2 * Copyright (c) 2003-2008 Chelsio, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 #include <linux/skbuff.h>
33 #include <linux/netdevice.h>
35 #include <linux/if_vlan.h>
36 #include <linux/jhash.h>
37 #include <net/neighbour.h>
40 #include "cxgb3_defs.h"
43 #include "firmware_exports.h"
45 #define VLAN_NONE 0xfff
48 * Module locking notes: There is a RW lock protecting the L2 table as a
49 * whole plus a spinlock per L2T entry. Entry lookups and allocations happen
50 * under the protection of the table lock, individual entry changes happen
51 * while holding that entry's spinlock. The table lock nests outside the
52 * entry locks. Allocations of new entries take the table lock as writers so
53 * no other lookups can happen while allocating new entries. Entry updates
54 * take the table lock as readers so multiple entries can be updated in
55 * parallel. An L2T entry can be dropped by decrementing its reference count
56 * and therefore can happen in parallel with entry allocation but no entry
57 * can change state or increment its ref count during allocation as both of
58 * these perform lookups.
61 static inline unsigned int vlan_prio(const struct l2t_entry *e)
66 static inline unsigned int arp_hash(u32 key, int ifindex,
67 const struct l2t_data *d)
69 return jhash_2words(key, ifindex, 0) & (d->nentries - 1);
72 static inline void neigh_replace(struct l2t_entry *e, struct neighbour *n)
76 neigh_release(e->neigh);
81 * Set up an L2T entry and send any packets waiting in the arp queue. The
82 * supplied skb is used for the CPL_L2T_WRITE_REQ. Must be called with the
85 static int setup_l2e_send_pending(struct t3cdev *dev, struct sk_buff *skb,
88 struct cpl_l2t_write_req *req;
92 skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
97 req = (struct cpl_l2t_write_req *)__skb_put(skb, sizeof(*req));
98 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
99 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, e->idx));
100 req->params = htonl(V_L2T_W_IDX(e->idx) | V_L2T_W_IFF(e->smt_idx) |
101 V_L2T_W_VLAN(e->vlan & VLAN_VID_MASK) |
102 V_L2T_W_PRIO(vlan_prio(e)));
103 memcpy(e->dmac, e->neigh->ha, sizeof(e->dmac));
104 memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
105 skb->priority = CPL_PRIORITY_CONTROL;
106 cxgb3_ofld_send(dev, skb);
108 skb_queue_walk_safe(&e->arpq, skb, tmp) {
109 __skb_unlink(skb, &e->arpq);
110 cxgb3_ofld_send(dev, skb);
112 e->state = L2T_STATE_VALID;
118 * Add a packet to the an L2T entry's queue of packets awaiting resolution.
119 * Must be called with the entry's lock held.
121 static inline void arpq_enqueue(struct l2t_entry *e, struct sk_buff *skb)
123 __skb_queue_tail(&e->arpq, skb);
126 int t3_l2t_send_slow(struct t3cdev *dev, struct sk_buff *skb,
131 case L2T_STATE_STALE: /* entry is stale, kick off revalidation */
132 neigh_event_send(e->neigh, NULL);
133 spin_lock_bh(&e->lock);
134 if (e->state == L2T_STATE_STALE)
135 e->state = L2T_STATE_VALID;
136 spin_unlock_bh(&e->lock);
137 case L2T_STATE_VALID: /* fast-path, send the packet on */
138 return cxgb3_ofld_send(dev, skb);
139 case L2T_STATE_RESOLVING:
140 spin_lock_bh(&e->lock);
141 if (e->state != L2T_STATE_RESOLVING) {
142 /* ARP already completed */
143 spin_unlock_bh(&e->lock);
146 arpq_enqueue(e, skb);
147 spin_unlock_bh(&e->lock);
150 * Only the first packet added to the arpq should kick off
151 * resolution. However, because the alloc_skb below can fail,
152 * we allow each packet added to the arpq to retry resolution
153 * as a way of recovering from transient memory exhaustion.
154 * A better way would be to use a work request to retry L2T
155 * entries when there's no memory.
