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         if (insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid)) {
 199                 cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
 200                 kfree(chp);
 201                 return ERR_PTR(-ENOMEM);
 202         }
 203
 204         if (ucontext) {
 205                 struct iwch_mm_entry *mm;
 206
 207                 mm = kmalloc(sizeof *mm, GFP_KERNEL);
 208                 if (!mm) {
 209                         iwch_destroy_cq(&chp->ibcq);
 210                         return ERR_PTR(-ENOMEM);
 211                 }
 212                 uresp.cqid = chp->cq.cqid;
 213                 uresp.size_log2 = chp->cq.size_log2;
 214                 spin_lock(&ucontext->mmap_lock);
 215                 uresp.key = ucontext->key;
 216                 ucontext->key += PAGE_SIZE;
 217                 spin_unlock(&ucontext->mmap_lock);
 218                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 219                         kfree(mm);
 220                         iwch_destroy_cq(&chp->ibcq);
 221                         return ERR_PTR(-EFAULT);
 222                 }
 223                 mm->key = uresp.key;
 224                 mm->addr = virt_to_phys(chp->cq.queue);
 225                 mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
 226                                              sizeof (struct t3_cqe));
 227                 insert_mmap(ucontext, mm);
 228         }
 229         PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
 230              chp->cq.cqid, chp, (1 << chp->cq.size_log2),
 231              (unsigned long long) chp->cq.dma_addr);
 232         return &chp->ibcq;
 233 }
 234
 235 static int iwch_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
 236 {
 237 #ifdef notyet
 238         struct iwch_cq *chp = to_iwch_cq(cq);
 239         struct t3_cq oldcq, newcq;
 240         int ret;
 241
 242         PDBG("%s ib_cq %p cqe %d\n", __func__, cq, cqe);
 243
 244         /* We don't downsize... */
 245         if (cqe <= cq->cqe)
 246                 return 0;
 247
 248         /* create new t3_cq with new size */
 249         cqe = roundup_pow_of_two(cqe+1);
 250         newcq.size_log2 = ilog2(cqe);
 251
 252         /* Dont allow resize to less than the current wce count */
 253         if (cqe < Q_COUNT(chp->cq.rptr, chp->cq.wptr)) {
 254                 return -ENOMEM;
 255         }
 256
 257         /* Quiesce all QPs using this CQ */
 258         ret = iwch_quiesce_qps(chp);
 259         if (ret) {
 260                 return ret;
 261         }
 262
 263         ret = cxio_create_cq(&chp->rhp->rdev, &newcq);
 264         if (ret) {
 265                 return ret;
 266         }
 267
 268         /* copy CQEs */
 269         memcpy(newcq.queue, chp->cq.queue, (1 << chp->cq.size_log2) *
 270                                         sizeof(struct t3_cqe));
 271
 272         /* old iwch_qp gets new t3_cq but keeps old cqid */
 273         oldcq = chp->cq;
 274         chp->cq = newcq;
 275         chp->cq.cqid = oldcq.cqid;
 276
 277         /* resize new t3_cq to update the HW context */
 278         ret = cxio_resize_cq(&chp->rhp->rdev, &chp->cq);
 279         if (ret) {
 280                 chp->cq = oldcq;
 281                 return ret;
 282         }
 283         chp->ibcq.cqe = (1<<chp->cq.size_log2) - 1;
 284
 285         /* destroy old t3_cq */
 286         oldcq.cqid = newcq.cqid;
 287         ret = cxio_destroy_cq(&chp->rhp->rdev, &oldcq);
 288         if (ret) {
 289                 printk(KERN_ERR MOD "%s - cxio_destroy_cq failed %d\n",
 290                         __func__, ret);
 291         }
 292
 293         /* add user hooks here */
 294
 295         /* resume qps */
 296         ret = iwch_resume_qps(chp);
 297         return ret;
 298 #else
 299         return -ENOSYS;
 300 #endif
 301 }
 302
 303 static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
 304 {
 305         struct iwch_dev *rhp;
 306         struct iwch_cq *chp;
 307         enum t3_cq_opcode cq_op;
 308         int err;
 309         unsigned long flag;
 310         u32 rptr;
 311
 312         chp = to_iwch_cq(ibcq);
 313         rhp = chp->rhp;
 314         if ((flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED)
 315                 cq_op = CQ_ARM_SE;
