2 * Copyright (c) 2012 Chelsio Communications, Inc.
4 * Written by: Navdeep Parhar <np@FreeBSD.org>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
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.
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
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
32 #include "opt_inet6.h"
34 #include <sys/param.h>
35 #include <sys/types.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.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 <sys/taskqueue.h>
46 #include <net/if_var.h>
47 #include <netinet/in.h>
48 #include <netinet/in_pcb.h>
49 #include <netinet/in_var.h>
50 #include <netinet/ip.h>
51 #include <netinet/ip6.h>
52 #include <netinet/tcp_var.h>
53 #include <netinet6/scope6_var.h>
55 #include <netinet/tcp_fsm.h>
56 #include <netinet/toecore.h>
59 #include "common/common.h"
60 #include "common/t4_msg.h"
61 #include "common/t4_regs.h"
62 #include "common/t4_regs_values.h"
63 #include "common/t4_tcb.h"
64 #include "tom/t4_tom_l2t.h"
65 #include "tom/t4_tom.h"
67 static struct protosw ddp_protosw;
68 static struct pr_usrreqs ddp_usrreqs;
70 static struct protosw ddp6_protosw;
71 static struct pr_usrreqs ddp6_usrreqs;
74 static int t4_tom_mod_load(void);
75 static int t4_tom_mod_unload(void);
76 static int t4_tom_modevent(module_t, int, void *);
78 /* ULD ops and helpers */
79 static int t4_tom_activate(struct adapter *);
80 static int t4_tom_deactivate(struct adapter *);
82 static struct uld_info tom_uld_info = {
84 .activate = t4_tom_activate,
85 .deactivate = t4_tom_deactivate,
88 static void queue_tid_release(struct adapter *, int);
89 static void release_offload_resources(struct toepcb *);
90 static int alloc_tid_tabs(struct tid_info *);
91 static void free_tid_tabs(struct tid_info *);
92 static int add_lip(struct adapter *, struct in6_addr *);
93 static int delete_lip(struct adapter *, struct in6_addr *);
94 static struct clip_entry *search_lip(struct tom_data *, struct in6_addr *);
95 static void init_clip_table(struct adapter *, struct tom_data *);
96 static void update_clip(struct adapter *, void *);
97 static void t4_clip_task(void *, int);
98 static void update_clip_table(struct adapter *, struct tom_data *);
99 static void destroy_clip_table(struct adapter *, struct tom_data *);
100 static void free_tom_data(struct adapter *, struct tom_data *);
102 static int in6_ifaddr_gen;
103 static eventhandler_tag ifaddr_evhandler;
104 static struct timeout_task clip_task;
107 alloc_toepcb(struct port_info *pi, int txqid, int rxqid, int flags)
109 struct adapter *sc = pi->adapter;
111 int tx_credits, txsd_total, len;
114 * The firmware counts tx work request credits in units of 16 bytes
115 * each. Reserve room for an ABORT_REQ so the driver never has to worry
116 * about tx credits if it wants to abort a connection.
118 tx_credits = sc->params.ofldq_wr_cred;
119 tx_credits -= howmany(sizeof(struct cpl_abort_req), 16);
122 * Shortest possible tx work request is a fw_ofld_tx_data_wr + 1 byte
123 * immediate payload, and firmware counts tx work request credits in
124 * units of 16 byte. Calculate the maximum work requests possible.
