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;
128 mbufq_len(const struct mbufq *q)
134 for (m = q->head; m != NULL; m = m->m_nextpkt)
141 alloc_toepcb(struct vi_info *vi, int txqid, int rxqid, int flags)
143 struct port_info *pi = vi->pi;
144 struct adapter *sc = pi->adapter;
146 int tx_credits, txsd_total, len;
149 * The firmware counts tx work request credits in units of 16 bytes
150 * each. Reserve room for an ABORT_REQ so the driver never has to worry
151 * about tx credits if it wants to abort a connection.
153 tx_credits = sc->params.ofldq_wr_cred;
154 tx_credits -= howmany(sizeof(struct cpl_abort_req), 16);
157 * Shortest possible tx work request is a fw_ofld_tx_data_wr + 1 byte
158 * immediate payload, and firmware counts tx work request credits in
159 * units of 16 byte. Calculate the maximum work requests possible.
161 txsd_total = tx_credits /
162 howmany(sizeof(struct fw_ofld_tx_data_wr) + 1, 16);
165 txqid = (arc4random() % vi->nofldtxq) + vi->first_ofld_txq;
166 KASSERT(txqid >= vi->first_ofld_txq &&
167 txqid < vi->first_ofld_txq + vi->nofldtxq,
168 ("%s: txqid %d for vi %p (first %d, n %d)", __func__, txqid, vi,
169 vi->first_ofld_txq, vi->nofldtxq));
172 rxqid = (arc4random() % vi->nofldrxq) + vi->first_ofld_rxq;
173 KASSERT(rxqid >= vi->first_ofld_rxq &&
174 rxqid < vi->first_ofld_rxq + vi->nofldrxq,
175 ("%s: rxqid %d for vi %p (first %d, n %d)", __func__, rxqid, vi,
176 vi->first_ofld_rxq, vi->nofldrxq));
178 len = offsetof(struct toepcb, txsd) +
179 txsd_total * sizeof(struct ofld_tx_sdesc);
181 toep = malloc(len, M_CXGBE, M_ZERO | flags);
185 toep->td = sc->tom_softc;
187 toep->tx_total = tx_credits;
188 toep->tx_credits = tx_credits;
189 toep->ofld_txq = &sc->sge.ofld_txq[txqid];
190 toep->ofld_rxq = &sc->sge.ofld_rxq[rxqid];
191 toep->ctrlq = &sc->sge.ctrlq[pi->port_id];
192 mbufq_init(&toep->ulp_pduq, INT_MAX);
193 mbufq_init(&toep->ulp_pdu_reclaimq, INT_MAX);
194 toep->txsd_total = txsd_total;
195 toep->txsd_avail = txsd_total;
203 free_toepcb(struct toepcb *toep)
206 KASSERT(!(toep->flags & TPF_ATTACHED),
207 ("%s: attached to an inpcb", __func__));
208 KASSERT(!(toep->flags & TPF_CPL_PENDING),
209 ("%s: CPL pending", __func__));
215 * Set up the socket for TCP offload.