 316         else
 317                 cq_op = CQ_ARM_AN;
 318         if (chp->user_rptr_addr) {
 319                 if (get_user(rptr, chp->user_rptr_addr))
 320                         return -EFAULT;
 321                 spin_lock_irqsave(&chp->lock, flag);
 322                 chp->cq.rptr = rptr;
 323         } else
 324                 spin_lock_irqsave(&chp->lock, flag);
 325         PDBG("%s rptr 0x%x\n", __func__, chp->cq.rptr);
 326         err = cxio_hal_cq_op(&rhp->rdev, &chp->cq, cq_op, 0);
 327         spin_unlock_irqrestore(&chp->lock, flag);
 328         if (err < 0)
 329                 printk(KERN_ERR MOD "Error %d rearming CQID 0x%x\n", err,
 330                        chp->cq.cqid);
 331         if (err > 0 && !(flags & IB_CQ_REPORT_MISSED_EVENTS))
 332                 err = 0;
 333         return err;
 334 }
 335
 336 static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 337 {
 338         int len = vma->vm_end - vma->vm_start;
 339         u32 key = vma->vm_pgoff << PAGE_SHIFT;
 340         struct cxio_rdev *rdev_p;
 341         int ret = 0;
 342         struct iwch_mm_entry *mm;
 343         struct iwch_ucontext *ucontext;
 344         u64 addr;
 345
 346         PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __func__, vma->vm_pgoff,
 347              key, len);
 348
 349         if (vma->vm_start & (PAGE_SIZE-1)) {
 350                 return -EINVAL;
 351         }
 352
 353         rdev_p = &(to_iwch_dev(context->device)->rdev);
 354         ucontext = to_iwch_ucontext(context);
 355
 356         mm = remove_mmap(ucontext, key, len);
 357         if (!mm)
 358                 return -EINVAL;
 359         addr = mm->addr;
 360         kfree(mm);
 361
 362         if ((addr >= rdev_p->rnic_info.udbell_physbase) &&
 363             (addr < (rdev_p->rnic_info.udbell_physbase +
 364                        rdev_p->rnic_info.udbell_len))) {
 365
 366                 /*
 367                  * Map T3 DB register.
 368                  */
 369                 if (vma->vm_flags & VM_READ) {
 370                         return -EPERM;
 371                 }
 372
 373                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 374                 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
 375                 vma->vm_flags &= ~VM_MAYREAD;
 376                 ret = io_remap_pfn_range(vma, vma->vm_start,
 377                                          addr >> PAGE_SHIFT,
 378                                          len, vma->vm_page_prot);
 379         } else {
 380
 381                 /*
 382                  * Map WQ or CQ contig dma memory...
 383                  */
 384                 ret = remap_pfn_range(vma, vma->vm_start,
 385                                       addr >> PAGE_SHIFT,
 386                                       len, vma->vm_page_prot);
 387         }
 388
 389         return ret;
 390 }
 391
 392 static int iwch_deallocate_pd(struct ib_pd *pd)
 393 {
 394         struct iwch_dev *rhp;
 395         struct iwch_pd *php;
 396
 397         php = to_iwch_pd(pd);
 398         rhp = php->rhp;
 399         PDBG("%s ibpd %p pdid 0x%x\n", __func__, pd, php->pdid);
 400         cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid);
 401         kfree(php);
 402         return 0;
 403 }
 404
 405 static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev,
 406                                struct ib_ucontext *context,
 407                                struct ib_udata *udata)
 408 {
 409         struct iwch_pd *php;
 410         u32 pdid;
 411         struct iwch_dev *rhp;
 412
 413         PDBG("%s ibdev %p\n", __func__, ibdev);
 414         rhp = (struct iwch_dev *) ibdev;
 415         pdid = cxio_hal_get_pdid(rhp->rdev.rscp);
 416         if (!pdid)
 417                 return ERR_PTR(-EINVAL);
 418         php = kzalloc(sizeof(*php), GFP_KERNEL);
 419         if (!php) {
 420                 cxio_hal_put_pdid(rhp->rdev.rscp, pdid);
 421                 return ERR_PTR(-ENOMEM);
 422         }
 423         php->pdid = pdid;
 424         php->rhp = rhp;
 425         if (context) {
 426                 if (ib_copy_to_udata(udata, &php->pdid, sizeof (__u32))) {
 427                         iwch_deallocate_pd(&php->ibpd);
 428                         return ERR_PTR(-EFAULT);
