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(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 flowi4 fl4;
 342
 343        rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
 344                                   peer_port, local_port, IPPROTO_TCP,
 345                                   tos, 0);
 346        if (IS_ERR(rt))
 347                return NULL;
 348        return rt;
 349}
 350
 351static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
 352{
 353        int i = 0;
 354
 355        while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
 356                ++i;
 357        return i;
 358}
 359
 360static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
 361{
 362        PDBG("%s t3cdev %p\n", __func__, dev);
 363        kfree_skb(skb);
 364}
 365
 366/*
 367 * Handle an ARP failure for an active open.
 368 */
 369static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
 370{
 371        printk(KERN_ERR MOD "ARP failure duing connect\n");
 372        kfree_skb(skb);
 373}
 374
 375/*
 376 * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
 377 * and send it along.
 378 */
 379static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
 380{
 381        struct cpl_abort_req *req = cplhdr(skb);
 382
 383        PDBG("%s t3cdev %p\n", __func__, dev);
 384        req->cmd = CPL_ABORT_NO_RST;
 385        iwch_cxgb3_ofld_send(dev, skb);
 386}
 387
 388static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
 389{
 390        struct cpl_close_con_req *req;
 391        struct sk_buff *skb;
 392
 393        PDBG("%s ep %p\n", __func__, ep);
 394        skb = get_skb(NULL, sizeof(*req), gfp);
 395        if (!skb) {
 396                printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
 397                return -ENOMEM;
 398        }
 399        skb->priority = CPL_PRIORITY_DATA;
 400        set_arp_failure_handler(skb, arp_failure_discard);
 401        req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
 402        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
 403        req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
 404        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
 405        return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 406}
 407
 408static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
 409{
 410        struct cpl_abort_req *req;
 411
 412        PDBG("%s ep %p\n", __func__, ep);
 413        skb = get_skb(skb, sizeof(*req), gfp);
 414        if (!skb) {
 415                printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
 416                       __func__);
 417                return -ENOMEM;
 418        }
 419        skb->priority = CPL_PRIORITY_DATA;
 420        set_arp_failure_handler(skb, abort_arp_failure);
 421        req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
 422        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
 423        req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
 424        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
 425        req->cmd = CPL_ABORT_SEND_RST;
 426        return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 427}
 428
 429static int send_connect(struct iwch_ep *ep)
 430{
 431        struct cpl_act_open_req *req;
 432        struct sk_buff *skb;
 433        u32 opt0h, opt0l, opt2;
 434        unsigned int mtu_idx;
 435        int wscale;
 436
 437        PDBG("%s ep %p\n", __func__, ep);
 438
 439        skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
 440        if (!skb) {
 441                printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
 442                       __func__);
 443                return -ENOMEM;
 444        }
 445        mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
 446        wscale = compute_wscale(rcv_win);
 447        opt0h = V_NAGLE(0) |
 448            V_NO_CONG(nocong) |
 449            V_KEEP_ALIVE(1) |
 450            F_TCAM_BYPASS |
 451            V_WND_SCALE(wscale) |
 452            V_MSS_IDX(mtu_idx) |
 453            V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
 454        opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
 455        opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
 456               V_CONG_CONTROL_FLAVOR(cong_flavor);
 457        skb->priority = CPL_PRIORITY_SETUP;
 458        set_arp_failure_handler(skb, act_open_req_arp_failure);
 459
 460        req = (struct cpl_act_open_req *) skb_put(skb, sizeof(*req));
 461        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 462        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ep->atid));
 463        req->local_port = ep->com.local_addr.sin_port;
 464        req->peer_port = ep->com.remote_addr.sin_port;
 465        req->local_ip = ep->com.local_addr.sin_addr.s_addr;
 466        req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
 467        req->opt0h = htonl(opt0h);
 468        req->opt0l = htonl(opt0l);
 469        req->params = 0;
 470        req->opt2 = htonl(opt2);
 471        return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 472}
 473
 474static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
 475{
 476        int mpalen;
 477        struct tx_data_wr *req;
 478        struct mpa_message *mpa;
 479        int len;
 480
 481        PDBG("%s ep %p pd_len %d\n", __func__, ep, ep->plen);
 482
 483        BUG_ON(skb_cloned(skb));
 484
 485        mpalen = sizeof(*mpa) + ep->plen;
 486        if (skb->data + mpalen + sizeof(*req) > skb_end_pointer(skb)) {
 487                kfree_skb(skb);
 488                skb=alloc_skb(mpalen + sizeof(*req), GFP_KERNEL);
 489                if (!skb) {
 490                        connect_reply_upcall(ep, -ENOMEM);
 491                        return;
 492                }
 493        }
 494        skb_trim(skb, 0);
 495        skb_reserve(skb, sizeof(*req));
 496        skb_put(skb, mpalen);
 497        skb->priority = CPL_PRIORITY_DATA;
 498        mpa = (struct mpa_message *) skb->data;
 499        memset(mpa, 0, sizeof(*mpa));
 500        memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
 501        mpa->flags = (crc_enabled ? MPA_CRC : 0) |
 502                     (markers_enabled ? MPA_MARKERS : 0);
 503        mpa->private_data_size = htons(ep->plen);
 504        mpa->revision = mpa_rev;
 505
 506        if (ep->plen)
 507                memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
 508
 509        /*
 510         * Reference the mpa skb.  This ensures the data area
 511         * will remain in memory until the hw acks the tx.
 512         * Function tx_ack() will deref it.
 513         */
 514        skb_get(skb);
 515        set_arp_failure_handler(skb, arp_failure_discard);
 516        skb_reset_transport_header(skb);
 517        len = skb->len;
 518        req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
 519        req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
 520        req->wr_lo = htonl(V_WR_TID(ep->hwtid));
 521        req->len = htonl(len);
 522        req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
 523                           V_TX_SNDBUF(snd_win>>15));
 524        req->flags = htonl(F_TX_INIT);
 525        req->sndseq = htonl(ep->snd_seq);
 526        BUG_ON(ep->mpa_skb);
 527        ep->mpa_skb = skb;
 528        iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 529        start_ep_timer(ep);
 530        state_set(&ep->com, MPA_REQ_SENT);
 531        return;
 532}
 533
 534static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
 535{
 536        int mpalen;
 537        struct tx_data_wr *req;
 538        struct mpa_message *mpa;
 539        struct sk_buff *skb;
 540
 541        PDBG("%s ep %p plen %d\n", __func__, ep, plen);
 542
 543        mpalen = sizeof(*mpa) + plen;
 544
 545        skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
 546        if (!skb) {
 547                printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
 548                return -ENOMEM;
 549        }
 550        skb_reserve(skb, sizeof(*req));
 551        mpa = (struct mpa_message *) skb_put(skb, mpalen);
 552        memset(mpa, 0, sizeof(*mpa));
 553        memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
 554        mpa->flags = MPA_REJECT;
 555        mpa->revision = mpa_rev;
 556        mpa->private_data_size = htons(plen);
 557        if (plen)
 558                memcpy(mpa->private_data, pdata, plen);
 559
 560        /*
 561         * Reference the mpa skb again.  This ensures the data area
 562         * will remain in memory until the hw acks the tx.
 563         * Function tx_ack() will deref it.
