1/*
2 * This file is part of the Chelsio T4 Ethernet driver for Linux.
3 *
4 * Copyright (c) 2003-2010 Chelsio Communications, Inc. All rights reserved.
5 *
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:
11 *
12 *     Redistribution and use in source and binary forms, with or
13 *     without modification, are permitted provided that the following
14 *     conditions are met:
15 *
16 *      - Redistributions of source code must retain the above
17 *        copyright notice, this list of conditions and the following
18 *        disclaimer.
19 *
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.
24 *
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
32 * SOFTWARE.
33 */
34
35#include <linux/skbuff.h>
36#include <linux/netdevice.h>
37#include <linux/if.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>
44#include "cxgb4.h"
45#include "l2t.h"
46#include "t4_msg.h"
47#include "t4fw_api.h"
48
49#define VLAN_NONE 0xfff
50
51/* identifies sync vs async L2T_WRITE_REQs */
52#define F_SYNC_WR    (1 << 12)
53
54enum {
55	L2T_STATE_VALID,      /* entry is up to date */
56	L2T_STATE_STALE,      /* entry may be used but needs revalidation */
57	L2T_STATE_RESOLVING,  /* entry needs address resolution */
58	L2T_STATE_SYNC_WRITE, /* synchronous write of entry underway */
59
60	/* when state is one of the below the entry is not hashed */
61	L2T_STATE_SWITCHING,  /* entry is being used by a switching filter */
62	L2T_STATE_UNUSED      /* entry not in use */
63};
64
65struct l2t_data {
66	rwlock_t lock;
67	atomic_t nfree;             /* number of free entries */
68	struct l2t_entry *rover;    /* starting point for next allocation */
69	struct l2t_entry l2tab[L2T_SIZE];
70};
71
72static inline unsigned int vlan_prio(const struct l2t_entry *e)
73{
74	return e->vlan >> 13;
75}
76
77static inline void l2t_hold(struct l2t_data *d, struct l2t_entry *e)
78{
79	if (atomic_add_return(1, &e->refcnt) == 1)  /* 0 -> 1 transition */
80		atomic_dec(&d->nfree);
81}
82
83/*
84 * To avoid having to check address families we do not allow v4 and v6
85 * neighbors to be on the same hash chain.  We keep v4 entries in the first
86 * half of available hash buckets and v6 in the second.
87 */
88enum {
89	L2T_SZ_HALF = L2T_SIZE / 2,
90	L2T_HASH_MASK = L2T_SZ_HALF - 1
91};
92
93static inline unsigned int arp_hash(const u32 *key, int ifindex)
94{
95	return jhash_2words(*key, ifindex, 0) & L2T_HASH_MASK;
96}
97
98static inline unsigned int ipv6_hash(const u32 *key, int ifindex)
99{
100	u32 xor = key[0] ^ key[1] ^ key[2] ^ key[3];
101
102	return L2T_SZ_HALF + (jhash_2words(xor, ifindex, 0) & L2T_HASH_MASK);
103}
104
105static unsigned int addr_hash(const u32 *addr, int addr_len, int ifindex)
106{
107	return addr_len == 4 ? arp_hash(addr, ifindex) :
108			       ipv6_hash(addr, ifindex);
109}
110
111/*
112 * Checks if an L2T entry is for the given IP/IPv6 address.  It does not check
113 * whether the L2T entry and the address are of the same address family.
114 * Callers ensure an address is only checked against L2T entries of the same
115 * family, something made trivial by the separation of IP and IPv6 hash chains
116 * mentioned above.  Returns 0 if there's a match,
117 */
118static int addreq(const struct l2t_entry *e, const u32 *addr)
119{
120	if (e->v6)
121		return (e->addr[0] ^ addr[0]) | (e->addr[1] ^ addr[1]) |
122		       (e->addr[2] ^ addr[2]) | (e->addr[3] ^ addr[3]);
123	return e->addr[0] ^ addr[0];
124}
125
126static void neigh_replace(struct l2t_entry *e, struct neighbour *n)
127{
128	neigh_hold(n);
129	if (e->neigh)
130		neigh_release(e->neigh);
131	e->neigh = n;
132}
133
134/*
135 * Write an L2T entry.  Must be called with the entry locked.
