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/limits.h>
40 #include <sys/module.h>
41 #include <sys/protosw.h>
42 #include <sys/domain.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/taskqueue.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 *);
101 static void reclaim_wr_resources(void *, int);
103 static int in6_ifaddr_gen;
104 static eventhandler_tag ifaddr_evhandler;
105 static struct timeout_task clip_task;
108 mbufq_init(struct mbufq *q, int limit)
111 q->head = q->tail = NULL;
115 mbufq_drain(struct mbufq *q)
119 while ((m = q->head) != NULL) {
120 q->head = m->m_nextpkt;
127 mbufq_len(const struct mbufq *q)
133 for (m = q->head; m != NULL; m = m->m_nextpkt)
139 alloc_toepcb(struct vi_info *vi, int txqid, int rxqid, int flags)
141 struct port_info *pi = vi->pi;
142 struct adapter *sc = pi->adapter;
144 int tx_credits, txsd_total, len;
147 * The firmware counts tx work request credits in units of 16 bytes
148 * each. Reserve room for an ABORT_REQ so the driver never has to worry
149 * about tx credits if it wants to abort a connection.
151 tx_credits = sc->params.ofldq_wr_cred;
152 tx_credits -= howmany(sizeof(struct cpl_abort_req), 16);
155 * Shortest possible tx work request is a fw_ofld_tx_data_wr + 1 byte
156 * immediate payload, and firmware counts tx work request credits in
157 * units of 16 byte. Calculate the maximum work requests possible.
159 txsd_total = tx_credits /
160 howmany(sizeof(struct fw_ofld_tx_data_wr) + 1, 16);
163 txqid = (arc4random() % vi->nofldtxq) + vi->first_ofld_txq;
164 KASSERT(txqid >= vi->first_ofld_txq &&
165 txqid < vi->first_ofld_txq + vi->nofldtxq,
166 ("%s: txqid %d for vi %p (first %d, n %d)", __func__, txqid, vi,
167 vi->first_ofld_txq, vi->nofldtxq));
170 rxqid = (arc4random() % vi->nofldrxq) + vi->first_ofld_rxq;
171 KASSERT(rxqid >= vi->first_ofld_rxq &&
172 rxqid < vi->first_ofld_rxq + vi->nofldrxq,
173 ("%s: rxqid %d for vi %p (first %d, n %d)", __func__, rxqid, vi,
174 vi->first_ofld_rxq, vi->nofldrxq));
176 len = offsetof(struct toepcb, txsd) +
177 txsd_total * sizeof(struct ofld_tx_sdesc);
179 toep = malloc(len, M_CXGBE, M_ZERO | flags);
183 toep->td = sc->tom_softc;
185 toep->tx_total = tx_credits;
186 toep->tx_credits = tx_credits;
187 toep->ofld_txq = &sc->sge.ofld_txq[txqid];
188 toep->ofld_rxq = &sc->sge.ofld_rxq[rxqid];
189 toep->ctrlq = &sc->sge.ctrlq[pi->port_id];
190 mbufq_init(&toep->ulp_pduq, INT_MAX);
191 mbufq_init(&toep->ulp_pdu_reclaimq, INT_MAX);
192 toep->txsd_total = txsd_total;
193 toep->txsd_avail = txsd_total;
201 free_toepcb(struct toepcb *toep)
204 KASSERT(!(toep->flags & TPF_ATTACHED),
205 ("%s: attached to an inpcb", __func__));
206 KASSERT(!(toep->flags & TPF_CPL_PENDING),
207 ("%s: CPL pending", __func__));
213 * Set up the socket for TCP offload.
