1 /*- 2 * Copyright (c) 2012 Chelsio Communications, Inc. 3 * All rights reserved. 4 * Written by: Navdeep Parhar <np@FreeBSD.org> 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28 #include <sys/cdefs.h> 29 __FBSDID("$FreeBSD$"); 30 31 #include "opt_inet.h" 32 #include "opt_inet6.h" 33 34 #include <sys/param.h> 35 #include <sys/types.h> 36 #include <sys/systm.h> 37 #include <sys/kernel.h> 38 #include <sys/ktr.h> 39 #include <sys/module.h> 40 #include <sys/protosw.h> 41 #include <sys/domain.h> 42 #include <sys/socket.h> 43 #include <sys/socketvar.h> 44 #include <net/if.h> 45 #include <netinet/in.h> 46 #include <netinet/in_pcb.h> 47 #include <netinet/in_var.h> 48 #include <netinet/ip.h> 49 #include <netinet/ip6.h> 50 #include <netinet/tcp_var.h> 51 #include <netinet6/scope6_var.h> 52 #define TCPSTATES 53 #include <netinet/tcp_fsm.h> 54 #include <netinet/toecore.h> 55 56 #ifdef TCP_OFFLOAD 57 #include "common/common.h" 58 #include "common/t4_msg.h" 59 #include "common/t4_regs.h" 60 #include "tom/t4_tom_l2t.h" 61 #include "tom/t4_tom.h" 62 63 static struct protosw ddp_protosw; 64 static struct pr_usrreqs ddp_usrreqs; 65 66 static struct protosw ddp6_protosw; 67 static struct pr_usrreqs ddp6_usrreqs; 68 69 /* Module ops */ 70 static int t4_tom_mod_load(void); 71 static int t4_tom_mod_unload(void); 72 static int t4_tom_modevent(module_t, int, void *); 73 74 /* ULD ops and helpers */ 75 static int t4_tom_activate(struct adapter *); 76 static int t4_tom_deactivate(struct adapter *); 77 78 static struct uld_info tom_uld_info = { 79 .uld_id = ULD_TOM, 80 .activate = t4_tom_activate, 81 .deactivate = t4_tom_deactivate, 82 }; 83 84 static void queue_tid_release(struct adapter *, int); 85 static void release_offload_resources(struct toepcb *); 86 static int alloc_tid_tabs(struct tid_info *); 87 static void free_tid_tabs(struct tid_info *); 88 static int add_lip(struct adapter *, struct in6_addr *); 89 static int delete_lip(struct adapter *, struct in6_addr *); 90 static struct clip_entry *search_lip(struct tom_data *, struct in6_addr *); 91 static void init_clip_table(struct adapter *, struct tom_data *); 92 static void destroy_clip_table(struct adapter *, struct tom_data *); 93 static void free_tom_data(struct adapter *, struct tom_data *); 94 95 struct toepcb * 96 alloc_toepcb(struct port_info *pi, int txqid, int rxqid, int flags) 97 { 98 struct adapter *sc = pi->adapter; 99 struct toepcb *toep; 100 int tx_credits, txsd_total, len; 101 102 /* 103 * The firmware counts tx work request credits in units of 16 bytes 104 * each. Reserve room for an ABORT_REQ so the driver never has to worry 105 * about tx credits if it wants to abort a connection. 106 */ 107 tx_credits = sc->params.ofldq_wr_cred; 108 tx_credits -= howmany(sizeof(struct cpl_abort_req), 16); 109 110 /* 111 * Shortest possible tx work request is a fw_ofld_tx_data_wr + 1 byte 112 * immediate payload, and firmware counts tx work request credits in 113 * units of 16 byte. Calculate the maximum work requests possible. 