t4_l2t.c revision 237263
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: head/sys/dev/cxgbe/t4_l2t.c 237263 2012-06-19 07:34:13Z np $"); 28 29#include "opt_inet.h" 30 31#include <sys/param.h> 32#include <sys/systm.h> 33#include <sys/kernel.h> 34#include <sys/module.h> 35#include <sys/bus.h> 36#include <sys/lock.h> 37#include <sys/mutex.h> 38#include <sys/rwlock.h> 39#include <sys/socket.h> 40#include <sys/sbuf.h> 41#include <netinet/in.h> 42 43#include "common/common.h" 44#include "common/jhash.h" 45#include "common/t4_msg.h" 46#include "t4_l2t.h" 47 48/* 49 * Module locking notes: There is a RW lock protecting the L2 table as a 50 * whole plus a spinlock per L2T entry. Entry lookups and allocations happen 51 * under the protection of the table lock, individual entry changes happen 52 * while holding that entry's spinlock. The table lock nests outside the 53 * entry locks. Allocations of new entries take the table lock as writers so 54 * no other lookups can happen while allocating new entries. Entry updates 55 * take the table lock as readers so multiple entries can be updated in 56 * parallel. An L2T entry can be dropped by decrementing its reference count 57 * and therefore can happen in parallel with entry allocation but no entry 58 * can change state or increment its ref count during allocation as both of 59 * these perform lookups. 60 * 61 * Note: We do not take refereces to ifnets in this module because both 62 * the TOE and the sockets already hold references to the interfaces and the 63 * lifetime of an L2T entry is fully contained in the lifetime of the TOE. 64 */ 65 66/* 67 * Allocate a free L2T entry. Must be called with l2t_data.lock held. 68 */ 69struct l2t_entry * 70t4_alloc_l2e(struct l2t_data *d) 71{ 72 struct l2t_entry *end, *e, **p; 73 74 rw_assert(&d->lock, RA_WLOCKED); 75 76 if (!atomic_load_acq_int(&d->nfree)) 77 return (NULL); 78 79 /* there's definitely a free entry */ 80 for (e = d->rover, end = &d->l2tab[L2T_SIZE]; e != end; ++e) 81 if (atomic_load_acq_int(&e->refcnt) == 0) 82 goto found; 83 84 for (e = d->l2tab; atomic_load_acq_int(&e->refcnt); ++e) 85 continue; 86found: 87 d->rover = e + 1; 88 atomic_subtract_int(&d->nfree, 1); 89 90 /* 91 * The entry we found may be an inactive entry that is 92 * presently in the hash table. We need to remove it. 93 */ 94 if (e->state < L2T_STATE_SWITCHING) { 95 for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next) { 96 if (*p == e) { 97 *p = e->next; 98 e->next = NULL; 99 break; 100 } 101 } 102 } 103 104 e->state = L2T_STATE_UNUSED; 105 return (e); 106} 107 108/* 109 * Write an L2T entry. Must be called with the entry locked. 110 * The write may be synchronous or asynchronous. 111 */ 112int 113t4_write_l2e(struct adapter *sc, struct l2t_entry *e, int sync) 114{ 115 struct wrqe *wr; 116 struct cpl_l2t_write_req *req; 117 118 mtx_assert(&e->lock, MA_OWNED); 119 120 wr = alloc_wrqe(sizeof(*req), &sc->sge.mgmtq); 121 if (wr == NULL) 122 return (ENOMEM); 123 req = wrtod(wr); 124 125 INIT_TP_WR(req, 0); 126 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, e->idx | 127 V_SYNC_WR(sync) | V_TID_QID(sc->sge.fwq.abs_id))); 128 req->params = htons(V_L2T_W_PORT(e->lport) | V_L2T_W_NOREPLY(!sync)); 129 req->l2t_idx = htons(e->idx); 130 req->vlan = htons(e->vlan); 131 memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac)); 132 133 t4_wrq_tx(sc, wr); 134 135 if (sync && e->state != L2T_STATE_SWITCHING) 136 e->state = L2T_STATE_SYNC_WRITE; 137 138 return (0); 139} 140 141/* 142 * Allocate an L2T entry for use by a switching rule. Such need to be 143 * explicitly freed and while busy they are not on any hash chain, so normal 144 * address resolution updates do not see them. 145 */ 146struct l2t_entry * 147t4_l2t_alloc_switching(struct l2t_data *d) 148{ 149 struct l2t_entry *e; 150 151 rw_rlock(&d->lock); 152 e = t4_alloc_l2e(d); 153 if (e) { 154 mtx_lock(&e->lock); /* avoid race with t4_l2t_free */ 155 e->state = L2T_STATE_SWITCHING; 156 atomic_store_rel_int(&e->refcnt, 1); 157 mtx_unlock(&e->lock); 158 } 159 rw_runlock(&d->lock); 160 return e; 161} 162 163/* 164 * Sets/updates the contents of a switching L2T entry that has been allocated 165 * with an earlier call to @t4_l2t_alloc_switching. 