1237263Snp/*- 2237263Snp * Copyright (c) 2012 Chelsio Communications, Inc. 3237263Snp * All rights reserved. 4237263Snp * Written by: Navdeep Parhar <np@FreeBSD.org> 5237263Snp * 6237263Snp * Redistribution and use in source and binary forms, with or without 7237263Snp * modification, are permitted provided that the following conditions 8237263Snp * are met: 9237263Snp * 1. Redistributions of source code must retain the above copyright 10237263Snp * notice, this list of conditions and the following disclaimer. 11237263Snp * 2. Redistributions in binary form must reproduce the above copyright 12237263Snp * notice, this list of conditions and the following disclaimer in the 13237263Snp * documentation and/or other materials provided with the distribution. 14237263Snp * 15237263Snp * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16237263Snp * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17237263Snp * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18237263Snp * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19237263Snp * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20237263Snp * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21237263Snp * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22237263Snp * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23237263Snp * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24237263Snp * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25237263Snp * SUCH DAMAGE. 26237263Snp */ 27237263Snp 28237263Snp#include <sys/cdefs.h> 29237263Snp__FBSDID("$FreeBSD$"); 30237263Snp 31237263Snp#include "opt_inet.h" 32237263Snp 33237263Snp#ifdef TCP_OFFLOAD 34237263Snp#include <sys/param.h> 35237263Snp#include <sys/types.h> 36237263Snp#include <sys/kernel.h> 37237263Snp#include <sys/ktr.h> 38237263Snp#include <sys/module.h> 39237263Snp#include <sys/protosw.h> 40237263Snp#include <sys/domain.h> 41237263Snp#include <sys/socket.h> 42237263Snp#include <sys/socketvar.h> 43237263Snp#include <sys/sglist.h> 44237263Snp#include <netinet/in.h> 45237263Snp#include <netinet/in_pcb.h> 46237263Snp#include <netinet/ip.h> 47237263Snp#include <netinet/tcp_var.h> 48237263Snp#define TCPSTATES 49237263Snp#include <netinet/tcp_fsm.h> 50237263Snp#include <netinet/tcp_seq.h> 51237263Snp#include <netinet/toecore.h> 52237263Snp 53237263Snp#include "common/common.h" 54237263Snp#include "common/t4_msg.h" 55237263Snp#include "common/t4_regs.h" 56239344Snp#include "common/t4_tcb.h" 57237263Snp#include "tom/t4_tom_l2t.h" 58237263Snp#include "tom/t4_tom.h" 59237263Snp 60237263SnpVNET_DECLARE(int, tcp_do_autosndbuf); 61237263Snp#define V_tcp_do_autosndbuf VNET(tcp_do_autosndbuf) 62237263SnpVNET_DECLARE(int, tcp_autosndbuf_inc); 63237263Snp#define V_tcp_autosndbuf_inc VNET(tcp_autosndbuf_inc) 64237263SnpVNET_DECLARE(int, tcp_autosndbuf_max); 65237263Snp#define V_tcp_autosndbuf_max VNET(tcp_autosndbuf_max) 66237263SnpVNET_DECLARE(int, tcp_do_autorcvbuf); 67237263Snp#define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 68237263SnpVNET_DECLARE(int, tcp_autorcvbuf_inc); 69237263Snp#define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 70237263SnpVNET_DECLARE(int, tcp_autorcvbuf_max); 71237263Snp#define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 72237263Snp 73237263Snpvoid 74237263Snpsend_flowc_wr(struct toepcb *toep, struct flowc_tx_params *ftxp) 75237263Snp{ 76241626Snp struct wrqe *wr; 77241626Snp struct fw_flowc_wr *flowc; 78241642Snp unsigned int nparams = ftxp ? 8 : 6, flowclen; 79237263Snp struct port_info *pi = toep->port; 80237263Snp struct adapter *sc = pi->adapter; 81237263Snp unsigned int pfvf = G_FW_VIID_PFN(pi->viid) << S_FW_VIID_PFN; 82237263Snp struct ofld_tx_sdesc *txsd = &toep->txsd[toep->txsd_pidx]; 83237263Snp 84239514Snp KASSERT(!(toep->flags & TPF_FLOWC_WR_SENT), 85237263Snp ("%s: flowc for tid %u sent already", __func__, toep->tid)); 86237263Snp 87237263Snp CTR2(KTR_CXGBE, "%s: tid %u", __func__, toep->tid); 88237263Snp 89237263Snp flowclen = sizeof(*flowc) + nparams * sizeof(struct fw_flowc_mnemval); 90237263Snp 91248925Snp wr = alloc_wrqe(roundup2(flowclen, 16), toep->ofld_txq); 92237263Snp if (wr == NULL) { 93237263Snp /* XXX */ 94237263Snp panic("%s: allocation failure.", __func__); 95237263Snp } 96237263Snp flowc = wrtod(wr); 97237263Snp memset(flowc, 0, wr->wr_len); 98237263Snp 99237263Snp flowc->op_to_nparams = htobe32(V_FW_WR_OP(FW_FLOWC_WR) | 100237263Snp V_FW_FLOWC_WR_NPARAMS(nparams)); 101237263Snp flowc->flowid_len16 = htonl(V_FW_WR_LEN16(howmany(flowclen, 16)) | 102237263Snp V_FW_WR_FLOWID(toep->tid)); 103237263Snp 104237263Snp flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN; 105241626Snp flowc->mnemval[0].val = htobe32(pfvf); 106241626Snp flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH; 107241626Snp flowc->mnemval[1].val = htobe32(pi->tx_chan); 108241626Snp flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT; 109241626Snp flowc->mnemval[2].val = htobe32(pi->tx_chan); 110241626Snp flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID; 111241626Snp flowc->mnemval[3].val = htobe32(toep->ofld_rxq->iq.abs_id); 112237263Snp if (ftxp) { 113237263Snp uint32_t sndbuf = min(ftxp->snd_space, sc->tt.sndbuf); 114237263Snp 115237263Snp flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT; 116237263Snp flowc->mnemval[4].val = htobe32(ftxp->snd_nxt); 117237263Snp flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT; 118237263Snp flowc->mnemval[5].val = htobe32(ftxp->rcv_nxt); 119237263Snp flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF; 120237263Snp flowc->mnemval[6].val = htobe32(sndbuf); 121237263Snp flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS; 122237263Snp flowc->mnemval[7].val = htobe32(ftxp->mss); 123241642Snp } else { 124241642Snp flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDBUF; 125241642Snp flowc->mnemval[4].val = htobe32(512); 126241642Snp flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_MSS; 127241642Snp flowc->mnemval[5].val = htobe32(512); 128237263Snp } 129237263Snp 130237263Snp txsd->tx_credits = howmany(flowclen, 16); 131237263Snp txsd->plen = 0; 132237263Snp KASSERT(toep->tx_credits >= txsd->tx_credits && toep->txsd_avail > 0, 133237263Snp ("%s: not enough credits (%d)", __func__, toep->tx_credits)); 134237263Snp toep->tx_credits -= txsd->tx_credits; 135237263Snp if (__predict_false(++toep->txsd_pidx == toep->txsd_total)) 136237263Snp toep->txsd_pidx = 0; 137237263Snp toep->txsd_avail--; 138237263Snp 139239514Snp toep->flags |= TPF_FLOWC_WR_SENT; 140237263Snp t4_wrq_tx(sc, wr); 141237263Snp} 142237263Snp 143237263Snpvoid 144237263Snpsend_reset(struct adapter *sc, struct toepcb *toep, uint32_t snd_nxt) 145237263Snp{ 146237263Snp struct wrqe *wr; 147237263Snp struct cpl_abort_req *req; 148237263Snp int tid = toep->tid; 149237263Snp struct inpcb *inp = toep->inp; 150237263Snp struct tcpcb *tp = intotcpcb(inp); /* don't use if INP_DROPPED */ 151237263Snp 152237263Snp INP_WLOCK_ASSERT(inp); 153237263Snp 154237263Snp CTR6(KTR_CXGBE, "%s: tid %d (%s), toep_flags 0x%x, inp_flags 0x%x%s", 155237263Snp __func__, toep->tid, 156237263Snp inp->inp_flags & INP_DROPPED ? "inp dropped" : 157237263Snp tcpstates[tp->t_state], 158237263Snp toep->flags, inp->inp_flags, 159239514Snp toep->flags & TPF_ABORT_SHUTDOWN ? 