1 /*-
2 * Copyright (c) 2012 Chelsio Communications, Inc.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26 #include <sys/cdefs.h>
27 __FBSDID("$FreeBSD$");
28
29 #include "opt_inet.h"
30 #include "opt_inet6.h"
31
32 #include <sys/param.h>
33 #include <sys/eventhandler.h>
34 #include <sys/systm.h>
35 #include <sys/kernel.h>
36 #include <sys/module.h>
37 #include <sys/bus.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/rwlock.h>
41 #include <sys/socket.h>
42 #include <sys/sbuf.h>
43 #include <netinet/in.h>
44
45 #include "common/common.h"
46 #include "common/t4_msg.h"
47 #include "t4_l2t.h"
48
49 /*
50 * Module locking notes: There is a RW lock protecting the L2 table as a
51 * whole plus a spinlock per L2T entry. Entry lookups and allocations happen
52 * under the protection of the table lock, individual entry changes happen
53 * while holding that entry's spinlock. The table lock nests outside the
54 * entry locks. Allocations of new entries take the table lock as writers so
55 * no other lookups can happen while allocating new entries. Entry updates
56 * take the table lock as readers so multiple entries can be updated in
57 * parallel. An L2T entry can be dropped by decrementing its reference count
58 * and therefore can happen in parallel with entry allocation but no entry
59 * can change state or increment its ref count during allocation as both of
60 * these perform lookups.
61 *
62 * Note: We do not take references to ifnets in this module because both
63 * the TOE and the sockets already hold references to the interfaces and the
64 * lifetime of an L2T entry is fully contained in the lifetime of the TOE.
65 */
66
67 /*
68 * Allocate a free L2T entry. Must be called with l2t_data.lock held.
69 */
70 struct l2t_entry *
t4_alloc_l2e(struct l2t_data * d)71 t4_alloc_l2e(struct l2t_data *d)
72 {
73 struct l2t_entry *end, *e, **p;
74
75 rw_assert(&d->lock, RA_WLOCKED);
76
77 if (!atomic_load_acq_int(&d->nfree))
78 return (NULL);
79
80 /* there's definitely a free entry */
81 for (e = d->rover, end = &d->l2tab[d->l2t_size]; e != end; ++e)
82 if (atomic_load_acq_int(&e->refcnt) == 0)
83 goto found;
84
85 for (e = d->l2tab; atomic_load_acq_int(&e->refcnt); ++e)
86 continue;
87 found:
88 d->rover = e + 1;
89 atomic_subtract_int(&d->nfree, 1);
90
91 /*
92 * The entry we found may be an inactive entry that is
93 * presently in the hash table. We need to remove it.
94 */
95 if (e->state < L2T_STATE_SWITCHING) {
96 for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next) {
97 if (*p == e) {
98 *p = e->next;
99 e->next = NULL;
100 break;
101 }
102 }
103 }
104
105 e->state = L2T_STATE_UNUSED;
106 return (e);
107 }
108
109 static struct l2t_entry *
find_or_alloc_l2e(struct l2t_data * d,uint16_t vlan,uint8_t port,uint8_t * dmac)110 find_or_alloc_l2e(struct l2t_data *d, uint16_t vlan, uint8_t port, uint8_t *dmac)
111 {
112 struct l2t_entry *end, *e, **p;
113 struct l2t_entry *first_free = NULL;
114
115 for (e = &d->l2tab[0], end = &d->l2tab[d->l2t_size]; e != end; ++e) {
116 if (atomic_load_acq_int(&e->refcnt) == 0) {
117 if (!first_free)
118 first_free = e;
119 } else if (e->state == L2T_STATE_SWITCHING &&
120 memcmp(e->dmac, dmac, ETHER_ADDR_LEN) == 0 &&
121 e->vlan == vlan && e->lport == port)
122 return (e); /* Found existing entry that matches. */
123 }
124
125 if (first_free == NULL)
126 return (NULL); /* No match and no room for a new entry. */
127
128 /*
129 * The entry we found may be an inactive entry that is
130 * presently in the hash table. We need to remove it.
131 */
132 e = first_free;
133 if (e->state < L2T_STATE_SWITCHING) {
134 for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next) {
135 if (*p == e) {
136 *p = e->next;
137 e->next = NULL;
138 break;
139 }
140 }
141 }
142 e->state = L2T_STATE_UNUSED;
143 return (e);
144 }
145
146
147 /*
148 * Write an L2T entry. Must be called with the entry locked.
