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                WARN(1, "%s timer stopped when its not running!  ep %p state %u\n",
 132                        __func__, ep, ep->com.state);
 133                return;
 134        }
 135        del_timer_sync(&ep->timer);
 136        put_ep(&ep->com);
 137}
 138
 139static int iwch_l2t_send(struct t3cdev *tdev, struct sk_buff *skb, struct l2t_entry *l2e)
 140{
 141        int     error = 0;
 142        struct cxio_rdev *rdev;
 143
 144        rdev = (struct cxio_rdev *)tdev->ulp;
 145        if (cxio_fatal_error(rdev)) {
 146                kfree_skb(skb);
 147                return -EIO;
 148        }
 149        error = l2t_send(tdev, skb, l2e);
 150        if (error < 0)
 151                kfree_skb(skb);
 152        return error < 0 ? error : 0;
 153}
 154
 155int iwch_cxgb3_ofld_send(struct t3cdev *tdev, struct sk_buff *skb)
 156{
 157        int     error = 0;
 158        struct cxio_rdev *rdev;
 159
 160        rdev = (struct cxio_rdev *)tdev->ulp;
 161        if (cxio_fatal_error(rdev)) {
 162                kfree_skb(skb);
 163                return -EIO;
 164        }
 165        error = cxgb3_ofld_send(tdev, skb);
 166        if (error < 0)
 167                kfree_skb(skb);
 168        return error < 0 ? error : 0;
 169}
 170
 171static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
 172{
 173        struct cpl_tid_release *req;
 174
 175        skb = get_skb(skb, sizeof *req, GFP_KERNEL);
 176        if (!skb)
 177                return;
 178        req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
 179        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 180        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
 181        skb->priority = CPL_PRIORITY_SETUP;
 182        iwch_cxgb3_ofld_send(tdev, skb);
 183        return;
 184}
 185
 186int iwch_quiesce_tid(struct iwch_ep *ep)
 187{
 188        struct cpl_set_tcb_field *req;
 189        struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
 190
 191        if (!skb)
 192                return -ENOMEM;
 193        req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
 194        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 195        req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
 196        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
 197        req->reply = 0;
 198        req->cpu_idx = 0;
 199        req->word = htons(W_TCB_RX_QUIESCE);
 200        req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
 201        req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
 202
 203        skb->priority = CPL_PRIORITY_DATA;
 204        return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
 205}
 206
 207int iwch_resume_tid(struct iwch_ep *ep)
 208{
 209        struct cpl_set_tcb_field *req;
 210        struct sk_buff *skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
 211
 212        if (!skb)
 213                return -ENOMEM;
 214        req = (struct cpl_set_tcb_field *) skb_put(skb, sizeof(*req));
 215        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 216        req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
 217        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, ep->hwtid));
 218        req->reply = 0;
 219        req->cpu_idx = 0;
 220        req->word = htons(W_TCB_RX_QUIESCE);
 221        req->mask = cpu_to_be64(1ULL << S_TCB_RX_QUIESCE);
 222        req->val = 0;
 223
 224        skb->priority = CPL_PRIORITY_DATA;
 225        return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
 226}
 227
 228static void set_emss(struct iwch_ep *ep, u16 opt)
 229{
 230        PDBG("%s ep %p opt %u\n", __func__, ep, opt);
 231        ep->emss = T3C_DATA(ep->com.tdev)->mtus[G_TCPOPT_MSS(opt)] - 40;
 232        if (G_TCPOPT_TSTAMP(opt))
 233                ep->emss -= 12;
 234        if (ep->emss < 128)
 235                ep->emss = 128;
 236        PDBG("emss=%d\n", ep->emss);
 237}
 238
 239static enum iwch_ep_state state_read(struct iwch_ep_common *epc)
 240{
 241        unsigned long flags;
 242        enum iwch_ep_state state;
 243
 244        spin_lock_irqsave(&epc->lock, flags);
 245        state = epc->state;
 246        spin_unlock_irqrestore(&epc->lock, flags);
 247        return state;
 248}
 249
 250static void __state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
 251{
 252        epc->state = new;
 253}
 254
 255static void state_set(struct iwch_ep_common *epc, enum iwch_ep_state new)
 256{
 257        unsigned long flags;
 258
 259        spin_lock_irqsave(&epc->lock, flags);
 260        PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
 261        __state_set(epc, new);
 262        spin_unlock_irqrestore(&epc->lock, flags);
 263        return;
 264}
 265
 266static void *alloc_ep(int size, gfp_t gfp)
 267{
 268        struct iwch_ep_common *epc;
 269
 270        epc = kzalloc(size, gfp);
 271        if (epc) {
 272                kref_init(&epc->kref);
 273                spin_lock_init(&epc->lock);
 274                init_waitqueue_head(&epc->waitq);
 275        }
 276        PDBG("%s alloc ep %p\n", __func__, epc);
 277        return epc;
 278}
 279
 280void __free_ep(struct kref *kref)
 281{
 282        struct iwch_ep *ep;
 283        ep = container_of(container_of(kref, struct iwch_ep_common, kref),
 284                          struct iwch_ep, com);
 285        PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
 286        if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
 287                cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
 288                dst_release(ep->dst);
 289                l2t_release(ep->com.tdev, ep->l2t);
 290        }
 291        kfree(ep);
 292}
 293
 294static void release_ep_resources(struct iwch_ep *ep)
 295{
 296        PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
 297        set_bit(RELEASE_RESOURCES, &ep->com.flags);
 298        put_ep(&ep->com);
 299}
 300
 301static int status2errno(int status)
 302{
 303        switch (status) {
 304        case CPL_ERR_NONE:
 305                return 0;
 306        case CPL_ERR_CONN_RESET:
 307                return -ECONNRESET;
 308        case CPL_ERR_ARP_MISS:
 309                return -EHOSTUNREACH;
 310        case CPL_ERR_CONN_TIMEDOUT:
 311                return -ETIMEDOUT;
 312        case CPL_ERR_TCAM_FULL:
 313                return -ENOMEM;
 314        case CPL_ERR_CONN_EXIST:
 315                return -EADDRINUSE;
 316        default:
 317                return -EIO;
 318        }
 319}
 320
 321/*
 322 * Try and reuse skbs already allocated...
