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 */
54 #define F_SYNC_WR (1 << 12)
57 L2T_STATE_VALID, /* entry is up to date */
58 L2T_STATE_STALE, /* entry may be used but needs revalidation */
59 L2T_STATE_RESOLVING, /* entry needs address resolution */
60 L2T_STATE_SYNC_WRITE, /* synchronous write of entry underway */
62 /* when state is one of the below the entry is not hashed */
63 L2T_STATE_SWITCHING, /* entry is being used by a switching filter */
64 L2T_STATE_UNUSED /* entry not in use */
69 atomic_t nfree; /* number of free entries */
70 struct l2t_entry *rover; /* starting point for next allocation */
71 struct l2t_entry l2tab[L2T_SIZE];
74 static inline unsigned int vlan_prio(const struct l2t_entry *e)
79 static inline void l2t_hold(struct l2t_data *d, struct l2t_entry *e)
81 if (atomic_add_return(1, &e->refcnt) == 1) /* 0 -> 1 transition */
82 atomic_dec(&d->nfree);
86 * To avoid having to check address families we do not allow v4 and v6
87 * neighbors to be on the same hash chain. We keep v4 entries in the first
88 * half of available hash buckets and v6 in the second.
91 L2T_SZ_HALF = L2T_SIZE / 2,
92 L2T_HASH_MASK = L2T_SZ_HALF - 1
95 static inline unsigned int arp_hash(const u32 *key, int ifindex)
97 return jhash_2words(*key, ifindex, 0) & L2T_HASH_MASK;
100 static inline unsigned int ipv6_hash(const u32 *key, int ifindex)
102 u32 xor = key[0] ^ key[1] ^ key[2] ^ key[3];
104 return L2T_SZ_HALF + (jhash_2words(xor, ifindex, 0) & L2T_HASH_MASK);
107 static unsigned int addr_hash(const u32 *addr, int addr_len, int ifindex)
109 return addr_len == 4 ? arp_hash(addr, ifindex) :
110 ipv6_hash(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)
143 struct cpl_l2t_write_req *req;
145 skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
149 req = (struct cpl_l2t_write_req *)__skb_put(skb, sizeof(*req));
152 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ,
153 e->idx | (sync ? F_SYNC_WR : 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(e->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));
162 set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
163 t4_ofld_send(adap, skb);
165 if (sync && e->state != L2T_STATE_SWITCHING)
166 e->state = L2T_STATE_SYNC_WRITE;
171 * Send packets waiting in an L2T entry's ARP queue. Must be called with the
174 static void send_pending(struct adapter *adap, struct l2t_entry *e)
176 while (e->arpq_head) {
177 struct sk_buff *skb = e->arpq_head;
179 e->arpq_head = skb->next;
181 t4_ofld_send(adap, skb);
187 * Process a CPL_L2T_WRITE_RPL. Wake up the ARP queue if it completes a
188 * synchronous L2T_WRITE. Note that the TID in the reply is really the L2T
189 * index it refers to.
191 void do_l2t_write_rpl(struct adapter *adap, const struct cpl_l2t_write_rpl *rpl)
193 unsigned int tid = GET_TID(rpl);
194 unsigned int idx = tid & (L2T_SIZE - 1);
196 if (unlikely(rpl->status != CPL_ERR_NONE)) {
197 dev_err(adap->pdev_dev,
198 "Unexpected L2T_WRITE_RPL status %u for entry %u\n",
203 if (tid & F_SYNC_WR) {
204 struct l2t_entry *e = &adap->l2t->l2tab[idx];
207 if (e->state != L2T_STATE_SWITCHING) {
208 send_pending(adap, e);
209 e->state = (e->neigh->nud_state & NUD_STALE) ?
210 L2T_STATE_STALE : L2T_STATE_VALID;
212 spin_unlock(&e->lock);
217 * Add a packet to an L2T entry's queue of packets awaiting resolution.
218 * Must be called with the entry's lock held.