136 * The write may be synchronous or asynchronous.
137 */
138static int write_l2e(struct adapter *adap, struct l2t_entry *e, int sync)
139{
140	struct sk_buff *skb;
141	struct cpl_l2t_write_req *req;
142
143	skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
144	if (!skb)
145		return -ENOMEM;
146
147	req = (struct cpl_l2t_write_req *)__skb_put(skb, sizeof(*req));
148	INIT_TP_WR(req, 0);
149
150	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ,
151					e->idx | (sync ? F_SYNC_WR : 0) |
152					TID_QID(adap->sge.fw_evtq.abs_id)));
153	req->params = htons(L2T_W_PORT(e->lport) | L2T_W_NOREPLY(!sync));
154	req->l2t_idx = htons(e->idx);
155	req->vlan = htons(e->vlan);
156	if (e->neigh)
157		memcpy(e->dmac, e->neigh->ha, sizeof(e->dmac));
158	memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
159
160	set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
161	t4_ofld_send(adap, skb);
162
163	if (sync && e->state != L2T_STATE_SWITCHING)
164		e->state = L2T_STATE_SYNC_WRITE;
165	return 0;
166}
167
168/*
169 * Send packets waiting in an L2T entry's ARP queue.  Must be called with the
170 * entry locked.
171 */
172static void send_pending(struct adapter *adap, struct l2t_entry *e)
173{
174	while (e->arpq_head) {
175		struct sk_buff *skb = e->arpq_head;
176
177		e->arpq_head = skb->next;
178		skb->next = NULL;
179		t4_ofld_send(adap, skb);
180	}
181	e->arpq_tail = NULL;
182}
183
184/*
185 * Process a CPL_L2T_WRITE_RPL.  Wake up the ARP queue if it completes a
186 * synchronous L2T_WRITE.  Note that the TID in the reply is really the L2T
187 * index it refers to.
188 */
189void do_l2t_write_rpl(struct adapter *adap, const struct cpl_l2t_write_rpl *rpl)
190{
191	unsigned int tid = GET_TID(rpl);
192	unsigned int idx = tid & (L2T_SIZE - 1);
193
194	if (unlikely(rpl->status != CPL_ERR_NONE)) {
195		dev_err(adap->pdev_dev,
196			"Unexpected L2T_WRITE_RPL status %u for entry %u\n",
197			rpl->status, idx);
198		return;
199	}
200
201	if (tid & F_SYNC_WR) {
202		struct l2t_entry *e = &adap->l2t->l2tab[idx];
203
204		spin_lock(&e->lock);
205		if (e->state != L2T_STATE_SWITCHING) {
206			send_pending(adap, e);
207			e->state = (e->neigh->nud_state & NUD_STALE) ?
208					L2T_STATE_STALE : L2T_STATE_VALID;
209		}
210		spin_unlock(&e->lock);
211	}
212}
213
214/*
215 * Add a packet to an L2T entry's queue of packets awaiting resolution.
216 * Must be called with the entry's lock held.
