t4_tom.c revision 331645
1237263Snp/*- 2330897Seadler * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3330897Seadler * 4237263Snp * Copyright (c) 2012 Chelsio Communications, Inc. 5237263Snp * All rights reserved. 6237263Snp * Written by: Navdeep Parhar <np@FreeBSD.org> 7237263Snp * 8237263Snp * Redistribution and use in source and binary forms, with or without 9237263Snp * modification, are permitted provided that the following conditions 10237263Snp * are met: 11237263Snp * 1. Redistributions of source code must retain the above copyright 12237263Snp * notice, this list of conditions and the following disclaimer. 13237263Snp * 2. Redistributions in binary form must reproduce the above copyright 14237263Snp * notice, this list of conditions and the following disclaimer in the 15237263Snp * documentation and/or other materials provided with the distribution. 16237263Snp * 17237263Snp * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18237263Snp * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19237263Snp * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20237263Snp * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21237263Snp * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22237263Snp * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23237263Snp * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24237263Snp * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25237263Snp * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26237263Snp * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27237263Snp * SUCH DAMAGE. 28237263Snp */ 29237263Snp 30237263Snp#include <sys/cdefs.h> 31237263Snp__FBSDID("$FreeBSD: stable/11/sys/dev/cxgbe/tom/t4_tom.c 331645 2018-03-27 20:14:22Z jhb $"); 32237263Snp 33237263Snp#include "opt_inet.h" 34245441Snp#include "opt_inet6.h" 35237263Snp 36237263Snp#include <sys/param.h> 37237263Snp#include <sys/types.h> 38237263Snp#include <sys/systm.h> 39237263Snp#include <sys/kernel.h> 40237263Snp#include <sys/ktr.h> 41286001Sae#include <sys/lock.h> 42292736Snp#include <sys/limits.h> 43237263Snp#include <sys/module.h> 44237263Snp#include <sys/protosw.h> 45237263Snp#include <sys/domain.h> 46299210Sjhb#include <sys/refcount.h> 47286001Sae#include <sys/rmlock.h> 48237263Snp#include <sys/socket.h> 49237263Snp#include <sys/socketvar.h> 50249627Snp#include <sys/taskqueue.h> 51245448Snp#include <net/if.h> 52257241Sglebius#include <net/if_var.h> 53237263Snp#include <netinet/in.h> 54237263Snp#include <netinet/in_pcb.h> 55245448Snp#include <netinet/in_var.h> 56237263Snp#include <netinet/ip.h> 57245441Snp#include <netinet/ip6.h> 58245448Snp#include <netinet6/scope6_var.h> 59237263Snp#define TCPSTATES 60237263Snp#include <netinet/tcp_fsm.h> 61330303Sjhb#include <netinet/tcp_timer.h> 62294889Sglebius#include <netinet/tcp_var.h> 63237263Snp#include <netinet/toecore.h> 64237263Snp 65237263Snp#ifdef TCP_OFFLOAD 66237263Snp#include "common/common.h" 67237263Snp#include "common/t4_msg.h" 68237263Snp#include "common/t4_regs.h" 69252705Snp#include "common/t4_regs_values.h" 70252716Snp#include "common/t4_tcb.h" 71237263Snp#include "tom/t4_tom_l2t.h" 72237263Snp#include "tom/t4_tom.h" 73237263Snp 74306661Sjhbstatic struct protosw toe_protosw; 75306661Sjhbstatic struct pr_usrreqs toe_usrreqs; 76239344Snp 77306661Sjhbstatic struct protosw toe6_protosw; 78306661Sjhbstatic struct pr_usrreqs toe6_usrreqs; 79245441Snp 80237263Snp/* Module ops */ 81237263Snpstatic int t4_tom_mod_load(void); 82237263Snpstatic int t4_tom_mod_unload(void); 83237263Snpstatic int t4_tom_modevent(module_t, int, void *); 84237263Snp 85237263Snp/* ULD ops and helpers */ 86237263Snpstatic int t4_tom_activate(struct adapter *); 87237263Snpstatic int t4_tom_deactivate(struct adapter *); 88237263Snp 89237263Snpstatic struct uld_info tom_uld_info = { 90237263Snp .uld_id = ULD_TOM, 91237263Snp .activate = t4_tom_activate, 92237263Snp .deactivate = t4_tom_deactivate, 93237263Snp}; 94237263Snp 95237263Snpstatic void queue_tid_release(struct adapter *, int); 96237263Snpstatic void release_offload_resources(struct toepcb *); 97237263Snpstatic int alloc_tid_tabs(struct tid_info *); 98237263Snpstatic void free_tid_tabs(struct tid_info *); 99245448Snpstatic int add_lip(struct adapter *, struct in6_addr *); 100245448Snpstatic int delete_lip(struct adapter *, struct in6_addr *); 101245448Snpstatic struct clip_entry *search_lip(struct tom_data *, struct in6_addr *); 102245448Snpstatic void init_clip_table(struct adapter *, struct tom_data *); 103249627Snpstatic void update_clip(struct adapter *, void *); 104249627Snpstatic void t4_clip_task(void *, int); 105249627Snpstatic void update_clip_table(struct adapter *, struct tom_data *); 106245448Snpstatic void destroy_clip_table(struct adapter *, struct tom_data *); 107237263Snpstatic void free_tom_data(struct adapter *, struct tom_data *); 108272719Snpstatic void reclaim_wr_resources(void *, int); 109237263Snp 110249627Snpstatic int in6_ifaddr_gen; 111249627Snpstatic eventhandler_tag ifaddr_evhandler; 112249627Snpstatic struct timeout_task clip_task; 113249627Snp 114237263Snpstruct toepcb * 115291665Sjhballoc_toepcb(struct vi_info *vi, int txqid, int rxqid, int flags) 116237263Snp{ 117291665Sjhb struct port_info *pi = vi->pi; 118237263Snp struct adapter *sc = pi->adapter; 119237263Snp struct toepcb *toep; 120237263Snp int tx_credits, txsd_total, len; 121237263Snp 122237263Snp /* 123237263Snp * The firmware counts tx work request credits in units of 16 bytes 124237263Snp * each. Reserve room for an ABORT_REQ so the driver never has to worry 125237263Snp * about tx credits if it wants to abort a connection. 