220 static inline void arpq_enqueue(struct l2t_entry *e, struct sk_buff *skb)
224 e->arpq_tail->next = skb;
230 int cxgb4_l2t_send(struct net_device *dev, struct sk_buff *skb,
233 struct adapter *adap = netdev2adap(dev);
237 case L2T_STATE_STALE: /* entry is stale, kick off revalidation */
238 neigh_event_send(e->neigh, NULL);
239 spin_lock_bh(&e->lock);
240 if (e->state == L2T_STATE_STALE)
241 e->state = L2T_STATE_VALID;
242 spin_unlock_bh(&e->lock);
243 case L2T_STATE_VALID: /* fast-path, send the packet on */
244 return t4_ofld_send(adap, skb);
245 case L2T_STATE_RESOLVING:
246 case L2T_STATE_SYNC_WRITE:
247 spin_lock_bh(&e->lock);
248 if (e->state != L2T_STATE_SYNC_WRITE &&
249 e->state != L2T_STATE_RESOLVING) {
250 spin_unlock_bh(&e->lock);
253 arpq_enqueue(e, skb);
254 spin_unlock_bh(&e->lock);
256 if (e->state == L2T_STATE_RESOLVING &&
257 !neigh_event_send(e->neigh, NULL)) {
258 spin_lock_bh(&e->lock);
259 if (e->state == L2T_STATE_RESOLVING && e->arpq_head)
260 write_l2e(adap, e, 1);
261 spin_unlock_bh(&e->lock);
266 EXPORT_SYMBOL(cxgb4_l2t_send);
269 * Allocate a free L2T entry. Must be called with l2t_data.lock held.
271 static struct l2t_entry *alloc_l2e(struct l2t_data *d)
273 struct l2t_entry *end, *e, **p;
275 if (!atomic_read(&d->nfree))
278 /* there's definitely a free entry */
279 for (e = d->rover, end = &d->l2tab[L2T_SIZE]; e != end; ++e)
280 if (atomic_read(&e->refcnt) == 0)
283 for (e = d->l2tab; atomic_read(&e->refcnt); ++e)
287 atomic_dec(&d->nfree);
290 * The entry we found may be an inactive entry that is
291 * presently in the hash table. We need to remove it.
293 if (e->state < L2T_STATE_SWITCHING)
294 for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next)
301 e->state = L2T_STATE_UNUSED;
306 * Called when an L2T entry has no more users.
308 static void t4_l2e_free(struct l2t_entry *e)
312 spin_lock_bh(&e->lock);
313 if (atomic_read(&e->refcnt) == 0) { /* hasn't been recycled */
315 neigh_release(e->neigh);
318 while (e->arpq_head) {
319 struct sk_buff *skb = e->arpq_head;
321 e->arpq_head = skb->next;
326 spin_unlock_bh(&e->lock);
328 d = container_of(e, struct l2t_data, l2tab[e->idx]);
329 atomic_inc(&d->nfree);
332 void cxgb4_l2t_release(struct l2t_entry *e)
334 if (atomic_dec_and_test(&e->refcnt))
337 EXPORT_SYMBOL(cxgb4_l2t_release);
340 * Update an L2T entry that was previously used for the same next hop as neigh.
341 * Must be called with softirqs disabled.
343 static void reuse_entry(struct l2t_entry *e, struct neighbour *neigh)
345 unsigned int nud_state;
347 spin_lock(&e->lock); /* avoid race with t4_l2t_free */
348 if (neigh != e->neigh)
349 neigh_replace(e, neigh);
350 nud_state = neigh->nud_state;
351 if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)) ||
352 !(nud_state & NUD_VALID))
353 e->state = L2T_STATE_RESOLVING;
354 else if (nud_state & NUD_CONNECTED)
355 e->state = L2T_STATE_VALID;
357 e->state = L2T_STATE_STALE;
358 spin_unlock(&e->lock);
361 struct l2t_entry *cxgb4_l2t_get(struct l2t_data *d, struct neighbour *neigh,
362 const struct net_device *physdev,
363 unsigned int priority)
368 int addr_len = neigh->tbl->key_len;
369 u32 *addr = (u32 *)neigh->primary_key;
370 int ifidx = neigh->dev->ifindex;
371 int hash = addr_hash(addr, addr_len, ifidx);
373 if (neigh->dev->flags & IFF_LOOPBACK)