218 offload_socket(struct socket *so, struct toepcb *toep)
220 struct tom_data *td = toep->td;
221 struct inpcb *inp = sotoinpcb(so);
222 struct tcpcb *tp = intotcpcb(inp);
225 INP_WLOCK_ASSERT(inp);
230 sb->sb_flags |= SB_NOCOALESCE;
234 sb->sb_flags |= SB_NOCOALESCE;
235 if (toep->ulp_mode == ULP_MODE_TCPDDP) {
236 if (inp->inp_vflag & INP_IPV6)
237 so->so_proto = &ddp6_protosw;
239 so->so_proto = &ddp_protosw;
246 tp->t_flags |= TF_TOE;
248 /* Install an extra hold on inp */
250 toep->flags |= TPF_ATTACHED;
253 /* Add the TOE PCB to the active list */
254 mtx_lock(&td->toep_list_lock);
255 TAILQ_INSERT_HEAD(&td->toep_list, toep, link);
256 mtx_unlock(&td->toep_list_lock);
259 /* This is _not_ the normal way to "unoffload" a socket. */
261 undo_offload_socket(struct socket *so)
263 struct inpcb *inp = sotoinpcb(so);
264 struct tcpcb *tp = intotcpcb(inp);
265 struct toepcb *toep = tp->t_toe;
266 struct tom_data *td = toep->td;
269 INP_WLOCK_ASSERT(inp);
273 sb->sb_flags &= ~SB_NOCOALESCE;
277 sb->sb_flags &= ~SB_NOCOALESCE;
282 tp->t_flags &= ~TF_TOE;
285 toep->flags &= ~TPF_ATTACHED;
286 if (in_pcbrele_wlocked(inp))
287 panic("%s: inp freed.", __func__);
289 mtx_lock(&td->toep_list_lock);
290 TAILQ_REMOVE(&td->toep_list, toep, link);
291 mtx_unlock(&td->toep_list_lock);
295 release_offload_resources(struct toepcb *toep)
297 struct tom_data *td = toep->td;
298 struct adapter *sc = td_adapter(td);
301 KASSERT(!(toep->flags & TPF_CPL_PENDING),
302 ("%s: %p has CPL pending.", __func__, toep));
303 KASSERT(!(toep->flags & TPF_ATTACHED),
304 ("%s: %p is still attached.", __func__, toep));
306 CTR5(KTR_CXGBE, "%s: toep %p (tid %d, l2te %p, ce %p)",
307 __func__, toep, tid, toep->l2te, toep->ce);
310 * These queues should have been emptied at approximately the same time
311 * that a normal connection's socket's so_snd would have been purged or
312 * drained. Do _not_ clean up here.
314 MPASS(mbufq_len(&toep->ulp_pduq) == 0);
315 MPASS(mbufq_len(&toep->ulp_pdu_reclaimq) == 0);
317 if (toep->ulp_mode == ULP_MODE_TCPDDP)
318 release_ddp_resources(toep);
321 t4_l2t_release(toep->l2te);
325 release_tid(sc, tid, toep->ctrlq);
329 release_lip(td, toep->ce);
331 mtx_lock(&td->toep_list_lock);
332 TAILQ_REMOVE(&td->toep_list, toep, link);
333 mtx_unlock(&td->toep_list_lock);
339 * The kernel is done with the TCP PCB and this is our opportunity to unhook the
340 * toepcb hanging off of it. If the TOE driver is also done with the toepcb (no
341 * pending CPL) then it is time to release all resources tied to the toepcb.
343 * Also gets called when an offloaded active open fails and the TOM wants the
344 * kernel to take the TCP PCB back.
347 t4_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp)
349 #if defined(KTR) || defined(INVARIANTS)
350 struct inpcb *inp = tp->t_inpcb;
352 struct toepcb *toep = tp->t_toe;
354 INP_WLOCK_ASSERT(inp);
356 KASSERT(toep != NULL, ("%s: toep is NULL", __func__));
357 KASSERT(toep->flags & TPF_ATTACHED,
358 ("%s: not attached", __func__));
361 if (tp->t_state == TCPS_SYN_SENT) {
362 CTR6(KTR_CXGBE, "%s: atid %d, toep %p (0x%x), inp %p (0x%x)",
363 __func__, toep->tid, toep, toep->flags, inp,
367 "t4_pcb_detach: tid %d (%s), toep %p (0x%x), inp %p (0x%x)",
368 toep->tid, tcpstates[tp->t_state], toep, toep->flags, inp,
374 tp->t_flags &= ~TF_TOE;
375 toep->flags &= ~TPF_ATTACHED;
377 if (!(toep->flags & TPF_CPL_PENDING))
378 release_offload_resources(toep);
382 * setsockopt handler.
385 t4_ctloutput(struct toedev *tod, struct tcpcb *tp, int dir, int name)
387 struct adapter *sc = tod->tod_softc;
388 struct toepcb *toep = tp->t_toe;
393 CTR4(KTR_CXGBE, "%s: tp %p, dir %u, name %u", __func__, tp, dir, name);
397 t4_set_tcb_field(sc, toep->ctrlq, toep->tid, W_TCB_T_FLAGS,
398 V_TF_NAGLE(1), V_TF_NAGLE(tp->t_flags & TF_NODELAY ? 0 : 1),
399 0, 0, toep->ofld_rxq->iq.abs_id);
407 * The TOE driver will not receive any more CPLs for the tid associated with the
408 * toepcb; release the hold on the inpcb.