 564         */
 565        skb_get(skb);
 566        skb->priority = CPL_PRIORITY_DATA;
 567        set_arp_failure_handler(skb, arp_failure_discard);
 568        skb_reset_transport_header(skb);
 569        req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
 570        req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
 571        req->wr_lo = htonl(V_WR_TID(ep->hwtid));
 572        req->len = htonl(mpalen);
 573        req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
 574                           V_TX_SNDBUF(snd_win>>15));
 575        req->flags = htonl(F_TX_INIT);
 576        req->sndseq = htonl(ep->snd_seq);
 577        BUG_ON(ep->mpa_skb);
 578        ep->mpa_skb = skb;
 579        return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 580}
 581
 582static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
 583{
 584        int mpalen;
 585        struct tx_data_wr *req;
 586        struct mpa_message *mpa;
 587        int len;
 588        struct sk_buff *skb;
 589
 590        PDBG("%s ep %p plen %d\n", __func__, ep, plen);
 591
 592        mpalen = sizeof(*mpa) + plen;
 593
 594        skb = get_skb(NULL, mpalen + sizeof(*req), GFP_KERNEL);
 595        if (!skb) {
 596                printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
 597                return -ENOMEM;
 598        }
 599        skb->priority = CPL_PRIORITY_DATA;
 600        skb_reserve(skb, sizeof(*req));
 601        mpa = (struct mpa_message *) skb_put(skb, mpalen);
 602        memset(mpa, 0, sizeof(*mpa));
 603        memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
 604        mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
 605                     (markers_enabled ? MPA_MARKERS : 0);
 606        mpa->revision = mpa_rev;
 607        mpa->private_data_size = htons(plen);
 608        if (plen)
 609                memcpy(mpa->private_data, pdata, plen);
 610
 611        /*
 612         * Reference the mpa skb.  This ensures the data area
 613         * will remain in memory until the hw acks the tx.
 614         * Function tx_ack() will deref it.
 615         */
 616        skb_get(skb);
 617        set_arp_failure_handler(skb, arp_failure_discard);
 618        skb_reset_transport_header(skb);
 619        len = skb->len;
 620        req = (struct tx_data_wr *) skb_push(skb, sizeof(*req));
 621        req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_TX_DATA)|F_WR_COMPL);
 622        req->wr_lo = htonl(V_WR_TID(ep->hwtid));
 623        req->len = htonl(len);
 624        req->param = htonl(V_TX_PORT(ep->l2t->smt_idx) |
 625                           V_TX_SNDBUF(snd_win>>15));
 626        req->flags = htonl(F_TX_INIT);
 627        req->sndseq = htonl(ep->snd_seq);
 628        ep->mpa_skb = skb;
 629        state_set(&ep->com, MPA_REP_SENT);
 630        return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 631}
 632
 633static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 634{
 635        struct iwch_ep *ep = ctx;
 636        struct cpl_act_establish *req = cplhdr(skb);
 637        unsigned int tid = GET_TID(req);
 638
 639        PDBG("%s ep %p tid %d\n", __func__, ep, tid);
 640
 641        dst_confirm(ep->dst);
 642
 643        /* setup the hwtid for this connection */
 644        ep->hwtid = tid;
 645        cxgb3_insert_tid(ep->com.tdev, &t3c_client, ep, tid);
 646
 647        ep->snd_seq = ntohl(req->snd_isn);
 648        ep->rcv_seq = ntohl(req->rcv_isn);
 649
 650        set_emss(ep, ntohs(req->tcp_opt));
 651
 652        /* dealloc the atid */
 653        cxgb3_free_atid(ep->com.tdev, ep->atid);
 654
 655        /* start MPA negotiation */
 656        send_mpa_req(ep, skb);
 657
 658        return 0;
 659}
 660
 661static void abort_connection(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
 662{
 663        PDBG("%s ep %p\n", __FILE__, ep);
 664        state_set(&ep->com, ABORTING);
 665        send_abort(ep, skb, gfp);
 666}
 667
 668static void close_complete_upcall(struct iwch_ep *ep)
 669{
 670        struct iw_cm_event event;
 671
 672        PDBG("%s ep %p\n", __func__, ep);
 673        memset(&event, 0, sizeof(event));
 674        event.event = IW_CM_EVENT_CLOSE;
 675        if (ep->com.cm_id) {
 676                PDBG("close complete delivered ep %p cm_id %p tid %d\n",
 677                     ep, ep->com.cm_id, ep->hwtid);
 678                ep->com.cm_id->event_handler(ep->com.cm_id, &event);
 679                ep->com.cm_id->rem_ref(ep->com.cm_id);
 680                ep->com.cm_id = NULL;
 681                ep->com.qp = NULL;
 682        }
 683}
 684
 685static void peer_close_upcall(struct iwch_ep *ep)
 686{
 687        struct iw_cm_event event;
 688
 689        PDBG("%s ep %p\n", __func__, ep);
 690        memset(&event, 0, sizeof(event));
 691        event.event = IW_CM_EVENT_DISCONNECT;
 692        if (ep->com.cm_id) {
 693                PDBG("peer close delivered ep %p cm_id %p tid %d\n",
 694                     ep, ep->com.cm_id, ep->hwtid);
 695                ep->com.cm_id->event_handler(ep->com.cm_id, &event);
 696        }
 697}
 698
 699static void peer_abort_upcall(struct iwch_ep *ep)
 700{
 701        struct iw_cm_event event;
 702
 703        PDBG("%s ep %p\n", __func__, ep);
 704        memset(&event, 0, sizeof(event));
 705        event.event = IW_CM_EVENT_CLOSE;
 706        event.status = -ECONNRESET;
 707        if (ep->com.cm_id) {
 708                PDBG("abort delivered ep %p cm_id %p tid %d\n", ep,
 709                     ep->com.cm_id, ep->hwtid);
 710                ep->com.cm_id->event_handler(ep->com.cm_id, &event);
 711                ep->com.cm_id->rem_ref(ep->com.cm_id);
 712                ep->com.cm_id = NULL;
 713                ep->com.qp = NULL;
 714        }
 715}
 716
 717static void connect_reply_upcall(struct iwch_ep *ep, int status)
 718{
 719        struct iw_cm_event event;
 720
 721        PDBG("%s ep %p status %d\n", __func__, ep, status);
 722        memset(&event, 0, sizeof(event));
 723        event.event = IW_CM_EVENT_CONNECT_REPLY;
 724        event.status = status;
 725        event.local_addr = ep->com.local_addr;
 726        event.remote_addr = ep->com.remote_addr;
 727
 728        if ((status == 0) || (status == -ECONNREFUSED)) {
 729                event.private_data_len = ep->plen;
 730                event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
 731        }
 732        if (ep->com.cm_id) {
 733                PDBG("%s ep %p tid %d status %d\n", __func__, ep,
 734                     ep->hwtid, status);
 735                ep->com.cm_id->event_handler(ep->com.cm_id, &event);
 736        }
 737        if (status < 0) {
 738                ep->com.cm_id->rem_ref(ep->com.cm_id);
 739                ep->com.cm_id = NULL;
 740                ep->com.qp = NULL;
 741        }
 742}
 743
 744static void connect_request_upcall(struct iwch_ep *ep)
 745{
 746        struct iw_cm_event event;
 747
 748        PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
 749        memset(&event, 0, sizeof(event));
 750        event.event = IW_CM_EVENT_CONNECT_REQUEST;
 751        event.local_addr = ep->com.local_addr;
 752        event.remote_addr = ep->com.remote_addr;
 753        event.private_data_len = ep->plen;
 754        event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
 755        event.provider_data = ep;
 756        /*
 757         * Until ird/ord negotiation via MPAv2 support is added, send max
 758         * supported values
 759         */
 760        event.ird = event.ord = 8;
 761        if (state_read(&ep->parent_ep->com) != DEAD) {
 762                get_ep(&ep->com);
 763                ep->parent_ep->com.cm_id->event_handler(
 764                                                ep->parent_ep->com.cm_id,
 765                                                &event);
 766        }
 767        put_ep(&ep->parent_ep->com);
 768        ep->parent_ep = NULL;
 769}
 770
 771static void established_upcall(struct iwch_ep *ep)
 772{
 773        struct iw_cm_event event;
 774
 775        PDBG("%s ep %p\n", __func__, ep);
 776        memset(&event, 0, sizeof(event));
 777        event.event = IW_CM_EVENT_ESTABLISHED;
 778        /*
 779         * Until ird/ord negotiation via MPAv2 support is added, send max
 780         * supported values
 781         */
 782        event.ird = event.ord = 8;
 783        if (ep->com.cm_id) {
 784                PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
 785                ep->com.cm_id->event_handler(ep->com.cm_id, &event);
 786        }
 787}
 788
 789static int update_rx_credits(struct iwch_ep *ep, u32 credits)
 790{
 791        struct cpl_rx_data_ack *req;
 792        struct sk_buff *skb;
 793
 794        PDBG("%s ep %p credits %u\n", __func__, ep, credits);
 795        skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
 796        if (!skb) {
 797                printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
 798                return 0;
 799        }
 800
 801        req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
 802        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 803        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
 804        req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
 805        skb->priority = CPL_PRIORITY_ACK;
 806        iwch_cxgb3_ofld_send(ep->com.tdev, skb);
 807        return credits;
 808}
 809
 810static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
 811{
 812        struct mpa_message *mpa;
 813        u16 plen;
 814        struct iwch_qp_attributes attrs;
 815        enum iwch_qp_attr_mask mask;
 816        int err;
 817
 818        PDBG("%s ep %p\n", __func__, ep);
 819
 820        /*
 821         * Stop mpa timer.  If it expired, then the state has
 822         * changed and we bail since ep_timeout already aborted
 823         * the connection.