126237263Snp */ 127237263Snp tx_credits = sc->params.ofldq_wr_cred; 128237263Snp tx_credits -= howmany(sizeof(struct cpl_abort_req), 16); 129237263Snp 130237263Snp /* 131237263Snp * Shortest possible tx work request is a fw_ofld_tx_data_wr + 1 byte 132237263Snp * immediate payload, and firmware counts tx work request credits in 133237263Snp * units of 16 byte. Calculate the maximum work requests possible. 134237263Snp */ 135237263Snp txsd_total = tx_credits / 136298482Spfg howmany(sizeof(struct fw_ofld_tx_data_wr) + 1, 16); 137237263Snp 138237263Snp if (txqid < 0) 139291665Sjhb txqid = (arc4random() % vi->nofldtxq) + vi->first_ofld_txq; 140291665Sjhb KASSERT(txqid >= vi->first_ofld_txq && 141291665Sjhb txqid < vi->first_ofld_txq + vi->nofldtxq, 142291665Sjhb ("%s: txqid %d for vi %p (first %d, n %d)", __func__, txqid, vi, 143291665Sjhb vi->first_ofld_txq, vi->nofldtxq)); 144237263Snp 145237263Snp if (rxqid < 0) 146291665Sjhb rxqid = (arc4random() % vi->nofldrxq) + vi->first_ofld_rxq; 147291665Sjhb KASSERT(rxqid >= vi->first_ofld_rxq && 148291665Sjhb rxqid < vi->first_ofld_rxq + vi->nofldrxq, 149291665Sjhb ("%s: rxqid %d for vi %p (first %d, n %d)", __func__, rxqid, vi, 150291665Sjhb vi->first_ofld_rxq, vi->nofldrxq)); 151237263Snp 152237263Snp len = offsetof(struct toepcb, txsd) + 153237263Snp txsd_total * sizeof(struct ofld_tx_sdesc); 154237263Snp 155237263Snp toep = malloc(len, M_CXGBE, M_ZERO | flags); 156237263Snp if (toep == NULL) 157237263Snp return (NULL); 158237263Snp 159299210Sjhb refcount_init(&toep->refcount, 1); 160237263Snp toep->td = sc->tom_softc; 161291665Sjhb toep->vi = vi; 162255411Snp toep->tx_total = tx_credits; 163237263Snp toep->tx_credits = tx_credits; 164237263Snp toep->ofld_txq = &sc->sge.ofld_txq[txqid]; 165237263Snp toep->ofld_rxq = &sc->sge.ofld_rxq[rxqid]; 166237263Snp toep->ctrlq = &sc->sge.ctrlq[pi->port_id]; 167292736Snp mbufq_init(&toep->ulp_pduq, INT_MAX); 168292736Snp mbufq_init(&toep->ulp_pdu_reclaimq, INT_MAX); 169237263Snp toep->txsd_total = txsd_total; 170237263Snp toep->txsd_avail = txsd_total; 171237263Snp toep->txsd_pidx = 0; 172237263Snp toep->txsd_cidx = 0; 173306661Sjhb aiotx_init_toep(toep); 174237263Snp 175237263Snp return (toep); 176237263Snp} 177237263Snp 178299210Sjhbstruct toepcb * 179299210Sjhbhold_toepcb(struct toepcb *toep) 180299210Sjhb{ 181299210Sjhb 182299210Sjhb refcount_acquire(&toep->refcount); 183299210Sjhb return (toep); 184299210Sjhb} 185299210Sjhb 186237263Snpvoid 187237263Snpfree_toepcb(struct toepcb *toep) 188237263Snp{ 189237263Snp 190299210Sjhb if (refcount_release(&toep->refcount) == 0) 191299210Sjhb return; 192299210Sjhb 193239514Snp KASSERT(!(toep->flags & TPF_ATTACHED), 194237263Snp ("%s: attached to an inpcb", __func__)); 195239514Snp KASSERT(!(toep->flags & TPF_CPL_PENDING), 196237263Snp ("%s: CPL pending", __func__)); 197237263Snp 198331645Sjhb if (toep->ulp_mode == ULP_MODE_TCPDDP) 199331645Sjhb ddp_uninit_toep(toep); 200237263Snp free(toep, M_CXGBE); 201237263Snp} 202237263Snp 203237263Snp/* 204237263Snp * Set up the socket for TCP offload. 205237263Snp */ 206237263Snpvoid 207237263Snpoffload_socket(struct socket *so, struct toepcb *toep) 208237263Snp{ 209237263Snp struct tom_data *td = toep->td; 210237263Snp struct inpcb *inp = sotoinpcb(so); 211237263Snp struct tcpcb *tp = intotcpcb(inp); 212237263Snp struct sockbuf *sb; 213237263Snp 214237263Snp INP_WLOCK_ASSERT(inp); 215237263Snp 216237263Snp /* Update socket */ 217237263Snp sb = &so->so_snd; 218237263Snp SOCKBUF_LOCK(sb); 219237263Snp sb->sb_flags |= SB_NOCOALESCE; 220237263Snp SOCKBUF_UNLOCK(sb); 221237263Snp sb = &so->so_rcv; 222237263Snp SOCKBUF_LOCK(sb); 223237263Snp sb->sb_flags |= SB_NOCOALESCE; 224306661Sjhb if (inp->inp_vflag & INP_IPV6) 225306661Sjhb so->so_proto = &toe6_protosw; 226306661Sjhb else 227306661Sjhb so->so_proto = &toe_protosw; 228237263Snp SOCKBUF_UNLOCK(sb); 229237263Snp 230237263Snp /* Update TCP PCB */ 231237263Snp tp->tod = &td->tod; 232237263Snp tp->t_toe = toep; 233237263Snp tp->t_flags |= TF_TOE; 234237263Snp 235237263Snp /* Install an extra hold on inp */ 236237263Snp toep->inp = inp; 237239514Snp toep->flags |= TPF_ATTACHED; 238237263Snp in_pcbref(inp); 239237263Snp 240237263Snp /* Add the TOE PCB to the active list */ 241237263Snp mtx_lock(&td->toep_list_lock); 242237263Snp TAILQ_INSERT_HEAD(&td->toep_list, toep, link); 243237263Snp mtx_unlock(&td->toep_list_lock); 244237263Snp} 245237263Snp 246237263Snp/* This is _not_ the normal way to "unoffload" a socket. */ 247237263Snpvoid 248237263Snpundo_offload_socket(struct socket *so) 249237263Snp{ 250237263Snp struct inpcb *inp = sotoinpcb(so); 251237263Snp struct tcpcb *tp = intotcpcb(inp); 252237263Snp struct toepcb *toep = tp->t_toe; 253237263Snp struct tom_data *td = toep->td; 254237263Snp struct sockbuf *sb; 255237263Snp 256237263Snp INP_WLOCK_ASSERT(inp); 257237263Snp 258237263Snp sb = &so->so_snd; 259237263Snp SOCKBUF_LOCK(sb); 260237263Snp sb->sb_flags &= ~SB_NOCOALESCE; 261237263Snp SOCKBUF_UNLOCK(sb); 262237263Snp sb = &so->so_rcv; 263237263Snp SOCKBUF_LOCK(sb); 264237263Snp sb->sb_flags &= ~SB_NOCOALESCE; 265237263Snp SOCKBUF_UNLOCK(sb); 266237263Snp 267237263Snp tp->tod = NULL; 268237263Snp tp->t_toe = NULL; 269237263Snp tp->t_flags &= ~TF_TOE; 270237263Snp 271237263Snp toep->inp = NULL; 272239514Snp toep->flags &= ~TPF_ATTACHED; 273237263Snp if (in_pcbrele_wlocked(inp)) 274237263Snp panic("%s: inp freed.", __func__); 275237263Snp 276237263Snp mtx_lock(&td->toep_list_lock); 277237263Snp TAILQ_REMOVE(&td->toep_list, toep, link); 278237263Snp mtx_unlock(&td->toep_list_lock); 279237263Snp} 280237263Snp 281237263Snpstatic void 282237263Snprelease_offload_resources(struct toepcb *toep) 283237263Snp{ 284237263Snp struct tom_data *td = toep->td; 285237263Snp struct adapter *sc = td_adapter(td); 286237263Snp int tid = toep->tid; 287237263Snp 288239514Snp KASSERT(!(toep->flags & TPF_CPL_PENDING), 289237263Snp ("%s: %p has CPL pending.", __func__, toep)); 290239514Snp KASSERT(!