126 txsd_total = tx_credits /
127 howmany((sizeof(struct fw_ofld_tx_data_wr) + 1), 16);
130 txqid = (arc4random() % pi->nofldtxq) + pi->first_ofld_txq;
131 KASSERT(txqid >= pi->first_ofld_txq &&
132 txqid < pi->first_ofld_txq + pi->nofldtxq,
133 ("%s: txqid %d for port %p (first %d, n %d)", __func__, txqid, pi,
134 pi->first_ofld_txq, pi->nofldtxq));
137 rxqid = (arc4random() % pi->nofldrxq) + pi->first_ofld_rxq;
138 KASSERT(rxqid >= pi->first_ofld_rxq &&
139 rxqid < pi->first_ofld_rxq + pi->nofldrxq,
140 ("%s: rxqid %d for port %p (first %d, n %d)", __func__, rxqid, pi,
141 pi->first_ofld_rxq, pi->nofldrxq));
143 len = offsetof(struct toepcb, txsd) +
144 txsd_total * sizeof(struct ofld_tx_sdesc);
146 toep = malloc(len, M_CXGBE, M_ZERO | flags);
150 toep->td = sc->tom_softc;
152 toep->tx_total = tx_credits;
153 toep->tx_credits = tx_credits;
154 toep->ofld_txq = &sc->sge.ofld_txq[txqid];
155 toep->ofld_rxq = &sc->sge.ofld_rxq[rxqid];
156 toep->ctrlq = &sc->sge.ctrlq[pi->port_id];
157 toep->txsd_total = txsd_total;
158 toep->txsd_avail = txsd_total;
166 free_toepcb(struct toepcb *toep)
169 KASSERT(!(toep->flags & TPF_ATTACHED),
170 ("%s: attached to an inpcb", __func__));
171 KASSERT(!(toep->flags & TPF_CPL_PENDING),
172 ("%s: CPL pending", __func__));
178 * Set up the socket for TCP offload.
181 offload_socket(struct socket *so, struct toepcb *toep)
183 struct tom_data *td = toep->td;
184 struct inpcb *inp = sotoinpcb(so);
185 struct tcpcb *tp = intotcpcb(inp);
188 INP_WLOCK_ASSERT(inp);
193 sb->sb_flags |= SB_NOCOALESCE;
197 sb->sb_flags |= SB_NOCOALESCE;
198 if (toep->ulp_mode == ULP_MODE_TCPDDP) {
199 if (inp->inp_vflag & INP_IPV6)
200 so->so_proto = &ddp6_protosw;
202 so->so_proto = &ddp_protosw;
209 tp->t_flags |= TF_TOE;
211 /* Install an extra hold on inp */
213 toep->flags |= TPF_ATTACHED;
216 /* Add the TOE PCB to the active list */
217 mtx_lock(&td->toep_list_lock);
218 TAILQ_INSERT_HEAD(&td->toep_list, toep, link);
219 mtx_unlock(&td->toep_list_lock);
222 /* This is _not_ the normal way to "unoffload" a socket. */
224 undo_offload_socket(struct socket *so)
226 struct inpcb *inp = sotoinpcb(so);
227 struct tcpcb *tp = intotcpcb(inp);
228 struct toepcb *toep = tp->t_toe;
229 struct tom_data *td = toep->td;
232 INP_WLOCK_ASSERT(inp);
236 sb->sb_flags &= ~SB_NOCOALESCE;
240 sb->sb_flags &= ~SB_NOCOALESCE;
245 tp->t_flags &= ~TF_TOE;
248 toep->flags &= ~TPF_ATTACHED;
249 if (in_pcbrele_wlocked(inp))
250 panic("%s: inp freed.", __func__);
252 mtx_lock(&td->toep_list_lock);
253 TAILQ_REMOVE(&td->toep_list, toep, link);
254 mtx_unlock(&td->toep_list_lock);
258 release_offload_resources(struct toepcb *toep)
260 struct tom_data *td = toep->td;
261 struct adapter *sc = td_adapter(td);
264 KASSERT(!(toep->flags & TPF_CPL_PENDING),
265 ("%s: %p has CPL pending.", __func__, toep));
266 KASSERT(!(toep->flags & TPF_ATTACHED),
267 ("%s: %p is still attached.", __func__, toep));
269 CTR5(KTR_CXGBE, "%s: toep %p (tid %d, l2te %p, ce %p)",
270 __func__, toep, tid, toep->l2te, toep->ce);
272 if (toep->ulp_mode == ULP_MODE_TCPDDP)
273 release_ddp_resources(toep);
276 t4_l2t_release(toep->l2te);
280 release_tid(sc, tid, toep->ctrlq);
284 release_lip(td, toep->ce);
286 mtx_lock(&td->toep_list_lock);
287 TAILQ_REMOVE(&td->toep_list, toep, link);
288 mtx_unlock(&td->toep_list_lock);
294 * The kernel is done with the TCP PCB and this is our opportunity to unhook the
295 * toepcb hanging off of it. If the TOE driver is also done with the toepcb (no
296 * pending CPL) then it is time to release all resources tied to the toepcb.
