drivers/net/ethernet/chelsio/cxgb3/cxgb3_offload.c

   1 /*
   2  * Copyright (c) 2006-2008 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
  33 #include <linux/list.h>
  34 #include <linux/slab.h>
  35 #include <net/neighbour.h>
  36 #include <linux/notifier.h>
  37 #include <linux/atomic.h>
  38 #include <linux/proc_fs.h>
  39 #include <linux/if_vlan.h>
  40 #include <net/netevent.h>
  41 #include <linux/highmem.h>
  42 #include <linux/vmalloc.h>
  43 #include <linux/export.h>
  44
  45 #include "common.h"
  46 #include "regs.h"
  47 #include "cxgb3_ioctl.h"
  48 #include "cxgb3_ctl_defs.h"
  49 #include "cxgb3_defs.h"
  50 #include "l2t.h"
  51 #include "firmware_exports.h"
  52 #include "cxgb3_offload.h"
  53
  54 static LIST_HEAD(client_list);
  55 static LIST_HEAD(ofld_dev_list);
  56 static DEFINE_MUTEX(cxgb3_db_lock);
  57
  58 static DEFINE_RWLOCK(adapter_list_lock);
  59 static LIST_HEAD(adapter_list);
  60
  61 static const unsigned int MAX_ATIDS = 64 * 1024;
  62 static const unsigned int ATID_BASE = 0x10000;
  63
  64 static void cxgb_neigh_update(struct neighbour *neigh);
  65 static void cxgb_redirect(struct dst_entry *old, struct dst_entry *new);
  66
  67 static inline int offload_activated(struct t3cdev *tdev)
  68 {
  69         const struct adapter *adapter = tdev2adap(tdev);
  70
  71         return test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
  72 }
  73
  74 /**
  75  *      cxgb3_register_client - register an offload client
  76  *      @client: the client
  77  *
  78  *      Add the client to the client list,
  79  *      and call backs the client for each activated offload device
  80  */
  81 void cxgb3_register_client(struct cxgb3_client *client)
  82 {
  83         struct t3cdev *tdev;
  84
  85         mutex_lock(&cxgb3_db_lock);
  86         list_add_tail(&client->client_list, &client_list);
  87
  88         if (client->add) {
  89                 list_for_each_entry(tdev, &ofld_dev_list, ofld_dev_list) {
  90                         if (offload_activated(tdev))
  91                                 client->add(tdev);
  92                 }
  93         }
  94         mutex_unlock(&cxgb3_db_lock);
  95 }
  96
  97 EXPORT_SYMBOL(cxgb3_register_client);
  98
  99 /**
 100  *      cxgb3_unregister_client - unregister an offload client
 101  *      @client: the client
 102  *
 103  *      Remove the client to the client list,
 104  *      and call backs the client for each activated offload device.
 105  */
 106 void cxgb3_unregister_client(struct cxgb3_client *client)
 107 {
 108         struct t3cdev *tdev;
 109
 110         mutex_lock(&cxgb3_db_lock);
 111         list_del(&client->client_list);
 112
 113         if (client->remove) {
 114                 list_for_each_entry(tdev, &ofld_dev_list, ofld_dev_list) {
 115                         if (offload_activated(tdev))
 116                                 client->remove(tdev);
 117                 }
 118         }
 119         mutex_unlock(&cxgb3_db_lock);
 120 }
 121
 122 EXPORT_SYMBOL(cxgb3_unregister_client);
 123
 124 /**
 125  *      cxgb3_add_clients - activate registered clients for an offload device
 126  *      @tdev: the offload device
 127  *
 128  *      Call backs all registered clients once a offload device is activated
 129  */
 130 void cxgb3_add_clients(struct t3cdev *tdev)
 131 {
 132         struct cxgb3_client *client;
 133
 134         mutex_lock(&cxgb3_db_lock);
 135         list_for_each_entry(client, &client_list, client_list) {
 136                 if (client->add)
 137                         client->add(tdev);
 138         }
 139         mutex_unlock(&cxgb3_db_lock);
 140 }
 141
 142 /**
 143  *      cxgb3_remove_clients - deactivates registered clients
 144  *                             for an offload device
 145  *      @tdev: the offload device
 146  *
 147  *      Call backs all registered clients once a offload device is deactivated
 148  */
 149 void cxgb3_remove_clients(struct t3cdev *tdev)
 150 {
 151         struct cxgb3_client *client;
 152
 153         mutex_lock(&cxgb3_db_lock);
 154         list_for_each_entry(client, &client_list, client_list) {
 155                 if (client->remove)
 156                         client->remove(tdev);
 157         }
 158         mutex_unlock(&cxgb3_db_lock);
 159 }
 160
 161 void cxgb3_event_notify(struct t3cdev *tdev, u32 event, u32 port)
 162 {
 163         struct cxgb3_client *client;
 164
 165         mutex_lock(&cxgb3_db_lock);
 166         list_for_each_entry(client, &client_list, client_list) {
 167                 if (client->event_handler)
 168                         client->event_handler(tdev, event, port);
 169         }
 170         mutex_unlock(&cxgb3_db_lock);
 171 }
 172
 173 static struct net_device *get_iff_from_mac(struct adapter *adapter,
 174                                            const unsigned char *mac,
 175                                            unsigned int vlan)
 176 {
 177         int i;
 178
 179         for_each_port(adapter, i) {
 180                 struct net_device *dev = adapter->port[i];
 181
 182                 if (!memcmp(dev->dev_addr, mac, ETH_ALEN)) {
 183                         if (vlan && vlan != VLAN_VID_MASK) {
 184                                 rcu_read_lock();
 185                                 dev = __vlan_find_dev_deep(dev, vlan);
 186                                 rcu_read_unlock();
 187                         } else if (netif_is_bond_slave(dev)) {
 188                                 while (dev->master)
 189                                         dev = dev->master;
 190                         }
 191                         return dev;
 192                 }
 193         }
 194         return NULL;
 195 }
 196
 197 static int cxgb_ulp_iscsi_ctl(struct adapter *adapter, unsigned int req,
 198                               void *data)
 199 {
 200         int i;
 201         int ret = 0;
 202         unsigned int val = 0;
 203         struct ulp_iscsi_info *uiip = data;
 204
 205         switch (req) {
 206         case ULP_ISCSI_GET_PARAMS:
 207                 uiip->pdev = adapter->pdev;
 208                 uiip->llimit = t3_read_reg(adapter, A_ULPRX_ISCSI_LLIMIT);
 209                 uiip->ulimit = t3_read_reg(adapter, A_ULPRX_ISCSI_ULIMIT);
 210                 uiip->tagmask = t3_read_reg(adapter, A_ULPRX_ISCSI_TAGMASK);
 211
 212                 val = t3_read_reg(adapter, A_ULPRX_ISCSI_PSZ);
 213                 for (i = 0; i < 4; i++, val >>= 8)
 214                         uiip->pgsz_factor[i] = val & 0xFF;
 215
 216                 val = t3_read_reg(adapter, A_TP_PARA_REG7);
 217                 uiip->max_txsz =
 218                 uiip->max_rxsz = min((val >> S_PMMAXXFERLEN0)&M_PMMAXXFERLEN0,
 219                                      (val >> S_PMMAXXFERLEN1)&M_PMMAXXFERLEN1);
 220                 /*
 221                  * On tx, the iscsi pdu has to be <= tx page size and has to
 222                  * fit into the Tx PM FIFO.