(toep->flags & TPF_ATTACHED), 291237263Snp ("%s: %p is still attached.", __func__, toep)); 292237263Snp 293245448Snp CTR5(KTR_CXGBE, "%s: toep %p (tid %d, l2te %p, ce %p)", 294245448Snp __func__, toep, tid, toep->l2te, toep->ce); 295237263Snp 296292736Snp /* 297292736Snp * These queues should have been emptied at approximately the same time 298292736Snp * that a normal connection's socket's so_snd would have been purged or 299292736Snp * drained. Do _not_ clean up here. 300292736Snp */ 301292736Snp MPASS(mbufq_len(&toep->ulp_pduq) == 0); 302292736Snp MPASS(mbufq_len(&toep->ulp_pdu_reclaimq) == 0); 303299210Sjhb#ifdef INVARIANTS 304331645Sjhb if (toep->ulp_mode == ULP_MODE_TCPDDP) 305331645Sjhb ddp_assert_empty(toep); 306299210Sjhb#endif 307292736Snp 308237263Snp if (toep->l2te) 309237263Snp t4_l2t_release(toep->l2te); 310237263Snp 311237263Snp if (tid >= 0) { 312312116Snp remove_tid(sc, tid, toep->ce ? 2 : 1); 313237263Snp release_tid(sc, tid, toep->ctrlq); 314237263Snp } 315237263Snp 316245448Snp if (toep->ce) 317245448Snp release_lip(td, toep->ce); 318245448Snp 319237263Snp mtx_lock(&td->toep_list_lock); 320237263Snp TAILQ_REMOVE(&td->toep_list, toep, link); 321237263Snp mtx_unlock(&td->toep_list_lock); 322237263Snp 323237263Snp free_toepcb(toep); 324237263Snp} 325237263Snp 326237263Snp/* 327237263Snp * The kernel is done with the TCP PCB and this is our opportunity to unhook the 328237263Snp * toepcb hanging off of it. If the TOE driver is also done with the toepcb (no 329237263Snp * pending CPL) then it is time to release all resources tied to the toepcb. 330237263Snp * 331237263Snp * Also gets called when an offloaded active open fails and the TOM wants the 332237263Snp * kernel to take the TCP PCB back. 333237263Snp */ 334237263Snpstatic void 335237263Snpt4_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp) 336237263Snp{ 337237263Snp#if defined(KTR) || defined(INVARIANTS) 338237263Snp struct inpcb *inp = tp->t_inpcb; 339237263Snp#endif 340237263Snp struct toepcb *toep = tp->t_toe; 341237263Snp 342237263Snp INP_WLOCK_ASSERT(inp); 343237263Snp 344237263Snp KASSERT(toep != NULL, ("%s: toep is NULL", __func__)); 345239514Snp KASSERT(toep->flags & TPF_ATTACHED, 346237263Snp ("%s: not attached", __func__)); 347237263Snp 348237263Snp#ifdef KTR 349237263Snp if (tp->t_state == TCPS_SYN_SENT) { 350237263Snp CTR6(KTR_CXGBE, "%s: atid %d, toep %p (0x%x), inp %p (0x%x)", 351237263Snp __func__, toep->tid, toep, toep->flags, inp, 352237263Snp inp->inp_flags); 353237263Snp } else { 354237263Snp CTR6(KTR_CXGBE, 355237263Snp "t4_pcb_detach: tid %d (%s), toep %p (0x%x), inp %p (0x%x)", 356237263Snp toep->tid, tcpstates[tp->t_state], toep, toep->flags, inp, 357237263Snp inp->inp_flags); 358237263Snp } 359237263Snp#endif 360237263Snp 361237263Snp tp->t_toe = NULL; 362237263Snp tp->t_flags &= ~TF_TOE; 363239514Snp toep->flags &= ~TPF_ATTACHED; 364237263Snp 365239514Snp if (!(toep->flags & TPF_CPL_PENDING)) 366237263Snp release_offload_resources(toep); 367237263Snp} 368237263Snp 369237263Snp/* 370252716Snp * setsockopt handler. 371252716Snp */ 372252716Snpstatic void 373252716Snpt4_ctloutput(struct toedev *tod, struct tcpcb *tp, int dir, int name) 374252716Snp{ 375252716Snp struct adapter *sc = tod->tod_softc; 376252716Snp struct toepcb *toep = tp->t_toe; 377252716Snp 378252716Snp if (dir == SOPT_GET) 379252716Snp return; 380252716Snp 381252716Snp CTR4(KTR_CXGBE, "%s: tp %p, dir %u, name %u", __func__, tp, dir, name); 382252716Snp 383252716Snp switch (name) { 384252716Snp case TCP_NODELAY: 385330307Snp if (tp->t_state != TCPS_ESTABLISHED) 386330307Snp break; 387302339Snp t4_set_tcb_field(sc, toep->ctrlq, toep->tid, W_TCB_T_FLAGS, 388302339Snp V_TF_NAGLE(1), V_TF_NAGLE(tp->t_flags & TF_NODELAY ? 0 : 1), 389302339Snp 0, 0, toep->ofld_rxq->iq.abs_id); 390252716Snp break; 391252716Snp default: 392252716Snp break; 393252716Snp } 394252716Snp} 395252716Snp 396252716Snp/* 397237263Snp * The TOE driver will not receive any more CPLs for the tid associated with the 398237263Snp * toepcb; release the hold on the inpcb. 399237263Snp */ 400237263Snpvoid 401237263Snpfinal_cpl_received(struct toepcb *toep) 402237263Snp{ 403237263Snp struct inpcb *inp = toep->inp; 404237263Snp 405237263Snp KASSERT(inp != NULL, ("%s: inp is NULL", __func__)); 406237263Snp INP_WLOCK_ASSERT(inp); 407239514Snp KASSERT(toep->flags & TPF_CPL_PENDING, 408237263Snp ("%s: CPL not pending already?", __func__)); 409237263Snp 410237263Snp CTR6(KTR_CXGBE, "%s: tid %d, toep %p (0x%x), inp %p (0x%x)", 411237263Snp __func__, toep->tid, toep, toep->flags, inp, inp->inp_flags); 412237263Snp 413299210Sjhb if (toep->ulp_mode == ULP_MODE_TCPDDP) 414299210Sjhb release_ddp_resources(toep); 415237263Snp toep->inp = NULL; 416239514Snp toep->flags &= ~TPF_CPL_PENDING; 417292736Snp mbufq_drain(&toep->ulp_pdu_reclaimq); 418237263Snp 419239514Snp if (!(toep->flags & TPF_ATTACHED)) 420237263Snp release_offload_resources(toep); 421237263Snp 422237263Snp if (!in_pcbrele_wlocked(inp)) 423237263Snp INP_WUNLOCK(inp); 424237263Snp} 425237263Snp 426237263Snpvoid 427312116Snpinsert_tid(struct adapter *sc, int tid, void *ctx, int ntids) 428237263Snp{ 429237263Snp struct tid_info *t = &sc->tids; 430237263Snp 431237263Snp t->tid_tab[tid] = ctx; 432312116Snp atomic_add_int(&t->tids_in_use, ntids); 433237263Snp} 434237263Snp 435237263Snpvoid * 436237263Snplookup_tid(struct adapter *sc, int tid) 437237263Snp{ 438237263Snp struct tid_info *t = &sc->tids; 439237263Snp 440237263Snp return (t->tid_tab[tid]); 441237263Snp} 442237263Snp 443237263Snpvoid 444237263Snpupdate_tid(struct adapter *sc, int tid, void *ctx) 445237263Snp{ 446237263Snp struct tid_info *t = &sc->tids; 447237263Snp 448237263Snp t->tid_tab[tid] = ctx; 449237263Snp} 450237263Snp 451237263Snpvoid 452312116Snpremove_tid(struct adapter *sc, int tid, int ntids) 453237263Snp{ 