298 * Also gets called when an offloaded active open fails and the TOM wants the
299 * kernel to take the TCP PCB back.
302 t4_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp)
304 #if defined(KTR) || defined(INVARIANTS)
305 struct inpcb *inp = tp->t_inpcb;
307 struct toepcb *toep = tp->t_toe;
309 INP_WLOCK_ASSERT(inp);
311 KASSERT(toep != NULL, ("%s: toep is NULL", __func__));
312 KASSERT(toep->flags & TPF_ATTACHED,
313 ("%s: not attached", __func__));
316 if (tp->t_state == TCPS_SYN_SENT) {
317 CTR6(KTR_CXGBE, "%s: atid %d, toep %p (0x%x), inp %p (0x%x)",
318 __func__, toep->tid, toep, toep->flags, inp,
322 "t4_pcb_detach: tid %d (%s), toep %p (0x%x), inp %p (0x%x)",
323 toep->tid, tcpstates[tp->t_state], toep, toep->flags, inp,
329 tp->t_flags &= ~TF_TOE;
330 toep->flags &= ~TPF_ATTACHED;
332 if (!(toep->flags & TPF_CPL_PENDING))
333 release_offload_resources(toep);
337 * setsockopt handler.
340 t4_ctloutput(struct toedev *tod, struct tcpcb *tp, int dir, int name)
342 struct adapter *sc = tod->tod_softc;
343 struct toepcb *toep = tp->t_toe;
348 CTR4(KTR_CXGBE, "%s: tp %p, dir %u, name %u", __func__, tp, dir, name);
352 t4_set_tcb_field(sc, toep, 1, W_TCB_T_FLAGS, V_TF_NAGLE(1),
353 V_TF_NAGLE(tp->t_flags & TF_NODELAY ? 0 : 1));
361 * The TOE driver will not receive any more CPLs for the tid associated with the
362 * toepcb; release the hold on the inpcb.
365 final_cpl_received(struct toepcb *toep)
367 struct inpcb *inp = toep->inp;
369 KASSERT(inp != NULL, ("%s: inp is NULL", __func__));
370 INP_WLOCK_ASSERT(inp);
371 KASSERT(toep->flags & TPF_CPL_PENDING,
372 ("%s: CPL not pending already?", __func__));
374 CTR6(KTR_CXGBE, "%s: tid %d, toep %p (0x%x), inp %p (0x%x)",
375 __func__, toep->tid, toep, toep->flags, inp, inp->inp_flags);
378 toep->flags &= ~TPF_CPL_PENDING;
380 if (!(toep->flags & TPF_ATTACHED))
381 release_offload_resources(toep);
383 if (!in_pcbrele_wlocked(inp))
388 insert_tid(struct adapter *sc, int tid, void *ctx)
390 struct tid_info *t = &sc->tids;
392 t->tid_tab[tid] = ctx;
393 atomic_add_int(&t->tids_in_use, 1);
397 lookup_tid(struct adapter *sc, int tid)
399 struct tid_info *t = &sc->tids;
401 return (t->tid_tab[tid]);
405 update_tid(struct adapter *sc, int tid, void *ctx)
407 struct tid_info *t = &sc->tids;
409 t->tid_tab[tid] = ctx;
413 remove_tid(struct adapter *sc, int tid)
415 struct tid_info *t = &sc->tids;
417 t->tid_tab[tid] = NULL;
418 atomic_subtract_int(&t->tids_in_use, 1);
422 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
425 struct cpl_tid_release *req;
427 wr = alloc_wrqe(sizeof(*req), ctrlq);
429 queue_tid_release(sc, tid); /* defer */
434 INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
440 queue_tid_release(struct adapter *sc, int tid)
443 CXGBE_UNIMPLEMENTED("deferred tid release");
447 * What mtu_idx to use, given a 4-tuple and/or an MSS cap
450 find_best_mtu_idx(struct adapter *sc, struct in_conninfo *inc, int pmss)
452 unsigned short *mtus = &sc->params.mtus[0];
455 KASSERT(inc != NULL || pmss > 0,
456 ("%s: at least one of inc/pmss must be specified", __func__));
458 mss = inc ? tcp_mssopt(inc) : pmss;
459 if (pmss > 0 && mss > pmss)
462 if (inc->inc_flags & INC_ISIPV6)
463 n = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
465 n = sizeof(struct ip) + sizeof(struct tcphdr);
467 for (i = 0; i < NMTUS - 1 && mtus[i + 1] <= mss + n; i++)
474 * Determine the receive window size for a socket.
