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>
40 #include <sys/module.h>
41 #include <sys/protosw.h>
42 #include <sys/domain.h>
43 #include <sys/rmlock.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/taskqueue.h>
48 #include <net/if_var.h>
49 #include <netinet/in.h>
50 #include <netinet/in_pcb.h>
51 #include <netinet/in_var.h>
52 #include <netinet/ip.h>
53 #include <netinet/ip6.h>
54 #include <netinet/tcp_var.h>
55 #include <netinet6/scope6_var.h>
57 #include <netinet/tcp_fsm.h>
58 #include <netinet/toecore.h>
61 #include "common/common.h"
62 #include "common/t4_msg.h"
63 #include "common/t4_regs.h"
64 #include "common/t4_regs_values.h"
65 #include "common/t4_tcb.h"
66 #include "tom/t4_tom_l2t.h"
67 #include "tom/t4_tom.h"
69 static struct protosw ddp_protosw;
70 static struct pr_usrreqs ddp_usrreqs;
72 static struct protosw ddp6_protosw;
73 static struct pr_usrreqs ddp6_usrreqs;
76 static int t4_tom_mod_load(void);
77 static int t4_tom_mod_unload(void);
78 static int t4_tom_modevent(module_t, int, void *);
80 /* ULD ops and helpers */
81 static int t4_tom_activate(struct adapter *);
82 static int t4_tom_deactivate(struct adapter *);
84 static struct uld_info tom_uld_info = {
86 .activate = t4_tom_activate,
87 .deactivate = t4_tom_deactivate,
90 static void queue_tid_release(struct adapter *, int);
91 static void release_offload_resources(struct toepcb *);
92 static int alloc_tid_tabs(struct tid_info *);
93 static void free_tid_tabs(struct tid_info *);
94 static int add_lip(struct adapter *, struct in6_addr *);
95 static int delete_lip(struct adapter *, struct in6_addr *);
96 static struct clip_entry *search_lip(struct tom_data *, struct in6_addr *);
97 static void init_clip_table(struct adapter *, struct tom_data *);
98 static void update_clip(struct adapter *, void *);
99 static void t4_clip_task(void *, int);
100 static void update_clip_table(struct adapter *, struct tom_data *);
101 static void destroy_clip_table(struct adapter *, struct tom_data *);
102 static void free_tom_data(struct adapter *, struct tom_data *);
103 static void reclaim_wr_resources(void *, int);
105 static int in6_ifaddr_gen;
106 static eventhandler_tag ifaddr_evhandler;
107 static struct timeout_task clip_task;
110 alloc_toepcb(struct port_info *pi, int txqid, int rxqid, int flags)
112 struct adapter *sc = pi->adapter;
114 int tx_credits, txsd_total, len;
117 * The firmware counts tx work request credits in units of 16 bytes
118 * each. Reserve room for an ABORT_REQ so the driver never has to worry
119 * about tx credits if it wants to abort a connection.
121 tx_credits = sc->params.ofldq_wr_cred;
122 tx_credits -= howmany(sizeof(struct cpl_abort_req), 16);
125 * Shortest possible tx work request is a fw_ofld_tx_data_wr + 1 byte
126 * immediate payload, and firmware counts tx work request credits in
127 * units of 16 byte. Calculate the maximum work requests possible.
129 txsd_total = tx_credits /
130 howmany((sizeof(struct fw_ofld_tx_data_wr) + 1), 16);
133 txqid = (arc4random() % pi->nofldtxq) + pi->first_ofld_txq;
134 KASSERT(txqid >= pi->first_ofld_txq &&
135 txqid < pi->first_ofld_txq + pi->nofldtxq,
136 ("%s: txqid %d for port %p (first %d, n %d)", __func__, txqid, pi,
137 pi->first_ofld_txq, pi->nofldtxq));
140 rxqid = (arc4random() % pi->nofldrxq) + pi->first_ofld_rxq;
141 KASSERT(rxqid >= pi->first_ofld_rxq &&
142 rxqid < pi->first_ofld_rxq + pi->nofldrxq,
143 ("%s: rxqid %d for port %p (first %d, n %d)", __func__, rxqid, pi,
144 pi->first_ofld_rxq, pi->nofldrxq));
146 len = offsetof(struct toepcb, txsd) +
147 txsd_total * sizeof(struct ofld_tx_sdesc);
149 toep = malloc(len, M_CXGBE, M_ZERO | flags);
153 toep->td = sc->tom_softc;
155 toep->tx_total = tx_credits;
156 toep->tx_credits = tx_credits;
157 toep->ofld_txq = &sc->sge.ofld_txq[txqid];
158 toep->ofld_rxq = &sc->sge.ofld_rxq[rxqid];
159 toep->ctrlq = &sc->sge.ctrlq[pi->port_id];
160 toep->txsd_total = txsd_total;
161 toep->txsd_avail = txsd_total;
169 free_toepcb(struct toepcb *toep)
172 KASSERT(!(toep->flags & TPF_ATTACHED),
173 ("%s: attached to an inpcb", __func__));
174 KASSERT(!(toep->flags & TPF_CPL_PENDING),
175 ("%s: CPL pending", __func__));
181 * Set up the socket for TCP offload.