 429                 }
 430         }
 431         PDBG("%s pdid 0x%0x ptr 0x%p\n", __func__, pdid, php);
 432         return &php->ibpd;
 433 }
 434
 435 static int iwch_dereg_mr(struct ib_mr *ib_mr)
 436 {
 437         struct iwch_dev *rhp;
 438         struct iwch_mr *mhp;
 439         u32 mmid;
 440
 441         PDBG("%s ib_mr %p\n", __func__, ib_mr);
 442         /* There can be no memory windows */
 443         if (atomic_read(&ib_mr->usecnt))
 444                 return -EINVAL;
 445
 446         mhp = to_iwch_mr(ib_mr);
 447         rhp = mhp->rhp;
 448         mmid = mhp->attr.stag >> 8;
 449         cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
 450                        mhp->attr.pbl_addr);
 451         iwch_free_pbl(mhp);
 452         remove_handle(rhp, &rhp->mmidr, mmid);
 453         if (mhp->kva)
 454                 kfree((void *) (unsigned long) mhp->kva);
 455         if (mhp->umem)
 456                 ib_umem_release(mhp->umem);
 457         PDBG("%s mmid 0x%x ptr %p\n", __func__, mmid, mhp);
 458         kfree(mhp);
 459         return 0;
 460 }
 461
 462 static struct ib_mr *iwch_register_phys_mem(struct ib_pd *pd,
 463                                         struct ib_phys_buf *buffer_list,
 464                                         int num_phys_buf,
 465                                         int acc,
 466                                         u64 *iova_start)
 467 {
 468         __be64 *page_list;
 469         int shift;
 470         u64 total_size;
 471         int npages;
 472         struct iwch_dev *rhp;
 473         struct iwch_pd *php;
 474         struct iwch_mr *mhp;
 475         int ret;
 476
 477         PDBG("%s ib_pd %p\n", __func__, pd);
 478         php = to_iwch_pd(pd);
 479         rhp = php->rhp;
 480
 481         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 482         if (!mhp)
 483                 return ERR_PTR(-ENOMEM);
 484
 485         mhp->rhp = rhp;
 486
 487         /* First check that we have enough alignment */
 488         if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
 489                 ret = -EINVAL;
 490                 goto err;
 491         }
 492
 493         if (num_phys_buf > 1 &&
 494             ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
 495                 ret = -EINVAL;
 496                 goto err;
 497         }
 498
 499         ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
 500                                    &total_size, &npages, &shift, &page_list);
 501         if (ret)
 502                 goto err;
 503
 504         ret = iwch_alloc_pbl(mhp, npages);
 505         if (ret) {
 506                 kfree(page_list);
 507                 goto err_pbl;
 508         }
 509
 510         ret = iwch_write_pbl(mhp, page_list, npages, 0);
 511         kfree(page_list);
 512         if (ret)
 513                 goto err_pbl;
 514
 515         mhp->attr.pdid = php->pdid;
 516         mhp->attr.zbva = 0;
 517
 518         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 519         mhp->attr.va_fbo = *iova_start;
 520         mhp->attr.page_size = shift - 12;
 521
 522         mhp->attr.len = (u32) total_size;
 523         mhp->attr.pbl_size = npages;
 524         ret = iwch_register_mem(rhp, php, mhp, shift);
 525         if (ret)
 526                 goto err_pbl;
 527
 528         return &mhp->ibmr;
 529
 530 err_pbl:
 531         iwch_free_pbl(mhp);
 532
 533 err:
 534         kfree(mhp);
 535         return ERR_PTR(ret);
 536
 537 }
 538
 539 static int iwch_reregister_phys_mem(struct ib_mr *mr,
 540                                      int mr_rereg_mask,
 541                                      struct ib_pd *pd,
 542                                      struct ib_phys_buf *buffer_list,
 543                                      int num_phys_buf,
 544                                      int acc, u64 * iova_start)
 545 {
 546
 547         struct iwch_mr mh, *mhp;
 548         struct iwch_pd *php;
 549         struct iwch_dev *rhp;
 550         __be64 *page_list = NULL;
 551         int shift = 0;
 552         u64 total_size;
 553         int npages;
 554         int ret;
 555
 556         PDBG("%s ib_mr %p ib_pd %p\n", __func__, mr, pd);
 557
 558         /* There can be no memory windows */
 559         if (atomic_read(&mr->usecnt))
