linux/drivers/infiniband/hw/cxgb3/iwch_cm.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
   3 *
   4 * This software is available to you under a choice of one of two
   5 * licenses.  You may choose to be licensed under the terms of the GNU
   6 * General Public License (GPL) Version 2, available from the file
   7 * COPYING in the main directory of this source tree, or the
   8 * OpenIB.org BSD license below:
   9 *
  10 *     Redistribution and use in source and binary forms, with or
  11 *     without modification, are permitted provided that the following
  12 *     conditions are met:
  13 *
  14 *      - Redistributions of source code must retain the above
  15 *        copyright notice, this list of conditions and the following
  16 *        disclaimer.
  17 *
  18 *      - Redistributions in binary form must reproduce the above
  19 *        copyright notice, this list of conditions and the following
  20 *        disclaimer in the documentation and/or other materials
  21 *        provided with the distribution.
  22 *
  23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30 * SOFTWARE.
  31 */
  32#include <linux/module.h>
  33#include <linux/list.h>
  34#include <linux/slab.h>
  35#include <linux/workqueue.h>
  36#include <linux/skbuff.h>
  37#include <linux/timer.h>
  38#include <linux/notifier.h>
  39#include <linux/inetdevice.h>
  40
  41#include <net/neighbour.h>
  42#include <net/netevent.h>
  43#include <net/route.h>
  44
  45#include "tcb.h"
  46#include "cxgb3_offload.h"
  47#include "iwch.h"
  48#include "iwch_provider.h"
  49#include "iwch_cm.h"
  50
  51static char *states[] = {
  52        "idle",
  53        "listen",
  54        "connecting",
  55        "mpa_wait_req",
  56        "mpa_req_sent",
  57        "mpa_req_rcvd",
  58        "mpa_rep_sent",
  59        "fpdu_mode",
  60        "aborting",
  61        "closing",
  62        "moribund",
  63        "dead",
  64        NULL,
  65};
  66
  67int peer2peer = 0;
  68module_param(peer2peer, int, 0644);
  69MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
  70
  71static int ep_timeout_secs = 60;
  72module_param(ep_timeout_secs, int, 0644);
  73MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
  74                                   "in seconds (default=60)");
  75
  76static int mpa_rev = 1;
  77module_param(mpa_rev, int, 0644);
  78MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
  79                 "1 is spec compliant. (default=1)");
  80
  81static int markers_enabled = 0;
  82module_param(markers_enabled, int, 0644);
  83MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
  84
  85static int crc_enabled = 1;
  86module_param(crc_enabled, int, 0644);
  87MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
  88
  89static int rcv_win = 256 * 1024;
  90module_param(rcv_win, int, 0644);
  91MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256)");
  92
  93static int snd_win = 32 * 1024;
  94module_param(snd_win, int, 0644);
  95MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=32KB)");
  96
  97static unsigned int nocong = 0;
  98module_param(nocong, uint, 0644);
  99MODULE_PARM_DESC(nocong, "Turn off congestion control (default=0)");
 100
 101static unsigned int cong_flavor = 1;
 102module_param(cong_flavor, uint, 0644);
 103MODULE_PARM_DESC(cong_flavor, "TCP Congestion control flavor (default=1)");
 104
 105static struct workqueue_struct *workq;
 106
 107static struct sk_buff_head rxq;
 108
 109static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
 110static void ep_timeout(unsigned long arg);
 111static void connect_reply_upcall(struct iwch_ep *ep, int status);
 112
 113static void start_ep_timer(struct iwch_ep *ep)
 114{
 115        PDBG("%s ep %p\n", __func__, ep);
 116        if (timer_pending(&ep->timer)) {
 117                PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
 118                del_timer_sync(&ep->timer);
 119        } else
 120                get_ep(&ep->com);
 121        ep->timer.expires = jiffies + ep_timeout_secs * HZ;
 122        ep->timer.data = (unsigned long)ep;
 123        ep->timer.function = ep_timeout;
 124        add_timer(&ep->timer);
 125}
 126
 127static void stop_ep_timer(struct iwch_ep *ep)
 128{
 129        PDBG("%s ep %p\n", __func__, ep);
 130        if (!timer_pending(&ep->timer)) {
 131                printk(KERN_ERR "%s timer stopped when its not running!  ep %p state %u\n",
 132                        __func__, ep, ep->com.state);
 133                WARN_ON(1);
 134                return;
 135        }
 136        del_timer_sync(&ep->timer);
 137        put_ep(&ep->com);
 138}
 139
 140static int iwch_l2t_send(struct t3cdev *tdev, struct sk_buff *skb, struct l2t_entry *l2e)
 141{
 142        int     error = 0;
 143        struct cxio_rdev *rdev;
 144
 145        rdev = (struct cxio_rdev *)tdev->ulp;
 146        if (cxio_fatal_error(rdev)) {
 147                kfree_skb(skb);
 148                return -EIO;
 149        }
 150        error = l2t_send(tdev, skb, l2e);
 151        if (error < 0)
 152                kfree_skb(skb);
 153        return error;
 154}
 155
 156int iwch_cxgb3_ofld_send(struct t3cdev *tdev, struct sk_buff *skb)
 157{
 158        int     error = 0;
 159        struct cxio_rdev *rdev;
 160
 161        rdev = (struct cxio_rdev *)tdev->ulp;
 162        if (cxio_fatal_error(rdev)) {
 163                kfree_skb(skb);
 164                return -EIO;
 165        }
 166        error = cxgb3_ofld_send(tdev, skb);
 167        if (error < 0)
 168                kfree_skb(skb);
 169        return error;
 170}
 171
 172static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
 173{
 174        struct cpl_tid_release *req;
 175
 176        skb = get_skb(skb, sizeof *req, GFP_KERNEL);
 177        if (!skb)
 178                return;
 179        req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
 180        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 181        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
 182        skb->priority = CPL_PRIORITY_SETUP;
 183        iwch_cxgb3_ofld_send(tdev, skb);
 184        return;
 185}
 186
 187int iwch_quiesce_tid(struct iwch_ep *ep)
 188{
 189        struct cpl_set_tcb_field *req;
 190        struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
 191
 192        if (!skb)
 193                return -ENOMEM;
 194        req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
 195        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 196        req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
 197        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
 198        req->reply = 0;
 199        req->cpu_idx = 0;
 200        req->word = htons(W_TCB_RX_QUIESCE);
 201        req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
 202        req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
 203
 204        skb->priority = CPL_PRIORITY_DATA;
 205        return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
 206}
 207
 208int iwch_resume_tid(struct iwch_ep *ep)
 209{
 210        struct cpl_set_tcb_field *req;
 211        struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
 212
 213        if (!skb)
 214                return -ENOMEM;
 215        req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
 216        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 217        req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
 218        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
 219        req->reply = 0;
 220        req->cpu_idx = 0;
 221        req->word = htons(W_TCB_RX_QUIESCE);
 222        req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
 223        req->val = 0;
 224
 225        skb->priority = CPL_PRIORITY_DATA;
 226        return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
 227}
 228
 229static void set_emss(struct iwch_ep *ep, u16 opt)
 230{
 231        PDBG("%s ep %p opt %u\n", __func__, ep, opt);
 232        ep->emss = T3C_DATA(ep->com.tdev)->mtus[G_TCPOPT_MSS(opt)] - 40;
 233        if (G_TCPOPT_TSTAMP(opt))
 234                ep->emss -= 12;
 235        if (ep->emss < 128)
 236                ep->emss = 128;
 237        PDBG("emss=%d\n", ep->emss);
 238}
 239
 240static enum iwch_ep_state state_read(struct iwch_ep_common *epc)
 241{
 242        unsigned long flags;
 243        enum iwch_ep_state state;
 244
 245        spin_lock_irqsave(&epc->lock, flags);
 246        state = epc->state;
 247        spin_unlock_irqrestore(&epc->lock, flags);
 248        return state;
 249}
 250
 251static void __state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
 252{
 253        epc->state = new;
 254}
 255
 256static void state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
 257{
 258        unsigned long flags;
 259
 260        spin_lock_irqsave(&epc->lock, flags);
 261        PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
 262        __state_set(epc, new);
 263        spin_unlock_irqrestore(&epc->lock, flags);
 264        return;
 265}
 266
 267static void *alloc_ep(int size, gfp_t gfp)
 268{
 269        struct iwch_ep_common *epc;
 270
 271        epc = kzalloc(size, gfp);
 272        if (epc) {
 273                kref_init(&epc->kref);
 274                spin_lock_init(&epc->lock);
 275                init_waitqueue_head(&epc->waitq);
 276        }
 277        PDBG("%s alloc ep %p\n", __func__, epc);
 278        return epc;
 279}
 280
 281void __free_ep(struct kref *kref)
 282{
 283        struct iwch_ep *ep;
 284        ep = container_of(container_of(kref, struct iwch_ep_common, kref),
 285                          struct iwch_ep, com);
 286        PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
 287        if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
 288                cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
 289                dst_release(ep->dst);
 290                l2t_release(L2DATA(ep->com.tdev), ep->l2t);
 291        }
 292        kfree(ep);
 293}
 294
 295static void release_ep_resources(struct iwch_ep *ep)
 296{
 297        PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
 298        set_bit(RELEASE_RESOURCES, &ep->com.flags);
 299        put_ep(&ep->com);
 300}
 301
 302static int status2errno(int status)
 303{
 304        switch (status) {
 305        case CPL_ERR_NONE:
 306                return 0;
 307        case CPL_ERR_CONN_RESET:
 308                return -ECONNRESET;
 309        case CPL_ERR_ARP_MISS:
 310                return -EHOSTUNREACH;
 311        case CPL_ERR_CONN_TIMEDOUT:
 312                return -ETIMEDOUT;
 313        case CPL_ERR_TCAM_FULL:
 314                return -ENOMEM;
 315        case CPL_ERR_CONN_EXIST:
 316                return -EADDRINUSE;
 317        default:
 318                return -EIO;
 319        }
 320}
 321
 322/*
 323 * Try and reuse skbs already allocated...