 223                  */
 224                 val = min(adapter->params.tp.tx_pg_size,
 225                           t3_read_reg(adapter, A_PM1_TX_CFG) >> 17);
 226                 uiip->max_txsz = min(val, uiip->max_txsz);
 227
 228                 /* set MaxRxData to 16224 */
 229                 val = t3_read_reg(adapter, A_TP_PARA_REG2);
 230                 if ((val >> S_MAXRXDATA) != 0x3f60) {
 231                         val &= (M_RXCOALESCESIZE << S_RXCOALESCESIZE);
 232                         val |= V_MAXRXDATA(0x3f60);
 233                         printk(KERN_INFO
 234                                 "%s, iscsi set MaxRxData to 16224 (0x%x).\n",
 235                                 adapter->name, val);
 236                         t3_write_reg(adapter, A_TP_PARA_REG2, val);
 237                 }
 238
 239                 /*
 240                  * on rx, the iscsi pdu has to be < rx page size and the
 241                  * the max rx data length programmed in TP
 242                  */
 243                 val = min(adapter->params.tp.rx_pg_size,
 244                           ((t3_read_reg(adapter, A_TP_PARA_REG2)) >>
 245                                 S_MAXRXDATA) & M_MAXRXDATA);
 246                 uiip->max_rxsz = min(val, uiip->max_rxsz);
 247                 break;
 248         case ULP_ISCSI_SET_PARAMS:
 249                 t3_write_reg(adapter, A_ULPRX_ISCSI_TAGMASK, uiip->tagmask);
 250                 /* program the ddp page sizes */
 251                 for (i = 0; i < 4; i++)
 252                         val |= (uiip->pgsz_factor[i] & 0xF) << (8 * i);
 253                 if (val && (val != t3_read_reg(adapter, A_ULPRX_ISCSI_PSZ))) {
 254                         printk(KERN_INFO
 255                                 "%s, setting iscsi pgsz 0x%x, %u,%u,%u,%u.\n",
 256                                 adapter->name, val, uiip->pgsz_factor[0],
 257                                 uiip->pgsz_factor[1], uiip->pgsz_factor[2],
 258                                 uiip->pgsz_factor[3]);
 259                         t3_write_reg(adapter, A_ULPRX_ISCSI_PSZ, val);
 260                 }
 261                 break;
 262         default:
 263                 ret = -EOPNOTSUPP;
 264         }
 265         return ret;
 266 }
 267
 268 /* Response queue used for RDMA events. */
 269 #define ASYNC_NOTIF_RSPQ 0
 270
 271 static int cxgb_rdma_ctl(struct adapter *adapter, unsigned int req, void *data)
 272 {
 273         int ret = 0;
 274
 275         switch (req) {
 276         case RDMA_GET_PARAMS: {
 277                 struct rdma_info *rdma = data;
 278                 struct pci_dev *pdev = adapter->pdev;
 279
 280                 rdma->udbell_physbase = pci_resource_start(pdev, 2);
 281                 rdma->udbell_len = pci_resource_len(pdev, 2);
 282                 rdma->tpt_base =
 283                         t3_read_reg(adapter, A_ULPTX_TPT_LLIMIT);
 284                 rdma->tpt_top = t3_read_reg(adapter, A_ULPTX_TPT_ULIMIT);
 285                 rdma->pbl_base =
 286                         t3_read_reg(adapter, A_ULPTX_PBL_LLIMIT);
 287                 rdma->pbl_top = t3_read_reg(adapter, A_ULPTX_PBL_ULIMIT);
 288                 rdma->rqt_base = t3_read_reg(adapter, A_ULPRX_RQ_LLIMIT);
 289                 rdma->rqt_top = t3_read_reg(adapter, A_ULPRX_RQ_ULIMIT);
 290                 rdma->kdb_addr = adapter->regs + A_SG_KDOORBELL;
 291                 rdma->pdev = pdev;
 292                 break;
 293         }
 294         case RDMA_CQ_OP:{
 295                 unsigned long flags;
 296                 struct rdma_cq_op *rdma = data;
 297
 298                 /* may be called in any context */
 299                 spin_lock_irqsave(&adapter->sge.reg_lock, flags);
 300                 ret = t3_sge_cqcntxt_op(adapter, rdma->id, rdma->op,
 301                                         rdma->credits);
 302                 spin_unlock_irqrestore(&adapter->sge.reg_lock, flags);
 303                 break;
 304         }
 305         case RDMA_GET_MEM:{
 306                 struct ch_mem_range *t = data;
 307                 struct mc7 *mem;
 308
 309                 if ((t->addr & 7) || (t->len & 7))
 310                         return -EINVAL;
 311                 if (t->mem_id == MEM_CM)
 312                         mem = &adapter->cm;
 313                 else if (t->mem_id == MEM_PMRX)
 314                         mem = &adapter->pmrx;
 315                 else if (t->mem_id == MEM_PMTX)
 316                         mem = &adapter->pmtx;
 317                 else
 318                         return -EINVAL;
 319
 320                 ret =
 321                         t3_mc7_bd_read(mem, t->addr / 8, t->len / 8,
 322                                         (u64 *) t->buf);
 323                 if (ret)
 324                         return ret;
 325                 break;
 326         }
 327         case RDMA_CQ_SETUP:{
 328                 struct rdma_cq_setup *rdma = data;
 329
 330                 spin_lock_irq(&adapter->sge.reg_lock);
 331                 ret =
 332                         t3_sge_init_cqcntxt(adapter, rdma->id,
 333                                         rdma->base_addr, rdma->size,
 334                                         ASYNC_NOTIF_RSPQ,
 335                                         rdma->ovfl_mode, rdma->credits,
 336                                         rdma->credit_thres);
 337                 spin_unlock_irq(&adapter->sge.reg_lock);
 338                 break;
 339         }
 340         case RDMA_CQ_DISABLE:
 341                 spin_lock_irq(&adapter->sge.reg_lock);
 342                 ret = t3_sge_disable_cqcntxt(adapter, *(unsigned int *)data);
 343                 spin_unlock_irq(&adapter->sge.reg_lock);
 344                 break;
 345         case RDMA_CTRL_QP_SETUP:{
 346                 struct rdma_ctrlqp_setup *rdma = data;
 347
 348                 spin_lock_irq(&adapter->sge.reg_lock);
 349                 ret = t3_sge_init_ecntxt(adapter, FW_RI_SGEEC_START, 0,
 350                                                 SGE_CNTXT_RDMA,
 351                                                 ASYNC_NOTIF_RSPQ,
 352                                                 rdma->base_addr, rdma->size,
 353                                                 FW_RI_TID_START, 1, 0);
 354                 spin_unlock_irq(&adapter->sge.reg_lock);
 355                 break;
 356         }
 357         case RDMA_GET_MIB: {
 358                 spin_lock(&adapter->stats_lock);
 359                 t3_tp_get_mib_stats(adapter, (struct tp_mib_stats *)data);
 360                 spin_unlock(&adapter->stats_lock);
 361                 break;
 362         }
 363         default:
 364                 ret = -EOPNOTSUPP;
 365         }
 366         return ret;
 367 }
 368
 369 static int cxgb_offload_ctl(struct t3cdev *tdev, unsigned int req, void *data)
 370 {
 371         struct adapter *adapter = tdev2adap(tdev);
 372         struct tid_range *tid;
 373         struct mtutab *mtup;
 374         struct iff_mac *iffmacp;
 375         struct ddp_params *ddpp;
 376         struct adap_ports *ports;
 377         struct ofld_page_info *rx_page_info;
 378         struct tp_params *tp = &adapter->params.tp;
 379         int i;
 380
 381         switch (req) {
 382         case GET_MAX_OUTSTANDING_WR:
 383                 *(unsigned int *)data = FW_WR_NUM;
 384                 break;
 385         case GET_WR_LEN:
 386                 *(unsigned int *)data = WR_FLITS;
 387                 break;
 388         case GET_TX_MAX_CHUNK:
 389                 *(unsigned int *)data = 1 << 20;        /* 1MB */
 390                 break;
 391         case GET_TID_RANGE:
 392                 tid = data;
 393                 tid->num = t3_mc5_size(&adapter->mc5) -
 394                     adapter->params.mc5.nroutes -
 395                     adapter->params.mc5.nfilters - adapter->params.mc5.nservers;
 396                 tid->base = 0;
 397                 break;
 398         case GET_STID_RANGE:
 399                 tid = data;
 400                 tid->num = adapter->params.mc5.nservers;
 401                 tid->base = t3_mc5_size(&adapter->mc5) - tid->num -
 402                     adapter->params.mc5.nfilters - adapter->params.mc5.nroutes;
 403                 break;
 404         case GET_L2T_CAPACITY:
 405                 *(unsigned int *)data = 2048;
 406                 break;
 407         case GET_MTUS:
 408                 mtup = data;
 409                 mtup->size = NMTUS;
 410                 mtup->mtus = adapter->params.mtus;
 411                 break;
 412         case GET_IFF_FROM_MAC:
 413                 iffmacp = data;
 414                 iffmacp->dev = get_iff_from_mac(adapter, iffmacp->mac_addr,
 415                                                 iffmacp->vlan_tag &
 416                                                 VLAN_VID_MASK);
 417                 break;
 418         case GET_DDP_PARAMS:
 419                 ddpp = data;
 420                 ddpp->llimit = t3_read_reg(adapter, A_ULPRX_TDDP_LLIMIT);
 421                 ddpp->ulimit = t3_read_reg(adapter, A_ULPRX_TDDP_ULIMIT);
 422                 ddpp->tag_mask = t3_read_reg(adapter, A_ULPRX_TDDP_TAGMASK);
 423                 break;
 424         case GET_PORTS:
 425                 ports = data;
 426                 ports->nports = adapter->params.nports;
 427                 for_each_port(adapter, i)
 428                         ports->lldevs[i] = adapter->port[i];
 429                 break;
 430         case ULP_ISCSI_GET_PARAMS:
 431         case ULP_ISCSI_SET_PARAMS:
 432                 if (!offload_running(adapter))
 433                         return -EAGAIN;
 434                 return cxgb_ulp_iscsi_ctl(adapter, req, data);
 435         case RDMA_GET_PARAMS:
 436         case RDMA_CQ_OP:
 437         case RDMA_CQ_SETUP:
 438         case RDMA_CQ_DISABLE:
 439         case RDMA_CTRL_QP_SETUP:
 440         case RDMA_GET_MEM:
 441         case RDMA_GET_MIB:
 442                 if (!offload_running(adapter))
 443                         return -EAGAIN;
 444                 return cxgb_rdma_ctl(adapter, req, data);
 445         case GET_RX_PAGE_INFO:
 446                 rx_page_info = data;
 447                 rx_page_info->page_size = tp->rx_pg_size;
 448                 rx_page_info->num = tp->rx_num_pgs;
 449                 break;
 450         case GET_ISCSI_IPV4ADDR: {
 451                 struct iscsi_ipv4addr *p = data;
 452                 struct port_info *pi = netdev_priv(p->dev);
 453                 p->ipv4addr = pi->iscsi_ipv4addr;
 454                 break;
 455         }
 456         case GET_EMBEDDED_INFO: {
 457                 struct ch_embedded_info *e = data;
 458
 459                 spin_lock(&adapter->stats_lock);
 460                 t3_get_fw_version(adapter, &e->fw_vers);
 461                 t3_get_tp_version(adapter, &e->tp_vers);
 462                 spin_unlock(&adapter->stats_lock);
 463                 break;
 464         }
 465         default:
 466                 return -EOPNOTSUPP;
 467         }
 468         return 0;
 469 }
 470
 471 /*
 472  * Dummy handler for Rx offload packets in case we get an offload packet before
 473  * proper processing is setup.  This complains and drops the packet as it isn't
 474  * normal to get offload packets at this stage.
 475  */
 476 static int rx_offload_blackhole(struct t3cdev *dev, struct sk_buff **skbs,
 477                                 int n)
 478 {
 479         while (n--)
 480                 dev_kfree_skb_any(skbs[n]);
 481         return 0;
 482 }
 483
 484 static void dummy_neigh_update(struct t3cdev *dev, struct neighbour *neigh)
 485 {
 486 }
 487
 488 void cxgb3_set_dummy_ops(struct t3cdev *dev)
 489 {
 490         dev->recv = rx_offload_blackhole;
 491         dev->neigh_update = dummy_neigh_update;
 492 }
 493
 494 /*
 495  * Free an active-open TID.
 496  */
 497 void *cxgb3_free_atid(struct t3cdev *tdev, int atid)
 498 {
 499         struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
 500         union active_open_entry *p = atid2entry(t, atid);
 501         void *ctx = p->t3c_tid.ctx;
 502
 503         spin_lock_bh(&t->atid_lock);
 504         p->next = t->afree;
 505         t->afree = p;
 506         t->atids_in_use--;
 507         spin_unlock_bh(&t->atid_lock);
 508
 509         return ctx;
 510 }
 511
 512 EXPORT_SYMBOL(cxgb3_free_atid);
 513
 514 /*
 515  * Free a server TID and return it to the free pool.
 516  */
 517 void cxgb3_free_stid(struct t3cdev *tdev, int stid)
 518 {
 519         struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
 520         union listen_entry *p = stid2entry(t, stid);
 521
 522         spin_lock_bh(&t->stid_lock);
 523         p->next = t->sfree;
 524         t->sfree = p;
 525         t->stids_in_use--;
 526         spin_unlock_bh(&t->stid_lock);
 527 }
 528
 529 EXPORT_SYMBOL(cxgb3_free_stid);
 530
 531 void cxgb3_insert_tid(struct t3cdev *tdev, struct cxgb3_client *client,
 532                       void *ctx, unsigned int tid)
 533 {
 534         struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
 535
 536         t->tid_tab[tid].client = client;
 537         t->tid_tab[tid].ctx = ctx;
 538         atomic_inc(&t->tids_in_use);
 539 }
 540
 541 EXPORT_SYMBOL(cxgb3_insert_tid);
 542
 543 /*
 544  * Populate a TID_RELEASE WR.  The skb must be already propely sized.
