git.oblomov.eu Git - linux-2.6/blob - drivers/infiniband/hw/cxgb3/iwch_provider.c

   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/moduleparam.h>
  34 #include <linux/device.h>
  35 #include <linux/netdevice.h>
  36 #include <linux/etherdevice.h>
  37 #include <linux/delay.h>
  38 #include <linux/errno.h>
  39 #include <linux/list.h>
  40 #include <linux/spinlock.h>
  41 #include <linux/ethtool.h>
  42 #include <linux/rtnetlink.h>
  43 #include <linux/inetdevice.h>
  44
  45 #include <asm/io.h>
  46 #include <asm/irq.h>
  47 #include <asm/byteorder.h>
  48
  49 #include <rdma/iw_cm.h>
  50 #include <rdma/ib_verbs.h>
  51 #include <rdma/ib_smi.h>
  52 #include <rdma/ib_umem.h>
  53 #include <rdma/ib_user_verbs.h>
  54
  55 #include "cxio_hal.h"
  56 #include "iwch.h"
  57 #include "iwch_provider.h"
  58 #include "iwch_cm.h"
  59 #include "iwch_user.h"
  60 #include "common.h"
  61
  62 static int iwch_modify_port(struct ib_device *ibdev,
  63                             u8 port, int port_modify_mask,
  64                             struct ib_port_modify *props)
  65 {
  66         return -ENOSYS;
  67 }
  68
  69 static struct ib_ah *iwch_ah_create(struct ib_pd *pd,
  70                                     struct ib_ah_attr *ah_attr)
  71 {
  72         return ERR_PTR(-ENOSYS);
  73 }
  74
  75 static int iwch_ah_destroy(struct ib_ah *ah)
  76 {
  77         return -ENOSYS;
  78 }
  79
  80 static int iwch_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
  81 {
  82         return -ENOSYS;
  83 }
  84
  85 static int iwch_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
  86 {
  87         return -ENOSYS;
  88 }
  89
  90 static int iwch_process_mad(struct ib_device *ibdev,
  91                             int mad_flags,
  92                             u8 port_num,
  93                             struct ib_wc *in_wc,
  94                             struct ib_grh *in_grh,
  95                             struct ib_mad *in_mad, struct ib_mad *out_mad)
  96 {
  97         return -ENOSYS;
  98 }
  99
 100 static int iwch_dealloc_ucontext(struct ib_ucontext *context)
 101 {
 102         struct iwch_dev *rhp = to_iwch_dev(context->device);
 103         struct iwch_ucontext *ucontext = to_iwch_ucontext(context);
 104         struct iwch_mm_entry *mm, *tmp;
 105
 106         PDBG("%s context %p\n", __func__, context);
 107         list_for_each_entry_safe(mm, tmp, &ucontext->mmaps, entry)
 108                 kfree(mm);
 109         cxio_release_ucontext(&rhp->rdev, &ucontext->uctx);
 110         kfree(ucontext);
 111         return 0;
 112 }
 113
 114 static struct ib_ucontext *iwch_alloc_ucontext(struct ib_device *ibdev,
 115                                         struct ib_udata *udata)
 116 {
 117         struct iwch_ucontext *context;
 118         struct iwch_dev *rhp = to_iwch_dev(ibdev);
 119
 120         PDBG("%s ibdev %p\n", __func__, ibdev);
 121         context = kzalloc(sizeof(*context), GFP_KERNEL);
 122         if (!context)
 123                 return ERR_PTR(-ENOMEM);
 124         cxio_init_ucontext(&rhp->rdev, &context->uctx);
 125         INIT_LIST_HEAD(&context->mmaps);
 126         spin_lock_init(&context->mmap_lock);
 127         return &context->ibucontext;
 128 }
 129
 130 static int iwch_destroy_cq(struct ib_cq *ib_cq)
 131 {
 132         struct iwch_cq *chp;
 133
 134         PDBG("%s ib_cq %p\n", __func__, ib_cq);
 135         chp = to_iwch_cq(ib_cq);
 136
 137         remove_handle(chp->rhp, &chp->rhp->cqidr, chp->cq.cqid);
 138         atomic_dec(&chp->refcnt);
 139         wait_event(chp->wait, !atomic_read(&chp->refcnt));
 140
 141         cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
 142         kfree(chp);
 143         return 0;
 144 }
 145
 146 static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries, int vector,
 147                              struct ib_ucontext *ib_context,
 148                              struct ib_udata *udata)
 149 {
 150         struct iwch_dev *rhp;
 151         struct iwch_cq *chp;
 152         struct iwch_create_cq_resp uresp;
 153         struct iwch_create_cq_req ureq;
 154         struct iwch_ucontext *ucontext = NULL;
 155
 156         PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
 157         rhp = to_iwch_dev(ibdev);
 158         chp = kzalloc(sizeof(*chp), GFP_KERNEL);
 159         if (!chp)
 160                 return ERR_PTR(-ENOMEM);
 161
 162         if (ib_context) {
 163                 ucontext = to_iwch_ucontext(ib_context);
 164                 if (!t3a_device(rhp)) {
 165                         if (ib_copy_from_udata(&ureq, udata, sizeof (ureq))) {
 166                                 kfree(chp);
 167                                 return ERR_PTR(-EFAULT);
 168                         }
 169                         chp->user_rptr_addr = (u32 __user *)(unsigned long)ureq.user_rptr_addr;
 170                 }
 171         }
 172
 173         if (t3a_device(rhp)) {
 174
 175                 /*
 176                  * T3A: Add some fluff to handle extra CQEs inserted
 177                  * for various errors.
 178                  * Additional CQE possibilities:
 179                  *      TERMINATE,
 180                  *      incoming RDMA WRITE Failures
 181                  *      incoming RDMA READ REQUEST FAILUREs
 182                  * NOTE: We cannot ensure the CQ won't overflow.