477 select_rcv_wnd(struct socket *so)
481 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
483 wnd = sbspace(&so->so_rcv);
484 if (wnd < MIN_RCV_WND)
487 return min(wnd, MAX_RCV_WND);
491 select_rcv_wscale(void)
494 unsigned long space = sb_max;
496 if (space > MAX_RCV_WND)
499 while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < space)
505 extern int always_keepalive;
506 #define VIID_SMACIDX(v) (((unsigned int)(v) & 0x7f) << 1)
509 * socket so could be a listening socket too.
512 calc_opt0(struct socket *so, struct port_info *pi, struct l2t_entry *e,
513 int mtu_idx, int rscale, int rx_credits, int ulp_mode)
517 KASSERT(rx_credits <= M_RCV_BUFSIZ,
518 ("%s: rcv_bufsiz too high", __func__));
520 opt0 = F_TCAM_BYPASS | V_WND_SCALE(rscale) | V_MSS_IDX(mtu_idx) |
521 V_ULP_MODE(ulp_mode) | V_RCV_BUFSIZ(rx_credits);
524 struct inpcb *inp = sotoinpcb(so);
525 struct tcpcb *tp = intotcpcb(inp);
526 int keepalive = always_keepalive ||
527 so_options_get(so) & SO_KEEPALIVE;
529 opt0 |= V_NAGLE((tp->t_flags & TF_NODELAY) == 0);
530 opt0 |= V_KEEP_ALIVE(keepalive != 0);
534 opt0 |= V_L2T_IDX(e->idx);
537 opt0 |= V_SMAC_SEL(VIID_SMACIDX(pi->viid));
538 opt0 |= V_TX_CHAN(pi->tx_chan);
541 return htobe64(opt0);
545 select_ntuple(struct port_info *pi, struct l2t_entry *e)
547 struct adapter *sc = pi->adapter;
548 struct tp_params *tp = &sc->params.tp;
549 uint16_t viid = pi->viid;
553 * Initialize each of the fields which we care about which are present
554 * in the Compressed Filter Tuple.
556 if (tp->vlan_shift >= 0 && e->vlan != CPL_L2T_VLAN_NONE)
557 ntuple |= (uint64_t)(F_FT_VLAN_VLD | e->vlan) << tp->vlan_shift;
559 if (tp->port_shift >= 0)
560 ntuple |= (uint64_t)e->lport << tp->port_shift;
562 if (tp->protocol_shift >= 0)
563 ntuple |= (uint64_t)IPPROTO_TCP << tp->protocol_shift;
565 if (tp->vnic_shift >= 0) {
566 uint32_t vf = G_FW_VIID_VIN(viid);
567 uint32_t pf = G_FW_VIID_PFN(viid);
568 uint32_t vld = G_FW_VIID_VIVLD(viid);
570 ntuple |= (uint64_t)(V_FT_VNID_ID_VF(vf) | V_FT_VNID_ID_PF(pf) |
571 V_FT_VNID_ID_VLD(vld)) << tp->vnic_shift;
575 return (htobe32((uint32_t)ntuple));
577 return (htobe64(V_FILTER_TUPLE(ntuple)));
581 set_tcpddp_ulp_mode(struct toepcb *toep)
584 toep->ulp_mode = ULP_MODE_TCPDDP;
585 toep->ddp_flags = DDP_OK;
586 toep->ddp_score = DDP_LOW_SCORE;
590 negative_advice(int status)
593 return (status == CPL_ERR_RTX_NEG_ADVICE ||
594 status == CPL_ERR_PERSIST_NEG_ADVICE ||
595 status == CPL_ERR_KEEPALV_NEG_ADVICE);
599 alloc_tid_tabs(struct tid_info *t)
604 size = t->ntids * sizeof(*t->tid_tab) +
605 t->natids * sizeof(*t->atid_tab) +