 545  */
 546 static inline void mk_tid_release(struct sk_buff *skb, unsigned int tid)
 547 {
 548         struct cpl_tid_release *req;
 549
 550         skb->priority = CPL_PRIORITY_SETUP;
 551         req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
 552         req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 553         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
 554 }
 555
 556 static void t3_process_tid_release_list(struct work_struct *work)
 557 {
 558         struct t3c_data *td = container_of(work, struct t3c_data,
 559                                            tid_release_task);
 560         struct sk_buff *skb;
 561         struct t3cdev *tdev = td->dev;
 562
 563
 564         spin_lock_bh(&td->tid_release_lock);
 565         while (td->tid_release_list) {
 566                 struct t3c_tid_entry *p = td->tid_release_list;
 567
 568                 td->tid_release_list = p->ctx;
 569                 spin_unlock_bh(&td->tid_release_lock);
 570
 571                 skb = alloc_skb(sizeof(struct cpl_tid_release),
 572                                 GFP_KERNEL);
 573                 if (!skb)
 574                         skb = td->nofail_skb;
 575                 if (!skb) {
 576                         spin_lock_bh(&td->tid_release_lock);
 577                         p->ctx = (void *)td->tid_release_list;
 578                         td->tid_release_list = (struct t3c_tid_entry *)p;
 579                         break;
 580                 }
 581                 mk_tid_release(skb, p - td->tid_maps.tid_tab);
 582                 cxgb3_ofld_send(tdev, skb);
 583                 p->ctx = NULL;
 584                 if (skb == td->nofail_skb)
 585                         td->nofail_skb =
 586                                 alloc_skb(sizeof(struct cpl_tid_release),
 587                                         GFP_KERNEL);
 588                 spin_lock_bh(&td->tid_release_lock);
 589         }
 590         td->release_list_incomplete = (td->tid_release_list == NULL) ? 0 : 1;
 591         spin_unlock_bh(&td->tid_release_lock);
 592
 593         if (!td->nofail_skb)
 594                 td->nofail_skb =
 595                         alloc_skb(sizeof(struct cpl_tid_release),
 596                                 GFP_KERNEL);
 597 }
 598
 599 /* use ctx as a next pointer in the tid release list */
 600 void cxgb3_queue_tid_release(struct t3cdev *tdev, unsigned int tid)
 601 {
 602         struct t3c_data *td = T3C_DATA(tdev);
 603         struct t3c_tid_entry *p = &td->tid_maps.tid_tab[tid];
 604
 605         spin_lock_bh(&td->tid_release_lock);
 606         p->ctx = (void *)td->tid_release_list;
 607         p->client = NULL;
 608         td->tid_release_list = p;
 609         if (!p->ctx || td->release_list_incomplete)
 610                 schedule_work(&td->tid_release_task);
 611         spin_unlock_bh(&td->tid_release_lock);
 612 }
 613
 614 EXPORT_SYMBOL(cxgb3_queue_tid_release);
 615
 616 /*
 617  * Remove a tid from the TID table.  A client may defer processing its last
 618  * CPL message if it is locked at the time it arrives, and while the message
 619  * sits in the client's backlog the TID may be reused for another connection.
 620  * To handle this we atomically switch the TID association if it still points
 621  * to the original client context.
 622  */
 623 void cxgb3_remove_tid(struct t3cdev *tdev, void *ctx, unsigned int tid)
 624 {
 625         struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
 626
 627         BUG_ON(tid >= t->ntids);
 628         if (tdev->type == T3A)
 629                 (void)cmpxchg(&t->tid_tab[tid].ctx, ctx, NULL);
 630         else {
 631                 struct sk_buff *skb;
 632
 633                 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
 634                 if (likely(skb)) {
 635                         mk_tid_release(skb, tid);
 636                         cxgb3_ofld_send(tdev, skb);
 637                         t->tid_tab[tid].ctx = NULL;
 638                 } else
 639                         cxgb3_queue_tid_release(tdev, tid);
 640         }
 641         atomic_dec(&t->tids_in_use);
 642 }
 643
 644 EXPORT_SYMBOL(cxgb3_remove_tid);
 645
 646 int cxgb3_alloc_atid(struct t3cdev *tdev, struct cxgb3_client *client,
 647                      void *ctx)
 648 {
 649         int atid = -1;
 650         struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
 651
 652         spin_lock_bh(&t->atid_lock);
 653         if (t->afree &&
 654             t->atids_in_use + atomic_read(&t->tids_in_use) + MC5_MIN_TIDS <=
 655             t->ntids) {
 656                 union active_open_entry *p = t->afree;
 657
 658                 atid = (p - t->atid_tab) + t->atid_base;
 659                 t->afree = p->next;
 660                 p->t3c_tid.ctx = ctx;
 661                 p->t3c_tid.client = client;
 662                 t->atids_in_use++;
 663         }
 664         spin_unlock_bh(&t->atid_lock);
 665         return atid;
 666 }
 667
 668 EXPORT_SYMBOL(cxgb3_alloc_atid);
 669
 670 int cxgb3_alloc_stid(struct t3cdev *tdev, struct cxgb3_client *client,
 671                      void *ctx)
 672 {
 673         int stid = -1;
 674         struct tid_info *t = &(T3C_DATA(tdev))->tid_maps;
 675
 676         spin_lock_bh(&t->stid_lock);
 677         if (t->sfree) {
 678                 union listen_entry *p = t->sfree;
 679
 680                 stid = (p - t->stid_tab) + t->stid_base;
 681                 t->sfree = p->next;
 682                 p->t3c_tid.ctx = ctx;
 683                 p->t3c_tid.client = client;
 684                 t->stids_in_use++;
 685         }
 686         spin_unlock_bh(&t->stid_lock);
 687         return stid;
 688 }
 689
 690 EXPORT_SYMBOL(cxgb3_alloc_stid);
 691
 692 /* Get the t3cdev associated with a net_device */
 693 struct t3cdev *dev2t3cdev(struct net_device *dev)
 694 {
 695         const struct port_info *pi = netdev_priv(dev);
 696
 697         return (struct t3cdev *)pi->adapter;
 698 }
 699
 700 EXPORT_SYMBOL(dev2t3cdev);
 701
 702 static int do_smt_write_rpl(struct t3cdev *dev, struct sk_buff *skb)
 703 {
 704         struct cpl_smt_write_rpl *rpl = cplhdr(skb);
 705
 706         if (rpl->status != CPL_ERR_NONE)
 707                 printk(KERN_ERR
 708                        "Unexpected SMT_WRITE_RPL status %u for entry %u\n",
 709                        rpl->status, GET_TID(rpl));
 710
 711         return CPL_RET_BUF_DONE;
 712 }
 713
 714 static int do_l2t_write_rpl(struct t3cdev *dev, struct sk_buff *skb)
 715 {
 716         struct cpl_l2t_write_rpl *rpl = cplhdr(skb);
 717
 718         if (rpl->status != CPL_ERR_NONE)
 719                 printk(KERN_ERR
 720                        "Unexpected L2T_WRITE_RPL status %u for entry %u\n",
 721                        rpl->status, GET_TID(rpl));
 722
 723         return CPL_RET_BUF_DONE;
 724 }
 725
 726 static int do_rte_write_rpl(struct t3cdev *dev, struct sk_buff *skb)
 727 {
 728         struct cpl_rte_write_rpl *rpl = cplhdr(skb);
 729
 730         if (rpl->status != CPL_ERR_NONE)
 731                 printk(KERN_ERR
 732                        "Unexpected RTE_WRITE_RPL status %u for entry %u\n",
 733                        rpl->status, GET_TID(rpl));
 734
 735         return CPL_RET_BUF_DONE;
 736 }
 737
 738 static int do_act_open_rpl(struct t3cdev *dev, struct sk_buff *skb)
 739 {
 740         struct cpl_act_open_rpl *rpl = cplhdr(skb);
 741         unsigned int atid = G_TID(ntohl(rpl->atid));
 742         struct t3c_tid_entry *t3c_tid;
 743
 744         t3c_tid = lookup_atid(&(T3C_DATA(dev))->tid_maps, atid);
 745         if (t3c_tid && t3c_tid->ctx && t3c_tid->client &&
 746             t3c_tid->client->handlers &&
 747             t3c_tid->client->handlers[CPL_ACT_OPEN_RPL]) {
 748                 return t3c_tid->client->handlers[CPL_ACT_OPEN_RPL] (dev, skb,
 749                                                                     t3c_tid->