184 offload_socket(struct socket *so, struct toepcb *toep)
186 struct tom_data *td = toep->td;
187 struct inpcb *inp = sotoinpcb(so);
188 struct tcpcb *tp = intotcpcb(inp);
191 INP_WLOCK_ASSERT(inp);
196 sb->sb_flags |= SB_NOCOALESCE;
200 sb->sb_flags |= SB_NOCOALESCE;
201 if (toep->ulp_mode == ULP_MODE_TCPDDP) {
202 if (inp->inp_vflag & INP_IPV6)
203 so->so_proto = &ddp6_protosw;
205 so->so_proto = &ddp_protosw;
212 tp->t_flags |= TF_TOE;
214 /* Install an extra hold on inp */
216 toep->flags |= TPF_ATTACHED;
219 /* Add the TOE PCB to the active list */
220 mtx_lock(&td->toep_list_lock);
221 TAILQ_INSERT_HEAD(&td->toep_list, toep, link);
222 mtx_unlock(&td->toep_list_lock);
225 /* This is _not_ the normal way to "unoffload" a socket. */
227 undo_offload_socket(struct socket *so)
229 struct inpcb *inp = sotoinpcb(so);
230 struct tcpcb *tp = intotcpcb(inp);
231 struct toepcb *toep = tp->t_toe;
232 struct tom_data *td = toep->td;
235 INP_WLOCK_ASSERT(inp);
239 sb->sb_flags &= ~SB_NOCOALESCE;
243 sb->sb_flags &= ~SB_NOCOALESCE;
248 tp->t_flags &= ~TF_TOE;
251 toep->flags &= ~TPF_ATTACHED;
252 if (in_pcbrele_wlocked(inp))
253 panic("%s: inp freed.", __func__);
255 mtx_lock(&td->toep_list_lock);
256 TAILQ_REMOVE(&td->toep_list, toep, link);
257 mtx_unlock(&td->toep_list_lock);
261 release_offload_resources(struct toepcb *toep)
263 struct tom_data *td = toep->td;
264 struct adapter *sc = td_adapter(td);
267 KASSERT(!(toep->flags & TPF_CPL_PENDING),
268 ("%s: %p has CPL pending.", __func__, toep));
269 KASSERT(!(toep->flags & TPF_ATTACHED),
270 ("%s: %p is still attached.", __func__, toep));
272 CTR5(KTR_CXGBE, "%s: toep %p (tid %d, l2te %p, ce %p)",
273 __func__, toep, tid, toep->l2te, toep->ce);
275 if (toep->ulp_mode == ULP_MODE_TCPDDP)
276 release_ddp_resources(toep);
279 t4_l2t_release(toep->l2te);
283 release_tid(sc, tid, toep->ctrlq);
287 release_lip(td, toep->ce);
289 mtx_lock(&td->toep_list_lock);
290 TAILQ_REMOVE(&td->toep_list, toep, link);
291 mtx_unlock(&td->toep_list_lock);
297 * The kernel is done with the TCP PCB and this is our opportunity to unhook the
298 * toepcb hanging off of it. If the TOE driver is also done with the toepcb (no
299 * pending CPL) then it is time to release all resources tied to the toepcb.
301 * Also gets called when an offloaded active open fails and the TOM wants the
302 * kernel to take the TCP PCB back.