606 t->nstids * sizeof(*t->stid_tab);
608 t->tid_tab = malloc(size, M_CXGBE, M_ZERO | M_NOWAIT);
609 if (t->tid_tab == NULL)
612 mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
613 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
614 t->afree = t->atid_tab;
616 for (i = 1; i < t->natids; i++)
617 t->atid_tab[i - 1].next = &t->atid_tab[i];
618 t->atid_tab[t->natids - 1].next = NULL;
620 mtx_init(&t->stid_lock, "stid lock", NULL, MTX_DEF);
621 t->stid_tab = (struct listen_ctx **)&t->atid_tab[t->natids];
623 TAILQ_INIT(&t->stids);
624 t->nstids_free_head = t->nstids;
626 atomic_store_rel_int(&t->tids_in_use, 0);
632 free_tid_tabs(struct tid_info *t)
634 KASSERT(t->tids_in_use == 0,
635 ("%s: %d tids still in use.", __func__, t->tids_in_use));
636 KASSERT(t->atids_in_use == 0,
637 ("%s: %d atids still in use.", __func__, t->atids_in_use));
638 KASSERT(t->stids_in_use == 0,
639 ("%s: %d tids still in use.", __func__, t->stids_in_use));
641 free(t->tid_tab, M_CXGBE);
644 if (mtx_initialized(&t->atid_lock))
645 mtx_destroy(&t->atid_lock);
646 if (mtx_initialized(&t->stid_lock))
647 mtx_destroy(&t->stid_lock);
651 add_lip(struct adapter *sc, struct in6_addr *lip)
653 struct fw_clip_cmd c;
655 ASSERT_SYNCHRONIZED_OP(sc);
656 /* mtx_assert(&td->clip_table_lock, MA_OWNED); */
658 memset(&c, 0, sizeof(c));
659 c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST |
661 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_ALLOC | FW_LEN16(c));
662 c.ip_hi = *(uint64_t *)&lip->s6_addr[0];
663 c.ip_lo = *(uint64_t *)&lip->s6_addr[8];
665 return (-t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c));
669 delete_lip(struct adapter *sc, struct in6_addr *lip)
671 struct fw_clip_cmd c;
673 ASSERT_SYNCHRONIZED_OP(sc);
674 /* mtx_assert(&td->clip_table_lock, MA_OWNED); */
676 memset(&c, 0, sizeof(c));
677 c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST |
679 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_FREE | FW_LEN16(c));
680 c.ip_hi = *(uint64_t *)&lip->s6_addr[0];
681 c.ip_lo = *(uint64_t *)&lip->s6_addr[8];
683 return (-t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c));
686 static struct clip_entry *
687 search_lip(struct tom_data *td, struct in6_addr *lip)
689 struct clip_entry *ce;
691 mtx_assert(&td->clip_table_lock, MA_OWNED);
693 TAILQ_FOREACH(ce, &td->clip_table, link) {
694 if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip))
702 hold_lip(struct tom_data *td, struct in6_addr *lip)
704 struct clip_entry *ce;
706 mtx_lock(&td->clip_table_lock);
707 ce = search_lip(td, lip);