 824         */
 825        stop_ep_timer(ep);
 826        if (state_read(&ep->com) != MPA_REQ_SENT)
 827                return;
 828
 829        /*
 830         * If we get more than the supported amount of private data
 831         * then we must fail this connection.
 832         */
 833        if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
 834                err = -EINVAL;
 835                goto err;
 836        }
 837
 838        /*
 839         * copy the new data into our accumulation buffer.
 840         */
 841        skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
 842                                  skb->len);
 843        ep->mpa_pkt_len += skb->len;
 844
 845        /*
 846         * if we don't even have the mpa message, then bail.
 847         */
 848        if (ep->mpa_pkt_len < sizeof(*mpa))
 849                return;
 850        mpa = (struct mpa_message *) ep->mpa_pkt;
 851
 852        /* Validate MPA header. */
 853        if (mpa->revision != mpa_rev) {
 854                err = -EPROTO;
 855                goto err;
 856        }
 857        if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
 858                err = -EPROTO;
 859                goto err;
 860        }
 861
 862        plen = ntohs(mpa->private_data_size);
 863
 864        /*
 865         * Fail if there's too much private data.
 866         */
 867        if (plen > MPA_MAX_PRIVATE_DATA) {
 868                err = -EPROTO;
 869                goto err;
 870        }
 871
 872        /*
 873         * If plen does not account for pkt size
 874         */
 875        if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
 876                err = -EPROTO;
 877                goto err;
 878        }
 879
 880        ep->plen = (u8) plen;
 881
 882        /*
 883         * If we don't have all the pdata yet, then bail.
 884         * We'll continue process when more data arrives.
 885         */
 886        if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
 887                return;
 888
 889        if (mpa->flags & MPA_REJECT) {
 890                err = -ECONNREFUSED;
 891                goto err;
 892        }
 893
 894        /*
 895         * If we get here we have accumulated the entire mpa
 896         * start reply message including private data. And
 897         * the MPA header is valid.
 898         */
 899        state_set(&ep->com, FPDU_MODE);
 900        ep->mpa_attr.initiator = 1;
 901        ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
 902        ep->mpa_attr.recv_marker_enabled = markers_enabled;
 903        ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
 904        ep->mpa_attr.version = mpa_rev;
 905        PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
 906             "xmit_marker_enabled=%d, version=%d\n", __func__,
 907             ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
 908             ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
 909
 910        attrs.mpa_attr = ep->mpa_attr;
 911        attrs.max_ird = ep->ird;
 912        attrs.max_ord = ep->ord;
 913        attrs.llp_stream_handle = ep;
 914        attrs.next_state = IWCH_QP_STATE_RTS;
 915
 916        mask = IWCH_QP_ATTR_NEXT_STATE |
 917            IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
 918            IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
 919
 920        /* bind QP and TID with INIT_WR */
 921        err = iwch_modify_qp(ep->com.qp->rhp,
 922                             ep->com.qp, mask, &attrs, 1);
 923        if (err)
 924                goto err;
 925
 926        if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
 927                iwch_post_zb_read(ep);
 928        }
 929
 930        goto out;
 931err:
 932        abort_connection(ep, skb, GFP_KERNEL);
 933out:
 934        connect_reply_upcall(ep, err);
 935        return;
 936}
 937
 938static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
 939{
 940        struct mpa_message *mpa;
 941        u16 plen;
 942
 943        PDBG("%s ep %p\n", __func__, ep);
 944
 945        /*
 946         * Stop mpa timer.  If it expired, then the state has
 947         * changed and we bail since ep_timeout already aborted
 948         * the connection.
 949         */
 950        stop_ep_timer(ep);
 951        if (state_read(&ep->com) != MPA_REQ_WAIT)
 952                return;
 953
 954        /*
 955         * If we get more than the supported amount of private data
 956         * then we must fail this connection.
 957         */
 958        if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
 959                abort_connection(ep, skb, GFP_KERNEL);
 960                return;
 961        }
 962
 963        PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
 964
 965        /*
 966         * Copy the new data into our accumulation buffer.
 967         */
 968        skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
 969                                  skb->len);
 970        ep->mpa_pkt_len += skb->len;
 971
 972        /*
 973         * If we don't even have the mpa message, then bail.
 974         * We'll continue process when more data arrives.
 975         */
 976        if (ep->mpa_pkt_len < sizeof(*mpa))
 977                return;
 978        PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
 979        mpa = (struct mpa_message *) ep->mpa_pkt;
 980
 981        /*
 982         * Validate MPA Header.
 983         */
 984        if (mpa->revision != mpa_rev) {
 985                abort_connection(ep, skb, GFP_KERNEL);
 986                return;
 987        }
 988
 989        if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
 990                abort_connection(ep, skb, GFP_KERNEL);
 991                return;
 992        }
 993
 994        plen = ntohs(mpa->private_data_size);
 995
 996        /*
 997         * Fail if there's too much private data.
 998         */
 999        if (plen > MPA_MAX_PRIVATE_DATA) {
1000                abort_connection(ep, skb, GFP_KERNEL);
1001                return;
1002        }
1003
1004        /*
1005         * If plen does not account for pkt size
1006         */
1007        if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1008                abort_connection(ep, skb, GFP_KERNEL);
1009                return;
1010        }
1011        ep->plen = (u8) plen;
1012
1013        /*
1014         * If we don't have all the pdata yet, then bail.
1015         */
1016        if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1017                return;
1018
1019        /*
1020         * If we get here we have accumulated the entire mpa
1021         * start reply message including private data.
1022         */
1023        ep->mpa_attr.initiator = 0;
1024        ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1025        ep->mpa_attr.recv_marker_enabled = markers_enabled;
1026        ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1027        ep->mpa_attr.version = mpa_rev;
1028        PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1029             "xmit_marker_enabled=%d, version=%d\n", __func__,
1030             ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1031             ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1032
1033        state_set(&ep->com, MPA_REQ_RCVD);
1034
1035        /* drive upcall */
1036        connect_request_upcall(ep);
1037        return;
1038}
1039
1040static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1041{
1042        struct iwch_ep *ep = ctx;
1043        struct cpl_rx_data *hdr = cplhdr(skb);
1044        unsigned int dlen = ntohs(hdr->len);
1045
1046        PDBG("%s ep %p dlen %u\n", __func__, ep, dlen);
1047
1048        skb_pull(skb, sizeof(*hdr));
1049        skb_trim(skb, dlen);
1050
1051        ep->rcv_seq += dlen;
1052        BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1053
1054        switch (state_read(&ep->com)) {
1055        case MPA_REQ_SENT:
1056                process_mpa_reply(ep, skb);
1057                break;
1058        case MPA_REQ_WAIT:
1059                process_mpa_request(ep, skb);
1060                break;
1061        case MPA_REP_SENT:
1062                break;
1063        default:
1064                printk(KERN_ERR MOD "%s Unexpected streaming data."
1065                       " ep %p state %d tid %d\n",
1066                       __func__, ep, state_read(&ep->com), ep->hwtid);
1067
1068                /*
1069                 * The ep will timeout and inform the ULP of the failure.
1070                 * See ep_timeout().
1071                 */
1072                break;
1073        }
1074
1075        /* update RX credits */
1076        update_rx_credits(ep, dlen);
1077
1078        return CPL_RET_BUF_DONE;
1079}
1080
1081/*
1082 * Upcall from the adapter indicating data has been transmitted.
