xref: /freebsd/sys/dev/cxgbe/tom/t4_tom.c (revision ab2043b81eaba0d7d7769b4a58b2b6d17bc464a3)
1 /*-
2  * Copyright (c) 2012 Chelsio Communications, Inc.
3  * All rights reserved.
4  * Written by: Navdeep Parhar <np@FreeBSD.org>
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_inet.h"
32 
33 #include <sys/param.h>
34 #include <sys/types.h>
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
37 #include <sys/ktr.h>
38 #include <sys/module.h>
39 #include <sys/protosw.h>
40 #include <sys/domain.h>
41 #include <sys/socket.h>
42 #include <sys/socketvar.h>
43 #include <netinet/in.h>
44 #include <netinet/in_pcb.h>
45 #include <netinet/ip.h>
46 #include <netinet/tcp_var.h>
47 #define TCPSTATES
48 #include <netinet/tcp_fsm.h>
49 #include <netinet/toecore.h>
50 
51 #ifdef TCP_OFFLOAD
52 #include "common/common.h"
53 #include "common/t4_msg.h"
54 #include "common/t4_regs.h"
55 #include "tom/t4_tom_l2t.h"
56 #include "tom/t4_tom.h"
57 
58 static struct protosw ddp_protosw;
59 static struct pr_usrreqs ddp_usrreqs;
60 
61 /* Module ops */
62 static int t4_tom_mod_load(void);
63 static int t4_tom_mod_unload(void);
64 static int t4_tom_modevent(module_t, int, void *);
65 
66 /* ULD ops and helpers */
67 static int t4_tom_activate(struct adapter *);
68 static int t4_tom_deactivate(struct adapter *);
69 
70 static struct uld_info tom_uld_info = {
71 	.uld_id = ULD_TOM,
72 	.activate = t4_tom_activate,
73 	.deactivate = t4_tom_deactivate,
74 };
75 
76 static void queue_tid_release(struct adapter *, int);
77 static void release_offload_resources(struct toepcb *);
78 static int alloc_tid_tabs(struct tid_info *);
79 static void free_tid_tabs(struct tid_info *);
80 static void free_tom_data(struct adapter *, struct tom_data *);
81 
82 struct toepcb *
83 alloc_toepcb(struct port_info *pi, int txqid, int rxqid, int flags)
84 {
85 	struct adapter *sc = pi->adapter;
86 	struct toepcb *toep;
87 	int tx_credits, txsd_total, len;
88 
89 	/*
90 	 * The firmware counts tx work request credits in units of 16 bytes
91 	 * each.  Reserve room for an ABORT_REQ so the driver never has to worry
92 	 * about tx credits if it wants to abort a connection.
93 	 */
94 	tx_credits = sc->params.ofldq_wr_cred;
95 	tx_credits -= howmany(sizeof(struct cpl_abort_req), 16);
96 
97 	/*
98 	 * Shortest possible tx work request is a fw_ofld_tx_data_wr + 1 byte
99 	 * immediate payload, and firmware counts tx work request credits in
100 	 * units of 16 byte.  Calculate the maximum work requests possible.
101 	 */
102 	txsd_total = tx_credits /
103 	    howmany((sizeof(struct fw_ofld_tx_data_wr) + 1), 16);
104 
105 	if (txqid < 0)
106 		txqid = (arc4random() % pi->nofldtxq) + pi->first_ofld_txq;
107 	KASSERT(txqid >= pi->first_ofld_txq &&
108 	    txqid < pi->first_ofld_txq + pi->nofldtxq,
109 	    ("%s: txqid %d for port %p (first %d, n %d)", __func__, txqid, pi,
110 		pi->first_ofld_txq, pi->nofldtxq));
111 
112 	if (rxqid < 0)
113 		rxqid = (arc4random() % pi->nofldrxq) + pi->first_ofld_rxq;
114 	KASSERT(rxqid >= pi->first_ofld_rxq &&
115 	    rxqid < pi->first_ofld_rxq + pi->nofldrxq,
116 	    ("%s: rxqid %d for port %p (first %d, n %d)", __func__, rxqid, pi,
117 		pi->first_ofld_rxq, pi->nofldrxq));
118 
119 	len = offsetof(struct toepcb, txsd) +
120 	    txsd_total * sizeof(struct ofld_tx_sdesc);
121 
122 	toep = malloc(len, M_CXGBE, M_ZERO | flags);
123 	if (toep == NULL)
124 		return (NULL);
125 
126 	toep->td = sc->tom_softc;
127 	toep->port = pi;
128 	toep->tx_credits = tx_credits;
129 	toep->ofld_txq = &sc->sge.ofld_txq[txqid];
130 	toep->ofld_rxq = &sc->sge.ofld_rxq[rxqid];
131 	toep->ctrlq = &sc->sge.ctrlq[pi->port_id];
132 	toep->txsd_total = txsd_total;
133 	toep->txsd_avail = txsd_total;
134 	toep->txsd_pidx = 0;
135 	toep->txsd_cidx = 0;
136 
137 	return (toep);
138 }
139 
140 void
141 free_toepcb(struct toepcb *toep)
142 {
143 
144 	KASSERT(!(toep->flags & TPF_ATTACHED),
145 	    ("%s: attached to an inpcb", __func__));
146 	KASSERT(!(toep->flags & TPF_CPL_PENDING),
147 	    ("%s: CPL pending", __func__));
148 
149 	free(toep, M_CXGBE);
150 }
151 
152 /*
153  * Set up the socket for TCP offload.