 183                  */
 184                 entries += 16;
 185         }
 186         entries = roundup_pow_of_two(entries);
 187         chp->cq.size_log2 = ilog2(entries);
 188
 189         if (cxio_create_cq(&rhp->rdev, &chp->cq)) {
 190                 kfree(chp);
 191                 return ERR_PTR(-ENOMEM);
 192         }
 193         chp->rhp = rhp;
 194         chp->ibcq.cqe = 1 << chp->cq.size_log2;
 195         spin_lock_init(&chp->lock);
 196         atomic_set(&chp->refcnt, 1);
 197         init_waitqueue_head(&chp->wait);
 198         insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid);
 199
 200         if (ucontext) {
 201                 struct iwch_mm_entry *mm;
 202
 203                 mm = kmalloc(sizeof *mm, GFP_KERNEL);
 204                 if (!mm) {
 205                         iwch_destroy_cq(&chp->ibcq);
 206                         return ERR_PTR(-ENOMEM);
 207                 }
 208                 uresp.cqid = chp->cq.cqid;
 209                 uresp.size_log2 = chp->cq.size_log2;
 210                 spin_lock(&ucontext->mmap_lock);
 211                 uresp.key = ucontext->key;
 212                 ucontext->key += PAGE_SIZE;
 213                 spin_unlock(&ucontext->mmap_lock);
 214                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 215                         kfree(mm);
 216                         iwch_destroy_cq(&chp->ibcq);
 217                         return ERR_PTR(-EFAULT);
 218                 }
 219                 mm->key = uresp.key;
 220                 mm->addr = virt_to_phys(chp->cq.queue);
 221                 mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
 222                                              sizeof (struct t3_cqe));
 223                 insert_mmap(ucontext, mm);
 224         }
 225         PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
 226              chp->cq.cqid, chp, (1 << chp->cq.size_log2),
 227              (unsigned long long) chp->cq.dma_addr);
 228         return &chp->ibcq;
 229 }
 230
 231 static int iwch_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
 232 {
 233 #ifdef notyet
 234         struct iwch_cq *chp = to_iwch_cq(cq);
 235         struct t3_cq oldcq, newcq;
 236         int ret;
 237
 238         PDBG("%s ib_cq %p cqe %d\n", __func__, cq, cqe);
 239
 240         /* We don't downsize... */
 241         if (cqe <= cq->cqe)
 242                 return 0;
 243
 244         /* create new t3_cq with new size */
 245         cqe = roundup_pow_of_two(cqe+1);
 246         newcq.size_log2 = ilog2(cqe);
 247
 248         /* Dont allow resize to less than the current wce count */
 249         if (cqe < Q_COUNT(chp->cq.rptr, chp->cq.wptr)) {
 250                 return -ENOMEM;
 251         }
 252
 253         /* Quiesce all QPs using this CQ */
 254         ret = iwch_quiesce_qps(chp);
 255         if (ret) {
 256                 return ret;
 257         }
 258
 259         ret = cxio_create_cq(&chp->rhp->rdev, &newcq);
 260         if (ret) {
 261                 return ret;
 262         }
 263
 264         /* copy CQEs */
 265         memcpy(newcq.queue, chp->cq.queue, (1 << chp->cq.size_log2) *
 266                                         sizeof(struct t3_cqe));
 267
 268         /* old iwch_qp gets new t3_cq but keeps old cqid */
 269         oldcq = chp->cq;
 270         chp->cq = newcq;
 271         chp->cq.cqid = oldcq.cqid;
 272
 273         /* resize new t3_cq to update the HW context */
 274         ret = cxio_resize_cq(&chp->rhp->rdev, &chp->cq);
 275         if (ret) {
 276                 chp->cq = oldcq;
 277                 return ret;
 278         }
 279         chp->ibcq.cqe = (1<<chp->cq.size_log2) - 1;
 280
 281         /* destroy old t3_cq */
 282         oldcq.cqid = newcq.cqid;
 283         ret = cxio_destroy_cq(&chp->rhp->rdev, &oldcq);
 284         if (ret) {
 285                 printk(KERN_ERR MOD "%s - cxio_destroy_cq failed %d\n",
 286                         __func__, ret);
 287         }
 288
 289         /* add user hooks here */
 290
 291         /* resume qps */
 292         ret = iwch_resume_qps(chp);
 293         return ret;
 294 #else
 295         return -ENOSYS;
 296 #endif
 297 }
 298
 299 static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
 300 {
 301         struct iwch_dev *rhp;
 302         struct iwch_cq *chp;
 303         enum t3_cq_opcode cq_op;
 304         int err;
 305         unsigned long flag;
 306         u32 rptr;
 307
 308         chp = to_iwch_cq(ibcq);
 309         rhp = chp->rhp;
 310         if ((flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED)
 311                 cq_op = CQ_ARM_SE;
 312         else
 313                 cq_op = CQ_ARM_AN;
 314         if (chp->user_rptr_addr) {
 315                 if (get_user(rptr, chp->user_rptr_addr))
 316                         return -EFAULT;
 317                 spin_lock_irqsave(&chp->lock, flag);
 318                 chp->cq.rptr = rptr;
 319         } else
 320                 spin_lock_irqsave(&chp->lock, flag);
 321         PDBG("%s rptr 0x%x\n", __func__, chp->cq.rptr);
 322         err = cxio_hal_cq_op(&rhp->rdev, &chp->cq, cq_op, 0);
 323         spin_unlock_irqrestore(&chp->lock, flag);
 324         if (err < 0)
 325                 printk(KERN_ERR MOD "Error %d rearming CQID 0x%x\n", err,
 326                        chp->cq.cqid);
 327         if (err > 0 && !(flags & IB_CQ_REPORT_MISSED_EVENTS))
 328                 err = 0;
 329         return err;
 330 }
 331
 332 static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 333 {
 334         int len = vma->vm_end - vma->vm_start;
 335         u32 key = vma->vm_pgoff << PAGE_SHIFT;
 336         struct cxio_rdev *rdev_p;
 337         int ret = 0;
 338         struct iwch_mm_entry *mm;
 339         struct iwch_ucontext *ucontext;
 340         u64 addr;
 341
 342         PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __func__, vma->vm_pgoff,
 343              key, len);
 344
 345         if (vma->vm_start & (PAGE_SIZE-1)) {
 346                 return -EINVAL;
 347         }
 348
 349         rdev_p = &(to_iwch_dev(context->device)->rdev);
 350         ucontext = to_iwch_ucontext(context);
 351
 352         mm = remove_mmap(ucontext, key, len);
 353         if (!mm)
 354                 return -EINVAL;
 355         addr = mm->addr;
 356         kfree(mm);
 357
 358         if ((addr >= rdev_p->rnic_info.udbell_physbase) &&
 359             (addr < (rdev_p->rnic_info.udbell_physbase +
 360                        rdev_p->rnic_info.udbell_len))) {
 361
 362                 /*
 363                  * Map T3 DB register.
 364                  */
 365                 if (vma->vm_flags & VM_READ) {
 366                         return -EPERM;
 367                 }
 368
 369                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 370                 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
 371                 vma->vm_flags &= ~VM_MAYREAD;
 372                 ret = io_remap_pfn_range(vma, vma->vm_start,
 373                                          addr >> PAGE_SHIFT,
 374                                          len, vma->vm_page_prot);
 375         } else {
 376
 377                 /*
 378                  * Map WQ or CQ contig dma memory...