411 final_cpl_received(struct toepcb *toep)
413 struct inpcb *inp = toep->inp;
415 KASSERT(inp != NULL, ("%s: inp is NULL", __func__));
416 INP_WLOCK_ASSERT(inp);
417 KASSERT(toep->flags & TPF_CPL_PENDING,
418 ("%s: CPL not pending already?", __func__));
420 CTR6(KTR_CXGBE, "%s: tid %d, toep %p (0x%x), inp %p (0x%x)",
421 __func__, toep->tid, toep, toep->flags, inp, inp->inp_flags);
424 toep->flags &= ~TPF_CPL_PENDING;
425 mbufq_drain(&toep->ulp_pdu_reclaimq);
427 if (!(toep->flags & TPF_ATTACHED))
428 release_offload_resources(toep);
430 if (!in_pcbrele_wlocked(inp))
435 insert_tid(struct adapter *sc, int tid, void *ctx)
437 struct tid_info *t = &sc->tids;
439 t->tid_tab[tid] = ctx;
440 atomic_add_int(&t->tids_in_use, 1);
444 lookup_tid(struct adapter *sc, int tid)
446 struct tid_info *t = &sc->tids;
448 return (t->tid_tab[tid]);
452 update_tid(struct adapter *sc, int tid, void *ctx)
454 struct tid_info *t = &sc->tids;
456 t->tid_tab[tid] = ctx;
460 remove_tid(struct adapter *sc, int tid)
462 struct tid_info *t = &sc->tids;
464 t->tid_tab[tid] = NULL;
465 atomic_subtract_int(&t->tids_in_use, 1);
469 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
472 struct cpl_tid_release *req;
474 wr = alloc_wrqe(sizeof(*req), ctrlq);
476 queue_tid_release(sc, tid); /* defer */
481 INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
487 queue_tid_release(struct adapter *sc, int tid)
490 CXGBE_UNIMPLEMENTED("deferred tid release");
494 * What mtu_idx to use, given a 4-tuple and/or an MSS cap
497 find_best_mtu_idx(struct adapter *sc, struct in_conninfo *inc, int pmss)
499 unsigned short *mtus = &sc->params.mtus[0];
502 KASSERT(inc != NULL || pmss > 0,
503 ("%s: at least one of inc/pmss must be specified", __func__));
505 mss = inc ? tcp_mssopt(inc) : pmss;
506 if (pmss > 0 && mss > pmss)
509 if (inc->inc_flags & INC_ISIPV6)
510 n = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
512 n = sizeof(struct ip) + sizeof(struct tcphdr);
514 for (i = 0; i < NMTUS - 1 && mtus[i + 1] <= mss + n; i++)
521 * Determine the receive window size for a socket.
524 select_rcv_wnd(struct socket *so)
528 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
530 wnd = sbspace(&so->so_rcv);
531 if (wnd < MIN_RCV_WND)
534 return min(wnd, MAX_RCV_WND);
538 select_rcv_wscale(void)
541 unsigned long space = sb_max;
543 if (space > MAX_RCV_WND)
546 while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < space)
552 extern int always_keepalive;
553 #define VIID_SMACIDX(v) (((unsigned int)(v) & 0x7f) << 1)
556 * socket so could be a listening socket too.