454237263Snp struct tid_info *t = &sc->tids; 455237263Snp 456237263Snp t->tid_tab[tid] = NULL; 457312116Snp atomic_subtract_int(&t->tids_in_use, ntids); 458237263Snp} 459237263Snp 460237263Snpvoid 461237263Snprelease_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq) 462237263Snp{ 463237263Snp struct wrqe *wr; 464237263Snp struct cpl_tid_release *req; 465237263Snp 466237263Snp wr = alloc_wrqe(sizeof(*req), ctrlq); 467237263Snp if (wr == NULL) { 468237263Snp queue_tid_release(sc, tid); /* defer */ 469237263Snp return; 470237263Snp } 471237263Snp req = wrtod(wr); 472237263Snp 473237263Snp INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid); 474237263Snp 475237263Snp t4_wrq_tx(sc, wr); 476237263Snp} 477237263Snp 478237263Snpstatic void 479237263Snpqueue_tid_release(struct adapter *sc, int tid) 480237263Snp{ 481237263Snp 482237263Snp CXGBE_UNIMPLEMENTED("deferred tid release"); 483237263Snp} 484237263Snp 485237263Snp/* 486237263Snp * What mtu_idx to use, given a 4-tuple and/or an MSS cap 487237263Snp */ 488237263Snpint 489237263Snpfind_best_mtu_idx(struct adapter *sc, struct in_conninfo *inc, int pmss) 490237263Snp{ 491237263Snp unsigned short *mtus = &sc->params.mtus[0]; 492245441Snp int i, mss, n; 493237263Snp 494237263Snp KASSERT(inc != NULL || pmss > 0, 495237263Snp ("%s: at least one of inc/pmss must be specified", __func__)); 496237263Snp 497237263Snp mss = inc ? tcp_mssopt(inc) : pmss; 498237263Snp if (pmss > 0 && mss > pmss) 499237263Snp mss = pmss; 500237263Snp 501245441Snp if (inc->inc_flags & INC_ISIPV6) 502245441Snp n = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 503245441Snp else 504245441Snp n = sizeof(struct ip) + sizeof(struct tcphdr); 505237263Snp 506245441Snp for (i = 0; i < NMTUS - 1 && mtus[i + 1] <= mss + n; i++) 507245441Snp continue; 508245441Snp 509237263Snp return (i); 510237263Snp} 511237263Snp 512237263Snp/* 513237263Snp * Determine the receive window size for a socket. 514237263Snp */ 515237263Snpu_long 516237263Snpselect_rcv_wnd(struct socket *so) 517237263Snp{ 518237263Snp unsigned long wnd; 519237263Snp 520237263Snp SOCKBUF_LOCK_ASSERT(&so->so_rcv); 521237263Snp 522237263Snp wnd = sbspace(&so->so_rcv); 523237263Snp if (wnd < MIN_RCV_WND) 524237263Snp wnd = MIN_RCV_WND; 525237263Snp 526237263Snp return min(wnd, MAX_RCV_WND); 527237263Snp} 528237263Snp 529237263Snpint 530237263Snpselect_rcv_wscale(void) 531237263Snp{ 532237263Snp int wscale = 0; 533237263Snp unsigned long space = sb_max; 534237263Snp 535237263Snp if (space > MAX_RCV_WND) 536237263Snp space = MAX_RCV_WND; 537237263Snp 538237263Snp while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < space) 539237263Snp wscale++; 540237263Snp 541237263Snp return (wscale); 542237263Snp} 543237263Snp 544237263Snp/* 545237263Snp * socket so could be a listening socket too. 546237263Snp */ 547237263Snpuint64_t 548291665Sjhbcalc_opt0(struct socket *so, struct vi_info *vi, struct l2t_entry *e, 549237263Snp int mtu_idx, int rscale, int rx_credits, int ulp_mode) 550237263Snp{ 551237263Snp uint64_t opt0; 552237263Snp 553237263Snp KASSERT(rx_credits <= M_RCV_BUFSIZ, 554237263Snp ("%s: rcv_bufsiz too high", __func__)); 555237263Snp 556237263Snp opt0 = F_TCAM_BYPASS | V_WND_SCALE(rscale) | V_MSS_IDX(mtu_idx) | 557237263Snp V_ULP_MODE(ulp_mode) | V_RCV_BUFSIZ(rx_credits); 558237263Snp 559237263Snp if (so != NULL) { 560237263Snp struct inpcb *inp = sotoinpcb(so); 561237263Snp struct tcpcb *tp = intotcpcb(inp); 562330303Sjhb int keepalive = tcp_always_keepalive || 563237263Snp so_options_get(so) & SO_KEEPALIVE; 564237263Snp 565237263Snp opt0 |= V_NAGLE((tp->t_flags & TF_NODELAY) == 0); 566237263Snp opt0 |= V_KEEP_ALIVE(keepalive != 0); 567237263Snp } 568237263Snp 569237263Snp if (e != NULL) 570237263Snp opt0 |= V_L2T_IDX(e->idx); 571237263Snp 572291665Sjhb if (vi != NULL) { 573309560Sjhb opt0 |= V_SMAC_SEL(vi->smt_idx); 574291665Sjhb opt0 |= V_TX_CHAN(vi->pi->tx_chan); 575237263Snp } 576237263Snp 577237263Snp return htobe64(opt0); 578237263Snp} 579237263Snp 580248925Snpuint64_t 581291665Sjhbselect_ntuple(struct vi_info *vi, struct l2t_entry *e) 582237263Snp{ 583291665Sjhb struct adapter *sc = vi->pi->adapter; 584252705Snp struct tp_params *tp = &sc->params.tp; 585291665Sjhb uint16_t viid = vi->viid; 586252705Snp uint64_t ntuple = 0; 587237263Snp 588252705Snp /* 589252705Snp * Initialize each of the fields which we care about which are present 590252705Snp * in the Compressed Filter Tuple. 591252705Snp */ 592252705Snp if (tp->vlan_shift >= 0 && e->vlan != CPL_L2T_VLAN_NONE) 593252705Snp ntuple |= (uint64_t)(F_FT_VLAN_VLD | e->vlan) << tp->vlan_shift; 594237263Snp 595252705Snp if (tp->port_shift >= 0) 596252705Snp ntuple |= (uint64_t)e->lport << tp->port_shift; 597252705Snp 598252705Snp if (tp->protocol_shift >= 0) 599252705Snp ntuple |= (uint64_t)IPPROTO_TCP << tp->protocol_shift; 600252705Snp 601252705Snp if (tp->vnic_shift >= 0) { 602252705Snp uint32_t vf = G_FW_VIID_VIN(viid); 603252705Snp uint32_t pf = G_FW_VIID_PFN(viid); 604252705Snp uint32_t vld = G_FW_VIID_VIVLD(viid); 605252705Snp 606252705Snp ntuple |= (uint64_t)(V_FT_VNID_ID_VF(vf) | V_FT_VNID_ID_PF(pf) | 607252705Snp V_FT_VNID_ID_VLD(vld)) << tp->vnic_shift; 608252705Snp } 609252705Snp 610252705Snp if (is_t4(sc)) 611252705Snp return (htobe32((uint32_t)ntuple)); 612248925Snp else 613248925Snp return (htobe64(V_FILTER_TUPLE(ntuple))); 614237263Snp} 615237263Snp 616245441Snpvoid 617245441Snpset_tcpddp_ulp_mode(struct toepcb *toep) 618245441Snp{ 619245441Snp 620245441Snp toep->ulp_mode = ULP_MODE_TCPDDP; 621331645Sjhb ddp_init_toep(toep); 622245441Snp} 623245441Snp 624245935Snpint 625245935Snpnegative_advice(int status) 626245935Snp{ 627245935Snp 628245935Snp return (status == CPL_ERR_RTX_NEG_ADVICE || 629245935Snp status == CPL_ERR_PERSIST_NEG_ADVICE || 630245935Snp status == CPL_ERR_KEEPALV_NEG_ADVICE); 631245935Snp} 632245935Snp 633237263Snpstatic int 634237263Snpalloc_tid_tabs(struct tid_info *t) 635237263Snp{ 636237263Snp size_t size; 637237263Snp unsigned int i; 638237263Snp 639237263Snp size = t->ntids * sizeof(*t->tid_tab) + 640237263Snp t->natids * sizeof(*t->atid_tab) + 641237263Snp t->nstids * sizeof(*t->stid_tab); 642237263Snp 643237263Snp t->tid_tab = malloc(size, M_CXGBE, M_ZERO | M_NOWAIT); 644237263Snp if (t->tid_tab == NULL) 645237263Snp return (ENOMEM); 646237263Snp 647237263Snp mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF); 648237263Snp t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids]; 649237263Snp t->afree = t->atid_tab; 650237263Snp t->atids_in_use = 0; 651237263Snp for (i = 1; i < t->natids; i++) 652237263Snp t->atid_tab[i - 1].next = &t->atid_tab[i]; 653237263Snp t->atid_tab[t->natids - 1].next = NULL; 654237263Snp 655237263Snp mtx_init(&t->stid_lock, "stid lock", NULL, MTX_DEF); 656245276Snp t->stid_tab = (struct listen_ctx **)&t->atid_tab[t->natids]; 657237263Snp t->stids_in_use = 0; 658245276Snp TAILQ_INIT(&t->stids); 659245276Snp t->nstids_free_head = t->nstids; 660237263Snp 661237263Snp atomic_store_rel_int(&t->tids_in_use, 0); 662237263Snp 663237263Snp return (0); 664237263Snp} 665237263Snp 666237263Snpstatic void 667237263Snpfree_tid_tabs(struct tid_info *t) 668237263Snp{ 669237263Snp KASSERT(t->tids_in_use == 0, 670237263Snp ("%s: %d tids still in use.", __func__, t->tids_in_use)); 671237263Snp KASSERT(t->atids_in_use == 0, 672237263Snp ("%s: %d atids still in use.", __func__, t->atids_in_use)); 673237263Snp KASSERT(t->stids_in_use == 0, 674237263Snp ("%s: %d tids still in use.", __func__, t->stids_in_use)); 675237263Snp 676237263Snp free(t->tid_tab, M_CXGBE); 677237263Snp t->tid_tab = NULL; 678237263Snp 679237263Snp if (mtx_initialized(&t->atid_lock)) 680237263Snp mtx_destroy(&t->atid_lock); 681237263Snp if (mtx_initialized(&t->stid_lock)) 682237263Snp mtx_destroy(&t->stid_lock); 683237263Snp} 684237263Snp 685245448Snpstatic int 686245448Snpadd_lip(struct adapter *sc, struct in6_addr *lip) 687245448Snp{ 688245448Snp struct fw_clip_cmd c; 689245448Snp 690245448Snp ASSERT_SYNCHRONIZED_OP(sc); 691245448Snp /* mtx_assert(&td->clip_table_lock, MA_OWNED); */ 692245448Snp 693245448Snp memset(&c, 0, sizeof(c)); 694245448Snp c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST | 695245448Snp F_FW_CMD_WRITE); 696245448Snp c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_ALLOC | FW_LEN16(c)); 697245448Snp c.ip_hi = *(uint64_t *)&lip->s6_addr[0]; 698245448Snp c.ip_lo = *(uint64_t *)&lip->s6_addr[8]; 699245448Snp 700249627Snp return (-t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c)); 701245448Snp} 702245448Snp 703245448Snpstatic int 704245448Snpdelete_lip(struct adapter *sc, struct in6_addr *lip) 705245448Snp{ 706245448Snp struct fw_clip_cmd c; 707245448Snp 708245448Snp ASSERT_SYNCHRONIZED_OP(sc); 709245448Snp /* mtx_assert(&td->clip_table_lock, MA_OWNED); */ 710245448Snp 711245448Snp memset(&c, 0, sizeof(c)); 712245448Snp c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST | 713245448Snp F_FW_CMD_READ); 714245448Snp c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_FREE | FW_LEN16(c)); 715245448Snp c.ip_hi = *(uint64_t *)&lip->s6_addr[0]; 716245448Snp c.ip_lo = *(uint64_t *)&lip->s6_addr[8]; 717245448Snp 718249627Snp return (-t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c)); 719245448Snp} 720245448Snp 721245448Snpstatic struct clip_entry * 722245448Snpsearch_lip(struct tom_data *td, struct in6_addr *lip) 723245448Snp{ 724245448Snp struct clip_entry *ce; 725245448Snp 726245448Snp mtx_assert(&td->clip_table_lock, MA_OWNED); 727245448Snp 728245448Snp TAILQ_FOREACH(ce, &td->clip_table, link) { 729245448Snp if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip)) 730245448Snp return (ce); 731245448Snp } 732245448Snp 733245448Snp return (NULL); 734245448Snp} 735245448Snp 736245448Snpstruct clip_entry * 737318803Snphold_lip(struct tom_data *td, struct in6_addr *lip, struct clip_entry *ce) 738245448Snp{ 739245448Snp 740245448Snp mtx_lock(&td->clip_table_lock); 741318803Snp if (ce == NULL) 742318803Snp ce = search_lip(td, lip); 743245448Snp if (ce != NULL) 744245448Snp ce->refcount++; 745245448Snp mtx_unlock(&td->clip_table_lock); 746245448Snp 747245448Snp return (ce); 748245448Snp} 749245448Snp 750245448Snpvoid 751245448Snprelease_lip(struct tom_data *td, struct clip_entry *ce) 752245448Snp{ 753245448Snp 754245448Snp mtx_lock(&td->clip_table_lock); 755245448Snp KASSERT(search_lip(td, &ce->lip) == ce, 756245448Snp ("%s: CLIP entry %p p not in CLIP table.", __func__, ce)); 757245448Snp KASSERT(ce->refcount > 0, 758245448Snp ("%s: CLIP entry %p has refcount 0", __func__, ce)); 759245448Snp --ce->refcount; 760245448Snp mtx_unlock(&td->clip_table_lock); 761245448Snp} 762245448Snp 763237263Snpstatic void 764245448Snpinit_clip_table(struct adapter *sc, struct tom_data *td) 765245448Snp{ 766245448Snp 767245448Snp ASSERT_SYNCHRONIZED_OP(sc); 768245448Snp 769245448Snp mtx_init(&td->clip_table_lock, "CLIP table lock", NULL, MTX_DEF); 770245448Snp TAILQ_INIT(&td->clip_table); 771249627Snp td->clip_gen = -1; 772245448Snp 773249627Snp update_clip_table(sc, td); 774249627Snp} 775249627Snp 776249627Snpstatic void 777249627Snpupdate_clip(struct adapter *sc, void *arg __unused) 778249627Snp{ 779249627Snp 780249627Snp if (begin_synchronized_op(sc, NULL, HOLD_LOCK, "t4tomuc")) 781249627Snp return; 782249627Snp 783278374Snp if (uld_active(sc, ULD_TOM)) 784249627Snp update_clip_table(sc, sc->tom_softc); 785249627Snp 786249627Snp end_synchronized_op(sc, LOCK_HELD); 787249627Snp} 788249627Snp 789249627Snpstatic void 790249627Snpt4_clip_task(void *arg, int count) 791249627Snp{ 792249627Snp 793249627Snp t4_iterate(update_clip, NULL); 794249627Snp} 