 379                  */
 380                 ret = remap_pfn_range(vma, vma->vm_start,
 381                                       addr >> PAGE_SHIFT,
 382                                       len, vma->vm_page_prot);
 383         }
 384
 385         return ret;
 386 }
 387
 388 static int iwch_deallocate_pd(struct ib_pd *pd)
 389 {
 390         struct iwch_dev *rhp;
 391         struct iwch_pd *php;
 392
 393         php = to_iwch_pd(pd);
 394         rhp = php->rhp;
 395         PDBG("%s ibpd %p pdid 0x%x\n", __func__, pd, php->pdid);
 396         cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid);
 397         kfree(php);
 398         return 0;
 399 }
 400
 401 static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev,
 402                                struct ib_ucontext *context,
 403                                struct ib_udata *udata)
 404 {
 405         struct iwch_pd *php;
 406         u32 pdid;
 407         struct iwch_dev *rhp;
 408
 409         PDBG("%s ibdev %p\n", __func__, ibdev);
 410         rhp = (struct iwch_dev *) ibdev;
 411         pdid = cxio_hal_get_pdid(rhp->rdev.rscp);
 412         if (!pdid)
 413                 return ERR_PTR(-EINVAL);
 414         php = kzalloc(sizeof(*php), GFP_KERNEL);
 415         if (!php) {
 416                 cxio_hal_put_pdid(rhp->rdev.rscp, pdid);
 417                 return ERR_PTR(-ENOMEM);
 418         }
 419         php->pdid = pdid;
 420         php->rhp = rhp;
 421         if (context) {
 422                 if (ib_copy_to_udata(udata, &php->pdid, sizeof (__u32))) {
 423                         iwch_deallocate_pd(&php->ibpd);
 424                         return ERR_PTR(-EFAULT);
 425                 }
 426         }
 427         PDBG("%s pdid 0x%0x ptr 0x%p\n", __func__, pdid, php);
 428         return &php->ibpd;
 429 }
 430
 431 static int iwch_dereg_mr(struct ib_mr *ib_mr)
 432 {
 433         struct iwch_dev *rhp;
 434         struct iwch_mr *mhp;
 435         u32 mmid;
 436
 437         PDBG("%s ib_mr %p\n", __func__, ib_mr);
 438         /* There can be no memory windows */
 439         if (atomic_read(&ib_mr->usecnt))
 440                 return -EINVAL;
 441
 442         mhp = to_iwch_mr(ib_mr);
 443         rhp = mhp->rhp;
 444         mmid = mhp->attr.stag >> 8;
 445         cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
 446                        mhp->attr.pbl_addr);
 447         iwch_free_pbl(mhp);
 448         remove_handle(rhp, &rhp->mmidr, mmid);
 449         if (mhp->kva)
 450                 kfree((void *) (unsigned long) mhp->kva);
 451         if (mhp->umem)
 452                 ib_umem_release(mhp->umem);
 453         PDBG("%s mmid 0x%x ptr %p\n", __func__, mmid, mhp);
 454         kfree(mhp);
 455         return 0;
 456 }
 457
 458 static struct ib_mr *iwch_register_phys_mem(struct ib_pd *pd,
 459                                         struct ib_phys_buf *buffer_list,
 460                                         int num_phys_buf,
 461                                         int acc,
 462                                         u64 *iova_start)
 463 {
 464         __be64 *page_list;
 465         int shift;
 466         u64 total_size;
 467         int npages;
 468         struct iwch_dev *rhp;
 469         struct iwch_pd *php;
 470         struct iwch_mr *mhp;
 471         int ret;
 472
 473         PDBG("%s ib_pd %p\n", __func__, pd);
 474         php = to_iwch_pd(pd);
 475         rhp = php->rhp;
 476
 477         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 478         if (!mhp)
 479                 return ERR_PTR(-ENOMEM);
 480
 481         mhp->rhp = rhp;
 482
 483         /* First check that we have enough alignment */
 484         if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
 485                 ret = -EINVAL;
 486                 goto err;
 487         }
 488
 489         if (num_phys_buf > 1 &&
 490             ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
 491                 ret = -EINVAL;
 492                 goto err;
 493         }
 494
 495         ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
 496                                    &total_size, &npages, &shift, &page_list);
 497         if (ret)
 498                 goto err;
 499
 500         ret = iwch_alloc_pbl(mhp, npages);
 501         if (ret) {
 502                 kfree(page_list);
 503                 goto err_pbl;
 504         }
 505
 506         ret = iwch_write_pbl(mhp, page_list, npages, 0);
 507         kfree(page_list);
 508         if (ret)
 509                 goto err_pbl;
 510
 511         mhp->attr.pdid = php->pdid;
 512         mhp->attr.zbva = 0;
 513
 514         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 515         mhp->attr.va_fbo = *iova_start;
 516         mhp->attr.page_size = shift - 12;
 517
 518         mhp->attr.len = (u32) total_size;
 519         mhp->attr.pbl_size = npages;
 520         ret = iwch_register_mem(rhp, php, mhp, shift);
 521         if (ret)
 522                 goto err_pbl;
 523
 524         return &mhp->ibmr;
 525
 526 err_pbl:
 527         iwch_free_pbl(mhp);
 528
 529 err:
 530         kfree(mhp);
 531         return ERR_PTR(ret);
 532
 533 }
 534
 535 static int iwch_reregister_phys_mem(struct ib_mr *mr,
 536                                      int mr_rereg_mask,
 537                                      struct ib_pd *pd,
 538                                      struct ib_phys_buf *buffer_list,
 539                                      int num_phys_buf,
 540                                      int acc, u64 * iova_start)
 541 {
 542
 543         struct iwch_mr mh, *mhp;
 544         struct iwch_pd *php;
 545         struct iwch_dev *rhp;
 546         __be64 *page_list = NULL;
 547         int shift = 0;
 548         u64 total_size;
 549         int npages;
 550         int ret;
 551
 552         PDBG("%s ib_mr %p ib_pd %p\n", __func__, mr, pd);
 553
 554         /* There can be no memory windows */
 555         if (atomic_read(&mr->usecnt))
 556                 return -EINVAL;
 557
 558         mhp = to_iwch_mr(mr);
 559         rhp = mhp->rhp;
 560         php = to_iwch_pd(mr->pd);
 561
 562         /* make sure we are on the same adapter */
 563         if (rhp != php->rhp)
 564                 return -EINVAL;
 565
 566         memcpy(&mh, mhp, sizeof *mhp);
 567
 568         if (mr_rereg_mask & IB_MR_REREG_PD)
 569                 php = to_iwch_pd(pd);
 570         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 571                 mh.attr.perms = iwch_ib_to_tpt_access(acc);
 572         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 573                 ret = build_phys_page_list(buffer_list, num_phys_buf,
 574                                            iova_start,
 575                                            &total_size, &npages,
 576                                            &shift, &page_list);
 577                 if (ret)
 578                         return ret;