710 mtx_unlock(&td->clip_table_lock);
716 release_lip(struct tom_data *td, struct clip_entry *ce)
719 mtx_lock(&td->clip_table_lock);
720 KASSERT(search_lip(td, &ce->lip) == ce,
721 ("%s: CLIP entry %p p not in CLIP table.", __func__, ce));
722 KASSERT(ce->refcount > 0,
723 ("%s: CLIP entry %p has refcount 0", __func__, ce));
725 mtx_unlock(&td->clip_table_lock);
729 init_clip_table(struct adapter *sc, struct tom_data *td)
732 ASSERT_SYNCHRONIZED_OP(sc);
734 mtx_init(&td->clip_table_lock, "CLIP table lock", NULL, MTX_DEF);
735 TAILQ_INIT(&td->clip_table);
738 update_clip_table(sc, td);
742 update_clip(struct adapter *sc, void *arg __unused)
745 if (begin_synchronized_op(sc, NULL, HOLD_LOCK, "t4tomuc"))
748 if (sc->flags & TOM_INIT_DONE)
749 update_clip_table(sc, sc->tom_softc);
751 end_synchronized_op(sc, LOCK_HELD);
755 t4_clip_task(void *arg, int count)
758 t4_iterate(update_clip, NULL);
762 update_clip_table(struct adapter *sc, struct tom_data *td)
764 struct in6_ifaddr *ia;
765 struct in6_addr *lip, tlip;
766 struct clip_head stale;
767 struct clip_entry *ce, *ce_temp;
768 int rc, gen = atomic_load_acq_int(&in6_ifaddr_gen);
770 ASSERT_SYNCHRONIZED_OP(sc);
773 mtx_lock(&td->clip_table_lock);
775 if (gen == td->clip_gen)
779 TAILQ_CONCAT(&stale, &td->clip_table, link);
781 TAILQ_FOREACH(ia, &V_in6_ifaddrhead, ia_link) {
782 lip = &ia->ia_addr.sin6_addr;
784 KASSERT(!IN6_IS_ADDR_MULTICAST(lip),
785 ("%s: mcast address in in6_ifaddr list", __func__));
787 if (IN6_IS_ADDR_LOOPBACK(lip))
789 if (IN6_IS_SCOPE_EMBED(lip)) {
790 /* Remove the embedded scope */
796 * XXX: how to weed out the link local address for the loopback
797 * interface? It's fe80::1 usually (always?).
801 * If it's in the main list then we already know it's not stale.
803 TAILQ_FOREACH(ce, &td->clip_table, link) {
804 if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip))
809 * If it's in the stale list we should move it to the main list.
811 TAILQ_FOREACH(ce, &stale, link) {
812 if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip)) {
813 TAILQ_REMOVE(&stale, ce, link);
814 TAILQ_INSERT_TAIL(&td->clip_table, ce, link);
819 /* A new IP6 address; add it to the CLIP table */
820 ce = malloc(sizeof(*ce), M_CXGBE, M_NOWAIT);
821 memcpy(&ce->lip, lip, sizeof(ce->lip));
823 rc = add_lip(sc, lip);
825 TAILQ_INSERT_TAIL(&td->clip_table, ce, link);
827 char ip[INET6_ADDRSTRLEN];
829 inet_ntop(AF_INET6, &ce->lip, &ip[0], sizeof(ip));
830 log(LOG_ERR, "%s: could not add %s (%d)\n",
839 * Remove stale addresses (those no longer in V_in6_ifaddrhead) that are
840 * no longer referenced by the driver.