 560                 return -EINVAL;
 561
 562         mhp = to_iwch_mr(mr);
 563         rhp = mhp->rhp;
 564         php = to_iwch_pd(mr->pd);
 565
 566         /* make sure we are on the same adapter */
 567         if (rhp != php->rhp)
 568                 return -EINVAL;
 569
 570         memcpy(&mh, mhp, sizeof *mhp);
 571
 572         if (mr_rereg_mask & IB_MR_REREG_PD)
 573                 php = to_iwch_pd(pd);
 574         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 575                 mh.attr.perms = iwch_ib_to_tpt_access(acc);
 576         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 577                 ret = build_phys_page_list(buffer_list, num_phys_buf,
 578                                            iova_start,
 579                                            &total_size, &npages,
 580                                            &shift, &page_list);
 581                 if (ret)
 582                         return ret;
 583         }
 584
 585         ret = iwch_reregister_mem(rhp, php, &mh, shift, npages);
 586         kfree(page_list);
 587         if (ret) {
 588                 return ret;
 589         }
 590         if (mr_rereg_mask & IB_MR_REREG_PD)
 591                 mhp->attr.pdid = php->pdid;
 592         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 593                 mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 594         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 595                 mhp->attr.zbva = 0;
 596                 mhp->attr.va_fbo = *iova_start;
 597                 mhp->attr.page_size = shift - 12;
 598                 mhp->attr.len = (u32) total_size;
 599                 mhp->attr.pbl_size = npages;
 600         }
 601
 602         return 0;
 603 }
 604
 605
 606 static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 607                                       u64 virt, int acc, struct ib_udata *udata)
 608 {
 609         __be64 *pages;
 610         int shift, n, len;
 611         int i, j, k;
 612         int err = 0;
 613         struct ib_umem_chunk *chunk;
 614         struct iwch_dev *rhp;
 615         struct iwch_pd *php;
 616         struct iwch_mr *mhp;
 617         struct iwch_reg_user_mr_resp uresp;
 618
 619         PDBG("%s ib_pd %p\n", __func__, pd);
 620
 621         php = to_iwch_pd(pd);
 622         rhp = php->rhp;
 623         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 624         if (!mhp)
 625                 return ERR_PTR(-ENOMEM);
 626
 627         mhp->rhp = rhp;
 628
 629         mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
 630         if (IS_ERR(mhp->umem)) {
 631                 err = PTR_ERR(mhp->umem);
 632                 kfree(mhp);
 633                 return ERR_PTR(err);
 634         }
 635
 636         shift = ffs(mhp->umem->page_size) - 1;
 637
 638         n = 0;
 639         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 640                 n += chunk->nents;
 641
 642         err = iwch_alloc_pbl(mhp, n);
 643         if (err)
 644                 goto err;
 645
 646         pages = (__be64 *) __get_free_page(GFP_KERNEL);
 647         if (!pages) {
 648                 err = -ENOMEM;
 649                 goto err_pbl;
 650         }
 651
 652         i = n = 0;
 653
 654         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 655                 for (j = 0; j < chunk->nmap; ++j) {
 656                         len = sg_dma_len(&chunk->page_list[j]) >> shift;
 657                         for (k = 0; k < len; ++k) {
 658                                 pages[i++] = cpu_to_be64(sg_dma_address(
 659                                         &chunk->page_list[j]) +
 660                                         mhp->umem->page_size * k);
 661                                 if (i == PAGE_SIZE / sizeof *pages) {
 662                                         err = iwch_write_pbl(mhp, pages, i, n);
 663                                         if (err)
 664                                                 goto pbl_done;
 665                                         n += i;
 666                                         i = 0;
 667                                 }
 668                         }
 669                 }
 670
 671         if (i)
 672                 err = iwch_write_pbl(mhp, pages, i, n);
 673
 674 pbl_done:
 675         free_page((unsigned long) pages);
 676         if (err)
 677                 goto err_pbl;
 678
 679         mhp->attr.pdid = php->pdid;
 680         mhp->attr.zbva = 0;
 681         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 682         mhp->attr.va_fbo = virt;
 683         mhp->attr.page_size = shift - 12;
 684         mhp->attr.len = (u32) length;
 685
 686         err = iwch_register_mem(rhp, php, mhp, shift);
 687         if (err)
 688                 goto err_pbl;
 689
 690         if (udata && !t3a_device(rhp)) {
 691                 uresp.pbl_addr = (mhp->attr.pbl_addr -
 692                                  rhp->rdev.rnic_info.pbl_base) >> 3;
 693                 PDBG("%s user resp pbl_addr 0x%x\n", __func__,
 694                      uresp.pbl_addr);
 695
 696                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 697                         iwch_dereg_mr(&mhp->ibmr);
 698                         err = -EFAULT;
 699                         goto err;
 700                 }
 701         }
 702
 703         return &mhp->ibmr;
 704
 705 err_pbl:
 706         iwch_free_pbl(mhp);
 707
 708 err:
 709         ib_umem_release(mhp->umem);
 710         kfree(mhp);
 711         return ERR_PTR(err);
 712 }
 713
 714 static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
 715 {
 716         struct ib_phys_buf bl;
 717         u64 kva;
 718         struct ib_mr *ibmr;
 719
 720         PDBG("%s ib_pd %p\n", __func__, pd);
 721
 722         /*
 723          * T3 only supports 32 bits of size.
 724          */
 725         bl.size = 0xffffffff;
 726         bl.addr = 0;
 727         kva = 0;
 728         ibmr = iwch_register_phys_mem(pd, &bl, 1, acc, &kva);
 729         return ibmr;
 730 }
 731
 732 static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
 733 {
 734         struct iwch_dev *rhp;
 735         struct iwch_pd *php;
 736         struct iwch_mw *mhp;
 737         u32 mmid;
 738         u32 stag = 0;
 739         int ret;
 740
 741         php = to_iwch_pd(pd);
 742         rhp = php->rhp;
 743         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 744         if (!mhp)
 745                 return ERR_PTR(-ENOMEM);
 746         ret = cxio_allocate_window(&rhp->rdev, &stag, php->pdid);
 747         if (ret) {
 748                 kfree(mhp);
 749                 return ERR_PTR(ret);
 750         }
 751         mhp->rhp = rhp;
 752         mhp->attr.pdid = php->pdid;
 753         mhp->attr.type = TPT_MW;
 754         mhp->attr.stag = stag;
 755         mmid = (stag) >> 8;
 756         mhp->ibmw.rkey = stag;
 757         if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) {
 758                 cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 759                 kfree(mhp);
 760                 return ERR_PTR(-ENOMEM);
 761         }
 762         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 763         return &(mhp->ibmw);
 764 }
 765
 766 static int iwch_dealloc_mw(struct ib_mw *mw)
 767 {
 768         struct iwch_dev *rhp;
 769         struct iwch_mw *mhp;
 770         u32 mmid;
 771
 772         mhp = to_iwch_mw(mw);
 773         rhp = mhp->rhp;
 774         mmid = (mw->rkey) >> 8;
 775         cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 776         remove_handle(rhp, &rhp->mmidr, mmid);
 777         kfree(mhp);
 778         PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
 779         return 0;
 780 }
 781
 782 static struct ib_mr *iwch_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth)
 783 {
 784         struct iwch_dev *rhp;
 785         struct iwch_pd *php;
 786         struct iwch_mr *mhp;
 787         u32 mmid;
 788         u32 stag = 0;
 789         int ret = 0;
 790
 791         php = to_iwch_pd(pd);
 792         rhp = php->rhp;
 793         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 794         if (!mhp)
 795                 goto err;
 796
 797         mhp->rhp = rhp;
 798         ret = iwch_alloc_pbl(mhp, pbl_depth);
 799         if (ret)
 800                 goto err1;
 801         mhp->attr.pbl_size = pbl_depth;
 802         ret = cxio_allocate_stag(&rhp->rdev, &stag, php->pdid,
 803                                  mhp->attr.pbl_size, mhp->attr.pbl_addr);
 804         if (ret)
 805                 goto err2;
