmac, e->neigh->ha, sizeof(e->dmac));
162 memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
164 t4_mgmt_tx(adap, skb);
166 if (sync && e->state != L2T_STATE_SWITCHING)
167 e->state = L2T_STATE_SYNC_WRITE;
172 * Send packets waiting in an L2T entry's ARP queue. Must be called with the
175 static void send_pending(struct adapter *adap, struct l2t_entry *e)
177 while (e->arpq_head) {
178 struct sk_buff *skb = e->arpq_head;
180 e->arpq_head = skb->next;
182 t4_ofld_send(adap, skb);
188 * Process a CPL_L2T_WRITE_RPL. Wake up the ARP queue if it completes a
189 * synchronous L2T_WRITE. Note that the TID in the reply is really the L2T
190 * index it refers to.
192 void do_l2t_write_rpl(struct adapter *adap, const struct cpl_l2t_write_rpl *rpl)
194 struct l2t_data *d = adap->l2t;
195 unsigned int tid = GET_TID(rpl);
196 unsigned int l2t_idx = tid % L2T_SIZE;
198 if (unlikely(rpl->status != CPL_ERR_NONE)) {
199 dev_err(adap->pdev_dev,
200 "Unexpected L2T_WRITE_RPL status %u for entry %u\n",
201 rpl->status, l2t_idx);
205 if (tid & SYNC_WR_F) {
206 struct l2t_entry *e = &d->l2tab[l2t_idx - d->l2t_start];
209 if (e->state != L2T_STATE_SWITCHING) {
210 send_pending(adap, e);
211 e->state = (e->neigh->nud_state & NUD_STALE) ?
212 L2T_STATE_STALE : L2T_STATE_VALID;
214 spin_unlock(&e->lock);
219 * Add a packet to an L2T entry's queue of packets awaiting resolution.
220 * Must be called with the entry's lock held.
222 static inline void arpq_enqueue(struct l2t_entry *e, struct sk_buff *skb)
226 e->arpq_tail->next = skb;
232 int cxgb4_l2t_send(struct net_device *dev, struct sk_buff *skb,
235 struct adapter *adap = netdev2adap(dev);
239 case L2T_STATE_STALE: /* entry is stale, kick off revalidation */
240 neigh_event_send(e->neigh, NULL);
241 spin_lock_bh(&e->lock);
242 if (e->state == L2T_STATE_STALE)
243 e->state = L2T_STATE_VALID;
244 spin_unlock_bh(&e->lock);
245 case L2T_STATE_VALID: /* fast-path, send the packet on */
246 return t4_ofld_send(adap, skb);
247 case L2T_STATE_RESOLVING:
248 case L2T_STATE_SYNC_WRITE:
249 spin_lock_bh(&e->lock);
250 if (e->state != L2T_STATE_SYNC_WRITE &&
251 e->state != L2T_STATE_RESOLVING) {
252 spin_unlock_bh(&e->lock);
255 arpq_enqueue(e, skb);
256 spin_unlock_bh(&e->lock);
258 if (e->state == L2T_STATE_RESOLVING &&
259 !neigh_event_send(e->neigh, NULL)) {
260 spin_lock_bh(&e->lock);
261 if (e->state == L2T_STATE_RESOLVING && e->arpq_head)
262 write_l2e(adap, e, 1);
263 spin_unlock_bh(&e->lock);
268 EXPORT_SYMBOL(cxgb4_l2t_send);
271 * Allocate a free L2T entry. Must be called with l2t_data.lock held.
273 static struct l2t_entry *alloc_l2e(struct l2t_data *d)
275 struct l2t_entry *end, *e, **p;
277 if (!atomic_read(&d->nfree))
280 /* there's definitely a free entry */
281 for (e = d->rover, end = &d->l2tab[d->l2t_size]; e != end; ++e)
282 if (atomic_read(&e->refcnt) == 0)
285 for (e = d->l2tab; atomic_read(&e->refcnt); ++e)
289 atomic_dec(&d->nfree);
292 * The entry we found may be an inactive entry that is
293 * presently in the hash table. We need to remove it.