114 */ 115 txsd_total = tx_credits / 116 howmany((sizeof(struct fw_ofld_tx_data_wr) + 1), 16); 117 118 if (txqid < 0) 119 txqid = (arc4random() % pi->nofldtxq) + pi->first_ofld_txq; 120 KASSERT(txqid >= pi->first_ofld_txq && 121 txqid < pi->first_ofld_txq + pi->nofldtxq, 122 ("%s: txqid %d for port %p (first %d, n %d)", __func__, txqid, pi, 123 pi->first_ofld_txq, pi->nofldtxq)); 124 125 if (rxqid < 0) 126 rxqid = (arc4random() % pi->nofldrxq) + pi->first_ofld_rxq; 127 KASSERT(rxqid >= pi->first_ofld_rxq && 128 rxqid < pi->first_ofld_rxq + pi->nofldrxq, 129 ("%s: rxqid %d for port %p (first %d, n %d)", __func__, rxqid, pi, 130 pi->first_ofld_rxq, pi->nofldrxq)); 131 132 len = offsetof(struct toepcb, txsd) + 133 txsd_total * sizeof(struct ofld_tx_sdesc); 134 135 toep = malloc(len, M_CXGBE, M_ZERO | flags); 136 if (toep == NULL) 137 return (NULL); 138 139 toep->td = sc->tom_softc; 140 toep->port = pi; 141 toep->tx_credits = tx_credits; 142 toep->ofld_txq = &sc->sge.ofld_txq[txqid]; 143 toep->ofld_rxq = &sc->sge.ofld_rxq[rxqid]; 144 toep->ctrlq = &sc->sge.ctrlq[pi->port_id]; 145 toep->txsd_total = txsd_total; 146 toep->txsd_avail = txsd_total; 147 toep->txsd_pidx = 0; 148 toep->txsd_cidx = 0; 149 150 return (toep); 151 } 152 153 void 154 free_toepcb(struct toepcb *toep) 155 { 156 157 KASSERT(!(toep->flags & TPF_ATTACHED), 158 ("%s: attached to an inpcb", __func__)); 159 KASSERT(!(toep->flags & TPF_CPL_PENDING), 160 ("%s: CPL pending", __func__)); 161 162 free(toep, M_CXGBE); 163 } 164 165 /* 166 * Set up the socket for TCP offload. 167 */ 168 void 169 offload_socket(struct socket *so, struct toepcb *toep) 170 { 171 struct tom_data *td = toep->td; 172 struct inpcb *inp = sotoinpcb(so); 173 struct tcpcb *tp = intotcpcb(inp); 174 struct sockbuf *sb; 175 176 INP_WLOCK_ASSERT(inp); 177 178 /* Update socket */ 179 sb = &so->so_snd; 180 SOCKBUF_LOCK(sb); 181 sb->sb_flags |= SB_NOCOALESCE; 182 SOCKBUF_UNLOCK(sb); 183 sb = &so->so_rcv; 184 SOCKBUF_LOCK(sb); 185 sb->sb_flags |= SB_NOCOALESCE; 186 if (toep->ulp_mode == ULP_MODE_TCPDDP) { 187 if (inp->inp_vflag & INP_IPV6) 188 so->so_proto = &ddp6_protosw; 189 else 190 so->so_proto = &ddp_protosw; 191 } 192 SOCKBUF_UNLOCK(sb); 193 194 /* Update TCP PCB */ 195 tp->tod = &td->tod; 196 tp->t_toe = toep; 197 tp->t_flags |= TF_TOE; 198 199 /* Install an extra hold on inp */ 200 toep->inp = inp; 201 toep->flags |= TPF_ATTACHED; 202 in_pcbref(inp); 203 204 /* Add the TOE PCB to the active list */ 205 mtx_lock(&td->toep_list_lock); 206 TAILQ_INSERT_HEAD(&td->toep_list, toep, link); 207 mtx_unlock(&td->toep_list_lock); 208 } 209 210 /* This is _not_ the normal way to "unoffload" a socket. */ 211 void 212 undo_offload_socket(struct socket *so) 213 { 214 struct inpcb *inp = sotoinpcb(so); 215 struct tcpcb *tp = intotcpcb(inp); 216 struct toepcb *toep = tp->t_toe; 217 struct tom_data *td = toep->td; 218 struct sockbuf *sb; 219 220 INP_WLOCK_ASSERT(inp); 221 222 sb = &so->so_snd; 223 SOCKBUF_LOCK(sb); 224 sb->sb_flags &= ~SB_NOCOALESCE; 225 SOCKBUF_UNLOCK(sb); 226 sb = &so->so_rcv; 227 SOCKBUF_LOCK(sb); 228 sb->sb_flags &= ~SB_NOCOALESCE; 229 SOCKBUF_UNLOCK(sb); 230 231 tp->tod = NULL; 232 tp->t_toe = NULL; 233 tp->t_flags &= ~TF_TOE; 234 235 toep->inp = NULL; 236 toep->flags &= ~TPF_ATTACHED; 237 if (in_pcbrele_wlocked(inp)) 238 panic("%s: inp freed.", __func__); 239 240 mtx_lock(&td->toep_list_lock); 241 TAILQ_REMOVE(&td->toep_list, toep, link); 242 mtx_unlock(&td->toep_list_lock); 243 } 244 245 static void 246 release_offload_resources(struct toepcb *toep) 247 { 248 struct tom_data *td = toep->td; 249 struct adapter *sc = td_adapter(td); 250 int tid = toep->tid; 251 252 KASSERT(!