 324 */
 325static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
 326{
 327        if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
 328                skb_trim(skb, 0);
 329                skb_get(skb);
 330        } else {
 331                skb = alloc_skb(len, gfp);
 332        }
 333        return skb;
 334}
 335
 336static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
 337                                 __be32 peer_ip, __be16 local_port,
 338                                 __be16 peer_port, u8 tos)
 339{
 340        struct rtable *rt;
 341        struct flowi fl = {
 342                .oif = 0,
 343                .nl_u = {
 344                         .ip4_u = {
 345                                   .daddr = peer_ip,
 346                                   .saddr = local_ip,
 347                                   .tos = tos}
 348                         },
 349                .proto = IPPROTO_TCP,
 350                .uli_u = {
 351                          .ports = {
 352                                    .sport = local_port,
 353                                    .dport = peer_port}
 354                          }
 355        };
 356
 357        if (ip_route_output_flow(&init_net, &rt, &fl, NULL, 0))
 358                return NULL;
 359        return rt;
 360}
 361
 362static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
 363{
 364        int i = 0;
 365
 366        while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
 367                ++i;
 368        return i;
 369}
 370
 371static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
 372{
 373        PDBG("%s t3cdev %p\n", __func__, dev);
 374        kfree_skb(skb);
 375}
 376
 377/*
 378 * Handle an ARP failure for an active open.
 379 */
 380static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
 381{
 382        printk(KERN_ERR MOD "ARP failure duing connect\n");
 383        kfree_skb(skb);
 384}
 385
 386/*
 387 * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
 388 * and send it along.
 389 */
 390static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
 391{
 392        struct cpl_abort_req *req = cplhdr(skb);
 393
 394        PDBG("%s t3cdev %p\n", __func__, dev);
 395        req->cmd = CPL_ABORT_NO_RST;
 396        iwch_cxgb3_ofld_send(dev, skb);
 397}
 398
 399static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
 400{
 401        struct cpl_close_con_req *req;
 402        struct sk_buff *skb;
 403
 404        PDBG("%s ep %p\n", __func__, ep);
 405        skb = get_skb(NULL, sizeof(*req), gfp);
 406        if (!skb) {
 407                printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
 408                return -ENOMEM;
 409        }
 410        skb->priority = CPL_PRIORITY_DATA;
 411        set_arp_failure_handler(skb, arp_failure_discard);
 412        req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
 413        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
 414        req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
 415        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
 416        return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 417}
 418
 419static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
 420{
 421        struct cpl_abort_req *req;
 422
 423        PDBG("%s ep %p\n", __func__, ep);
 424        skb = get_skb(skb, sizeof(*req), gfp);
 425        if (!skb) {
 426                printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
 427                       __func__);
 428                return -ENOMEM;
 429        }
 430        skb->priority = CPL_PRIORITY_DATA;
 431        set_arp_failure_handler(skb, abort_arp_failure);
 432        req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
 433        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
 434        req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
 435        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
 436        req->cmd = CPL_ABORT_SEND_RST;
 437        return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 438}
 439
 440static int send_connect(struct iwch_ep *ep)
 441{
 442        struct cpl_act_open_req *req;
 443        struct sk_buff *skb;
 444        u32 opt0h, opt0l, opt2;
 445        unsigned int mtu_idx;
 446        int wscale;
 447
 448        PDBG("%s ep %p\n", __func__, ep);
 449
 450        skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
 451        if (!skb) {
 452                printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
 453                       __func__);
 454                return -ENOMEM;
 455        }
 456        mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
 457        wscale = compute_wscale(rcv_win);
 458        opt0h = V_NAGLE(0) |
 459            V_NO_CONG(nocong) |
 460            V_KEEP_ALIVE(1) |
 461            F_TCAM_BYPASS |
 462            V_WND_SCALE(wscale) |
 463            V_MSS_IDX(mtu_idx) |
 464            V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
 465        opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
 466        opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
 467               V_CONG_CONTROL_FLAVOR(cong_flavor);
 468        skb->priority = CPL_PRIORITY_SETUP;
 469        set_arp_failure_handler(skb, act_open_req_arp_failure);
 470
 471        req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
 472        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 473        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
 474        req->local_port = ep->com.local_addr.sin_port;
 475        req->peer_port = ep->com.remote_addr.sin_port;
 476        req->local_ip = ep->com.local_addr.sin_addr.s_addr;
 477        req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
 478        req->opt0h = htonl(opt0h);
 479        req->opt0l = htonl(opt0l);
 480        req->params = 0;
 481        req->opt2 = htonl(opt2);
 482        return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 483}
 484
 485static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
 486{
 487        int mpalen;
 488        struct tx_data_wr *req;
 489        struct mpa_message *mpa;
 490        int len;
 491
 492        PDBG("%s ep %p pd_len %d\n", __func__, ep, ep->plen);
 493
 494        BUG_ON(skb_cloned(skb));
 495
 496        mpalen = sizeof(*mpa) + ep->plen;
 497        if (skb->data + mpalen + sizeof(*req) > skb_end_pointer(skb)) {
 498                kfree_skb(skb);
 499                skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
 500                if (!skb) {
 501                        connect_reply_upcall(ep, -ENOMEM);
 502                        return;
 503                }
 504        }
 505        skb_trim(skb, 0);
 506        skb_reserve(skb, sizeof(*req));
 507        skb_put(skb, mpalen);
 508        skb->priority = CPL_PRIORITY_DATA;
 509        mpa = (struct mpa_message *) skb->data;
 510        memset(mpa, 0, sizeof(*mpa));
 511        memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
 512        mpa->flags = (crc_enabled ? MPA_CRC : 0) |
 513                     (markers_enabled ? MPA_MARKERS : 0);
 514        mpa->private_data_size = htons(ep->plen);
 515        mpa->revision = mpa_rev;
 516
 517        if (ep->plen)
 518                memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
 519
 520        /*
 521         * Reference the mpa skb.  This ensures the data area
 522         * will remain in memory until the hw acks the tx.
 523         * Function tx_ack() will deref it.
 524         */
 525        skb_get(skb);
 526        set_arp_failure_handler(skb, arp_failure_discard);
 527        skb_reset_transport_header(skb);
 528        len = skb->len;
 529        req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
 530        req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
 531        req->wr_lo = htonl(V_WR_TID(ep->hwtid));
 532        req->len = htonl(len);
 533        req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
 534                           V_TX_SNDBUF(snd_win>>15));
 535        req->flags = htonl(F_TX_INIT);
 536        req->sndseq = htonl(ep->snd_seq);
 537        BUG_ON(ep->mpa_skb);
 538        ep->mpa_skb = skb;
 539        iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 540        start_ep_timer(ep);
 541        state_set(&ep->com, MPA_REQ_SENT);
 542        return;
 543}
 544
 545static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
 546{
 547        int mpalen;
 548        struct tx_data_wr *req;
 549        struct mpa_message *mpa;
 550        struct sk_buff *skb;
 551
 552        PDBG("%s ep %p plen %d\n", __func__, ep, plen);
 553
 554        mpalen = sizeof(*mpa) + plen;
 555
 556        skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
 557        if (!skb) {
 558                printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
 559                return -ENOMEM;
 560        }
 561        skb_reserve(skb, sizeof(*req));
 562        mpa = (struct mpa_message *) skb_put(skb, mpalen);
 563        memset(mpa, 0, sizeof(*mpa));
 564        memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
 565        mpa->flags = MPA_REJECT;
 566        mpa->revision = mpa_rev;
 567        mpa->private_data_size = htons(plen);
 568        if (plen)
 569                memcpy(mpa->private_data, pdata, plen);
 570
 571        /*
 572         * Reference the mpa skb again.  This ensures the data area
 573         * will remain in memory until the hw acks the tx.
 574         * Function tx_ack() will deref it.