1083 * For us its just the single MPA request or reply.  We can now free
1084 * the skb holding the mpa message.
1085 */
1086static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1087{
1088        struct iwch_ep *ep = ctx;
1089        struct cpl_wr_ack *hdr = cplhdr(skb);
1090        unsigned int credits = ntohs(hdr->credits);
1091        unsigned long flags;
1092        int post_zb = 0;
1093
1094        PDBG("%s ep %p credits %u\n", __func__, ep, credits);
1095
1096        if (credits == 0) {
1097                PDBG("%s 0 credit ack  ep %p state %u\n",
1098                     __func__, ep, state_read(&ep->com));
1099                return CPL_RET_BUF_DONE;
1100        }
1101
1102        spin_lock_irqsave(&ep->com.lock, flags);
1103        BUG_ON(credits != 1);
1104        dst_confirm(ep->dst);
1105        if (!ep->mpa_skb) {
1106                PDBG("%s rdma_init wr_ack ep %p state %u\n",
1107                        __func__, ep, ep->com.state);
1108                if (ep->mpa_attr.initiator) {
1109                        PDBG("%s initiator ep %p state %u\n",
1110                                __func__, ep, ep->com.state);
1111                        if (peer2peer && ep->com.state == FPDU_MODE)
1112                                post_zb = 1;
1113                } else {
1114                        PDBG("%s responder ep %p state %u\n",
1115                                __func__, ep, ep->com.state);
1116                        if (ep->com.state == MPA_REQ_RCVD) {
1117                                ep->com.rpl_done = 1;
1118                                wake_up(&ep->com.waitq);
1119                        }
1120                }
1121        } else {
1122                PDBG("%s lsm ack ep %p state %u freeing skb\n",
1123                        __func__, ep, ep->com.state);
1124                kfree_skb(ep->mpa_skb);
1125                ep->mpa_skb = NULL;
1126        }
1127        spin_unlock_irqrestore(&ep->com.lock, flags);
1128        if (post_zb)
1129                iwch_post_zb_read(ep);
1130        return CPL_RET_BUF_DONE;
1131}
1132
1133static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1134{
1135        struct iwch_ep *ep = ctx;
1136        unsigned long flags;
1137        int release = 0;
1138
1139        PDBG("%s ep %p\n", __func__, ep);
1140        BUG_ON(!ep);
1141
1142        /*
1143         * We get 2 abort replies from the HW.  The first one must
1144         * be ignored except for scribbling that we need one more.
1145         */
1146        if (!test_and_set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) {
1147                return CPL_RET_BUF_DONE;
1148        }
1149
1150        spin_lock_irqsave(&ep->com.lock, flags);
1151        switch (ep->com.state) {
1152        case ABORTING:
1153                close_complete_upcall(ep);
1154                __state_set(&ep->com, DEAD);
1155                release = 1;
1156                break;
1157        default:
1158                printk(KERN_ERR "%s ep %p state %d\n",
1159                     __func__, ep, ep->com.state);
1160                break;
1161        }
1162        spin_unlock_irqrestore(&ep->com.lock, flags);
1163
1164        if (release)
1165                release_ep_resources(ep);
1166        return CPL_RET_BUF_DONE;
1167}
1168
1169/*
1170 * Return whether a failed active open has allocated a TID
1171 */
1172static inline int act_open_has_tid(int status)
1173{
1174        return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1175               status != CPL_ERR_ARP_MISS;
1176}
1177
1178static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1179{
1180        struct iwch_ep *ep = ctx;
1181        struct cpl_act_open_rpl *rpl = cplhdr(skb);
1182
1183        PDBG("%s ep %p status %u errno %d\n", __func__, ep, rpl->status,
1184             status2errno(rpl->status));
1185        connect_reply_upcall(ep, status2errno(rpl->status));
1186        state_set(&ep->com, DEAD);
1187        if (ep->com.tdev->type != T3A && act_open_has_tid(rpl->status))
1188                release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1189        cxgb3_free_atid(ep->com.tdev, ep->atid);
1190        dst_release(ep->dst);
1191        l2t_release(ep->com.tdev, ep->l2t);
1192        put_ep(&ep->com);
1193        return CPL_RET_BUF_DONE;
1194}
1195
1196static int listen_start(struct iwch_listen_ep *ep)
1197{
1198        struct sk_buff *skb;
1199        struct cpl_pass_open_req *req;
1200
1201        PDBG("%s ep %p\n", __func__, ep);
1202        skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1203        if (!skb) {
1204                printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1205                return -ENOMEM;
1206        }
1207
1208        req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1209        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1210        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1211        req->local_port = ep->com.local_addr.sin_port;
1212        req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1213        req->peer_port = 0;
1214        req->peer_ip = 0;
1215        req->peer_netmask = 0;
1216        req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1217        req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1218        req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1219
1220        skb->priority = 1;
1221        return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1222}
1223
1224static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1225{
1226        struct iwch_listen_ep *ep = ctx;
1227        struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1228
1229        PDBG("%s ep %p status %d error %d\n", __func__, ep,
1230             rpl->status, status2errno(rpl->status));
1231        ep->com.rpl_err = status2errno(rpl->status);
1232        ep->com.rpl_done = 1;
1233        wake_up(&ep->com.waitq);
1234
1235        return CPL_RET_BUF_DONE;
1236}
1237
1238static int listen_stop(struct iwch_listen_ep *ep)
1239{
1240        struct sk_buff *skb;
1241        struct cpl_close_listserv_req *req;
1242
1243        PDBG("%s ep %p\n", __func__, ep);
1244        skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1245        if (!skb) {
1246                printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1247                return -ENOMEM;
1248        }
1249        req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1250        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1251        req->cpu_idx = 0;
1252        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1253        skb->priority = 1;
1254        return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1255}
1256
1257static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1258                             void *ctx)
1259{
1260        struct iwch_listen_ep *ep = ctx;
1261        struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1262
1263        PDBG("%s ep %p\n", __func__, ep);
1264        ep->com.rpl_err = status2errno(rpl->status);
1265        ep->com.rpl_done = 1;
1266        wake_up(&ep->com.waitq);
1267        return CPL_RET_BUF_DONE;
1268}
1269
1270static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1271{
1272        struct cpl_pass_accept_rpl *rpl;
1273        unsigned int mtu_idx;
1274        u32 opt0h, opt0l, opt2;
1275        int wscale;
1276
1277        PDBG("%s ep %p\n", __func__, ep);
1278        BUG_ON(skb_cloned(skb));
1279        skb_trim(skb, sizeof(*rpl));
1280        skb_get(skb);
1281        mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1282        wscale = compute_wscale(rcv_win);
1283        opt0h = V_NAGLE(0) |
1284            V_NO_CONG(nocong) |
1285            V_KEEP_ALIVE(1) |
1286            F_TCAM_BYPASS |
1287            V_WND_SCALE(wscale) |
1288            V_MSS_IDX(mtu_idx) |
1289            V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1290        opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1291        opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
1292               V_CONG_CONTROL_FLAVOR(cong_flavor);
1293
1294        rpl = cplhdr(skb);
1295        rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1296        OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1297        rpl->peer_ip = peer_ip;
1298        rpl->opt0h = htonl(opt0h);
1299        rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1300        rpl->opt2 = htonl(opt2);
1301        rpl->rsvd = rpl->opt2;  /* workaround for HW bug */
1302        skb->priority = CPL_PRIORITY_SETUP;
1303        iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
1304
1305        return;
1306}
1307
1308static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1309                      struct sk_buff *skb)
1310{
1311        PDBG("%s t3cdev %p tid %u peer_ip %x\n", __func__, tdev, hwtid,
1312             peer_ip);
1313        BUG_ON(skb_cloned(skb));
1314        skb_trim(skb, sizeof(struct cpl_tid_release));
1315        skb_get(skb);
1316
1317        if (tdev->type != T3A)
1318                release_tid(tdev, hwtid, skb);
1319        else {
1320                struct cpl_pass_accept_rpl *rpl;
1321
1322                rpl = cplhdr(skb);
1323                skb->priority = CPL_PRIORITY_SETUP;
1324                rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1325                OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1326                                                      hwtid));
1327                rpl->peer_ip = peer_ip;
1328                rpl->opt0h = htonl(F_TCAM_BYPASS);
1329                rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1330                rpl->opt2 = 0;
1331                rpl->rsvd = rpl->opt2;
1332                iwch_cxgb3_ofld_send(tdev, skb);
1333        }
1334}
1335
1336static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1337{
1338        struct iwch_ep *child_ep, *parent_ep = ctx;
1339        struct cpl_pass_accept_req *req = cplhdr(skb);
1340        unsigned int hwtid = GET_TID(req);
1341        struct neighbour *neigh;
1342        struct dst_entry *dst;
1343        struct l2t_entry *l2t;
1344        struct rtable *rt;
1345        struct iff_mac tim;
1346
1347        PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1348
1349        if (state_read(&parent_ep->com) != LISTEN) {
1350                printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1351                       __func__);
1352                goto reject;
1353        }
1354
1355        /*
1356         * Find the netdev for this connection request.