149 * The write may be synchronous or asynchronous.
150 */
151 int
t4_write_l2e(struct l2t_entry * e,int sync)152 t4_write_l2e(struct l2t_entry *e, int sync)
153 {
154 struct sge_wrq *wrq;
155 struct adapter *sc;
156 struct wrq_cookie cookie;
157 struct cpl_l2t_write_req *req;
158 int idx;
159
160 mtx_assert(&e->lock, MA_OWNED);
161 MPASS(e->wrq != NULL);
162
163 wrq = e->wrq;
164 sc = wrq->adapter;
165
166 req = start_wrq_wr(wrq, howmany(sizeof(*req), 16), &cookie);
167 if (req == NULL)
168 return (ENOMEM);
169
170 idx = e->idx + sc->vres.l2t.start;
171 INIT_TP_WR(req, 0);
172 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, idx |
173 V_SYNC_WR(sync) | V_TID_QID(e->iqid)));
174 req->params = htons(V_L2T_W_PORT(e->lport) | V_L2T_W_NOREPLY(!sync));
175 req->l2t_idx = htons(idx);
176 req->vlan = htons(e->vlan);
177 memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
178
179 commit_wrq_wr(wrq, req, &cookie);
180
181 if (sync && e->state != L2T_STATE_SWITCHING)
182 e->state = L2T_STATE_SYNC_WRITE;
183
184 return (0);
185 }
186
187 /*
188 * Allocate an L2T entry for use by a switching rule. Such need to be
189 * explicitly freed and while busy they are not on any hash chain, so normal
190 * address resolution updates do not see them.
191 */
192 struct l2t_entry *
t4_l2t_alloc_switching(struct adapter * sc,uint16_t vlan,uint8_t port,uint8_t * eth_addr)193 t4_l2t_alloc_switching(struct adapter *sc, uint16_t vlan, uint8_t port,
194 uint8_t *eth_addr)
195 {
196 struct l2t_data *d = sc->l2t;
197 struct l2t_entry *e;
198 int rc;
199
200 rw_wlock(&d->lock);
201 e = find_or_alloc_l2e(d, vlan, port, eth_addr);
202 if (e) {
203 if (atomic_load_acq_int(&e->refcnt) == 0) {
204 mtx_lock(&e->lock); /* avoid race with t4_l2t_free */
205 e->wrq = &sc->sge.ctrlq[0];
206 e->iqid = sc->sge.fwq.abs_id;
207 e->state = L2T_STATE_SWITCHING;
208 e->vlan = vlan;
209 e->lport = port;
210 memcpy(e->dmac, eth_addr, ETHER_ADDR_LEN);
211 atomic_store_rel_int(&e->refcnt, 1);
212 atomic_subtract_int(&d->nfree, 1);
213 rc = t4_write_l2e(e, 0);
214 mtx_unlock(&e->lock);
215 if (rc != 0)
216 e = NULL;
217 } else {
218 MPASS(e->vlan == vlan);
219 MPASS(e->lport == port);
220 atomic_add_int(&e->refcnt, 1);
221 }
222 }
223 rw_wunlock(&d->lock);
224 return (e);
225 }
226
227 int
t4_init_l2t(struct adapter * sc,int flags)228 t4_init_l2t(struct adapter *sc, int flags)
229 {
230 int i, l2t_size;
231 struct l2t_data *d;
232
233 l2t_size = sc->vres.l2t.size;
234 if (l2t_size < 2) /* At least 1 bucket for IP and 1 for IPv6 */
235 return (EINVAL);
236
237 d = malloc(sizeof(*d) + l2t_size * sizeof (struct l2t_entry), M_CXGBE,
238 M_ZERO | flags);
239 if (!d)
240 return (ENOMEM);
241
242 d->l2t_size = l2t_size;
243 d->rover = d->l2tab;
244 atomic_store_rel_int(&d->nfree, l2t_size);
245 rw_init(&d->lock, "L2T");
246
247 for (i = 0; i < l2t_size; i++) {
248 struct l2t_entry *e = &d->l2tab[i];
249
250 e->idx = i;
251 e->state = L2T_STATE_UNUSED;
252 mtx_init(&e->lock, "L2T_E", NULL, MTX_DEF);