 750                                                                     ctx);
 751         } else {
 752                 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
 753                        dev->name, CPL_ACT_OPEN_RPL);
 754                 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
 755         }
 756 }
 757
 758 static int do_stid_rpl(struct t3cdev *dev, struct sk_buff *skb)
 759 {
 760         union opcode_tid *p = cplhdr(skb);
 761         unsigned int stid = G_TID(ntohl(p->opcode_tid));
 762         struct t3c_tid_entry *t3c_tid;
 763
 764         t3c_tid = lookup_stid(&(T3C_DATA(dev))->tid_maps, stid);
 765         if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
 766             t3c_tid->client->handlers[p->opcode]) {
 767                 return t3c_tid->client->handlers[p->opcode] (dev, skb,
 768                                                              t3c_tid->ctx);
 769         } else {
 770                 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
 771                        dev->name, p->opcode);
 772                 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
 773         }
 774 }
 775
 776 static int do_hwtid_rpl(struct t3cdev *dev, struct sk_buff *skb)
 777 {
 778         union opcode_tid *p = cplhdr(skb);
 779         unsigned int hwtid = G_TID(ntohl(p->opcode_tid));
 780         struct t3c_tid_entry *t3c_tid;
 781
 782         t3c_tid = lookup_tid(&(T3C_DATA(dev))->tid_maps, hwtid);
 783         if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
 784             t3c_tid->client->handlers[p->opcode]) {
 785                 return t3c_tid->client->handlers[p->opcode]
 786                     (dev, skb, t3c_tid->ctx);
 787         } else {
 788                 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
 789                        dev->name, p->opcode);
 790                 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
 791         }
 792 }
 793
 794 static int do_cr(struct t3cdev *dev, struct sk_buff *skb)
 795 {
 796         struct cpl_pass_accept_req *req = cplhdr(skb);
 797         unsigned int stid = G_PASS_OPEN_TID(ntohl(req->tos_tid));
 798         struct tid_info *t = &(T3C_DATA(dev))->tid_maps;
 799         struct t3c_tid_entry *t3c_tid;
 800         unsigned int tid = GET_TID(req);
 801
 802         if (unlikely(tid >= t->ntids)) {
 803                 printk("%s: passive open TID %u too large\n",
 804                        dev->name, tid);
 805                 t3_fatal_err(tdev2adap(dev));
 806                 return CPL_RET_BUF_DONE;
 807         }
 808
 809         t3c_tid = lookup_stid(t, stid);
 810         if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
 811             t3c_tid->client->handlers[CPL_PASS_ACCEPT_REQ]) {
 812                 return t3c_tid->client->handlers[CPL_PASS_ACCEPT_REQ]
 813                     (dev, skb, t3c_tid->ctx);
 814         } else {
 815                 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
 816                        dev->name, CPL_PASS_ACCEPT_REQ);
 817                 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
 818         }
 819 }
 820
 821 /*
 822  * Returns an sk_buff for a reply CPL message of size len.  If the input
 823  * sk_buff has no other users it is trimmed and reused, otherwise a new buffer
 824  * is allocated.  The input skb must be of size at least len.  Note that this
 825  * operation does not destroy the original skb data even if it decides to reuse
 826  * the buffer.
 827  */
 828 static struct sk_buff *cxgb3_get_cpl_reply_skb(struct sk_buff *skb, size_t len,
 829                                                gfp_t gfp)
 830 {
 831         if (likely(!skb_cloned(skb))) {
 832                 BUG_ON(skb->len < len);
 833                 __skb_trim(skb, len);
 834                 skb_get(skb);
 835         } else {
 836                 skb = alloc_skb(len, gfp);
 837                 if (skb)
 838                         __skb_put(skb, len);
 839         }
 840         return skb;
 841 }
 842
 843 static int do_abort_req_rss(struct t3cdev *dev, struct sk_buff *skb)
 844 {
 845         union opcode_tid *p = cplhdr(skb);
 846         unsigned int hwtid = G_TID(ntohl(p->opcode_tid));
 847         struct t3c_tid_entry *t3c_tid;
 848
 849         t3c_tid = lookup_tid(&(T3C_DATA(dev))->tid_maps, hwtid);
 850         if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
 851             t3c_tid->client->handlers[p->opcode]) {
 852                 return t3c_tid->client->handlers[p->opcode]
 853                     (dev, skb, t3c_tid->ctx);
 854         } else {
 855                 struct cpl_abort_req_rss *req = cplhdr(skb);
 856                 struct cpl_abort_rpl *rpl;
 857                 struct sk_buff *reply_skb;
 858                 unsigned int tid = GET_TID(req);
 859                 u8 cmd = req->status;
 860
 861                 if (req->status == CPL_ERR_RTX_NEG_ADVICE ||
 862                     req->status == CPL_ERR_PERSIST_NEG_ADVICE)
 863                         goto out;
 864
 865                 reply_skb = cxgb3_get_cpl_reply_skb(skb,
 866                                                     sizeof(struct
 867                                                            cpl_abort_rpl),
 868                                                     GFP_ATOMIC);
 869
 870                 if (!reply_skb) {
 871                         printk("do_abort_req_rss: couldn't get skb!\n");
 872                         goto out;
 873                 }
 874                 reply_skb->priority = CPL_PRIORITY_DATA;
 875                 __skb_put(reply_skb, sizeof(struct cpl_abort_rpl));
 876                 rpl = cplhdr(reply_skb);
 877                 rpl->wr.wr_hi =
 878                     htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_RPL));
 879                 rpl->wr.wr_lo = htonl(V_WR_TID(tid));
 880                 OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, tid));
 881                 rpl->cmd = cmd;
 882                 cxgb3_ofld_send(dev, reply_skb);
 883 out:
 884                 return CPL_RET_BUF_DONE;
 885         }
 886 }
 887
 888 static int do_act_establish(struct t3cdev *dev, struct sk_buff *skb)
 889 {
 890         struct cpl_act_establish *req = cplhdr(skb);
 891         unsigned int atid = G_PASS_OPEN_TID(ntohl(req->tos_tid));
 892         struct tid_info *t = &(T3C_DATA(dev))->tid_maps;
 893         struct t3c_tid_entry *t3c_tid;
 894         unsigned int tid = GET_TID(req);
 895
 896         if (unlikely(tid >= t->ntids)) {
 897                 printk("%s: active establish TID %u too large\n",
 898                        dev->name, tid);
 899                 t3_fatal_err(tdev2adap(dev));
 900                 return CPL_RET_BUF_DONE;
 901         }
 902
 903         t3c_tid = lookup_atid(t, atid);
 904         if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
 905             t3c_tid->client->handlers[CPL_ACT_ESTABLISH]) {
 906                 return t3c_tid->client->handlers[CPL_ACT_ESTABLISH]
 907                     (dev, skb, t3c_tid->ctx);
 908         } else {
 909                 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
 910                        dev->name, CPL_ACT_ESTABLISH);
 911                 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
 912         }
 913 }
 914
 915 static int do_trace(struct t3cdev *dev, struct sk_buff *skb)
 916 {
 917         struct cpl_trace_pkt *p = cplhdr(skb);
 918
 919         skb->protocol = htons(0xffff);
 920         skb->dev = dev->lldev;
 921         skb_pull(skb, sizeof(*p));
 922         skb_reset_mac_header(skb);
 923         netif_receive_skb(skb);
 924         return 0;
 925 }
 926
 927 /*
 928  * That skb would better have come from process_responses() where we abuse
 929  * ->priority and ->csum to carry our data.  NB: if we get to per-arch
 930  * ->csum, the things might get really interesting here.