795249627Snp 796249627Snpstatic void 797249627Snpupdate_clip_table(struct adapter *sc, struct tom_data *td) 798249627Snp{ 799286001Sae struct rm_priotracker in6_ifa_tracker; 800249627Snp struct in6_ifaddr *ia; 801249627Snp struct in6_addr *lip, tlip; 802249627Snp struct clip_head stale; 803249627Snp struct clip_entry *ce, *ce_temp; 804312116Snp struct vi_info *vi; 805312116Snp int rc, gen, i, j; 806312116Snp uintptr_t last_vnet; 807249627Snp 808249627Snp ASSERT_SYNCHRONIZED_OP(sc); 809249627Snp 810286001Sae IN6_IFADDR_RLOCK(&in6_ifa_tracker); 811249627Snp mtx_lock(&td->clip_table_lock); 812249627Snp 813312116Snp gen = atomic_load_acq_int(&in6_ifaddr_gen); 814249627Snp if (gen == td->clip_gen) 815249627Snp goto done; 816249627Snp 817249627Snp TAILQ_INIT(&stale); 818249627Snp TAILQ_CONCAT(&stale, &td->clip_table, link); 819249627Snp 820312116Snp /* 821312116Snp * last_vnet optimizes the common cases where all if_vnet = NULL (no 822312116Snp * VIMAGE) or all if_vnet = vnet0. 823312116Snp */ 824312116Snp last_vnet = (uintptr_t)(-1); 825312116Snp for_each_port(sc, i) 826312116Snp for_each_vi(sc->port[i], j, vi) { 827312116Snp if (last_vnet == (uintptr_t)vi->ifp->if_vnet) 828312116Snp continue; 829245448Snp 830312116Snp /* XXX: races with if_vmove */ 831312116Snp CURVNET_SET(vi->ifp->if_vnet); 832312116Snp TAILQ_FOREACH(ia, &V_in6_ifaddrhead, ia_link) { 833312116Snp lip = &ia->ia_addr.sin6_addr; 834245448Snp 835312116Snp KASSERT(!IN6_IS_ADDR_MULTICAST(lip), 836312116Snp ("%s: mcast address in in6_ifaddr list", __func__)); 837245448Snp 838312116Snp if (IN6_IS_ADDR_LOOPBACK(lip)) 839312116Snp continue; 840312116Snp if (IN6_IS_SCOPE_EMBED(lip)) { 841312116Snp /* Remove the embedded scope */ 842312116Snp tlip = *lip; 843312116Snp lip = &tlip; 844312116Snp in6_clearscope(lip); 845312116Snp } 846312116Snp /* 847312116Snp * XXX: how to weed out the link local address for the 848312116Snp * loopback interface? It's fe80::1 usually (always?). 849312116Snp */ 850249627Snp 851312116Snp /* 852312116Snp * If it's in the main list then we already know it's 853312116Snp * not stale. 854312116Snp */ 855312116Snp TAILQ_FOREACH(ce, &td->clip_table, link) { 856312116Snp if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip)) 857312116Snp goto next; 858249627Snp } 859249627Snp 860312116Snp /* 861312116Snp * If it's in the stale list we should move it to the 862312116Snp * main list. 863312116Snp */ 864312116Snp TAILQ_FOREACH(ce, &stale, link) { 865312116Snp if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip)) { 866312116Snp TAILQ_REMOVE(&stale, ce, link); 867312116Snp TAILQ_INSERT_TAIL(&td->clip_table, ce, 868312116Snp link); 869312116Snp goto next; 870312116Snp } 871312116Snp } 872249627Snp 873312116Snp /* A new IP6 address; add it to the CLIP table */ 874312116Snp ce = malloc(sizeof(*ce), M_CXGBE, M_NOWAIT); 875312116Snp memcpy(&ce->lip, lip, sizeof(ce->lip)); 876312116Snp ce->refcount = 0; 877312116Snp rc = add_lip(sc, lip); 878312116Snp if (rc == 0) 879312116Snp TAILQ_INSERT_TAIL(&td->clip_table, ce, link); 880312116Snp else { 881312116Snp char ip[INET6_ADDRSTRLEN]; 882312116Snp 883312116Snp inet_ntop(AF_INET6, &ce->lip, &ip[0], 884312116Snp sizeof(ip)); 885312116Snp log(LOG_ERR, "%s: could not add %s (%d)\n", 886312116Snp __func__, ip, rc); 887312116Snp free(ce, M_CXGBE); 888312116Snp } 889312116Snpnext: 890312116Snp continue; 891249627Snp } 892312116Snp CURVNET_RESTORE(); 893312116Snp last_vnet = (uintptr_t)vi->ifp->if_vnet; 894249627Snp } 895249627Snp 896249627Snp /* 897249627Snp * Remove stale addresses (those no longer in V_in6_ifaddrhead) that are 898249627Snp * no longer referenced by the driver. 899249627Snp */ 900249627Snp TAILQ_FOREACH_SAFE(ce, &stale, link, ce_temp) { 901249627Snp if (ce->refcount == 0) { 902249627Snp rc = delete_lip(sc, &ce->lip); 903249627Snp if (rc == 0) { 904249627Snp TAILQ_REMOVE(&stale, ce, link); 905245448Snp free(ce, M_CXGBE); 906249627Snp } else { 907249627Snp char ip[INET6_ADDRSTRLEN]; 908249627Snp 909249627Snp inet_ntop(AF_INET6, &ce->lip, &ip[0], 910249627Snp sizeof(ip)); 911249627Snp log(LOG_ERR, "%s: could not delete %s (%d)\n", 912249627Snp __func__, ip, rc); 913249627Snp } 914245448Snp } 915245448Snp } 916249627Snp /* The ones that are still referenced need to stay in the CLIP table */ 917249627Snp TAILQ_CONCAT(&td->clip_table, &stale, link); 918249627Snp 919249627Snp td->clip_gen = gen; 920249627Snpdone: 921249627Snp mtx_unlock(&td->clip_table_lock); 922286001Sae IN6_IFADDR_RUNLOCK(&in6_ifa_tracker); 923245448Snp} 924245448Snp 925245448Snpstatic void 926245448Snpdestroy_clip_table(struct adapter *sc, struct tom_data *td) 927245448Snp{ 928245448Snp struct clip_entry *ce, *ce_temp; 929245448Snp 930245448Snp if (mtx_initialized(&td->clip_table_lock)) { 931245448Snp mtx_lock(&td->clip_table_lock); 932245448Snp TAILQ_FOREACH_SAFE(ce, &td->clip_table, link, ce_temp) { 933245448Snp KASSERT(ce->refcount == 0, 934245448Snp ("%s: CLIP entry %p still in use (%d)", __func__, 935245448Snp ce, ce->refcount)); 936245448Snp TAILQ_REMOVE(&td->clip_table, ce, link); 937245448Snp delete_lip(sc, &ce->lip); 938245448Snp free(ce, M_CXGBE); 939245448Snp } 940245448Snp mtx_unlock(&td->clip_table_lock); 941245448Snp mtx_destroy(&td->clip_table_lock); 942245448Snp } 943245448Snp} 944245448Snp 945245448Snpstatic void 946237263Snpfree_tom_data(struct adapter *sc, struct tom_data *td) 947237263Snp{ 948245448Snp 949245448Snp ASSERT_SYNCHRONIZED_OP(sc); 950245448Snp 951237263Snp KASSERT(TAILQ_EMPTY(&td->toep_list), 952237263Snp ("%s: TOE PCB list is not empty.", __func__)); 953237263Snp KASSERT(td->lctx_count == 0, 954237263Snp ("%s: lctx hash table is not empty.", __func__)); 955237263Snp 956309555Sjhb t4_free_ppod_region(&td->pr); 957245448Snp destroy_clip_table(sc, td); 958237263Snp 959237263Snp if (td->listen_mask != 