 575         */
 576        skb_get(skb);
 577        skb->priority = CPL_PRIORITY_DATA;
 578        set_arp_failure_handler(skb, arp_failure_discard);
 579        skb_reset_transport_header(skb);
 580        req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
 581        req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
 582        req->wr_lo = htonl(V_WR_TID(ep->hwtid));
 583        req->len = htonl(mpalen);
 584        req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
 585                           V_TX_SNDBUF(snd_win>>15));
 586        req->flags = htonl(F_TX_INIT);
 587        req->sndseq = htonl(ep->snd_seq);
 588        BUG_ON(ep->mpa_skb);
 589        ep->mpa_skb = skb;
 590        return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 591}
 592
 593static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
 594{
 595        int mpalen;
 596        struct tx_data_wr *req;
 597        struct mpa_message *mpa;
 598        int len;
 599        struct sk_buff *skb;
 600
 601        PDBG("%s ep %p plen %d\n", __func__, ep, plen);
 602
 603        mpalen = sizeof(*mpa) + plen;
 604
 605        skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
 606        if (!skb) {
 607                printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
 608                return -ENOMEM;
 609        }
 610        skb->priority = CPL_PRIORITY_DATA;
 611        skb_reserve(skb, sizeof(*req));
 612        mpa = (struct mpa_message *) skb_put(skb, mpalen);
 613        memset(mpa, 0, sizeof(*mpa));
 614        memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
 615        mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
 616                     (markers_enabled ? MPA_MARKERS : 0);
 617        mpa->revision = mpa_rev;
 618        mpa->private_data_size = htons(plen);
 619        if (plen)
 620                memcpy(mpa->private_data, pdata, plen);
 621
 622        /*
 623         * Reference the mpa skb.  This ensures the data area
 624         * will remain in memory until the hw acks the tx.
 625         * Function tx_ack() will deref it.
 626         */
 627        skb_get(skb);
 628        set_arp_failure_handler(skb, arp_failure_discard);
 629        skb_reset_transport_header(skb);
 630        len = skb->len;
 631        req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
 632        req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
 633        req->wr_lo = htonl(V_WR_TID(ep->hwtid));
 634        req->len = htonl(len);
 635        req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
 636                           V_TX_SNDBUF(snd_win>>15));
 637        req->flags = htonl(F_TX_INIT);
 638        req->sndseq = htonl(ep->snd_seq);
 639        ep->mpa_skb = skb;
 640        state_set(&ep->com, MPA_REP_SENT);
 641        return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 642}
 643
 644static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 645{
 646        struct iwch_ep *ep = ctx;
 647        struct cpl_act_establish *req = cplhdr(skb);
 648        unsigned int tid = GET_TID(req);
 649
 650        PDBG("%s ep %p tid %d\n", __func__, ep, tid);
 651
 652        dst_confirm(ep->dst);
 653
 654        /* setup the hwtid for this connection */
 655        ep->hwtid = tid;
 656        cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
 657
 658        ep->snd_seq = ntohl(req->snd_isn);
 659        ep->rcv_seq = ntohl(req->rcv_isn);
 660
 661        set_emss(ep, ntohs(req->tcp_opt));
 662
 663        /* dealloc the atid */
 664        cxgb3_free_atid(ep->com.tdev, ep->atid);
 665
 666        /* start MPA negotiation */
 667        send_mpa_req(ep, skb);
 668
 669        return 0;
 670}
 671
 672static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
 673{
 674        PDBG("%s ep %p\n", __FILE__, ep);
 675        state_set(&ep->com, ABORTING);
 676        send_abort(ep, skb, gfp);
 677}
 678
 679static void close_complete_upcall(struct iwch_ep *ep)
 680{
 681        struct iw_cm_event event;
 682
 683        PDBG("%s ep %p\n", __func__, ep);
 684        memset(&event, 0, sizeof(event));
 685        event.event = IW_CM_EVENT_CLOSE;
 686        if (ep->com.cm_id) {
 687                PDBG("close complete delivered ep %p cm_id %p tid %d\n",
 688                     ep, ep->com.cm_id, ep->hwtid);
 689                ep->com.cm_id->event_handler(ep->com.cm_id, &event);
 690                ep->com.cm_id->rem_ref(ep->com.cm_id);
 691                ep->com.cm_id = NULL;
 692                ep->com.qp = NULL;
 693        }
 694}
 695
 696static void peer_close_upcall(struct iwch_ep *ep)
 697{
 698        struct iw_cm_event event;
 699
 700        PDBG("%s ep %p\n", __func__, ep);
 701        memset(&event, 0, sizeof(event));
 702        event.event = IW_CM_EVENT_DISCONNECT;
 703        if (ep->com.cm_id) {
 704                PDBG("peer close delivered ep %p cm_id %p tid %d\n",
 705                     ep, ep->com.cm_id, ep->hwtid);
 706                ep->com.cm_id->event_handler(ep->com.cm_id, &event);
 707        }
 708}
 709
 710static void peer_abort_upcall(struct iwch_ep *ep)
 711{
 712        struct iw_cm_event event;
 713
 714        PDBG("%s ep %p\n", __func__, ep);
 715        memset(&event, 0, sizeof(event));
 716        event.event = IW_CM_EVENT_CLOSE;
 717        event.status = -ECONNRESET;
 718        if (ep->com.cm_id) {
 719                PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
 720                     ep->com.cm_id, ep->hwtid);
 721                ep->com.cm_id->event_handler(ep->com.cm_id, &event);
 722                ep->com.cm_id->rem_ref(ep->com.cm_id);
 723                ep->com.cm_id = NULL;
 724                ep->com.qp = NULL;
 725        }
 726}
 727
 728static void connect_reply_upcall(struct iwch_ep *ep, int status)
 729{
 730        struct iw_cm_event event;
 731
 732        PDBG("%s ep %p status %d\n", __func__, ep, status);
 733        memset(&event, 0, sizeof(event));
 734        event.event = IW_CM_EVENT_CONNECT_REPLY;
 735        event.status = status;
 736        event.local_addr = ep->com.local_addr;
 737        event.remote_addr = ep->com.remote_addr;
 738
 739        if ((status == 0) || (status == -ECONNREFUSED)) {
 740                event.private_data_len = ep->plen;
 741                event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
 742        }
 743        if (ep->com.cm_id) {
 744                PDBG("%s ep %p tid %d status %d\n", __func__, ep,
 745                     ep->hwtid, status);
 746                ep->com.cm_id->event_handler(ep->com.cm_id, &event);
 747        }
 748        if (status < 0) {
 749                ep->com.cm_id->rem_ref(ep->com.cm_id);
 750                ep->com.cm_id = NULL;
 751                ep->com.qp = NULL;
 752        }
 753}
 754
 755static void connect_request_upcall(struct iwch_ep *ep)
 756{
 757        struct iw_cm_event event;
 758
 759        PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
 760        memset(&event, 0, sizeof(event));
 761        event.event = IW_CM_EVENT_CONNECT_REQUEST;
 762        event.local_addr = ep->com.local_addr;
 763        event.remote_addr = ep->com.remote_addr;
 764        event.private_data_len = ep->plen;
 765        event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
 766        event.provider_data = ep;
 767        if (state_read(&ep->parent_ep->com) != DEAD) {
 768                get_ep(&ep->com);
 769                ep->parent_ep->com.cm_id->event_handler(
 770                                                ep->parent_ep->com.cm_id,
 771                                                &event);
 772        }
 773        put_ep(&ep->parent_ep->com);
 774        ep->parent_ep = NULL;
 775}
 776
 777static void established_upcall(struct iwch_ep *ep)
 778{
 779        struct iw_cm_event event;
 780
 781        PDBG("%s ep %p\n", __func__, ep);
 782        memset(&event, 0, sizeof(event));
 783        event.event = IW_CM_EVENT_ESTABLISHED;
 784        if (ep->com.cm_id) {
 785                PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
 786                ep->com.cm_id->event_handler(ep->com.cm_id, &event);
 787        }
 788}
 789
 790static int update_rx_credits(struct iwch_ep *ep, u32 credits)
 791{
 792        struct cpl_rx_data_ack *req;
 793        struct sk_buff *skb;
 794
 795        PDBG("%s ep %p credits %u\n", __func__, ep, credits);
 796        skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
 797        if (!skb) {
 798                printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
 799                return 0;
 800        }
 801
 802        req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
 803        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 804        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
 805        req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
 806        skb->priority = CPL_PRIORITY_ACK;
 807        iwch_cxgb3_ofld_send(ep->com.tdev, skb);
 808        return credits;
 809}
 810
 811static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
 812{
 813        struct mpa_message *mpa;
 814        u16 plen;
 815        struct iwch_qp_attributes attrs;
 816        enum iwch_qp_attr_mask mask;
 817        int err;
 818
 819        PDBG("%s ep %p\n", __func__, ep);
 820
 821        /*
 822         * Stop mpa timer.  If it expired, then the state has
 823         * changed and we bail since ep_timeout already aborted
 824         * the connection.
 825         */
 826        stop_ep_timer(ep);
 827        if (state_read(&ep->com) != MPA_REQ_SENT)
 828                return;
 829
 830        /*
 831         * If we get more than the supported amount of private data
 832         * then we must fail this connection.
 833         */
 834        if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
 835                err = -EINVAL;
 836                goto err;
 837        }
 838
 839        /*
 840         * copy the new data into our accumulation buffer.
 841         */
 842        skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
 843                                  skb->len);
 844        ep->mpa_pkt_len += skb->len;
 845
 846        /*
 847         * if we don't even have the mpa message, then bail.