1357         */
1358        tim.mac_addr = req->dst_mac;
1359        tim.vlan_tag = ntohs(req->vlan_tag);
1360        if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1361                printk(KERN_ERR "%s bad dst mac %pM\n",
1362                        __func__, req->dst_mac);
1363                goto reject;
1364        }
1365
1366        /* Find output route */
1367        rt = find_route(tdev,
1368                        req->local_ip,
1369                        req->peer_ip,
1370                        req->local_port,
1371                        req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1372        if (!rt) {
1373                printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1374                       __func__);
1375                goto reject;
1376        }
1377        dst = &rt->dst;
1378        rcu_read_lock();
1379        neigh = dst_get_neighbour(dst);
1380        l2t = t3_l2t_get(tdev, neigh, neigh->dev);
1381        rcu_read_unlock();
1382        if (!l2t) {
1383                printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1384                       __func__);
1385                dst_release(dst);
1386                goto reject;
1387        }
1388        child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1389        if (!child_ep) {
1390                printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1391                       __func__);
1392                l2t_release(tdev, l2t);
1393                dst_release(dst);
1394                goto reject;
1395        }
1396        state_set(&child_ep->com, CONNECTING);
1397        child_ep->com.tdev = tdev;
1398        child_ep->com.cm_id = NULL;
1399        child_ep->com.local_addr.sin_family = PF_INET;
1400        child_ep->com.local_addr.sin_port = req->local_port;
1401        child_ep->com.local_addr.sin_addr.s_addr = req->local_ip;
1402        child_ep->com.remote_addr.sin_family = PF_INET;
1403        child_ep->com.remote_addr.sin_port = req->peer_port;
1404        child_ep->com.remote_addr.sin_addr.s_addr = req->peer_ip;
1405        get_ep(&parent_ep->com);
1406        child_ep->parent_ep = parent_ep;
1407        child_ep->tos = G_PASS_OPEN_TOS(ntohl(req->tos_tid));
1408        child_ep->l2t = l2t;
1409        child_ep->dst = dst;
1410        child_ep->hwtid = hwtid;
1411        init_timer(&child_ep->timer);
1412        cxgb3_insert_tid(tdev, &t3c_client, child_ep, hwtid);
1413        accept_cr(child_ep, req->peer_ip, skb);
1414        goto out;
1415reject:
1416        reject_cr(tdev, hwtid, req->peer_ip, skb);
1417out:
1418        return CPL_RET_BUF_DONE;
1419}
1420
1421static int pass_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1422{
1423        struct iwch_ep *ep = ctx;
1424        struct cpl_pass_establish *req = cplhdr(skb);
1425
1426        PDBG("%s ep %p\n", __func__, ep);
1427        ep->snd_seq = ntohl(req->snd_isn);
1428        ep->rcv_seq = ntohl(req->rcv_isn);
1429
1430        set_emss(ep, ntohs(req->tcp_opt));
1431
1432        dst_confirm(ep->dst);
1433        state_set(&ep->com, MPA_REQ_WAIT);
1434        start_ep_timer(ep);
1435
1436        return CPL_RET_BUF_DONE;
1437}
1438
1439static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1440{
1441        struct iwch_ep *ep = ctx;
1442        struct iwch_qp_attributes attrs;
1443        unsigned long flags;
1444        int disconnect = 1;
1445        int release = 0;
1446
1447        PDBG("%s ep %p\n", __func__, ep);
1448        dst_confirm(ep->dst);
1449
1450        spin_lock_irqsave(&ep->com.lock, flags);
1451        switch (ep->com.state) {
1452        case MPA_REQ_WAIT:
1453                __state_set(&ep->com, CLOSING);
1454                break;
1455        case MPA_REQ_SENT:
1456                __state_set(&ep->com, CLOSING);
1457                connect_reply_upcall(ep, -ECONNRESET);
1458                break;
1459        case MPA_REQ_RCVD:
1460
1461                /*
1462                 * We're gonna mark this puppy DEAD, but keep
1463                 * the reference on it until the ULP accepts or
1464                 * rejects the CR. Also wake up anyone waiting
1465                 * in rdma connection migration (see iwch_accept_cr()).
1466                 */
1467                __state_set(&ep->com, CLOSING);
1468                ep->com.rpl_done = 1;
1469                ep->com.rpl_err = -ECONNRESET;
1470                PDBG("waking up ep %p\n", ep);
1471                wake_up(&ep->com.waitq);
1472                break;
1473        case MPA_REP_SENT:
1474                __state_set(&ep->com, CLOSING);
1475                ep->com.rpl_done = 1;
1476                ep->com.rpl_err = -ECONNRESET;
1477                PDBG("waking up ep %p\n", ep);
1478                wake_up(&ep->com.waitq);
1479                break;
1480        case FPDU_MODE:
1481                start_ep_timer(ep);
1482                __state_set(&ep->com, CLOSING);
1483                attrs.next_state = IWCH_QP_STATE_CLOSING;
1484                iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1485                               IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1486                peer_close_upcall(ep);
1487                break;
1488        case ABORTING:
1489                disconnect = 0;
1490                break;
1491        case CLOSING:
1492                __state_set(&ep->com, MORIBUND);
1493                disconnect = 0;
1494                break;
1495        case MORIBUND:
1496                stop_ep_timer(ep);
1497                if (ep->com.cm_id && ep->com.qp) {
1498                        attrs.next_state = IWCH_QP_STATE_IDLE;
1499                        iwch_modify_qp(ep->com.qp->rhp, ep->com.qp,
1500                                       IWCH_QP_ATTR_NEXT_STATE, &attrs, 1);
1501                }
1502                close_complete_upcall(ep);
1503                __state_set(&ep->com, DEAD);
1504                release = 1;
1505                disconnect = 0;
1506                break;
1507        case DEAD:
1508                disconnect = 0;
1509                break;
1510        default:
1511                BUG_ON(1);
1512        }
1513        spin_unlock_irqrestore(&ep->com.lock, flags);
1514        if (disconnect)
1515                iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1516        if (release)
1517                release_ep_resources(ep);
1518        return CPL_RET_BUF_DONE;
1519}
1520
1521/*
1522 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1523 */
1524static int is_neg_adv_abort(unsigned int status)
1525{
1526        return status == CPL_ERR_RTX_NEG_ADVICE ||
1527               status == CPL_ERR_PERSIST_NEG_ADVICE;
1528}
1529
1530static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1531{
1532        struct cpl_abort_req_rss *req = cplhdr(skb);
1533        struct iwch_ep *ep = ctx;
1534        struct cpl_abort_rpl *rpl;
1535        struct sk_buff *rpl_skb;
1536        struct iwch_qp_attributes attrs;
1537        int ret;
1538        int release = 0;
1539        unsigned long flags;
1540
1541        if (is_neg_adv_abort(req->status)) {
1542                PDBG("%s neg_adv_abort ep %p tid %d\n", __func__, ep,
1543                     ep->hwtid);
1544                t3_l2t_send_event(ep->com.tdev, ep->l2t);
1545                return CPL_RET_BUF_DONE;
1546        }
1547
1548        /*
1549         * We get 2 peer aborts from the HW.  The first one must
1550         * be ignored except for scribbling that we need one more.
