2 * Copyright (c) 2003-2007 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;
91 skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
96 req = (struct cpl_l2t_write_req *)__skb_put(skb, sizeof(*req));
97 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
98 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, e->idx));
99 req->params = htonl(V_L2T_W_IDX(e->idx) | V_L2T_W_IFF(e->smt_idx) |
100 V_L2T_W_VLAN(e->vlan & VLAN_VID_MASK) |
101 V_L2T_W_PRIO(vlan_prio(e)));
102 memcpy(e->dmac, e->neigh->ha, sizeof(e->dmac));
103 memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
104 skb->priority = CPL_PRIORITY_CONTROL;
105 cxgb3_ofld_send(dev, skb);
106 while (e->arpq_head) {
108 e->arpq_head = skb->next;
110 cxgb3_ofld_send(dev, skb);
113 e->state = L2T_STATE_VALID;
119 * Add a packet to the an L2T entry's queue of packets awaiting resolution.
120 * Must be called with the entry's lock held.
122 static inline void arpq_enqueue(struct l2t_entry *e, struct sk_buff *skb)
126 e->arpq_tail->next = skb;
132 int t3_l2t_send_slow(struct t3cdev *dev, struct sk_buff *skb,
137 case L2T_STATE_STALE: /* entry is stale, kick off revalidation */
138 neigh_event_send(e->neigh, NULL);
139 spin_lock_bh(&e->lock);
140 if (e->state == L2T_STATE_STALE)
141 e->state = L2T_STATE_VALID;
142 spin_unlock_bh(&e->lock);
143 case L2T_STATE_VALID: /* fast-path, send the packet on */
144 return cxgb3_ofld_send(dev, skb);
145 case L2T_STATE_RESOLVING:
146 spin_lock_bh(&e->lock);
147 if (e->state != L2T_STATE_RESOLVING) {
148 /* ARP already completed */
149 spin_unlock_bh(&e->lock);
152 arpq_enqueue(e, skb);
153 spin_unlock_bh(&e->lock);
156 * Only the first packet added to the arpq should kick off
157 * resolution. However, because the alloc_skb below can fail,
158 * we allow each packet added to the arpq to retry resolution
159 * as a way of recovering from transient memory exhaustion.
160 * A better way would be to use a work request to retry L2T
161 * entries when there's no memory.
163 if (!neigh_event_send(e->neigh, NULL)) {
164 skb = alloc_skb(sizeof(struct cpl_l2t_write_req),
169 spin_lock_bh(&e->lock);
171 setup_l2e_send_pending(dev, skb, e);
172 else /* we lost the race */
174 spin_unlock_bh(&e->lock);
180 EXPORT_SYMBOL(t3_l2t_send_slow);
182 void t3_l2t_send_event(struct t3cdev *dev, struct l2t_entry *e)
186 case L2T_STATE_STALE: /* entry is stale, kick off revalidation */
187 neigh_event_send(e->neigh, NULL);
188 spin_lock_bh(&e->lock);
189 if (e->state == L2T_STATE_STALE) {
190 e->state = L2T_STATE_VALID;
192 spin_unlock_bh(&e->lock);
194 case L2T_STATE_VALID: /* fast-path, send the packet on */
196 case L2T_STATE_RESOLVING:
197 spin_lock_bh(&e->lock);
198 if (e->state != L2T_STATE_RESOLVING) {
199 /* ARP already completed */
200 spin_unlock_bh(&e->lock);
203 spin_unlock_bh(&e->lock);
206 * Only the first packet added to the arpq should kick off
207 * resolution. However, because the alloc_skb below can fail,
208 * we allow each packet added to the arpq to retry resolution
209 * as a way of recovering from transient memory exhaustion.
210 * A better way would be to use a work request to retry L2T
211 * entries when there's no memory.
213 neigh_event_send(e->neigh, NULL);
218 EXPORT_SYMBOL(t3_l2t_send_event);
221 * Allocate a free L2T entry. Must be called with l2t_data.lock held.
223 static struct l2t_entry *alloc_l2e(struct l2t_data *d)
225 struct l2t_entry *end, *e, **p;
227 if (!atomic_read(&d->nfree))
230 /* there's definitely a free entry */
231 for (e = d->rover, end = &d->l2tab[d->nentries]; e != end; ++e)
232 if (atomic_read(&e->refcnt) == 0)
235 for (e = &d->l2tab[1]; atomic_read(&e->refcnt); ++e) ;
238 atomic_dec(&d->nfree);
241 * The entry we found may be an inactive entry that is
242 * presently in the hash table. We need to remove it.
244 if (e->state != L2T_STATE_UNUSED) {
245 int hash = arp_hash(e->addr, e->ifindex, d);
247 for (p = &d->l2tab[hash].first; *p; p = &(*p)->next)
252 e->state = L2T_STATE_UNUSED;
258 * Called when an L2T entry has no more users. The entry is left in the hash
259 * table since it is likely to be reused but we also bump nfree to indicate
260 * that the entry can be reallocated for a different neighbor. We also drop
261 * the existing neighbor reference in case the neighbor is going away and is
262 * waiting on our reference.