160237263Snp " (abort already in progress)" : ""); 161237263Snp 162239514Snp if (toep->flags & TPF_ABORT_SHUTDOWN) 163237263Snp return; /* abort already in progress */ 164237263Snp 165239514Snp toep->flags |= TPF_ABORT_SHUTDOWN; 166237263Snp 167239514Snp KASSERT(toep->flags & TPF_FLOWC_WR_SENT, 168237263Snp ("%s: flowc_wr not sent for tid %d.", __func__, tid)); 169237263Snp 170237263Snp wr = alloc_wrqe(sizeof(*req), toep->ofld_txq); 171237263Snp if (wr == NULL) { 172237263Snp /* XXX */ 173237263Snp panic("%s: allocation failure.", __func__); 174237263Snp } 175237263Snp req = wrtod(wr); 176237263Snp 177237263Snp INIT_TP_WR_MIT_CPL(req, CPL_ABORT_REQ, tid); 178237263Snp if (inp->inp_flags & INP_DROPPED) 179237263Snp req->rsvd0 = htobe32(snd_nxt); 180237263Snp else 181237263Snp req->rsvd0 = htobe32(tp->snd_nxt); 182239514Snp req->rsvd1 = !(toep->flags & TPF_TX_DATA_SENT); 183237263Snp req->cmd = CPL_ABORT_SEND_RST; 184237263Snp 185237263Snp /* 186237263Snp * XXX: What's the correct way to tell that the inp hasn't been detached 187237263Snp * from its socket? Should I even be flushing the snd buffer here? 188237263Snp */ 189237263Snp if ((inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) == 0) { 190237263Snp struct socket *so = inp->inp_socket; 191237263Snp 192237263Snp if (so != NULL) /* because I'm not sure. See comment above */ 193237263Snp sbflush(&so->so_snd); 194237263Snp } 195237263Snp 196237263Snp t4_l2t_send(sc, wr, toep->l2te); 197237263Snp} 198237263Snp 199237263Snp/* 200237263Snp * Called when a connection is established to translate the TCP options 201237263Snp * reported by HW to FreeBSD's native format. 202237263Snp */ 203237263Snpstatic void 204237263Snpassign_rxopt(struct tcpcb *tp, unsigned int opt) 205237263Snp{ 206237263Snp struct toepcb *toep = tp->t_toe; 207237263Snp struct adapter *sc = td_adapter(toep->td); 208237263Snp 209237263Snp INP_LOCK_ASSERT(tp->t_inpcb); 210237263Snp 211237263Snp tp->t_maxseg = tp->t_maxopd = sc->params.mtus[G_TCPOPT_MSS(opt)] - 40; 212237263Snp 213237263Snp if (G_TCPOPT_TSTAMP(opt)) { 214237263Snp tp->t_flags |= TF_RCVD_TSTMP; /* timestamps ok */ 215237263Snp tp->ts_recent = 0; /* hmmm */ 216237263Snp tp->ts_recent_age = tcp_ts_getticks(); 217237263Snp tp->t_maxseg -= TCPOLEN_TSTAMP_APPA; 218237263Snp } 219237263Snp 220237263Snp if (G_TCPOPT_SACK(opt)) 221237263Snp tp->t_flags |= TF_SACK_PERMIT; /* should already be set */ 222237263Snp else 223237263Snp tp->t_flags &= ~TF_SACK_PERMIT; /* sack disallowed by peer */ 224237263Snp 225237263Snp if (G_TCPOPT_WSCALE_OK(opt)) 226237263Snp tp->t_flags |= TF_RCVD_SCALE; 227237263Snp 228237263Snp /* Doing window scaling? */ 229237263Snp if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 230237263Snp (TF_RCVD_SCALE | TF_REQ_SCALE)) { 231237263Snp tp->rcv_scale = tp->request_r_scale; 232237263Snp tp->snd_scale = G_TCPOPT_SND_WSCALE(opt); 233237263Snp } 234237263Snp} 235237263Snp 236237263Snp/* 237237263Snp * Completes some final bits of initialization for just established connections 238237263Snp * and changes their state to TCPS_ESTABLISHED. 239237263Snp * 240237263Snp * The ISNs are from after the exchange of SYNs. i.e., the true ISN + 1. 241237263Snp */ 242237263Snpvoid 243237263Snpmake_established(struct toepcb *toep, uint32_t snd_isn, uint32_t rcv_isn, 244237263Snp uint16_t opt) 245237263Snp{ 246237263Snp struct inpcb *inp = toep->inp; 247237263Snp struct socket *so = inp->inp_socket; 248237263Snp struct tcpcb *tp = intotcpcb(inp); 249237263Snp long bufsize; 250237263Snp uint32_t iss = be32toh(snd_isn) - 1; /* true ISS */ 251237263Snp uint32_t irs = be32toh(rcv_isn) - 1; /* true IRS */ 252237263Snp uint16_t tcpopt = be16toh(opt); 253237263Snp struct flowc_tx_params ftxp; 254237263Snp 255237263Snp INP_WLOCK_ASSERT(inp); 256237263Snp KASSERT(tp->t_state == TCPS_SYN_SENT || 257237263Snp tp->t_state == TCPS_SYN_RECEIVED, 258237263Snp ("%s: TCP state %s", __func__, tcpstates[tp->t_state])); 259237263Snp 260237263Snp CTR4(KTR_CXGBE, "%s: tid %d, toep %p, inp %p", 261237263Snp __func__, toep->tid, toep, inp); 262237263Snp 263237263Snp tp->t_state = TCPS_ESTABLISHED; 264237263Snp tp->t_starttime = ticks; 265237263Snp TCPSTAT_INC(tcps_connects); 266237263Snp 267237263Snp tp->irs = irs; 268237263Snp tcp_rcvseqinit(tp); 269237263Snp tp->rcv_wnd = toep->rx_credits << 10; 270237263Snp tp->rcv_adv += tp->rcv_wnd; 271237263Snp tp->last_ack_sent = tp->rcv_nxt; 272237263Snp 273237263Snp /* 274237263Snp * If we were unable to send all rx credits via opt0, save the remainder 275237263Snp * in rx_credits so that they can be handed over with the next credit 276237263Snp * update. 277237263Snp */ 278237263Snp SOCKBUF_LOCK(&so->so_rcv); 279237263Snp bufsize = select_rcv_wnd(so); 280237263Snp SOCKBUF_UNLOCK(&so->so_rcv); 281237263Snp toep->rx_credits = bufsize - tp->rcv_wnd; 282237263Snp 283237263Snp tp->iss = iss; 284237263Snp tcp_sendseqinit(tp); 285237263Snp tp->snd_una = iss + 1; 286237263Snp tp->snd_nxt = iss + 1; 287237263Snp tp->snd_max = iss + 1; 288237263Snp 289237263Snp assign_rxopt(tp, tcpopt); 290237263Snp 291237263Snp SOCKBUF_LOCK(&so->so_snd); 292237263Snp if (so->so_snd.sb_flags & SB_AUTOSIZE && V_tcp_do_autosndbuf) 293237263Snp bufsize = V_tcp_autosndbuf_max; 294237263Snp else 295237263Snp bufsize = sbspace(&so->so_snd); 296237263Snp SOCKBUF_UNLOCK(&so->so_snd); 297237263Snp 298237263Snp ftxp.snd_nxt = tp->snd_nxt; 299237263Snp ftxp.rcv_nxt = tp->rcv_nxt; 300237263Snp ftxp.snd_space = bufsize; 301237263Snp ftxp.mss = tp->t_maxseg; 302237263Snp send_flowc_wr(toep, &ftxp); 303237263Snp 304237263Snp soisconnected(so); 305237263Snp} 306237263Snp 307237263Snpstatic int 308239344Snpsend_rx_credits(struct adapter *sc, struct toepcb *toep, int credits) 309237263Snp{ 310237263Snp struct wrqe *wr; 311237263Snp struct cpl_rx_data_ack *req; 312237263Snp uint32_t dack = F_RX_DACK_CHANGE | V_RX_DACK_MODE(1); 313237263Snp 314239344Snp KASSERT(credits >= 0, ("%s: %d credits", __func__, credits)); 315239344Snp 316237263Snp wr = alloc_wrqe(sizeof(*req), toep->ctrlq); 317237263Snp if (wr == NULL) 318237263Snp return (0); 319237263Snp req = wrtod(wr); 320237263Snp 321237263Snp INIT_TP_WR_MIT_CPL(req, CPL_RX_DATA_ACK, toep->tid); 322237263Snp req->credit_dack = htobe32(dack | V_RX_CREDITS(credits)); 323237263Snp 324237263Snp t4_wrq_tx(sc, wr); 325237263Snp return (credits); 326237263Snp} 327237263Snp 328237263Snpvoid 329237263Snpt4_rcvd(struct toedev *tod, struct tcpcb *tp) 330237263Snp{ 331237263Snp struct adapter *sc = tod->tod_softc; 332237263Snp struct inpcb *inp = tp->t_inpcb; 333237263Snp struct socket *so = inp->inp_socket; 334239344Snp struct sockbuf *sb = &so->so_rcv; 335237263Snp struct toepcb *toep = tp->t_toe; 336239344Snp int credits; 337237263Snp 338237263Snp INP_WLOCK_ASSERT(inp); 339237263Snp 340239344Snp SOCKBUF_LOCK(sb); 341239344Snp KASSERT(toep->sb_cc >= sb->sb_cc, 342239344Snp ("%s: sb %p has more data (%d) than last time (%d).", 343239344Snp __func__, sb, sb->sb_cc, toep->sb_cc)); 344239344Snp toep->rx_credits += toep->sb_cc - sb->sb_cc; 345239344Snp toep->sb_cc = sb->sb_cc; 346239344Snp credits = toep->rx_credits; 347239344Snp SOCKBUF_UNLOCK(sb); 348237263Snp 349239344Snp if (credits > 0 && 350239344Snp (credits + 16384 >= tp->rcv_wnd || credits >= 15 * 1024)) { 351237263Snp 352239344Snp credits = send_rx_credits(sc, toep, credits); 353239344Snp SOCKBUF_LOCK(sb); 354237263Snp toep->rx_credits -= credits; 355239344Snp SOCKBUF_UNLOCK(sb); 356237263Snp tp->rcv_wnd += credits; 357237263Snp tp->rcv_adv += credits; 358237263Snp } 359237263Snp} 360237263Snp 361237263Snp/* 362237263Snp * Close a connection by sending a CPL_CLOSE_CON_REQ message. 