 931  */
 932
 933 static inline u32 get_hwtid(struct sk_buff *skb)
 934 {
 935         return ntohl((__force __be32)skb->priority) >> 8 & 0xfffff;
 936 }
 937
 938 static inline u32 get_opcode(struct sk_buff *skb)
 939 {
 940         return G_OPCODE(ntohl((__force __be32)skb->csum));
 941 }
 942
 943 static int do_term(struct t3cdev *dev, struct sk_buff *skb)
 944 {
 945         unsigned int hwtid = get_hwtid(skb);
 946         unsigned int opcode = get_opcode(skb);
 947         struct t3c_tid_entry *t3c_tid;
 948
 949         t3c_tid = lookup_tid(&(T3C_DATA(dev))->tid_maps, hwtid);
 950         if (t3c_tid && t3c_tid->ctx && t3c_tid->client->handlers &&
 951             t3c_tid->client->handlers[opcode]) {
 952                 return t3c_tid->client->handlers[opcode] (dev, skb,
 953                                                           t3c_tid->ctx);
 954         } else {
 955                 printk(KERN_ERR "%s: received clientless CPL command 0x%x\n",
 956                        dev->name, opcode);
 957                 return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
 958         }
 959 }
 960
 961 static int nb_callback(struct notifier_block *self, unsigned long event,
 962                        void *ctx)
 963 {
 964         switch (event) {
 965         case (NETEVENT_NEIGH_UPDATE):{
 966                 cxgb_neigh_update((struct neighbour *)ctx);
 967                 break;
 968         }
 969         case (NETEVENT_REDIRECT):{
 970                 struct netevent_redirect *nr = ctx;
 971                 cxgb_redirect(nr->old, nr->new);
 972                 cxgb_neigh_update(dst_get_neighbour_noref(nr->new));
 973                 break;
 974         }
 975         default:
 976                 break;
 977         }
 978         return 0;
 979 }
 980
 981 static struct notifier_block nb = {
 982         .notifier_call = nb_callback
 983 };
 984
 985 /*
 986  * Process a received packet with an unknown/unexpected CPL opcode.
 987  */
 988 static int do_bad_cpl(struct t3cdev *dev, struct sk_buff *skb)
 989 {
 990         printk(KERN_ERR "%s: received bad CPL command 0x%x\n", dev->name,
 991                *skb->data);
 992         return CPL_RET_BUF_DONE | CPL_RET_BAD_MSG;
 993 }
 994
 995 /*
 996  * Handlers for each CPL opcode
 997  */
 998 static cpl_handler_func cpl_handlers[NUM_CPL_CMDS];
 999
1000 /*
1001  * Add a new handler to the CPL dispatch table.  A NULL handler may be supplied
1002  * to unregister an existing handler.
1003  */
1004 void t3_register_cpl_handler(unsigned int opcode, cpl_handler_func h)
1005 {
1006         if (opcode < NUM_CPL_CMDS)
1007                 cpl_handlers[opcode] = h ? h : do_bad_cpl;
1008         else
1009                 printk(KERN_ERR "T3C: handler registration for "
1010                        "opcode %x failed\n", opcode);
1011 }
1012
1013 EXPORT_SYMBOL(t3_register_cpl_handler);
1014
1015 /*
1016  * T3CDEV's receive method.
1017  */
1018 static int process_rx(struct t3cdev *dev, struct sk_buff **skbs, int n)
1019 {
1020         while (n--) {
1021                 struct sk_buff *skb = *skbs++;
1022                 unsigned int opcode = get_opcode(skb);
1023                 int ret = cpl_handlers[opcode] (dev, skb);
1024
1025 #if VALIDATE_TID
1026                 if (ret & CPL_RET_UNKNOWN_TID) {
1027                         union opcode_tid *p = cplhdr(skb);
1028
1029                         printk(KERN_ERR "%s: CPL message (opcode %u) had "
1030                                "unknown TID %u\n", dev->name, opcode,
1031                                G_TID(ntohl(p->opcode_tid)));
1032                 }
1033 #endif
1034                 if (ret & CPL_RET_BUF_DONE)
1035                         kfree_skb(skb);
1036         }
1037         return 0;
1038 }
1039
1040 /*
1041  * Sends an sk_buff to a T3C driver after dealing with any active network taps.