842 TAILQ_FOREACH_SAFE(ce, &stale, link, ce_temp) {
843 if (ce->refcount == 0) {
844 rc = delete_lip(sc, &ce->lip);
846 TAILQ_REMOVE(&stale, ce, link);
849 char ip[INET6_ADDRSTRLEN];
851 inet_ntop(AF_INET6, &ce->lip, &ip[0],
853 log(LOG_ERR, "%s: could not delete %s (%d)\n",
858 /* The ones that are still referenced need to stay in the CLIP table */
859 TAILQ_CONCAT(&td->clip_table, &stale, link);
863 mtx_unlock(&td->clip_table_lock);
864 IN6_IFADDR_RUNLOCK();
868 destroy_clip_table(struct adapter *sc, struct tom_data *td)
870 struct clip_entry *ce, *ce_temp;
872 if (mtx_initialized(&td->clip_table_lock)) {
873 mtx_lock(&td->clip_table_lock);
874 TAILQ_FOREACH_SAFE(ce, &td->clip_table, link, ce_temp) {
875 KASSERT(ce->refcount == 0,
876 ("%s: CLIP entry %p still in use (%d)", __func__,
878 TAILQ_REMOVE(&td->clip_table, ce, link);
879 delete_lip(sc, &ce->lip);
882 mtx_unlock(&td->clip_table_lock);
883 mtx_destroy(&td->clip_table_lock);
888 free_tom_data(struct adapter *sc, struct tom_data *td)
891 ASSERT_SYNCHRONIZED_OP(sc);
893 KASSERT(TAILQ_EMPTY(&td->toep_list),
894 ("%s: TOE PCB list is not empty.", __func__));
895 KASSERT(td->lctx_count == 0,
896 ("%s: lctx hash table is not empty.", __func__));
898 t4_uninit_l2t_cpl_handlers(sc);
899 t4_uninit_cpl_io_handlers(sc);
900 t4_uninit_ddp(sc, td);
901 destroy_clip_table(sc, td);
903 if (td->listen_mask != 0)
904 hashdestroy(td->listen_hash, M_CXGBE, td->listen_mask);
906 if (mtx_initialized(&td->lctx_hash_lock))
907 mtx_destroy(&td->lctx_hash_lock);
908 if (mtx_initialized(&td->toep_list_lock))
909 mtx_destroy(&td->toep_list_lock);
911 free_tid_tabs(&sc->tids);
916 * Ground control to Major TOM
917 * Commencing countdown, engines on
920 t4_tom_activate(struct adapter *sc)
926 ASSERT_SYNCHRONIZED_OP(sc);
928 /* per-adapter softc for TOM */
929 td = malloc(sizeof(*td), M_CXGBE, M_ZERO | M_NOWAIT);
933 /* List of TOE PCBs and associated lock */
934 mtx_init(&td->toep_list_lock, "PCB list lock", NULL, MTX_DEF);
935 TAILQ_INIT(&td->toep_list);
938 mtx_init(&td->lctx_hash_lock, "lctx hash lock", NULL, MTX_DEF);
939 td->listen_hash = hashinit_flags(LISTEN_HASH_SIZE, M_CXGBE,
940 &td->listen_mask, HASH_NOWAIT);
943 rc = alloc_tid_tabs(&sc->tids);
947 /* DDP page pods and CPL handlers */
950 /* CLIP table for IPv6 offload */
951 init_clip_table(sc, td);
954 t4_init_connect_cpl_handlers(sc);
955 t4_init_l2t_cpl_handlers(sc);
956 t4_init_listen_cpl_handlers(sc);
957 t4_init_cpl_io_handlers(sc);
963 tod->tod_connect = t4_connect;
964 tod->tod_listen_start = t4_listen_start;
965 tod->tod_listen_stop = t4_listen_stop;
966 tod->tod_rcvd = t4_rcvd;
967 tod->tod_output = t4_tod_output;
968 tod->tod_send_rst = t4_send_rst;
969 tod->tod_send_fin = t4_send_fin;
970 tod->tod_pcb_detach = t4_pcb_detach;
971 tod->tod_l2_update = t4_l2_update;
972 tod->tod_syncache_added = t4_syncache_added;
973 tod->tod_syncache_removed = t4_syncache_removed;
974 tod->tod_syncache_respond = t4_syncache_respond;
975 tod->tod_offload_socket = t4_offload_socket;