305 t4_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp)
307 #if defined(KTR) || defined(INVARIANTS)
308 struct inpcb *inp = tp->t_inpcb;
310 struct toepcb *toep = tp->t_toe;
312 INP_WLOCK_ASSERT(inp);
314 KASSERT(toep != NULL, ("%s: toep is NULL", __func__));
315 KASSERT(toep->flags & TPF_ATTACHED,
316 ("%s: not attached", __func__));
319 if (tp->t_state == TCPS_SYN_SENT) {
320 CTR6(KTR_CXGBE, "%s: atid %d, toep %p (0x%x), inp %p (0x%x)",
321 __func__, toep->tid, toep, toep->flags, inp,
325 "t4_pcb_detach: tid %d (%s), toep %p (0x%x), inp %p (0x%x)",
326 toep->tid, tcpstates[tp->t_state], toep, toep->flags, inp,
332 tp->t_flags &= ~TF_TOE;
333 toep->flags &= ~TPF_ATTACHED;
335 if (!(toep->flags & TPF_CPL_PENDING))
336 release_offload_resources(toep);
340 * setsockopt handler.
343 t4_ctloutput(struct toedev *tod, struct tcpcb *tp, int dir, int name)
345 struct adapter *sc = tod->tod_softc;
346 struct toepcb *toep = tp->t_toe;
351 CTR4(KTR_CXGBE, "%s: tp %p, dir %u, name %u", __func__, tp, dir, name);
355 t4_set_tcb_field(sc, toep, 1, W_TCB_T_FLAGS, V_TF_NAGLE(1),
356 V_TF_NAGLE(tp->t_flags & TF_NODELAY ? 0 : 1));
364 * The TOE driver will not receive any more CPLs for the tid associated with the
365 * toepcb; release the hold on the inpcb.
368 final_cpl_received(struct toepcb *toep)
370 struct inpcb *inp = toep->inp;
372 KASSERT(inp != NULL, ("%s: inp is NULL", __func__));
373 INP_WLOCK_ASSERT(inp);
374 KASSERT(toep->flags & TPF_CPL_PENDING,
375 ("%s: CPL not pending already?", __func__));
377 CTR6(KTR_CXGBE, "%s: tid %d, toep %p (0x%x), inp %p (0x%x)",
378 __func__, toep->tid, toep, toep->flags, inp, inp->inp_flags);
381 toep->flags &= ~TPF_CPL_PENDING;
383 if (!(toep->flags & TPF_ATTACHED))
384 release_offload_resources(toep);
386 if (!in_pcbrele_wlocked(inp))
391 insert_tid(struct adapter *sc, int tid, void *ctx)
393 struct tid_info *t = &sc->tids;
395 t->tid_tab[tid] = ctx;
396 atomic_add_int(&t->tids_in_use, 1);
400 lookup_tid(struct adapter *sc, int tid)
402 struct tid_info *t = &sc->tids;
404 return (t->tid_tab[tid]);
408 update_tid(struct adapter *sc, int tid, void *ctx)
410 struct tid_info *t = &sc->tids;
412 t->tid_tab[tid] = ctx;
416 remove_tid(struct adapter *sc, int tid)
418 struct tid_info *t = &sc->tids;
420 t->tid_tab[tid] = NULL;
421 atomic_subtract_int(&t->tids_in_use, 1);
425 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
428 struct cpl_tid_release *req;
430 wr = alloc_wrqe(sizeof(*req), ctrlq);
432 queue_tid_release(sc, tid); /* defer */
437 INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
443 queue_tid_release(struct adapter *sc, int tid)
446 CXGBE_UNIMPLEMENTED("deferred tid release");
450 * What mtu_idx to use, given a 4-tuple and/or an MSS cap
453 find_best_mtu_idx(struct adapter *sc, struct in_conninfo *inc, int pmss)
455 unsigned short *mtus = &sc->params.mtus[0];
458 KASSERT(inc != NULL || pmss > 0,
459 ("%s: at least one of inc/pmss must be specified", __func__));
461 mss = inc ? tcp_mssopt(inc) : pmss;
462 if (pmss > 0 && mss > pmss)
465 if (inc->inc_flags & INC_ISIPV6)
466 n = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
468 n = sizeof(struct ip) + sizeof(struct tcphdr);
470 for (i = 0; i < NMTUS - 1 && mtus[i + 1] <= mss + n; i++)
477 * Determine the receive window size for a socket.