1042  */
1043 int cxgb3_ofld_send(struct t3cdev *dev, struct sk_buff *skb)
1044 {
1045         int r;
1046
1047         local_bh_disable();
1048         r = dev->send(dev, skb);
1049         local_bh_enable();
1050         return r;
1051 }
1052
1053 EXPORT_SYMBOL(cxgb3_ofld_send);
1054
1055 static int is_offloading(struct net_device *dev)
1056 {
1057         struct adapter *adapter;
1058         int i;
1059
1060         read_lock_bh(&adapter_list_lock);
1061         list_for_each_entry(adapter, &adapter_list, adapter_list) {
1062                 for_each_port(adapter, i) {
1063                         if (dev == adapter->port[i]) {
1064                                 read_unlock_bh(&adapter_list_lock);
1065                                 return 1;
1066                         }
1067                 }
1068         }
1069         read_unlock_bh(&adapter_list_lock);
1070         return 0;
1071 }
1072
1073 static void cxgb_neigh_update(struct neighbour *neigh)
1074 {
1075         struct net_device *dev;
1076
1077         if (!neigh)
1078                 return;
1079         dev = neigh->dev;
1080         if (dev && (is_offloading(dev))) {
1081                 struct t3cdev *tdev = dev2t3cdev(dev);
1082
1083                 BUG_ON(!tdev);
1084                 t3_l2t_update(tdev, neigh);
1085         }
1086 }
1087
1088 static void set_l2t_ix(struct t3cdev *tdev, u32 tid, struct l2t_entry *e)
1089 {
1090         struct sk_buff *skb;
1091         struct cpl_set_tcb_field *req;
1092
1093         skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
1094         if (!skb) {
1095                 printk(KERN_ERR "%s: cannot allocate skb!\n", __func__);
1096                 return;
1097         }
1098         skb->priority = CPL_PRIORITY_CONTROL;
1099         req = (struct cpl_set_tcb_field *)skb_put(skb, sizeof(*req));
1100         req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1101         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
1102         req->reply = 0;
1103         req->cpu_idx = 0;
1104         req->word = htons(W_TCB_L2T_IX);
1105         req->mask = cpu_to_be64(V_TCB_L2T_IX(M_TCB_L2T_IX));
1106         req->val = cpu_to_be64(V_TCB_L2T_IX(e->idx));
1107         tdev->send(tdev, skb);
1108 }
1109
1110 static void cxgb_redirect(struct dst_entry *old, struct dst_entry *new)
1111 {
1112         struct net_device *olddev, *newdev;
1113         struct neighbour *n;
1114         struct tid_info *ti;
1115         struct t3cdev *tdev;
1116         u32 tid;
1117         int update_tcb;
1118         struct l2t_entry *e;
1119         struct t3c_tid_entry *te;
1120
1121         n = dst_get_neighbour_noref(old);
1122         if (!n)
1123                 return;
1124         olddev = n->dev;
1125
1126         n = dst_get_neighbour_noref(new);
1127         if (!n)
1128                 return;
1129         newdev = n->dev;
1130
1131         if (!is_offloading(olddev))
1132                 return;
1133         if (!is_offloading(newdev)) {
1134                 printk(KERN_WARNING "%s: Redirect to non-offload "
1135                        "device ignored.\n", __func__);
1136                 return;
1137         }
1138         tdev = dev2t3cdev(olddev);
1139         BUG_ON(!tdev);
1140         if (tdev != dev2t3cdev(newdev)) {
1141                 printk(KERN_WARNING "%s: Redirect to different "
1142                        "offload device ignored.\n", __func__);
1143                 return;
1144         }
1145
1146         /* Add new L2T entry */
1147         e = t3_l2t_get(tdev, new, newdev);
1148         if (!e) {
1149                 printk(KERN_ERR "%s: couldn't allocate new l2t entry!\n",
1150                        __func__);
1151                 return;
1152         }
1153
1154         /* Walk tid table and notify clients of dst change. */
1155         ti = &(T3C_DATA(tdev))->tid_maps;
1156         for (tid = 0; tid < ti->ntids; tid++) {
1157                 te = lookup_tid(ti, tid);
1158                 BUG_ON(!te);
1159                 if (te && te->ctx && te->client && te->client->redirect) {
1160                         update_tcb = te->client->redirect(te->ctx, old, new, e);
1161                         if (update_tcb) {
1162                                 rcu_read_lock();
1163                                 l2t_hold(L2DATA(tdev), e);
1164                                 rcu_read_unlock();
1165                                 set_l2t_ix(tdev, tid, e);
1166                         }
1167                 }
1168         }
1169         l2t_release(tdev, e);
1170 }
1171
1172 /*
1173  * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
1174  * The allocated memory is cleared.
1175  */
1176 void *cxgb_alloc_mem(unsigned long size)
1177 {
1178         void *p = kzalloc(size, GFP_KERNEL);
1179
1180         if (!p)
1181                 p = vzalloc(size);
1182         return p;
1183 }
1184
1185 /*
1186  * Free memory allocated through t3_alloc_mem().
1187  */
1188 void cxgb_free_mem(void *addr)
1189 {
1190         if (is_vmalloc_addr(addr))
1191                 vfree(addr);
1192         else
1193                 kfree(addr);
1194 }
1195
1196 /*
1197  * Allocate and initialize the TID tables.  Returns 0 on success.
1198  */
1199 static int init_tid_tabs(struct tid_info *t, unsigned int ntids,
1200                          unsigned int natids, unsigned int nstids,
1201                          unsigned int atid_base, unsigned int stid_base)
1202 {
1203         unsigned long size = ntids * sizeof(*t->tid_tab) +
1204             natids * sizeof(*t->atid_tab) + nstids * sizeof(*t->stid_tab);
1205
1206         t->tid_tab = cxgb_alloc_mem(size);
1207         if (!t->tid_tab)
1208                 return -ENOMEM;
1209
1210         t->stid_tab = (union listen_entry *)&t->tid_tab[ntids];
1211         t->atid_tab = (union active_open_entry *)&t->stid_tab[nstids];
1212         t->ntids = ntids;
1213         t->nstids = nstids;
1214         t->stid_base = stid_base;
1215         t->sfree = NULL;
1216         t->natids = natids;
1217         t->atid_base = atid_base;
1218         t->afree = NULL;
1219         t->stids_in_use = t->atids_in_use = 0;
1220         atomic_set(&t->tids_in_use, 0);
1221         spin_lock_init(&t->stid_lock);
1222         spin_lock_init(&t->atid_lock);
1223
1224         /*
1225          * Setup the free lists for stid_tab and atid_tab.
1226          */
1227         if (nstids) {
1228                 while (--nstids)
1229                         t->stid_tab[nstids - 1].next = &t->stid_tab[nstids];
1230                 t->sfree = t->stid_tab;
1231         }
1232         if (natids) {
1233                 while (--natids)
1234                         t->atid_tab[natids - 1].next = &t->atid_tab[natids];
1235                 t->afree = t->atid_tab;
1236         }
1237         return 0;
1238 }
1239
1240 static void free_tid_maps(struct tid_info *t)
1241 {
1242         cxgb_free_mem(t->tid_tab);
1243 }
1244
1245 static inline void add_adapter(struct adapter *adap)
1246 {
1247         write_lock_bh(&adapter_list_lock);
1248         list_add_tail(&adap->adapter_list, &adapter_list);
1249         write_unlock_bh(&adapter_list_lock);
