xref: /freebsd/sys/dev/cxgbe/t4_l2t.c (revision c6ec7d31830ab1c80edae95ad5e4b9dba10c47ac)
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/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/module.h>
36 #include <sys/bus.h>
37 #include <sys/lock.h>
38 #include <sys/mutex.h>
39 #include <sys/rwlock.h>
40 #include <sys/socket.h>
41 #include <sys/sbuf.h>
42 #include <netinet/in.h>
43 
44 #include "common/common.h"
45 #include "common/jhash.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 refereces 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 *
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[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 /*
110  * Write an L2T entry.  Must be called with the entry locked.
111  * The write may be synchronous or asynchronous.
112  */
113 int
114 t4_write_l2e(struct adapter *sc, struct l2t_entry *e, int sync)
115 {
116 	struct wrqe *wr;
117 	struct cpl_l2t_write_req *req;
118 
119 	mtx_assert(&e->lock, MA_OWNED);
120 
121 	wr = alloc_wrqe(sizeof(*req), &sc->sge.mgmtq);
122 	if (wr == NULL)
123 		return (ENOMEM);
124 	req = wrtod(wr);
125 
126 	INIT_TP_WR(req, 0);
127 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, e->idx |
128 	    V_SYNC_WR(sync) | V_TID_QID(sc->sge.fwq.abs_id)));
129 	req->params = htons(V_L2T_W_PORT(e->lport) | V_L2T_W_NOREPLY(!sync));
130 	req->l2t_idx = htons(e->idx);
131 	req->vlan = htons(e->vlan);
132 	memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
133 
134 	t4_wrq_tx(sc, wr);
135 
136 	if (sync && e->state != L2T_STATE_SWITCHING)
137 		e->state = L2T_STATE_SYNC_WRITE;
138 
139 	return (0);
140 }
141 
142 /*
143  * Allocate an L2T entry for use by a switching rule.  Such need to be
144  * explicitly freed and while busy they are not on any hash chain, so normal
145  * address resolution updates do not see them.
146  */
147 struct l2t_entry *
148 t4_l2t_alloc_switching(struct l2t_data *d)
149 {
150 	struct l2t_entry *e;
151 
152 	rw_wlock(&d->lock);
153 	e = t4_alloc_l2e(d);
154 	if (e) {
155 		mtx_lock(&e->lock);          /* avoid race with t4_l2t_free */
156 		e->state = L2T_STATE_SWITCHING;
157 		atomic_store_rel_int(&e->refcnt, 1);
158 		mtx_unlock(&e->lock);
159 	}
160 	rw_wunlock(&d->lock);
161 	return e;
162 }
163 
164 /*
165  * Sets/updates the contents of a switching L2T entry that has been allocated
166  * with an earlier call to @t4_l2t_alloc_switching.
167  */
168 int
169 t4_l2t_set_switching(struct adapter *sc, struct l2t_entry *e, uint16_t vlan,
170     uint8_t port, uint8_t *eth_addr)
171 {
172 	int rc;
173 
174 	e->vlan = vlan;
175 	e->lport = port;
176 	memcpy(e->dmac, eth_addr, ETHER_ADDR_LEN);
177 	mtx_lock(&e->lock);
178 	rc = t4_write_l2e(sc, e, 0);
179 	mtx_unlock(&e->lock);
180 	return (rc);
181 }
182 
183 int
184 t4_init_l2t(struct adapter *sc, int flags)
185 {
186 	int i;
187 	struct l2t_data *d;
188 
189 	d = malloc(sizeof(*d), M_CXGBE, M_ZERO | flags);
190 	if (!d)
191 		return (ENOMEM);
192 
193 	d->rover = d->l2tab;
194 	atomic_store_rel_int(&d->nfree, L2T_SIZE);
195 	rw_init(&d->lock, "L2T");
196 
197 	for (i = 0; i < L2T_SIZE; i++) {