253 STAILQ_INIT(&e->wr_list);
254 atomic_store_rel_int(&e->refcnt, 0);
255 }
256
257 sc->l2t = d;
258
259 return (0);
260 }
261
262 int
t4_free_l2t(struct l2t_data * d)263 t4_free_l2t(struct l2t_data *d)
264 {
265 int i;
266
267 for (i = 0; i < d->l2t_size; i++)
268 mtx_destroy(&d->l2tab[i].lock);
269 rw_destroy(&d->lock);
270 free(d, M_CXGBE);
271
272 return (0);
273 }
274
275 int
do_l2t_write_rpl(struct sge_iq * iq,const struct rss_header * rss,struct mbuf * m)276 do_l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss,
277 struct mbuf *m)
278 {
279 const struct cpl_l2t_write_rpl *rpl = (const void *)(rss + 1);
280 unsigned int tid = GET_TID(rpl);
281 unsigned int idx = tid % L2T_SIZE;
282
283 if (__predict_false(rpl->status != CPL_ERR_NONE)) {
284 log(LOG_ERR,
285 "Unexpected L2T_WRITE_RPL (%u) for entry at hw_idx %u\n",
286 rpl->status, idx);
287 return (EINVAL);
288 }
289
290 return (0);
291 }
292
293 static inline unsigned int
vlan_prio(const struct l2t_entry * e)294 vlan_prio(const struct l2t_entry *e)
295 {
296 return e->vlan >> 13;
297 }
298
299 static char
l2e_state(const struct l2t_entry * e)300 l2e_state(const struct l2t_entry *e)
301 {
302 switch (e->state) {
303 case L2T_STATE_VALID: return 'V'; /* valid, fast-path entry */
304 case L2T_STATE_STALE: return 'S'; /* needs revalidation, but usable */
305 case L2T_STATE_SYNC_WRITE: return 'W';
306 case L2T_STATE_RESOLVING: return STAILQ_EMPTY(&e->wr_list) ? 'R' : 'A';
307 case L2T_STATE_SWITCHING: return 'X';
308 default: return 'U';
309 }
310 }
311
312 int
sysctl_l2t(SYSCTL_HANDLER_ARGS)313 sysctl_l2t(SYSCTL_HANDLER_ARGS)
314 {
315 struct adapter *sc = arg1;
316 struct l2t_data *l2t = sc->l2t;
317 struct l2t_entry *e;
318 struct sbuf *sb;
319 int rc, i, header = 0;
320 char ip[INET6_ADDRSTRLEN];
321
322 if (l2t == NULL)
323 return (ENXIO);
324
325 rc = sysctl_wire_old_buffer(req, 0);
326 if (rc != 0)
327 return (rc);
328
329 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
330 if (sb == NULL)
331 return (ENOMEM);
332
333 e = &l2t->l2tab[0];
334 for (i = 0; i < l2t->l2t_size; i++, e++) {
335 mtx_lock(&e->lock);
336 if (e->state == L2T_STATE_UNUSED)
337 goto skip;
338
339 if (header == 0) {
340 sbuf_printf(sb, " Idx IP address "
341 "Ethernet address VLAN/P LP State Users Port");
342 header = 1;
343 }
344 if (e->state == L2T_STATE_SWITCHING)
345 ip[0] = 0;
346 else {
347 inet_ntop(e->ipv6 ? AF_INET6 : AF_INET, &e->addr[0],
348 &ip[0], sizeof(ip));
349 }
350
351 /*
352 * XXX: IPv6 addresses may not align properly in the output.
353 */
354 sbuf_printf(sb, "\n%4u %-15s %02x:%02x:%02x:%02x:%02x:%02x %4d"
355 " %u %2u %c %5u %s",
356 e->idx, ip, e->dmac[0], e->dmac[1], e->dmac[2],
357 e->dmac[3], e->dmac[4], e->dmac[5],
358 e->vlan & 0xfff, vlan_prio(e), e->lport,
359 l2e_state(e), atomic_load_acq_int(&e->refcnt),
360 e->ifp ? e->ifp->if_xname : "-");
361 skip:
362 mtx_unlock(&e->lock);
363 }
364
365 rc = sbuf_finish(sb);
366 sbuf_delete(sb);
367
368 return (rc);
369 }
370