 323 */
 324static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
 325{
 326        if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
 327                skb_trim(skb, 0);
 328                skb_get(skb);
 329        } else {
 330                skb = alloc_skb(len, gfp);
 331        }
 332        return skb;
 333}
 334
 335static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
 336                                 __be32 peer_ip, __be16 local_port,
 337                                 __be16 peer_port, u8 tos)
 338{
 339        struct rtable *rt;
 340        struct flowi4 fl4;
 341
 342        rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
 343                                   peer_port, local_port, IPPROTO_TCP,
 344                                   tos, 0);
 345        if (IS_ERR(rt))
 346                return NULL;
 347        return rt;
 348}
 349
 350static unsigned int find_best_mtu(const struct t3c_data *d, unsigned short mtu)
 351{
 352        int i = 0;
 353
 354        while (i < d->nmtus - 1 && d->mtus[i + 1] <= mtu)
 355                ++i;
 356        return i;
 357}
 358
 359static void arp_failure_discard(struct t3cdev *dev, struct sk_buff *skb)
 360{
 361        PDBG("%s t3cdev %p\n", __func__, dev);
 362        kfree_skb(skb);
 363}
 364
 365/*
 366 * Handle an ARP failure for an active open.
 367 */
 368static void act_open_req_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
 369{
 370        printk(KERN_ERR MOD "ARP failure during connect\n");
 371        kfree_skb(skb);
 372}
 373
 374/*
 375 * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
 376 * and send it along.
 377 */
 378static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
 379{
 380        struct cpl_abort_req *req = cplhdr(skb);
 381
 382        PDBG("%s t3cdev %p\n", __func__, dev);
 383        req->cmd = CPL_ABORT_NO_RST;
 384        iwch_cxgb3_ofld_send(dev, skb);
 385}
 386
 387static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
 388{
 389        struct cpl_close_con_req *req;
 390        struct sk_buff *skb;
 391
 392        PDBG("%s ep %p\n", __func__, ep);
 393        skb = get_skb(NULL, sizeof(*req), gfp);
 394        if (!skb) {
 395                printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
 396                return -ENOMEM;
 397        }
 398        skb->priority = CPL_PRIORITY_DATA;
 399        set_arp_failure_handler(skb, arp_failure_discard);
 400        req = (struct cpl_close_con_req *) skb_put(skb, sizeof(*req));
 401        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
 402        req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
 403        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
 404        return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 405}
 406
 407static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
 408{
 409        struct cpl_abort_req *req;
 410
 411        PDBG("%s ep %p\n", __func__, ep);
 412        skb = get_skb(skb, sizeof(*req), gfp);
 413        if (!skb) {
 414                printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
 415                       __func__);
 416                return -ENOMEM;
 417        }
 418        skb->priority = CPL_PRIORITY_DATA;
 419        set_arp_failure_handler(skb, abort_arp_failure);
 420        req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
 421        memset(req, 0, 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        memcpy(&event.local_addr, &ep->com.local_addr,
 726               sizeof(ep->com.local_addr));
 727        memcpy(&event.remote_addr, &ep->com.remote_addr,
 728               sizeof(ep->com.remote_addr));
 729
 730        if ((status == 0) || (status == -ECONNREFUSED)) {
 731                event.private_data_len = ep->plen;
 732                event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
 733        }
 734        if (ep->com.cm_id) {
 735                PDBG("%s ep %p tid %d status %d\n", __func__, ep,
 736                     ep->hwtid, status);
 737                ep->com.cm_id->event_handler(ep->com.cm_id, &event);
 738        }
 739        if (status < 0) {
 740                ep->com.cm_id->rem_ref(ep->com.cm_id);
 741                ep->com.cm_id = NULL;
 742                ep->com.qp = NULL;
 743        }
 744}
 745
 746static void connect_request_upcall(struct iwch_ep *ep)
 747{
 748        struct iw_cm_event event;
 749
 750        PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
 751        memset(&event, 0, sizeof(event));
 752        event.event = IW_CM_EVENT_CONNECT_REQUEST;
 753        memcpy(&event.local_addr, &ep->com.local_addr,
 754               sizeof(ep->com.local_addr));
 755        memcpy(&event.remote_addr, &ep->com.remote_addr,
 756               sizeof(ep->com.local_addr));
 757        event.private_data_len = ep->plen;
 758        event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
 759        event.provider_data = ep;
 760        /*
 761         * Until ird/ord negotiation via MPAv2 support is added, send max
 762         * supported values
 763         */
 764        event.ird = event.ord = 8;
 765        if (state_read(&ep->parent_ep->com) != DEAD) {
 766                get_ep(&ep->com);
 767                ep->parent_ep->com.cm_id->event_handler(
 768                                                ep->parent_ep->com.cm_id,
 769                                                &event);
 770        }
 771        put_ep(&ep->parent_ep->com);
 772        ep->parent_ep = NULL;
 773}
 774
 775static void established_upcall(struct iwch_ep *ep)
 776{
 777        struct iw_cm_event event;
 778
 779        PDBG("%s ep %p\n", __func__, ep);
 780        memset(&event, 0, sizeof(event));
 781        event.event = IW_CM_EVENT_ESTABLISHED;
 782        /*
 783         * Until ird/ord negotiation via MPAv2 support is added, send max
 784         * supported values
 785         */
 786        event.ird = event.ord = 8;
 787        if (ep->com.cm_id) {
 788                PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
 789                ep->com.cm_id->event_handler(ep->com.cm_id, &event);
 790        }
 791}
 792
 793static int update_rx_credits(struct iwch_ep *ep, u32 credits)
 794{
 795        struct cpl_rx_data_ack *req;
 796        struct sk_buff *skb;
 797
 798        PDBG("%s ep %p credits %u\n", __func__, ep, credits);
 799        skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
 800        if (!skb) {
 801                printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
 802                return 0;
 803        }
 804
 805        req = (struct cpl_rx_data_ack *) skb_put(skb, sizeof(*req));
 806        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 807        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
 808        req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
 809        skb->priority = CPL_PRIORITY_ACK;
 810        iwch_cxgb3_ofld_send(ep->com.tdev, skb);
 811        return credits;
 812}
 813
 814static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
 815{
 816        struct mpa_message *mpa;
 817        u16 plen;
 818        struct iwch_qp_attributes attrs;
 819        enum iwch_qp_attr_mask mask;
 820        int err;
 821
 822        PDBG("%s ep %p\n", __func__, ep);
 823
 824        /*
 825         * Stop mpa timer.  If it expired, then the state has
 826         * changed and we bail since ep_timeout already aborted
 827         * the connection.