264 * Because entries can be reallocated to other neighbors once their ref count
265 * drops to 0 we need to take the entry's lock to avoid races with a new
268 void t3_l2e_free(struct l2t_data *d, struct l2t_entry *e)
270 spin_lock_bh(&e->lock);
271 if (atomic_read(&e->refcnt) == 0) { /* hasn't been recycled */
273 neigh_release(e->neigh);
277 spin_unlock_bh(&e->lock);
278 atomic_inc(&d->nfree);
281 EXPORT_SYMBOL(t3_l2e_free);
284 * Update an L2T entry that was previously used for the same next hop as neigh.
285 * Must be called with softirqs disabled.
287 static inline void reuse_entry(struct l2t_entry *e, struct neighbour *neigh)
289 unsigned int nud_state;
291 spin_lock(&e->lock); /* avoid race with t3_l2t_free */
293 if (neigh != e->neigh)
294 neigh_replace(e, neigh);
295 nud_state = neigh->nud_state;
296 if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)) ||
297 !(nud_state & NUD_VALID))
298 e->state = L2T_STATE_RESOLVING;
299 else if (nud_state & NUD_CONNECTED)
300 e->state = L2T_STATE_VALID;
302 e->state = L2T_STATE_STALE;
303 spin_unlock(&e->lock);
306 struct l2t_entry *t3_l2t_get(struct t3cdev *cdev, struct neighbour *neigh,
307 struct net_device *dev)
310 struct l2t_data *d = L2DATA(cdev);
311 u32 addr = *(u32 *) neigh->primary_key;
312 int ifidx = neigh->dev->ifindex;
313 int hash = arp_hash(addr, ifidx, d);
314 struct port_info *p = netdev_priv(dev);
315 int smt_idx = p->port_id;
317 write_lock_bh(&d->lock);
318 for (e = d->l2tab[hash].first; e; e = e->next)
319 if (e->addr == addr && e->ifindex == ifidx &&
320 e->smt_idx == smt_idx) {
322 if (atomic_read(&e->refcnt) == 1)
323 reuse_entry(e, neigh);
327 /* Need to allocate a new entry */
330 spin_lock(&e->lock); /* avoid race with t3_l2t_free */
331 e->next = d->l2tab[hash].first;
332 d->l2tab[hash].first = e;
333 e->state = L2T_STATE_RESOLVING;
336 e->smt_idx = smt_idx;
337 atomic_set(&e->refcnt, 1);
338 neigh_replace(e, neigh);
339 if (neigh->dev->priv_flags & IFF_802_1Q_VLAN)
340 e->vlan = VLAN_DEV_INFO(neigh->dev)->vlan_id;
343 spin_unlock(&e->lock);
346 write_unlock_bh(&d->lock);
350 EXPORT_SYMBOL(t3_l2t_get);
353 * Called when address resolution fails for an L2T entry to handle packets
354 * on the arpq head. If a packet specifies a failure handler it is invoked,
355 * otherwise the packets is sent to the offload device.
357 * XXX: maybe we should abandon the latter behavior and just require a failure
360 static void handle_failed_resolution(struct t3cdev *dev, struct sk_buff *arpq)
363 struct sk_buff *skb = arpq;
364 struct l2t_skb_cb *cb = L2T_SKB_CB(skb);
368 if (cb->arp_failure_handler)
369 cb->arp_failure_handler(dev, skb);
371 cxgb3_ofld_send(dev, skb);
376 * Called when the host's ARP layer makes a change to some entry that is
377 * loaded into the HW L2 table.
379 void t3_l2t_update(struct t3cdev *dev, struct neighbour *neigh)
382 struct sk_buff *arpq = NULL;
383 struct l2t_data *d = L2DATA(dev);
384 u32 addr = *(u32 *) neigh->primary_key;
385 int ifidx = neigh->dev->ifindex;
386 int hash = arp_hash(addr, ifidx, d);
388 read_lock_bh(&d->lock);
389 for (e = d->l2tab[hash].first; e; e = e->next)
390 if (e->addr == addr && e->ifindex == ifidx) {
394 read_unlock_bh(&d->lock);
398 read_unlock(&d->lock);
399 if (atomic_read(&e->refcnt)) {
400 if (neigh != e->neigh)
401 neigh_replace(e, neigh);
403 if (e->state == L2T_STATE_RESOLVING) {
404 if (neigh->nud_state & NUD_FAILED) {
406 e->arpq_head = e->arpq_tail = NULL;
407 } else if (neigh_is_connected(neigh))
408 setup_l2e_send_pending(dev, NULL, e);
410 e->state = neigh_is_connected(neigh) ?
411 L2T_STATE_VALID : L2T_STATE_STALE;
412 if (memcmp(e->dmac, neigh->ha, 6))
413 setup_l2e_send_pending(dev, NULL, e);
416 spin_unlock_bh(&e->lock);
419 handle_failed_resolution(dev, arpq);
422 struct l2t_data *t3_init_l2t(unsigned int l2t_capacity)
425 int i, size = sizeof(*d) + l2t_capacity * sizeof(struct l2t_entry);
427 d = cxgb_alloc_mem(size);
431 d->nentries = l2t_capacity;
432 d->rover = &d->l2tab[1]; /* entry 0 is not used */
433 atomic_set(&d->nfree, l2t_capacity - 1);
434 rwlock_init(&d->lock);
436 for (i = 0; i < l2t_capacity; ++i) {
438 d->l2tab[i].state = L2T_STATE_UNUSED;
439 spin_lock_init(&d->l2tab[i].lock);
440 atomic_set(&d->l2tab[i].refcnt, 0);
445 void t3_free_l2t(struct l2t_data *d)