480 select_rcv_wnd(struct socket *so)
484 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
486 wnd = sbspace(&so->so_rcv);
487 if (wnd < MIN_RCV_WND)
490 return min(wnd, MAX_RCV_WND);
494 select_rcv_wscale(void)
497 unsigned long space = sb_max;
499 if (space > MAX_RCV_WND)
502 while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < space)
508 extern int always_keepalive;
509 #define VIID_SMACIDX(v) (((unsigned int)(v) & 0x7f) << 1)
512 * socket so could be a listening socket too.
515 calc_opt0(struct socket *so, struct port_info *pi, struct l2t_entry *e,
516 int mtu_idx, int rscale, int rx_credits, int ulp_mode)
520 KASSERT(rx_credits <= M_RCV_BUFSIZ,
521 ("%s: rcv_bufsiz too high", __func__));
523 opt0 = F_TCAM_BYPASS | V_WND_SCALE(rscale) | V_MSS_IDX(mtu_idx) |
524 V_ULP_MODE(ulp_mode) | V_RCV_BUFSIZ(rx_credits);
527 struct inpcb *inp = sotoinpcb(so);
528 struct tcpcb *tp = intotcpcb(inp);
529 int keepalive = always_keepalive ||
530 so_options_get(so) & SO_KEEPALIVE;
532 opt0 |= V_NAGLE((tp->t_flags & TF_NODELAY) == 0);
533 opt0 |= V_KEEP_ALIVE(keepalive != 0);
537 opt0 |= V_L2T_IDX(e->idx);
540 opt0 |= V_SMAC_SEL(VIID_SMACIDX(pi->viid));
541 opt0 |= V_TX_CHAN(pi->tx_chan);
544 return htobe64(opt0);
548 select_ntuple(struct port_info *pi, struct l2t_entry *e)
550 struct adapter *sc = pi->adapter;
551 struct tp_params *tp = &sc->params.tp;
552 uint16_t viid = pi->viid;
556 * Initialize each of the fields which we care about which are present
557 * in the Compressed Filter Tuple.
559 if (tp->vlan_shift >= 0 && e->vlan != CPL_L2T_VLAN_NONE)
560 ntuple |= (uint64_t)(F_FT_VLAN_VLD | e->vlan) << tp->vlan_shift;
562 if (tp->port_shift >= 0)
563 ntuple |= (uint64_t)e->lport << tp->port_shift;
565 if (tp->protocol_shift >= 0)
566 ntuple |= (uint64_t)IPPROTO_TCP << tp->protocol_shift;
568 if (tp->vnic_shift >= 0) {
569 uint32_t vf = G_FW_VIID_VIN(viid);
570 uint32_t pf = G_FW_VIID_PFN(viid);
571 uint32_t vld = G_FW_VIID_VIVLD(viid);
573 ntuple |= (uint64_t)(V_FT_VNID_ID_VF(vf) | V_FT_VNID_ID_PF(pf) |
574 V_FT_VNID_ID_VLD(vld)) << tp->vnic_shift;
578 return (htobe32((uint32_t)ntuple));
580 return (htobe64(V_FILTER_TUPLE(ntuple)));
584 set_tcpddp_ulp_mode(struct toepcb *toep)
587 toep->ulp_mode = ULP_MODE_TCPDDP;
588 toep->ddp_flags = DDP_OK;
589 toep->ddp_score = DDP_LOW_SCORE;
593 negative_advice(int status)
596 return (status == CPL_ERR_RTX_NEG_ADVICE ||
597 status == CPL_ERR_PERSIST_NEG_ADVICE ||
598 status == CPL_ERR_KEEPALV_NEG_ADVICE);
602 alloc_tid_tabs(struct tid_info *t)
607 size = t->ntids * sizeof(*t->tid_tab) +
608 t->natids * sizeof(*t->atid_tab) +
609 t->nstids * sizeof(*t->stid_tab);
611 t->tid_tab = malloc(size, M_CXGBE, M_ZERO | M_NOWAIT);
612 if (t->tid_tab == NULL)
615 mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
616 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
617 t->afree = t->atid_tab;
619 for (i = 1; i < t->natids; i++)
620 t->atid_tab[i - 1].next = &t->atid_tab[i];
621 t->atid_tab[t->natids - 1].next = NULL;
623 mtx_init(&t->stid_lock, "stid lock", NULL, MTX_DEF);
624 t->stid_tab = (struct listen_ctx **)&t->atid_tab[t->natids];
626 TAILQ_INIT(&t->stids);
627 t->nstids_free_head = t->nstids;
629 atomic_store_rel_int(&t->tids_in_use, 0);
635 free_tid_tabs(struct tid_info *t)
637 KASSERT(t->tids_in_use == 0,
638 ("%s: %d tids still in use.", __func__, t->tids_in_use));
639 KASSERT(t->atids_in_use == 0,
640 ("%s: %d atids still in use.", __func__, t->atids_in_use));