 828         */
 829        stop_ep_timer(ep);
 830        if (state_read(&ep->com) != MPA_REQ_SENT)
 831                return;
 832
 833        /*
 834         * If we get more than the supported amount of private data
 835         * then we must fail this connection.
 836         */
 837        if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
 838                err = -EINVAL;
 839                goto err;
 840        }
 841
 842        /*
 843         * copy the new data into our accumulation buffer.
 844         */
 845        skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
 846                                  skb->len);
 847        ep->mpa_pkt_len += skb->len;
 848
 849        /*
 850         * if we don't even have the mpa message, then bail.
 851         */
 852        if (ep->mpa_pkt_len < sizeof(*mpa))
 853                return;
 854        mpa = (struct mpa_message *) ep->mpa_pkt;
 855
 856        /* Validate MPA header. */
 857        if (mpa->revision != mpa_rev) {
 858                err = -EPROTO;
 859                goto err;
 860        }
 861        if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
 862                err = -EPROTO;
 863                goto err;
 864        }
 865
 866        plen = ntohs(mpa->private_data_size);
 867
 868        /*
 869         * Fail if there's too much private data.
 870         */
 871        if (plen > MPA_MAX_PRIVATE_DATA) {
 872                err = -EPROTO;
 873                goto err;
 874        }
 875
 876        /*
 877         * If plen does not account for pkt size
 878         */
 879        if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
 880                err = -EPROTO;
 881                goto err;
 882        }
 883
 884        ep->plen = (u8) plen;
 885
 886        /*
 887         * If we don't have all the pdata yet, then bail.
 888         * We'll continue process when more data arrives.
 889         */
 890        if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
 891                return;
 892
 893        if (mpa->flags & MPA_REJECT) {
 894                err = -ECONNREFUSED;
 895                goto err;
 896        }
 897
 898        /*
 899         * If we get here we have accumulated the entire mpa
 900         * start reply message including private data. And
 901         * the MPA header is valid.
 902         */
 903        state_set(&ep->com, FPDU_MODE);
 904        ep->mpa_attr.initiator = 1;
 905        ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
 906        ep->mpa_attr.recv_marker_enabled = markers_enabled;
 907        ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
 908        ep->mpa_attr.version = mpa_rev;
 909        PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
 910             "xmit_marker_enabled=%d, version=%d\n", __func__,
 911             ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
 912             ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
 913
 914        attrs.mpa_attr = ep->mpa_attr;
 915        attrs.max_ird = ep->ird;
 916        attrs.max_ord = ep->ord;
 917        attrs.llp_stream_handle = ep;
 918        attrs.next_state = IWCH_QP_STATE_RTS;
 919
 920        mask = IWCH_QP_ATTR_NEXT_STATE |
 921            IWCH_QP_ATTR_LLP_STREAM_HANDLE | IWCH_QP_ATTR_MPA_ATTR |
 922            IWCH_QP_ATTR_MAX_IRD | IWCH_QP_ATTR_MAX_ORD;
 923
 924        /* bind QP and TID with INIT_WR */
 925        err = iwch_modify_qp(ep->com.qp->rhp,
 926                             ep->com.qp, mask, &attrs, 1);
 927        if (err)
 928                goto err;
 929
 930        if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
 931                iwch_post_zb_read(ep);
 932        }
 933
 934        goto out;
 935err:
 936        abort_connection(ep, skb, GFP_KERNEL);
 937out:
 938        connect_reply_upcall(ep, err);
 939        return;
 940}
 941
 942static void process_mpa_request(struct iwch_ep *ep, struct sk_buff *skb)
 943{
 944        struct mpa_message *mpa;
 945        u16 plen;
 946
 947        PDBG("%s ep %p\n", __func__, ep);
 948
 949        /*
 950         * Stop mpa timer.  If it expired, then the state has
 951         * changed and we bail since ep_timeout already aborted
 952         * the connection.
 953         */
 954        stop_ep_timer(ep);
 955        if (state_read(&ep->com) != MPA_REQ_WAIT)
 956                return;
 957
 958        /*
 959         * If we get more than the supported amount of private data
 960         * then we must fail this connection.
 961         */
 962        if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
 963                abort_connection(ep, skb, GFP_KERNEL);
 964                return;
 965        }
 966
 967        PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
 968
 969        /*
 970         * Copy the new data into our accumulation buffer.
 971         */
 972        skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
 973                                  skb->len);
 974        ep->mpa_pkt_len += skb->len;
 975
 976        /*
 977         * If we don't even have the mpa message, then bail.
 978         * We'll continue process when more data arrives.
 979         */
 980        if (ep->mpa_pkt_len < sizeof(*mpa))
 981                return;
 982        PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
 983        mpa = (struct mpa_message *) ep->mpa_pkt;
 984
 985        /*
 986         * Validate MPA Header.
 987         */
 988        if (mpa->revision != mpa_rev) {
 989                abort_connection(ep, skb, GFP_KERNEL);
 990                return;
 991        }
 992
 993        if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
 994                abort_connection(ep, skb, GFP_KERNEL);
 995                return;
 996        }
 997
 998        plen = ntohs(mpa->private_data_size);
 999
1000        /*
1001         * Fail if there's too much private data.