1250 }
1251
1252 static inline void remove_adapter(struct adapter *adap)
1253 {
1254         write_lock_bh(&adapter_list_lock);
1255         list_del(&adap->adapter_list);
1256         write_unlock_bh(&adapter_list_lock);
1257 }
1258
1259 int cxgb3_offload_activate(struct adapter *adapter)
1260 {
1261         struct t3cdev *dev = &adapter->tdev;
1262         int natids, err;
1263         struct t3c_data *t;
1264         struct tid_range stid_range, tid_range;
1265         struct mtutab mtutab;
1266         unsigned int l2t_capacity;
1267
1268         t = kzalloc(sizeof(*t), GFP_KERNEL);
1269         if (!t)
1270                 return -ENOMEM;
1271
1272         err = -EOPNOTSUPP;
1273         if (dev->ctl(dev, GET_TX_MAX_CHUNK, &t->tx_max_chunk) < 0 ||
1274             dev->ctl(dev, GET_MAX_OUTSTANDING_WR, &t->max_wrs) < 0 ||
1275             dev->ctl(dev, GET_L2T_CAPACITY, &l2t_capacity) < 0 ||
1276             dev->ctl(dev, GET_MTUS, &mtutab) < 0 ||
1277             dev->ctl(dev, GET_TID_RANGE, &tid_range) < 0 ||
1278             dev->ctl(dev, GET_STID_RANGE, &stid_range) < 0)
1279                 goto out_free;
1280
1281         err = -ENOMEM;
1282         RCU_INIT_POINTER(dev->l2opt, t3_init_l2t(l2t_capacity));
1283         if (!L2DATA(dev))
1284                 goto out_free;
1285
1286         natids = min(tid_range.num / 2, MAX_ATIDS);
1287         err = init_tid_tabs(&t->tid_maps, tid_range.num, natids,
1288                             stid_range.num, ATID_BASE, stid_range.base);
1289         if (err)
1290                 goto out_free_l2t;
1291
1292         t->mtus = mtutab.mtus;
1293         t->nmtus = mtutab.size;
1294
1295         INIT_WORK(&t->tid_release_task, t3_process_tid_release_list);
1296         spin_lock_init(&t->tid_release_lock);
1297         INIT_LIST_HEAD(&t->list_node);
1298         t->dev = dev;
1299
1300         T3C_DATA(dev) = t;
1301         dev->recv = process_rx;
1302         dev->neigh_update = t3_l2t_update;
1303
1304         /* Register netevent handler once */
1305         if (list_empty(&adapter_list))
1306                 register_netevent_notifier(&nb);
1307
1308         t->nofail_skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_KERNEL);
1309         t->release_list_incomplete = 0;
1310
1311         add_adapter(adapter);
1312         return 0;
1313
1314 out_free_l2t:
1315         t3_free_l2t(L2DATA(dev));
1316         RCU_INIT_POINTER(dev->l2opt, NULL);
1317 out_free:
1318         kfree(t);
1319         return err;
1320 }
1321
1322 static void clean_l2_data(struct rcu_head *head)
1323 {
1324         struct l2t_data *d = container_of(head, struct l2t_data, rcu_head);
1325         t3_free_l2t(d);
1326 }
1327
1328
1329 void cxgb3_offload_deactivate(struct adapter *adapter)
1330 {
1331         struct t3cdev *tdev = &adapter->tdev;
1332         struct t3c_data *t = T3C_DATA(tdev);
1333         struct l2t_data *d;
1334
1335         remove_adapter(adapter);
1336         if (list_empty(&adapter_list))
1337                 unregister_netevent_notifier(&nb);
1338
1339         free_tid_maps(&t->tid_maps);
1340         T3C_DATA(tdev) = NULL;
1341         rcu_read_lock();
1342         d = L2DATA(tdev);
1343         rcu_read_unlock();
1344         RCU_INIT_POINTER(tdev->l2opt, NULL);
1345         call_rcu(&d->rcu_head, clean_l2_data);
1346         if (t->nofail_skb)
1347                 kfree_skb(t->nofail_skb);
1348         kfree(t);
1349 }
1350
1351 static inline void register_tdev(struct t3cdev *tdev)
1352 {
1353         static int unit;
1354
1355         mutex_lock(&cxgb3_db_lock);
1356         snprintf(tdev->name, sizeof(tdev->name), "ofld_dev%d", unit++);
1357         list_add_tail(&tdev->ofld_dev_list, &ofld_dev_list);
1358         mutex_unlock(&cxgb3_db_lock);
1359 }
1360
1361 static inline void unregister_tdev(struct t3cdev *tdev)
1362 {
1363         mutex_lock(&cxgb3_db_lock);
1364         list_del(&tdev->ofld_dev_list);
1365         mutex_unlock(&cxgb3_db_lock);
1366 }
1367
1368 static inline int adap2type(struct adapter *adapter)
1369 {
1370         int type = 0;
1371
1372         switch (adapter->params.rev) {
1373         case T3_REV_A:
1374                 type = T3A;
1375                 break;
1376         case T3_REV_B:
1377         case T3_REV_B2:
1378                 type = T3B;
1379                 break;
1380         case T3_REV_C:
1381                 type = T3C;
1382                 break;
1383         }
1384         return type;
1385 }
1386
1387 void __devinit cxgb3_adapter_ofld(struct adapter *adapter)
1388 {
1389         struct t3cdev *tdev = &adapter->tdev;
1390
1391         INIT_LIST_HEAD(&tdev->ofld_dev_list);
1392
1393         cxgb3_set_dummy_ops(tdev);
1394         tdev->send = t3_offload_tx;
1395         tdev->ctl = cxgb_offload_ctl;
1396         tdev->type = adap2type(adapter);
1397
1398         register_tdev(tdev);
1399 }
1400
1401 void __devexit cxgb3_adapter_unofld(struct adapter *adapter)
1402 {
1403         struct t3cdev *tdev = &adapter->tdev;
1404
1405         tdev->recv = NULL;
1406         tdev->neigh_update = NULL;
1407
1408         unregister_tdev(tdev);
1409 }
1410
1411 void __init cxgb3_offload_init(void)
1412 {
1413         int i;
1414
1415         for (i = 0; i < NUM_CPL_CMDS; ++i)
1416                 cpl_handlers[i] = do_bad_cpl;
1417
1418         t3_register_cpl_handler(CPL_SMT_WRITE_RPL, do_smt_write_rpl);
1419         t3_register_cpl_handler(CPL_L2T_WRITE_RPL, do_l2t_write_rpl);
1420         t3_register_cpl_handler(CPL_RTE_WRITE_RPL, do_rte_write_rpl);
1421         t3_register_cpl_handler(CPL_PASS_OPEN_RPL, do_stid_rpl);
1422         t3_register_cpl_handler(CPL_CLOSE_LISTSRV_RPL, do_stid_rpl);
1423         t3_register_cpl_handler(CPL_PASS_ACCEPT_REQ, do_cr);
1424         t3_register_cpl_handler(CPL_PASS_ESTABLISH, do_hwtid_rpl);
1425         t3_register_cpl_handler(CPL_ABORT_RPL_RSS, do_hwtid_rpl);
1426         t3_register_cpl_handler(CPL_ABORT_RPL, do_hwtid_rpl);
1427         t3_register_cpl_handler(CPL_RX_URG_NOTIFY, do_hwtid_rpl);
1428         t3_register_cpl_handler(CPL_RX_DATA, do_hwtid_rpl);
1429         t3_register_cpl_handler(CPL_TX_DATA_ACK, do_hwtid_rpl);
1430         t3_register_cpl_handler(CPL_TX_DMA_ACK, do_hwtid_rpl);
1431         t3_register_cpl_handler(CPL_ACT_OPEN_RPL, do_act_open_rpl);
1432         t3_register_cpl_handler(CPL_PEER_CLOSE, do_hwtid_rpl);
1433         t3_register_cpl_handler(CPL_CLOSE_CON_RPL, do_hwtid_rpl);
1434         t3_register_cpl_handler(CPL_ABORT_REQ_RSS, do_abort_req_rss);
1435         t3_register_cpl_handler(CPL_ACT_ESTABLISH, do_act_establish);
1436         t3_register_cpl_handler(CPL_SET_TCB_RPL, do_hwtid_rpl);
1437         t3_register_cpl_handler(CPL_GET_TCB_RPL, do_hwtid_rpl);
1438         t3_register_cpl_handler(CPL_RDMA_TERMINATE, do_term);
1439         t3_register_cpl_handler(CPL_RDMA_EC_STATUS, do_hwtid_rpl);
1440         t3_register_cpl_handler(CPL_TRACE_PKT, do_trace);
1441         t3_register_cpl_handler(CPL_RX_DATA_DDP, do_hwtid_rpl);
1442         t3_register_cpl_handler(CPL_RX_DDP_COMPLETE, do_hwtid_rpl);
1443         t3_register_cpl_handler(CPL_ISCSI_HDR, do_hwtid_rpl);
1444 }