(toep->flags & TPF_CPL_PENDING), 253 ("%s: %p has CPL pending.", __func__, toep)); 254 KASSERT(!(toep->flags & TPF_ATTACHED), 255 ("%s: %p is still attached.", __func__, toep)); 256 257 CTR5(KTR_CXGBE, "%s: toep %p (tid %d, l2te %p, ce %p)", 258 __func__, toep, tid, toep->l2te, toep->ce); 259 260 if (toep->ulp_mode == ULP_MODE_TCPDDP) 261 release_ddp_resources(toep); 262 263 if (toep->l2te) 264 t4_l2t_release(toep->l2te); 265 266 if (tid >= 0) { 267 remove_tid(sc, tid); 268 release_tid(sc, tid, toep->ctrlq); 269 } 270 271 if (toep->ce) 272 release_lip(td, toep->ce); 273 274 mtx_lock(&td->toep_list_lock); 275 TAILQ_REMOVE(&td->toep_list, toep, link); 276 mtx_unlock(&td->toep_list_lock); 277 278 free_toepcb(toep); 279 } 280 281 /* 282 * The kernel is done with the TCP PCB and this is our opportunity to unhook the 283 * toepcb hanging off of it. If the TOE driver is also done with the toepcb (no 284 * pending CPL) then it is time to release all resources tied to the toepcb. 285 * 286 * Also gets called when an offloaded active open fails and the TOM wants the 287 * kernel to take the TCP PCB back. 288 */ 289 static void 290 t4_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp) 291 { 292 #if defined(KTR) || defined(INVARIANTS) 293 struct inpcb *inp = tp->t_inpcb; 294 #endif 295 struct toepcb *toep = tp->t_toe; 296 297 INP_WLOCK_ASSERT(inp); 298 299 KASSERT(toep != NULL, ("%s: toep is NULL", __func__)); 300 KASSERT(toep->flags & TPF_ATTACHED, 301 ("%s: not attached", __func__)); 302 303 #ifdef KTR 304 if (tp->t_state == TCPS_SYN_SENT) { 305 CTR6(KTR_CXGBE, "%s: atid %d, toep %p (0x%x), inp %p (0x%x)", 306 __func__, toep->tid, toep, toep->flags, inp, 307 inp->inp_flags); 308 } else { 309 CTR6(KTR_CXGBE, 310 "t4_pcb_detach: tid %d (%s), toep %p (0x%x), inp %p (0x%x)", 311 toep->tid, tcpstates[tp->t_state], toep, toep->flags, inp, 312 inp->inp_flags); 313 } 314 #endif 315 316 tp->t_toe = NULL; 317 tp->t_flags &= ~TF_TOE; 318 toep->flags &= ~TPF_ATTACHED; 319 320 if (!(toep->flags & TPF_CPL_PENDING)) 321 release_offload_resources(toep); 322 } 323 324 /* 325 * The TOE driver will not receive any more CPLs for the tid associated with the 326 * toepcb; release the hold on the inpcb. 327 */ 328 void 329 final_cpl_received(struct toepcb *toep) 330 { 331 struct inpcb *inp = toep->inp; 332 333 KASSERT(inp != NULL, ("%s: inp is NULL", __func__)); 334 INP_WLOCK_ASSERT(inp); 335 KASSERT(toep->flags & TPF_CPL_PENDING, 336 ("%s: CPL not pending already?", __func__)); 337 338 CTR6(KTR_CXGBE, "%s: tid %d, toep %p (0x%x), inp %p (0x%x)", 339 __func__, toep->tid, toep, toep->flags, inp, inp->inp_flags); 340 341 toep->inp = NULL; 342 toep->flags &= ~TPF_CPL_PENDING; 343 344 if (!