1002         */
1003        if (plen > MPA_MAX_PRIVATE_DATA) {
1004                abort_connection(ep, skb, GFP_KERNEL);
1005                return;
1006        }
1007
1008        /*
1009         * If plen does not account for pkt size
1010         */
1011        if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1012                abort_connection(ep, skb, GFP_KERNEL);
1013                return;
1014        }
1015        ep->plen = (u8) plen;
1016
1017        /*
1018         * If we don't have all the pdata yet, then bail.
1019         */
1020        if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1021                return;
1022
1023        /*
1024         * If we get here we have accumulated the entire mpa
1025         * start reply message including private data.
1026         */
1027        ep->mpa_attr.initiator = 0;
1028        ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1029        ep->mpa_attr.recv_marker_enabled = markers_enabled;
1030        ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1031        ep->mpa_attr.version = mpa_rev;
1032        PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1033             "xmit_marker_enabled=%d, version=%d\n", __func__,
1034             ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1035             ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1036
1037        state_set(&ep->com, MPA_REQ_RCVD);
1038
1039        /* drive upcall */
1040        connect_request_upcall(ep);
1041        return;
1042}
1043
1044static int rx_data(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1045{
1046        struct iwch_ep *ep = ctx;
1047        struct cpl_rx_data *hdr = cplhdr(skb);
1048        unsigned int dlen = ntohs(hdr->len);
1049
1050        PDBG("%s ep %p dlen %u\n", __func__, ep, dlen);
1051
1052        skb_pull(skb, sizeof(*hdr));
1053        skb_trim(skb, dlen);
1054
1055        ep->rcv_seq += dlen;
1056        BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1057
1058        switch (state_read(&ep->com)) {
1059        case MPA_REQ_SENT:
1060                process_mpa_reply(ep, skb);
1061                break;
1062        case MPA_REQ_WAIT:
1063                process_mpa_request(ep, skb);
1064                break;
1065        case MPA_REP_SENT:
1066                break;
1067        default:
1068                printk(KERN_ERR MOD "%s Unexpected streaming data."
1069                       " ep %p state %d tid %d\n",
1070                       __func__, ep, state_read(&ep->com), ep->hwtid);
1071
1072                /*
1073                 * The ep will timeout and inform the ULP of the failure.
1074                 * See ep_timeout().
1075                 */
1076                break;
1077        }
1078
1079        /* update RX credits */
1080        update_rx_credits(ep, dlen);
1081
1082        return CPL_RET_BUF_DONE;
1083}
1084
1085/*
1086 * Upcall from the adapter indicating data has been transmitted.
1087 * For us its just the single MPA request or reply.  We can now free
1088 * the skb holding the mpa message.
1089 */
1090static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1091{
1092        struct iwch_ep *ep = ctx;
1093        struct cpl_wr_ack *hdr = cplhdr(skb);
1094        unsigned int credits = ntohs(hdr->credits);
1095        unsigned long flags;
1096        int post_zb = 0;
1097
1098        PDBG("%s ep %p credits %u\n", __func__, ep, credits);
1099
1100        if (credits == 0) {
1101                PDBG("%s 0 credit ack  ep %p state %u\n",
1102                     __func__, ep, state_read(&ep->com));
1103                return CPL_RET_BUF_DONE;
1104        }
1105
1106        spin_lock_irqsave(&ep->com.lock, flags);
1107        BUG_ON(credits != 1);
1108        dst_confirm(ep->dst);
1109        if (!ep->mpa_skb) {
1110                PDBG("%s rdma_init wr_ack ep %p state %u\n",
1111                        __func__, ep, ep->com.state);
1112                if (ep->mpa_attr.initiator) {
1113                        PDBG("%s initiator ep %p state %u\n",
1114                                __func__, ep, ep->com.state);
1115                        if (peer2peer && ep->com.state == FPDU_MODE)
1116                                post_zb = 1;
1117                } else {
1118                        PDBG("%s responder ep %p state %u\n",
1119                                __func__, ep, ep->com.state);
1120                        if (ep->com.state == MPA_REQ_RCVD) {
1121                                ep->com.rpl_done = 1;
1122                                wake_up(&ep->com.waitq);
1123                        }
1124                }
1125        } else {
1126                PDBG("%s lsm ack ep %p state %u freeing skb\n",
1127                        __func__, ep, ep->com.state);
1128                kfree_skb(ep->mpa_skb);
1129                ep->mpa_skb = NULL;
1130        }
1131        spin_unlock_irqrestore(&ep->com.lock, flags);
1132        if (post_zb)
1133                iwch_post_zb_read(ep);
1134        return CPL_RET_BUF_DONE;
1135}
1136
1137static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1138{
1139        struct iwch_ep *ep = ctx;
1140        unsigned long flags;
1141        int release = 0;
1142
1143        PDBG("%s ep %p\n", __func__, ep);
1144        BUG_ON(!ep);
1145
1146        /*
1147         * We get 2 abort replies from the HW.  The first one must
1148         * be ignored except for scribbling that we need one more.