641 KASSERT(t->stids_in_use == 0,
642 ("%s: %d tids still in use.", __func__, t->stids_in_use));
644 free(t->tid_tab, M_CXGBE);
647 if (mtx_initialized(&t->atid_lock))
648 mtx_destroy(&t->atid_lock);
649 if (mtx_initialized(&t->stid_lock))
650 mtx_destroy(&t->stid_lock);
654 add_lip(struct adapter *sc, struct in6_addr *lip)
656 struct fw_clip_cmd c;
658 ASSERT_SYNCHRONIZED_OP(sc);
659 /* mtx_assert(&td->clip_table_lock, MA_OWNED); */
661 memset(&c, 0, sizeof(c));
662 c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST |
664 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_ALLOC | FW_LEN16(c));
665 c.ip_hi = *(uint64_t *)&lip->s6_addr[0];
666 c.ip_lo = *(uint64_t *)&lip->s6_addr[8];
668 return (-t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c));
672 delete_lip(struct adapter *sc, struct in6_addr *lip)
674 struct fw_clip_cmd c;
676 ASSERT_SYNCHRONIZED_OP(sc);
677 /* mtx_assert(&td->clip_table_lock, MA_OWNED); */
679 memset(&c, 0, sizeof(c));
680 c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST |
682 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_FREE | FW_LEN16(c));
683 c.ip_hi = *(uint64_t *)&lip->s6_addr[0];
684 c.ip_lo = *(uint64_t *)&lip->s6_addr[8];
686 return (-t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c));
689 static struct clip_entry *
690 search_lip(struct tom_data *td, struct in6_addr *lip)
692 struct clip_entry *ce;
694 mtx_assert(&td->clip_table_lock, MA_OWNED);
696 TAILQ_FOREACH(ce, &td->clip_table, link) {
697 if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip))
705 hold_lip(struct tom_data *td, struct in6_addr *lip)
707 struct clip_entry *ce;
709 mtx_lock(&td->clip_table_lock);
710 ce = search_lip(td, lip);
713 mtx_unlock(&td->clip_table_lock);
719 release_lip(struct tom_data *td, struct clip_entry *ce)
722 mtx_lock(&td->clip_table_lock);
723 KASSERT(search_lip(td, &ce->lip) == ce,
724 ("%s: CLIP entry %p p not in CLIP table.", __func__, ce));
725 KASSERT(ce->refcount > 0,
726 ("%s: CLIP entry %p has refcount 0", __func__, ce));
728 mtx_unlock(&td->clip_table_lock);
732 init_clip_table(struct adapter *sc, struct tom_data *td)
735 ASSERT_SYNCHRONIZED_OP(sc);
737 mtx_init(&td->clip_table_lock, "CLIP table lock", NULL, MTX_DEF);
738 TAILQ_INIT(&td->clip_table);
741 update_clip_table(sc, td);
745 update_clip(struct adapter *sc, void *arg __unused)
748 if (begin_synchronized_op(sc, NULL, HOLD_LOCK, "t4tomuc"))
751 if (uld_active(sc, ULD_TOM))
752 update_clip_table(sc, sc->tom_softc);
754 end_synchronized_op(sc, LOCK_HELD);
758 t4_clip_task(void *arg, int count)
761 t4_iterate(update_clip, NULL);
765 update_clip_table(struct adapter *sc, struct tom_data *td)
767 struct rm_priotracker in6_ifa_tracker;
768 struct in6_ifaddr *ia;
769 struct in6_addr *lip, tlip;
770 struct clip_head stale;
771 struct clip_entry *ce, *ce_temp;
772 int rc, gen = atomic_load_acq_int(&in6_ifaddr_gen);
774 ASSERT_SYNCHRONIZED_OP(sc);
776 IN6_IFADDR_RLOCK(&in6_ifa_tracker);
777 mtx_lock(&td->clip_table_lock);
779 if (gen == td->clip_gen)
783 TAILQ_CONCAT(&stale, &td->clip_table, link);
785 TAILQ_FOREACH(ia, &V_in6_ifaddrhead, ia_link) {
786 lip = &ia->ia_addr.sin6_addr;
788 KASSERT(!IN6_IS_ADDR_MULTICAST(lip),
789 ("%s: mcast address in in6_ifaddr list", __func__));
791 if (IN6_IS_ADDR_LOOPBACK(lip))
793 if (IN6_IS_SCOPE_EMBED(lip)) {
794 /* Remove the embedded scope */
800 * XXX: how to weed out the link local address for the loopback
801 * interface? It's fe80::1 usually (always?).