1551         */
1552        if (!test_and_set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) {
1553                return CPL_RET_BUF_DONE;
1554        }
1555
1556        spin_lock_irqsave(&ep->com.lock, flags);
1557        PDBG("%s ep %p state %u\n", __func__, ep, ep->com.state);
1558        switch (ep->com.state) {
1559        case CONNECTING:
1560                break;
1561        case MPA_REQ_WAIT:
1562                stop_ep_timer(ep);
1563                break;
1564        case MPA_REQ_SENT:
1565                stop_ep_timer(ep);
1566                connect_reply_upcall(ep, -ECONNRESET);
1567                break;
1568        case MPA_REP_SENT:
1569                ep->com.rpl_done = 1;
1570                ep->com.rpl_err = -ECONNRESET;
1571                PDBG("waking up ep %p\n", ep);
1572                wake_up(&ep->com.waitq);
1573                break;
1574        case MPA_REQ_RCVD:
1575
1576                /*
1577                 * We're gonna mark this puppy DEAD, but keep
1578                 * the reference on it until the ULP accepts or
1579                 * rejects the CR. Also wake up anyone waiting
1580                 * in rdma connection migration (see iwch_accept_cr()).
1581                 */
1582                ep->com.rpl_done = 1;
1583                ep->com.rpl_err = -ECONNRESET;
1584                PDBG("waking up ep %p\n", ep);
1585                wake_up(&ep->com.waitq);
1586                break;
1587        case MORIBUND:
1588        case CLOSING:
1589                stop_ep_timer(ep);
1590                /*FALLTHROUGH*/
1591        case FPDU_MODE:
1592                if (ep->com.cm_id && ep->com.qp) {
1593                        attrs.next_state = IWCH_QP_STATE_ERROR;
1594                        ret = iwch_modify_qp(ep->com.qp->rhp,
1595                                     ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1596                                     &attrs, 1);
1597                        if (ret)
1598                                printk(KERN_ERR MOD
1599                                       "%s - qp <- error failed!\n",
1600                                       __func__);
1601                }
1602                peer_abort_upcall(ep);
1603                break;
1604        case ABORTING:
1605                break;
1606        case DEAD:
1607                PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
1608                spin_unlock_irqrestore(&ep->com.lock, flags);
1609                return CPL_RET_BUF_DONE;
1610        default:
1611                BUG_ON(1);
1612                break;
1613        }
1614        dst_confirm(ep->dst);
1615        if (ep->com.state != ABORTING) {
1616                __state_set(&ep->com, DEAD);
1617                release = 1;
1618        }
1619        spin_unlock_irqrestore(&ep->com.lock, flags);
1620
1621        rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
1622        if (!rpl_skb) {
1623                printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
1624                       __func__);
1625                release = 1;
1626                goto out;
1627        }
1628        rpl_skb->priority = CPL_PRIORITY_DATA;
1629        rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
1630        rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
1631        rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
1632        OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
1633        rpl->cmd = CPL_ABORT_NO_RST;
1634        iwch_cxgb3_ofld_send(ep->com.tdev, rpl_skb);
1635out:
1636        if (release)
1637                release_ep_resources(ep);
1638        return CPL_RET_BUF_DONE;
1639}
1640
1641static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1642{
1643        struct iwch_ep *ep = ctx;
1644        struct iwch_qp_attributes attrs;
1645        unsigned long flags;
1646        int release = 0;
1647
1648        PDBG("%s ep %p\n", __func__, ep);
1649        BUG_ON(!ep);
1650
1651        /* The cm_id may be null if we failed to connect */
1652        spin_lock_irqsave(&ep->com.lock, flags);
1653        switch (ep->com.state) {
1654        case CLOSING:
1655                __state_set(&ep->com, MORIBUND);
1656                break;
1657        case MORIBUND:
1658                stop_ep_timer(ep);
1659                if ((ep->com.cm_id) && (ep->com.qp)) {
1660                        attrs.next_state = IWCH_QP_STATE_IDLE;
1661                        iwch_modify_qp(ep->com.qp->rhp,
1662                                             ep->com.qp,
1663                                             IWCH_QP_ATTR_NEXT_STATE,
1664                                             &attrs, 1);
1665                }
1666                close_complete_upcall(ep);
1667                __state_set(&ep->com, DEAD);
1668                release = 1;
1669                break;
1670        case ABORTING:
1671        case DEAD:
1672                break;
1673        default:
1674                BUG_ON(1);
1675                break;
1676        }
1677        spin_unlock_irqrestore(&ep->com.lock, flags);
1678        if (release)
1679                release_ep_resources(ep);
1680        return CPL_RET_BUF_DONE;
1681}
1682
1683/*
1684 * T3A does 3 things when a TERM is received:
1685 * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
1686 * 2) generate an async event on the QP with the TERMINATE opcode
1687 * 3) post a TERMINATE opcde cqe into the associated CQ.
1688 *
1689 * For (1), we save the message in the qp for later consumer consumption.
1690 * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
1691 * For (3), we toss the CQE in cxio_poll_cq().
1692 *
1693 * terminate() handles case (1)...
1694 */
1695static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1696{
1697        struct iwch_ep *ep = ctx;
1698
1699        if (state_read(&ep->com) != FPDU_MODE)
1700                return CPL_RET_BUF_DONE;
1701
1702        PDBG("%s ep %p\n", __func__, ep);
1703        skb_pull(skb, sizeof(struct cpl_rdma_terminate));
1704        PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
1705        skb_copy_from_linear_data(skb, ep->com.qp->attr.terminate_buffer,
1706                                  skb->len);
1707        ep->com.qp->attr.terminate_msg_len = skb->len;
1708        ep->com.qp->attr.is_terminate_local = 0;
1709        return CPL_RET_BUF_DONE;
1710}
1711
1712static int ec_status(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1713{
1714        struct cpl_rdma_ec_status *rep = cplhdr(skb);
1715        struct iwch_ep *ep = ctx;
1716
1717        PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid,
1718             rep->status);
1719        if (rep->status) {
1720                struct iwch_qp_attributes attrs;
1721
1722                printk(KERN_ERR MOD "%s BAD CLOSE - Aborting tid %u\n",
1723                       __func__, ep->hwtid);
1724                stop_ep_timer(ep);
1725                attrs.next_state = IWCH_QP_STATE_ERROR;
1726                iwch_modify_qp(ep->com.qp->rhp,
1727                               ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1728                               &attrs, 1);
1729                abort_connection(ep, NULL, GFP_KERNEL);
1730        }
1731        return CPL_RET_BUF_DONE;
1732}
1733
1734static void ep_timeout(unsigned long arg)
1735{
1736        struct iwch_ep *ep = (struct iwch_ep *)arg;
1737        struct iwch_qp_attributes attrs;
1738        unsigned long flags;
1739        int abort = 1;
1740
1741        spin_lock_irqsave(&ep->com.lock, flags);
1742        PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
1743             ep->com.state);
1744        switch (ep->com.state) {
1745        case MPA_REQ_SENT:
1746                __state_set(&ep->com, ABORTING);
1747                connect_reply_upcall(ep, -ETIMEDOUT);
1748                break;
1749        case MPA_REQ_WAIT:
1750                __state_set(&ep->com, ABORTING);
1751                break;
1752        case CLOSING:
1753        case MORIBUND:
1754                if (ep->com.cm_id && ep->com.qp) {
1755                        attrs.next_state = IWCH_QP_STATE_ERROR;
1756                        iwch_modify_qp(ep->com.qp->rhp,
1757                                     ep->com.qp, IWCH_QP_ATTR_NEXT_STATE,
1758                                     &attrs, 1);
1759                }
1760                __state_set(&ep->com, ABORTING);