559 calc_opt0(struct socket *so, struct vi_info *vi, struct l2t_entry *e,
560 int mtu_idx, int rscale, int rx_credits, int ulp_mode)
564 KASSERT(rx_credits <= M_RCV_BUFSIZ,
565 ("%s: rcv_bufsiz too high", __func__));
567 opt0 = F_TCAM_BYPASS | V_WND_SCALE(rscale) | V_MSS_IDX(mtu_idx) |
568 V_ULP_MODE(ulp_mode) | V_RCV_BUFSIZ(rx_credits);
571 struct inpcb *inp = sotoinpcb(so);
572 struct tcpcb *tp = intotcpcb(inp);
573 int keepalive = always_keepalive ||
574 so_options_get(so) & SO_KEEPALIVE;
576 opt0 |= V_NAGLE((tp->t_flags & TF_NODELAY) == 0);
577 opt0 |= V_KEEP_ALIVE(keepalive != 0);
581 opt0 |= V_L2T_IDX(e->idx);
584 opt0 |= V_SMAC_SEL(VIID_SMACIDX(vi->viid));
585 opt0 |= V_TX_CHAN(vi->pi->tx_chan);
588 return htobe64(opt0);
592 select_ntuple(struct vi_info *vi, struct l2t_entry *e)
594 struct adapter *sc = vi->pi->adapter;
595 struct tp_params *tp = &sc->params.tp;
596 uint16_t viid = vi->viid;
600 * Initialize each of the fields which we care about which are present
601 * in the Compressed Filter Tuple.
603 if (tp->vlan_shift >= 0 && e->vlan != CPL_L2T_VLAN_NONE)
604 ntuple |= (uint64_t)(F_FT_VLAN_VLD | e->vlan) << tp->vlan_shift;
606 if (tp->port_shift >= 0)
607 ntuple |= (uint64_t)e->lport << tp->port_shift;
609 if (tp->protocol_shift >= 0)
610 ntuple |= (uint64_t)IPPROTO_TCP << tp->protocol_shift;
612 if (tp->vnic_shift >= 0) {
613 uint32_t vf = G_FW_VIID_VIN(viid);
614 uint32_t pf = G_FW_VIID_PFN(viid);
615 uint32_t vld = G_FW_VIID_VIVLD(viid);
617 ntuple |= (uint64_t)(V_FT_VNID_ID_VF(vf) | V_FT_VNID_ID_PF(pf) |
618 V_FT_VNID_ID_VLD(vld)) << tp->vnic_shift;
622 return (htobe32((uint32_t)ntuple));
624 return (htobe64(V_FILTER_TUPLE(ntuple)));
628 set_tcpddp_ulp_mode(struct toepcb *toep)
631 toep->ulp_mode = ULP_MODE_TCPDDP;
632 toep->ddp_flags = DDP_OK;
633 toep->ddp_score = DDP_LOW_SCORE;
637 negative_advice(int status)
640 return (status == CPL_ERR_RTX_NEG_ADVICE ||
641 status == CPL_ERR_PERSIST_NEG_ADVICE ||
642 status == CPL_ERR_KEEPALV_NEG_ADVICE);
646 alloc_tid_tabs(struct tid_info *t)
651 size = t->ntids * sizeof(*t->tid_tab) +
652 t->natids * sizeof(*t->atid_tab) +
653 t->nstids * sizeof(*t->stid_tab);
655 t->tid_tab = malloc(size, M_CXGBE, M_ZERO | M_NOWAIT);
656 if (t->tid_tab == NULL)
659 mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
660 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
661 t->afree = t->atid_tab;
663 for (i = 1; i < t->natids; i++)
664 t->atid_tab[i - 1].next = &t->atid_tab[i];
665 t->atid_tab[t->natids - 1].next = NULL;
667 mtx_init(&t->stid_lock, "stid lock", NULL, MTX_DEF);
668 t->stid_tab = (struct listen_ctx **)&t->atid_tab[t->natids];
670 TAILQ_INIT(&t->stids);
671 t->nstids_free_head = t->nstids;
673 atomic_store_rel_int(&t->tids_in_use, 0);
679 free_tid_tabs(struct tid_info *t)
681 KASSERT(t->tids_in_use == 0,
682 ("%s: %d tids still in use.", __func__, t->tids_in_use));
683 KASSERT(t->atids_in_use == 0,
684 ("%s: %d atids still in use.", __func__, t->atids_in_use));
685 KASSERT(t->stids_in_use == 0,
686 ("%s: %d tids still in use.", __func__, t->stids_in_use));
688 free(t->tid_tab, M_CXGBE);
691 if (mtx_initialized(&t->atid_lock))
692 mtx_destroy(&t->atid_lock);