216 offload_socket(struct socket *so, struct toepcb *toep)
218 struct tom_data *td = toep->td;
219 struct inpcb *inp = sotoinpcb(so);
220 struct tcpcb *tp = intotcpcb(inp);
223 INP_WLOCK_ASSERT(inp);
228 sb->sb_flags |= SB_NOCOALESCE;
232 sb->sb_flags |= SB_NOCOALESCE;
233 if (toep->ulp_mode == ULP_MODE_TCPDDP) {
234 if (inp->inp_vflag & INP_IPV6)
235 so->so_proto = &ddp6_protosw;
237 so->so_proto = &ddp_protosw;
244 tp->t_flags |= TF_TOE;
246 /* Install an extra hold on inp */
248 toep->flags |= TPF_ATTACHED;
251 /* Add the TOE PCB to the active list */
252 mtx_lock(&td->toep_list_lock);
253 TAILQ_INSERT_HEAD(&td->toep_list, toep, link);
254 mtx_unlock(&td->toep_list_lock);
257 /* This is _not_ the normal way to "unoffload" a socket. */
259 undo_offload_socket(struct socket *so)
261 struct inpcb *inp = sotoinpcb(so);
262 struct tcpcb *tp = intotcpcb(inp);
263 struct toepcb *toep = tp->t_toe;
264 struct tom_data *td = toep->td;
267 INP_WLOCK_ASSERT(inp);
271 sb->sb_flags &= ~SB_NOCOALESCE;
275 sb->sb_flags &= ~SB_NOCOALESCE;
280 tp->t_flags &= ~TF_TOE;
283 toep->flags &= ~TPF_ATTACHED;
284 if (in_pcbrele_wlocked(inp))
285 panic("%s: inp freed.", __func__);
287 mtx_lock(&td->toep_list_lock);
288 TAILQ_REMOVE(&td->toep_list, toep, link);
289 mtx_unlock(&td->toep_list_lock);
293 release_offload_resources(struct toepcb *toep)
295 struct tom_data *td = toep->td;
296 struct adapter *sc = td_adapter(td);
299 KASSERT(!(toep->flags & TPF_CPL_PENDING),
300 ("%s: %p has CPL pending.", __func__, toep));
301 KASSERT(!(toep->flags & TPF_ATTACHED),
302 ("%s: %p is still attached.", __func__, toep));
304 CTR5(KTR_CXGBE, "%s: toep %p (tid %d, l2te %p, ce %p)",
305 __func__, toep, tid, toep->l2te, toep->ce);
308 * These queues should have been emptied at approximately the same time
309 * that a normal connection's socket's so_snd would have been purged or
310 * drained. Do _not_ clean up here.
312 MPASS(mbufq_len(&toep->ulp_pduq) == 0);
313 MPASS(mbufq_len(&toep->ulp_pdu_reclaimq) == 0);
315 if (toep->ulp_mode == ULP_MODE_TCPDDP)
316 release_ddp_resources(toep);
319 t4_l2t_release(toep->l2te);
323 release_tid(sc, tid, toep->ctrlq);
327 release_lip(td, toep->ce);
329 mtx_lock(&td->toep_list_lock);
330 TAILQ_REMOVE(&td->toep_list, toep, link);
331 mtx_unlock(&td->toep_list_lock);
337 * The kernel is done with the TCP PCB and this is our opportunity to unhook the
338 * toepcb hanging off of it. If the TOE driver is also done with the toepcb (no
339 * pending CPL) then it is time to release all resources tied to the toepcb.
341 * Also gets called when an offloaded active open fails and the TOM wants the
342 * kernel to take the TCP PCB back.
345 t4_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp)
347 #if defined(KTR) || defined(INVARIANTS)
348 struct inpcb *inp = tp->t_inpcb;
350 struct toepcb *toep = tp->t_toe;
352 INP_WLOCK_ASSERT(inp);
354 KASSERT(toep != NULL, ("%s: toep is NULL", __func__));
355 KASSERT(toep->flags & TPF_ATTACHED,
356 ("%s: not attached", __func__));
359 if (tp->t_state == TCPS_SYN_SENT) {
360 CTR6(KTR_CXGBE, "%s: atid %d, toep %p (0x%x), inp %p (0x%x)",
361 __func__, toep->tid, toep, toep->flags, inp,
365 "t4_pcb_detach: tid %d (%s), toep %p (0x%x), inp %p (0x%x)",
366 toep->tid, tcpstates[tp->t_state], toep, toep->flags, inp,
372 tp->t_flags &= ~TF_TOE;
373 toep->flags &= ~TPF_ATTACHED;
375 if (!(toep->flags & TPF_CPL_PENDING))
376 release_offload_resources(toep);
380 * setsockopt handler.