 848         */
 849        if (ep->mpa_pkt_len < sizeof(*mpa))
 850                return;
 851        mpa = (struct mpa_message *) ep->mpa_pkt;
 852
 853        /* Validate MPA header. */
 854        if (mpa->revision != mpa_rev) {
 855                err = -EPROTO;
 856                goto err;
 857        }
 858        if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
 859                err = -EPROTO;
 860                goto err;
 861        }
 862
 863        plen = ntohs(mpa->private_data_size);
 864
 865        /*
 866         * Fail if there's too much private data.
 867         */
 868        if (plen > MPA_MAX_PRIVATE_DATA) {
 869                err = -EPROTO;
 870                goto err;
 871        }
 872
 873        /*
 874         * If plen does not account for pkt size
 875         */
 876        if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
 877                err = -EPROTO;
 878                goto err;
 879        }
 880
 881        ep->plen = (u8) plen;
 882
 883        /*
 884         * If we don't have all the pdata yet, then bail.
 885         * We'll continue process when more data arrives.
 886         */
 887        if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
 888                return;
 889
 890        if (mpa->flags & MPA_REJECT) {
 891                err = -ECONNREFUSED;
 892                goto err;
 893        }
 894
 895        /*
 896         * If we get here we have accumulated the entire mpa
 897         * start reply message including private data. And
 898         * the MPA header is valid.
 899         */
 900        state_set(&ep->com, FPDU_MODE);
 901        ep->mpa_attr.initiator = 1;
 902        ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
 903        ep->mpa_attr.recv_marker_enabled = markers_enabled;
 904        ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
 905        ep->mpa_attr.version = mpa_rev;
 906        PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
 907             "xmit_marker_enabled=%d, version=%d\n", __func__,
 908             ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
 909             ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
 910
 911        attrs.mpa_attr = ep->mpa_attr;
 912        attrs.max_ird = ep->ird;
 913        attrs.max_ord = ep->ord;
 914        attrs.llp_stream_handle = ep;
 915        attrs.next_state = IWCH_QP_STATE_RTS;
 916
 917        mask = IWCH_QP_ATTR_NEXT_STATE |
 918            IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
 919            IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
 920
 921        /* bind QP and TID with INIT_WR */
 922        err = iwch_modify_qp(ep->com.qp->rhp,
 923                             ep->com.qp, mask, &attrs, 1);
 924        if (err)
 925                goto err;
 926
 927        if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
 928                iwch_post_zb_read(ep->com.qp);
 929        }
 930
 931        goto out;
 932err:
 933        abort_connection(ep, skb, GFP_KERNEL);
 934out:
 935        connect_reply_upcall(ep, err);
 936        return;
 937}
 938
 939static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
 940{
 941        struct mpa_message *mpa;
 942        u16 plen;
 943
 944        PDBG("%s ep %p\n", __func__, ep);
 945
 946        /*
 947         * Stop mpa timer.  If it expired, then the state has
 948         * changed and we bail since ep_timeout already aborted
 949         * the connection.
 950         */
 951        stop_ep_timer(ep);
 952        if (state_read(&ep->com) != MPA_REQ_WAIT)
 953                return;
 954
 955        /*
 956         * If we get more than the supported amount of private data
 957         * then we must fail this connection.
 958         */
 959        if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
 960                abort_connection(ep, skb, GFP_KERNEL);
 961                return;
 962        }
 963
 964        PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
 965
 966        /*
 967         * Copy the new data into our accumulation buffer.
 968         */
 969        skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
 970                                  skb->len);
 971        ep->mpa_pkt_len += skb->len;
 972
 973        /*
 974         * If we don't even have the mpa message, then bail.
 975         * We'll continue process when more data arrives.
 976         */
 977        if (ep->mpa_pkt_len < sizeof(*mpa))
 978                return;
 979        PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
 980        mpa = (struct mpa_message *) ep->mpa_pkt;
 981
 982        /*
 983         * Validate MPA Header.
 984         */
 985        if (mpa->revision != mpa_rev) {
 986                abort_connection(ep, skb, GFP_KERNEL);
 987                return;
 988        }
 989
 990        if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
 991                abort_connection(ep, skb, GFP_KERNEL);
 992                return;
 993        }
 994
 995        plen = ntohs(mpa->private_data_size);
 996
 997        /*
 998         * Fail if there's too much private data.
 999         */
1000        if (plen > MPA_MAX_PRIVATE_DATA) {
1001                abort_connection(ep, skb, GFP_KERNEL);
1002                return;
1003        }
1004
1005        /*
1006         * If plen does not account for pkt size
1007         */
1008        if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1009                abort_connection(ep, skb, GFP_KERNEL);
1010                return;
1011        }
1012        ep->plen = (u8) plen;
1013
1014        /*
1015         * If we don't have all the pdata yet, then bail.
1016         */
1017        if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1018                return;
1019
1020        /*
1021         * If we get here we have accumulated the entire mpa
1022         * start reply message including private data.
1023         */
1024        ep->mpa_attr.initiator = 0;
1025        ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1026        ep->mpa_attr.recv_marker_enabled = markers_enabled;
1027        ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1028        ep->mpa_attr.version = mpa_rev;
1029        PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1030             "xmit_marker_enabled=%d, version=%d\n", __func__,
1031             ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1032             ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1033
1034        state_set(&ep->com, MPA_REQ_RCVD);
1035
1036        /* drive upcall */
1037        connect_request_upcall(ep);
1038        return;
1039}
1040
1041static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1042{
1043        struct iwch_ep *ep = ctx;
1044        struct cpl_rx_data *hdr = cplhdr(skb);
1045        unsigned int dlen = ntohs(hdr->len);
1046
1047        PDBG("%s ep %p dlen %u\n", __func__, ep, dlen);
1048
1049        skb_pull(skb, sizeof(*hdr));
1050        skb_trim(skb, dlen);
1051
1052        ep->rcv_seq += dlen;
1053        BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1054
1055        switch (state_read(&ep->com)) {
1056        case MPA_REQ_SENT:
1057                process_mpa_reply(ep, skb);
1058                break;
1059        case MPA_REQ_WAIT:
1060                process_mpa_request(ep, skb);
1061                break;
1062        case MPA_REP_SENT:
1063                break;
1064        default:
1065                printk(KERN_ERR MOD "%s Unexpected streaming data."
1066                       " ep %p state %d tid %d\n",
1067                       __func__, ep, state_read(&ep->com), ep->hwtid);
1068
1069                /*
1070                 * The ep will timeout and inform the ULP of the failure.
1071                 * See ep_timeout().
1072                 */
1073                break;
1074        }
1075
1076        /* update RX credits */
1077        update_rx_credits(ep, dlen);
1078
1079        return CPL_RET_BUF_DONE;
1080}
1081
1082/*
1083 * Upcall from the adapter indicating data has been transmitted.
1084 * For us its just the single MPA request or reply.  We can now free
1085 * the skb holding the mpa message.
1086 */
1087static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1088{
1089        struct iwch_ep *ep = ctx;
1090        struct cpl_wr_ack *hdr = cplhdr(skb);
1091        unsigned int credits = ntohs(hdr->credits);
1092
1093        PDBG("%s ep %p credits %u\n", __func__, ep, credits);
1094
1095        if (credits == 0) {
1096                PDBG("%s 0 credit ack  ep %p state %u\n",
1097                     __func__, ep, state_read(&ep->com));
1098                return CPL_RET_BUF_DONE;
1099        }
1100
1101        BUG_ON(credits != 1);
1102        dst_confirm(ep->dst);
1103        if (!ep->mpa_skb) {
1104                PDBG("%s rdma_init wr_ack ep %p state %u\n",
1105                        __func__, ep, state_read(&ep->com));
1106                if (ep->mpa_attr.initiator) {
1107                        PDBG("%s initiator ep %p state %u\n",
1108                                __func__, ep, state_read(&ep->com));
1109                        if (peer2peer)
1110                                iwch_post_zb_read(ep->com.qp);
1111                } else {
1112                        PDBG("%s responder ep %p state %u\n",
1113                                __func__, ep, state_read(&ep->com));
1114                        ep->com.rpl_done = 1;
1115                        wake_up(&ep->com.waitq);
1116                }
1117        } else {
1118                PDBG("%s lsm ack ep %p state %u freeing skb\n",
1119                        __func__, ep, state_read(&ep->com));
1120                kfree_skb(ep->mpa_skb);
1121                ep->mpa_skb = NULL;
1122        }
1123        return CPL_RET_BUF_DONE;
1124}
1125
1126static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1127{
1128        struct iwch_ep *ep = ctx;
1129        unsigned long flags;
1130        int release = 0;
1131
1132        PDBG("%s ep %p\n", __func__, ep);
1133        BUG_ON(!ep);
1134
1135        /*
1136         * We get 2 abort replies from the HW.  The first one must
1137         * be ignored except for scribbling that we need one more.