(toep->flags & TPF_ATTACHED)) 345 release_offload_resources(toep); 346 347 if (!in_pcbrele_wlocked(inp)) 348 INP_WUNLOCK(inp); 349 } 350 351 void 352 insert_tid(struct adapter *sc, int tid, void *ctx) 353 { 354 struct tid_info *t = &sc->tids; 355 356 t->tid_tab[tid] = ctx; 357 atomic_add_int(&t->tids_in_use, 1); 358 } 359 360 void * 361 lookup_tid(struct adapter *sc, int tid) 362 { 363 struct tid_info *t = &sc->tids; 364 365 return (t->tid_tab[tid]); 366 } 367 368 void 369 update_tid(struct adapter *sc, int tid, void *ctx) 370 { 371 struct tid_info *t = &sc->tids; 372 373 t->tid_tab[tid] = ctx; 374 } 375 376 void 377 remove_tid(struct adapter *sc, int tid) 378 { 379 struct tid_info *t = &sc->tids; 380 381 t->tid_tab[tid] = NULL; 382 atomic_subtract_int(&t->tids_in_use, 1); 383 } 384 385 void 386 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq) 387 { 388 struct wrqe *wr; 389 struct cpl_tid_release *req; 390 391 wr = alloc_wrqe(sizeof(*req), ctrlq); 392 if (wr == NULL) { 393 queue_tid_release(sc, tid); /* defer */ 394 return; 395 } 396 req = wrtod(wr); 397 398 INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid); 399 400 t4_wrq_tx(sc, wr); 401 } 402 403 static void 404 queue_tid_release(struct adapter *sc, int tid) 405 { 406 407 CXGBE_UNIMPLEMENTED("deferred tid release"); 408 } 409 410 /* 411 * What mtu_idx to use, given a 4-tuple and/or an MSS cap 412 */ 413 int 414 find_best_mtu_idx(struct adapter *sc, struct in_conninfo *inc, int pmss) 415 { 416 unsigned short *mtus = &sc->params.mtus[0]; 417 int i, mss, n; 418 419 KASSERT(inc != NULL || pmss > 0, 420 ("%s: at least one of inc/pmss must be specified", __func__)); 421 422 mss = inc ? tcp_mssopt(inc) : pmss; 423 if (pmss > 0 && mss > pmss) 424 mss = pmss; 425 426 if (inc->inc_flags & INC_ISIPV6) 427 n = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 428 else 429 n = sizeof(struct ip) + sizeof(struct tcphdr); 430 431 for (i = 0; i < NMTUS - 1 && mtus[i + 1] <= mss + n; i++) 432 continue; 433 434 return (i); 435 } 436 437 /* 438 * Determine the receive window size for a socket. 439 */ 440 u_long 441 select_rcv_wnd(struct socket *so) 442 { 443 unsigned long wnd; 444 445 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 446 447 wnd = sbspace(&so->so_rcv); 448 if (wnd < MIN_RCV_WND) 449 wnd = MIN_RCV_WND; 450 451 return min(wnd, MAX_RCV_WND); 452 } 453 454 int 455 select_rcv_wscale(void) 456 { 457 int wscale = 0; 458 unsigned long space = sb_max; 459 460 if (space > MAX_RCV_WND) 461 space = MAX_RCV_WND; 462 463 while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < space) 464 wscale++; 465 466 return (wscale); 467 } 468 469 extern int always_keepalive; 470 #define VIID_SMACIDX(v) (((unsigned int)(v) & 0x7f) << 1) 471 472 /* 473 * socket so could be a listening socket too. 474 */ 475 uint64_t 476 calc_opt0(struct socket *so, struct port_info *pi, struct l2t_entry *e, 477 int mtu_idx, int rscale, int rx_credits, int ulp_mode) 478 { 479 uint64_t opt0; 480 481 KASSERT(rx_credits <= M_RCV_BUFSIZ, 482 ("%s: rcv_bufsiz too high", __func__)); 483 484 opt0 = F_TCAM_BYPASS | V_WND_SCALE(rscale) | V_MSS_IDX(mtu_idx) | 485 V_ULP_MODE(ulp_mode) | V_RCV_BUFSIZ(rx_credits); 486 487 if (so != NULL) { 488 struct inpcb *inp = sotoinpcb(so); 489 struct tcpcb *tp = intotcpcb(inp); 490 int keepalive = always_keepalive || 491 so_options_get(so) & SO_KEEPALIVE; 492 493 opt0 |= V_NAGLE((tp->t_flags & TF_NODELAY) == 0); 494 opt0 |= V_KEEP_ALIVE(keepalive != 0); 495 } 496 497 if (e != NULL) 498 opt0 |= V_L2T_IDX(e->idx); 499 500 if (pi != NULL) { 501 opt0 |= V_SMAC_SEL(VIID_SMACIDX(pi->viid)); 