217 */
218static inline void arpq_enqueue(struct l2t_entry *e, struct sk_buff *skb)
219{
220	skb->next = NULL;
221	if (e->arpq_head)
222		e->arpq_tail->next = skb;
223	else
224		e->arpq_head = skb;
225	e->arpq_tail = skb;
226}
227
228int cxgb4_l2t_send(struct net_device *dev, struct sk_buff *skb,
229		   struct l2t_entry *e)
230{
231	struct adapter *adap = netdev2adap(dev);
232
233again:
234	switch (e->state) {
235	case L2T_STATE_STALE:     /* entry is stale, kick off revalidation */
236		neigh_event_send(e->neigh, NULL);
237		spin_lock_bh(&e->lock);
238		if (e->state == L2T_STATE_STALE)
239			e->state = L2T_STATE_VALID;
240		spin_unlock_bh(&e->lock);
241	case L2T_STATE_VALID:     /* fast-path, send the packet on */
242		return t4_ofld_send(adap, skb);
243	case L2T_STATE_RESOLVING:
244	case L2T_STATE_SYNC_WRITE:
245		spin_lock_bh(&e->lock);
246		if (e->state != L2T_STATE_SYNC_WRITE &&
247		    e->state != L2T_STATE_RESOLVING) {
248			spin_unlock_bh(&e->lock);
249			goto again;
250		}
251		arpq_enqueue(e, skb);
252		spin_unlock_bh(&e->lock);
253
254		if (e->state == L2T_STATE_RESOLVING &&
255		    !neigh_event_send(e->neigh, NULL)) {
256			spin_lock_bh(&e->lock);
257			if (e->state == L2T_STATE_RESOLVING && e->arpq_head)
258				write_l2e(adap, e, 1);
259			spin_unlock_bh(&e->lock);
260		}
261	}
262	return 0;
263}
264EXPORT_SYMBOL(cxgb4_l2t_send);
265
266/*
267 * Allocate a free L2T entry.  Must be called with l2t_data.lock held.
268 */
269static struct l2t_entry *alloc_l2e(struct l2t_data *d)
270{
271	struct l2t_entry *end, *e, **p;
272
273	if (!atomic_read(&d->nfree))
274		return NULL;
275
276	/* there's definitely a free entry */
277	for (e = d->rover, end = &d->l2tab[L2T_SIZE]; e != end; ++e)
278		if (atomic_read(&e->refcnt) == 0)
279			goto found;
280
281	for (e = d->l2tab; atomic_read(&e->refcnt); ++e)
282		;
283found:
284	d->rover = e + 1;
285	atomic_dec(&d->nfree);
286
287	/*
288	 * The entry we found may be an inactive entry that is
289	 * presently in the hash table.  We need to remove it.
290	 */
291	if (e->state < L2T_STATE_SWITCHING)
292		for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next)
293			if (*p == e) {
294				*p = e->next;
295				e->next = NULL;
296				break;
297			}
298
299	e->state = L2T_STATE_UNUSED;
300	return e;
301}
302
303/*
304 * Called when an L2T entry has no more users.
305 */
306static void t4_l2e_free(struct l2t_entry *e)
307{
308	struct l2t_data *d;
309
310	spin_lock_bh(&e->lock);
311	if (atomic_read(&e->refcnt) == 0) {  /* hasn't been recycled */
312		if (e->neigh) {
313			neigh_release(e->neigh);
314			e->neigh = NULL;
315		}
316		while (e->arpq_head) {
317			struct sk_buff *skb = e->arpq_head;
318
319			e->arpq_head = skb->next;
320			kfree_skb(skb);
321		}
322		e->arpq_tail = NULL;
323	}
324	spin_unlock_bh(&e->lock);
325
326	d = container_of(e, struct l2t_data, l2tab[e->idx]);
327	atomic_inc(&d->nfree);
328}
329
330void cxgb4_l2t_release(struct l2t_entry *e)
331{
332	if (atomic_dec_and_test(&e->refcnt))
333		t4_l2e_free(e);
334}
335EXPORT_SYMBOL(cxgb4_l2t_release);
336
337/*
338 * Update an L2T entry that was previously used for the same next hop as neigh.
339 * Must be called with softirqs disabled.