154  */
155 void
156 offload_socket(struct socket *so, struct toepcb *toep)
157 {
158 	struct tom_data *td = toep->td;
159 	struct inpcb *inp = sotoinpcb(so);
160 	struct tcpcb *tp = intotcpcb(inp);
161 	struct sockbuf *sb;
162 
163 	INP_WLOCK_ASSERT(inp);
164 
165 	/* Update socket */
166 	sb = &so->so_snd;
167 	SOCKBUF_LOCK(sb);
168 	sb->sb_flags |= SB_NOCOALESCE;
169 	SOCKBUF_UNLOCK(sb);
170 	sb = &so->so_rcv;
171 	SOCKBUF_LOCK(sb);
172 	sb->sb_flags |= SB_NOCOALESCE;
173 	if (toep->ulp_mode == ULP_MODE_TCPDDP)
174 		so->so_proto = &ddp_protosw;
175 	SOCKBUF_UNLOCK(sb);
176 
177 	/* Update TCP PCB */
178 	tp->tod = &td->tod;
179 	tp->t_toe = toep;
180 	tp->t_flags |= TF_TOE;
181 
182 	/* Install an extra hold on inp */
183 	toep->inp = inp;
184 	toep->flags |= TPF_ATTACHED;
185 	in_pcbref(inp);
186 
187 	/* Add the TOE PCB to the active list */
188 	mtx_lock(&td->toep_list_lock);
189 	TAILQ_INSERT_HEAD(&td->toep_list, toep, link);
190 	mtx_unlock(&td->toep_list_lock);
191 }
192 
193 /* This is _not_ the normal way to "unoffload" a socket. */
194 void
195 undo_offload_socket(struct socket *so)
196 {
197 	struct inpcb *inp = sotoinpcb(so);
198 	struct tcpcb *tp = intotcpcb(inp);
199 	struct toepcb *toep = tp->t_toe;
200 	struct tom_data *td = toep->td;
201 	struct sockbuf *sb;
202 
203 	INP_WLOCK_ASSERT(inp);
204 
205 	sb = &so->so_snd;
206 	SOCKBUF_LOCK(sb);
207 	sb->sb_flags &= ~SB_NOCOALESCE;
208 	SOCKBUF_UNLOCK(sb);
209 	sb = &so->so_rcv;
210 	SOCKBUF_LOCK(sb);
211 	sb->sb_flags &= ~SB_NOCOALESCE;
212 	SOCKBUF_UNLOCK(sb);
213 
214 	tp->tod = NULL;
215 	tp->t_toe = NULL;
216 	tp->t_flags &= ~TF_TOE;
217 
218 	toep->inp = NULL;
219 	toep->flags &= ~TPF_ATTACHED;
220 	if (in_pcbrele_wlocked(inp))
221 		panic("%s: inp freed.", __func__);
222 
223 	mtx_lock(&td->toep_list_lock);
224 	TAILQ_REMOVE(&td->toep_list, toep, link);
225 	mtx_unlock(&td->toep_list_lock);
226 }
227 
228 static void
229 release_offload_resources(struct toepcb *toep)
230 {
231 	struct tom_data *td = toep->td;
232 	struct adapter *sc = td_adapter(td);
233 	int tid = toep->tid;
234 
235 	KASSERT(!(toep->flags & TPF_CPL_PENDING),
236 	    ("%s: %p has CPL pending.", __func__, toep));