805 * If it's in the main list then we already know it's not stale.
807 TAILQ_FOREACH(ce, &td->clip_table, link) {
808 if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip))
813 * If it's in the stale list we should move it to the main list.
815 TAILQ_FOREACH(ce, &stale, link) {
816 if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip)) {
817 TAILQ_REMOVE(&stale, ce, link);
818 TAILQ_INSERT_TAIL(&td->clip_table, ce, link);
823 /* A new IP6 address; add it to the CLIP table */
824 ce = malloc(sizeof(*ce), M_CXGBE, M_NOWAIT);
825 memcpy(&ce->lip, lip, sizeof(ce->lip));
827 rc = add_lip(sc, lip);
829 TAILQ_INSERT_TAIL(&td->clip_table, ce, link);
831 char ip[INET6_ADDRSTRLEN];
833 inet_ntop(AF_INET6, &ce->lip, &ip[0], sizeof(ip));
834 log(LOG_ERR, "%s: could not add %s (%d)\n",
843 * Remove stale addresses (those no longer in V_in6_ifaddrhead) that are
844 * no longer referenced by the driver.
846 TAILQ_FOREACH_SAFE(ce, &stale, link, ce_temp) {
847 if (ce->refcount == 0) {
848 rc = delete_lip(sc, &ce->lip);
850 TAILQ_REMOVE(&stale, ce, link);
853 char ip[INET6_ADDRSTRLEN];
855 inet_ntop(AF_INET6, &ce->lip, &ip[0],
857 log(LOG_ERR, "%s: could not delete %s (%d)\n",
862 /* The ones that are still referenced need to stay in the CLIP table */
863 TAILQ_CONCAT(&td->clip_table, &stale, link);
867 mtx_unlock(&td->clip_table_lock);
868 IN6_IFADDR_RUNLOCK(&in6_ifa_tracker);
872 destroy_clip_table(struct adapter *sc, struct tom_data *td)
874 struct clip_entry *ce, *ce_temp;
876 if (mtx_initialized(&td->clip_table_lock)) {
877 mtx_lock(&td->clip_table_lock);
878 TAILQ_FOREACH_SAFE(ce, &td->clip_table, link, ce_temp) {
879 KASSERT(ce->refcount == 0,
880 ("%s: CLIP entry %p still in use (%d)", __func__,
882 TAILQ_REMOVE(&td->clip_table, ce, link);
883 delete_lip(sc, &ce->lip);
886 mtx_unlock(&td->clip_table_lock);
887 mtx_destroy(&td->clip_table_lock);
892 free_tom_data(struct adapter *sc, struct tom_data *td)
895 ASSERT_SYNCHRONIZED_OP(sc);
897 KASSERT(TAILQ_EMPTY(&td->toep_list),
898 ("%s: TOE PCB list is not empty.", __func__));
899 KASSERT(td->lctx_count == 0,
900 ("%s: lctx hash table is not empty.", __func__));
902 t4_uninit_l2t_cpl_handlers(sc);
903 t4_uninit_cpl_io_handlers(sc);
904 t4_uninit_ddp(sc, td);
905 destroy_clip_table(sc, td);
907 if (td->listen_mask != 0)
908 hashdestroy(td->listen_hash, M_CXGBE, td->listen_mask);
910 if (mtx_initialized(&td->unsent_wr_lock))
911 mtx_destroy(&td->unsent_wr_lock);
912 if (mtx_initialized(&td->lctx_hash_lock))
913 mtx_destroy(&td->lctx_hash_lock);
914 if (mtx_initialized(&td->toep_list_lock))
915 mtx_destroy(&td->toep_list_lock);
917 free_tid_tabs(&sc->tids);
922 reclaim_wr_resources(void *arg, int count)
924 struct tom_data *td = arg;
925 STAILQ_HEAD(, wrqe) twr_list = STAILQ_HEAD_INITIALIZER(twr_list);