693 if (mtx_initialized(&t->stid_lock))
694 mtx_destroy(&t->stid_lock);
698 add_lip(struct adapter *sc, struct in6_addr *lip)
700 struct fw_clip_cmd c;
702 ASSERT_SYNCHRONIZED_OP(sc);
703 /* mtx_assert(&td->clip_table_lock, MA_OWNED); */
705 memset(&c, 0, sizeof(c));
706 c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST |
708 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_ALLOC | FW_LEN16(c));
709 c.ip_hi = *(uint64_t *)&lip->s6_addr[0];
710 c.ip_lo = *(uint64_t *)&lip->s6_addr[8];
712 return (-t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c));
716 delete_lip(struct adapter *sc, struct in6_addr *lip)
718 struct fw_clip_cmd c;
720 ASSERT_SYNCHRONIZED_OP(sc);
721 /* mtx_assert(&td->clip_table_lock, MA_OWNED); */
723 memset(&c, 0, sizeof(c));
724 c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST |
726 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_FREE | FW_LEN16(c));
727 c.ip_hi = *(uint64_t *)&lip->s6_addr[0];
728 c.ip_lo = *(uint64_t *)&lip->s6_addr[8];
730 return (-t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c));
733 static struct clip_entry *
734 search_lip(struct tom_data *td, struct in6_addr *lip)
736 struct clip_entry *ce;
738 mtx_assert(&td->clip_table_lock, MA_OWNED);
740 TAILQ_FOREACH(ce, &td->clip_table, link) {
741 if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip))
749 hold_lip(struct tom_data *td, struct in6_addr *lip)
751 struct clip_entry *ce;
753 mtx_lock(&td->clip_table_lock);
754 ce = search_lip(td, lip);
757 mtx_unlock(&td->clip_table_lock);
763 release_lip(struct tom_data *td, struct clip_entry *ce)
766 mtx_lock(&td->clip_table_lock);
767 KASSERT(search_lip(td, &ce->lip) == ce,
768 ("%s: CLIP entry %p p not in CLIP table.", __func__, ce));
769 KASSERT(ce->refcount > 0,
770 ("%s: CLIP entry %p has refcount 0", __func__, ce));
772 mtx_unlock(&td->clip_table_lock);
776 init_clip_table(struct adapter *sc, struct tom_data *td)
779 ASSERT_SYNCHRONIZED_OP(sc);
781 mtx_init(&td->clip_table_lock, "CLIP table lock", NULL, MTX_DEF);
782 TAILQ_INIT(&td->clip_table);
785 update_clip_table(sc, td);
789 update_clip(struct adapter *sc, void *arg __unused)
792 if (begin_synchronized_op(sc, NULL, HOLD_LOCK, "t4tomuc"))
795 if (uld_active(sc, ULD_TOM))
796 update_clip_table(sc, sc->tom_softc);
798 end_synchronized_op(sc, LOCK_HELD);
802 t4_clip_task(void *arg, int count)
805 t4_iterate(update_clip, NULL);
809 update_clip_table(struct adapter *sc, struct tom_data *td)
811 struct in6_ifaddr *ia;
812 struct in6_addr *lip, tlip;
813 struct clip_head stale;
814 struct clip_entry *ce, *ce_temp;
815 int rc, gen = atomic_load_acq_int(&in6_ifaddr_gen);
817 ASSERT_SYNCHRONIZED_OP(sc);
820 mtx_lock(&td->clip_table_lock);
822 if (gen == td->clip_gen)
826 TAILQ_CONCAT(&stale, &td->clip_table, link);
828 TAILQ_FOREACH(ia, &V_in6_ifaddrhead, ia_link) {
829 lip = &ia->ia_addr.sin6_addr;
831 KASSERT(!IN6_IS_ADDR_MULTICAST(lip),
832 ("%s: mcast address in in6_ifaddr list", __func__));
834 if (IN6_IS_ADDR_LOOPBACK(lip))
836 if (IN6_IS_SCOPE_EMBED(lip)) {
837 /* Remove the embedded scope */
843 * XXX: how to weed out the link local address for the loopback
844 * interface? It's fe80::1 usually (always?).