383 t4_ctloutput(struct toedev *tod, struct tcpcb *tp, int dir, int name)
385 struct adapter *sc = tod->tod_softc;
386 struct toepcb *toep = tp->t_toe;
391 CTR4(KTR_CXGBE, "%s: tp %p, dir %u, name %u", __func__, tp, dir, name);
395 t4_set_tcb_field(sc, toep, 1, W_TCB_T_FLAGS, V_TF_NAGLE(1),
396 V_TF_NAGLE(tp->t_flags & TF_NODELAY ? 0 : 1));
404 * The TOE driver will not receive any more CPLs for the tid associated with the
405 * toepcb; release the hold on the inpcb.
408 final_cpl_received(struct toepcb *toep)
410 struct inpcb *inp = toep->inp;
412 KASSERT(inp != NULL, ("%s: inp is NULL", __func__));
413 INP_WLOCK_ASSERT(inp);
414 KASSERT(toep->flags & TPF_CPL_PENDING,
415 ("%s: CPL not pending already?", __func__));
417 CTR6(KTR_CXGBE, "%s: tid %d, toep %p (0x%x), inp %p (0x%x)",
418 __func__, toep->tid, toep, toep->flags, inp, inp->inp_flags);
421 toep->flags &= ~TPF_CPL_PENDING;
422 mbufq_drain(&toep->ulp_pdu_reclaimq);
424 if (!(toep->flags & TPF_ATTACHED))
425 release_offload_resources(toep);
427 if (!in_pcbrele_wlocked(inp))
432 insert_tid(struct adapter *sc, int tid, void *ctx)
434 struct tid_info *t = &sc->tids;
436 t->tid_tab[tid] = ctx;
437 atomic_add_int(&t->tids_in_use, 1);
441 lookup_tid(struct adapter *sc, int tid)
443 struct tid_info *t = &sc->tids;
445 return (t->tid_tab[tid]);
449 update_tid(struct adapter *sc, int tid, void *ctx)
451 struct tid_info *t = &sc->tids;
453 t->tid_tab[tid] = ctx;
457 remove_tid(struct adapter *sc, int tid)
459 struct tid_info *t = &sc->tids;
461 t->tid_tab[tid] = NULL;
462 atomic_subtract_int(&t->tids_in_use, 1);
466 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
469 struct cpl_tid_release *req;
471 wr = alloc_wrqe(sizeof(*req), ctrlq);
473 queue_tid_release(sc, tid); /* defer */
478 INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
484 queue_tid_release(struct adapter *sc, int tid)
487 CXGBE_UNIMPLEMENTED("deferred tid release");
491 * What mtu_idx to use, given a 4-tuple and/or an MSS cap
494 find_best_mtu_idx(struct adapter *sc, struct in_conninfo *inc, int pmss)
496 unsigned short *mtus = &sc->params.mtus[0];
499 KASSERT(inc != NULL || pmss > 0,
500 ("%s: at least one of inc/pmss must be specified", __func__));
502 mss = inc ? tcp_mssopt(inc) : pmss;
503 if (pmss > 0 && mss > pmss)
506 if (inc->inc_flags & INC_ISIPV6)
507 n = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
509 n = sizeof(struct ip) + sizeof(struct tcphdr);
511 for (i = 0; i < NMTUS - 1 && mtus[i + 1] <= mss + n; i++)
518 * Determine the receive window size for a socket.
521 select_rcv_wnd(struct socket *so)
525 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
527 wnd = sbspace(&so->so_rcv);
528 if (wnd < MIN_RCV_WND)
531 return min(wnd, MAX_RCV_WND);
535 select_rcv_wscale(void)
538 unsigned long space = sb_max;
540 if (space > MAX_RCV_WND)
543 while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < space)
549 extern int always_keepalive;
550 #define VIID_SMACIDX(v) (((unsigned int)(v) & 0x7f) << 1)
553 * socket so could be a listening socket too.