237 	KASSERT(!(toep->flags & TPF_ATTACHED),
238 	    ("%s: %p is still attached.", __func__, toep));
239 
240 	CTR4(KTR_CXGBE, "%s: toep %p (tid %d, l2te %p)",
241 	    __func__, toep, tid, toep->l2te);
242 
243 	if (toep->ulp_mode == ULP_MODE_TCPDDP)
244 		release_ddp_resources(toep);
245 
246 	if (toep->l2te)
247 		t4_l2t_release(toep->l2te);
248 
249 	if (tid >= 0) {
250 		remove_tid(sc, tid);
251 		release_tid(sc, tid, toep->ctrlq);
252 	}
253 
254 	mtx_lock(&td->toep_list_lock);
255 	TAILQ_REMOVE(&td->toep_list, toep, link);
256 	mtx_unlock(&td->toep_list_lock);
257 
258 	free_toepcb(toep);
259 }
260 
261 /*
262  * The kernel is done with the TCP PCB and this is our opportunity to unhook the
263  * toepcb hanging off of it.  If the TOE driver is also done with the toepcb (no
264  * pending CPL) then it is time to release all resources tied to the toepcb.
265  *
266  * Also gets called when an offloaded active open fails and the TOM wants the
267  * kernel to take the TCP PCB back.
268  */
269 static void
270 t4_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp)
271 {
272 #if defined(KTR) || defined(INVARIANTS)
273 	struct inpcb *inp = tp->t_inpcb;
274 #endif
275 	struct toepcb *toep = tp->t_toe;
276 
277 	INP_WLOCK_ASSERT(inp);
278 
279 	KASSERT(toep != NULL, ("%s: toep is NULL", __func__));
280 	KASSERT(toep->flags & TPF_ATTACHED,
281 	    ("%s: not attached", __func__));
282 
283 #ifdef KTR
284 	if (tp->t_state == TCPS_SYN_SENT) {
285 		CTR6(KTR_CXGBE, "%s: atid %d, toep %p (0x%x), inp %p (0x%x)",
286 		    __func__, toep->tid, toep, toep->flags, inp,
287 		    inp->inp_flags);
288 	} else {
289 		CTR6(KTR_CXGBE,
290 		    "t4_pcb_detach: tid %d (%s), toep %p (0x%x), inp %p (0x%x)",
291 		    toep->tid, tcpstates[tp->t_state], toep, toep->flags, inp,
292 		    inp->inp_flags);
293 	}
294 #endif
295 
296 	tp->t_toe = NULL;
297 	tp->t_flags &= ~TF_TOE;
298 	toep->flags &= ~TPF_ATTACHED;
299 
300 	if (!(toep->flags & TPF_CPL_PENDING))
301 		release_offload_resources(toep);
302 }
303 
304 /*
305  * The TOE driver will not receive any more CPLs for the tid associated with the
306  * toepcb; release the hold on the inpcb.