 579         }
 580
 581         ret = iwch_reregister_mem(rhp, php, &mh, shift, npages);
 582         kfree(page_list);
 583         if (ret) {
 584                 return ret;
 585         }
 586         if (mr_rereg_mask & IB_MR_REREG_PD)
 587                 mhp->attr.pdid = php->pdid;
 588         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 589                 mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 590         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 591                 mhp->attr.zbva = 0;
 592                 mhp->attr.va_fbo = *iova_start;
 593                 mhp->attr.page_size = shift - 12;
 594                 mhp->attr.len = (u32) total_size;
 595                 mhp->attr.pbl_size = npages;
 596         }
 597
 598         return 0;
 599 }
 600
 601
 602 static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 603                                       u64 virt, int acc, struct ib_udata *udata)
 604 {
 605         __be64 *pages;
 606         int shift, n, len;
 607         int i, j, k;
 608         int err = 0;
 609         struct ib_umem_chunk *chunk;
 610         struct iwch_dev *rhp;
 611         struct iwch_pd *php;
 612         struct iwch_mr *mhp;
 613         struct iwch_reg_user_mr_resp uresp;
 614
 615         PDBG("%s ib_pd %p\n", __func__, pd);
 616
 617         php = to_iwch_pd(pd);
 618         rhp = php->rhp;
 619         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 620         if (!mhp)
 621                 return ERR_PTR(-ENOMEM);
 622
 623         mhp->rhp = rhp;
 624
 625         mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
 626         if (IS_ERR(mhp->umem)) {
 627                 err = PTR_ERR(mhp->umem);
 628                 kfree(mhp);
 629                 return ERR_PTR(err);
 630         }
 631
 632         shift = ffs(mhp->umem->page_size) - 1;
 633
 634         n = 0;
 635         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 636                 n += chunk->nents;
 637
 638         err = iwch_alloc_pbl(mhp, n);
 639         if (err)
 640                 goto err;
 641
 642         pages = (__be64 *) __get_free_page(GFP_KERNEL);
 643         if (!pages) {
 644                 err = -ENOMEM;
 645                 goto err_pbl;
 646         }
 647
 648         i = n = 0;
 649
 650         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 651                 for (j = 0; j < chunk->nmap; ++j) {
 652                         len = sg_dma_len(&chunk->page_list[j]) >> shift;
 653                         for (k = 0; k < len; ++k) {
 654                                 pages[i++] = cpu_to_be64(sg_dma_address(
 655                                         &chunk->page_list[j]) +
 656                                         mhp->umem->page_size * k);
 657                                 if (i == PAGE_SIZE / sizeof *pages) {
 658                                         err = iwch_write_pbl(mhp, pages, i, n);
 659                                         if (err)
 660                                                 goto pbl_done;
 661                                         n += i;
 662                                         i = 0;
 663                                 }
 664                         }
 665                 }
 666
 667         if (i)
 668                 err = iwch_write_pbl(mhp, pages, i, n);
 669
 670 pbl_done:
 671         free_page((unsigned long) pages);
 672         if (err)
 673                 goto err_pbl;
 674
 675         mhp->attr.pdid = php->pdid;
 676         mhp->attr.zbva = 0;
 677         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 678         mhp->attr.va_fbo = virt;
 679         mhp->attr.page_size = shift - 12;
 680         mhp->attr.len = (u32) length;
 681
 682         err = iwch_register_mem(rhp, php, mhp, shift);
 683         if (err)
 684                 goto err_pbl;
 685
 686         if (udata && !t3a_device(rhp)) {
 687                 uresp.pbl_addr = (mhp->attr.pbl_addr -
 688                                  rhp->rdev.rnic_info.pbl_base) >> 3;
 689                 PDBG("%s user resp pbl_addr 0x%x\n", __func__,
 690                      uresp.pbl_addr);
 691
 692                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 693                         iwch_dereg_mr(&mhp->ibmr);
 694                         err = -EFAULT;
 695                         goto err;
 696                 }
 697         }
 698
 699         return &mhp->ibmr;
 700
 701 err_pbl:
 702         iwch_free_pbl(mhp);
 703
 704 err:
 705         ib_umem_release(mhp->umem);
 706         kfree(mhp);
 707         return ERR_PTR(err);
 708 }
 709
 710 static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
 711 {
 712         struct ib_phys_buf bl;
 713         u64 kva;
 714         struct ib_mr *ibmr;
 715
 716         PDBG("%s ib_pd %p\n", __func__, pd);
 717
 718         /*
 719          * T3 only supports 32 bits of size.
 720          */
 721         bl.size = 0xffffffff;
 722         bl.addr = 0;
 723         kva = 0;
 724         ibmr = iwch_register_phys_mem(pd, &bl, 1, acc, &kva);
 725         return ibmr;
 726 }
 727
 728 static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
 729 {
 730         struct iwch_dev *rhp;
 731         struct iwch_pd *php;
 732         struct iwch_mw *mhp;
 733         u32 mmid;
 734         u32 stag = 0;
 735         int ret;
 736
 737         php = to_iwch_pd(pd);
 738         rhp = php->rhp;
 739         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 740         if (!mhp)
 741                 return ERR_PTR(-ENOMEM);
 742         ret = cxio_allocate_window(&rhp->rdev, &stag, php->pdid);
 743         if (ret) {
 744                 kfree(mhp);
 745                 return ERR_PTR(ret);
 746         }
 747         mhp->rhp = rhp;
 748         mhp->attr.pdid = php->pdid;
 749         mhp->attr.type = TPT_MW;
 750         mhp->attr.stag = stag;
 751         mmid = (stag) >> 8;
 752         mhp->ibmw.rkey = stag;
 753         insert_handle(rhp, &rhp->mmidr, mhp, mmid);
 754         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 755         return &(mhp->ibmw);
 756 }
 757
 758 static int iwch_dealloc_mw(struct ib_mw *mw)
 759 {
 760         struct iwch_dev *rhp;
 761         struct iwch_mw *mhp;
 762         u32 mmid;
 763
 764         mhp = to_iwch_mw(mw);
 765         rhp = mhp->rhp;
 766         mmid = (mw->rkey) >> 8;
 767         cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 768         remove_handle(rhp, &rhp->mmidr, mmid);
 769         kfree(mhp);
 770         PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
 771         return 0;
 772 }
 773
 774 static struct ib_mr *iwch_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth)
 775 {
 776         struct iwch_dev *rhp;
 777         struct iwch_pd *php;
 778         struct iwch_mr *mhp;
 779         u32 mmid;
 780         u32 stag = 0;
 781         int ret;
 782
 783         php = to_iwch_pd(pd);