926 struct cpl_act_open_req *cpl;
931 mtx_lock(&td->unsent_wr_lock);
932 STAILQ_SWAP(&td->unsent_wr_list, &twr_list, wrqe);
933 mtx_unlock(&td->unsent_wr_lock);
935 while ((wr = STAILQ_FIRST(&twr_list)) != NULL) {
936 STAILQ_REMOVE_HEAD(&twr_list, link);
939 opcode = GET_OPCODE(cpl);
942 case CPL_ACT_OPEN_REQ:
943 case CPL_ACT_OPEN_REQ6:
944 atid = G_TID_TID(be32toh(OPCODE_TID(cpl)));
947 CTR2(KTR_CXGBE, "%s: atid %u ", __func__, atid);
948 act_open_failure_cleanup(sc, atid, EHOSTUNREACH);
952 log(LOG_ERR, "%s: leaked work request %p, wr_len %d, "
953 "opcode %x\n", __func__, wr, wr->wr_len, opcode);
954 /* WR not freed here; go look at it with a debugger. */
960 * Ground control to Major TOM
961 * Commencing countdown, engines on
964 t4_tom_activate(struct adapter *sc)
970 ASSERT_SYNCHRONIZED_OP(sc);
972 /* per-adapter softc for TOM */
973 td = malloc(sizeof(*td), M_CXGBE, M_ZERO | M_NOWAIT);
977 /* List of TOE PCBs and associated lock */
978 mtx_init(&td->toep_list_lock, "PCB list lock", NULL, MTX_DEF);
979 TAILQ_INIT(&td->toep_list);
982 mtx_init(&td->lctx_hash_lock, "lctx hash lock", NULL, MTX_DEF);
983 td->listen_hash = hashinit_flags(LISTEN_HASH_SIZE, M_CXGBE,
984 &td->listen_mask, HASH_NOWAIT);
986 /* List of WRs for which L2 resolution failed */
987 mtx_init(&td->unsent_wr_lock, "Unsent WR list lock", NULL, MTX_DEF);
988 STAILQ_INIT(&td->unsent_wr_list);
989 TASK_INIT(&td->reclaim_wr_resources, 0, reclaim_wr_resources, td);
992 rc = alloc_tid_tabs(&sc->tids);
996 /* DDP page pods and CPL handlers */
999 /* CLIP table for IPv6 offload */
1000 init_clip_table(sc, td);
1003 t4_init_connect_cpl_handlers(sc);
1004 t4_init_l2t_cpl_handlers(sc);
1005 t4_init_listen_cpl_handlers(sc);
1006 t4_init_cpl_io_handlers(sc);
1011 tod->tod_softc = sc;
1012 tod->tod_connect = t4_connect;
1013 tod->tod_listen_start = t4_listen_start;
1014 tod->tod_listen_stop = t4_listen_stop;
1015 tod->tod_rcvd = t4_rcvd;
1016 tod->tod_output = t4_tod_output;
1017 tod->tod_send_rst = t4_send_rst;
1018 tod->tod_send_fin = t4_send_fin;
1019 tod->tod_pcb_detach = t4_pcb_detach;
1020 tod->tod_l2_update = t4_l2_update;
1021 tod->tod_syncache_added = t4_syncache_added;
1022 tod->tod_syncache_removed = t4_syncache_removed;
1023 tod->tod_syncache_respond = t4_syncache_respond;
1024 tod->tod_offload_socket = t4_offload_socket;
1025 tod->tod_ctloutput = t4_ctloutput;
1027 for_each_port(sc, i)
1028 TOEDEV(sc->port[i]->ifp) = &td->tod;
1031 register_toedev(sc->tom_softc);
1035 free_tom_data(sc, td);
1040 t4_tom_deactivate(struct adapter *sc)
1043 struct tom_data *td = sc->tom_softc;
1045 ASSERT_SYNCHRONIZED_OP(sc);
1048 return (0); /* XXX. KASSERT? */
1050 if (sc->offload_map != 0)
1051 return (EBUSY); /* at least one port has IFCAP_TOE enabled */
1053 if (uld_active(sc, ULD_IWARP) || uld_active(sc, ULD_ISCSI))