307  */
308 void
309 final_cpl_received(struct toepcb *toep)
310 {
311 	struct inpcb *inp = toep->inp;
312 
313 	KASSERT(inp != NULL, ("%s: inp is NULL", __func__));
314 	INP_WLOCK_ASSERT(inp);
315 	KASSERT(toep->flags & TPF_CPL_PENDING,
316 	    ("%s: CPL not pending already?", __func__));
317 
318 	CTR6(KTR_CXGBE, "%s: tid %d, toep %p (0x%x), inp %p (0x%x)",
319 	    __func__, toep->tid, toep, toep->flags, inp, inp->inp_flags);
320 
321 	toep->inp = NULL;
322 	toep->flags &= ~TPF_CPL_PENDING;
323 
324 	if (!(toep->flags & TPF_ATTACHED))
325 		release_offload_resources(toep);
326 
327 	if (!in_pcbrele_wlocked(inp))
328 		INP_WUNLOCK(inp);
329 }
330 
331 void
332 insert_tid(struct adapter *sc, int tid, void *ctx)
333 {
334 	struct tid_info *t = &sc->tids;
335 
336 	t->tid_tab[tid] = ctx;
337 	atomic_add_int(&t->tids_in_use, 1);
338 }
339 
340 void *
341 lookup_tid(struct adapter *sc, int tid)
342 {
343 	struct tid_info *t = &sc->tids;
344 
345 	return (t->tid_tab[tid]);
346 }
347 
348 void
349 update_tid(struct adapter *sc, int tid, void *ctx)
350 {
351 	struct tid_info *t = &sc->tids;
352 
353 	t->tid_tab[tid] = ctx;
354 }
355 
356 void
357 remove_tid(struct adapter *sc, int tid)
358 {
359 	struct tid_info *t = &sc->tids;
360 
361 	t->tid_tab[tid] = NULL;
362 	atomic_subtract_int(&t->tids_in_use, 1);
363 }
364 
365 void
366 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
367 {
368 	struct wrqe *wr;
369 	struct cpl_tid_release *req;
370 
371 	wr = alloc_wrqe(sizeof(*req), ctrlq);
372 	if (wr == NULL) {
373 		queue_tid_release(sc, tid);	/* defer */
374 		return;
375 	}
376 	req = wrtod(wr);
377 
378 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
379 
380 	t4_wrq_tx(sc, wr);
381 }
382 
383 static void
384 queue_tid_release(struct adapter *sc, int tid)
385 {
386 
387 	CXGBE_UNIMPLEMENTED("deferred tid release");
388 }
389 
390 /*
391  * What mtu_idx to use, given a 4-tuple and/or an MSS cap
392  */
393 int
394 find_best_mtu_idx(struct adapter *sc, struct in_conninfo *inc, int pmss)
395 {
396 	unsigned short *mtus = &sc->params.mtus[0];
397 	int i = 0, mss;
398 
399 	KASSERT(inc != NULL || pmss > 0,
400 	    ("%s: at least one of inc/pmss must be specified", __func__));
401 
402 	mss = inc ? tcp_mssopt(inc) : pmss;
403 	if (pmss > 0 && mss > pmss)
404 		mss = pmss;
405 
406 	while (i < NMTUS - 1 && mtus[i + 1] <= mss + 40)
407 		++i;
408 
409 	return (i);
410 }
411 
412 /*
413  * Determine the receive window size for a socket.
414  */
415 u_long
416 select_rcv_wnd(struct socket *so)
417 {
418 	unsigned long wnd;
419 
420 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
421 
422 	wnd = sbspace(&so->so_rcv);
423 	if (wnd < MIN_RCV_WND)
424 		wnd = MIN_RCV_WND;
425 
426 	return min(wnd, MAX_RCV_WND);
427 }
428 
429 int
430 select_rcv_wscale(void)
431 {
432 	int wscale = 0;
433 	unsigned long space = sb_max;
434 
435 	if (space > MAX_RCV_WND)
436 		space = MAX_RCV_WND;
437 
438 	while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < space)
439 		wscale++;
440 
441 	return (wscale);
442 }
443 
444 extern int always_keepalive;
445 #define VIID_SMACIDX(v)	(((unsigned int)(v) & 0x7f) << 1)
446 
447 /*
448  * socket so could be a listening socket too.