502 opt0 |= V_TX_CHAN(pi->tx_chan); 503 } 504 505 return htobe64(opt0); 506 } 507 508 #define FILTER_SEL_WIDTH_P_FC (3 + 1) 509 #define FILTER_SEL_WIDTH_VIN_P_FC (6 + 7 + FILTER_SEL_WIDTH_P_FC) 510 #define FILTER_SEL_WIDTH_TAG_P_FC (3 + FILTER_SEL_WIDTH_VIN_P_FC) 511 #define FILTER_SEL_WIDTH_VLD_TAG_P_FC (1 + FILTER_SEL_WIDTH_TAG_P_FC) 512 #define VLAN_NONE 0xfff 513 #define FILTER_SEL_VLAN_NONE 0xffff 514 515 uint64_t 516 select_ntuple(struct port_info *pi, struct l2t_entry *e, uint32_t filter_mode) 517 { 518 uint16_t viid = pi->viid; 519 uint32_t ntuple = 0; 520 521 if (filter_mode == HW_TPL_FR_MT_PR_IV_P_FC) { 522 if (e->vlan == VLAN_NONE) 523 ntuple |= FILTER_SEL_VLAN_NONE << FILTER_SEL_WIDTH_P_FC; 524 else { 525 ntuple |= e->vlan << FILTER_SEL_WIDTH_P_FC; 526 ntuple |= 1 << FILTER_SEL_WIDTH_VLD_TAG_P_FC; 527 } 528 ntuple |= e->lport << S_PORT; 529 ntuple |= IPPROTO_TCP << FILTER_SEL_WIDTH_VLD_TAG_P_FC; 530 } else if (filter_mode == HW_TPL_FR_MT_PR_OV_P_FC) { 531 ntuple |= G_FW_VIID_VIN(viid) << FILTER_SEL_WIDTH_P_FC; 532 ntuple |= G_FW_VIID_PFN(viid) << FILTER_SEL_WIDTH_VIN_P_FC; 533 ntuple |= G_FW_VIID_VIVLD(viid) << FILTER_SEL_WIDTH_TAG_P_FC; 534 ntuple |= e->lport << S_PORT; 535 ntuple |= IPPROTO_TCP << FILTER_SEL_WIDTH_VLD_TAG_P_FC; 536 } 537 538 if (is_t4(pi->adapter)) 539 return (htobe32(ntuple)); 540 else 541 return (htobe64(V_FILTER_TUPLE(ntuple))); 542 } 543 544 void 545 set_tcpddp_ulp_mode(struct toepcb *toep) 546 { 547 548 toep->ulp_mode = ULP_MODE_TCPDDP; 549 toep->ddp_flags = DDP_OK; 550 toep->ddp_score = DDP_LOW_SCORE; 551 } 552 553 int 554 negative_advice(int status) 555 { 556 557 return (status == CPL_ERR_RTX_NEG_ADVICE || 558 status == CPL_ERR_PERSIST_NEG_ADVICE || 559 status == CPL_ERR_KEEPALV_NEG_ADVICE); 560 } 561 562 static int 563 alloc_tid_tabs(struct tid_info *t) 564 { 565 size_t size; 566 unsigned int i; 567 568 size = t->ntids * sizeof(*t->tid_tab) + 569 t->natids * sizeof(*t->atid_tab) + 570 t->nstids * sizeof(*t->stid_tab); 571 572 t->tid_tab = malloc(size, M_CXGBE, M_ZERO | M_NOWAIT); 573 if (t->tid_tab == NULL) 574 return (ENOMEM); 575 576 mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF); 577 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids]; 578 t->afree = t->atid_tab; 579 t->atids_in_use = 0; 580 for (i = 1; i < t->natids; i++) 581 t->atid_tab[i - 1].next = &t->atid_tab[i]; 582 t->atid_tab[t->natids - 1].next = NULL; 583 584 mtx_init(&t->stid_lock, "stid lock", NULL, MTX_DEF); 585 t->stid_tab = (struct listen_ctx **)&t->atid_tab[t->natids]; 586 t->stids_in_use = 0; 587 TAILQ_INIT(&t->stids); 588 t->nstids_free_head = t->nstids; 589 590 atomic_store_rel_int(&t->tids_in_use, 0); 591 592 return (0); 593 } 594 595 static void 596 free_tid_tabs(struct tid_info *t) 597 { 598 KASSERT(t->tids_in_use == 0, 599 ("%s: %d tids still in use.", __func__, t->tids_in_use)); 600 KASSERT(t->atids_in_use == 0, 601 ("%s: %d atids still in use.", __func__, t->atids_in_use)); 602 KASSERT(t->stids_in_use == 0, 603 ("%s: %d tids still in use.", __func__, t->stids_in_use)); 604 605 free(t->tid_tab, M_CXGBE); 606 t->tid_tab = NULL; 607 608 if (mtx_initialized(&t->atid_lock)) 609 mtx_destroy(&t->atid_lock); 610 if (mtx_initialized(&t->stid_lock)) 611 