1761                break;
1762        default:
1763                printk(KERN_ERR "%s unexpected state ep %p state %u\n",
1764                        __func__, ep, ep->com.state);
1765                WARN_ON(1);
1766                abort = 0;
1767        }
1768        spin_unlock_irqrestore(&ep->com.lock, flags);
1769        if (abort)
1770                abort_connection(ep, NULL, GFP_ATOMIC);
1771        put_ep(&ep->com);
1772}
1773
1774int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1775{
1776        int err;
1777        struct iwch_ep *ep = to_ep(cm_id);
1778        PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1779
1780        if (state_read(&ep->com) == DEAD) {
1781                put_ep(&ep->com);
1782                return -ECONNRESET;
1783        }
1784        BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1785        if (mpa_rev == 0)
1786                abort_connection(ep, NULL, GFP_KERNEL);
1787        else {
1788                err = send_mpa_reject(ep, pdata, pdata_len);
1789                err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1790        }
1791        put_ep(&ep->com);
1792        return 0;
1793}
1794
1795int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1796{
1797        int err;
1798        struct iwch_qp_attributes attrs;
1799        enum iwch_qp_attr_mask mask;
1800        struct iwch_ep *ep = to_ep(cm_id);
1801        struct iwch_dev *h = to_iwch_dev(cm_id->device);
1802        struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1803
1804        PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1805        if (state_read(&ep->com) == DEAD) {
1806                err = -ECONNRESET;
1807                goto err;
1808        }
1809
1810        BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1811        BUG_ON(!qp);
1812
1813        if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1814            (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1815                abort_connection(ep, NULL, GFP_KERNEL);
1816                err = -EINVAL;
1817                goto err;
1818        }
1819
1820        cm_id->add_ref(cm_id);
1821        ep->com.cm_id = cm_id;
1822        ep->com.qp = qp;
1823
1824        ep->ird = conn_param->ird;
1825        ep->ord = conn_param->ord;
1826
1827        if (peer2peer && ep->ird == 0)
1828                ep->ird = 1;
1829
1830        PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1831
1832        /* bind QP to EP and move to RTS */
1833        attrs.mpa_attr = ep->mpa_attr;
1834        attrs.max_ird = ep->ird;
1835        attrs.max_ord = ep->ord;
1836        attrs.llp_stream_handle = ep;
1837        attrs.next_state = IWCH_QP_STATE_RTS;
1838
1839        /* bind QP and TID with INIT_WR */
1840        mask = IWCH_QP_ATTR_NEXT_STATE |
1841                             IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1842                             IWCH_QP_ATTR_MPA_ATTR |
1843                             IWCH_QP_ATTR_MAX_IRD |
1844                             IWCH_QP_ATTR_MAX_ORD;
1845
1846        err = iwch_modify_qp(ep->com.qp->rhp,
1847                             ep->com.qp, mask, &attrs, 1);
1848        if (err)
1849                goto err1;
1850
1851        /* if needed, wait for wr_ack */
1852        if (iwch_rqes_posted(qp)) {
1853                wait_event(ep->com.waitq, ep->com.rpl_done);
1854                err = ep->com.rpl_err;
1855                if (err)
1856                        goto err1;
1857        }
1858
1859        err = send_mpa_reply(ep, conn_param->private_data,
1860                             conn_param->private_data_len);
1861        if (err)
1862                goto err1;
1863
1864
1865        state_set(&ep->com, FPDU_MODE);
1866        established_upcall(ep);
1867        put_ep(&ep->com);
1868        return 0;
1869err1:
1870        ep->com.cm_id = NULL;
1871        ep->com.qp = NULL;
1872        cm_id->rem_ref(cm_id);
1873err:
1874        put_ep(&ep->com);
1875        return err;
1876}
1877
1878static int is_loopback_dst(struct iw_cm_id *cm_id)
1879{
1880        struct net_device *dev;
1881
1882        dev = ip_dev_find(&init_net, cm_id->remote_addr.sin_addr.s_addr);
1883        if (!dev)
1884                return 0;
1885        dev_put(dev);
1886        return 1;
1887}
1888
1889int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1890{
1891        struct iwch_dev *h = to_iwch_dev(cm_id->device);
1892        struct neighbour *neigh;
1893        struct iwch_ep *ep;
1894        struct rtable *rt;
1895        int err = 0;
1896
1897        if (is_loopback_dst(cm_id)) {
1898                err = -ENOSYS;
1899                goto out;
1900        }
1901
1902        ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1903        if (!ep) {
1904                printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1905                err = -ENOMEM;
1906                goto out;
1907        }
1908        init_timer(&ep->timer);
1909        ep->plen = conn_param->private_data_len;
1910        if (ep->plen)
1911                memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1912                       conn_param->private_data, ep->plen);
1913        ep->ird = conn_param->ird;
1914        ep->ord = conn_param->ord;
1915
1916        if (peer2peer && ep->ord == 0)
1917                ep->ord = 1;
1918
1919        ep->com.tdev = h->rdev.t3cdev_p;
1920
1921        cm_id->add_ref(cm_id);
1922        ep->com.cm_id = cm_id;
1923        ep->com.qp = get_qhp(h, conn_param->qpn);
1924        BUG_ON(!ep->com.qp);
1925        PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
1926             ep->com.qp, cm_id);
1927
1928        /*
1929         * Allocate an active TID to initiate a TCP connection.
1930         */
1931        ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1932        if (ep->atid == -1) {
1933                printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1934                err = -ENOMEM;
1935                goto fail2;
1936        }
1937
1938        /* find a route */
1939        rt = find_route(h->rdev.t3cdev_p,
1940                        cm_id->local_addr.sin_addr.s_addr,
1941                        cm_id->remote_addr.sin_addr.s_addr,
1942                        cm_id->local_addr.sin_port,
1943                        cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
1944        if (!rt) {
1945                printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1946                err = -EHOSTUNREACH;
1947                goto fail3;
1948        }
1949        ep->dst = &rt->dst;
1950
1951        rcu_read_lock();
1952        neigh = dst_get_neighbour(ep->dst);
1953
1954        /* get a l2t entry */
1955        ep->l2t = t3_l2t_get(ep->com.tdev, neigh, neigh->dev);
1956        rcu_read_unlock();
1957        if (!ep->l2t) {
1958                printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1959                err = -ENOMEM;
1960                goto fail4;
1961        }
1962
1963        state_set(&ep->com, CONNECTING);
1964        ep->tos = IPTOS_LOWDELAY;
1965        ep->com.local_addr = cm_id->local_addr;
1966        ep->com.remote_addr = cm_id->remote_addr;
1967
1968        /* send connect request to rnic */
1969        err = send_connect(ep);
1970        if (!err)
1971                goto out;
1972
1973        l2t_release(h->rdev.t3cdev_p, ep->l2t);
1974fail4:
1975        dst_release(ep->dst);
1976fail3:
1977        cxgb3_free_atid(ep->com.tdev, ep->atid);
1978fail2:
1979        cm_id->rem_ref(cm_id);
1980        put_ep(&ep->com);
1981out:
1982        return err;
1983}
1984
1985int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1986{
1987        int err = 0;
1988        struct iwch_dev *h = to_iwch_dev(cm_id->device);
1989        struct iwch_listen_ep *ep;
1990
1991
1992        might_sleep();
1993
1994        ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1995        if (!ep) {
1996                printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1997                err = -ENOMEM;
1998                goto fail1;
1999        }
2000        PDBG("%s ep %p\n", __func__, ep);
2001        ep->com.tdev = h->rdev.t3cdev_p;
2002        cm_id->add_ref(cm_id);
2003        ep->com.cm_id = cm_id;
2004        ep->backlog = backlog;
2005        ep->com.local_addr = cm_id->local_addr;
2006
2007        /*
2008         * Allocate a server TID.