1054 return (EBUSY); /* both iWARP and iSCSI rely on the TOE. */
1056 mtx_lock(&td->toep_list_lock);
1057 if (!TAILQ_EMPTY(&td->toep_list))
1059 mtx_unlock(&td->toep_list_lock);
1061 mtx_lock(&td->lctx_hash_lock);
1062 if (td->lctx_count > 0)
1064 mtx_unlock(&td->lctx_hash_lock);
1066 taskqueue_drain(taskqueue_thread, &td->reclaim_wr_resources);
1067 mtx_lock(&td->unsent_wr_lock);
1068 if (!STAILQ_EMPTY(&td->unsent_wr_list))
1070 mtx_unlock(&td->unsent_wr_lock);
1073 unregister_toedev(sc->tom_softc);
1074 free_tom_data(sc, td);
1075 sc->tom_softc = NULL;
1082 t4_tom_ifaddr_event(void *arg __unused, struct ifnet *ifp)
1085 atomic_add_rel_int(&in6_ifaddr_gen, 1);
1086 taskqueue_enqueue_timeout(taskqueue_thread, &clip_task, -hz / 4);
1090 t4_tom_mod_load(void)
1093 struct protosw *tcp_protosw, *tcp6_protosw;
1095 tcp_protosw = pffindproto(PF_INET, IPPROTO_TCP, SOCK_STREAM);
1096 if (tcp_protosw == NULL)
1097 return (ENOPROTOOPT);
1098 bcopy(tcp_protosw, &ddp_protosw, sizeof(ddp_protosw));
1099 bcopy(tcp_protosw->pr_usrreqs, &ddp_usrreqs, sizeof(ddp_usrreqs));
1100 ddp_usrreqs.pru_soreceive = t4_soreceive_ddp;
1101 ddp_protosw.pr_usrreqs = &ddp_usrreqs;
1103 tcp6_protosw = pffindproto(PF_INET6, IPPROTO_TCP, SOCK_STREAM);
1104 if (tcp6_protosw == NULL)
1105 return (ENOPROTOOPT);
1106 bcopy(tcp6_protosw, &ddp6_protosw, sizeof(ddp6_protosw));
1107 bcopy(tcp6_protosw->pr_usrreqs, &ddp6_usrreqs, sizeof(ddp6_usrreqs));
1108 ddp6_usrreqs.pru_soreceive = t4_soreceive_ddp;
1109 ddp6_protosw.pr_usrreqs = &ddp6_usrreqs;
1111 TIMEOUT_TASK_INIT(taskqueue_thread, &clip_task, 0, t4_clip_task, NULL);
1112 ifaddr_evhandler = EVENTHANDLER_REGISTER(ifaddr_event,
1113 t4_tom_ifaddr_event, NULL, EVENTHANDLER_PRI_ANY);
1115 rc = t4_register_uld(&tom_uld_info);
1117 t4_tom_mod_unload();
1123 tom_uninit(struct adapter *sc, void *arg __unused)
1125 if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tomun"))
1128 /* Try to free resources (works only if no port has IFCAP_TOE) */
1129 if (uld_active(sc, ULD_TOM))
1130 t4_deactivate_uld(sc, ULD_TOM);
1132 end_synchronized_op(sc, 0);
1136 t4_tom_mod_unload(void)
1138 t4_iterate(tom_uninit, NULL);
1140 if (t4_unregister_uld(&tom_uld_info) == EBUSY)
1143 if (ifaddr_evhandler) {
1144 EVENTHANDLER_DEREGISTER(ifaddr_event, ifaddr_evhandler);
1145 taskqueue_cancel_timeout(taskqueue_thread, &clip_task, NULL);
1150 #endif /* TCP_OFFLOAD */
1153 t4_tom_modevent(module_t mod, int cmd, void *arg)
1160 rc = t4_tom_mod_load();
1164 rc = t4_tom_mod_unload();
1171 printf("t4_tom: compiled without TCP_OFFLOAD support.\n");
1177 static moduledata_t t4_tom_moddata= {
1183 MODULE_VERSION(t4_tom, 1);
1184 MODULE_DEPEND(t4_tom, toecore, 1, 1, 1);
1185 MODULE_DEPEND(t4_tom, t4nex, 1, 1, 1);
1186 DECLARE_MODULE(t4_tom, t4_tom_moddata, SI_SUB_EXEC, SI_ORDER_ANY);