556 calc_opt0(struct socket *so, struct vi_info *vi, struct l2t_entry *e,
557 int mtu_idx, int rscale, int rx_credits, int ulp_mode)
561 KASSERT(rx_credits <= M_RCV_BUFSIZ,
562 ("%s: rcv_bufsiz too high", __func__));
564 opt0 = F_TCAM_BYPASS | V_WND_SCALE(rscale) | V_MSS_IDX(mtu_idx) |
565 V_ULP_MODE(ulp_mode) | V_RCV_BUFSIZ(rx_credits);
568 struct inpcb *inp = sotoinpcb(so);
569 struct tcpcb *tp = intotcpcb(inp);
570 int keepalive = always_keepalive ||
571 so_options_get(so) & SO_KEEPALIVE;
573 opt0 |= V_NAGLE((tp->t_flags & TF_NODELAY) == 0);
574 opt0 |= V_KEEP_ALIVE(keepalive != 0);
578 opt0 |= V_L2T_IDX(e->idx);
581 opt0 |= V_SMAC_SEL(VIID_SMACIDX(vi->viid));
582 opt0 |= V_TX_CHAN(vi->pi->tx_chan);
585 return htobe64(opt0);
589 select_ntuple(struct vi_info *vi, struct l2t_entry *e)
591 struct adapter *sc = vi->pi->adapter;
592 struct tp_params *tp = &sc->params.tp;
593 uint16_t viid = vi->viid;
597 * Initialize each of the fields which we care about which are present
598 * in the Compressed Filter Tuple.
600 if (tp->vlan_shift >= 0 && e->vlan != CPL_L2T_VLAN_NONE)
601 ntuple |= (uint64_t)(F_FT_VLAN_VLD | e->vlan) << tp->vlan_shift;
603 if (tp->port_shift >= 0)
604 ntuple |= (uint64_t)e->lport << tp->port_shift;
606 if (tp->protocol_shift >= 0)
607 ntuple |= (uint64_t)IPPROTO_TCP << tp->protocol_shift;
609 if (tp->vnic_shift >= 0) {
610 uint32_t vf = G_FW_VIID_VIN(viid);
611 uint32_t pf = G_FW_VIID_PFN(viid);
612 uint32_t vld = G_FW_VIID_VIVLD(viid);
614 ntuple |= (uint64_t)(V_FT_VNID_ID_VF(vf) | V_FT_VNID_ID_PF(pf) |
615 V_FT_VNID_ID_VLD(vld)) << tp->vnic_shift;
619 return (htobe32((uint32_t)ntuple));
621 return (htobe64(V_FILTER_TUPLE(ntuple)));
625 set_tcpddp_ulp_mode(struct toepcb *toep)
628 toep->ulp_mode = ULP_MODE_TCPDDP;
629 toep->ddp_flags = DDP_OK;
630 toep->ddp_score = DDP_LOW_SCORE;
634 negative_advice(int status)
637 return (status == CPL_ERR_RTX_NEG_ADVICE ||
638 status == CPL_ERR_PERSIST_NEG_ADVICE ||
639 status == CPL_ERR_KEEPALV_NEG_ADVICE);
643 alloc_tid_tabs(struct tid_info *t)
648 size = t->ntids * sizeof(*t->tid_tab) +
649 t->natids * sizeof(*t->atid_tab) +
650 t->nstids * sizeof(*t->stid_tab);
652 t->tid_tab = malloc(size, M_CXGBE, M_ZERO | M_NOWAIT);
653 if (t->tid_tab == NULL)
656 mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
657 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
658 t->afree = t->atid_tab;
660 for (i = 1; i < t->natids; i++)
661 t->atid_tab[i - 1].next = &t->atid_tab[i];
662 t->atid_tab[t->natids - 1].next = NULL;
664 mtx_init(&t->stid_lock, "stid lock", NULL, MTX_DEF);
665 t->stid_tab = (struct listen_ctx **)&t->atid_tab[t->natids];
667 TAILQ_INIT(&t->stids);
668 t->nstids_free_head = t->nstids;
670 atomic_store_rel_int(&t->tids_in_use, 0);
676 free_tid_tabs(struct tid_info *t)
678 KASSERT(t->tids_in_use == 0,
679 ("%s: %d tids still in use.", __func__, t->tids_in_use));
680 KASSERT(t->atids_in_use == 0,
681 ("%s: %d atids still in use.", __func__, t->atids_in_use));
682 KASSERT(t->stids_in_use == 0,
683 ("%s: %d tids still in use.", __func__, t->stids_in_use));
685 free(t->tid_tab, M_CXGBE);
688 if (mtx_initialized(&t->atid_lock))
689 mtx_destroy(&t->atid_lock);