1138         */
1139        if (!test_and_set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) {
1140                return CPL_RET_BUF_DONE;
1141        }
1142
1143        spin_lock_irqsave(&ep->com.lock, flags);
1144        switch (ep->com.state) {
1145        case ABORTING:
1146                close_complete_upcall(ep);
1147                __state_set(&ep->com, DEAD);
1148                release = 1;
1149                break;
1150        default:
1151                printk(KERN_ERR "%s ep %p state %d\n",
1152                     __func__, ep, ep->com.state);
1153                break;
1154        }
1155        spin_unlock_irqrestore(&ep->com.lock, flags);
1156
1157        if (release)
1158                release_ep_resources(ep);
1159        return CPL_RET_BUF_DONE;
1160}
1161
1162/*
1163 * Return whether a failed active open has allocated a TID
1164 */
1165static inline int act_open_has_tid(int status)
1166{
1167        return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1168               status != CPL_ERR_ARP_MISS;
1169}
1170
1171static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1172{
1173        struct iwch_ep *ep = ctx;
1174        struct cpl_act_open_rpl *rpl = cplhdr(skb);
1175
1176        PDBG("%s ep %p status %u errno %d\n", __func__, ep, rpl->status,
1177             status2errno(rpl->status));
1178        connect_reply_upcall(ep, status2errno(rpl->status));
1179        state_set(&ep->com, DEAD);
1180        if (ep->com.tdev->type != T3A && act_open_has_tid(rpl->status))
1181                release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1182        cxgb3_free_atid(ep->com.tdev, ep->atid);
1183        dst_release(ep->dst);
1184        l2t_release(L2DATA(ep->com.tdev), ep->l2t);
1185        put_ep(&ep->com);
1186        return CPL_RET_BUF_DONE;
1187}
1188
1189static int listen_start(struct iwch_listen_ep *ep)
1190{
1191        struct sk_buff *skb;
1192        struct cpl_pass_open_req *req;
1193
1194        PDBG("%s ep %p\n", __func__, ep);
1195        skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1196        if (!skb) {
1197                printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1198                return -ENOMEM;
1199        }
1200
1201        req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1202        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1203        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1204        req->local_port = ep->com.local_addr.sin_port;
1205        req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1206        req->peer_port = 0;
1207        req->peer_ip = 0;
1208        req->peer_netmask = 0;
1209        req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1210        req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1211        req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1212
1213        skb->priority = 1;
1214        return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1215}
1216
1217static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1218{
1219        struct iwch_listen_ep *ep = ctx;
1220        struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1221
1222        PDBG("%s ep %p status %d error %d\n", __func__, ep,
1223             rpl->status, status2errno(rpl->status));
1224        ep->com.rpl_err = status2errno(rpl->status);
1225        ep->com.rpl_done = 1;
1226        wake_up(&ep->com.waitq);
1227
1228        return CPL_RET_BUF_DONE;
1229}
1230
1231static int listen_stop(struct iwch_listen_ep *ep)
1232{
1233        struct sk_buff *skb;
1234        struct cpl_close_listserv_req *req;
1235
1236        PDBG("%s ep %p\n", __func__, ep);
1237        skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1238        if (!skb) {
1239                printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1240                return -ENOMEM;
1241        }
1242        req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1243        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1244        req->cpu_idx = 0;
1245        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1246        skb->priority = 1;
1247        return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1248}
1249
1250static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1251                             void *ctx)
1252{
1253        struct iwch_listen_ep *ep = ctx;
1254        struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1255
1256        PDBG("%s ep %p\n", __func__, ep);
1257        ep->com.rpl_err = status2errno(rpl->status);
1258        ep->com.rpl_done = 1;
1259        wake_up(&ep->com.waitq);
1260        return CPL_RET_BUF_DONE;
1261}
1262
1263static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1264{
1265        struct cpl_pass_accept_rpl *rpl;
1266        unsigned int mtu_idx;
1267        u32 opt0h, opt0l, opt2;
1268        int wscale;
1269
1270        PDBG("%s ep %p\n", __func__, ep);
1271        BUG_ON(skb_cloned(skb));
1272        skb_trim(skb, sizeof(*rpl));
1273        skb_get(skb);
1274        mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1275        wscale = compute_wscale(rcv_win);
1276        opt0h = V_NAGLE(0) |
1277            V_NO_CONG(nocong) |
1278            V_KEEP_ALIVE(1) |
1279            F_TCAM_BYPASS |
1280            V_WND_SCALE(wscale) |
1281            V_MSS_IDX(mtu_idx) |
1282            V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1283        opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1284        opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
1285               V_CONG_CONTROL_FLAVOR(cong_flavor);
1286
1287        rpl = cplhdr(skb);
1288        rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1289        OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1290        rpl->peer_ip = peer_ip;
1291        rpl->opt0h = htonl(opt0h);
1292        rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1293        rpl->opt2 = htonl(opt2);
1294        rpl->rsvd = rpl->opt2;  /* workaround for HW bug */
1295        skb->priority = CPL_PRIORITY_SETUP;
1296        iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
1297
1298        return;
1299}
1300
1301static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1302                      struct sk_buff *skb)
1303{
1304        PDBG("%s t3cdev %p tid %u peer_ip %x\n", __func__, tdev, hwtid,
1305             peer_ip);
1306        BUG_ON(skb_cloned(skb));
1307        skb_trim(skb, sizeof(struct cpl_tid_release));
1308        skb_get(skb);
1309
1310        if (tdev->type != T3A)
1311                release_tid(tdev, hwtid, skb);
1312        else {
1313                struct cpl_pass_accept_rpl *rpl;
1314
1315                rpl = cplhdr(skb);
1316                skb->priority = CPL_PRIORITY_SETUP;
1317                rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1318                OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1319                                                      hwtid));
1320                rpl->peer_ip = peer_ip;
1321                rpl->opt0h = htonl(F_TCAM_BYPASS);
1322                rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1323                rpl->opt2 = 0;
1324                rpl->rsvd = rpl->opt2;
1325                iwch_cxgb3_ofld_send(tdev, skb);
1326        }
1327}
1328
1329static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1330{
1331        struct iwch_ep *child_ep, *parent_ep = ctx;
1332        struct cpl_pass_accept_req *req = cplhdr(skb);
1333        unsigned int hwtid = GET_TID(req);
1334        struct dst_entry *dst;
1335        struct l2t_entry *l2t;
1336        struct rtable *rt;
1337        struct iff_mac tim;
1338
1339        PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1340
1341        if (state_read(&parent_ep->com) != LISTEN) {
1342                printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1343                       __func__);
1344                goto reject;
1345        }
1346
1347        /*
1348         * Find the netdev for this connection request.
1349         */
1350        tim.mac_addr = req->dst_mac;
1351        tim.vlan_tag = ntohs(req->vlan_tag);
1352        if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1353                printk(KERN_ERR "%s bad dst mac %pM\n",
1354                        __func__, req->dst_mac);
1355                goto reject;
1356        }
1357
1358        /* Find output route */
1359        rt = find_route(tdev,
1360                        req->local_ip,
1361                        req->peer_ip,
1362                        req->local_port,
1363                        req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1364        if (!rt) {
1365                printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1366                       __func__);
1367                goto reject;
1368        }
1369        dst = &rt->dst;
1370        l2t = t3_l2t_get(tdev, dst->neighbour, dst->neighbour->dev);
1371        if (!l2t) {
1372                printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1373                       __func__);
1374                dst_release(dst);
1375                goto reject;
1376        }
1377        child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1378        if (!child_ep) {
1379                printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1380                       __func__);
1381                l2t_release(L2DATA(tdev), l2t);
1382                dst_release(dst);
1383                goto reject;
1384        }
1385        state_set(&child_ep->com, CONNECTING);
1386        child_ep->com.tdev = tdev;
1387        child_ep->com.cm_id = NULL;
1388        child_ep->com.local_addr.sin_family = PF_INET;
1389        child_ep->com.local_addr.sin_port = req->local_port;
1390        child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1391        child_ep->com.remote_addr.sin_family = PF_INET;
1392        child_ep->com.remote_addr.sin_port = req->peer_port;
1393        child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1394        get_ep(&parent_ep->com);
1395        child_ep->parent_ep = parent_ep;
1396        child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1397        child_ep->l2t = l2t;
1398        child_ep->dst = dst;
1399        child_ep->hwtid = hwtid;
1400        init_timer(&child_ep->timer);
1401        cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1402        accept_cr(child_ep, req->peer_ip, skb);
1403        goto out;
1404reject:
1405        reject_cr(tdev, hwtid, req->peer_ip, skb);
1406out:
1407        return CPL_RET_BUF_DONE;
1408}
1409
1410static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1411{
1412        struct iwch_ep *ep = ctx;
1413        struct cpl_pass_establish *req = cplhdr(skb);
1414
1415        PDBG("%s ep %p\n", __func__, ep);
1416        ep->snd_seq = ntohl(req->snd_isn);
1417        ep->rcv_seq = ntohl(req->rcv_isn);
1418
1419        set_emss(ep, ntohs(req->tcp_opt));
1420
1421        dst_confirm(ep->dst);
1422        state_set(&ep->com, MPA_REQ_WAIT);
1423        start_ep_timer(ep);
1424
1425        return CPL_RET_BUF_DONE;
1426}
1427
1428static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1429{
1430        struct iwch_ep *ep = ctx;
1431        struct iwch_qp_attributes attrs;
1432        unsigned long flags;
1433        int disconnect = 1;
1434        int release = 0;
1435
1436        PDBG("%s ep %p\n", __func__, ep);
1437        dst_confirm(ep->dst);
1438
1439        spin_lock_irqsave(&ep->com.lock, flags);
1440        switch (ep->com.state) {
1441        case MPA_REQ_WAIT:
1442                __state_set(&ep->com, CLOSING);
1443                break;
1444        case MPA_REQ_SENT:
1445                __state_set(&ep->com, CLOSING);
1446                connect_reply_upcall(ep, -ECONNRESET);
1447                break;
1448        case MPA_REQ_RCVD:
1449
1450                /*
1451                 * We're gonna mark this puppy DEAD, but keep
1452                 * the reference on it until the ULP accepts or
1453                 * rejects the CR. Also wake up anyone waiting
1454                 * in rdma connection migration (see iwch_accept_cr()).