976 tod->tod_ctloutput = t4_ctloutput;
979 TOEDEV(sc->port[i]->ifp) = &td->tod;
982 sc->flags |= TOM_INIT_DONE;
983 register_toedev(sc->tom_softc);
987 free_tom_data(sc, td);
992 t4_tom_deactivate(struct adapter *sc)
995 struct tom_data *td = sc->tom_softc;
997 ASSERT_SYNCHRONIZED_OP(sc);
1000 return (0); /* XXX. KASSERT? */
1002 if (sc->offload_map != 0)
1003 return (EBUSY); /* at least one port has IFCAP_TOE enabled */
1005 mtx_lock(&td->toep_list_lock);
1006 if (!TAILQ_EMPTY(&td->toep_list))
1008 mtx_unlock(&td->toep_list_lock);
1010 mtx_lock(&td->lctx_hash_lock);
1011 if (td->lctx_count > 0)
1013 mtx_unlock(&td->lctx_hash_lock);
1016 unregister_toedev(sc->tom_softc);
1017 free_tom_data(sc, td);
1018 sc->tom_softc = NULL;
1019 sc->flags &= ~TOM_INIT_DONE;
1026 t4_tom_ifaddr_event(void *arg __unused, struct ifnet *ifp)
1029 atomic_add_rel_int(&in6_ifaddr_gen, 1);
1030 taskqueue_enqueue_timeout(taskqueue_thread, &clip_task, -hz / 4);
1034 t4_tom_mod_load(void)
1037 struct protosw *tcp_protosw, *tcp6_protosw;
1039 tcp_protosw = pffindproto(PF_INET, IPPROTO_TCP, SOCK_STREAM);
1040 if (tcp_protosw == NULL)
1041 return (ENOPROTOOPT);
1042 bcopy(tcp_protosw, &ddp_protosw, sizeof(ddp_protosw));
1043 bcopy(tcp_protosw->pr_usrreqs, &ddp_usrreqs, sizeof(ddp_usrreqs));
1044 ddp_usrreqs.pru_soreceive = t4_soreceive_ddp;
1045 ddp_protosw.pr_usrreqs = &ddp_usrreqs;
1047 tcp6_protosw = pffindproto(PF_INET6, IPPROTO_TCP, SOCK_STREAM);
1048 if (tcp6_protosw == NULL)
1049 return (ENOPROTOOPT);
1050 bcopy(tcp6_protosw, &ddp6_protosw, sizeof(ddp6_protosw));
1051 bcopy(tcp6_protosw->pr_usrreqs, &ddp6_usrreqs, sizeof(ddp6_usrreqs));
1052 ddp6_usrreqs.pru_soreceive = t4_soreceive_ddp;
1053 ddp6_protosw.pr_usrreqs = &ddp6_usrreqs;
1055 TIMEOUT_TASK_INIT(taskqueue_thread, &clip_task, 0, t4_clip_task, NULL);
1056 ifaddr_evhandler = EVENTHANDLER_REGISTER(ifaddr_event,
1057 t4_tom_ifaddr_event, NULL, EVENTHANDLER_PRI_ANY);
1059 rc = t4_register_uld(&tom_uld_info);
1061 t4_tom_mod_unload();
1067 tom_uninit(struct adapter *sc, void *arg __unused)
1069 if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tomun"))
1072 /* Try to free resources (works only if no port has IFCAP_TOE) */
1073 if (sc->flags & TOM_INIT_DONE)
1074 t4_deactivate_uld(sc, ULD_TOM);
1076 end_synchronized_op(sc, 0);
1080 t4_tom_mod_unload(void)
1082 t4_iterate(tom_uninit, NULL);
1084 if (t4_unregister_uld(&tom_uld_info) == EBUSY)
1087 if (ifaddr_evhandler) {
1088 EVENTHANDLER_DEREGISTER(ifaddr_event, ifaddr_evhandler);
1089 taskqueue_cancel_timeout(taskqueue_thread, &clip_task, NULL);
1094 #endif /* TCP_OFFLOAD */
1097 t4_tom_modevent(module_t mod, int cmd, void *arg)
1104 rc = t4_tom_mod_load();
1108 rc = t4_tom_mod_unload();
1115 printf("t4_tom: compiled without TCP_OFFLOAD support.\n");
1121 static moduledata_t t4_tom_moddata= {
1127 MODULE_VERSION(t4_tom, 1);
1128 MODULE_DEPEND(t4_tom, toecore, 1, 1, 1);
1129 MODULE_DEPEND(t4_tom, t4nex, 1, 1, 1);
1130 DECLARE_MODULE(t4_tom, t4_tom_moddata, SI_SUB_EXEC, SI_ORDER_ANY);