690 if (mtx_initialized(&t->stid_lock))
691 mtx_destroy(&t->stid_lock);
695 add_lip(struct adapter *sc, struct in6_addr *lip)
697 struct fw_clip_cmd c;
699 ASSERT_SYNCHRONIZED_OP(sc);
700 /* mtx_assert(&td->clip_table_lock, MA_OWNED); */
702 memset(&c, 0, sizeof(c));
703 c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST |
705 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_ALLOC | FW_LEN16(c));
706 c.ip_hi = *(uint64_t *)&lip->s6_addr[0];
707 c.ip_lo = *(uint64_t *)&lip->s6_addr[8];
709 return (-t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c));
713 delete_lip(struct adapter *sc, struct in6_addr *lip)
715 struct fw_clip_cmd c;
717 ASSERT_SYNCHRONIZED_OP(sc);
718 /* mtx_assert(&td->clip_table_lock, MA_OWNED); */
720 memset(&c, 0, sizeof(c));
721 c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST |
723 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_FREE | FW_LEN16(c));
724 c.ip_hi = *(uint64_t *)&lip->s6_addr[0];
725 c.ip_lo = *(uint64_t *)&lip->s6_addr[8];
727 return (-t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c));
730 static struct clip_entry *
731 search_lip(struct tom_data *td, struct in6_addr *lip)
733 struct clip_entry *ce;
735 mtx_assert(&td->clip_table_lock, MA_OWNED);
737 TAILQ_FOREACH(ce, &td->clip_table, link) {
738 if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip))
746 hold_lip(struct tom_data *td, struct in6_addr *lip)
748 struct clip_entry *ce;
750 mtx_lock(&td->clip_table_lock);
751 ce = search_lip(td, lip);
754 mtx_unlock(&td->clip_table_lock);
760 release_lip(struct tom_data *td, struct clip_entry *ce)
763 mtx_lock(&td->clip_table_lock);
764 KASSERT(search_lip(td, &ce->lip) == ce,
765 ("%s: CLIP entry %p p not in CLIP table.", __func__, ce));
766 KASSERT(ce->refcount > 0,
767 ("%s: CLIP entry %p has refcount 0", __func__, ce));
769 mtx_unlock(&td->clip_table_lock);
773 init_clip_table(struct adapter *sc, struct tom_data *td)
776 ASSERT_SYNCHRONIZED_OP(sc);
778 mtx_init(&td->clip_table_lock, "CLIP table lock", NULL, MTX_DEF);
779 TAILQ_INIT(&td->clip_table);
782 update_clip_table(sc, td);
786 update_clip(struct adapter *sc, void *arg __unused)
789 if (begin_synchronized_op(sc, NULL, HOLD_LOCK, "t4tomuc"))
792 if (uld_active(sc, ULD_TOM))
793 update_clip_table(sc, sc->tom_softc);
795 end_synchronized_op(sc, LOCK_HELD);
799 t4_clip_task(void *arg, int count)
802 t4_iterate(update_clip, NULL);
806 update_clip_table(struct adapter *sc, struct tom_data *td)
808 struct in6_ifaddr *ia;
809 struct in6_addr *lip, tlip;
810 struct clip_head stale;
811 struct clip_entry *ce, *ce_temp;
812 int rc, gen = atomic_load_acq_int(&in6_ifaddr_gen);
814 ASSERT_SYNCHRONIZED_OP(sc);
817 mtx_lock(&td->clip_table_lock);
819 if (gen == td->clip_gen)
823 TAILQ_CONCAT(&stale, &td->clip_table, link);
825 TAILQ_FOREACH(ia, &V_in6_ifaddrhead, ia_link) {
826 lip = &ia->ia_addr.sin6_addr;
828 KASSERT(!IN6_IS_ADDR_MULTICAST(lip),
829 ("%s: mcast address in in6_ifaddr list", __func__));
831 if (IN6_IS_ADDR_LOOPBACK(lip))
833 if (IN6_IS_SCOPE_EMBED(lip)) {
834 /* Remove the embedded scope */
840 * XXX: how to weed out the link local address for the loopback
841 * interface? It's fe80::1 usually (always?).