 784         rhp = php->rhp;
 785         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 786         if (!mhp)
 787                 return ERR_PTR(-ENOMEM);
 788
 789         mhp->rhp = rhp;
 790         ret = iwch_alloc_pbl(mhp, pbl_depth);
 791         if (ret) {
 792                 kfree(mhp);
 793                 return ERR_PTR(ret);
 794         }
 795         mhp->attr.pbl_size = pbl_depth;
 796         ret = cxio_allocate_stag(&rhp->rdev, &stag, php->pdid,
 797                                  mhp->attr.pbl_size, mhp->attr.pbl_addr);
 798         if (ret) {
 799                 iwch_free_pbl(mhp);
 800                 kfree(mhp);
 801                 return ERR_PTR(ret);
 802         }
 803         mhp->attr.pdid = php->pdid;
 804         mhp->attr.type = TPT_NON_SHARED_MR;
 805         mhp->attr.stag = stag;
 806         mhp->attr.state = 1;
 807         mmid = (stag) >> 8;
 808         mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
 809         insert_handle(rhp, &rhp->mmidr, mhp, mmid);
 810         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 811         return &(mhp->ibmr);
 812 }
 813
 814 static struct ib_fast_reg_page_list *iwch_alloc_fastreg_pbl(
 815                                         struct ib_device *device,
 816                                         int page_list_len)
 817 {
 818         struct ib_fast_reg_page_list *page_list;
 819
 820         page_list = kmalloc(sizeof *page_list + page_list_len * sizeof(u64),
 821                             GFP_KERNEL);
 822         if (!page_list)
 823                 return ERR_PTR(-ENOMEM);
 824
 825         page_list->page_list = (u64 *)(page_list + 1);
 826         page_list->max_page_list_len = page_list_len;
 827
 828         return page_list;
 829 }
 830
 831 static void iwch_free_fastreg_pbl(struct ib_fast_reg_page_list *page_list)
 832 {
 833         kfree(page_list);
 834 }
 835
 836 static int iwch_destroy_qp(struct ib_qp *ib_qp)
 837 {
 838         struct iwch_dev *rhp;
 839         struct iwch_qp *qhp;
 840         struct iwch_qp_attributes attrs;
 841         struct iwch_ucontext *ucontext;
 842
 843         qhp = to_iwch_qp(ib_qp);
 844         rhp = qhp->rhp;
 845
 846         attrs.next_state = IWCH_QP_STATE_ERROR;
 847         iwch_modify_qp(rhp, qhp, IWCH_QP_ATTR_NEXT_STATE, &attrs, 0);
 848         wait_event(qhp->wait, !qhp->ep);
 849
 850         remove_handle(rhp, &rhp->qpidr, qhp->wq.qpid);
 851
 852         atomic_dec(&qhp->refcnt);
 853         wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
 854
 855         ucontext = ib_qp->uobject ? to_iwch_ucontext(ib_qp->uobject->context)
 856                                   : NULL;
 857         cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 858                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 859
 860         PDBG("%s ib_qp %p qpid 0x%0x qhp %p\n", __func__,
 861              ib_qp, qhp->wq.qpid, qhp);
 862         kfree(qhp);
 863         return 0;
 864 }
 865
 866 static struct ib_qp *iwch_create_qp(struct ib_pd *pd,
 867                              struct ib_qp_init_attr *attrs,
 868                              struct ib_udata *udata)
 869 {
 870         struct iwch_dev *rhp;
 871         struct iwch_qp *qhp;
 872         struct iwch_pd *php;
 873         struct iwch_cq *schp;
 874         struct iwch_cq *rchp;
 875         struct iwch_create_qp_resp uresp;
 876         int wqsize, sqsize, rqsize;
 877         struct iwch_ucontext *ucontext;
 878
 879         PDBG("%s ib_pd %p\n", __func__, pd);
 880         if (attrs->qp_type != IB_QPT_RC)
 881                 return ERR_PTR(-EINVAL);
 882         php = to_iwch_pd(pd);
 883         rhp = php->rhp;
 884         schp = get_chp(rhp, ((struct iwch_cq *) attrs->send_cq)->cq.cqid);
 885         rchp = get_chp(rhp, ((struct iwch_cq *) attrs->recv_cq)->cq.cqid);
 886         if (!schp || !rchp)
 887                 return ERR_PTR(-EINVAL);
 888
 889         /* The RQT size must be # of entries + 1 rounded up to a power of two */
 890         rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr);
 891         if (rqsize == attrs->cap.max_recv_wr)
 892                 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr+1);
 893
 894         /* T3 doesn't support RQT depth < 16 */
 895         if (rqsize < 16)
 896                 rqsize = 16;
 897
 898         if (rqsize > T3_MAX_RQ_SIZE)
 899                 return ERR_PTR(-EINVAL);
 900
 901         if (attrs->cap.max_inline_data > T3_MAX_INLINE)
 902                 return ERR_PTR(-EINVAL);
 903
 904         /*
 905          * NOTE: The SQ and total WQ sizes don't need to be
 906          * a power of two.  However, all the code assumes
 907          * they are. EG: Q_FREECNT() and friends.
 908          */
 909         sqsize = roundup_pow_of_two(attrs->cap.max_send_wr);
 910         wqsize = roundup_pow_of_two(rqsize + sqsize);
 911
 912         /*
 913          * Kernel users need more wq space for fastreg WRs which can take
 914          * 2 WR fragments.
 915          */
 916         ucontext = pd->uobject ? to_iwch_ucontext(pd->uobject->context) : NULL;
 917         if (!ucontext && wqsize < (rqsize + (2 * sqsize)))
 918                 wqsize = roundup_pow_of_two(rqsize +
 919                                 roundup_pow_of_two(attrs->cap.max_send_wr * 2));
 920         PDBG("%s wqsize %d sqsize %d rqsize %d\n", __func__,
 921              wqsize, sqsize, rqsize);
 922         qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
 923         if (!qhp)
 924                 return ERR_PTR(-ENOMEM);
 925         qhp->wq.size_log2 = ilog2(wqsize);
 926         qhp->wq.rq_size_log2 = ilog2(rqsize);
 927         qhp->wq.sq_size_log2 = ilog2(sqsize);
 928         if (cxio_create_qp(&rhp->rdev, !udata, &qhp->wq,
 929                            ucontext ? &ucontext->uctx : &rhp->rdev.uctx)) {
 930                 kfree(qhp);
 931                 return ERR_PTR(-ENOMEM);
 932         }
 933
 934         attrs->cap.max_recv_wr = rqsize - 1;
 935         attrs->cap.max_send_wr = sqsize;
 936         attrs->cap.max_inline_data = T3_MAX_INLINE;
 937
 938         qhp->rhp = rhp;
 939         qhp->attr.pd = php->pdid;
 940         qhp->attr.scq = ((struct iwch_cq *) attrs->send_cq)->cq.cqid;
 941         qhp->attr.rcq = ((struct iwch_cq *) attrs->recv_cq)->cq.cqid;
 942         qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
 943         qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
 944         qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
 945         qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
 946         qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
 947         qhp->attr.state = IWCH_QP_STATE_IDLE;
 948         qhp->attr.next_state = IWCH_QP_STATE_IDLE;
 949
 950         /*
 951          * XXX - These don't get passed in from the openib user
 952          * at create time.  The CM sets them via a QP modify.