374 lport = netdev2pinfo(physdev)->tx_chan + 4;
376 lport = netdev2pinfo(physdev)->lport;
378 if (neigh->dev->priv_flags & IFF_802_1Q_VLAN)
379 vlan = vlan_dev_vlan_id(neigh->dev);
383 write_lock_bh(&d->lock);
384 for (e = d->l2tab[hash].first; e; e = e->next)
385 if (!addreq(e, addr) && e->ifindex == ifidx &&
386 e->vlan == vlan && e->lport == lport) {
388 if (atomic_read(&e->refcnt) == 1)
389 reuse_entry(e, neigh);
393 /* Need to allocate a new entry */
396 spin_lock(&e->lock); /* avoid race with t4_l2t_free */
397 e->state = L2T_STATE_RESOLVING;
398 if (neigh->dev->flags & IFF_LOOPBACK)
399 memcpy(e->dmac, physdev->dev_addr, sizeof(e->dmac));
400 memcpy(e->addr, addr, addr_len);
404 e->v6 = addr_len == 16;
405 atomic_set(&e->refcnt, 1);
406 neigh_replace(e, neigh);
408 e->next = d->l2tab[hash].first;
409 d->l2tab[hash].first = e;
410 spin_unlock(&e->lock);
413 write_unlock_bh(&d->lock);
416 EXPORT_SYMBOL(cxgb4_l2t_get);
418 u64 cxgb4_select_ntuple(struct net_device *dev,
419 const struct l2t_entry *l2t)
421 struct adapter *adap = netdev2adap(dev);
422 struct tp_params *tp = &adap->params.tp;
425 /* Initialize each of the fields which we care about which are present
426 * in the Compressed Filter Tuple.
428 if (tp->vlan_shift >= 0 && l2t->vlan != VLAN_NONE)
429 ntuple |= (u64)(FT_VLAN_VLD_F | l2t->vlan) << tp->vlan_shift;
431 if (tp->port_shift >= 0)
432 ntuple |= (u64)l2t->lport << tp->port_shift;
434 if (tp->protocol_shift >= 0)
435 ntuple |= (u64)IPPROTO_TCP << tp->protocol_shift;
437 if (tp->vnic_shift >= 0) {
438 u32 viid = cxgb4_port_viid(dev);
439 u32 vf = FW_VIID_VIN_G(viid);
440 u32 pf = FW_VIID_PFN_G(viid);
441 u32 vld = FW_VIID_VIVLD_G(viid);
443 ntuple |= (u64)(FT_VNID_ID_VF_V(vf) |
444 FT_VNID_ID_PF_V(pf) |
445 FT_VNID_ID_VLD_V(vld)) << tp->vnic_shift;
450 EXPORT_SYMBOL(cxgb4_select_ntuple);
453 * Called when address resolution fails for an L2T entry to handle packets
454 * on the arpq head. If a packet specifies a failure handler it is invoked,
455 * otherwise the packet is sent to the device.
457 static void handle_failed_resolution(struct adapter *adap, struct sk_buff *arpq)
460 struct sk_buff *skb = arpq;
461 const struct l2t_skb_cb *cb = L2T_SKB_CB(skb);
465 if (cb->arp_err_handler)
466 cb->arp_err_handler(cb->handle, skb);
468 t4_ofld_send(adap, skb);
473 * Called when the host's neighbor layer makes a change to some entry that is
474 * loaded into the HW L2 table.
476 void t4_l2t_update(struct adapter *adap, struct neighbour *neigh)
479 struct sk_buff *arpq = NULL;
480 struct l2t_data *d = adap->l2t;
481 int addr_len = neigh->tbl->key_len;
482 u32 *addr = (u32 *) neigh->primary_key;
483 int ifidx = neigh->dev->ifindex;
484 int hash = addr_hash(addr, addr_len, ifidx);
486 read_lock_bh(&d->lock);
487 for (e = d->l2tab[hash].first; e; e = e->next)
488 if (!addreq(e, addr) && e->ifindex == ifidx) {
490 if (atomic_read(&e->refcnt))
492 spin_unlock(&e->lock);
495 read_unlock_bh(&d->lock);
499 read_unlock(&d->lock);
501 if (neigh != e->neigh)
502 neigh_replace(e, neigh);
504 if (e->state == L2T_STATE_RESOLVING) {
505 if (neigh->nud_state & NUD_FAILED) {
507 e->arpq_head = e->arpq_tail = NULL;
508 } else if ((neigh->nud_state & (NUD_CONNECTED | NUD_STALE)) &&
510 write_l2e(adap, e, 1);
513 e->state = neigh->nud_state & NUD_CONNECTED ?