363237263Snp */ 364237263Snpstatic int 365237263Snpclose_conn(struct adapter *sc, struct toepcb *toep) 366237263Snp{ 367237263Snp struct wrqe *wr; 368237263Snp struct cpl_close_con_req *req; 369237263Snp unsigned int tid = toep->tid; 370237263Snp 371237263Snp CTR3(KTR_CXGBE, "%s: tid %u%s", __func__, toep->tid, 372239514Snp toep->flags & TPF_FIN_SENT ? ", IGNORED" : ""); 373237263Snp 374239514Snp if (toep->flags & TPF_FIN_SENT) 375237263Snp return (0); 376237263Snp 377239514Snp KASSERT(toep->flags & TPF_FLOWC_WR_SENT, 378237263Snp ("%s: flowc_wr not sent for tid %u.", __func__, tid)); 379237263Snp 380237263Snp wr = alloc_wrqe(sizeof(*req), toep->ofld_txq); 381237263Snp if (wr == NULL) { 382237263Snp /* XXX */ 383237263Snp panic("%s: allocation failure.", __func__); 384237263Snp } 385237263Snp req = wrtod(wr); 386237263Snp 387237263Snp req->wr.wr_hi = htonl(V_FW_WR_OP(FW_TP_WR) | 388237263Snp V_FW_WR_IMMDLEN(sizeof(*req) - sizeof(req->wr))); 389237263Snp req->wr.wr_mid = htonl(V_FW_WR_LEN16(howmany(sizeof(*req), 16)) | 390237263Snp V_FW_WR_FLOWID(tid)); 391237263Snp req->wr.wr_lo = cpu_to_be64(0); 392237263Snp OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, tid)); 393237263Snp req->rsvd = 0; 394237263Snp 395239514Snp toep->flags |= TPF_FIN_SENT; 396239514Snp toep->flags &= ~TPF_SEND_FIN; 397237263Snp t4_l2t_send(sc, wr, toep->l2te); 398237263Snp 399237263Snp return (0); 400237263Snp} 401237263Snp 402237263Snp#define MAX_OFLD_TX_CREDITS (SGE_MAX_WR_LEN / 16) 403237263Snp#define MIN_OFLD_TX_CREDITS (howmany(sizeof(struct fw_ofld_tx_data_wr) + 1, 16)) 404237263Snp 405237263Snp/* Maximum amount of immediate data we could stuff in a WR */ 406237263Snpstatic inline int 407237263Snpmax_imm_payload(int tx_credits) 408237263Snp{ 409237263Snp const int n = 2; /* Use only up to 2 desc for imm. data WR */ 410237263Snp 411237263Snp KASSERT(tx_credits >= 0 && 412237263Snp tx_credits <= MAX_OFLD_TX_CREDITS, 413237263Snp ("%s: %d credits", __func__, tx_credits)); 414237263Snp 415237263Snp if (tx_credits < MIN_OFLD_TX_CREDITS) 416237263Snp return (0); 417237263Snp 418237263Snp if (tx_credits >= (n * EQ_ESIZE) / 16) 419237263Snp return ((n * EQ_ESIZE) - sizeof(struct fw_ofld_tx_data_wr)); 420237263Snp else 421237263Snp return (tx_credits * 16 - sizeof(struct fw_ofld_tx_data_wr)); 422237263Snp} 423237263Snp 424237263Snp/* Maximum number of SGL entries we could stuff in a WR */ 425237263Snpstatic inline int 426237263Snpmax_dsgl_nsegs(int tx_credits) 427237263Snp{ 428237263Snp int nseg = 1; /* ulptx_sgl has room for 1, rest ulp_tx_sge_pair */ 429237263Snp int sge_pair_credits = tx_credits - MIN_OFLD_TX_CREDITS; 430237263Snp 431237263Snp KASSERT(tx_credits >= 0 && 432237263Snp tx_credits <= MAX_OFLD_TX_CREDITS, 433237263Snp ("%s: %d credits", __func__, tx_credits)); 434237263Snp 435237263Snp if (tx_credits < MIN_OFLD_TX_CREDITS) 436237263Snp return (0); 437237263Snp 438237263Snp nseg += 2 * (sge_pair_credits * 16 / 24); 439237263Snp if ((sge_pair_credits * 16) % 24 == 16) 440237263Snp nseg++; 441237263Snp 442237263Snp return (nseg); 443237263Snp} 444237263Snp 445237263Snpstatic inline void 446237263Snpwrite_tx_wr(void *dst, struct toepcb *toep, unsigned int immdlen, 447255411Snp unsigned int plen, uint8_t credits, int shove) 448237263Snp{ 449237263Snp struct fw_ofld_tx_data_wr *txwr = dst; 450237263Snp 451237263Snp txwr->op_to_immdlen = htobe32(V_WR_OP(FW_OFLD_TX_DATA_WR) | 452255411Snp V_FW_WR_IMMDLEN(immdlen)); 453237263Snp txwr->flowid_len16 = htobe32(V_FW_WR_FLOWID(toep->tid) | 454237263Snp V_FW_WR_LEN16(credits)); 455256791Snp txwr->lsodisable_to_proxy = 456237263Snp htobe32(V_FW_OFLD_TX_DATA_WR_ULPMODE(toep->ulp_mode) | 457237263Snp V_FW_OFLD_TX_DATA_WR_URGENT(0) | /* XXX */ 458237263Snp V_FW_OFLD_TX_DATA_WR_SHOVE(shove)); 459237263Snp txwr->plen = htobe32(plen); 460237263Snp} 461237263Snp 462237263Snp/* 463237263Snp * Generate a DSGL from a starting mbuf. The total number of segments and the 464237263Snp * maximum segments in any one mbuf are provided. 465237263Snp */ 466237263Snpstatic void 467237263Snpwrite_tx_sgl(void *dst, struct mbuf *start, struct mbuf *stop, int nsegs, int n) 468237263Snp{ 469237263Snp struct mbuf *m; 470237263Snp struct ulptx_sgl *usgl = dst; 471237263Snp int i, j, rc; 472237263Snp struct sglist sg; 473237263Snp struct sglist_seg segs[n]; 474237263Snp 475237263Snp KASSERT(nsegs > 0, ("%s: nsegs 0", __func__)); 476237263Snp 477237263Snp sglist_init(&sg, n, segs); 478237263Snp usgl->cmd_nsge = htobe32(V_ULPTX_CMD(ULP_TX_SC_DSGL) | 479237263Snp V_ULPTX_NSGE(nsegs)); 480237263Snp 481237263Snp i = -1; 482237263Snp for (m = start; m != stop; m = m->m_next) { 483237263Snp rc = sglist_append(&sg, mtod(m, void *), m->m_len); 484237263Snp if (__predict_false(rc != 0)) 485237263Snp panic("%s: sglist_append %d", __func__, rc); 486237263Snp 487237263Snp for (j = 0; j < sg.sg_nseg; i++, j++) { 488237263Snp if (i < 0) { 489237263Snp usgl->len0 = htobe32(segs[j].ss_len); 490237263Snp usgl->addr0 = htobe64(segs[j].ss_paddr); 491237263Snp } else { 492237263Snp usgl->sge[i / 2].len[i & 1] = 493237263Snp htobe32(segs[j].ss_len); 494237263Snp usgl->sge[i / 2].addr[i & 1] = 495237263Snp htobe64(segs[j].ss_paddr); 496237263Snp } 497237263Snp#ifdef INVARIANTS 498237263Snp nsegs--; 499237263Snp#endif 500237263Snp } 501237263Snp sglist_reset(&sg); 502237263Snp } 503237263Snp if (i & 1) 504237263Snp usgl->sge[i / 2].len[1] = htobe32(0); 505237263Snp KASSERT(nsegs == 0, ("%s: nsegs %d, start %p, stop %p", 506237263Snp __func__, nsegs, start, stop)); 507237263Snp} 508237263Snp 509237263Snp/* 510237263Snp * Max number of SGL entries an offload tx work request can have. This is 41 511237263Snp * (1 + 40) for a full 512B work request. 512237263Snp * fw_ofld_tx_data_wr(16B) + ulptx_sgl(16B, 1) + ulptx_sge_pair(480B, 40) 513237263Snp */ 514237263Snp#define OFLD_SGL_LEN (41) 515237263Snp 516237263Snp/* 517237263Snp * Send data and/or a FIN to the peer. 518237263Snp * 519237263Snp * The socket's so_snd buffer consists of a stream of data starting with sb_mb 520237263Snp * and linked together with m_next. sb_sndptr, if set, is the last mbuf that 521237263Snp * was transmitted. 522255411Snp * 523255411Snp * drop indicates the number of bytes that should be dropped from the head of 524255411Snp * the send buffer. It is an optimization that lets do_fw4_ack avoid creating 525255411Snp * contention on the send buffer lock (before this change it used to do 526255411Snp * sowwakeup and then t4_push_frames right after that when recovering from tx 527255411Snp * stalls). When drop is set this function MUST drop the bytes and wake up any 528255411Snp * writers. 