1149         */
1150        if (!test_and_set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) {
1151                return CPL_RET_BUF_DONE;
1152        }
1153
1154        spin_lock_irqsave(&ep->com.lock, flags);
1155        switch (ep->com.state) {
1156        case ABORTING:
1157                close_complete_upcall(ep);
1158                __state_set(&ep->com, DEAD);
1159                release = 1;
1160                break;
1161        default:
1162                printk(KERN_ERR "%s ep %p state %d\n",
1163                     __func__, ep, ep->com.state);
1164                break;
1165        }
1166        spin_unlock_irqrestore(&ep->com.lock, flags);
1167
1168        if (release)
1169                release_ep_resources(ep);
1170        return CPL_RET_BUF_DONE;
1171}
1172
1173/*
1174 * Return whether a failed active open has allocated a TID
1175 */
1176static inline int act_open_has_tid(int status)
1177{
1178        return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1179               status != CPL_ERR_ARP_MISS;
1180}
1181
1182static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1183{
1184        struct iwch_ep *ep = ctx;
1185        struct cpl_act_open_rpl *rpl = cplhdr(skb);
1186
1187        PDBG("%s ep %p status %u errno %d\n", __func__, ep, rpl->status,
1188             status2errno(rpl->status));
1189        connect_reply_upcall(ep, status2errno(rpl->status));
1190        state_set(&ep->com, DEAD);
1191        if (ep->com.tdev->type != T3A && act_open_has_tid(rpl->status))
1192                release_tid(ep->com.tdev, GET_TID(rpl), NULL);
1193        cxgb3_free_atid(ep->com.tdev, ep->atid);
1194        dst_release(ep->dst);
1195        l2t_release(ep->com.tdev, ep->l2t);
1196        put_ep(&ep->com);
1197        return CPL_RET_BUF_DONE;
1198}
1199
1200static int listen_start(struct iwch_listen_ep *ep)
1201{
1202        struct sk_buff *skb;
1203        struct cpl_pass_open_req *req;
1204
1205        PDBG("%s ep %p\n", __func__, ep);
1206        skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1207        if (!skb) {
1208                printk(KERN_ERR MOD "t3c_listen_start failed to alloc skb!\n");
1209                return -ENOMEM;
1210        }
1211
1212        req = (struct cpl_pass_open_req *) skb_put(skb, sizeof(*req));
1213        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1214        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, ep->stid));
1215        req->local_port = ep->com.local_addr.sin_port;
1216        req->local_ip = ep->com.local_addr.sin_addr.s_addr;
1217        req->peer_port = 0;
1218        req->peer_ip = 0;
1219        req->peer_netmask = 0;
1220        req->opt0h = htonl(F_DELACK | F_TCAM_BYPASS);
1221        req->opt0l = htonl(V_RCV_BUFSIZ(rcv_win>>10));
1222        req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
1223
1224        skb->priority = 1;
1225        return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1226}
1227
1228static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1229{
1230        struct iwch_listen_ep *ep = ctx;
1231        struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1232
1233        PDBG("%s ep %p status %d error %d\n", __func__, ep,
1234             rpl->status, status2errno(rpl->status));
1235        ep->com.rpl_err = status2errno(rpl->status);
1236        ep->com.rpl_done = 1;
1237        wake_up(&ep->com.waitq);
1238
1239        return CPL_RET_BUF_DONE;
1240}
1241
1242static int listen_stop(struct iwch_listen_ep *ep)
1243{
1244        struct sk_buff *skb;
1245        struct cpl_close_listserv_req *req;
1246
1247        PDBG("%s ep %p\n", __func__, ep);
1248        skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1249        if (!skb) {
1250                printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1251                return -ENOMEM;
1252        }
1253        req = (struct cpl_close_listserv_req *) skb_put(skb, sizeof(*req));
1254        req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1255        req->cpu_idx = 0;
1256        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
1257        skb->priority = 1;
1258        return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
1259}
1260
1261static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
1262                             void *ctx)
1263{
1264        struct iwch_listen_ep *ep = ctx;
1265        struct cpl_close_listserv_rpl *rpl = cplhdr(skb);
1266
1267        PDBG("%s ep %p\n", __func__, ep);
1268        ep->com.rpl_err = status2errno(rpl->status);
1269        ep->com.rpl_done = 1;
1270        wake_up(&ep->com.waitq);
1271        return CPL_RET_BUF_DONE;
1272}
1273
1274static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
1275{
1276        struct cpl_pass_accept_rpl *rpl;
1277        unsigned int mtu_idx;
1278        u32 opt0h, opt0l, opt2;
1279        int wscale;
1280
1281        PDBG("%s ep %p\n", __func__, ep);
1282        BUG_ON(skb_cloned(skb));
1283        skb_trim(skb, sizeof(*rpl));
1284        skb_get(skb);
1285        mtu_idx = find_best_mtu(T3C_DATA(ep->com.tdev), dst_mtu(ep->dst));
1286        wscale = compute_wscale(rcv_win);
1287        opt0h = V_NAGLE(0) |
1288            V_NO_CONG(nocong) |
1289            V_KEEP_ALIVE(1) |
1290            F_TCAM_BYPASS |
1291            V_WND_SCALE(wscale) |
1292            V_MSS_IDX(mtu_idx) |
1293            V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
1294        opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
1295        opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
1296               V_CONG_CONTROL_FLAVOR(cong_flavor);
1297
1298        rpl = cplhdr(skb);
1299        rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1300        OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, ep->hwtid));
1301        rpl->peer_ip = peer_ip;
1302        rpl->opt0h = htonl(opt0h);
1303        rpl->opt0l_status = htonl(opt0l | CPL_PASS_OPEN_ACCEPT);
1304        rpl->opt2 = htonl(opt2);
1305        rpl->rsvd = rpl->opt2;  /* workaround for HW bug */
1306        skb->priority = CPL_PRIORITY_SETUP;
1307        iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
1308
1309        return;
1310}
1311
1312static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
1313                      struct sk_buff *skb)
1314{
1315        PDBG("%s t3cdev %p tid %u peer_ip %x\n", __func__, tdev, hwtid,
1316             peer_ip);
1317        BUG_ON(skb_cloned(skb));
1318        skb_trim(skb, sizeof(struct cpl_tid_release));
1319        skb_get(skb);
1320
1321        if (tdev->type != T3A)
1322                release_tid(tdev, hwtid, skb);
1323        else {
1324                struct cpl_pass_accept_rpl *rpl;
1325
1326                rpl = cplhdr(skb);
1327                skb->priority = CPL_PRIORITY_SETUP;
1328                rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1329                OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1330                                                      hwtid));
1331                rpl->peer_ip = peer_ip;
1332                rpl->opt0h = htonl(F_TCAM_BYPASS);
1333                rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
1334                rpl->opt2 = 0;
1335                rpl->rsvd = rpl->opt2;
1336                iwch_cxgb3_ofld_send(tdev, skb);
1337        }
1338}
1339
1340static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
1341{
1342        struct iwch_ep *child_ep, *parent_ep = ctx;
1343        struct cpl_pass_accept_req *req = cplhdr(skb);
1344        unsigned int hwtid = GET_TID(req);
1345        struct dst_entry *dst;
1346        struct l2t_entry *l2t;
1347        struct rtable *rt;
1348        struct iff_mac tim;
1349
1350        PDBG("%s parent ep %p tid %u\n", __func__, parent_ep, hwtid);
1351
1352        if (state_read(&parent_ep->com) != LISTEN) {
1353                printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1354                       __func__);
1355                goto reject;
1356        }
1357
1358        /*
1359         * Find the netdev for this connection request.