295 if (e->state < L2T_STATE_SWITCHING)
296 for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next)
303 e->state = L2T_STATE_UNUSED;
307 static struct l2t_entry *find_or_alloc_l2e(struct l2t_data *d, u16 vlan,
310 struct l2t_entry *end, *e, **p;
311 struct l2t_entry *first_free = NULL;
313 for (e = &d->l2tab[0], end = &d->l2tab[d->l2t_size]; e != end; ++e) {
314 if (atomic_read(&e->refcnt) == 0) {
318 if (e->state == L2T_STATE_SWITCHING) {
319 if (ether_addr_equal(e->dmac, dmac) &&
320 (e->vlan == vlan) && (e->lport == port))
334 /* The entry we found may be an inactive entry that is
335 * presently in the hash table. We need to remove it.
337 if (e->state < L2T_STATE_SWITCHING)
338 for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next)
344 e->state = L2T_STATE_UNUSED;
350 /* Called when an L2T entry has no more users. The entry is left in the hash
351 * table since it is likely to be reused but we also bump nfree to indicate
352 * that the entry can be reallocated for a different neighbor. We also drop
353 * the existing neighbor reference in case the neighbor is going away and is
354 * waiting on our reference.
356 * Because entries can be reallocated to other neighbors once their ref count
357 * drops to 0 we need to take the entry's lock to avoid races with a new
360 static void _t4_l2e_free(struct l2t_entry *e)
364 if (atomic_read(&e->refcnt) == 0) { /* hasn't been recycled */
366 neigh_release(e->neigh);
369 while (e->arpq_head) {
370 struct sk_buff *skb = e->arpq_head;
372 e->arpq_head = skb->next;
378 d = container_of(e, struct l2t_data, l2tab[e->idx]);
379 atomic_inc(&d->nfree);
382 /* Locked version of _t4_l2e_free */
383 static void t4_l2e_free(struct l2t_entry *e)
387 spin_lock_bh(&e->lock);
388 if (atomic_read(&e->refcnt) == 0) { /* hasn't been recycled */
390 neigh_release(e->neigh);
393 while (e->arpq_head) {
394 struct sk_buff *skb = e->arpq_head;
396 e->arpq_head = skb->next;
401 spin_unlock_bh(&e->lock);
403 d = container_of(e, struct l2t_data, l2tab[e->idx]);
404 atomic_inc(&d->nfree);
407 void cxgb4_l2t_release(struct l2t_entry *e)
409 if (atomic_dec_and_test(&e->refcnt))
412 EXPORT_SYMBOL(cxgb4_l2t_release);
415 * Update an L2T entry that was previously used for the same next hop as neigh.
416 * Must be called with softirqs disabled.
418 static void reuse_entry(struct l2t_entry *e, struct neighbour *neigh)
420 unsigned int nud_state;
422 spin_lock(&e->lock); /* avoid race with t4_l2t_free */
423 if (neigh != e->neigh)
424 neigh_replace(e, neigh);
425 nud_state = neigh->nud_state;
426 if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)) ||
427 !(nud_state & NUD_VALID))
428 e->state = L2T_STATE_RESOLVING;
429 else if (nud_state & NUD_CONNECTED)
430 e->state = L2T_STATE_VALID;
432 e->state = L2T_STATE_STALE;
433 spin_unlock(&e->lock);
436 struct l2t_entry *cxgb4_l2t_get(struct l2t_data *d, struct neighbour *neigh,
437 const struct net_device *physdev,
438 unsigned int priority)
443 int addr_len = neigh->tbl->key_len;
444 u32 *addr = (u32 *)neigh->primary_key;
445 int ifidx = neigh->dev->ifindex;
446 int hash = addr_hash(d, addr, addr_len, ifidx);
448 if (neigh->dev->flags & IFF_LOOPBACK)
449 lport = netdev2pinfo(physdev)->tx_chan + 4;
451 lport = netdev2pinfo(physdev)->lport;
453 if (neigh->dev->priv_flags & IFF_802_1Q_VLAN)
454 vlan = vlan_dev_vlan_id(neigh->dev);
458 write_lock_bh(&d->lock);
459 for (e = d->l2tab[hash].first; e; e = e->next)
460 if (!addreq(e, addr) && e->ifindex == ifidx &&
461 e->vlan == vlan && e->lport == lport) {
463 if (atomic_read(&e->refcnt) == 1)
464 reuse_entry(e, neigh);
468 /* Need to allocate a new entry */
471 spin_lock(&e->lock); /* avoid race with t4_l2t_free */
472 e->state = L2T_STATE_RESOLVING;
473 if (neigh->dev->flags & IFF_LOOPBACK)
474 memcpy(e->dmac, physdev->dev_addr, sizeof(e->dmac));
475 memcpy(e->addr, addr, addr_len);
479 e->v6 = addr_len == 16;
480 atomic_set(&e->refcnt, 1);
481 neigh_replace(e, neigh);
483 e->next = d->l2tab[hash].first;
484 d->l2tab[hash].first = e;
485 spin_unlock(&e->lock);
488 write_unlock_bh(&d->lock);
491 EXPORT_SYMBOL(cxgb4_l2t_get);
493 u64 cxgb4_select_ntuple(struct net_device *dev,
494 const struct l2t_entry *l2t)
496 struct adapter *adap = netdev2adap(dev);
497 struct tp_params *tp = &adap->params.tp;
500 /* Initialize each of the fields which we care about which are present
501 * in the Compressed Filter Tuple.