0) 960237263Snp hashdestroy(td->listen_hash, M_CXGBE, td->listen_mask); 961237263Snp 962272719Snp if (mtx_initialized(&td->unsent_wr_lock)) 963272719Snp mtx_destroy(&td->unsent_wr_lock); 964237263Snp if (mtx_initialized(&td->lctx_hash_lock)) 965237263Snp mtx_destroy(&td->lctx_hash_lock); 966237263Snp if (mtx_initialized(&td->toep_list_lock)) 967237263Snp mtx_destroy(&td->toep_list_lock); 968237263Snp 969237263Snp free_tid_tabs(&sc->tids); 970237263Snp free(td, M_CXGBE); 971237263Snp} 972237263Snp 973272719Snpstatic void 974272719Snpreclaim_wr_resources(void *arg, int count) 975272719Snp{ 976272719Snp struct tom_data *td = arg; 977272719Snp STAILQ_HEAD(, wrqe) twr_list = STAILQ_HEAD_INITIALIZER(twr_list); 978272719Snp struct cpl_act_open_req *cpl; 979272719Snp u_int opcode, atid; 980272719Snp struct wrqe *wr; 981272719Snp struct adapter *sc; 982272719Snp 983272719Snp mtx_lock(&td->unsent_wr_lock); 984272719Snp STAILQ_SWAP(&td->unsent_wr_list, &twr_list, wrqe); 985272719Snp mtx_unlock(&td->unsent_wr_lock); 986272719Snp 987272719Snp while ((wr = STAILQ_FIRST(&twr_list)) != NULL) { 988272719Snp STAILQ_REMOVE_HEAD(&twr_list, link); 989272719Snp 990272719Snp cpl = wrtod(wr); 991272719Snp opcode = GET_OPCODE(cpl); 992272719Snp 993272719Snp switch (opcode) { 994272719Snp case CPL_ACT_OPEN_REQ: 995272719Snp case CPL_ACT_OPEN_REQ6: 996272719Snp atid = G_TID_TID(be32toh(OPCODE_TID(cpl))); 997272719Snp sc = td_adapter(td); 998272719Snp 999272719Snp CTR2(KTR_CXGBE, "%s: atid %u ", __func__, atid); 1000272719Snp act_open_failure_cleanup(sc, atid, EHOSTUNREACH); 1001272719Snp free(wr, M_CXGBE); 1002272719Snp break; 1003272719Snp default: 1004272719Snp log(LOG_ERR, "%s: leaked work request %p, wr_len %d, " 1005272719Snp "opcode %x\n", __func__, wr, wr->wr_len, opcode); 1006272719Snp /* WR not freed here; go look at it with a debugger. */ 1007272719Snp } 1008272719Snp } 1009272719Snp} 1010272719Snp 1011237263Snp/* 1012237263Snp * Ground control to Major TOM 1013237263Snp * Commencing countdown, engines on 1014237263Snp */ 1015237263Snpstatic int 1016237263Snpt4_tom_activate(struct adapter *sc) 1017237263Snp{ 1018237263Snp struct tom_data *td; 1019237263Snp struct toedev *tod; 1020291665Sjhb struct vi_info *vi; 1021302339Snp struct sge_ofld_rxq *ofld_rxq; 1022302339Snp int i, j, rc, v; 1023237263Snp 1024245274Snp ASSERT_SYNCHRONIZED_OP(sc); 1025237263Snp 1026237263Snp /* per-adapter softc for TOM */ 1027237263Snp td = malloc(sizeof(*td), M_CXGBE, M_ZERO | M_NOWAIT); 1028237263Snp if (td == NULL) 1029237263Snp return (ENOMEM); 1030237263Snp 1031237263Snp /* List of TOE PCBs and associated lock */ 1032237263Snp mtx_init(&td->toep_list_lock, "PCB list lock", NULL, MTX_DEF); 1033237263Snp TAILQ_INIT(&td->toep_list); 1034237263Snp 1035237263Snp /* Listen context */ 1036237263Snp mtx_init(&td->lctx_hash_lock, "lctx hash lock", NULL, MTX_DEF); 1037237263Snp td->listen_hash = hashinit_flags(LISTEN_HASH_SIZE, M_CXGBE, 1038237263Snp &td->listen_mask, HASH_NOWAIT); 1039237263Snp 1040272719Snp /* List of WRs for which L2 resolution failed */ 1041272719Snp mtx_init(&td->unsent_wr_lock, "Unsent WR list lock", NULL, MTX_DEF); 1042272719Snp STAILQ_INIT(&td->unsent_wr_list); 1043272719Snp TASK_INIT(&td->reclaim_wr_resources, 0, reclaim_wr_resources, td); 1044272719Snp 1045237263Snp /* TID tables */ 1046237263Snp rc = alloc_tid_tabs(&sc->tids); 1047237263Snp if (rc != 0) 1048237263Snp goto done; 1049237263Snp 1050309555Sjhb rc = t4_init_ppod_region(&td->pr, &sc->vres.ddp, 1051309555Sjhb t4_read_reg(sc, A_ULP_RX_TDDP_PSZ), "TDDP page pods"); 1052309555Sjhb if (rc != 0) 1053309555Sjhb goto done; 1054309555Sjhb t4_set_reg_field(sc, A_ULP_RX_TDDP_TAGMASK, 1055309555Sjhb V_TDDPTAGMASK(M_TDDPTAGMASK), td->pr.pr_tag_mask); 1056239344Snp 1057245448Snp /* CLIP table for IPv6 offload */ 1058245448Snp init_clip_table(sc, td); 1059245448Snp 1060237263Snp /* toedev ops */ 1061237263Snp tod = &td->tod; 1062237263Snp init_toedev(tod); 1063237263Snp tod->tod_softc = sc; 1064237263Snp tod->tod_connect = t4_connect; 1065237263Snp tod->tod_listen_start = t4_listen_start; 1066237263Snp tod->tod_listen_stop = t4_listen_stop; 1067237263Snp tod->tod_rcvd = t4_rcvd; 1068237263Snp tod->tod_output = t4_tod_output; 1069237263Snp tod->tod_send_rst = t4_send_rst; 1070237263Snp tod->tod_send_fin = t4_send_fin; 1071237263Snp tod->tod_pcb_detach = t4_pcb_detach; 1072237263Snp tod->tod_l2_update = t4_l2_update; 1073237263Snp tod->tod_syncache_added = t4_syncache_added; 1074237263Snp tod->tod_syncache_removed = t4_syncache_removed; 1075237263Snp tod->tod_syncache_respond = t4_syncache_respond; 1076237263Snp tod->tod_offload_socket = t4_offload_socket; 1077252716Snp tod->tod_ctloutput = t4_ctloutput; 1078237263Snp 1079291665Sjhb for_each_port(sc, i) { 1080291665Sjhb for_each_vi(sc->port[i], v, vi) { 1081291665Sjhb TOEDEV(vi->ifp) = &td->tod; 1082302339Snp for_each_ofld_rxq(vi, j, ofld_rxq) { 1083302339Snp ofld_rxq->iq.set_tcb_rpl = do_set_tcb_rpl; 1084302339Snp ofld_rxq->iq.l2t_write_rpl = do_l2t_write_rpl2; 1085302339Snp } 1086291665Sjhb } 1087291665Sjhb } 1088237263Snp 1089237263Snp sc->tom_softc = td; 1090237263Snp register_toedev(sc->tom_softc); 1091237263Snp 1092237263Snpdone: 1093237263Snp if (rc != 0) 1094237263Snp free_tom_data(sc, td); 1095237263Snp return (rc); 1096237263Snp} 1097237263Snp 1098237263Snpstatic int 1099237263Snpt4_tom_deactivate(struct adapter *sc) 1100237263Snp{ 1101237263Snp int rc = 0; 1102237263Snp struct tom_data *td = sc->tom_softc; 1103237263Snp 1104245274Snp ASSERT_SYNCHRONIZED_OP(sc); 1105237263Snp 1106237263Snp if (td == NULL) 1107237263Snp return (0); /* XXX. KASSERT? */ 1108237263Snp 1109237263Snp if (sc->offload_map != 0) 1110237263Snp return (EBUSY); /* at least one port has IFCAP_TOE enabled */ 1111237263Snp 1112278374Snp if (uld_active(sc, ULD_IWARP) || uld_active(sc, ULD_ISCSI)) 