529237263Snp */ 530237263Snpstatic void 531255411Snpt4_push_frames(struct adapter *sc, struct toepcb *toep, int drop) 532237263Snp{ 533237263Snp struct mbuf *sndptr, *m, *sb_sndptr; 534237263Snp struct fw_ofld_tx_data_wr *txwr; 535237263Snp struct wrqe *wr; 536255411Snp u_int plen, nsegs, credits, max_imm, max_nsegs, max_nsegs_1mbuf; 537237263Snp struct inpcb *inp = toep->inp; 538237263Snp struct tcpcb *tp = intotcpcb(inp); 539237263Snp struct socket *so = inp->inp_socket; 540237263Snp struct sockbuf *sb = &so->so_snd; 541255411Snp int tx_credits, shove, compl, space, sowwakeup; 542237263Snp struct ofld_tx_sdesc *txsd = &toep->txsd[toep->txsd_pidx]; 543237263Snp 544237263Snp INP_WLOCK_ASSERT(inp); 545239514Snp KASSERT(toep->flags & TPF_FLOWC_WR_SENT, 546237263Snp ("%s: flowc_wr not sent for tid %u.", __func__, toep->tid)); 547237263Snp 548255005Snp KASSERT(toep->ulp_mode == ULP_MODE_NONE || 549255005Snp toep->ulp_mode == ULP_MODE_TCPDDP || 550255005Snp toep->ulp_mode == ULP_MODE_RDMA, 551255005Snp ("%s: ulp_mode %u for toep %p", __func__, toep->ulp_mode, toep)); 552237263Snp 553237263Snp /* 554237263Snp * This function doesn't resume by itself. Someone else must clear the 555237263Snp * flag and call this function. 556237263Snp */ 557255411Snp if (__predict_false(toep->flags & TPF_TX_SUSPENDED)) { 558255411Snp KASSERT(drop == 0, 559255411Snp ("%s: drop (%d) != 0 but tx is suspended", __func__, drop)); 560237263Snp return; 561255411Snp } 562237263Snp 563237263Snp do { 564237263Snp tx_credits = min(toep->tx_credits, MAX_OFLD_TX_CREDITS); 565237263Snp max_imm = max_imm_payload(tx_credits); 566237263Snp max_nsegs = max_dsgl_nsegs(tx_credits); 567237263Snp 568237263Snp SOCKBUF_LOCK(sb); 569255411Snp sowwakeup = drop; 570255411Snp if (drop) { 571255411Snp sbdrop_locked(sb, drop); 572255411Snp drop = 0; 573255411Snp } 574237263Snp sb_sndptr = sb->sb_sndptr; 575237263Snp sndptr = sb_sndptr ? sb_sndptr->m_next : sb->sb_mb; 576237263Snp plen = 0; 577237263Snp nsegs = 0; 578237263Snp max_nsegs_1mbuf = 0; /* max # of SGL segments in any one mbuf */ 579237263Snp for (m = sndptr; m != NULL; m = m->m_next) { 580237263Snp int n = sglist_count(mtod(m, void *), m->m_len); 581237263Snp 582237263Snp nsegs += n; 583237263Snp plen += m->m_len; 584237263Snp 585237263Snp /* This mbuf sent us _over_ the nsegs limit, back out */ 586237263Snp if (plen > max_imm && nsegs > max_nsegs) { 587237263Snp nsegs -= n; 588237263Snp plen -= m->m_len; 589237263Snp if (plen == 0) { 590237263Snp /* Too few credits */ 591239514Snp toep->flags |= TPF_TX_SUSPENDED; 592255411Snp if (sowwakeup) 593255411Snp sowwakeup_locked(so); 594255411Snp else 595255411Snp SOCKBUF_UNLOCK(sb); 596255411Snp SOCKBUF_UNLOCK_ASSERT(sb); 597237263Snp return; 598237263Snp } 599237263Snp break; 600237263Snp } 601237263Snp 602237263Snp if (max_nsegs_1mbuf < n) 603237263Snp max_nsegs_1mbuf = n; 604237263Snp sb_sndptr = m; /* new sb->sb_sndptr if all goes well */ 605237263Snp 606237263Snp /* This mbuf put us right at the max_nsegs limit */ 607237263Snp if (plen > max_imm && nsegs == max_nsegs) { 608237263Snp m = m->m_next; 609237263Snp break; 610237263Snp } 611237263Snp } 612237263Snp 613255411Snp shove = m == NULL && !(tp->t_flags & TF_MORETOCOME); 614255411Snp space = sbspace(sb); 615255411Snp 616255411Snp if (space <= sb->sb_hiwat * 3 / 8 && 617255411Snp toep->plen_nocompl + plen >= sb->sb_hiwat / 4) 618255411Snp compl = 1; 619255411Snp else 620255411Snp compl = 0; 621255411Snp 622237263Snp if (sb->sb_flags & SB_AUTOSIZE && 623237263Snp V_tcp_do_autosndbuf && 624237263Snp sb->sb_hiwat < V_tcp_autosndbuf_max && 625255411Snp space < sb->sb_hiwat / 8) { 626237263Snp int newsize = min(sb->sb_hiwat + V_tcp_autosndbuf_inc, 627237263Snp V_tcp_autosndbuf_max); 628237263Snp 629237263Snp if (!sbreserve_locked(sb, newsize, so, NULL)) 630237263Snp sb->sb_flags &= ~SB_AUTOSIZE; 631255411Snp else 632255411Snp sowwakeup = 1; /* room available */ 633237263Snp } 634255411Snp if (sowwakeup) 635255411Snp sowwakeup_locked(so); 636255411Snp else 637255411Snp SOCKBUF_UNLOCK(sb); 638255411Snp SOCKBUF_UNLOCK_ASSERT(sb); 639237263Snp 640237263Snp /* nothing to send */ 641237263Snp if (plen == 0) { 642237263Snp KASSERT(m == NULL, 643237263Snp ("%s: nothing to send, but m != NULL", __func__)); 644237263Snp break; 645237263Snp } 646237263Snp 647239514Snp if (__predict_false(toep->flags & TPF_FIN_SENT)) 648237263Snp panic("%s: excess tx.", __func__); 649237263Snp 650237263Snp if (plen <= max_imm) { 651237263Snp 652237263Snp /* Immediate data tx */ 653237263Snp 654248925Snp wr = alloc_wrqe(roundup2(sizeof(*txwr) + plen, 16), 655237263Snp toep->ofld_txq); 656237263Snp if (wr == NULL) { 657237263Snp /* XXX: how will we recover from this? */ 658239514Snp toep->flags |= TPF_TX_SUSPENDED; 659237263Snp return; 660237263Snp } 661237263Snp txwr = wrtod(wr); 662237263Snp credits = howmany(wr->wr_len, 16); 663255411Snp write_tx_wr(txwr, toep, plen, plen, credits, shove); 664237263Snp m_copydata(sndptr, 0, plen, (void *)(txwr + 1)); 665255411Snp nsegs = 0; 666237263Snp } else { 667237263Snp int wr_len; 668237263Snp 669237263Snp /* DSGL tx */ 670237263Snp 671237263Snp wr_len = sizeof(*txwr) + sizeof(struct ulptx_sgl) + 672237263Snp ((3 * (nsegs - 1)) / 2 + ((nsegs - 1) & 1)) * 8; 673248925Snp wr = alloc_wrqe(roundup2(wr_len, 16), toep->ofld_txq); 674237263Snp if (wr == NULL) { 675237263Snp /* XXX: how will we recover from this? */ 676239514Snp toep->flags |= TPF_TX_SUSPENDED; 677237263Snp return; 678237263Snp } 679237263Snp txwr = wrtod(wr); 680237263Snp credits = howmany(wr_len, 16); 681255411Snp write_tx_wr(txwr, toep, 0, plen, credits, shove); 682237263Snp write_tx_sgl(txwr + 1, sndptr, m, nsegs, 683237263Snp max_nsegs_1mbuf); 684237263Snp if (wr_len & 0xf) { 685237263Snp uint64_t *pad = (uint64_t *) 686237263Snp ((uintptr_t)txwr + wr_len); 687237263Snp *pad = 0; 688237263Snp } 689237263Snp } 690237263Snp 691237263Snp KASSERT(toep->tx_credits >= credits, 692237263Snp ("%s: not enough credits", __func__)); 693237263Snp 694237263Snp toep->tx_credits -= credits; 695255411Snp toep->tx_nocompl += credits; 696255411Snp toep->plen_nocompl += plen; 697255411Snp if (toep->tx_credits <= toep->tx_total * 3 / 8 && 698255411Snp toep->tx_nocompl >= toep->tx_total / 4) 699255411Snp compl = 1; 700237263Snp 701255411Snp if (compl) { 702255411Snp txwr->op_to_immdlen |= htobe32(F_FW_WR_COMPL); 703255411Snp toep->tx_nocompl = 0; 704255411Snp toep->plen_nocompl = 0; 705255411Snp } 706255411Snp 707237263Snp tp->snd_nxt += plen; 708237263Snp tp->snd_max += plen; 709237263Snp 710237263Snp SOCKBUF_LOCK(sb); 711237263Snp KASSERT(sb_sndptr, ("%s: sb_sndptr is NULL", __func__)); 712237263Snp sb->sb_sndptr = sb_sndptr; 713237263Snp SOCKBUF_UNLOCK(sb); 714237263Snp 715239514Snp toep->flags |= TPF_TX_DATA_SENT; 716255411Snp if (toep->tx_credits < MIN_OFLD_TX_CREDITS) 717255411Snp toep->flags |= TPF_TX_SUSPENDED; 718237263Snp 719237263Snp KASSERT(toep->txsd_avail > 0, ("%s: no txsd", __func__)); 720237263Snp txsd->plen = plen; 721237263Snp txsd->tx_credits = credits; 722237263Snp txsd++; 723237263Snp if (__predict_false(++toep->txsd_pidx == toep->txsd_total)) { 724237263Snp toep->txsd_pidx = 0; 725237263Snp txsd = &toep->txsd[0]; 726237263Snp } 727237263Snp toep->txsd_avail--; 728237263Snp 729237263Snp t4_l2t_send(sc, wr, toep->l2te); 730237263Snp } while (m != NULL); 731237263Snp 732237263Snp /* Send