1455                 */
1456                __state_set(&ep->com, CLOSING);
1457                ep->com.rpl_done = 1;
1458                ep->com.rpl_err = -ECONNRESET;
1459                PDBG("waking up ep %p\n", ep);
1460                wake_up(&ep->com.waitq);
1461                break;
1462        case MPA_REP_SENT:
1463                __state_set(&ep->com, CLOSING);
1464                ep->com.rpl_done = 1;
1465                ep->com.rpl_err = -ECONNRESET;
1466                PDBG("waking up ep %p\n", ep);
1467                wake_up(&ep->com.waitq);
1468                break;
1469        case FPDU_MODE:
1470                start_ep_timer(ep);
1471                __state_set(&ep->com, CLOSING);
1472                attrs.next_state = IWCH_QP_STATE_CLOSING;
1473                iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1474                               IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1475                peer_close_upcall(ep);
1476                break;
1477        case ABORTING:
1478                disconnect = 0;
1479                break;
1480        case CLOSING:
1481                __state_set(&ep->com, MORIBUND);
1482                disconnect = 0;
1483                break;
1484        case MORIBUND:
1485                stop_ep_timer(ep);
1486                if (ep->com.cm_id && ep->com.qp) {
1487                        attrs.next_state = IWCH_QP_STATE_IDLE;
1488                        iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1489                                       IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1490                }
1491                close_complete_upcall(ep);
1492                __state_set(&ep->com, DEAD);
1493                release = 1;
1494                disconnect = 0;
1495                break;
1496        case DEAD:
1497                disconnect = 0;
1498                break;
1499        default:
1500                BUG_ON(1);
1501        }
1502        spin_unlock_irqrestore(&ep->com.lock, flags);
1503        if (disconnect)
1504                iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1505        if (release)
1506                release_ep_resources(ep);
1507        return CPL_RET_BUF_DONE;
1508}
1509
1510/*
1511 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1512 */
1513static int is_neg_adv_abort(unsigned int status)
1514{
1515        return status == CPL_ERR_RTX_NEG_ADVICE ||
1516               status == CPL_ERR_PERSIST_NEG_ADVICE;
1517}
1518
1519static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1520{
1521        struct cpl_abort_req_rss *req = cplhdr(skb);
1522        struct iwch_ep *ep = ctx;
1523        struct cpl_abort_rpl *rpl;
1524        struct sk_buff *rpl_skb;
1525        struct iwch_qp_attributes attrs;
1526        int ret;
1527        int release = 0;
1528        unsigned long flags;
1529
1530        if (is_neg_adv_abort(req->status)) {
1531                PDBG("%s neg_adv_abort ep %p tid %d\n", __func__, ep,
1532                     ep->hwtid);
1533                t3_l2t_send_event(ep->com.tdev, ep->l2t);
1534                return CPL_RET_BUF_DONE;
1535        }
1536
1537        /*
1538         * We get 2 peer aborts from the HW.  The first one must
1539         * be ignored except for scribbling that we need one more.
1540         */
1541        if (!test_and_set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) {
1542                return CPL_RET_BUF_DONE;
1543        }
1544
1545        spin_lock_irqsave(&ep->com.lock, flags);
1546        PDBG("%s ep %p state %u\n", __func__, ep, ep->com.state);
1547        switch (ep->com.state) {
1548        case CONNECTING:
1549                break;
1550        case MPA_REQ_WAIT:
1551                stop_ep_timer(ep);
1552                break;
1553        case MPA_REQ_SENT:
1554                stop_ep_timer(ep);
1555                connect_reply_upcall(ep, -ECONNRESET);
1556                break;
1557        case MPA_REP_SENT:
1558                ep->com.rpl_done = 1;
1559                ep->com.rpl_err = -ECONNRESET;
1560                PDBG("waking up ep %p\n", ep);
1561                wake_up(&ep->com.waitq);
1562                break;
1563        case MPA_REQ_RCVD:
1564
1565                /*
1566                 * We're gonna mark this puppy DEAD, but keep
1567                 * the reference on it until the ULP accepts or
1568                 * rejects the CR. Also wake up anyone waiting
1569                 * in rdma connection migration (see iwch_accept_cr()).
1570                 */
1571                ep->com.rpl_done = 1;
1572                ep->com.rpl_err = -ECONNRESET;
1573                PDBG("waking up ep %p\n", ep);
1574                wake_up(&ep->com.waitq);
1575                break;
1576        case MORIBUND:
1577        case CLOSING:
1578                stop_ep_timer(ep);
1579                /*FALLTHROUGH*/
1580        case FPDU_MODE:
1581                if (ep->com.cm_id && ep->com.qp) {
1582                        attrs.next_state = IWCH_QP_STATE_ERROR;
1583                        ret = iwch_modify_qp(ep->com.qp->rhp,
1584                                     ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1585                                     &attrs, 1);
1586                        if (ret)
1587                                printk(KERN_ERR MOD
1588                                       "%s - qp <- error failed!\n",
1589                                       __func__);
1590                }
1591                peer_abort_upcall(ep);
1592                break;
1593        case ABORTING:
1594                break;
1595        case DEAD:
1596                PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
1597                spin_unlock_irqrestore(&ep->com.lock, flags);
1598                return CPL_RET_BUF_DONE;
1599        default:
1600                BUG_ON(1);
1601                break;
1602        }
1603        dst_confirm(ep->dst);
1604        if (ep->com.state != ABORTING) {
1605                __state_set(&ep->com, DEAD);
1606                release = 1;
1607        }
1608        spin_unlock_irqrestore(&ep->com.lock, flags);
1609
1610        rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1611        if (!rpl_skb) {
1612                printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1613                       __func__);
1614                release = 1;
1615                goto out;
1616        }
1617        rpl_skb->priority = CPL_PRIORITY_DATA;
1618        rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1619        rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1620        rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1621        OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1622        rpl->cmd = CPL_ABORT_NO_RST;
1623        iwch_cxgb3_ofld_send(ep->com.tdev, rpl_skb);
1624out:
1625        if (release)
1626                release_ep_resources(ep);
1627        return CPL_RET_BUF_DONE;
1628}
1629
1630static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1631{
1632        struct iwch_ep *ep = ctx;
1633        struct iwch_qp_attributes attrs;
1634        unsigned long flags;
1635        int release = 0;
1636
1637        PDBG("%s ep %p\n", __func__, ep);
1638        BUG_ON(!ep);
1639
1640        /* The cm_id may be null if we failed to connect */
1641        spin_lock_irqsave(&ep->com.lock, flags);
1642        switch (ep->com.state) {
1643        case CLOSING:
1644                __state_set(&ep->com, MORIBUND);
1645                break;
1646        case MORIBUND:
1647                stop_ep_timer(ep);
1648                if ((ep->com.cm_id) && (ep->com.qp)) {
1649                        attrs.next_state = IWCH_QP_STATE_IDLE;
1650                        iwch_modify_qp(ep->com.qp->rhp,
1651                                             ep->com.qp,
1652                                             IWCH_QP_ATTR_NEXT_STATE,
1653                                             &attrs, 1);
1654                }
1655                close_complete_upcall(ep);
1656                __state_set(&ep->com, DEAD);
1657                release = 1;
1658                break;
1659        case ABORTING:
1660        case DEAD:
1661                break;
1662        default:
1663                BUG_ON(1);
1664                break;
1665        }
1666        spin_unlock_irqrestore(&ep->com.lock, flags);
1667        if (release)
1668                release_ep_resources(ep);
1669        return CPL_RET_BUF_DONE;
1670}
1671
1672/*
1673 * T3A does 3 things when a TERM is received:
1674 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1675 * 2) generate an async event on the QP with the TERMINATE opcode
1676 * 3) post a TERMINATE opcde cqe into the associated CQ.
1677 *
1678 * For (1), we save the message in the qp for later consumer consumption.
1679 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1680 * For (3), we toss the CQE in cxio_poll_cq().
1681 *
1682 * terminate() handles case (1)...