503 if (tp->vlan_shift >= 0 && l2t->vlan != VLAN_NONE)
504 ntuple |= (u64)(FT_VLAN_VLD_F | l2t->vlan) << tp->vlan_shift;
506 if (tp->port_shift >= 0)
507 ntuple |= (u64)l2t->lport << tp->port_shift;
509 if (tp->protocol_shift >= 0)
510 ntuple |= (u64)IPPROTO_TCP << tp->protocol_shift;
512 if (tp->vnic_shift >= 0) {
513 u32 viid = cxgb4_port_viid(dev);
514 u32 vf = FW_VIID_VIN_G(viid);
515 u32 pf = FW_VIID_PFN_G(viid);
516 u32 vld = FW_VIID_VIVLD_G(viid);
518 ntuple |= (u64)(FT_VNID_ID_VF_V(vf) |
519 FT_VNID_ID_PF_V(pf) |
520 FT_VNID_ID_VLD_V(vld)) << tp->vnic_shift;
525 EXPORT_SYMBOL(cxgb4_select_ntuple);
528 * Called when address resolution fails for an L2T entry to handle packets
529 * on the arpq head. If a packet specifies a failure handler it is invoked,
530 * otherwise the packet is sent to the device.
532 static void handle_failed_resolution(struct adapter *adap, struct sk_buff *arpq)
535 struct sk_buff *skb = arpq;
536 const struct l2t_skb_cb *cb = L2T_SKB_CB(skb);
540 if (cb->arp_err_handler)
541 cb->arp_err_handler(cb->handle, skb);
543 t4_ofld_send(adap, skb);
548 * Called when the host's neighbor layer makes a change to some entry that is
549 * loaded into the HW L2 table.
551 void t4_l2t_update(struct adapter *adap, struct neighbour *neigh)
554 struct sk_buff *arpq = NULL;
555 struct l2t_data *d = adap->l2t;
556 int addr_len = neigh->tbl->key_len;
557 u32 *addr = (u32 *) neigh->primary_key;
558 int ifidx = neigh->dev->ifindex;
559 int hash = addr_hash(d, addr, addr_len, ifidx);
561 read_lock_bh(&d->lock);
562 for (e = d->l2tab[hash].first; e; e = e->next)
563 if (!addreq(e, addr) && e->ifindex == ifidx) {
565 if (atomic_read(&e->refcnt))
567 spin_unlock(&e->lock);
570 read_unlock_bh(&d->lock);
574 read_unlock(&d->lock);
576 if (neigh != e->neigh)
577 neigh_replace(e, neigh);
579 if (e->state == L2T_STATE_RESOLVING) {
580 if (neigh->nud_state & NUD_FAILED) {
582 e->arpq_head = e->arpq_tail = NULL;
583 } else if ((neigh->nud_state & (NUD_CONNECTED | NUD_STALE)) &&
585 write_l2e(adap, e, 1);
588 e->state = neigh->nud_state & NUD_CONNECTED ?
589 L2T_STATE_VALID : L2T_STATE_STALE;
590 if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)))
591 write_l2e(adap, e, 0);
594 spin_unlock_bh(&e->lock);
597 handle_failed_resolution(adap, arpq);
600 /* Allocate an L2T entry for use by a switching rule. Such need to be
601 * explicitly freed and while busy they are not on any hash chain, so normal
602 * address resolution updates do not see them.