2009         */
2010        ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
2011        if (ep->stid == -1) {
2012                printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2013                err = -ENOMEM;
2014                goto fail2;
2015        }
2016
2017        state_set(&ep->com, LISTEN);
2018        err = listen_start(ep);
2019        if (err)
2020                goto fail3;
2021
2022        /* wait for pass_open_rpl */
2023        wait_event(ep->com.waitq, ep->com.rpl_done);
2024        err = ep->com.rpl_err;
2025        if (!err) {
2026                cm_id->provider_data = ep;
2027                goto out;
2028        }
2029fail3:
2030        cxgb3_free_stid(ep->com.tdev, ep->stid);
2031fail2:
2032        cm_id->rem_ref(cm_id);
2033        put_ep(&ep->com);
2034fail1:
2035out:
2036        return err;
2037}
2038
2039int iwch_destroy_listen(struct iw_cm_id *cm_id)
2040{
2041        int err;
2042        struct iwch_listen_ep *ep = to_listen_ep(cm_id);
2043
2044        PDBG("%s ep %p\n", __func__, ep);
2045
2046        might_sleep();
2047        state_set(&ep->com, DEAD);
2048        ep->com.rpl_done = 0;
2049        ep->com.rpl_err = 0;
2050        err = listen_stop(ep);
2051        if (err)
2052                goto done;
2053        wait_event(ep->com.waitq, ep->com.rpl_done);
2054        cxgb3_free_stid(ep->com.tdev, ep->stid);
2055done:
2056        err = ep->com.rpl_err;
2057        cm_id->rem_ref(cm_id);
2058        put_ep(&ep->com);
2059        return err;
2060}
2061
2062int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
2063{
2064        int ret=0;
2065        unsigned long flags;
2066        int close = 0;
2067        int fatal = 0;
2068        struct t3cdev *tdev;
2069        struct cxio_rdev *rdev;
2070
2071        spin_lock_irqsave(&ep->com.lock, flags);
2072
2073        PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2074             states[ep->com.state], abrupt);
2075
2076        tdev = (struct t3cdev *)ep->com.tdev;
2077        rdev = (struct cxio_rdev *)tdev->ulp;
2078        if (cxio_fatal_error(rdev)) {
2079                fatal = 1;
2080                close_complete_upcall(ep);
2081                ep->com.state = DEAD;
2082        }
2083        switch (ep->com.state) {
2084        case MPA_REQ_WAIT:
2085        case MPA_REQ_SENT:
2086        case MPA_REQ_RCVD:
2087        case MPA_REP_SENT:
2088        case FPDU_MODE:
2089                close = 1;
2090                if (abrupt)
2091                        ep->com.state = ABORTING;
2092                else {
2093                        ep->com.state = CLOSING;
2094                        start_ep_timer(ep);
2095                }
2096                set_bit(CLOSE_SENT, &ep->com.flags);
2097                break;
2098        case CLOSING:
2099                if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2100                        close = 1;
2101                        if (abrupt) {
2102                                stop_ep_timer(ep);
2103                                ep->com.state = ABORTING;
2104                        } else
2105                                ep->com.state = MORIBUND;
2106                }
2107                break;
2108        case MORIBUND:
2109        case ABORTING:
2110        case DEAD:
2111                PDBG("%s ignoring disconnect ep %p state %u\n",
2112                     __func__, ep, ep->com.state);
2113                break;
2114        default:
2115                BUG();
2116                break;
2117        }
2118
2119        spin_unlock_irqrestore(&ep->com.lock, flags);
2120        if (close) {
2121                if (abrupt)
2122                        ret = send_abort(ep, NULL, gfp);
2123                else
2124                        ret = send_halfclose(ep, gfp);
2125                if (ret)
2126                        fatal = 1;
2127        }
2128        if (fatal)
2129                release_ep_resources(ep);
2130        return ret;
2131}
2132
2133int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2134                     struct l2t_entry *l2t)
2135{
2136        struct iwch_ep *ep = ctx;
2137
2138        if (ep->dst != old)
2139                return 0;
2140
2141        PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2142             l2t);
2143        dst_hold(new);
2144        l2t_release(ep->com.tdev, ep->l2t);
2145        ep->l2t = l2t;
2146        dst_release(old);
2147        ep->dst = new;
2148        return 1;
2149}
2150
2151/*
2152 * All the CM events are handled on a work queue to have a safe context.
2153 * These are the real handlers that are called from the work queue.
2154 */
2155static const cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS] = {
2156        [CPL_ACT_ESTABLISH]     = act_establish,
2157        [CPL_ACT_OPEN_RPL]      = act_open_rpl,
2158        [CPL_RX_DATA]           = rx_data,
2159        [CPL_TX_DMA_ACK]        = tx_ack,
2160        [CPL_ABORT_RPL_RSS]     = abort_rpl,
2161        [CPL_ABORT_RPL]         = abort_rpl,
2162        [CPL_PASS_OPEN_RPL]     = pass_open_rpl,
2163        [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2164        [CPL_PASS_ACCEPT_REQ]   = pass_accept_req,
2165        [CPL_PASS_ESTABLISH]    = pass_establish,
2166        [CPL_PEER_CLOSE]        = peer_close,
2167        [CPL_ABORT_REQ_RSS]     = peer_abort,
2168        [CPL_CLOSE_CON_RPL]     = close_con_rpl,
2169        [CPL_RDMA_TERMINATE]    = terminate,
2170        [CPL_RDMA_EC_STATUS]    = ec_status,
2171};
2172
2173static void process_work(struct work_struct *work)
2174{
2175        struct sk_buff *skb = NULL;
2176        void *ep;
2177        struct t3cdev *tdev;
2178        int ret;
2179
2180        while ((skb = skb_dequeue(&rxq))) {
2181                ep = *((void **) (skb->cb));
2182                tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
2183                ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
2184                if (ret & CPL_RET_BUF_DONE)
2185                        kfree_skb(skb);
2186
2187                /*
2188                 * ep was referenced in sched(), and is freed here.
2189                 */
2190                put_ep((struct iwch_ep_common *)ep);
2191        }
2192}
2193
2194static DECLARE_WORK(skb_work, process_work);
2195
2196static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2197{
2198        struct iwch_ep_common *epc = ctx;
2199
2200        get_ep(epc);
2201
2202        /*
2203         * Save ctx and tdev in the skb->cb area.
2204         */
2205        *((void **) skb->cb) = ctx;
2206        *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2207
2208        /*
2209         * Queue the skb and schedule the worker thread.
2210         */
2211        skb_queue_tail(&rxq, skb);
2212        queue_work(workq, &skb_work);
2213        return 0;
2214}
2215
2216static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2217{
2218        struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2219
2220        if (rpl->status != CPL_ERR_NONE) {
2221                printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2222                       "for tid %u\n", rpl->status, GET_TID(rpl));
2223        }
2224        return CPL_RET_BUF_DONE;
2225}
2226
2227/*
2228 * All upcalls from the T3 Core go to sched() to schedule the
2229 * processing on a work queue.
2230 */
2231cxgb3_cpl_handler_func t3c_handlers[NUM_CPL_CMDS] = {
2232        [CPL_ACT_ESTABLISH]     = sched,
2233        [CPL_ACT_OPEN_RPL]      = sched,
2234        [CPL_RX_DATA]           = sched,
2235        [CPL_TX_DMA_ACK]        = sched,
2236        [CPL_ABORT_RPL_RSS]     = sched,
2237        [CPL_ABORT_RPL]         = sched,
2238        [CPL_PASS_OPEN_RPL]     = sched,
2239        [CPL_CLOSE_LISTSRV_RPL] = sched,
2240        [CPL_PASS_ACCEPT_REQ]   = sched,
2241        [CPL_PASS_ESTABLISH]    = sched,
2242        [CPL_PEER_CLOSE]        = sched,
2243        [CPL_CLOSE_CON_RPL]     = sched,
2244        [CPL_ABORT_REQ_RSS]     = sched,
2245        [CPL_RDMA_TERMINATE]    = sched,
2246        [CPL_RDMA_EC_STATUS]    = sched,
2247        [CPL_SET_TCB_RPL]       = set_tcb_rpl,
2248};
2249
2250int __init iwch_cm_init(void)
2251{
2252        skb_queue_head_init(&rxq);
2253
2254        workq = create_singlethread_workqueue("iw_cxgb3");
2255        if (!workq)
2256                return -ENOMEM;
2257
2258        return 0;
2259}
2260
2261void __exit iwch_cm_term(void)
2262{
2263        flush_workqueue(workq);
2264        destroy_workqueue(workq);
2265}
2266
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.