1683 */
1684static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1685{
1686        struct iwch_ep *ep = ctx;
1687
1688        if (state_read(&ep->com) != FPDU_MODE)
1689                return CPL_RET_BUF_DONE;
1690
1691        PDBG("%s ep %p\n", __func__, ep);
1692        skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1693        PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
1694        skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1695                                  skb->len);
1696        ep->com.qp->attr.terminate_msg_len = skb->len;
1697        ep->com.qp->attr.is_terminate_local = 0;
1698        return CPL_RET_BUF_DONE;
1699}
1700
1701static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1702{
1703        struct cpl_rdma_ec_status *rep = cplhdr(skb);
1704        struct iwch_ep *ep = ctx;
1705
1706        PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid,
1707             rep->status);
1708        if (rep->status) {
1709                struct iwch_qp_attributes attrs;
1710
1711                printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1712                       __func__, ep->hwtid);
1713                stop_ep_timer(ep);
1714                attrs.next_state = IWCH_QP_STATE_ERROR;
1715                iwch_modify_qp(ep->com.qp->rhp,
1716                               ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1717                               &attrs, 1);
1718                abort_connection(ep, NULL, GFP_KERNEL);
1719        }
1720        return CPL_RET_BUF_DONE;
1721}
1722
1723static void ep_timeout(unsigned long arg)
1724{
1725        struct iwch_ep *ep = (struct iwch_ep *)arg;
1726        struct iwch_qp_attributes attrs;
1727        unsigned long flags;
1728        int abort = 1;
1729
1730        spin_lock_irqsave(&ep->com.lock, flags);
1731        PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
1732             ep->com.state);
1733        switch (ep->com.state) {
1734        case MPA_REQ_SENT:
1735                __state_set(&ep->com, ABORTING);
1736                connect_reply_upcall(ep, -ETIMEDOUT);
1737                break;
1738        case MPA_REQ_WAIT:
1739                __state_set(&ep->com, ABORTING);
1740                break;
1741        case CLOSING:
1742        case MORIBUND:
1743                if (ep->com.cm_id && ep->com.qp) {
1744                        attrs.next_state = IWCH_QP_STATE_ERROR;
1745                        iwch_modify_qp(ep->com.qp->rhp,
1746                                     ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1747                                     &attrs, 1);
1748                }
1749                __state_set(&ep->com, ABORTING);
1750                break;
1751        default:
1752                printk(KERN_ERR "%s unexpected state ep %p state %u\n",
1753                        __func__, ep, ep->com.state);
1754                WARN_ON(1);
1755                abort = 0;
1756        }
1757        spin_unlock_irqrestore(&ep->com.lock, flags);
1758        if (abort)
1759                abort_connection(ep, NULL, GFP_ATOMIC);
1760        put_ep(&ep->com);
1761}
1762
1763int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1764{
1765        int err;
1766        struct iwch_ep *ep = to_ep(cm_id);
1767        PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1768
1769        if (state_read(&ep->com) == DEAD) {
1770                put_ep(&ep->com);
1771                return -ECONNRESET;
1772        }
1773        BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1774        if (mpa_rev == 0)
1775                abort_connection(ep, NULL, GFP_KERNEL);
1776        else {
1777                err = send_mpa_reject(ep, pdata, pdata_len);
1778                err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1779        }
1780        put_ep(&ep->com);
1781        return 0;
1782}
1783
1784int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1785{
1786        int err;
1787        struct iwch_qp_attributes attrs;
1788        enum iwch_qp_attr_mask mask;
1789        struct iwch_ep *ep = to_ep(cm_id);
1790        struct iwch_dev *h = to_iwch_dev(cm_id->device);
1791        struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1792
1793        PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1794        if (state_read(&ep->com) == DEAD) {
1795                err = -ECONNRESET;
1796                goto err;
1797        }
1798
1799        BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1800        BUG_ON(!qp);
1801
1802        if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1803            (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1804                abort_connection(ep, NULL, GFP_KERNEL);
1805                err = -EINVAL;
1806                goto err;
1807        }
1808
1809        cm_id->add_ref(cm_id);
1810        ep->com.cm_id = cm_id;
1811        ep->com.qp = qp;
1812
1813        ep->ird = conn_param->ird;
1814        ep->ord = conn_param->ord;
1815
1816        if (peer2peer && ep->ird == 0)
1817                ep->ird = 1;
1818
1819        PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1820
1821        /* bind QP to EP and move to RTS */
1822        attrs.mpa_attr = ep->mpa_attr;
1823        attrs.max_ird = ep->ird;
1824        attrs.max_ord = ep->ord;
1825        attrs.llp_stream_handle = ep;
1826        attrs.next_state = IWCH_QP_STATE_RTS;
1827
1828        /* bind QP and TID with INIT_WR */
1829        mask = IWCH_QP_ATTR_NEXT_STATE |
1830                             IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1831                             IWCH_QP_ATTR_MPA_ATTR |
1832                             IWCH_QP_ATTR_MAX_IRD |
1833                             IWCH_QP_ATTR_MAX_ORD;
1834
1835        err = iwch_modify_qp(ep->com.qp->rhp,
1836                             ep->com.qp, mask, &attrs, 1);
1837        if (err)
1838                goto err1;
1839
1840        /* if needed, wait for wr_ack */
1841        if (iwch_rqes_posted(qp)) {
1842                wait_event(ep->com.waitq, ep->com.rpl_done);
1843                err = ep->com.rpl_err;
1844                if (err)
1845                        goto err1;
1846        }
1847
1848        err = send_mpa_reply(ep, conn_param->private_data,
1849                             conn_param->private_data_len);
1850        if (err)
1851                goto err1;
1852
1853
1854        state_set(&ep->com, FPDU_MODE);
1855        established_upcall(ep);
1856        put_ep(&ep->com);
1857        return 0;
1858err1:
1859        ep->com.cm_id = NULL;
1860        ep->com.qp = NULL;
1861        cm_id->rem_ref(cm_id);
1862err:
1863        put_ep(&ep->com);
1864        return err;
1865}
1866
1867static int is_loopback_dst(struct iw_cm_id *cm_id)
1868{
1869        struct net_device *dev;
1870
1871        dev = ip_dev_find(&init_net, cm_id->remote_addr.sin_addr.s_addr);
1872        if (!dev)
1873                return 0;
1874        dev_put(dev);
1875        return 1;
1876}
1877
1878int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1879{
1880        int err = 0;
1881        struct iwch_dev *h = to_iwch_dev(cm_id->device);
1882        struct iwch_ep *ep;
1883        struct rtable *rt;
1884
1885        if (is_loopback_dst(cm_id)) {
1886                err = -ENOSYS;
1887                goto out;
1888        }
1889
1890        ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1891        if (!ep) {
1892                printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1893                err = -ENOMEM;
1894                goto out;
1895        }
1896        init_timer(&ep->timer);
1897        ep->plen = conn_param->private_data_len;
1898        if (ep->plen)
1899                memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1900                       conn_param->private_data, ep->plen);
1901        ep->ird = conn_param->ird;
1902        ep->ord = conn_param->ord;
1903
1904        if (peer2peer && ep->ord == 0)
1905                ep->ord = 1;
1906
1907        ep->com.tdev = h->rdev.t3cdev_p;
1908
1909        cm_id->add_ref(cm_id);
1910        ep->com.cm_id = cm_id;
1911        ep->com.qp = get_qhp(h, conn_param->qpn);
1912        BUG_ON(!ep->com.qp);
1913        PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
1914             ep->com.qp, cm_id);
1915
1916        /*
1917         * Allocate an active TID to initiate a TCP connection.
1918         */
1919        ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1920        if (ep->atid == -1) {
1921                printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1922                err = -ENOMEM;
1923                goto fail2;
1924        }
1925
1926        /* find a route */
1927        rt = find_route(h->rdev.t3cdev_p,
1928                        cm_id->local_addr.sin_addr.s_addr,
1929                        cm_id->remote_addr.sin_addr.s_addr,
1930                        cm_id->local_addr.sin_port,
1931                        cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1932        if (!rt) {
1933                printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1934                err = -EHOSTUNREACH;
1935                goto fail3;
1936        }
1937        ep->dst = &rt->dst;
1938
1939        /* get a l2t entry */
1940        ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst->neighbour,
1941                             ep->dst->neighbour->dev);
1942        if (!ep->l2t) {
1943                printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1944                err = -ENOMEM;
1945                goto fail4;
1946        }
1947
1948        state_set(&ep->com, CONNECTING);
1949        ep->tos = IPTOS_LOWDELAY;
1950        ep->com.local_addr = cm_id->local_addr;
1951        ep->com.remote_addr = cm_id->remote_addr;
1952
1953        /* send connect request to rnic */
1954        err = send_connect(ep);
1955        if (!err)
1956                goto out;
1957
1958        l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
1959fail4:
1960        dst_release(ep->dst);
1961fail3:
1962        cxgb3_free_atid(ep->com.tdev, ep->atid);
1963fail2:
1964        cm_id->rem_ref(cm_id);
1965        put_ep(&ep->com);
1966out:
1967        return err;
1968}
1969
1970int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1971{
1972        int err = 0;
1973        struct iwch_dev *h = to_iwch_dev(cm_id->device);
1974        struct iwch_listen_ep *ep;
1975
1976
1977        might_sleep();
1978
1979        ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1980        if (!ep) {
1981                printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1982                err = -ENOMEM;
1983                goto fail1;
1984        }
1985        PDBG("%s ep %p\n", __func__, ep);
1986        ep->com.tdev = h->rdev.t3cdev_p;
1987        cm_id->add_ref(cm_id);
1988        ep->com.cm_id = cm_id;
1989        ep->backlog = backlog;
1990        ep->com.local_addr = cm_id->local_addr;
1991
1992        /*
1993         * Allocate a server TID.