848 * If it's in the main list then we already know it's not stale.
850 TAILQ_FOREACH(ce, &td->clip_table, link) {
851 if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip))
856 * If it's in the stale list we should move it to the main list.
858 TAILQ_FOREACH(ce, &stale, link) {
859 if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip)) {
860 TAILQ_REMOVE(&stale, ce, link);
861 TAILQ_INSERT_TAIL(&td->clip_table, ce, link);
866 /* A new IP6 address; add it to the CLIP table */
867 ce = malloc(sizeof(*ce), M_CXGBE, M_NOWAIT);
868 memcpy(&ce->lip, lip, sizeof(ce->lip));
870 rc = add_lip(sc, lip);
872 TAILQ_INSERT_TAIL(&td->clip_table, ce, link);
874 char ip[INET6_ADDRSTRLEN];
876 inet_ntop(AF_INET6, &ce->lip, &ip[0], sizeof(ip));
877 log(LOG_ERR, "%s: could not add %s (%d)\n",
886 * Remove stale addresses (those no longer in V_in6_ifaddrhead) that are
887 * no longer referenced by the driver.
889 TAILQ_FOREACH_SAFE(ce, &stale, link, ce_temp) {
890 if (ce->refcount == 0) {
891 rc = delete_lip(sc, &ce->lip);
893 TAILQ_REMOVE(&stale, ce, link);
896 char ip[INET6_ADDRSTRLEN];
898 inet_ntop(AF_INET6, &ce->lip, &ip[0],
900 log(LOG_ERR, "%s: could not delete %s (%d)\n",
905 /* The ones that are still referenced need to stay in the CLIP table */
906 TAILQ_CONCAT(&td->clip_table, &stale, link);
910 mtx_unlock(&td->clip_table_lock);
911 IN6_IFADDR_RUNLOCK();
915 destroy_clip_table(struct adapter *sc, struct tom_data *td)
917 struct clip_entry *ce, *ce_temp;
919 if (mtx_initialized(&td->clip_table_lock)) {
920 mtx_lock(&td->clip_table_lock);
921 TAILQ_FOREACH_SAFE(ce, &td->clip_table, link, ce_temp) {
922 KASSERT(ce->refcount == 0,
923 ("%s: CLIP entry %p still in use (%d)", __func__,
925 TAILQ_REMOVE(&td->clip_table, ce, link);
926 delete_lip(sc, &ce->lip);
929 mtx_unlock(&td->clip_table_lock);
930 mtx_destroy(&td->clip_table_lock);
935 free_tom_data(struct adapter *sc, struct tom_data *td)
938 ASSERT_SYNCHRONIZED_OP(sc);
940 KASSERT(TAILQ_EMPTY(&td->toep_list),
941 ("%s: TOE PCB list is not empty.", __func__));
942 KASSERT(td->lctx_count == 0,
943 ("%s: lctx hash table is not empty.", __func__));
945 t4_free_ppod_region(&td->pr);
946 destroy_clip_table(sc, td);
948 if (td->listen_mask != 0)
949 hashdestroy(td->listen_hash, M_CXGBE, td->listen_mask);
951 if (mtx_initialized(&td->unsent_wr_lock))
952 mtx_destroy(&td->unsent_wr_lock);
953 if (mtx_initialized(&td->lctx_hash_lock))
954 mtx_destroy(&td->lctx_hash_lock);
955 if (mtx_initialized(&td->toep_list_lock))
956 mtx_destroy(&td->toep_list_lock);
958 free_tid_tabs(&sc->tids);
963 reclaim_wr_resources(void *arg, int count)
965 struct tom_data *td = arg;
966 STAILQ_HEAD(, wrqe) twr_list = STAILQ_HEAD_INITIALIZER(twr_list);