449  */
450 uint64_t
451 calc_opt0(struct socket *so, struct port_info *pi, struct l2t_entry *e,
452     int mtu_idx, int rscale, int rx_credits, int ulp_mode)
453 {
454 	uint64_t opt0;
455 
456 	KASSERT(rx_credits <= M_RCV_BUFSIZ,
457 	    ("%s: rcv_bufsiz too high", __func__));
458 
459 	opt0 = F_TCAM_BYPASS | V_WND_SCALE(rscale) | V_MSS_IDX(mtu_idx) |
460 	    V_ULP_MODE(ulp_mode) | V_RCV_BUFSIZ(rx_credits);
461 
462 	if (so != NULL) {
463 		struct inpcb *inp = sotoinpcb(so);
464 		struct tcpcb *tp = intotcpcb(inp);
465 		int keepalive = always_keepalive ||
466 		    so_options_get(so) & SO_KEEPALIVE;
467 
468 		opt0 |= V_NAGLE((tp->t_flags & TF_NODELAY) == 0);
469 		opt0 |= V_KEEP_ALIVE(keepalive != 0);
470 	}
471 
472 	if (e != NULL)
473 		opt0 |= V_L2T_IDX(e->idx);
474 
475 	if (pi != NULL) {
476 		opt0 |= V_SMAC_SEL(VIID_SMACIDX(pi->viid));
477 		opt0 |= V_TX_CHAN(pi->tx_chan);
478 	}
479 
480 	return htobe64(opt0);
481 }
482 
483 #define FILTER_SEL_WIDTH_P_FC (3 + 1)
484 #define FILTER_SEL_WIDTH_VIN_P_FC (6 + 7 + FILTER_SEL_WIDTH_P_FC)
485 #define FILTER_SEL_WIDTH_TAG_P_FC (3 + FILTER_SEL_WIDTH_VIN_P_FC)
486 #define FILTER_SEL_WIDTH_VLD_TAG_P_FC (1 + FILTER_SEL_WIDTH_TAG_P_FC)
487 #define VLAN_NONE 0xfff
488 #define FILTER_SEL_VLAN_NONE 0xffff
489 
490 uint32_t
491 select_ntuple(struct port_info *pi, struct l2t_entry *e, uint32_t filter_mode)
492 {
493 	uint16_t viid = pi->viid;
494 	uint32_t ntuple = 0;
495 
496 	if (filter_mode == HW_TPL_FR_MT_PR_IV_P_FC) {
497                 if (e->vlan == VLAN_NONE)
498 			ntuple |= FILTER_SEL_VLAN_NONE << FILTER_SEL_WIDTH_P_FC;
499                 else {
500                         ntuple |= e->vlan << FILTER_SEL_WIDTH_P_FC;
501                         ntuple |= 1 << FILTER_SEL_WIDTH_VLD_TAG_P_FC;
502                 }
503                 ntuple |= e->lport << S_PORT;
504 		ntuple |= IPPROTO_TCP << FILTER_SEL_WIDTH_VLD_TAG_P_FC;
505 	} else if (filter_mode == HW_TPL_FR_MT_PR_OV_P_FC) {
506                 ntuple |= G_FW_VIID_VIN(viid) << FILTER_SEL_WIDTH_P_FC;
507                 ntuple |= G_FW_VIID_PFN(viid) << FILTER_SEL_WIDTH_VIN_P_FC;
508                 ntuple |= G_FW_VIID_VIVLD(viid) << FILTER_SEL_WIDTH_TAG_P_FC;
509                 ntuple |= e->lport << S_PORT;
510 		ntuple |= IPPROTO_TCP << FILTER_SEL_WIDTH_VLD_TAG_P_FC;
511         }
512 
513 	return (htobe32(ntuple));
514 }
515 
516 static int
517 alloc_tid_tabs(struct tid_info *t)
518 {
519 	size_t size;
520 	unsigned int i;
521 
522 	size = t->ntids * sizeof(*t->tid_tab) +
523 	    t->natids * sizeof(*t->atid_tab) +
524 	    t->nstids * sizeof(*t->stid_tab);
525 
526 	t->tid_tab = malloc(size, M_CXGBE, M_ZERO | M_NOWAIT);
527 	if (t->tid_tab == NULL)
528 		return (ENOMEM);
529 
530 	mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
531 	t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
532 	t->afree = t->atid_tab;
533 	t->atids_in_use = 0;
534 	for (i = 1; i < t->natids; i++)
535 		t->atid_tab[i - 1].next = &t->atid_tab[i];
536 	t->atid_tab[t->natids - 1].next = NULL;
537 
538 	mtx_init(&t->stid_lock, "stid lock", NULL, MTX_DEF);
539 	t->stid_tab = (union serv_entry *)&t->atid_tab[t->natids];