a FIN if requested, but only if there's no more data to send */ 733239514Snp if (m == NULL && toep->flags & TPF_SEND_FIN) 734237263Snp close_conn(sc, toep); 735237263Snp} 736237263Snp 737237263Snpint 738237263Snpt4_tod_output(struct toedev *tod, struct tcpcb *tp) 739237263Snp{ 740237263Snp struct adapter *sc = tod->tod_softc; 741237263Snp#ifdef INVARIANTS 742237263Snp struct inpcb *inp = tp->t_inpcb; 743237263Snp#endif 744237263Snp struct toepcb *toep = tp->t_toe; 745237263Snp 746237263Snp INP_WLOCK_ASSERT(inp); 747237263Snp KASSERT((inp->inp_flags & INP_DROPPED) == 0, 748237263Snp ("%s: inp %p dropped.", __func__, inp)); 749237263Snp KASSERT(toep != NULL, ("%s: toep is NULL", __func__)); 750237263Snp 751255411Snp t4_push_frames(sc, toep, 0); 752237263Snp 753237263Snp return (0); 754237263Snp} 755237263Snp 756237263Snpint 757237263Snpt4_send_fin(struct toedev *tod, struct tcpcb *tp) 758237263Snp{ 759237263Snp struct adapter *sc = tod->tod_softc; 760237263Snp#ifdef INVARIANTS 761237263Snp struct inpcb *inp = tp->t_inpcb; 762237263Snp#endif 763237263Snp struct toepcb *toep = tp->t_toe; 764237263Snp 765237263Snp INP_WLOCK_ASSERT(inp); 766237263Snp KASSERT((inp->inp_flags & INP_DROPPED) == 0, 767237263Snp ("%s: inp %p dropped.", __func__, inp)); 768237263Snp KASSERT(toep != NULL, ("%s: toep is NULL", __func__)); 769237263Snp 770239514Snp toep->flags |= TPF_SEND_FIN; 771255411Snp if (tp->t_state >= TCPS_ESTABLISHED) 772255411Snp t4_push_frames(sc, toep, 0); 773237263Snp 774237263Snp return (0); 775237263Snp} 776237263Snp 777237263Snpint 778237263Snpt4_send_rst(struct toedev *tod, struct tcpcb *tp) 779237263Snp{ 780237263Snp struct adapter *sc = tod->tod_softc; 781237263Snp#if defined(INVARIANTS) 782237263Snp struct inpcb *inp = tp->t_inpcb; 783237263Snp#endif 784237263Snp struct toepcb *toep = tp->t_toe; 785237263Snp 786237263Snp INP_WLOCK_ASSERT(inp); 787237263Snp KASSERT((inp->inp_flags & INP_DROPPED) == 0, 788237263Snp ("%s: inp %p dropped.", __func__, inp)); 789237263Snp KASSERT(toep != NULL, ("%s: toep is NULL", __func__)); 790237263Snp 791237263Snp /* hmmmm */ 792239514Snp KASSERT(toep->flags & TPF_FLOWC_WR_SENT, 793237263Snp ("%s: flowc for tid %u [%s] not sent already", 794237263Snp __func__, toep->tid, tcpstates[tp->t_state])); 795237263Snp 796237263Snp send_reset(sc, toep, 0); 797237263Snp return (0); 798237263Snp} 799237263Snp 800237263Snp/* 801237263Snp * Peer has sent us a FIN. 802237263Snp */ 803237263Snpstatic int 804237263Snpdo_peer_close(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) 805237263Snp{ 806237263Snp struct adapter *sc = iq->adapter; 807237263Snp const struct cpl_peer_close *cpl = (const void *)(rss + 1); 808237263Snp unsigned int tid = GET_TID(cpl); 809237263Snp struct toepcb *toep = lookup_tid(sc, tid); 810237263Snp struct inpcb *inp = toep->inp; 811237263Snp struct tcpcb *tp = NULL; 812239344Snp struct socket *so; 813239344Snp struct sockbuf *sb; 814237263Snp#ifdef INVARIANTS 815237263Snp unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl))); 816237263Snp#endif 817237263Snp 818237263Snp KASSERT(opcode == CPL_PEER_CLOSE, 819237263Snp ("%s: unexpected opcode 0x%x", __func__, opcode)); 820237263Snp KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 821243680Snp 822243680Snp if (__predict_false(toep->flags & TPF_SYNQE)) { 823243680Snp#ifdef INVARIANTS 824243680Snp struct synq_entry *synqe = (void *)toep; 825243680Snp 826243680Snp INP_WLOCK(synqe->lctx->inp); 827243680Snp if (synqe->flags & TPF_SYNQE_HAS_L2TE) { 828243680Snp KASSERT(synqe->flags & TPF_ABORT_SHUTDOWN, 829243680Snp ("%s: listen socket closed but tid %u not aborted.", 830243680Snp __func__, tid)); 831243680Snp } else { 832243680Snp /* 833243680Snp * do_pass_accept_req is still running and will 834243680Snp * eventually take care of this tid. 835243680Snp */ 836243680Snp } 837243680Snp INP_WUNLOCK(synqe->lctx->inp); 838243680Snp#endif 839243680Snp CTR4(KTR_CXGBE, "%s: tid %u, synqe %p (0x%x)", __func__, tid, 840243680Snp toep, toep->flags); 841243680Snp return (0); 842243680Snp } 843243680Snp 844237263Snp KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__)); 845237263Snp 846237263Snp INP_INFO_WLOCK(&V_tcbinfo); 847237263Snp INP_WLOCK(inp); 848237263Snp tp = intotcpcb(inp); 849237263Snp 850237263Snp CTR5(KTR_CXGBE, "%s: tid %u (%s), toep_flags 0x%x, inp %p", __func__, 851237263Snp tid, tp ? tcpstates[tp->t_state] : "no tp", toep->flags, inp); 852237263Snp 853239514Snp if (toep->flags & TPF_ABORT_SHUTDOWN) 854237263Snp goto done; 855237263Snp 856239344Snp tp->rcv_nxt++; /* FIN */ 857239344Snp 858237263Snp so = inp->inp_socket; 859239344Snp sb = &so->so_rcv; 860239344Snp SOCKBUF_LOCK(sb); 861239344Snp if (__predict_false(toep->ddp_flags & (DDP_BUF0_ACTIVE | DDP_BUF1_ACTIVE))) { 862250218Snp m = get_ddp_mbuf(be32toh(cpl->rcv_nxt) - tp->rcv_nxt); 863239344Snp tp->rcv_nxt = be32toh(cpl->rcv_nxt); 864239344Snp toep->ddp_flags &= ~(DDP_BUF0_ACTIVE | DDP_BUF1_ACTIVE); 865239344Snp 866239344Snp KASSERT(toep->sb_cc >= sb->sb_cc, 867239344Snp ("%s: sb %p has more data (%d) than last time (%d).", 868239344Snp __func__, sb, sb->sb_cc, toep->sb_cc)); 869239344Snp toep->rx_credits += toep->sb_cc - sb->sb_cc; 870239344Snp#ifdef USE_DDP_RX_FLOW_CONTROL 871239344Snp toep->rx_credits -= m->m_len; /* adjust for F_RX_FC_DDP */ 872239344Snp#endif 873239344Snp sbappendstream_locked(sb, m); 874239344Snp toep->sb_cc = sb->sb_cc; 875239344Snp } 876239344Snp socantrcvmore_locked(so); /* unlocks the sockbuf */ 877239344Snp 878255005Snp if (toep->ulp_mode != ULP_MODE_RDMA) { 879255005Snp KASSERT(tp->rcv_nxt == be32toh(cpl->rcv_nxt), 880255005Snp ("%s: rcv_nxt mismatch: %u %u", __func__, tp->rcv_nxt, 881255005Snp be32toh(cpl->rcv_nxt))); 882255005Snp } 883237263Snp 884237263Snp switch (tp->t_state) { 885237263Snp case TCPS_SYN_RECEIVED: 886237263Snp tp->t_starttime = ticks; 887237263Snp /* FALLTHROUGH */ 888237263Snp 889237263Snp case TCPS_ESTABLISHED: 890237263Snp tp->t_state = TCPS_CLOSE_WAIT; 891237263Snp break; 892237263Snp 893237263Snp case TCPS_FIN_WAIT_1: 894237263Snp tp->t_state = TCPS_CLOSING; 895237263Snp break; 896237263Snp 897237263Snp case TCPS_FIN_WAIT_2: 898237263Snp tcp_twstart(tp); 899237263Snp INP_UNLOCK_ASSERT(inp); /* safe, we have a ref on the inp */ 900237263Snp INP_INFO_WUNLOCK(&V_tcbinfo); 901237263Snp 902237263Snp INP_WLOCK(inp); 903237263Snp final_cpl_received(toep); 904237263Snp return (0); 905237263Snp 906237263Snp default: 907237263Snp log(LOG_ERR, "%s: TID %u received CPL_PEER_CLOSE in state %d\n", 908237263Snp __func__, tid, tp->t_state); 909237263Snp } 910237263Snpdone: 911237263Snp INP_WUNLOCK(inp); 912237263Snp INP_INFO_WUNLOCK(&V_tcbinfo); 913237263Snp return (0); 914237263Snp} 915237263Snp 916237263Snp/* 917237263Snp * Peer has ACK'd our FIN. 918237263Snp */ 919237263Snpstatic int 920237263Snpdo_close_con_rpl(struct sge_iq *iq, const struct rss_header *rss, 921237263Snp struct mbuf *m) 922237263Snp{ 923237263Snp struct adapter *sc = iq->adapter; 924237263Snp const struct cpl_close_con_rpl *cpl = (const void *)(rss + 1); 925237263Snp unsigned int tid = GET_TID(cpl); 926237263Snp struct toepcb *toep = lookup_tid(sc, tid); 927237263Snp struct inpcb *inp = toep->inp; 928237263Snp struct tcpcb *tp = NULL; 