mtx_destroy(&t->stid_lock); 612 } 613 614 static int 615 add_lip(struct adapter *sc, struct in6_addr *lip) 616 { 617 struct fw_clip_cmd c; 618 619 ASSERT_SYNCHRONIZED_OP(sc); 620 /* mtx_assert(&td->clip_table_lock, MA_OWNED); */ 621 622 memset(&c, 0, sizeof(c)); 623 c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST | 624 F_FW_CMD_WRITE); 625 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_ALLOC | FW_LEN16(c)); 626 c.ip_hi = *(uint64_t *)&lip->s6_addr[0]; 627 c.ip_lo = *(uint64_t *)&lip->s6_addr[8]; 628 629 return (t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c)); 630 } 631 632 static int 633 delete_lip(struct adapter *sc, struct in6_addr *lip) 634 { 635 struct fw_clip_cmd c; 636 637 ASSERT_SYNCHRONIZED_OP(sc); 638 /* mtx_assert(&td->clip_table_lock, MA_OWNED); */ 639 640 memset(&c, 0, sizeof(c)); 641 c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST | 642 F_FW_CMD_READ); 643 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_FREE | FW_LEN16(c)); 644 c.ip_hi = *(uint64_t *)&lip->s6_addr[0]; 645 c.ip_lo = *(uint64_t *)&lip->s6_addr[8]; 646 647 return (t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c)); 648 } 649 650 static struct clip_entry * 651 search_lip(struct tom_data *td, struct in6_addr *lip) 652 { 653 struct clip_entry *ce; 654 655 mtx_assert(&td->clip_table_lock, MA_OWNED); 656 657 TAILQ_FOREACH(ce, &td->clip_table, link) { 658 if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip)) 659 return (ce); 660 } 661 662 return (NULL); 663 } 664 665 struct clip_entry * 666 hold_lip(struct tom_data *td, struct in6_addr *lip) 667 { 668 struct clip_entry *ce; 669 670 mtx_lock(&td->clip_table_lock); 671 ce = search_lip(td, lip); 672 if (ce != NULL) 673 ce->refcount++; 674 mtx_unlock(&td->clip_table_lock); 675 676 return (ce); 677 } 678 679 void 680 release_lip(struct tom_data *td, struct clip_entry *ce) 681 { 682 683 mtx_lock(&td->clip_table_lock); 684 KASSERT(search_lip(td, &ce->lip) == ce, 685 ("%s: CLIP entry %p p not in CLIP table.", __func__, ce)); 686 KASSERT(ce->refcount > 0, 687 ("%s: CLIP entry %p has refcount 0", __func__, ce)); 688 --ce->refcount; 689 mtx_unlock(&td->clip_table_lock); 690 } 691 692 static void 693 init_clip_table(struct adapter *sc, struct tom_data *td) 694 { 695 struct in6_ifaddr *ia; 696 struct in6_addr *lip, tlip; 697 struct clip_entry *ce; 698 699 ASSERT_SYNCHRONIZED_OP(sc); 700 701 mtx_init(&td->clip_table_lock, "CLIP table lock", NULL, MTX_DEF); 702 TAILQ_INIT(&td->clip_table); 703 704 IN6_IFADDR_RLOCK(); 705 TAILQ_FOREACH(ia, &V_in6_ifaddrhead, ia_link) { 706 lip = &ia->ia_addr.sin6_addr; 707 708 KASSERT(!IN6_IS_ADDR_MULTICAST(lip), 709 ("%s: mcast address in in6_ifaddr list", __func__)); 710 711 if (IN6_IS_ADDR_LOOPBACK(lip)) 712 continue; 713 if (IN6_IS_SCOPE_EMBED(lip)) { 714 /* Remove the embedded scope */ 715 tlip = *lip; 716 lip = &tlip; 717 in6_clearscope(lip); 718 } 719 /* 720 * XXX: how to weed out the link local address for the loopback 721 * interface? It's fe80::1 usually (always?). 