1360         */
1361        tim.mac_addr = req->dst_mac;
1362        tim.vlan_tag = ntohs(req->vlan_tag);
1363        if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
1364                printk(KERN_ERR "%s bad dst mac %pM\n",
1365                        __func__, req->dst_mac);
1366                goto reject;
1367        }
1368
1369        /* Find output route */
1370        rt = find_route(tdev,
1371                        req->local_ip,
1372                        req->peer_ip,
1373                        req->local_port,
1374                        req->peer_port, G_PASS_OPEN_TOS(ntohl(req->tos_tid)));
1375        if (!rt) {
1376                printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1377                       __func__);
1378                goto reject;
1379        }
1380        dst = &rt->dst;
1381        l2t = t3_l2t_get(tdev, dst, NULL, &req->peer_ip);
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 = AF_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 = AF_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 opcode 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                WARN(1, "%s unexpected state ep %p state %u\n",
1764                        __func__, ep, ep->com.state);
1765                abort = 0;
1766        }
1767        spin_unlock_irqrestore(&ep->com.lock, flags);
1768        if (abort)
1769                abort_connection(ep, NULL, GFP_ATOMIC);
1770        put_ep(&ep->com);
1771}
1772
1773int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1774{
1775        int err;
1776        struct iwch_ep *ep = to_ep(cm_id);
1777        PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1778
1779        if (state_read(&ep->com) == DEAD) {
1780                put_ep(&ep->com);
1781                return -ECONNRESET;
1782        }
1783        BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1784        if (mpa_rev == 0)
1785                abort_connection(ep, NULL, GFP_KERNEL);
1786        else {
1787                err = send_mpa_reject(ep, pdata, pdata_len);
1788                err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
1789        }
1790        put_ep(&ep->com);
1791        return 0;
1792}
1793
1794int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1795{
1796        int err;
1797        struct iwch_qp_attributes attrs;
1798        enum iwch_qp_attr_mask mask;
1799        struct iwch_ep *ep = to_ep(cm_id);
1800        struct iwch_dev *h = to_iwch_dev(cm_id->device);
1801        struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
1802
1803        PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1804        if (state_read(&ep->com) == DEAD) {
1805                err = -ECONNRESET;
1806                goto err;
1807        }
1808
1809        BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1810        BUG_ON(!qp);
1811
1812        if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
1813            (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
1814                abort_connection(ep, NULL, GFP_KERNEL);
1815                err = -EINVAL;
1816                goto err;
1817        }
1818
1819        cm_id->add_ref(cm_id);
1820        ep->com.cm_id = cm_id;
1821        ep->com.qp = qp;
1822
1823        ep->ird = conn_param->ird;
1824        ep->ord = conn_param->ord;
1825
1826        if (peer2peer && ep->ird == 0)
1827                ep->ird = 1;
1828
1829        PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
1830
1831        /* bind QP to EP and move to RTS */
1832        attrs.mpa_attr = ep->mpa_attr;
1833        attrs.max_ird = ep->ird;
1834        attrs.max_ord = ep->ord;
1835        attrs.llp_stream_handle = ep;
1836        attrs.next_state = IWCH_QP_STATE_RTS;
1837
1838        /* bind QP and TID with INIT_WR */
1839        mask = IWCH_QP_ATTR_NEXT_STATE |
1840                             IWCH_QP_ATTR_LLP_STREAM_HANDLE |
1841                             IWCH_QP_ATTR_MPA_ATTR |
1842                             IWCH_QP_ATTR_MAX_IRD |
1843                             IWCH_QP_ATTR_MAX_ORD;
1844
1845        err = iwch_modify_qp(ep->com.qp->rhp,
1846                             ep->com.qp, mask, &attrs, 1);
1847        if (err)
1848                goto err1;
1849
1850        /* if needed, wait for wr_ack */
1851        if (iwch_rqes_posted(qp)) {
1852                wait_event(ep->com.waitq, ep->com.rpl_done);
1853                err = ep->com.rpl_err;
1854                if (err)
1855                        goto err1;
1856        }
1857
1858        err = send_mpa_reply(ep, conn_param->private_data,
1859                             conn_param->private_data_len);
1860        if (err)
1861                goto err1;
1862
1863
1864        state_set(&ep->com, FPDU_MODE);
1865        established_upcall(ep);
1866        put_ep(&ep->com);
1867        return 0;
1868err1:
1869        ep->com.cm_id = NULL;
1870        ep->com.qp = NULL;
1871        cm_id->rem_ref(cm_id);
1872err:
1873        put_ep(&ep->com);
1874        return err;
1875}
1876
1877static int is_loopback_dst(struct iw_cm_id *cm_id)
1878{
1879        struct net_device *dev;
1880        struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
1881
1882        dev = ip_dev_find(&init_net, raddr->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 iwch_ep *ep;
1893        struct rtable *rt;
1894        int err = 0;
1895        struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
1896        struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
1897
1898        if (cm_id->m_remote_addr.ss_family != PF_INET) {
1899                err = -ENOSYS;
1900                goto out;
1901        }
1902
1903        if (is_loopback_dst(cm_id)) {
1904                err = -ENOSYS;
1905                goto out;
1906        }
1907
1908        ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
1909        if (!ep) {
1910                printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
1911                err = -ENOMEM;
1912                goto out;
1913        }
1914        init_timer(&ep->timer);
1915        ep->plen = conn_param->private_data_len;
1916        if (ep->plen)
1917                memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
1918                       conn_param->private_data, ep->plen);
1919        ep->ird = conn_param->ird;
1920        ep->ord = conn_param->ord;
1921
1922        if (peer2peer && ep->ord == 0)
1923                ep->ord = 1;
1924
1925        ep->com.tdev = h->rdev.t3cdev_p;
1926
1927        cm_id->add_ref(cm_id);
1928        ep->com.cm_id = cm_id;
1929        ep->com.qp = get_qhp(h, conn_param->qpn);
1930        BUG_ON(!ep->com.qp);
1931        PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
1932             ep->com.qp, cm_id);
1933
1934        /*
1935         * Allocate an active TID to initiate a TCP connection.