929237263Snp struct socket *so = NULL; 930237263Snp#ifdef INVARIANTS 931237263Snp unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl))); 932237263Snp#endif 933237263Snp 934237263Snp KASSERT(opcode == CPL_CLOSE_CON_RPL, 935237263Snp ("%s: unexpected opcode 0x%x", __func__, opcode)); 936237263Snp KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 937237263Snp KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__)); 938237263Snp 939237263Snp INP_INFO_WLOCK(&V_tcbinfo); 940237263Snp INP_WLOCK(inp); 941237263Snp tp = intotcpcb(inp); 942237263Snp 943237263Snp CTR4(KTR_CXGBE, "%s: tid %u (%s), toep_flags 0x%x", 944237263Snp __func__, tid, tp ? tcpstates[tp->t_state] : "no tp", toep->flags); 945237263Snp 946239514Snp if (toep->flags & TPF_ABORT_SHUTDOWN) 947237263Snp goto done; 948237263Snp 949237263Snp so = inp->inp_socket; 950237263Snp tp->snd_una = be32toh(cpl->snd_nxt) - 1; /* exclude FIN */ 951237263Snp 952237263Snp switch (tp->t_state) { 953237263Snp case TCPS_CLOSING: /* see TCPS_FIN_WAIT_2 in do_peer_close too */ 954237263Snp tcp_twstart(tp); 955237263Snprelease: 956237263Snp INP_UNLOCK_ASSERT(inp); /* safe, we have a ref on the inp */ 957237263Snp INP_INFO_WUNLOCK(&V_tcbinfo); 958237263Snp 959237263Snp INP_WLOCK(inp); 960237263Snp final_cpl_received(toep); /* no more CPLs expected */ 961237263Snp 962237263Snp return (0); 963237263Snp case TCPS_LAST_ACK: 964237263Snp if (tcp_close(tp)) 965237263Snp INP_WUNLOCK(inp); 966237263Snp goto release; 967237263Snp 968237263Snp case TCPS_FIN_WAIT_1: 969237263Snp if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 970237263Snp soisdisconnected(so); 971237263Snp tp->t_state = TCPS_FIN_WAIT_2; 972237263Snp break; 973237263Snp 974237263Snp default: 975237263Snp log(LOG_ERR, 976237263Snp "%s: TID %u received CPL_CLOSE_CON_RPL in state %s\n", 977237263Snp __func__, tid, tcpstates[tp->t_state]); 978237263Snp } 979237263Snpdone: 980237263Snp INP_WUNLOCK(inp); 981237263Snp INP_INFO_WUNLOCK(&V_tcbinfo); 982237263Snp return (0); 983237263Snp} 984237263Snp 985237263Snpvoid 986237263Snpsend_abort_rpl(struct adapter *sc, struct sge_wrq *ofld_txq, int tid, 987237263Snp int rst_status) 988237263Snp{ 989237263Snp struct wrqe *wr; 990237263Snp struct cpl_abort_rpl *cpl; 991237263Snp 992237263Snp wr = alloc_wrqe(sizeof(*cpl), ofld_txq); 993237263Snp if (wr == NULL) { 994237263Snp /* XXX */ 995237263Snp panic("%s: allocation failure.", __func__); 996237263Snp } 997237263Snp cpl = wrtod(wr); 998237263Snp 999237263Snp INIT_TP_WR_MIT_CPL(cpl, CPL_ABORT_RPL, tid); 1000237263Snp cpl->cmd = rst_status; 1001237263Snp 1002237263Snp t4_wrq_tx(sc, wr); 1003237263Snp} 1004237263Snp 1005237263Snpstatic int 1006237263Snpabort_status_to_errno(struct tcpcb *tp, unsigned int abort_reason) 1007237263Snp{ 1008237263Snp switch (abort_reason) { 1009237263Snp case CPL_ERR_BAD_SYN: 1010237263Snp case CPL_ERR_CONN_RESET: 1011237263Snp return (tp->t_state == TCPS_CLOSE_WAIT ? EPIPE : ECONNRESET); 1012237263Snp case CPL_ERR_XMIT_TIMEDOUT: 1013237263Snp case CPL_ERR_PERSIST_TIMEDOUT: 1014237263Snp case CPL_ERR_FINWAIT2_TIMEDOUT: 1015237263Snp case CPL_ERR_KEEPALIVE_TIMEDOUT: 1016237263Snp return (ETIMEDOUT); 1017237263Snp default: 1018237263Snp return (EIO); 1019237263Snp } 1020237263Snp} 1021237263Snp 1022237263Snp/* 1023237263Snp * TCP RST from the peer, timeout, or some other such critical error. 1024237263Snp */ 1025237263Snpstatic int 1026237263Snpdo_abort_req(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) 1027237263Snp{ 1028237263Snp struct adapter *sc = iq->adapter; 1029237263Snp const struct cpl_abort_req_rss *cpl = (const void *)(rss + 1); 1030237263Snp unsigned int tid = GET_TID(cpl); 1031237263Snp struct toepcb *toep = lookup_tid(sc, tid); 1032237263Snp struct sge_wrq *ofld_txq = toep->ofld_txq; 1033237263Snp struct inpcb *inp; 1034237263Snp struct tcpcb *tp; 1035237263Snp#ifdef INVARIANTS 1036237263Snp unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl))); 1037237263Snp#endif 1038237263Snp 1039237263Snp KASSERT(opcode == CPL_ABORT_REQ_RSS, 1040237263Snp ("%s: unexpected opcode 0x%x", __func__, opcode)); 1041237263Snp KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 1042237263Snp 1043239514Snp if (toep->flags & TPF_SYNQE) 1044237263Snp return (do_abort_req_synqe(iq, rss, m)); 1045237263Snp 1046237263Snp KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__)); 1047237263Snp 1048245935Snp if (negative_advice(cpl->status)) { 1049237263Snp CTR4(KTR_CXGBE, "%s: negative advice %d for tid %d (0x%x)", 1050237263Snp __func__, cpl->status, tid, toep->flags); 1051237263Snp return (0); /* Ignore negative advice */ 1052237263Snp } 1053237263Snp 1054237263Snp inp = toep->inp; 1055237263Snp INP_INFO_WLOCK(&V_tcbinfo); /* for tcp_close */ 1056237263Snp INP_WLOCK(inp); 1057237263Snp 1058237263Snp tp = intotcpcb(inp); 1059237263Snp 1060237263Snp CTR6(KTR_CXGBE, 1061237263Snp "%s: tid %d (%s), toep_flags 0x%x, inp_flags 0x%x, status %d", 1062239528Snp __func__, tid, tp ? tcpstates[tp->t_state] : "no tp", toep->flags, 1063239528Snp inp->inp_flags, cpl->status); 1064237263Snp 1065237263Snp /* 1066237263Snp * If we'd initiated an abort earlier the reply to it is responsible for 1067237263Snp * cleaning up resources. Otherwise we tear everything down right here 1068237263Snp * right now. We owe the T4 a CPL_ABORT_RPL no matter what. 1069237263Snp */ 1070239514Snp if (toep->flags & TPF_ABORT_SHUTDOWN) { 1071237263Snp INP_WUNLOCK(inp); 1072237263Snp goto done; 1073237263Snp } 1074239514Snp toep->flags |= TPF_ABORT_SHUTDOWN; 1075237263Snp 1076242671Snp if ((inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) == 0) { 1077242671Snp struct socket *so = inp->inp_socket; 1078237263Snp 1079242671Snp if (so != NULL) 1080242671Snp so_error_set(so, abort_status_to_errno(tp, 1081242671Snp cpl->status)); 1082242671Snp tp = tcp_close(tp); 1083242671Snp if (tp == NULL) 1084242671Snp INP_WLOCK(inp); /* re-acquire */ 1085242671Snp } 1086242671Snp 1087237263Snp final_cpl_received(toep); 1088237263Snpdone: 1089237263Snp INP_INFO_WUNLOCK(&V_tcbinfo); 1090237263Snp send_abort_rpl(sc, ofld_txq, tid, CPL_ABORT_NO_RST); 1091237263Snp return (0); 1092237263Snp} 1093237263Snp 1094237263Snp/* 1095237263Snp * Reply to the CPL_ABORT_REQ (send_reset) 1096237263Snp */ 1097237263Snpstatic int 1098237263Snpdo_abort_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) 1099237263Snp{ 1100237263Snp struct adapter *sc = iq->adapter; 1101237263Snp const struct cpl_abort_rpl_rss *cpl = (const void *)(rss + 1); 1102237263Snp unsigned int tid = GET_TID(cpl); 1103237263Snp struct toepcb *toep = lookup_tid(sc, tid); 1104237263Snp struct inpcb *inp = toep->inp; 1105237263Snp#ifdef INVARIANTS 1106237263Snp unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl))); 1107237263Snp#endif 1108237263Snp 1109237263Snp KASSERT(opcode == CPL_ABORT_RPL_RSS, 1110237263Snp ("%s: unexpected opcode 0x%x", __func__, opcode)); 1111237263Snp KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 1112237263Snp 1113239514Snp if (toep->flags & TPF_SYNQE) 1114237263Snp return (do_abort_rpl_synqe(iq, rss, m)); 1115237263Snp 1116237263Snp KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__)); 1117237263Snp 1118237263Snp CTR5(KTR_CXGBE, "%s: tid %u, toep %p, inp %p, status %d", 1119237263Snp __func__, tid, toep, inp, cpl->status); 1120237263Snp 1121239514Snp KASSERT(toep->flags & TPF_ABORT_SHUTDOWN, 1122237263Snp ("%s: wasn't expecting abort reply", __func__)); 1123237263Snp 1124237263Snp INP_WLOCK(inp); 1125237263Snp final_cpl_received(toep); 1126237263Snp 1127237263Snp return (0); 1128237263Snp} 1129237263Snp 1130237263Snpstatic int 1131237263Snpdo_rx_data(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) 1132237263Snp{ 1133237263Snp struct adapter *sc = iq->adapter; 1134237263Snp const struct cpl_rx_data *cpl = mtod(m, const void *); 1135237263Snp unsigned int tid = GET_TID(cpl); 1136237263Snp struct toepcb *toep = lookup_tid(sc, tid); 1137237263Snp struct inpcb *inp = toep->inp; 1138237263Snp struct tcpcb *tp; 1139237263Snp struct socket *so; 1140239344Snp struct sockbuf *sb; 1141239344Snp int len; 1142243681Snp uint32_t ddp_placed = 0; 1143237263Snp 1144239514Snp if (__predict_false(toep->flags & TPF_SYNQE)) { 1145243680Snp#ifdef INVARIANTS 1146243680Snp struct synq_entry *synqe = (void *)toep; 1147243680Snp 1148243680Snp INP_WLOCK(synqe->lctx->inp); 1149243680Snp if (synqe->flags & TPF_SYNQE_HAS_L2TE) { 1150243680Snp KASSERT(synqe->flags & TPF_ABORT_SHUTDOWN, 1151243680Snp ("%s: listen socket closed but tid %u not aborted.", 1152243680Snp __func__, tid)); 1153243680Snp } else { 1154243680Snp /* 1155243680Snp * do_pass_accept_req is still running and will 1156243680Snp * eventually take care of this tid. 1157243680Snp */ 1158243680Snp } 1159243680Snp INP_WUNLOCK(synqe->lctx->inp); 1160243680Snp#endif 1161243680Snp CTR4(KTR_CXGBE, "%s: tid %u, synqe %p (0x%x)", __func__, tid, 1162243680Snp toep, toep->flags); 1163237263Snp m_freem(m); 1164237263Snp return (0); 1165237263Snp } 1166237263Snp 1167237263Snp KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__)); 1168237263Snp 1169237263Snp /* strip off CPL header */ 1170237263Snp m_adj(m, sizeof(*cpl)); 1171239344Snp len = m->m_pkthdr.len; 1172237263Snp 1173237263Snp INP_WLOCK(inp); 1174237263Snp if (inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) { 1175237263Snp CTR4(KTR_CXGBE, "%s: tid %u, rx (%d bytes), inp_flags 0x%x", 1176239344Snp __func__, tid, len, inp->inp_flags); 1177237263Snp INP_WUNLOCK(inp); 1178237263Snp m_freem(m); 1179237263Snp return (0); 1180237263Snp } 1181237263Snp 1182237263Snp tp = intotcpcb(inp); 1183237263Snp 1184243681Snp if (__predict_false(tp->rcv_nxt != be32toh(cpl->seq))) 1185243681Snp ddp_placed = be32toh(cpl->seq) - tp->rcv_nxt; 1186237263Snp 1187239344Snp tp->rcv_nxt += len; 1188239344Snp KASSERT(tp->rcv_wnd >= len, ("%s: negative window size", __func__)); 1189239344Snp tp->rcv_wnd -= len; 1190237263Snp tp->t_rcvtime = ticks; 1191237263Snp 1192237263Snp so = inp_inpcbtosocket(inp); 1193239344Snp sb = &so->so_rcv; 1194239344Snp SOCKBUF_LOCK(sb); 1195237263Snp 1196239344Snp if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) { 1197237263Snp CTR3(KTR_CXGBE, "%s: tid %u, excess rx (%d bytes)", 1198239344Snp __func__, tid, len); 1199237263Snp m_freem(m); 1200239344Snp SOCKBUF_UNLOCK(sb); 1201237263Snp INP_WUNLOCK(inp); 1202237263Snp 1203237263Snp INP_INFO_WLOCK(&V_tcbinfo); 1204237263Snp INP_WLOCK(inp); 1205237263Snp tp = tcp_drop(tp, ECONNRESET); 1206237263Snp if (tp) 1207237263Snp INP_WUNLOCK(inp); 1208237263Snp INP_INFO_WUNLOCK(&V_tcbinfo); 1209237263Snp 1210237263Snp return (0); 1211237263Snp } 1212237263Snp 1213237263Snp /* receive buffer autosize */ 1214239344Snp if (sb->sb_flags & SB_AUTOSIZE && 1215237263Snp V_tcp_do_autorcvbuf && 1216239344Snp sb->sb_hiwat < V_tcp_autorcvbuf_max && 1217239344Snp len > (sbspace(sb) / 8 * 7)) { 1218239344Snp unsigned int hiwat = sb->sb_hiwat; 1219237263Snp unsigned int newsize = min(hiwat + V_tcp_autorcvbuf_inc, 1220237263Snp V_tcp_autorcvbuf_max); 1221237263Snp 1222239344Snp if (!sbreserve_locked(sb, newsize, so, NULL)) 1223239344Snp sb->sb_flags &= ~SB_AUTOSIZE; 1224237263Snp else 1225237263Snp toep->rx_credits += newsize - hiwat; 1226237263Snp } 1227239344Snp 1228239344Snp if (toep->ulp_mode == ULP_MODE_TCPDDP) { 1229239344Snp int changed = !(toep->ddp_flags & DDP_ON) ^ cpl->ddp_off; 1230239344Snp 1231239344Snp if (changed) { 1232243681Snp if (toep->ddp_flags & DDP_SC_REQ) 1233243681Snp toep->ddp_flags ^= DDP_ON | DDP_SC_REQ; 1234243681Snp else { 1235243681Snp KASSERT(cpl->ddp_off == 1, 1236243681Snp ("%s: DDP switched on by itself.", 1237243681Snp __func__)); 1238243681Snp 1239243681Snp /* Fell out of DDP mode */ 1240243681Snp toep->ddp_flags &= ~(DDP_ON | DDP_BUF0_ACTIVE | 1241243681Snp DDP_BUF1_ACTIVE); 1242243681Snp 1243243681Snp if (ddp_placed) 1244243681Snp insert_ddp_data(toep, ddp_placed); 1245239344Snp } 1246239344Snp } 1247239344Snp 1248239344Snp if ((toep->ddp_flags & DDP_OK) == 0 && 1249239344Snp time_uptime >= toep->ddp_disabled + DDP_RETRY_WAIT) { 1250239344Snp toep->ddp_score = DDP_LOW_SCORE; 1251239344Snp toep->ddp_flags |= DDP_OK; 1252239344Snp CTR3(KTR_CXGBE, "%s: tid %u DDP_OK @ %u", 1253239344Snp __func__, tid, time_uptime); 1254239344Snp } 1255239344Snp 1256239344Snp if (toep->ddp_flags & DDP_ON) { 1257239344Snp 1258239344Snp /* 1259239344Snp * CPL_RX_DATA with DDP on can only be an indicate. Ask 1260239344Snp * soreceive to post a buffer or disable DDP. The 1261239344Snp * payload that arrived in this indicate is appended to 1262239344Snp * the socket buffer as usual. 1263239344Snp */ 1264239344Snp 1265239344Snp#if 0 1266239344Snp CTR5(KTR_CXGBE, 1267239344Snp "%s: tid %u (0x%x) DDP indicate (seq 0x%x, len %d)", 1268239344Snp __func__, tid, toep->flags, be32toh(cpl->seq), len); 1269239344Snp#endif 1270239344Snp sb->sb_flags |= SB_DDP_INDICATE; 1271239344Snp } else if ((toep->ddp_flags & (DDP_OK|DDP_SC_REQ)) == DDP_OK && 1272239344Snp tp->rcv_wnd > DDP_RSVD_WIN && len >= sc->tt.ddp_thres) { 1273239344Snp 1274239344Snp /* 1275239344Snp * DDP allowed but isn't on (and a request to switch it 1276239344Snp * on isn't pending either), and conditions are ripe for 1277239344Snp * it to work. Switch it on. 1278239344Snp */ 1279239344Snp 1280239344Snp enable_ddp(sc, toep); 1281239344Snp } 1282239344Snp } 1283239344Snp 1284239344Snp KASSERT(toep->sb_cc >= sb->sb_cc, 1285239344Snp ("%s: sb %p has more data (%d) than last time (%d).", 1286239344Snp __func__, sb, sb->sb_cc, toep->sb_cc)); 1287239344Snp toep->rx_credits += toep->sb_cc - sb->sb_cc; 1288239344Snp sbappendstream_locked(sb, m); 1289239344Snp toep->sb_cc = sb->sb_cc; 1290237263Snp sorwakeup_locked(so); 1291239344Snp SOCKBUF_UNLOCK_ASSERT(sb); 1292237263Snp 1293237263Snp INP_WUNLOCK(inp); 1294237263Snp return (0); 1295237263Snp} 1296237263Snp 1297237263Snp#define S_CPL_FW4_ACK_OPCODE 24 1298237263Snp#define M_CPL_FW4_ACK_OPCODE 0xff 1299237263Snp#define V_CPL_FW4_ACK_OPCODE(x) ((x) << S_CPL_FW4_ACK_OPCODE) 1300237263Snp#define G_CPL_FW4_ACK_OPCODE(x) \ 1301237263Snp (((x) >> S_CPL_FW4_ACK_OPCODE) & M_CPL_FW4_ACK_OPCODE) 1302237263Snp 1303237263Snp#define S_CPL_FW4_ACK_FLOWID 0 1304237263Snp#define M_CPL_FW4_ACK_FLOWID 0xffffff 1305237263Snp#define V_CPL_FW4_ACK_FLOWID(x) ((x) << S_CPL_FW4_ACK_FLOWID) 1306237263Snp#define G_CPL_FW4_ACK_FLOWID(x) \ 1307237263Snp (((x) >> S_CPL_FW4_ACK_FLOWID) & M_CPL_FW4_ACK_FLOWID) 1308237263Snp 1309237263Snp#define S_CPL_FW4_ACK_CR 