540 	t->sfree = t->stid_tab;
541 	t->stids_in_use = 0;
542 	for (i = 1; i < t->nstids; i++)
543 		t->stid_tab[i - 1].next = &t->stid_tab[i];
544 	t->stid_tab[t->nstids - 1].next = NULL;
545 
546 	atomic_store_rel_int(&t->tids_in_use, 0);
547 
548 	return (0);
549 }
550 
551 static void
552 free_tid_tabs(struct tid_info *t)
553 {
554 	KASSERT(t->tids_in_use == 0,
555 	    ("%s: %d tids still in use.", __func__, t->tids_in_use));
556 	KASSERT(t->atids_in_use == 0,
557 	    ("%s: %d atids still in use.", __func__, t->atids_in_use));
558 	KASSERT(t->stids_in_use == 0,
559 	    ("%s: %d tids still in use.", __func__, t->stids_in_use));
560 
561 	free(t->tid_tab, M_CXGBE);
562 	t->tid_tab = NULL;
563 
564 	if (mtx_initialized(&t->atid_lock))
565 		mtx_destroy(&t->atid_lock);
566 	if (mtx_initialized(&t->stid_lock))
567 		mtx_destroy(&t->stid_lock);
568 }
569 
570 static void
571 free_tom_data(struct adapter *sc, struct tom_data *td)
572 {
573 	KASSERT(TAILQ_EMPTY(&td->toep_list),
574 	    ("%s: TOE PCB list is not empty.", __func__));
575 	KASSERT(td->lctx_count == 0,
576 	    ("%s: lctx hash table is not empty.", __func__));
577 
578 	t4_uninit_l2t_cpl_handlers(sc);
579 	t4_uninit_cpl_io_handlers(sc);
580 	t4_uninit_ddp(sc, td);
581 
582 	if (td->listen_mask != 0)
583 		hashdestroy(td->listen_hash, M_CXGBE, td->listen_mask);
584 
585 	if (mtx_initialized(&td->lctx_hash_lock))
586 		mtx_destroy(&td->lctx_hash_lock);
587 	if (mtx_initialized(&td->toep_list_lock))
588 		mtx_destroy(&td->toep_list_lock);
589 
590 	free_tid_tabs(&sc->tids);
591 	free(td, M_CXGBE);
592 }
593 
594 /*
595  * Ground control to Major TOM
596  * Commencing countdown, engines on
597  */
598 static int
599 t4_tom_activate(struct adapter *sc)
600 {
601 	struct tom_data *td;
602 	struct toedev *tod;
603 	int i, rc;
604 
605 	ADAPTER_LOCK_ASSERT_OWNED(sc);	/* for sc->flags */
606 
607 	/* per-adapter softc for TOM */
608 	td = malloc(sizeof(*td), M_CXGBE, M_ZERO | M_NOWAIT);
609 	if (td == NULL)
610 		return (ENOMEM);
611 
612 	/* List of TOE PCBs and associated lock */
613 	mtx_init(&td->toep_list_lock, "PCB list lock", NULL, MTX_DEF);
614 	TAILQ_INIT(&td->toep_list);
615 
616 	/* Listen context */
617 	mtx_init(&td->lctx_hash_lock, "lctx hash lock", NULL, MTX_DEF);
618 	td->listen_hash = hashinit_flags(LISTEN_HASH_SIZE, M_CXGBE,
619 	    &td->listen_mask, HASH_NOWAIT);
620 
621 	/* TID tables */
622 	rc = alloc_tid_tabs(&sc->tids);
623 	if (rc != 0)
624 		goto done;
625 
626 	t4_init_ddp(sc, td);
627 
628 	/* CPL handlers */
629 	t4_init_connect_cpl_handlers(sc);
630 	t4_init_l2t_cpl_handlers(sc);
631 	t4_init_listen_cpl_handlers(sc);
632 	t4_init_cpl_io_handlers(sc);
633 
634 	/* toedev ops */
635 	tod = &td->tod;
636 	init_toedev(tod);
637 	tod->tod_softc = sc;
638 	tod->tod_connect = t4_connect;
639 	tod->tod_listen_start = t4_listen_start;
640 	tod->tod_listen_stop = t4_listen_stop;
641 	tod->tod_rcvd = t4_rcvd;
642 	tod->tod_output = t4_tod_output;
643 	tod->tod_send_rst = t4_send_rst;
644 	tod->tod_send_fin = t4_send_fin;
645 	tod->tod_pcb_detach = t4_pcb_detach;