604 struct l2t_entry *t4_l2t_alloc_switching(struct adapter *adap, u16 vlan,
605 u8 port, u8 *eth_addr)
607 struct l2t_data *d = adap->l2t;
611 write_lock_bh(&d->lock);
612 e = find_or_alloc_l2e(d, vlan, port, eth_addr);
614 spin_lock(&e->lock); /* avoid race with t4_l2t_free */
615 if (!atomic_read(&e->refcnt)) {
616 e->state = L2T_STATE_SWITCHING;
619 ether_addr_copy(e->dmac, eth_addr);
620 atomic_set(&e->refcnt, 1);
621 ret = write_l2e(adap, e, 0);
624 spin_unlock(&e->lock);
625 write_unlock_bh(&d->lock);
629 atomic_inc(&e->refcnt);
632 spin_unlock(&e->lock);
634 write_unlock_bh(&d->lock);
639 * @dev: net_device pointer
641 * @port: Associated port
642 * @dmac: Destination MAC address to add to L2T
643 * Returns pointer to the allocated l2t entry
645 * Allocates an L2T entry for use by switching rule of a filter
647 struct l2t_entry *cxgb4_l2t_alloc_switching(struct net_device *dev, u16 vlan,
650 struct adapter *adap = netdev2adap(dev);
652 return t4_l2t_alloc_switching(adap, vlan, port, dmac);
654 EXPORT_SYMBOL(cxgb4_l2t_alloc_switching);
656 struct l2t_data *t4_init_l2t(unsigned int l2t_start, unsigned int l2t_end)
658 unsigned int l2t_size;
662 if (l2t_start >= l2t_end || l2t_end >= L2T_SIZE)
664 l2t_size = l2t_end - l2t_start + 1;
665 if (l2t_size < L2T_MIN_HASH_BUCKETS)
668 d = t4_alloc_mem(sizeof(*d) + l2t_size * sizeof(struct l2t_entry));
672 d->l2t_start = l2t_start;
673 d->l2t_size = l2t_size;
676 atomic_set(&d->nfree, l2t_size);
677 rwlock_init(&d->lock);
679 for (i = 0; i < d->l2t_size; ++i) {
681 d->l2tab[i].state = L2T_STATE_UNUSED;
682 spin_lock_init(&d->l2tab[i].lock);
683 atomic_set(&d->l2tab[i].refcnt, 0);
688 static inline void *l2t_get_idx(struct seq_file *seq, loff_t pos)
690 struct l2t_data *d = seq->private;
692 return pos >= d->l2t_size ? NULL : &d->l2tab[pos];
695 static void *l2t_seq_start(struct seq_file *seq, loff_t *pos)
697 return *pos ? l2t_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
700 static void *l2t_seq_next(struct seq_file *seq, void *v, loff_t *pos)
702 v = l2t_get_idx(seq, *pos);
708 static void l2t_seq_stop(struct seq_file *seq, void *v)
712 static char l2e_state(const struct l2t_entry *e)
715 case L2T_STATE_VALID: return 'V';
716 case L2T_STATE_STALE: return 'S';
717 case L2T_STATE_SYNC_WRITE: return 'W';
718 case L2T_STATE_RESOLVING: return e->arpq_head ? 'A' : 'R';
719 case L2T_STATE_SWITCHING: return 'X';
725 static int l2t_seq_show(struct seq_file *seq, void *v)
727 if (v == SEQ_START_TOKEN)
728 seq_puts(seq, " Idx IP address "
729 "Ethernet address VLAN/P LP State Users Port\n");
732 struct l2t_data *d = seq->private;
733 struct l2t_entry *e = v;
735 spin_lock_bh(&e->lock);
736 if (e->state == L2T_STATE_SWITCHING)
739 sprintf(ip, e->v6 ? "%pI6c" : "%pI4", e->addr);
740 seq_printf(seq, "%4u %-25s %17pM %4d %u %2u %c %5u %s\n",
741 e->idx + d->l2t_start, ip, e->dmac,
742 e->vlan & VLAN_VID_MASK, vlan_prio(e), e->lport,
743 l2e_state(e), atomic_read(&e->refcnt),
744 e->neigh ? e->neigh->dev->name : "");
745 spin_unlock_bh(&e->lock);
750 static const struct seq_operations l2t_seq_ops = {
751 .start = l2t_seq_start,
752 .next = l2t_seq_next,
753 .stop = l2t_seq_stop,
757 static int l2t_seq_open(struct inode *inode, struct file *file)
759 int rc = seq_open(file, &l2t_seq_ops);
762 struct adapter *adap = inode->i_private;
763 struct seq_file *seq = file->private_data;
765 seq->private = adap->l2t;
770 const struct file_operations t4_l2t_fops = {
771 .owner = THIS_MODULE,
772 .open = l2t_seq_open,
775 .release = seq_release,