340 */
341static void reuse_entry(struct l2t_entry *e, struct neighbour *neigh)
342{
343	unsigned int nud_state;
344
345	spin_lock(&e->lock);                /* avoid race with t4_l2t_free */
346	if (neigh != e->neigh)
347		neigh_replace(e, neigh);
348	nud_state = neigh->nud_state;
349	if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)) ||
350	    !(nud_state & NUD_VALID))
351		e->state = L2T_STATE_RESOLVING;
352	else if (nud_state & NUD_CONNECTED)
353		e->state = L2T_STATE_VALID;
354	else
355		e->state = L2T_STATE_STALE;
356	spin_unlock(&e->lock);
357}
358
359struct l2t_entry *cxgb4_l2t_get(struct l2t_data *d, struct neighbour *neigh,
360				const struct net_device *physdev,
361				unsigned int priority)
362{
363	u8 lport;
364	u16 vlan;
365	struct l2t_entry *e;
366	int addr_len = neigh->tbl->key_len;
367	u32 *addr = (u32 *)neigh->primary_key;
368	int ifidx = neigh->dev->ifindex;
369	int hash = addr_hash(addr, addr_len, ifidx);
370
371	if (neigh->dev->flags & IFF_LOOPBACK)
372		lport = netdev2pinfo(physdev)->tx_chan + 4;
373	else
374		lport = netdev2pinfo(physdev)->lport;
375
376	if (neigh->dev->priv_flags & IFF_802_1Q_VLAN)
377		vlan = vlan_dev_vlan_id(neigh->dev);
378	else
379		vlan = VLAN_NONE;
380
381	write_lock_bh(&d->lock);
382	for (e = d->l2tab[hash].first; e; e = e->next)
383		if (!addreq(e, addr) && e->ifindex == ifidx &&
384		    e->vlan == vlan && e->lport == lport) {
385			l2t_hold(d, e);
386			if (atomic_read(&e->refcnt) == 1)
387				reuse_entry(e, neigh);
388			goto done;
389		}
390
391	/* Need to allocate a new entry */
392	e = alloc_l2e(d);
393	if (e) {
394		spin_lock(&e->lock);          /* avoid race with t4_l2t_free */
395		e->state = L2T_STATE_RESOLVING;
396		memcpy(e->addr, addr, addr_len);
397		e->ifindex = ifidx;
398		e->hash = hash;
399		e->lport = lport;
400		e->v6 = addr_len == 16;
401		atomic_set(&e->refcnt, 1);
402		neigh_replace(e, neigh);
403		e->vlan = vlan;
404		e->next = d->l2tab[hash].first;
405		d->l2tab[hash].first = e;
406		spin_unlock(&e->lock);
407	}
408done:
409	write_unlock_bh(&d->lock);
410	return e;
411}
412EXPORT_SYMBOL(cxgb4_l2t_get);
413
414/*
415 * Called when address resolution fails for an L2T entry to handle packets
416 * on the arpq head.  If a packet specifies a failure handler it is invoked,
417 * otherwise the packet is sent to the device.
418 */
419static void handle_failed_resolution(struct adapter *adap, struct sk_buff *arpq)
420{
421	while (arpq) {
422		struct sk_buff *skb = arpq;
423		const struct l2t_skb_cb *cb = L2T_SKB_CB(skb);
424
425		arpq = skb->next;
426		skb->next = NULL;
427		if (cb->arp_err_handler)
428			cb->arp_err_handler(cb->handle, skb);
429		else
430			t4_ofld_send(adap, skb);
431	}
432}
433
434/*
435 * Called when the host's neighbor layer makes a change to some entry that is
436 * loaded into the HW L2 table.
437 */
438void t4_l2t_update(struct adapter *adap, struct neighbour *neigh)
439{
440	struct l2t_entry *e;
441	struct sk_buff *arpq = NULL;
442	struct l2t_data *d = adap->l2t;
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(addr, addr_len, ifidx);
447
448	read_lock_bh(&d->lock);
449	for (e = d->l2tab[hash].first; e; e = e->next)
450		if (!addreq(e, addr) && e->ifindex == ifidx) {
451			spin_lock(&e->lock);
452			if (atomic_read(&e->refcnt))
453				goto found;
454			spin_unlock(&e->lock);
455			break;
456		}
457	read_unlock_bh(&d->lock);
458	return;
459
460 found:
461	read_unlock(&d->lock);
462
463	if (neigh != e->neigh)
464		neigh_replace(e, neigh);
465
466	if (e->state == L2T_STATE_RESOLVING) {
467		if (neigh->nud_state & NUD_FAILED) {
468			arpq = e->arpq_head;
469			e->arpq_head = e->arpq_tail = NULL;
470		} else if ((neigh->nud_state & (NUD_CONNECTED | NUD_STALE)) &&
471			   e->arpq_head) {
472			write_l2e(adap, e, 1);
473		}
474	} else {
475		e->state = neigh->nud_state & NUD_CONNECTED ?