24 1310237263Snp#define M_CPL_FW4_ACK_CR 0xff 1311237263Snp#define V_CPL_FW4_ACK_CR(x) ((x) << S_CPL_FW4_ACK_CR) 1312237263Snp#define G_CPL_FW4_ACK_CR(x) (((x) >> S_CPL_FW4_ACK_CR) & M_CPL_FW4_ACK_CR) 1313237263Snp 1314237263Snp#define S_CPL_FW4_ACK_SEQVAL 0 1315237263Snp#define M_CPL_FW4_ACK_SEQVAL 0x1 1316237263Snp#define V_CPL_FW4_ACK_SEQVAL(x) ((x) << S_CPL_FW4_ACK_SEQVAL) 1317237263Snp#define G_CPL_FW4_ACK_SEQVAL(x) \ 1318237263Snp (((x) >> S_CPL_FW4_ACK_SEQVAL) & M_CPL_FW4_ACK_SEQVAL) 1319237263Snp#define F_CPL_FW4_ACK_SEQVAL V_CPL_FW4_ACK_SEQVAL(1U) 1320237263Snp 1321237263Snpstatic int 1322237263Snpdo_fw4_ack(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) 1323237263Snp{ 1324237263Snp struct adapter *sc = iq->adapter; 1325237263Snp const struct cpl_fw4_ack *cpl = (const void *)(rss + 1); 1326237263Snp unsigned int tid = G_CPL_FW4_ACK_FLOWID(be32toh(OPCODE_TID(cpl))); 1327237263Snp struct toepcb *toep = lookup_tid(sc, tid); 1328237263Snp struct inpcb *inp; 1329237263Snp struct tcpcb *tp; 1330237263Snp struct socket *so; 1331237263Snp uint8_t credits = cpl->credits; 1332237263Snp struct ofld_tx_sdesc *txsd; 1333237263Snp int plen; 1334237263Snp#ifdef INVARIANTS 1335237263Snp unsigned int opcode = G_CPL_FW4_ACK_OPCODE(be32toh(OPCODE_TID(cpl))); 1336237263Snp#endif 1337237263Snp 1338237263Snp /* 1339237263Snp * Very unusual case: we'd sent a flowc + abort_req for a synq entry and 1340237263Snp * now this comes back carrying the credits for the flowc. 1341237263Snp */ 1342239514Snp if (__predict_false(toep->flags & TPF_SYNQE)) { 1343239514Snp KASSERT(toep->flags & TPF_ABORT_SHUTDOWN, 1344237263Snp ("%s: credits for a synq entry %p", __func__, toep)); 1345237263Snp return (0); 1346237263Snp } 1347237263Snp 1348237263Snp inp = toep->inp; 1349237263Snp 1350237263Snp KASSERT(opcode == CPL_FW4_ACK, 1351237263Snp ("%s: unexpected opcode 0x%x", __func__, opcode)); 1352237263Snp KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 1353237263Snp KASSERT(toep->tid == tid, ("%s: toep tid mismatch", __func__)); 1354237263Snp 1355237263Snp INP_WLOCK(inp); 1356237263Snp 1357239514Snp if (__predict_false(toep->flags & TPF_ABORT_SHUTDOWN)) { 1358237263Snp INP_WUNLOCK(inp); 1359237263Snp return (0); 1360237263Snp } 1361237263Snp 1362237263Snp KASSERT((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) == 0, 1363237263Snp ("%s: inp_flags 0x%x", __func__, inp->inp_flags)); 1364237263Snp 1365237263Snp tp = intotcpcb(inp); 1366237263Snp 1367237436Snp if (cpl->flags & CPL_FW4_ACK_FLAGS_SEQVAL) { 1368237263Snp tcp_seq snd_una = be32toh(cpl->snd_una); 1369237263Snp 1370237263Snp#ifdef INVARIANTS 1371237263Snp if (__predict_false(SEQ_LT(snd_una, tp->snd_una))) { 1372237263Snp log(LOG_ERR, 1373237263Snp "%s: unexpected seq# %x for TID %u, snd_una %x\n", 1374237263Snp __func__, snd_una, toep->tid, tp->snd_una); 1375237263Snp } 1376237263Snp#endif 1377237263Snp 1378237263Snp if (tp->snd_una != snd_una) { 1379237263Snp tp->snd_una = snd_una; 1380237263Snp tp->ts_recent_age = tcp_ts_getticks(); 1381237263Snp } 1382237263Snp } 1383237263Snp 1384237263Snp so = inp->inp_socket; 1385237263Snp txsd = &toep->txsd[toep->txsd_cidx]; 1386237263Snp plen = 0; 1387237263Snp while (credits) { 1388237263Snp KASSERT(credits >= txsd->tx_credits, 1389237263Snp ("%s: too many (or partial) credits", __func__)); 1390237263Snp credits -= txsd->tx_credits; 1391237263Snp toep->tx_credits += txsd->tx_credits; 1392237263Snp plen += txsd->plen; 1393237263Snp txsd++; 1394237263Snp toep->txsd_avail++; 1395237263Snp KASSERT(toep->txsd_avail <= toep->txsd_total, 1396237263Snp ("%s: txsd avail > total", __func__)); 1397237263Snp if (__predict_false(++toep->txsd_cidx == toep->txsd_total)) { 1398237263Snp txsd = &toep->txsd[0]; 1399237263Snp toep->txsd_cidx = 0; 1400237263Snp } 1401237263Snp } 1402237263Snp 1403255411Snp if (toep->tx_credits == toep->tx_total) { 1404255411Snp toep->tx_nocompl = 0; 1405255411Snp toep->plen_nocompl = 0; 1406255411Snp } 1407255411Snp 1408255411Snp if (toep->flags & TPF_TX_SUSPENDED && 1409255411Snp toep->tx_credits >= toep->tx_total / 4) { 1410255411Snp toep->flags &= ~TPF_TX_SUSPENDED; 1411255411Snp t4_push_frames(sc, toep, plen); 1412255411Snp } else if (plen > 0) { 1413237263Snp struct sockbuf *sb = &so->so_snd; 1414237263Snp 1415237263Snp SOCKBUF_LOCK(sb); 1416237263Snp sbdrop_locked(sb, plen); 1417237263Snp sowwakeup_locked(so); 1418237263Snp SOCKBUF_UNLOCK_ASSERT(sb); 1419237263Snp } 1420237263Snp 1421237263Snp INP_WUNLOCK(inp); 1422237263Snp 1423237263Snp return (0); 1424237263Snp} 1425237263Snp 1426239338Snpstatic int 1427239338Snpdo_set_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) 1428239338Snp{ 1429239338Snp struct adapter *sc = iq->adapter; 1430239338Snp const struct cpl_set_tcb_rpl *cpl = (const void *)(rss + 1); 1431239338Snp unsigned int tid = GET_TID(cpl); 1432239338Snp#ifdef INVARIANTS 1433239338Snp unsigned int opcode = G_CPL_OPCODE(be32toh(OPCODE_TID(cpl))); 1434239338Snp#endif 1435239338Snp 1436239338Snp KASSERT(opcode == CPL_SET_TCB_RPL, 1437239338Snp ("%s: unexpected opcode 0x%x", __func__, opcode)); 1438239338Snp KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 1439239338Snp 1440239338Snp if (tid >= sc->tids.ftid_base && 1441239338Snp tid < sc->tids.ftid_base + sc->tids.nftids) 1442239338Snp return (t4_filter_rpl(iq, rss, m)); /* TCB is a filter */ 1443239338Snp 1444239338Snp CXGBE_UNIMPLEMENTED(__func__); 1445239338Snp} 1446239338Snp 1447237263Snpvoid 1448251638Snpt4_set_tcb_field(struct adapter *sc, struct toepcb *toep, int ctrl, 1449251638Snp uint16_t word, uint64_t mask, uint64_t val) 1450239338Snp{ 1451239338Snp struct wrqe *wr; 1452239338Snp struct cpl_set_tcb_field *req; 1453239338Snp 1454251638Snp wr = alloc_wrqe(sizeof(*req), ctrl ? toep->ctrlq : toep->ofld_txq); 1455239338Snp if (wr == NULL) { 1456239338Snp /* XXX */ 1457239338Snp panic("%s: allocation failure.", __func__); 1458239338Snp } 1459239338Snp req = wrtod(wr); 1460239338Snp 1461239338Snp INIT_TP_WR_MIT_CPL(req, CPL_SET_TCB_FIELD, toep->tid); 1462239338Snp req->reply_ctrl = htobe16(V_NO_REPLY(1) | 1463239338Snp V_QUEUENO(toep->ofld_rxq->iq.abs_id)); 1464239338Snp req->word_cookie = htobe16(V_WORD(word) | V_COOKIE(0)); 1465239338Snp req->mask = htobe64(mask); 1466239338Snp req->val = htobe64(val); 1467239338Snp 1468239338Snp t4_wrq_tx(sc, wr); 1469239338Snp} 1470239338Snp 1471239338Snpvoid 1472237263Snpt4_init_cpl_io_handlers(struct adapter *sc) 1473237263Snp{ 1474237263Snp 1475237263Snp t4_register_cpl_handler(sc, CPL_PEER_CLOSE, do_peer_close); 1476237263Snp t4_register_cpl_handler(sc, CPL_CLOSE_CON_RPL, do_close_con_rpl); 1477237263Snp t4_register_cpl_handler(sc, CPL_ABORT_REQ_RSS, do_abort_req); 1478237263Snp t4_register_cpl_handler(sc, CPL_ABORT_RPL_RSS, do_abort_rpl); 1479237263Snp t4_register_cpl_handler(sc, CPL_RX_DATA, do_rx_data); 1480237263Snp t4_register_cpl_handler(sc, CPL_FW4_ACK, do_fw4_ack); 1481239338Snp t4_register_cpl_handler(sc, CPL_SET_TCB_RPL, do_set_tcb_rpl); 1482237263Snp} 1483239338Snp 1484239338Snpvoid 1485239338Snpt4_uninit_cpl_io_handlers(struct adapter *sc) 1486239338Snp{ 1487239338Snp 1488239338Snp t4_register_cpl_handler(sc, CPL_SET_TCB_RPL, t4_filter_rpl); 1489239338Snp} 1490237263Snp#endif 1491