166 */ 167int 168t4_l2t_set_switching(struct adapter *sc, struct l2t_entry *e, uint16_t vlan, 169 uint8_t port, uint8_t *eth_addr) 170{ 171 int rc; 172 173 e->vlan = vlan; 174 e->lport = port; 175 memcpy(e->dmac, eth_addr, ETHER_ADDR_LEN); 176 mtx_lock(&e->lock); 177 rc = t4_write_l2e(sc, e, 0); 178 mtx_unlock(&e->lock); 179 return (rc); 180} 181 182int 183t4_init_l2t(struct adapter *sc, int flags) 184{ 185 int i; 186 struct l2t_data *d; 187 188 d = malloc(sizeof(*d), M_CXGBE, M_ZERO | flags); 189 if (!d) 190 return (ENOMEM); 191 192 d->rover = d->l2tab; 193 atomic_store_rel_int(&d->nfree, L2T_SIZE); 194 rw_init(&d->lock, "L2T"); 195 196 for (i = 0; i < L2T_SIZE; i++) { 197 struct l2t_entry *e = &d->l2tab[i]; 198 199 e->idx = i; 200 e->state = L2T_STATE_UNUSED; 201 mtx_init(&e->lock, "L2T_E", NULL, MTX_DEF); 202 STAILQ_INIT(&e->wr_list); 203 atomic_store_rel_int(&e->refcnt, 0); 204 } 205 206 sc->l2t = d; 207 t4_register_cpl_handler(sc, CPL_L2T_WRITE_RPL, do_l2t_write_rpl); 208 209 return (0); 210} 211 212int 213t4_free_l2t(struct l2t_data *d) 214{ 215 int i; 216 217 for (i = 0; i < L2T_SIZE; i++) 218 mtx_destroy(&d->l2tab[i].lock); 219 rw_destroy(&d->lock); 220 free(d, M_CXGBE); 221 222 return (0); 223} 224 225int 226do_l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss, 227 struct mbuf *m) 228{ 229 const struct cpl_l2t_write_rpl *rpl = (const void *)(rss + 1); 230 unsigned int tid = GET_TID(rpl); 231 unsigned int idx = tid & (L2T_SIZE - 1); 232 233 if (__predict_false(rpl->status != CPL_ERR_NONE)) { 234 log(LOG_ERR, 235 "Unexpected L2T_WRITE_RPL status %u for entry %u\n", 236 rpl->status, idx); 237 return (EINVAL); 238 } 239 240 return (0); 241} 242 243#ifdef SBUF_DRAIN 244static inline unsigned int 245vlan_prio(const struct l2t_entry *e) 246{ 247 return e->vlan >> 13; 248} 249 250static char 251l2e_state(const struct l2t_entry *e) 252{ 253 switch (e->state) { 254 case L2T_STATE_VALID: return 'V'; /* valid, fast-path entry */ 255 case L2T_STATE_STALE: return 'S'; /* needs revalidation, but usable */ 256 case L2T_STATE_SYNC_WRITE: return 'W'; 257 case L2T_STATE_RESOLVING: return STAILQ_EMPTY(&e->wr_list) ? 'R' : 'A'; 258 case L2T_STATE_SWITCHING: return 'X'; 259 default: return 'U'; 260 } 261} 262 263int 264sysctl_l2t(SYSCTL_HANDLER_ARGS) 265{ 266 struct adapter *sc = arg1; 267 struct l2t_data *l2t = sc->l2t; 268 struct l2t_entry *e; 269 struct sbuf *sb; 270 int rc, i, header = 0; 271 char ip[60]; 272 273 if (l2t == NULL) 274 return (ENXIO); 275 276 rc = sysctl_wire_old_buffer(req, 0); 277 if (rc != 0) 278 return (rc); 279 280 sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req); 281 if (sb == NULL) 282 return (ENOMEM); 283 284 e = &l2t->l2tab[0]; 285 for (i = 0; i < L2T_SIZE; i++, e++) { 286 mtx_lock(&e->lock); 287 if (e->state == L2T_STATE_UNUSED) 288 goto skip; 289 290 if (header == 0) { 291 sbuf_printf(sb, " Idx IP address " 292 "Ethernet address VLAN/P LP State Users Port"); 293 header = 1; 294 } 295 if (e->state == L2T_STATE_SWITCHING) 296 ip[0] = 0; 297 else 298 snprintf(ip, sizeof(ip), "%s", 299 inet_ntoa(*(struct in_addr *)&e->addr)); 300 301 /* XXX: e->ifp may not be around */ 302 sbuf_printf(sb, "\n%4u %-15s %02x:%02x:%02x:%02x:%02x:%02x %4d" 303 " %u %2u %c %5u %s", 304 e->idx, ip, e->dmac[0], e->dmac[1], e->dmac[2], 305 e->dmac[3], e->dmac[4], e->dmac[5], 306 e->vlan & 0xfff, vlan_prio(e), e->lport, 307 l2e_state(e), atomic_load_acq_int(&e->refcnt), 308 e->ifp->if_xname); 309skip: 310 mtx_unlock(&e->lock); 311 } 312 313 rc = sbuf_finish(sb); 314 sbuf_delete(sb); 315 316 return (rc); 317} 318#endif 319