1113278374Snp return (EBUSY); /* both iWARP and iSCSI rely on the TOE. */ 1114278374Snp 1115237263Snp mtx_lock(&td->toep_list_lock); 1116237263Snp if (!TAILQ_EMPTY(&td->toep_list)) 1117237263Snp rc = EBUSY; 1118237263Snp mtx_unlock(&td->toep_list_lock); 1119237263Snp 1120237263Snp mtx_lock(&td->lctx_hash_lock); 1121237263Snp if (td->lctx_count > 0) 1122237263Snp rc = EBUSY; 1123237263Snp mtx_unlock(&td->lctx_hash_lock); 1124237263Snp 1125272719Snp taskqueue_drain(taskqueue_thread, &td->reclaim_wr_resources); 1126272719Snp mtx_lock(&td->unsent_wr_lock); 1127272719Snp if (!STAILQ_EMPTY(&td->unsent_wr_list)) 1128272719Snp rc = EBUSY; 1129272719Snp mtx_unlock(&td->unsent_wr_lock); 1130272719Snp 1131237263Snp if (rc == 0) { 1132237263Snp unregister_toedev(sc->tom_softc); 1133237263Snp free_tom_data(sc, td); 1134237263Snp sc->tom_softc = NULL; 1135237263Snp } 1136237263Snp 1137237263Snp return (rc); 1138237263Snp} 1139237263Snp 1140249627Snpstatic void 1141249627Snpt4_tom_ifaddr_event(void *arg __unused, struct ifnet *ifp) 1142249627Snp{ 1143249627Snp 1144249627Snp atomic_add_rel_int(&in6_ifaddr_gen, 1); 1145249627Snp taskqueue_enqueue_timeout(taskqueue_thread, &clip_task, -hz / 4); 1146249627Snp} 1147249627Snp 1148237263Snpstatic int 1149306661Sjhbt4_aio_queue_tom(struct socket *so, struct kaiocb *job) 1150306661Sjhb{ 1151306661Sjhb struct tcpcb *tp = so_sototcpcb(so); 1152306661Sjhb struct toepcb *toep = tp->t_toe; 1153306661Sjhb int error; 1154306661Sjhb 1155306661Sjhb if (toep->ulp_mode == ULP_MODE_TCPDDP) { 1156306661Sjhb error = t4_aio_queue_ddp(so, job); 1157306661Sjhb if (error != EOPNOTSUPP) 1158306661Sjhb return (error); 1159306661Sjhb } 1160306661Sjhb 1161306661Sjhb return (t4_aio_queue_aiotx(so, job)); 1162306661Sjhb} 1163306661Sjhb 1164306661Sjhbstatic int 1165237263Snpt4_tom_mod_load(void) 1166237263Snp{ 1167237263Snp int rc; 1168245441Snp struct protosw *tcp_protosw, *tcp6_protosw; 1169237263Snp 1170302339Snp /* CPL handlers */ 1171302339Snp t4_init_connect_cpl_handlers(); 1172302339Snp t4_init_listen_cpl_handlers(); 1173302339Snp t4_init_cpl_io_handlers(); 1174302339Snp 1175299210Sjhb rc = t4_ddp_mod_load(); 1176299210Sjhb if (rc != 0) 1177299210Sjhb return (rc); 1178299210Sjhb 1179239344Snp tcp_protosw = pffindproto(PF_INET, IPPROTO_TCP, SOCK_STREAM); 1180239344Snp if (tcp_protosw == NULL) 1181239344Snp return (ENOPROTOOPT); 1182306661Sjhb bcopy(tcp_protosw, &toe_protosw, sizeof(toe_protosw)); 1183306661Sjhb bcopy(tcp_protosw->pr_usrreqs, &toe_usrreqs, sizeof(toe_usrreqs)); 1184306661Sjhb toe_usrreqs.pru_aio_queue = t4_aio_queue_tom; 1185306661Sjhb toe_protosw.pr_usrreqs = &toe_usrreqs; 1186239344Snp 1187245441Snp tcp6_protosw = pffindproto(PF_INET6, IPPROTO_TCP, SOCK_STREAM); 1188245441Snp if (tcp6_protosw == NULL) 1189245441Snp return (ENOPROTOOPT); 1190306661Sjhb bcopy(tcp6_protosw, &toe6_protosw, sizeof(toe6_protosw)); 1191306661Sjhb bcopy(tcp6_protosw->pr_usrreqs, &toe6_usrreqs, sizeof(toe6_usrreqs)); 1192306661Sjhb toe6_usrreqs.pru_aio_queue = t4_aio_queue_tom; 1193306661Sjhb toe6_protosw.pr_usrreqs = &toe6_usrreqs; 1194245441Snp 1195249627Snp TIMEOUT_TASK_INIT(taskqueue_thread, &clip_task, 0, t4_clip_task, NULL); 1196249627Snp ifaddr_evhandler = EVENTHANDLER_REGISTER(ifaddr_event, 1197249627Snp t4_tom_ifaddr_event, NULL, EVENTHANDLER_PRI_ANY); 1198249627Snp 1199237263Snp rc = t4_register_uld(&tom_uld_info); 1200237263Snp if (rc != 0) 1201237263Snp t4_tom_mod_unload(); 1202237263Snp 1203237263Snp return (rc); 1204237263Snp} 1205237263Snp 1206237263Snpstatic void 1207237263Snptom_uninit(struct adapter *sc, void *arg __unused) 1208237263Snp{ 1209255006Snp if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tomun")) 1210245274Snp return; 1211245274Snp 1212237263Snp /* Try to free resources (works only if no port has IFCAP_TOE) */ 1213278374Snp if (uld_active(sc, ULD_TOM)) 1214237263Snp t4_deactivate_uld(sc, ULD_TOM); 1215245274Snp 1216255006Snp end_synchronized_op(sc, 0); 1217237263Snp} 1218237263Snp 1219237263Snpstatic int 1220237263Snpt4_tom_mod_unload(void) 1221237263Snp{ 1222237263Snp t4_iterate(tom_uninit, NULL); 1223237263Snp 1224237263Snp if (t4_unregister_uld(&tom_uld_info) == EBUSY) 1225237263Snp return (EBUSY); 1226237263Snp 1227249627Snp if (ifaddr_evhandler) { 1228249627Snp EVENTHANDLER_DEREGISTER(ifaddr_event, ifaddr_evhandler); 1229249627Snp taskqueue_cancel_timeout(taskqueue_thread, &clip_task, NULL); 1230249627Snp } 1231249627Snp 1232299210Sjhb t4_ddp_mod_unload(); 1233299210Sjhb 1234313178Sjhb t4_uninit_connect_cpl_handlers(); 1235313178Sjhb t4_uninit_listen_cpl_handlers(); 1236313178Sjhb t4_uninit_cpl_io_handlers(); 1237313178Sjhb 1238237263Snp return (0); 1239237263Snp} 1240237263Snp#endif /* TCP_OFFLOAD */ 1241237263Snp 1242237263Snpstatic int 1243237263Snpt4_tom_modevent(module_t mod, int cmd, void *arg) 1244237263Snp{ 1245237263Snp int rc = 0; 1246237263Snp 1247237263Snp#ifdef TCP_OFFLOAD 1248237263Snp switch (cmd) { 1249237263Snp case MOD_LOAD: 1250237263Snp rc = t4_tom_mod_load(); 1251237263Snp break; 1252237263Snp 1253237263Snp case MOD_UNLOAD: 1254237263Snp rc = t4_tom_mod_unload(); 1255237263Snp break; 1256237263Snp 1257237263Snp default: 1258237263Snp rc = EINVAL; 1259237263Snp } 1260237263Snp#else 1261237263Snp printf("t4_tom: compiled without TCP_OFFLOAD support.\n"); 1262237263Snp rc = EOPNOTSUPP; 1263237263Snp#endif 1264237263Snp return (rc); 1265237263Snp} 1266237263Snp 1267237263Snpstatic moduledata_t t4_tom_moddata= { 1268237263Snp "t4_tom", 1269237263Snp t4_tom_modevent, 1270241394Skevlo 0 1271237263Snp}; 1272237263Snp 1273237263SnpMODULE_VERSION(t4_tom, 1); 1274237263SnpMODULE_DEPEND(t4_tom, toecore, 1, 1, 1); 1275237263SnpMODULE_DEPEND(t4_tom, t4nex, 1, 1, 1); 1276237263SnpDECLARE_MODULE(t4_tom, t4_tom_moddata, SI_SUB_EXEC, SI_ORDER_ANY); 1277