722 */ 723 724 mtx_lock(&td->clip_table_lock); 725 if (search_lip(td, lip) == NULL) { 726 ce = malloc(sizeof(*ce), M_CXGBE, M_NOWAIT); 727 memcpy(&ce->lip, lip, sizeof(ce->lip)); 728 ce->refcount = 0; 729 if (add_lip(sc, lip) == 0) 730 TAILQ_INSERT_TAIL(&td->clip_table, ce, link); 731 else 732 free(ce, M_CXGBE); 733 } 734 mtx_unlock(&td->clip_table_lock); 735 } 736 IN6_IFADDR_RUNLOCK(); 737 } 738 739 static void 740 destroy_clip_table(struct adapter *sc, struct tom_data *td) 741 { 742 struct clip_entry *ce, *ce_temp; 743 744 if (mtx_initialized(&td->clip_table_lock)) { 745 mtx_lock(&td->clip_table_lock); 746 TAILQ_FOREACH_SAFE(ce, &td->clip_table, link, ce_temp) { 747 KASSERT(ce->refcount == 0, 748 ("%s: CLIP entry %p still in use (%d)", __func__, 749 ce, ce->refcount)); 750 TAILQ_REMOVE(&td->clip_table, ce, link); 751 delete_lip(sc, &ce->lip); 752 free(ce, M_CXGBE); 753 } 754 mtx_unlock(&td->clip_table_lock); 755 mtx_destroy(&td->clip_table_lock); 756 } 757 } 758 759 static void 760 free_tom_data(struct adapter *sc, struct tom_data *td) 761 { 762 763 ASSERT_SYNCHRONIZED_OP(sc); 764 765 KASSERT(TAILQ_EMPTY(&td->toep_list), 766 ("%s: TOE PCB list is not empty.", __func__)); 767 KASSERT(td->lctx_count == 0, 768 ("%s: lctx hash table is not empty.", __func__)); 769 770 t4_uninit_l2t_cpl_handlers(sc); 771 t4_uninit_cpl_io_handlers(sc); 772 t4_uninit_ddp(sc, td); 773 destroy_clip_table(sc, td); 774 775 if (td->listen_mask != 0) 776 hashdestroy(td->listen_hash, M_CXGBE, td->listen_mask); 777 778 if (mtx_initialized(&td->lctx_hash_lock)) 779 mtx_destroy(&td->lctx_hash_lock); 780 if (mtx_initialized(&td->toep_list_lock)) 781 mtx_destroy(&td->toep_list_lock); 782 783 free_tid_tabs(&sc->tids); 784 free(td, M_CXGBE); 785 } 786 787 /* 788 * Ground control to Major TOM 789 * Commencing countdown, engines on 790 */ 791 static int 792 t4_tom_activate(struct adapter *sc) 793 { 794 struct tom_data *td; 795 struct toedev *tod; 796 int i, rc; 797 798 ASSERT_SYNCHRONIZED_OP(sc); 799 800 /* per-adapter softc for TOM */ 801 td = malloc(sizeof(*td), M_CXGBE, M_ZERO | M_NOWAIT); 802 if (td == NULL) 803 return (ENOMEM); 804 805 /* List of TOE PCBs and associated lock */ 806 mtx_init(&td->toep_list_lock, "PCB list lock", NULL, MTX_DEF); 807 TAILQ_INIT(&td->toep_list); 808 809 /* Listen context */ 810 mtx_init(&td->lctx_hash_lock, "lctx hash lock", NULL, MTX_DEF); 811 td->listen_hash = hashinit_flags(LISTEN_HASH_SIZE, M_CXGBE, 812 &td->listen_mask, HASH_NOWAIT); 813 814 /* TID tables */ 815 rc = alloc_tid_tabs(&sc->tids); 816 if (rc != 0) 817 goto done; 818 819 /* DDP page pods and CPL handlers */ 820 t4_init_ddp(sc, td); 821 822 /* CLIP table for IPv6 offload */ 823 init_clip_table(sc, td); 824 825 /* CPL handlers */ 826 t4_init_connect_cpl_handlers(sc); 827 t4_init_l2t_cpl_handlers(sc); 828 t4_init_listen_cpl_handlers(sc); 829 t4_init_cpl_io_handlers(sc); 830 831 /* toedev ops */ 832 tod = &td->tod; 833 init_toedev(tod); 834 tod->tod_softc = sc; 835 tod->tod_connect = t4_connect; 836 tod->tod_listen_start = t4_listen_start; 837 tod->tod_listen_stop = t4_listen_stop; 838 tod->tod_rcvd = t4_rcvd; 839 tod->tod_output = t4_tod_output; 840 tod->tod_send_rst = t4_send_rst; 841 tod->tod_send_fin = t4_send_fin; 842 tod->tod_pcb_detach = t4_pcb_detach; 843 tod->tod_l2_update = t4_l2_update; 844 tod->tod_syncache_added = t4_syncache_added; 845 tod->tod_syncache_removed = t4_syncache_removed; 846 tod->tod_syncache_respond = t4_syncache_respond; 847 tod->tod_offload_socket = t4_offload_socket; 848 849 for_each_port(sc, i) 850 TOEDEV(sc->port[i]->ifp) = &td->tod; 851 852 sc->tom_softc = td; 853 sc->flags |= TOM_INIT_DONE; 854 register_toedev(sc->tom_softc); 855 856 done: 857 if (rc != 0) 858 free_tom_data(sc, td); 859 return (rc); 860 } 861 862 static int 863 t4_tom_deactivate(struct adapter *sc) 864 { 865 int rc = 0; 866 struct tom_data *td = sc->tom_softc; 867 868 ASSERT_SYNCHRONIZED_OP(sc); 869 870 if (td == NULL) 871 return (0); /* XXX. KASSERT? */ 872 873 if (sc->offload_map != 0) 874 return (EBUSY); /* at least one port has IFCAP_TOE enabled */ 875 876 mtx_lock(&td->toep_list_lock); 877 if (!TAILQ_EMPTY(&td->toep_list)) 878 rc = EBUSY; 879 mtx_unlock(&td->toep_list_lock); 880 881 mtx_lock(&td->lctx_hash_lock); 882 if (td->lctx_count > 0) 883 rc = EBUSY; 884 mtx_unlock(&td->lctx_hash_lock); 885 886 if (rc == 0) { 887 unregister_toedev(sc->tom_softc); 888 free_tom_data(sc, td); 889 sc->tom_softc = NULL; 890 sc->flags &= ~TOM_INIT_DONE; 891 } 892 893 return (rc); 894 } 895 896 static int 897 t4_tom_mod_load(void) 898 { 899 int rc; 900 struct protosw *tcp_protosw, *tcp6_protosw; 901 902 tcp_protosw = pffindproto(PF_INET, IPPROTO_TCP, SOCK_STREAM); 903 if (tcp_protosw == NULL) 904 return (ENOPROTOOPT); 905 bcopy(tcp_protosw, &ddp_protosw, sizeof(ddp_protosw)); 906 bcopy(tcp_protosw->pr_usrreqs, &ddp_usrreqs, sizeof(ddp_usrreqs)); 907 ddp_usrreqs.pru_soreceive = t4_soreceive_ddp; 908 ddp_protosw.pr_usrreqs = &ddp_usrreqs; 909 910 tcp6_protosw = pffindproto(PF_INET6, IPPROTO_TCP, SOCK_STREAM); 911 if (tcp6_protosw == NULL) 912 return (ENOPROTOOPT); 913 bcopy(tcp6_protosw, &ddp6_protosw, sizeof(ddp6_protosw)); 914 bcopy(tcp6_protosw->pr_usrreqs, &ddp6_usrreqs, sizeof(ddp6_usrreqs)); 915 ddp6_usrreqs.pru_soreceive = t4_soreceive_ddp; 916 ddp6_protosw.pr_usrreqs = &ddp6_usrreqs; 917 918 rc = t4_register_uld(&tom_uld_info); 919 if (rc != 0) 920 t4_tom_mod_unload(); 921 922 return (rc); 923 } 924 925 static void 926 tom_uninit(struct adapter *sc, void *arg __unused) 927 { 928 if (begin_synchronized_op(sc, NULL, HOLD_LOCK, "t4tomun")) 929 return; 930 931 /* Try to free resources (works only if no port has IFCAP_TOE) */ 932 if (sc->flags & TOM_INIT_DONE) 933 t4_deactivate_uld(sc, ULD_TOM); 934 935 end_synchronized_op(sc, LOCK_HELD); 936 } 937 938 static int 939 t4_tom_mod_unload(void) 940 { 941 t4_iterate(tom_uninit, NULL); 942 943 if (t4_unregister_uld(&tom_uld_info) == EBUSY) 944 return (EBUSY); 945 946 return (0); 947 } 948 #endif /* TCP_OFFLOAD */ 949 950 static int 951 t4_tom_modevent(module_t mod, int cmd, void *arg) 952 { 953 int rc = 0; 954 955 #ifdef TCP_OFFLOAD 956 switch (cmd) { 957 case MOD_LOAD: 958 rc = t4_tom_mod_load(); 959 break; 960 961 case MOD_UNLOAD: 962 rc = t4_tom_mod_unload(); 963 break; 964 965 default: 966 rc = EINVAL; 967 } 968 #else 969 printf("t4_tom: compiled without TCP_OFFLOAD support.\n"); 970 rc = EOPNOTSUPP; 971 #endif 972 return (rc); 973 } 974 975 static moduledata_t t4_tom_moddata= { 976 "t4_tom", 977 t4_tom_modevent, 978 0 979 }; 980 981 MODULE_VERSION(t4_tom, 1); 982 MODULE_DEPEND(t4_tom, toecore, 1, 1, 1); 983 MODULE_DEPEND(t4_tom, t4nex, 1, 1, 1); 984 DECLARE_MODULE(t4_tom, t4_tom_moddata, SI_SUB_EXEC, SI_ORDER_ANY); 985