157 if (!neigh_event_send(e->neigh, NULL)) {
158 skb = alloc_skb(sizeof(struct cpl_l2t_write_req),
163 spin_lock_bh(&e->lock);
164 if (!skb_queue_empty(&e->arpq))
165 setup_l2e_send_pending(dev, skb, e);
166 else /* we lost the race */
168 spin_unlock_bh(&e->lock);
174 EXPORT_SYMBOL(t3_l2t_send_slow);
176 void t3_l2t_send_event(struct t3cdev *dev, struct l2t_entry *e)
180 case L2T_STATE_STALE: /* entry is stale, kick off revalidation */
181 neigh_event_send(e->neigh, NULL);
182 spin_lock_bh(&e->lock);
183 if (e->state == L2T_STATE_STALE) {
184 e->state = L2T_STATE_VALID;
186 spin_unlock_bh(&e->lock);
188 case L2T_STATE_VALID: /* fast-path, send the packet on */
190 case L2T_STATE_RESOLVING:
191 spin_lock_bh(&e->lock);
192 if (e->state != L2T_STATE_RESOLVING) {
193 /* ARP already completed */
194 spin_unlock_bh(&e->lock);
197 spin_unlock_bh(&e->lock);
200 * Only the first packet added to the arpq should kick off
201 * resolution. However, because the alloc_skb below can fail,
202 * we allow each packet added to the arpq to retry resolution
203 * as a way of recovering from transient memory exhaustion.
204 * A better way would be to use a work request to retry L2T
205 * entries when there's no memory.
207 neigh_event_send(e->neigh, NULL);
212 EXPORT_SYMBOL(t3_l2t_send_event);
215 * Allocate a free L2T entry. Must be called with l2t_data.lock held.
217 static struct l2t_entry *alloc_l2e(struct l2t_data *d)
219 struct l2t_entry *end, *e, **p;
221 if (!atomic_read(&d->nfree))
224 /* there's definitely a free entry */
225 for (e = d->rover, end = &d->l2tab[d->nentries]; e != end; ++e)
226 if (atomic_read(&e->refcnt) == 0)
229 for (e = &d->l2tab[1]; atomic_read(&e->refcnt); ++e) ;
232 atomic_dec(&d->nfree);
235 * The entry we found may be an inactive entry that is
236 * presently in the hash table. We need to remove it.
238 if (e->state != L2T_STATE_UNUSED) {
239 int hash = arp_hash(e->addr, e->ifindex, d);
241 for (p = &d->l2tab[hash].first; *p; p = &(*p)->next)
246 e->state = L2T_STATE_UNUSED;
252 * Called when an L2T entry has no more users. The entry is left in the hash
253 * table since it is likely to be reused but we also bump nfree to indicate
254 * that the entry can be reallocated for a different neighbor. We also drop
255 * the existing neighbor reference in case the neighbor is going away and is
256 * waiting on our reference.
258 * Because entries can be reallocated to other neighbors once their ref count
259 * drops to 0 we need to take the entry's lock to avoid races with a new
262 void t3_l2e_free(struct l2t_data *d, struct l2t_entry *e)
264 spin_lock_bh(&e->lock);
265 if (atomic_read(&e->refcnt) == 0) { /* hasn't been recycled */
267 neigh_release(e->neigh);
271 spin_unlock_bh(&e->lock);
272 atomic_inc(&d->nfree);
275 EXPORT_SYMBOL(t3_l2e_free);
278 * Update an L2T entry that was previously used for the same next hop as neigh.
279 * Must be called with softirqs disabled.