476			L2T_STATE_VALID : L2T_STATE_STALE;
477		if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)))
478			write_l2e(adap, e, 0);
479	}
480
481	spin_unlock_bh(&e->lock);
482
483	if (arpq)
484		handle_failed_resolution(adap, arpq);
485}
486
487struct l2t_data *t4_init_l2t(void)
488{
489	int i;
490	struct l2t_data *d;
491
492	d = t4_alloc_mem(sizeof(*d));
493	if (!d)
494		return NULL;
495
496	d->rover = d->l2tab;
497	atomic_set(&d->nfree, L2T_SIZE);
498	rwlock_init(&d->lock);
499
500	for (i = 0; i < L2T_SIZE; ++i) {
501		d->l2tab[i].idx = i;
502		d->l2tab[i].state = L2T_STATE_UNUSED;
503		spin_lock_init(&d->l2tab[i].lock);
504		atomic_set(&d->l2tab[i].refcnt, 0);
505	}
506	return d;
507}
508
509static inline void *l2t_get_idx(struct seq_file *seq, loff_t pos)
510{
511	struct l2t_entry *l2tab = seq->private;
512
513	return pos >= L2T_SIZE ? NULL : &l2tab[pos];
514}
515
516static void *l2t_seq_start(struct seq_file *seq, loff_t *pos)
517{
518	return *pos ? l2t_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
519}
520
521static void *l2t_seq_next(struct seq_file *seq, void *v, loff_t *pos)
522{
523	v = l2t_get_idx(seq, *pos);
524	if (v)
525		++*pos;
526	return v;
527}
528
529static void l2t_seq_stop(struct seq_file *seq, void *v)
530{
531}
532
533static char l2e_state(const struct l2t_entry *e)
534{
535	switch (e->state) {
536	case L2T_STATE_VALID: return 'V';
537	case L2T_STATE_STALE: return 'S';
538	case L2T_STATE_SYNC_WRITE: return 'W';
539	case L2T_STATE_RESOLVING: return e->arpq_head ? 'A' : 'R';
540	case L2T_STATE_SWITCHING: return 'X';
541	default:
542		return 'U';
543	}
544}
545
546static int l2t_seq_show(struct seq_file *seq, void *v)
547{
548	if (v == SEQ_START_TOKEN)
549		seq_puts(seq, " Idx IP address                "
550			 "Ethernet address  VLAN/P LP State Users Port\n");
551	else {
552		char ip[60];
553		struct l2t_entry *e = v;
554
555		spin_lock_bh(&e->lock);
556		if (e->state == L2T_STATE_SWITCHING)
557			ip[0] = '\0';
558		else
559			sprintf(ip, e->v6 ? "%pI6c" : "%pI4", e->addr);
560		seq_printf(seq, "%4u %-25s %17pM %4d %u %2u   %c   %5u %s\n",
561			   e->idx, ip, e->dmac,
562			   e->vlan & VLAN_VID_MASK, vlan_prio(e), e->lport,
563			   l2e_state(e), atomic_read(&e->refcnt),
564			   e->neigh ? e->neigh->dev->name : "");
565		spin_unlock_bh(&e->lock);
566	}
567	return 0;
568}
569
570static const struct seq_operations l2t_seq_ops = {
571	.start = l2t_seq_start,
572	.next = l2t_seq_next,
573	.stop = l2t_seq_stop,
574	.show = l2t_seq_show
575};
576
577static int l2t_seq_open(struct inode *inode, struct file *file)
578{
579	int rc = seq_open(file, &l2t_seq_ops);
580
581	if (!rc) {
582		struct adapter *adap = inode->i_private;
583		struct seq_file *seq = file->private_data;
584
585		seq->private = adap->l2t->l2tab;
586	}
587	return rc;
588}
589
590const struct file_operations t4_l2t_fops = {
591	.owner = THIS_MODULE,
592	.open = l2t_seq_open,
593	.read = seq_read,
594	.llseek = seq_lseek,
595	.release = seq_release,
596};
597