1936         */
1937        ep->atid = cxgb3_alloc_atid(h->rdev.t3cdev_p, &t3c_client, ep);
1938        if (ep->atid == -1) {
1939                printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1940                err = -ENOMEM;
1941                goto fail2;
1942        }
1943
1944        /* find a route */
1945        rt = find_route(h->rdev.t3cdev_p, laddr->sin_addr.s_addr,
1946                        raddr->sin_addr.s_addr, laddr->sin_port,
1947                        raddr->sin_port, IPTOS_LOWDELAY);
1948        if (!rt) {
1949                printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
1950                err = -EHOSTUNREACH;
1951                goto fail3;
1952        }
1953        ep->dst = &rt->dst;
1954        ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst, NULL,
1955                             &raddr->sin_addr.s_addr);
1956        if (!ep->l2t) {
1957                printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
1958                err = -ENOMEM;
1959                goto fail4;
1960        }
1961
1962        state_set(&ep->com, CONNECTING);
1963        ep->tos = IPTOS_LOWDELAY;
1964        memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
1965               sizeof(ep->com.local_addr));
1966        memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
1967               sizeof(ep->com.remote_addr));
1968
1969        /* send connect request to rnic */
1970        err = send_connect(ep);
1971        if (!err)
1972                goto out;
1973
1974        l2t_release(h->rdev.t3cdev_p, ep->l2t);
1975fail4:
1976        dst_release(ep->dst);
1977fail3:
1978        cxgb3_free_atid(ep->com.tdev, ep->atid);
1979fail2:
1980        cm_id->rem_ref(cm_id);
1981        put_ep(&ep->com);
1982out:
1983        return err;
1984}
1985
1986int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
1987{
1988        int err = 0;
1989        struct iwch_dev *h = to_iwch_dev(cm_id->device);
1990        struct iwch_listen_ep *ep;
1991
1992
1993        might_sleep();
1994
1995        if (cm_id->m_local_addr.ss_family != PF_INET) {
1996                err = -ENOSYS;
1997                goto fail1;
1998        }
1999
2000        ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2001        if (!ep) {
2002                printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2003                err = -ENOMEM;
2004                goto fail1;
2005        }
2006        PDBG("%s ep %p\n", __func__, ep);
2007        ep->com.tdev = h->rdev.t3cdev_p;
2008        cm_id->add_ref(cm_id);
2009        ep->com.cm_id = cm_id;
2010        ep->backlog = backlog;
2011        memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
2012               sizeof(ep->com.local_addr));
2013
2014        /*
2015         * Allocate a server TID.
2016         */
2017        ep->stid = cxgb3_alloc_stid(h->rdev.t3cdev_p, &t3c_client, ep);
2018        if (ep->stid == -1) {
2019                printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2020                err = -ENOMEM;
2021                goto fail2;
2022        }
2023
2024        state_set(&ep->com, LISTEN);
2025        err = listen_start(ep);
2026        if (err)
2027                goto fail3;
2028
2029        /* wait for pass_open_rpl */
2030        wait_event(ep->com.waitq, ep->com.rpl_done);
2031        err = ep->com.rpl_err;
2032        if (!err) {
2033                cm_id->provider_data = ep;
2034                goto out;
2035        }
2036fail3:
2037        cxgb3_free_stid(ep->com.tdev, ep->stid);
2038fail2:
2039        cm_id->rem_ref(cm_id);
2040        put_ep(&ep->com);
2041fail1:
2042out:
2043        return err;
2044}
2045
2046int iwch_destroy_listen(struct iw_cm_id *cm_id)
2047{
2048        int err;
2049        struct iwch_listen_ep *ep = to_listen_ep(cm_id);
2050
2051        PDBG("%s ep %p\n", __func__, ep);
2052
2053        might_sleep();
2054        state_set(&ep->com, DEAD);
2055        ep->com.rpl_done = 0;
2056        ep->com.rpl_err = 0;
2057        err = listen_stop(ep);
2058        if (err)
2059                goto done;
2060        wait_event(ep->com.waitq, ep->com.rpl_done);
2061        cxgb3_free_stid(ep->com.tdev, ep->stid);
2062done:
2063        err = ep->com.rpl_err;
2064        cm_id->rem_ref(cm_id);
2065        put_ep(&ep->com);
2066        return err;
2067}
2068
2069int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
2070{
2071        int ret=0;
2072        unsigned long flags;
2073        int close = 0;
2074        int fatal = 0;
2075        struct t3cdev *tdev;
2076        struct cxio_rdev *rdev;
2077
2078        spin_lock_irqsave(&ep->com.lock, flags);
2079
2080        PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2081             states[ep->com.state], abrupt);
2082
2083        tdev = (struct t3cdev *)ep->com.tdev;
2084        rdev = (struct cxio_rdev *)tdev->ulp;
2085        if (cxio_fatal_error(rdev)) {
2086                fatal = 1;
2087                close_complete_upcall(ep);
2088                ep->com.state = DEAD;
2089        }
2090        switch (ep->com.state) {
2091        case MPA_REQ_WAIT:
2092        case MPA_REQ_SENT:
2093        case MPA_REQ_RCVD:
2094        case MPA_REP_SENT:
2095        case FPDU_MODE:
2096                close = 1;
2097                if (abrupt)
2098                        ep->com.state = ABORTING;
2099                else {
2100                        ep->com.state = CLOSING;
2101                        start_ep_timer(ep);
2102                }
2103                set_bit(CLOSE_SENT, &ep->com.flags);
2104                break;
2105        case CLOSING:
2106                if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2107                        close = 1;
2108                        if (abrupt) {
2109                                stop_ep_timer(ep);
2110                                ep->com.state = ABORTING;
2111                        } else
2112                                ep->com.state = MORIBUND;
2113                }
2114                break;
2115        case MORIBUND:
2116        case ABORTING:
2117        case DEAD:
2118                PDBG("%s ignoring disconnect ep %p state %u\n",
2119                     __func__, ep, ep->com.state);
2120                break;
2121        default:
2122                BUG();
2123                break;
2124        }
2125
2126        spin_unlock_irqrestore(&ep->com.lock, flags);
2127        if (close) {
2128                if (abrupt)
2129                        ret = send_abort(ep, NULL, gfp);
2130                else
2131                        ret = send_halfclose(ep, gfp);
2132                if (ret)
2133                        fatal = 1;
2134        }
2135        if (fatal)
2136                release_ep_resources(ep);
2137        return ret;
2138}
2139
2140int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2141                     struct l2t_entry *l2t)
2142{
2143        struct iwch_ep *ep = ctx;
2144
2145        if (ep->dst != old)
2146                return 0;
2147
2148        PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2149             l2t);
2150        dst_hold(new);
2151        l2t_release(ep->com.tdev, ep->l2t);
2152        ep->l2t = l2t;
2153        dst_release(old);
2154        ep->dst = new;
2155        return 1;
2156}
2157
2158/*
2159 * All the CM events are handled on a work queue to have a safe context.