281 static inline void reuse_entry(struct l2t_entry *e, struct neighbour *neigh)
283 unsigned int nud_state;
285 spin_lock(&e->lock); /* avoid race with t3_l2t_free */
287 if (neigh != e->neigh)
288 neigh_replace(e, neigh);
289 nud_state = neigh->nud_state;
290 if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)) ||
291 !(nud_state & NUD_VALID))
292 e->state = L2T_STATE_RESOLVING;
293 else if (nud_state & NUD_CONNECTED)
294 e->state = L2T_STATE_VALID;
296 e->state = L2T_STATE_STALE;
297 spin_unlock(&e->lock);
300 struct l2t_entry *t3_l2t_get(struct t3cdev *cdev, struct neighbour *neigh,
301 struct net_device *dev)
304 struct l2t_data *d = L2DATA(cdev);
305 u32 addr = *(u32 *) neigh->primary_key;
306 int ifidx = neigh->dev->ifindex;
307 int hash = arp_hash(addr, ifidx, d);
308 struct port_info *p = netdev_priv(dev);
309 int smt_idx = p->port_id;
311 write_lock_bh(&d->lock);
312 for (e = d->l2tab[hash].first; e; e = e->next)
313 if (e->addr == addr && e->ifindex == ifidx &&
314 e->smt_idx == smt_idx) {
316 if (atomic_read(&e->refcnt) == 1)
317 reuse_entry(e, neigh);
321 /* Need to allocate a new entry */
324 spin_lock(&e->lock); /* avoid race with t3_l2t_free */
325 e->next = d->l2tab[hash].first;
326 d->l2tab[hash].first = e;
327 e->state = L2T_STATE_RESOLVING;
330 e->smt_idx = smt_idx;
331 atomic_set(&e->refcnt, 1);
332 neigh_replace(e, neigh);
333 if (neigh->dev->priv_flags & IFF_802_1Q_VLAN)
334 e->vlan = vlan_dev_vlan_id(neigh->dev);
337 spin_unlock(&e->lock);
340 write_unlock_bh(&d->lock);
344 EXPORT_SYMBOL(t3_l2t_get);
347 * Called when address resolution fails for an L2T entry to handle packets
348 * on the arpq head. If a packet specifies a failure handler it is invoked,
349 * otherwise the packets is sent to the offload device.
351 * XXX: maybe we should abandon the latter behavior and just require a failure
354 static void handle_failed_resolution(struct t3cdev *dev, struct sk_buff_head *arpq)
356 struct sk_buff *skb, *tmp;
358 skb_queue_walk_safe(arpq, skb, tmp) {
359 struct l2t_skb_cb *cb = L2T_SKB_CB(skb);
361 __skb_unlink(skb, arpq);
362 if (cb->arp_failure_handler)
363 cb->arp_failure_handler(dev, skb);
365 cxgb3_ofld_send(dev, skb);
370 * Called when the host's ARP layer makes a change to some entry that is
371 * loaded into the HW L2 table.
373 void t3_l2t_update(struct t3cdev *dev, struct neighbour *neigh)
375 struct sk_buff_head arpq;
377 struct l2t_data *d = L2DATA(dev);
378 u32 addr = *(u32 *) neigh->primary_key;
379 int ifidx = neigh->dev->ifindex;
380 int hash = arp_hash(addr, ifidx, d);
382 read_lock_bh(&d->lock);
383 for (e = d->l2tab[hash].first; e; e = e->next)
384 if (e->addr == addr && e->ifindex == ifidx) {
388 read_unlock_bh(&d->lock);
392 __skb_queue_head_init(&arpq);
394 read_unlock(&d->lock);
395 if (atomic_read(&e->refcnt)) {
396 if (neigh != e->neigh)
397 neigh_replace(e, neigh);
399 if (e->state == L2T_STATE_RESOLVING) {
400 if (neigh->nud_state & NUD_FAILED) {
401 skb_queue_splice_init(&e->arpq, &arpq);
402 } else if (neigh->nud_state & (NUD_CONNECTED|NUD_STALE))
403 setup_l2e_send_pending(dev, NULL, e);
405 e->state = neigh->nud_state & NUD_CONNECTED ?
406 L2T_STATE_VALID : L2T_STATE_STALE;
407 if (memcmp(e->dmac, neigh->ha, 6))
408 setup_l2e_send_pending(dev, NULL, e);
411 spin_unlock_bh(&e->lock);
413 if (!skb_queue_empty(&arpq))
414 handle_failed_resolution(dev, &arpq);
417 struct l2t_data *t3_init_l2t(unsigned int l2t_capacity)
420 int i, size = sizeof(*d) + l2t_capacity * sizeof(struct l2t_entry);
422 d = cxgb_alloc_mem(size);
426 d->nentries = l2t_capacity;
427 d->rover = &d->l2tab[1]; /* entry 0 is not used */
428 atomic_set(&d->nfree, l2t_capacity - 1);
429 rwlock_init(&d->lock);
431 for (i = 0; i < l2t_capacity; ++i) {
433 d->l2tab[i].state = L2T_STATE_UNUSED;
434 spin_lock_init(&d->l2tab[i].lock);
435 atomic_set(&d->l2tab[i].refcnt, 0);
440 void t3_free_l2t(struct l2t_data *d)