1994         */
1995        ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
1996        if (ep->stid == -1) {
1997                printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1998                err = -ENOMEM;
1999                goto fail2;
2000        }
2001
2002        state_set(&ep->com, LISTEN);
2003        err = listen_start(ep);
2004        if (err)
2005                goto fail3;
2006
2007        /* wait for pass_open_rpl */
2008        wait_event(ep->com.waitq, ep->com.rpl_done);
2009        err = ep->com.rpl_err;
2010        if (!err) {
2011                cm_id->provider_data = ep;
2012                goto out;
2013        }
2014fail3:
2015        cxgb3_free_stid(ep->com.tdev, ep->stid);
2016fail2:
2017        cm_id->rem_ref(cm_id);
2018        put_ep(&ep->com);
2019fail1:
2020out:
2021        return err;
2022}
2023
2024int iwch_destroy_listen(struct iw_cm_id *cm_id)
2025{
2026        int err;
2027        struct iwch_listen_ep *ep = to_listen_ep(cm_id);
2028
2029        PDBG("%s ep %p\n", __func__, ep);
2030
2031        might_sleep();
2032        state_set(&ep->com, DEAD);
2033        ep->com.rpl_done = 0;
2034        ep->com.rpl_err = 0;
2035        err = listen_stop(ep);
2036        if (err)
2037                goto done;
2038        wait_event(ep->com.waitq, ep->com.rpl_done);
2039        cxgb3_free_stid(ep->com.tdev, ep->stid);
2040done:
2041        err = ep->com.rpl_err;
2042        cm_id->rem_ref(cm_id);
2043        put_ep(&ep->com);
2044        return err;
2045}
2046
2047int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
2048{
2049        int ret=0;
2050        unsigned long flags;
2051        int close = 0;
2052        int fatal = 0;
2053        struct t3cdev *tdev;
2054        struct cxio_rdev *rdev;
2055
2056        spin_lock_irqsave(&ep->com.lock, flags);
2057
2058        PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2059             states[ep->com.state], abrupt);
2060
2061        tdev = (struct t3cdev *)ep->com.tdev;
2062        rdev = (struct cxio_rdev *)tdev->ulp;
2063        if (cxio_fatal_error(rdev)) {
2064                fatal = 1;
2065                close_complete_upcall(ep);
2066                ep->com.state = DEAD;
2067        }
2068        switch (ep->com.state) {
2069        case MPA_REQ_WAIT:
2070        case MPA_REQ_SENT:
2071        case MPA_REQ_RCVD:
2072        case MPA_REP_SENT:
2073        case FPDU_MODE:
2074                close = 1;
2075                if (abrupt)
2076                        ep->com.state = ABORTING;
2077                else {
2078                        ep->com.state = CLOSING;
2079                        start_ep_timer(ep);
2080                }
2081                set_bit(CLOSE_SENT, &ep->com.flags);
2082                break;
2083        case CLOSING:
2084                if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2085                        close = 1;
2086                        if (abrupt) {
2087                                stop_ep_timer(ep);
2088                                ep->com.state = ABORTING;
2089                        } else
2090                                ep->com.state = MORIBUND;
2091                }
2092                break;
2093        case MORIBUND:
2094        case ABORTING:
2095        case DEAD:
2096                PDBG("%s ignoring disconnect ep %p state %u\n",
2097                     __func__, ep, ep->com.state);
2098                break;
2099        default:
2100                BUG();
2101                break;
2102        }
2103
2104        spin_unlock_irqrestore(&ep->com.lock, flags);
2105        if (close) {
2106                if (abrupt)
2107                        ret = send_abort(ep, NULL, gfp);
2108                else
2109                        ret = send_halfclose(ep, gfp);
2110                if (ret)
2111                        fatal = 1;
2112        }
2113        if (fatal)
2114                release_ep_resources(ep);
2115        return ret;
2116}
2117
2118int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2119                     struct l2t_entry *l2t)
2120{
2121        struct iwch_ep *ep = ctx;
2122
2123        if (ep->dst != old)
2124                return 0;
2125
2126        PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2127             l2t);
2128        dst_hold(new);
2129        l2t_release(L2DATA(ep->com.tdev), ep->l2t);
2130        ep->l2t = l2t;
2131        dst_release(old);
2132        ep->dst = new;
2133        return 1;
2134}
2135
2136/*
2137 * All the CM events are handled on a work queue to have a safe context.
2138 * These are the real handlers that are called from the work queue.
2139 */
2140static const cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS] = {
2141        [CPL_ACT_ESTABLISH]     = act_establish,
2142        [CPL_ACT_OPEN_RPL]      = act_open_rpl,
2143        [CPL_RX_DATA]           = rx_data,
2144        [CPL_TX_DMA_ACK]        = tx_ack,
2145        [CPL_ABORT_RPL_RSS]     = abort_rpl,
2146        [CPL_ABORT_RPL]         = abort_rpl,
2147        [CPL_PASS_OPEN_RPL]     = pass_open_rpl,
2148        [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2149        [CPL_PASS_ACCEPT_REQ]   = pass_accept_req,
2150        [CPL_PASS_ESTABLISH]    = pass_establish,
2151        [CPL_PEER_CLOSE]        = peer_close,
2152        [CPL_ABORT_REQ_RSS]     = peer_abort,
2153        [CPL_CLOSE_CON_RPL]     = close_con_rpl,
2154        [CPL_RDMA_TERMINATE]    = terminate,
2155        [CPL_RDMA_EC_STATUS]    = ec_status,
2156};
2157
2158static void process_work(struct work_struct *work)
2159{
2160        struct sk_buff *skb = NULL;
2161        void *ep;
2162        struct t3cdev *tdev;
2163        int ret;
2164
2165        while ((skb = skb_dequeue(&rxq))) {
2166                ep = *((void **) (skb->cb));
2167                tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
2168                ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
2169                if (ret & CPL_RET_BUF_DONE)
2170                        kfree_skb(skb);
2171
2172                /*
2173                 * ep was referenced in sched(), and is freed here.
2174                 */
2175                put_ep((struct iwch_ep_common *)ep);
2176        }
2177}
2178
2179static DECLARE_WORK(skb_work, process_work);
2180
2181static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2182{
2183        struct iwch_ep_common *epc = ctx;
2184
2185        get_ep(epc);
2186
2187        /*
2188         * Save ctx and tdev in the skb->cb area.
2189         */
2190        *((void **) skb->cb) = ctx;
2191        *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2192
2193        /*
2194         * Queue the skb and schedule the worker thread.
2195         */
2196        skb_queue_tail(&rxq, skb);
2197        queue_work(workq, &skb_work);
2198        return 0;
2199}
2200
2201static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2202{
2203        struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2204
2205        if (rpl->status != CPL_ERR_NONE) {
2206                printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2207                       "for tid %u\n", rpl->status, GET_TID(rpl));
2208        }
2209        return CPL_RET_BUF_DONE;
2210}
2211
2212/*
2213 * All upcalls from the T3 Core go to sched() to schedule the
2214 * processing on a work queue.
2215 */
2216cxgb3_cpl_handler_func t3c_handlers[NUM_CPL_CMDS] = {
2217        [CPL_ACT_ESTABLISH]     = sched,
2218        [CPL_ACT_OPEN_RPL]      = sched,
2219        [CPL_RX_DATA]           = sched,
2220        [CPL_TX_DMA_ACK]        = sched,
2221        [CPL_ABORT_RPL_RSS]     = sched,
2222        [CPL_ABORT_RPL]         = sched,
2223        [CPL_PASS_OPEN_RPL]     = sched,
2224        [CPL_CLOSE_LISTSRV_RPL] = sched,
2225        [CPL_PASS_ACCEPT_REQ]   = sched,
2226        [CPL_PASS_ESTABLISH]    = sched,
2227        [CPL_PEER_CLOSE]        = sched,
2228        [CPL_CLOSE_CON_RPL]     = sched,
2229        [CPL_ABORT_REQ_RSS]     = sched,
2230        [CPL_RDMA_TERMINATE]    = sched,
2231        [CPL_RDMA_EC_STATUS]    = sched,
2232        [CPL_SET_TCB_RPL]       = set_tcb_rpl,
2233};
2234
2235int __init iwch_cm_init(void)
2236{
2237        skb_queue_head_init(&rxq);
2238
2239        workq = create_singlethread_workqueue("iw_cxgb3");
2240        if (!workq)
2241                return -ENOMEM;
2242
2243        return 0;
2244}
2245
2246void __exit iwch_cm_term(void)
2247{
2248        flush_workqueue(workq);
2249        destroy_workqueue(workq);
2250}
2251