 806         mhp->attr.pdid = php->pdid;
 807         mhp->attr.type = TPT_NON_SHARED_MR;
 808         mhp->attr.stag = stag;
 809         mhp->attr.state = 1;
 810         mmid = (stag) >> 8;
 811         mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
 812         if (insert_handle(rhp, &rhp->mmidr, mhp, mmid))
 813                 goto err3;
 814
 815         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 816         return &(mhp->ibmr);
 817 err3:
 818         cxio_dereg_mem(&rhp->rdev, stag, mhp->attr.pbl_size,
 819                        mhp->attr.pbl_addr);
 820 err2:
 821         iwch_free_pbl(mhp);
 822 err1:
 823         kfree(mhp);
 824 err:
 825         return ERR_PTR(ret);
 826 }
 827
 828 static struct ib_fast_reg_page_list *iwch_alloc_fastreg_pbl(
 829                                         struct ib_device *device,
 830                                         int page_list_len)
 831 {
 832         struct ib_fast_reg_page_list *page_list;
 833
 834         page_list = kmalloc(sizeof *page_list + page_list_len * sizeof(u64),
 835                             GFP_KERNEL);
 836         if (!page_list)
 837                 return ERR_PTR(-ENOMEM);
 838
 839         page_list->page_list = (u64 *)(page_list + 1);
 840         page_list->max_page_list_len = page_list_len;
 841
 842         return page_list;
 843 }
 844
 845 static void iwch_free_fastreg_pbl(struct ib_fast_reg_page_list *page_list)
 846 {
 847         kfree(page_list);
 848 }
 849
 850 static int iwch_destroy_qp(struct ib_qp *ib_qp)
 851 {
 852         struct iwch_dev *rhp;
 853         struct iwch_qp *qhp;
 854         struct iwch_qp_attributes attrs;
 855         struct iwch_ucontext *ucontext;
 856
 857         qhp = to_iwch_qp(ib_qp);
 858         rhp = qhp->rhp;
 859
 860         attrs.next_state = IWCH_QP_STATE_ERROR;
 861         iwch_modify_qp(rhp, qhp, IWCH_QP_ATTR_NEXT_STATE, &attrs, 0);
 862         wait_event(qhp->wait, !qhp->ep);
 863
 864         remove_handle(rhp, &rhp->qpidr, qhp->wq.qpid);
 865
 866         atomic_dec(&qhp->refcnt);
 867         wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
 868
 869         ucontext = ib_qp->uobject ? to_iwch_ucontext(ib_qp->uobject->context)
 870                                   : NULL;
 871         cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 872                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 873
 874         PDBG("%s ib_qp %p qpid 0x%0x qhp %p\n", __func__,
 875              ib_qp, qhp->wq.qpid, qhp);
 876         kfree(qhp);
 877         return 0;
 878 }
 879
 880 static struct ib_qp *iwch_create_qp(struct ib_pd *pd,
 881                              struct ib_qp_init_attr *attrs,
 882                              struct ib_udata *udata)
 883 {
 884         struct iwch_dev *rhp;
 885         struct iwch_qp *qhp;
 886         struct iwch_pd *php;
 887         struct iwch_cq *schp;
 888         struct iwch_cq *rchp;
 889         struct iwch_create_qp_resp uresp;
 890         int wqsize, sqsize, rqsize;
 891         struct iwch_ucontext *ucontext;
 892
 893         PDBG("%s ib_pd %p\n", __func__, pd);
 894         if (attrs->qp_type != IB_QPT_RC)
 895                 return ERR_PTR(-EINVAL);
 896         php = to_iwch_pd(pd);
 897         rhp = php->rhp;
 898         schp = get_chp(rhp, ((struct iwch_cq *) attrs->send_cq)->cq.cqid);
 899         rchp = get_chp(rhp, ((struct iwch_cq *) attrs->recv_cq)->cq.cqid);
 900         if (!schp || !rchp)
 901                 return ERR_PTR(-EINVAL);
 902
 903         /* The RQT size must be # of entries + 1 rounded up to a power of two */
 904         rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr);
 905         if (rqsize == attrs->cap.max_recv_wr)
 906                 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr+1);
 907
 908         /* T3 doesn't support RQT depth < 16 */
 909         if (rqsize < 16)
 910                 rqsize = 16;
 911
 912         if (rqsize > T3_MAX_RQ_SIZE)
 913                 return ERR_PTR(-EINVAL);
 914
 915         if (attrs->cap.max_inline_data > T3_MAX_INLINE)
 916                 return ERR_PTR(-EINVAL);
 917
 918         /*
 919          * NOTE: The SQ and total WQ sizes don't need to be
 920          * a power of two.  However, all the code assumes
 921          * they are. EG: Q_FREECNT() and friends.