 953          * Need to fix...  I think the CM should
 954          */
 955         qhp->attr.enable_rdma_read = 1;
 956         qhp->attr.enable_rdma_write = 1;
 957         qhp->attr.enable_bind = 1;
 958         qhp->attr.max_ord = 1;
 959         qhp->attr.max_ird = 1;
 960
 961         spin_lock_init(&qhp->lock);
 962         init_waitqueue_head(&qhp->wait);
 963         atomic_set(&qhp->refcnt, 1);
 964         insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid);
 965
 966         if (udata) {
 967
 968                 struct iwch_mm_entry *mm1, *mm2;
 969
 970                 mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
 971                 if (!mm1) {
 972                         iwch_destroy_qp(&qhp->ibqp);
 973                         return ERR_PTR(-ENOMEM);
 974                 }
 975
 976                 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
 977                 if (!mm2) {
 978                         kfree(mm1);
 979                         iwch_destroy_qp(&qhp->ibqp);
 980                         return ERR_PTR(-ENOMEM);
 981                 }
 982
 983                 uresp.qpid = qhp->wq.qpid;
 984                 uresp.size_log2 = qhp->wq.size_log2;
 985                 uresp.sq_size_log2 = qhp->wq.sq_size_log2;
 986                 uresp.rq_size_log2 = qhp->wq.rq_size_log2;
 987                 spin_lock(&ucontext->mmap_lock);
 988                 uresp.key = ucontext->key;
 989                 ucontext->key += PAGE_SIZE;
 990                 uresp.db_key = ucontext->key;
 991                 ucontext->key += PAGE_SIZE;
 992                 spin_unlock(&ucontext->mmap_lock);
 993                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 994                         kfree(mm1);
 995                         kfree(mm2);
 996                         iwch_destroy_qp(&qhp->ibqp);
 997                         return ERR_PTR(-EFAULT);
 998                 }
 999                 mm1->key = uresp.key;
1000                 mm1->addr = virt_to_phys(qhp->wq.queue);
1001                 mm1->len = PAGE_ALIGN(wqsize * sizeof (union t3_wr));
1002                 insert_mmap(ucontext, mm1);
1003                 mm2->key = uresp.db_key;
1004                 mm2->addr = qhp->wq.udb & PAGE_MASK;
1005                 mm2->len = PAGE_SIZE;
1006                 insert_mmap(ucontext, mm2);
1007         }
1008         qhp->ibqp.qp_num = qhp->wq.qpid;
1009         init_timer(&(qhp->timer));
1010         PDBG("%s sq_num_entries %d, rq_num_entries %d "
1011              "qpid 0x%0x qhp %p dma_addr 0x%llx size %d rq_addr 0x%x\n",
1012              __func__, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
1013              qhp->wq.qpid, qhp, (unsigned long long) qhp->wq.dma_addr,
1014              1 << qhp->wq.size_log2, qhp->wq.rq_addr);
1015         return &qhp->ibqp;
1016 }
1017
1018 static int iwch_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1019                       int attr_mask, struct ib_udata *udata)
1020 {
1021         struct iwch_dev *rhp;
1022         struct iwch_qp *qhp;
1023         enum iwch_qp_attr_mask mask = 0;
1024         struct iwch_qp_attributes attrs;
1025
1026         PDBG("%s ib_qp %p\n", __func__, ibqp);
1027
1028         /* iwarp does not support the RTR state */
1029         if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
1030                 attr_mask &= ~IB_QP_STATE;
1031
1032         /* Make sure we still have something left to do */
1033         if (!attr_mask)
1034                 return 0;
1035
1036         memset(&attrs, 0, sizeof attrs);
1037         qhp = to_iwch_qp(ibqp);
1038         rhp = qhp->rhp;
1039
1040         attrs.next_state = iwch_convert_state(attr->qp_state);
1041         attrs.enable_rdma_read = (attr->qp_access_flags &
1042                                IB_ACCESS_REMOTE_READ) ?  1 : 0;
1043         attrs.enable_rdma_write = (attr->qp_access_flags &
1044                                 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
1045         attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
1046
1047
1048         mask |= (attr_mask & IB_QP_STATE) ? IWCH_QP_ATTR_NEXT_STATE : 0;
1049         mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
1050                         (IWCH_QP_ATTR_ENABLE_RDMA_READ |
1051                          IWCH_QP_ATTR_ENABLE_RDMA_WRITE |
1052                          IWCH_QP_ATTR_ENABLE_RDMA_BIND) : 0;
1053
1054         return iwch_modify_qp(rhp, qhp, mask, &attrs, 0);
1055 }
1056
1057 void iwch_qp_add_ref(struct ib_qp *qp)
1058 {
1059         PDBG("%s ib_qp %p\n", __func__, qp);
1060         atomic_inc(&(to_iwch_qp(qp)->refcnt));
1061 }
1062
1063 void iwch_qp_rem_ref(struct ib_qp *qp)
1064 {
1065         PDBG("%s ib_qp %p\n", __func__, qp);
1066         if (atomic_dec_and_test(&(to_iwch_qp(qp)->refcnt)))
1067                 wake_up(&(to_iwch_qp(qp)->wait));
1068 }
1069
1070 static struct ib_qp *iwch_get_qp(struct ib_device *dev, int qpn)
1071 {
1072         PDBG("%s ib_dev %p qpn 0x%x\n", __func__, dev, qpn);
1073         return (struct ib_qp *)get_qhp(to_iwch_dev(dev), qpn);
1074 }
1075
1076
1077 static int iwch_query_pkey(struct ib_device *ibdev,
1078                            u8 port, u16 index, u16 * pkey)
1079 {
1080         PDBG("%s ibdev %p\n", __func__, ibdev);
1081         *pkey = 0;
1082         return 0;
1083 }
1084
1085 static int iwch_query_gid(struct ib_device *ibdev, u8 port,
1086                           int index, union ib_gid *gid)
1087 {
1088         struct iwch_dev *dev;
1089
1090         PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
1091                __func__, ibdev, port, index, gid);
1092         dev = to_iwch_dev(ibdev);
1093         BUG_ON(port == 0 || port > 2);
1094         memset(&(gid->raw[0]), 0, sizeof(gid->raw));
1095         memcpy(&(gid->raw[0]), dev->rdev.port_info.lldevs[port-1]->dev_addr, 6);
1096         return 0;
1097 }
1098
1099 static u64 fw_vers_string_to_u64(struct iwch_dev *iwch_dev)
1100 {
1101         struct ethtool_drvinfo info;
1102         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1103         char *cp, *next;
1104         unsigned fw_maj, fw_min, fw_mic;
1105
1106         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1107
1108         next = info.fw_version + 1;
1109         cp = strsep(&next, ".");
1110         sscanf(cp, "%i", &fw_maj);
1111         cp = strsep(&next, ".");
1112         sscanf(cp, "%i", &fw_min);
1113         cp = strsep(&next, ".");
1114         sscanf(cp, "%i", &fw_mic);
1115
1116         return (((u64)fw_maj & 0xffff) << 32) | ((fw_min & 0xffff) << 16) |
1117                (fw_mic & 0xffff);
1118 }
1119
1120 static int iwch_query_device(struct ib_device *ibdev,
1121                              struct ib_device_attr *props)
1122 {
1123
1124         struct iwch_dev *dev;
1125         PDBG("%s ibdev %p\n", __func__, ibdev);
1126
1127         dev = to_iwch_dev(ibdev);
1128         memset(props, 0, sizeof *props);
1129         memcpy(&props->sys_image_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1130         props->hw_ver = dev->rdev.t3cdev_p->type;