2160 * These are the real handlers that are called from the work queue.
2161 */
2162static const cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS] = {
2163        [CPL_ACT_ESTABLISH]     = act_establish,
2164        [CPL_ACT_OPEN_RPL]      = act_open_rpl,
2165        [CPL_RX_DATA]           = rx_data,
2166        [CPL_TX_DMA_ACK]        = tx_ack,
2167        [CPL_ABORT_RPL_RSS]     = abort_rpl,
2168        [CPL_ABORT_RPL]         = abort_rpl,
2169        [CPL_PASS_OPEN_RPL]     = pass_open_rpl,
2170        [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2171        [CPL_PASS_ACCEPT_REQ]   = pass_accept_req,
2172        [CPL_PASS_ESTABLISH]    = pass_establish,
2173        [CPL_PEER_CLOSE]        = peer_close,
2174        [CPL_ABORT_REQ_RSS]     = peer_abort,
2175        [CPL_CLOSE_CON_RPL]     = close_con_rpl,
2176        [CPL_RDMA_TERMINATE]    = terminate,
2177        [CPL_RDMA_EC_STATUS]    = ec_status,
2178};
2179
2180static void process_work(struct work_struct *work)
2181{
2182        struct sk_buff *skb = NULL;
2183        void *ep;
2184        struct t3cdev *tdev;
2185        int ret;
2186
2187        while ((skb = skb_dequeue(&rxq))) {
2188                ep = *((void **) (skb->cb));
2189                tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
2190                ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
2191                if (ret & CPL_RET_BUF_DONE)
2192                        kfree_skb(skb);
2193
2194                /*
2195                 * ep was referenced in sched(), and is freed here.
2196                 */
2197                put_ep((struct iwch_ep_common *)ep);
2198        }
2199}
2200
2201static DECLARE_WORK(skb_work, process_work);
2202
2203static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2204{
2205        struct iwch_ep_common *epc = ctx;
2206
2207        get_ep(epc);
2208
2209        /*
2210         * Save ctx and tdev in the skb->cb area.
2211         */
2212        *((void **) skb->cb) = ctx;
2213        *((struct t3cdev **) (skb->cb + sizeof(void *))) = tdev;
2214
2215        /*
2216         * Queue the skb and schedule the worker thread.
2217         */
2218        skb_queue_tail(&rxq, skb);
2219        queue_work(workq, &skb_work);
2220        return 0;
2221}
2222
2223static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
2224{
2225        struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
2226
2227        if (rpl->status != CPL_ERR_NONE) {
2228                printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
2229                       "for tid %u\n", rpl->status, GET_TID(rpl));
2230        }
2231        return CPL_RET_BUF_DONE;
2232}
2233
2234/*
2235 * All upcalls from the T3 Core go to sched() to schedule the
2236 * processing on a work queue.
2237 */
2238cxgb3_cpl_handler_func t3c_handlers[NUM_CPL_CMDS] = {
2239        [CPL_ACT_ESTABLISH]     = sched,
2240        [CPL_ACT_OPEN_RPL]      = sched,
2241        [CPL_RX_DATA]           = sched,
2242        [CPL_TX_DMA_ACK]        = sched,
2243        [CPL_ABORT_RPL_RSS]     = sched,
2244        [CPL_ABORT_RPL]         = sched,
2245        [CPL_PASS_OPEN_RPL]     = sched,
2246        [CPL_CLOSE_LISTSRV_RPL] = sched,
2247        [CPL_PASS_ACCEPT_REQ]   = sched,
2248        [CPL_PASS_ESTABLISH]    = sched,
2249        [CPL_PEER_CLOSE]        = sched,
2250        [CPL_CLOSE_CON_RPL]     = sched,
2251        [CPL_ABORT_REQ_RSS]     = sched,
2252        [CPL_RDMA_TERMINATE]    = sched,
2253        [CPL_RDMA_EC_STATUS]    = sched,
2254        [CPL_SET_TCB_RPL]       = set_tcb_rpl,
2255};
2256
2257int __init iwch_cm_init(void)
2258{
2259        skb_queue_head_init(&rxq);
2260
2261        workq = alloc_ordered_workqueue("iw_cxgb3", WQ_MEM_RECLAIM);
2262        if (!workq)
2263                return -ENOMEM;
2264
2265        return 0;
2266}
2267
2268void __exit iwch_cm_term(void)
2269{
2270        flush_workqueue(workq);
2271        destroy_workqueue(workq);
2272}
2273