646 	tod->tod_l2_update = t4_l2_update;
647 	tod->tod_syncache_added = t4_syncache_added;
648 	tod->tod_syncache_removed = t4_syncache_removed;
649 	tod->tod_syncache_respond = t4_syncache_respond;
650 	tod->tod_offload_socket = t4_offload_socket;
651 
652 	for_each_port(sc, i)
653 		TOEDEV(sc->port[i]->ifp) = &td->tod;
654 
655 	sc->tom_softc = td;
656 	sc->flags |= TOM_INIT_DONE;
657 	register_toedev(sc->tom_softc);
658 
659 done:
660 	if (rc != 0)
661 		free_tom_data(sc, td);
662 	return (rc);
663 }
664 
665 static int
666 t4_tom_deactivate(struct adapter *sc)
667 {
668 	int rc = 0;
669 	struct tom_data *td = sc->tom_softc;
670 
671 	ADAPTER_LOCK_ASSERT_OWNED(sc);	/* for sc->flags */
672 
673 	if (td == NULL)
674 		return (0);	/* XXX. KASSERT? */
675 
676 	if (sc->offload_map != 0)
677 		return (EBUSY);	/* at least one port has IFCAP_TOE enabled */
678 
679 	mtx_lock(&td->toep_list_lock);
680 	if (!TAILQ_EMPTY(&td->toep_list))
681 		rc = EBUSY;
682 	mtx_unlock(&td->toep_list_lock);
683 
684 	mtx_lock(&td->lctx_hash_lock);
685 	if (td->lctx_count > 0)
686 		rc = EBUSY;
687 	mtx_unlock(&td->lctx_hash_lock);
688 
689 	if (rc == 0) {
690 		unregister_toedev(sc->tom_softc);
691 		free_tom_data(sc, td);
692 		sc->tom_softc = NULL;
693 		sc->flags &= ~TOM_INIT_DONE;
694 	}
695 
696 	return (rc);
697 }
698 
699 static int
700 t4_tom_mod_load(void)
701 {
702 	int rc;
703 	struct protosw *tcp_protosw;
704 
705 	tcp_protosw = pffindproto(PF_INET, IPPROTO_TCP, SOCK_STREAM);
706 	if (tcp_protosw == NULL)
707 		return (ENOPROTOOPT);
708 
709 	bcopy(tcp_protosw, &ddp_protosw, sizeof(ddp_protosw));
710 	bcopy(tcp_protosw->pr_usrreqs, &ddp_usrreqs, sizeof(ddp_usrreqs));
711 	ddp_usrreqs.pru_soreceive = t4_soreceive_ddp;
712 	ddp_protosw.pr_usrreqs = &ddp_usrreqs;
713 
714 	rc = t4_register_uld(&tom_uld_info);
715 	if (rc != 0)
716 		t4_tom_mod_unload();
717 
718 	return (rc);
719 }
720 
721 static void
722 tom_uninit(struct adapter *sc, void *arg __unused)
723 {
724 	/* Try to free resources (works only if no port has IFCAP_TOE) */
725 	ADAPTER_LOCK(sc);
726 	if (sc->flags & TOM_INIT_DONE)
727 		t4_deactivate_uld(sc, ULD_TOM);
728 	ADAPTER_UNLOCK(sc);
729 }
730 
731 static int
732 t4_tom_mod_unload(void)
733 {
734 	t4_iterate(tom_uninit, NULL);
735 
736 	if (t4_unregister_uld(&tom_uld_info) == EBUSY)
737 		return (EBUSY);
738 
739 	return (0);
740 }
741 #endif	/* TCP_OFFLOAD */
742 
743 static int
744 t4_tom_modevent(module_t mod, int cmd, void *arg)
745 {
746 	int rc = 0;
747 
748 #ifdef TCP_OFFLOAD
749 	switch (cmd) {
750 	case MOD_LOAD:
751 		rc = t4_tom_mod_load();
752 		break;
753 
754 	case MOD_UNLOAD:
755 		rc = t4_tom_mod_unload();
756 		break;
757 
758 	default:
759 		rc = EINVAL;
760 	}
761 #else
762 	printf("t4_tom: compiled without TCP_OFFLOAD support.\n");
763 	rc = EOPNOTSUPP;
764 #endif
765 	return (rc);
766 }
767 
768 static moduledata_t t4_tom_moddata= {
769 	"t4_tom",
770 	t4_tom_modevent,
771 	0
772 };
773 
774 MODULE_VERSION(t4_tom, 1);
775 MODULE_DEPEND(t4_tom, toecore, 1, 1, 1);
776 MODULE_DEPEND(t4_tom, t4nex, 1, 1, 1);
777 DECLARE_MODULE(t4_tom, t4_tom_moddata, SI_SUB_EXEC, SI_ORDER_ANY);
778