967 struct cpl_act_open_req *cpl;
972 mtx_lock(&td->unsent_wr_lock);
973 STAILQ_SWAP(&td->unsent_wr_list, &twr_list, wrqe);
974 mtx_unlock(&td->unsent_wr_lock);
976 while ((wr = STAILQ_FIRST(&twr_list)) != NULL) {
977 STAILQ_REMOVE_HEAD(&twr_list, link);
980 opcode = GET_OPCODE(cpl);
983 case CPL_ACT_OPEN_REQ:
984 case CPL_ACT_OPEN_REQ6:
985 atid = G_TID_TID(be32toh(OPCODE_TID(cpl)));
988 CTR2(KTR_CXGBE, "%s: atid %u ", __func__, atid);
989 act_open_failure_cleanup(sc, atid, EHOSTUNREACH);
993 log(LOG_ERR, "%s: leaked work request %p, wr_len %d, "
994 "opcode %x\n", __func__, wr, wr->wr_len, opcode);
995 /* WR not freed here; go look at it with a debugger. */
1001 * Ground control to Major TOM
1002 * Commencing countdown, engines on
1005 t4_tom_activate(struct adapter *sc)
1007 struct tom_data *td;
1010 struct sge_ofld_rxq *ofld_rxq;
1013 ASSERT_SYNCHRONIZED_OP(sc);
1015 /* per-adapter softc for TOM */
1016 td = malloc(sizeof(*td), M_CXGBE, M_ZERO | M_NOWAIT);
1020 /* List of TOE PCBs and associated lock */
1021 mtx_init(&td->toep_list_lock, "PCB list lock", NULL, MTX_DEF);
1022 TAILQ_INIT(&td->toep_list);
1024 /* Listen context */
1025 mtx_init(&td->lctx_hash_lock, "lctx hash lock", NULL, MTX_DEF);
1026 td->listen_hash = hashinit_flags(LISTEN_HASH_SIZE, M_CXGBE,
1027 &td->listen_mask, HASH_NOWAIT);
1029 /* List of WRs for which L2 resolution failed */
1030 mtx_init(&td->unsent_wr_lock, "Unsent WR list lock", NULL, MTX_DEF);
1031 STAILQ_INIT(&td->unsent_wr_list);
1032 TASK_INIT(&td->reclaim_wr_resources, 0, reclaim_wr_resources, td);
1035 rc = alloc_tid_tabs(&sc->tids);
1039 rc = t4_init_ppod_region(&td->pr, &sc->vres.ddp,
1040 t4_read_reg(sc, A_ULP_RX_TDDP_PSZ), "TDDP page pods");
1043 t4_set_reg_field(sc, A_ULP_RX_TDDP_TAGMASK,
1044 V_TDDPTAGMASK(M_TDDPTAGMASK), td->pr.pr_tag_mask);
1046 /* CLIP table for IPv6 offload */
1047 init_clip_table(sc, td);
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;
1071 for_each_ofld_rxq(vi, j, ofld_rxq) {
1072 ofld_rxq->iq.set_tcb_rpl = do_set_tcb_rpl;
1073 ofld_rxq->iq.l2t_write_rpl = do_l2t_write_rpl2;
1079 register_toedev(sc->tom_softc);
1083 free_tom_data(sc, td);
1088 t4_tom_deactivate(struct adapter *sc)
1091 struct tom_data *td = sc->tom_softc;
1093 ASSERT_SYNCHRONIZED_OP(sc);
1096 return (0); /* XXX. KASSERT? */
1098 if (sc->offload_map != 0)
1099 return (EBUSY); /* at least one port has IFCAP_TOE enabled */
1101 if (uld_active(sc, ULD_IWARP) || uld_active(sc, ULD_ISCSI))
1102 return (EBUSY); /* both iWARP and iSCSI rely on the TOE. */
1104 mtx_lock(&td->toep_list_lock);
1105 if (!TAILQ_EMPTY(&td->toep_list))
1107 mtx_unlock(&td->toep_list_lock);
1109 mtx_lock(&td->lctx_hash_lock);
1110 if (td->lctx_count > 0)