1131         props->fw_ver = fw_vers_string_to_u64(dev);
1132         props->device_cap_flags = dev->device_cap_flags;
1133         props->page_size_cap = dev->attr.mem_pgsizes_bitmask;
1134         props->vendor_id = (u32)dev->rdev.rnic_info.pdev->vendor;
1135         props->vendor_part_id = (u32)dev->rdev.rnic_info.pdev->device;
1136         props->max_mr_size = dev->attr.max_mr_size;
1137         props->max_qp = dev->attr.max_qps;
1138         props->max_qp_wr = dev->attr.max_wrs;
1139         props->max_sge = dev->attr.max_sge_per_wr;
1140         props->max_sge_rd = 1;
1141         props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
1142         props->max_qp_init_rd_atom = dev->attr.max_rdma_reads_per_qp;
1143         props->max_cq = dev->attr.max_cqs;
1144         props->max_cqe = dev->attr.max_cqes_per_cq;
1145         props->max_mr = dev->attr.max_mem_regs;
1146         props->max_pd = dev->attr.max_pds;
1147         props->local_ca_ack_delay = 0;
1148         props->max_fast_reg_page_list_len = T3_MAX_FASTREG_DEPTH;
1149
1150         return 0;
1151 }
1152
1153 static int iwch_query_port(struct ib_device *ibdev,
1154                            u8 port, struct ib_port_attr *props)
1155 {
1156         struct iwch_dev *dev;
1157         struct net_device *netdev;
1158         struct in_device *inetdev;
1159
1160         PDBG("%s ibdev %p\n", __func__, ibdev);
1161
1162         dev = to_iwch_dev(ibdev);
1163         netdev = dev->rdev.port_info.lldevs[port-1];
1164
1165         memset(props, 0, sizeof(struct ib_port_attr));
1166         props->max_mtu = IB_MTU_4096;
1167         if (netdev->mtu >= 4096)
1168                 props->active_mtu = IB_MTU_4096;
1169         else if (netdev->mtu >= 2048)
1170                 props->active_mtu = IB_MTU_2048;
1171         else if (netdev->mtu >= 1024)
1172                 props->active_mtu = IB_MTU_1024;
1173         else if (netdev->mtu >= 512)
1174                 props->active_mtu = IB_MTU_512;
1175         else
1176                 props->active_mtu = IB_MTU_256;
1177
1178         if (!netif_carrier_ok(netdev))
1179                 props->state = IB_PORT_DOWN;
1180         else {
1181                 inetdev = in_dev_get(netdev);
1182                 if (inetdev->ifa_list)
1183                         props->state = IB_PORT_ACTIVE;
1184                 else
1185                         props->state = IB_PORT_INIT;
1186                 in_dev_put(inetdev);
1187         }
1188
1189         props->port_cap_flags =
1190             IB_PORT_CM_SUP |
1191             IB_PORT_SNMP_TUNNEL_SUP |
1192             IB_PORT_REINIT_SUP |
1193             IB_PORT_DEVICE_MGMT_SUP |
1194             IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
1195         props->gid_tbl_len = 1;
1196         props->pkey_tbl_len = 1;
1197         props->active_width = 2;
1198         props->active_speed = 2;
1199         props->max_msg_sz = -1;
1200
1201         return 0;
1202 }
1203
1204 static ssize_t show_rev(struct device *dev, struct device_attribute *attr,
1205                         char *buf)
1206 {
1207         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1208                                                  ibdev.dev);
1209         PDBG("%s dev 0x%p\n", __func__, dev);
1210         return sprintf(buf, "%d\n", iwch_dev->rdev.t3cdev_p->type);
1211 }
1212
1213 static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, char *buf)
1214 {
1215         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1216                                                  ibdev.dev);
1217         struct ethtool_drvinfo info;
1218         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1219
1220         PDBG("%s dev 0x%p\n", __func__, dev);
1221         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1222         return sprintf(buf, "%s\n", info.fw_version);
1223 }
1224
1225 static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
1226                         char *buf)
1227 {
1228         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1229                                                  ibdev.dev);
1230         struct ethtool_drvinfo info;
1231         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1232
1233         PDBG("%s dev 0x%p\n", __func__, dev);
1234         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1235         return sprintf(buf, "%s\n", info.driver);
1236 }
1237
1238 static ssize_t show_board(struct device *dev, struct device_attribute *attr,
1239                           char *buf)
1240 {
1241         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1242                                                  ibdev.dev);
1243         PDBG("%s dev 0x%p\n", __func__, dev);
1244         return sprintf(buf, "%x.%x\n", iwch_dev->rdev.rnic_info.pdev->vendor,
1245                        iwch_dev->rdev.rnic_info.pdev->device);
1246 }
1247
1248 static int iwch_get_mib(struct ib_device *ibdev,
1249                         union rdma_protocol_stats *stats)
1250 {
1251         struct iwch_dev *dev;
1252         struct tp_mib_stats m;
1253         int ret;
1254
1255         PDBG("%s ibdev %p\n", __func__, ibdev);
1256         dev = to_iwch_dev(ibdev);
1257         ret = dev->rdev.t3cdev_p->ctl(dev->rdev.t3cdev_p, RDMA_GET_MIB, &m);
1258         if (ret)
1259                 return -ENOSYS;
1260
1261         memset(stats, 0, sizeof *stats);
1262         stats->iw.ipInReceives = ((u64) m.ipInReceive_hi << 32) +
1263                                 m.ipInReceive_lo;
1264         stats->iw.ipInHdrErrors = ((u64) m.ipInHdrErrors_hi << 32) +
1265                                   m.ipInHdrErrors_lo;
1266         stats->iw.ipInAddrErrors = ((u64) m.ipInAddrErrors_hi << 32) +
1267                                    m.ipInAddrErrors_lo;
1268         stats->iw.ipInUnknownProtos = ((u64) m.ipInUnknownProtos_hi << 32) +
1269                                       m.ipInUnknownProtos_lo;
1270         stats->iw.ipInDiscards = ((u64) m.ipInDiscards_hi << 32) +
1271                                  m.ipInDiscards_lo;
1272         stats->iw.ipInDelivers = ((u64) m.ipInDelivers_hi << 32) +
1273                                  m.ipInDelivers_lo;
1274         stats->iw.ipOutRequests = ((u64) m.ipOutRequests_hi << 32) +
1275                                   m.ipOutRequests_lo;
1276         stats->iw.ipOutDiscards = ((u64) m.ipOutDiscards_hi << 32) +
1277                                   m.ipOutDiscards_lo;
1278         stats->iw.ipOutNoRoutes = ((u64) m.ipOutNoRoutes_hi << 32) +
1279                                   m.ipOutNoRoutes_lo;
1280         stats->iw.ipReasmTimeout = (u64) m.ipReasmTimeout;
1281         stats->iw.ipReasmReqds = (u64) m.ipReasmReqds;
1282         stats->iw.ipReasmOKs = (u64) m.ipReasmOKs;
1283         stats->iw.ipReasmFails = (u64) m.ipReasmFails;
1284         stats->iw.tcpActiveOpens = (u64) m.tcpActiveOpens;
1285         stats->iw.tcpPassiveOpens = (u64) m.tcpPassiveOpens;
1286         stats->iw.tcpAttemptFails = (u64) m.tcpAttemptFails;
1287         stats->iw.tcpEstabResets = (u64) m.tcpEstabResets;