845 * If it's in the main list then we already know it's not stale.
847 TAILQ_FOREACH(ce, &td->clip_table, link) {
848 if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip))
853 * If it's in the stale list we should move it to the main list.
855 TAILQ_FOREACH(ce, &stale, link) {
856 if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip)) {
857 TAILQ_REMOVE(&stale, ce, link);
858 TAILQ_INSERT_TAIL(&td->clip_table, ce, link);
863 /* A new IP6 address; add it to the CLIP table */
864 ce = malloc(sizeof(*ce), M_CXGBE, M_NOWAIT);
865 memcpy(&ce->lip, lip, sizeof(ce->lip));
867 rc = add_lip(sc, lip);
869 TAILQ_INSERT_TAIL(&td->clip_table, ce, link);
871 char ip[INET6_ADDRSTRLEN];
873 inet_ntop(AF_INET6, &ce->lip, &ip[0], sizeof(ip));
874 log(LOG_ERR, "%s: could not add %s (%d)\n",
883 * Remove stale addresses (those no longer in V_in6_ifaddrhead) that are
884 * no longer referenced by the driver.
886 TAILQ_FOREACH_SAFE(ce, &stale, link, ce_temp) {
887 if (ce->refcount == 0) {
888 rc = delete_lip(sc, &ce->lip);
890 TAILQ_REMOVE(&stale, ce, link);
893 char ip[INET6_ADDRSTRLEN];
895 inet_ntop(AF_INET6, &ce->lip, &ip[0],
897 log(LOG_ERR, "%s: could not delete %s (%d)\n",
902 /* The ones that are still referenced need to stay in the CLIP table */
903 TAILQ_CONCAT(&td->clip_table, &stale, link);
907 mtx_unlock(&td->clip_table_lock);
908 IN6_IFADDR_RUNLOCK();
912 destroy_clip_table(struct adapter *sc, struct tom_data *td)
914 struct clip_entry *ce, *ce_temp;
916 if (mtx_initialized(&td->clip_table_lock)) {
917 mtx_lock(&td->clip_table_lock);
918 TAILQ_FOREACH_SAFE(ce, &td->clip_table, link, ce_temp) {
919 KASSERT(ce->refcount == 0,
920 ("%s: CLIP entry %p still in use (%d)", __func__,
922 TAILQ_REMOVE(&td->clip_table, ce, link);
923 delete_lip(sc, &ce->lip);
926 mtx_unlock(&td->clip_table_lock);
927 mtx_destroy(&td->clip_table_lock);
932 free_tom_data(struct adapter *sc, struct tom_data *td)
935 ASSERT_SYNCHRONIZED_OP(sc);
937 KASSERT(TAILQ_EMPTY(&td->toep_list),
938 ("%s: TOE PCB list is not empty.", __func__));
939 KASSERT(td->lctx_count == 0,
940 ("%s: lctx hash table is not empty.", __func__));
942 t4_uninit_l2t_cpl_handlers(sc);
943 t4_uninit_cpl_io_handlers(sc);
944 t4_uninit_ddp(sc, td);
945 destroy_clip_table(sc, td);
947 if (td->listen_mask != 0)
948 hashdestroy(td->listen_hash, M_CXGBE, td->listen_mask);
950 if (mtx_initialized(&td->unsent_wr_lock))
951 mtx_destroy(&td->unsent_wr_lock);
952 if (mtx_initialized(&td->lctx_hash_lock))
953 mtx_destroy(&td->lctx_hash_lock);
954 if (mtx_initialized(&td->toep_list_lock))
955 mtx_destroy(&td->toep_list_lock);
957 free_tid_tabs(&sc->tids);
962 reclaim_wr_resources(void *arg, int count)
964 struct tom_data *td = arg;
965 STAILQ_HEAD(, wrqe) twr_list = STAILQ_HEAD_INITIALIZER(twr_list);
966 struct cpl_act_open_req *cpl;
971 mtx_lock(&td->unsent_wr_lock);