1112 mtx_unlock(&td->lctx_hash_lock);
1114 taskqueue_drain(taskqueue_thread, &td->reclaim_wr_resources);
1115 mtx_lock(&td->unsent_wr_lock);
1116 if (!STAILQ_EMPTY(&td->unsent_wr_list))
1118 mtx_unlock(&td->unsent_wr_lock);
1121 unregister_toedev(sc->tom_softc);
1122 free_tom_data(sc, td);
1123 sc->tom_softc = NULL;
1130 t4_tom_ifaddr_event(void *arg __unused, struct ifnet *ifp)
1133 atomic_add_rel_int(&in6_ifaddr_gen, 1);
1134 taskqueue_enqueue_timeout(taskqueue_thread, &clip_task, -hz / 4);
1138 t4_tom_mod_load(void)
1141 struct protosw *tcp_protosw, *tcp6_protosw;
1144 t4_init_connect_cpl_handlers();
1145 t4_init_listen_cpl_handlers();
1146 t4_init_cpl_io_handlers();
1148 rc = t4_ddp_mod_load();
1152 tcp_protosw = pffindproto(PF_INET, IPPROTO_TCP, SOCK_STREAM);
1153 if (tcp_protosw == NULL)
1154 return (ENOPROTOOPT);
1155 bcopy(tcp_protosw, &ddp_protosw, sizeof(ddp_protosw));
1156 bcopy(tcp_protosw->pr_usrreqs, &ddp_usrreqs, sizeof(ddp_usrreqs));
1157 ddp_usrreqs.pru_soreceive = t4_soreceive_ddp;
1158 ddp_protosw.pr_usrreqs = &ddp_usrreqs;
1160 tcp6_protosw = pffindproto(PF_INET6, IPPROTO_TCP, SOCK_STREAM);
1161 if (tcp6_protosw == NULL)
1162 return (ENOPROTOOPT);
1163 bcopy(tcp6_protosw, &ddp6_protosw, sizeof(ddp6_protosw));
1164 bcopy(tcp6_protosw->pr_usrreqs, &ddp6_usrreqs, sizeof(ddp6_usrreqs));
1165 ddp6_usrreqs.pru_soreceive = t4_soreceive_ddp;
1166 ddp6_protosw.pr_usrreqs = &ddp6_usrreqs;
1168 TIMEOUT_TASK_INIT(taskqueue_thread, &clip_task, 0, t4_clip_task, NULL);
1169 ifaddr_evhandler = EVENTHANDLER_REGISTER(ifaddr_event,
1170 t4_tom_ifaddr_event, NULL, EVENTHANDLER_PRI_ANY);
1172 rc = t4_register_uld(&tom_uld_info);
1174 t4_tom_mod_unload();
1180 tom_uninit(struct adapter *sc, void *arg __unused)
1182 if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tomun"))
1185 /* Try to free resources (works only if no port has IFCAP_TOE) */
1186 if (uld_active(sc, ULD_TOM))
1187 t4_deactivate_uld(sc, ULD_TOM);
1189 end_synchronized_op(sc, 0);
1193 t4_tom_mod_unload(void)
1195 t4_iterate(tom_uninit, NULL);
1197 if (t4_unregister_uld(&tom_uld_info) == EBUSY)
1200 if (ifaddr_evhandler) {
1201 EVENTHANDLER_DEREGISTER(ifaddr_event, ifaddr_evhandler);
1202 taskqueue_cancel_timeout(taskqueue_thread, &clip_task, NULL);
1205 t4_ddp_mod_unload();
1209 #endif /* TCP_OFFLOAD */
1212 t4_tom_modevent(module_t mod, int cmd, void *arg)
1219 rc = t4_tom_mod_load();
1223 rc = t4_tom_mod_unload();
1230 printf("t4_tom: compiled without TCP_OFFLOAD support.\n");
1236 static moduledata_t t4_tom_moddata= {
1242 MODULE_VERSION(t4_tom, 1);
1243 MODULE_DEPEND(t4_tom, toecore, 1, 1, 1);
1244 MODULE_DEPEND(t4_tom, t4nex, 1, 1, 1);
1245 DECLARE_MODULE(t4_tom, t4_tom_moddata, SI_SUB_EXEC, SI_ORDER_ANY);