198 		struct l2t_entry *e = &d->l2tab[i];
199 
200 		e->idx = i;
201 		e->state = L2T_STATE_UNUSED;
202 		mtx_init(&e->lock, "L2T_E", NULL, MTX_DEF);
203 		STAILQ_INIT(&e->wr_list);
204 		atomic_store_rel_int(&e->refcnt, 0);
205 	}
206 
207 	sc->l2t = d;
208 	t4_register_cpl_handler(sc, CPL_L2T_WRITE_RPL, do_l2t_write_rpl);
209 
210 	return (0);
211 }
212 
213 int
214 t4_free_l2t(struct l2t_data *d)
215 {
216 	int i;
217 
218 	for (i = 0; i < L2T_SIZE; i++)
219 		mtx_destroy(&d->l2tab[i].lock);
220 	rw_destroy(&d->lock);
221 	free(d, M_CXGBE);
222 
223 	return (0);
224 }
225 
226 int
227 do_l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss,
228     struct mbuf *m)
229 {
230 	const struct cpl_l2t_write_rpl *rpl = (const void *)(rss + 1);
231 	unsigned int tid = GET_TID(rpl);
232 	unsigned int idx = tid & (L2T_SIZE - 1);
233 
234 	if (__predict_false(rpl->status != CPL_ERR_NONE)) {
235 		log(LOG_ERR,
236 		    "Unexpected L2T_WRITE_RPL status %u for entry %u\n",
237 		    rpl->status, idx);
238 		return (EINVAL);
239 	}
240 
241 	return (0);
242 }
243 
244 #ifdef SBUF_DRAIN
245 static inline unsigned int
246 vlan_prio(const struct l2t_entry *e)
247 {
248 	return e->vlan >> 13;
249 }
250 
251 static char
252 l2e_state(const struct l2t_entry *e)
253 {
254 	switch (e->state) {
255 	case L2T_STATE_VALID: return 'V';  /* valid, fast-path entry */
256 	case L2T_STATE_STALE: return 'S';  /* needs revalidation, but usable */
257 	case L2T_STATE_SYNC_WRITE: return 'W';
258 	case L2T_STATE_RESOLVING: return STAILQ_EMPTY(&e->wr_list) ? 'R' : 'A';
259 	case L2T_STATE_SWITCHING: return 'X';
260 	default: return 'U';
261 	}
262 }
263 
264 int
265 sysctl_l2t(SYSCTL_HANDLER_ARGS)
266 {
267 	struct adapter *sc = arg1;
268 	struct l2t_data *l2t = sc->l2t;
269 	struct l2t_entry *e;
270 	struct sbuf *sb;
271 	int rc, i, header = 0;
272 	char ip[60];
273 
274 	if (l2t == NULL)
275 		return (ENXIO);
276 
277 	rc = sysctl_wire_old_buffer(req, 0);
278 	if (rc != 0)
279 		return (rc);
280 
281 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
282 	if (sb == NULL)
283 		return (ENOMEM);
284 
285 	e = &l2t->l2tab[0];
286 	for (i = 0; i < L2T_SIZE; i++, e++) {
287 		mtx_lock(&e->lock);
288 		if (e->state == L2T_STATE_UNUSED)
289 			goto skip;
290 
291 		if (header == 0) {
292 			sbuf_printf(sb, " Idx IP address      "
293 			    "Ethernet address  VLAN/P LP State Users Port");
294 			header = 1;
295 		}
296 		if (e->state == L2T_STATE_SWITCHING)
297 			ip[0] = 0;
298 		else
299 			snprintf(ip, sizeof(ip), "%s",
300 			    inet_ntoa(*(struct in_addr *)&e->addr));
301 
302 		/* XXX: e->ifp may not be around */
303 		sbuf_printf(sb, "\n%4u %-15s %02x:%02x:%02x:%02x:%02x:%02x %4d"
304 			   " %u %2u   %c   %5u %s",
305 			   e->idx, ip, e->dmac[0], e->dmac[1], e->dmac[2],
306 			   e->dmac[3], e->dmac[4], e->dmac[5],
307 			   e->vlan & 0xfff, vlan_prio(e), e->lport,
308 			   l2e_state(e), atomic_load_acq_int(&e->refcnt),
309 			   e->ifp->if_xname);
310 skip:
311 		mtx_unlock(&e->lock);
312 	}
313 
314 	rc = sbuf_finish(sb);
315 	sbuf_delete(sb);
316 
317 	return (rc);
318 }
319 #endif
320