972 STAILQ_SWAP(&td->unsent_wr_list, &twr_list, wrqe);
973 mtx_unlock(&td->unsent_wr_lock);
975 while ((wr = STAILQ_FIRST(&twr_list)) != NULL) {
976 STAILQ_REMOVE_HEAD(&twr_list, link);
979 opcode = GET_OPCODE(cpl);
982 case CPL_ACT_OPEN_REQ:
983 case CPL_ACT_OPEN_REQ6:
984 atid = G_TID_TID(be32toh(OPCODE_TID(cpl)));
987 CTR2(KTR_CXGBE, "%s: atid %u ", __func__, atid);
988 act_open_failure_cleanup(sc, atid, EHOSTUNREACH);
992 log(LOG_ERR, "%s: leaked work request %p, wr_len %d, "
993 "opcode %x\n", __func__, wr, wr->wr_len, opcode);
994 /* WR not freed here; go look at it with a debugger. */
1000 * Ground control to Major TOM
1001 * Commencing countdown, engines on
1004 t4_tom_activate(struct adapter *sc)
1006 struct tom_data *td;
1011 ASSERT_SYNCHRONIZED_OP(sc);
1013 /* per-adapter softc for TOM */
1014 td = malloc(sizeof(*td), M_CXGBE, M_ZERO | M_NOWAIT);
1018 /* List of TOE PCBs and associated lock */
1019 mtx_init(&td->toep_list_lock, "PCB list lock", NULL, MTX_DEF);
1020 TAILQ_INIT(&td->toep_list);
1022 /* Listen context */
1023 mtx_init(&td->lctx_hash_lock, "lctx hash lock", NULL, MTX_DEF);
1024 td->listen_hash = hashinit_flags(LISTEN_HASH_SIZE, M_CXGBE,
1025 &td->listen_mask, HASH_NOWAIT);
1027 /* List of WRs for which L2 resolution failed */
1028 mtx_init(&td->unsent_wr_lock, "Unsent WR list lock", NULL, MTX_DEF);
1029 STAILQ_INIT(&td->unsent_wr_list);
1030 TASK_INIT(&td->reclaim_wr_resources, 0, reclaim_wr_resources, td);
1033 rc = alloc_tid_tabs(&sc->tids);
1037 /* DDP page pods and CPL handlers */
1038 t4_init_ddp(sc, td);
1040 /* CLIP table for IPv6 offload */
1041 init_clip_table(sc, td);
1044 t4_init_connect_cpl_handlers(sc);
1045 t4_init_l2t_cpl_handlers(sc);
1046 t4_init_listen_cpl_handlers(sc);
1047 t4_init_cpl_io_handlers(sc);
1052 tod->tod_softc = sc;
1053 tod->tod_connect = t4_connect;
1054 tod->tod_listen_start = t4_listen_start;
1055 tod->tod_listen_stop = t4_listen_stop;
1056 tod->tod_rcvd = t4_rcvd;
1057 tod->tod_output = t4_tod_output;
1058 tod->tod_send_rst = t4_send_rst;
1059 tod->tod_send_fin = t4_send_fin;
1060 tod->tod_pcb_detach = t4_pcb_detach;
1061 tod->tod_l2_update = t4_l2_update;
1062 tod->tod_syncache_added = t4_syncache_added;
1063 tod->tod_syncache_removed = t4_syncache_removed;
1064 tod->tod_syncache_respond = t4_syncache_respond;
1065 tod->tod_offload_socket = t4_offload_socket;
1066 tod->tod_ctloutput = t4_ctloutput;
1068 for_each_port(sc, i) {
1069 for_each_vi(sc->port[i], v, vi) {
1070 TOEDEV(vi->ifp) = &td->tod;
1075 register_toedev(sc->tom_softc);
1079 free_tom_data(sc, td);
1084 t4_tom_deactivate(struct adapter *sc)
1087 struct tom_data *td = sc->tom_softc;
1089 ASSERT_SYNCHRONIZED_OP(sc);
1092 return (0); /* XXX. KASSERT? */
1094 if (sc->offload_map != 0)
1095 return (EBUSY); /* at least one port has IFCAP_TOE enabled */