514 L2T_STATE_VALID : L2T_STATE_STALE;
515 if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)))
516 write_l2e(adap, e, 0);
519 spin_unlock_bh(&e->lock);
522 handle_failed_resolution(adap, arpq);
525 /* Allocate an L2T entry for use by a switching rule. Such need to be
526 * explicitly freed and while busy they are not on any hash chain, so normal
527 * address resolution updates do not see them.
529 struct l2t_entry *t4_l2t_alloc_switching(struct l2t_data *d)
533 write_lock_bh(&d->lock);
536 spin_lock(&e->lock); /* avoid race with t4_l2t_free */
537 e->state = L2T_STATE_SWITCHING;
538 atomic_set(&e->refcnt, 1);
539 spin_unlock(&e->lock);
541 write_unlock_bh(&d->lock);
545 /* Sets/updates the contents of a switching L2T entry that has been allocated
546 * with an earlier call to @t4_l2t_alloc_switching.
548 int t4_l2t_set_switching(struct adapter *adap, struct l2t_entry *e, u16 vlan,
549 u8 port, u8 *eth_addr)
553 memcpy(e->dmac, eth_addr, ETH_ALEN);
554 return write_l2e(adap, e, 0);
557 struct l2t_data *t4_init_l2t(void)
562 d = t4_alloc_mem(sizeof(*d));
567 atomic_set(&d->nfree, L2T_SIZE);
568 rwlock_init(&d->lock);
570 for (i = 0; i < L2T_SIZE; ++i) {
572 d->l2tab[i].state = L2T_STATE_UNUSED;
573 spin_lock_init(&d->l2tab[i].lock);
574 atomic_set(&d->l2tab[i].refcnt, 0);
579 static inline void *l2t_get_idx(struct seq_file *seq, loff_t pos)
581 struct l2t_entry *l2tab = seq->private;
583 return pos >= L2T_SIZE ? NULL : &l2tab[pos];
586 static void *l2t_seq_start(struct seq_file *seq, loff_t *pos)
588 return *pos ? l2t_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
591 static void *l2t_seq_next(struct seq_file *seq, void *v, loff_t *pos)
593 v = l2t_get_idx(seq, *pos);
599 static void l2t_seq_stop(struct seq_file *seq, void *v)
603 static char l2e_state(const struct l2t_entry *e)
606 case L2T_STATE_VALID: return 'V';
607 case L2T_STATE_STALE: return 'S';
608 case L2T_STATE_SYNC_WRITE: return 'W';
609 case L2T_STATE_RESOLVING: return e->arpq_head ? 'A' : 'R';
610 case L2T_STATE_SWITCHING: return 'X';
616 static int l2t_seq_show(struct seq_file *seq, void *v)
618 if (v == SEQ_START_TOKEN)
619 seq_puts(seq, " Idx IP address "
620 "Ethernet address VLAN/P LP State Users Port\n");
623 struct l2t_entry *e = v;
625 spin_lock_bh(&e->lock);
626 if (e->state == L2T_STATE_SWITCHING)
629 sprintf(ip, e->v6 ? "%pI6c" : "%pI4", e->addr);
630 seq_printf(seq, "%4u %-25s %17pM %4d %u %2u %c %5u %s\n",
632 e->vlan & VLAN_VID_MASK, vlan_prio(e), e->lport,
633 l2e_state(e), atomic_read(&e->refcnt),
634 e->neigh ? e->neigh->dev->name : "");
635 spin_unlock_bh(&e->lock);
640 static const struct seq_operations l2t_seq_ops = {
641 .start = l2t_seq_start,
642 .next = l2t_seq_next,
643 .stop = l2t_seq_stop,
647 static int l2t_seq_open(struct inode *inode, struct file *file)
649 int rc = seq_open(file, &l2t_seq_ops);
652 struct adapter *adap = inode->i_private;
653 struct seq_file *seq = file->private_data;
655 seq->private = adap->l2t->l2tab;
660 const struct file_operations t4_l2t_fops = {
661 .owner = THIS_MODULE,
662 .open = l2t_seq_open,
665 .release = seq_release,