1288         stats->iw.tcpOutRsts = (u64) m.tcpOutRsts;
1289         stats->iw.tcpCurrEstab = (u64) m.tcpCurrEstab;
1290         stats->iw.tcpInSegs = ((u64) m.tcpInSegs_hi << 32) +
1291                               m.tcpInSegs_lo;
1292         stats->iw.tcpOutSegs = ((u64) m.tcpOutSegs_hi << 32) +
1293                                m.tcpOutSegs_lo;
1294         stats->iw.tcpRetransSegs = ((u64) m.tcpRetransSeg_hi << 32) +
1295                                   m.tcpRetransSeg_lo;
1296         stats->iw.tcpInErrs = ((u64) m.tcpInErrs_hi << 32) +
1297                               m.tcpInErrs_lo;
1298         stats->iw.tcpRtoMin = (u64) m.tcpRtoMin;
1299         stats->iw.tcpRtoMax = (u64) m.tcpRtoMax;
1300         return 0;
1301 }
1302
1303 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1304 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1305 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1306 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1307
1308 static struct device_attribute *iwch_class_attributes[] = {
1309         &dev_attr_hw_rev,
1310         &dev_attr_fw_ver,
1311         &dev_attr_hca_type,
1312         &dev_attr_board_id,
1313 };
1314
1315 int iwch_register_device(struct iwch_dev *dev)
1316 {
1317         int ret;
1318         int i;
1319
1320         PDBG("%s iwch_dev %p\n", __func__, dev);
1321         strlcpy(dev->ibdev.name, "cxgb3_%d", IB_DEVICE_NAME_MAX);
1322         memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
1323         memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1324         dev->ibdev.owner = THIS_MODULE;
1325         dev->device_cap_flags = IB_DEVICE_LOCAL_DMA_LKEY |
1326                                 IB_DEVICE_MEM_WINDOW |
1327                                 IB_DEVICE_MEM_MGT_EXTENSIONS;
1328
1329         /* cxgb3 supports STag 0. */
1330         dev->ibdev.local_dma_lkey = 0;
1331
1332         dev->ibdev.uverbs_cmd_mask =
1333             (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1334             (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1335             (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1336             (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1337             (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1338             (1ull << IB_USER_VERBS_CMD_REG_MR) |
1339             (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1340             (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1341             (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1342             (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1343             (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1344             (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1345             (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1346             (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1347             (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1348             (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1349             (1ull << IB_USER_VERBS_CMD_POST_RECV);
1350         dev->ibdev.node_type = RDMA_NODE_RNIC;
1351         memcpy(dev->ibdev.node_desc, IWCH_NODE_DESC, sizeof(IWCH_NODE_DESC));
1352         dev->ibdev.phys_port_cnt = dev->rdev.port_info.nports;
1353         dev->ibdev.num_comp_vectors = 1;
1354         dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
1355         dev->ibdev.query_device = iwch_query_device;
1356         dev->ibdev.query_port = iwch_query_port;
1357         dev->ibdev.modify_port = iwch_modify_port;
1358         dev->ibdev.query_pkey = iwch_query_pkey;
1359         dev->ibdev.query_gid = iwch_query_gid;
1360         dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
1361         dev->ibdev.dealloc_ucontext = iwch_dealloc_ucontext;
1362         dev->ibdev.mmap = iwch_mmap;
1363         dev->ibdev.alloc_pd = iwch_allocate_pd;
1364         dev->ibdev.dealloc_pd = iwch_deallocate_pd;
1365         dev->ibdev.create_ah = iwch_ah_create;
1366         dev->ibdev.destroy_ah = iwch_ah_destroy;
1367         dev->ibdev.create_qp = iwch_create_qp;
1368         dev->ibdev.modify_qp = iwch_ib_modify_qp;
1369         dev->ibdev.destroy_qp = iwch_destroy_qp;
1370         dev->ibdev.create_cq = iwch_create_cq;
1371         dev->ibdev.destroy_cq = iwch_destroy_cq;
1372         dev->ibdev.resize_cq = iwch_resize_cq;
1373         dev->ibdev.poll_cq = iwch_poll_cq;
1374         dev->ibdev.get_dma_mr = iwch_get_dma_mr;
1375         dev->ibdev.reg_phys_mr = iwch_register_phys_mem;
1376         dev->ibdev.rereg_phys_mr = iwch_reregister_phys_mem;
1377         dev->ibdev.reg_user_mr = iwch_reg_user_mr;
1378         dev->ibdev.dereg_mr = iwch_dereg_mr;
1379         dev->ibdev.alloc_mw = iwch_alloc_mw;
1380         dev->ibdev.bind_mw = iwch_bind_mw;
1381         dev->ibdev.dealloc_mw = iwch_dealloc_mw;
1382         dev->ibdev.alloc_fast_reg_mr = iwch_alloc_fast_reg_mr;
1383         dev->ibdev.alloc_fast_reg_page_list = iwch_alloc_fastreg_pbl;
1384         dev->ibdev.free_fast_reg_page_list = iwch_free_fastreg_pbl;
1385         dev->ibdev.attach_mcast = iwch_multicast_attach;
1386         dev->ibdev.detach_mcast = iwch_multicast_detach;
1387         dev->ibdev.process_mad = iwch_process_mad;
1388         dev->ibdev.req_notify_cq = iwch_arm_cq;
1389         dev->ibdev.post_send = iwch_post_send;
1390         dev->ibdev.post_recv = iwch_post_receive;
1391         dev->ibdev.get_protocol_stats = iwch_get_mib;
1392
1393         dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
1394         if (!dev->ibdev.iwcm)
1395                 return -ENOMEM;
1396
1397         dev->ibdev.iwcm->connect = iwch_connect;
1398         dev->ibdev.iwcm->accept = iwch_accept_cr;
1399         dev->ibdev.iwcm->reject = iwch_reject_cr;
1400         dev->ibdev.iwcm->create_listen = iwch_create_listen;
1401         dev->ibdev.iwcm->destroy_listen = iwch_destroy_listen;
1402         dev->ibdev.iwcm->add_ref = iwch_qp_add_ref;
1403         dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
1404         dev->ibdev.iwcm->get_qp = iwch_get_qp;
1405
1406         ret = ib_register_device(&dev->ibdev);
1407         if (ret)
1408                 goto bail1;
1409
1410         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i) {
1411                 ret = device_create_file(&dev->ibdev.dev,
1412                                          iwch_class_attributes[i]);
1413                 if (ret) {
1414                         goto bail2;
1415                 }
1416         }
1417         return 0;
1418 bail2:
1419         ib_unregister_device(&dev->ibdev);
1420 bail1:
1421         return ret;
1422 }
1423
1424 void iwch_unregister_device(struct iwch_dev *dev)
1425 {
1426         int i;
1427
1428         PDBG("%s iwch_dev %p\n", __func__, dev);
1429         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i)
1430                 device_remove_file(&dev->ibdev.dev,
1431                                    iwch_class_attributes[i]);
1432         ib_unregister_device(&dev->ibdev);
1433         return;
1434 }