1097 if (uld_active(sc, ULD_IWARP) || uld_active(sc, ULD_ISCSI))
1098 return (EBUSY); /* both iWARP and iSCSI rely on the TOE. */
1100 mtx_lock(&td->toep_list_lock);
1101 if (!TAILQ_EMPTY(&td->toep_list))
1103 mtx_unlock(&td->toep_list_lock);
1105 mtx_lock(&td->lctx_hash_lock);
1106 if (td->lctx_count > 0)
1108 mtx_unlock(&td->lctx_hash_lock);
1110 taskqueue_drain(taskqueue_thread, &td->reclaim_wr_resources);
1111 mtx_lock(&td->unsent_wr_lock);
1112 if (!STAILQ_EMPTY(&td->unsent_wr_list))
1114 mtx_unlock(&td->unsent_wr_lock);
1117 unregister_toedev(sc->tom_softc);
1118 free_tom_data(sc, td);
1119 sc->tom_softc = NULL;
1126 t4_tom_ifaddr_event(void *arg __unused, struct ifnet *ifp)
1129 atomic_add_rel_int(&in6_ifaddr_gen, 1);
1130 taskqueue_enqueue_timeout(taskqueue_thread, &clip_task, -hz / 4);
1134 t4_tom_mod_load(void)
1137 struct protosw *tcp_protosw, *tcp6_protosw;
1139 tcp_protosw = pffindproto(PF_INET, IPPROTO_TCP, SOCK_STREAM);
1140 if (tcp_protosw == NULL)
1141 return (ENOPROTOOPT);
1142 bcopy(tcp_protosw, &ddp_protosw, sizeof(ddp_protosw));
1143 bcopy(tcp_protosw->pr_usrreqs, &ddp_usrreqs, sizeof(ddp_usrreqs));
1144 ddp_usrreqs.pru_soreceive = t4_soreceive_ddp;
1145 ddp_protosw.pr_usrreqs = &ddp_usrreqs;
1147 tcp6_protosw = pffindproto(PF_INET6, IPPROTO_TCP, SOCK_STREAM);
1148 if (tcp6_protosw == NULL)
1149 return (ENOPROTOOPT);
1150 bcopy(tcp6_protosw, &ddp6_protosw, sizeof(ddp6_protosw));
1151 bcopy(tcp6_protosw->pr_usrreqs, &ddp6_usrreqs, sizeof(ddp6_usrreqs));
1152 ddp6_usrreqs.pru_soreceive = t4_soreceive_ddp;
1153 ddp6_protosw.pr_usrreqs = &ddp6_usrreqs;
1155 TIMEOUT_TASK_INIT(taskqueue_thread, &clip_task, 0, t4_clip_task, NULL);
1156 ifaddr_evhandler = EVENTHANDLER_REGISTER(ifaddr_event,
1157 t4_tom_ifaddr_event, NULL, EVENTHANDLER_PRI_ANY);
1159 rc = t4_register_uld(&tom_uld_info);
1161 t4_tom_mod_unload();
1167 tom_uninit(struct adapter *sc, void *arg __unused)
1169 if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tomun"))
1172 /* Try to free resources (works only if no port has IFCAP_TOE) */
1173 if (uld_active(sc, ULD_TOM))
1174 t4_deactivate_uld(sc, ULD_TOM);
1176 end_synchronized_op(sc, 0);
1180 t4_tom_mod_unload(void)
1182 t4_iterate(tom_uninit, NULL);
1184 if (t4_unregister_uld(&tom_uld_info) == EBUSY)
1187 if (ifaddr_evhandler) {
1188 EVENTHANDLER_DEREGISTER(ifaddr_event, ifaddr_evhandler);
1189 taskqueue_cancel_timeout(taskqueue_thread, &clip_task, NULL);
1194 #endif /* TCP_OFFLOAD */
1197 t4_tom_modevent(module_t mod, int cmd, void *arg)
1204 rc = t4_tom_mod_load();
1208 rc = t4_tom_mod_unload();
1215 printf("t4_tom: compiled without TCP_OFFLOAD support.\n");
1221 static moduledata_t t4_tom_moddata= {
1227 MODULE_VERSION(t4_tom, 1);
1228 MODULE_DEPEND(t4_tom, toecore, 1, 1, 1);
1229 MODULE_DEPEND(t4_tom, t4nex, 1, 1, 1);
1230 DECLARE_MODULE(t4_tom, t4_tom_moddata, SI_SUB_EXEC, SI_ORDER_ANY);