 922          */
 923         sqsize = roundup_pow_of_two(attrs->cap.max_send_wr);
 924         wqsize = roundup_pow_of_two(rqsize + sqsize);
 925
 926         /*
 927          * Kernel users need more wq space for fastreg WRs which can take
 928          * 2 WR fragments.
 929          */
 930         ucontext = pd->uobject ? to_iwch_ucontext(pd->uobject->context) : NULL;
 931         if (!ucontext && wqsize < (rqsize + (2 * sqsize)))
 932                 wqsize = roundup_pow_of_two(rqsize +
 933                                 roundup_pow_of_two(attrs->cap.max_send_wr * 2));
 934         PDBG("%s wqsize %d sqsize %d rqsize %d\n", __func__,
 935              wqsize, sqsize, rqsize);
 936         qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
 937         if (!qhp)
 938                 return ERR_PTR(-ENOMEM);
 939         qhp->wq.size_log2 = ilog2(wqsize);
 940         qhp->wq.rq_size_log2 = ilog2(rqsize);
 941         qhp->wq.sq_size_log2 = ilog2(sqsize);
 942         if (cxio_create_qp(&rhp->rdev, !udata, &qhp->wq,
 943                            ucontext ? &ucontext->uctx : &rhp->rdev.uctx)) {
 944                 kfree(qhp);
 945                 return ERR_PTR(-ENOMEM);
 946         }
 947
 948         attrs->cap.max_recv_wr = rqsize - 1;
 949         attrs->cap.max_send_wr = sqsize;
 950         attrs->cap.max_inline_data = T3_MAX_INLINE;
 951
 952         qhp->rhp = rhp;
 953         qhp->attr.pd = php->pdid;
 954         qhp->attr.scq = ((struct iwch_cq *) attrs->send_cq)->cq.cqid;
 955         qhp->attr.rcq = ((struct iwch_cq *) attrs->recv_cq)->cq.cqid;
 956         qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
 957         qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
 958         qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
 959         qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
 960         qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
 961         qhp->attr.state = IWCH_QP_STATE_IDLE;
 962         qhp->attr.next_state = IWCH_QP_STATE_IDLE;
 963
 964         /*
 965          * XXX - These don't get passed in from the openib user
 966          * at create time.  The CM sets them via a QP modify.
 967          * Need to fix...  I think the CM should
 968          */
 969         qhp->attr.enable_rdma_read = 1;
 970         qhp->attr.enable_rdma_write = 1;
 971         qhp->attr.enable_bind = 1;
 972         qhp->attr.max_ord = 1;
 973         qhp->attr.max_ird = 1;
 974
 975         spin_lock_init(&qhp->lock);
 976         init_waitqueue_head(&qhp->wait);
 977         atomic_set(&qhp->refcnt, 1);
 978
 979         if (insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid)) {
 980                 cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 981                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 982                 kfree(qhp);
 983                 return ERR_PTR(-ENOMEM);
 984         }
 985
 986         if (udata) {
 987
 988                 struct iwch_mm_entry *mm1, *mm2;
 989
 990                 mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
 991                 if (!mm1) {
 992                         iwch_destroy_qp(&qhp->ibqp);
 993                         return ERR_PTR(-ENOMEM);
 994                 }
 995
 996                 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
 997                 if (!mm2) {
 998                         kfree(mm1);
 999                         iwch_destroy_qp(&qhp->ibqp);
1000                         return ERR_PTR(-ENOMEM);
1001                 }
1002
1003                 uresp.qpid = qhp->wq.qpid;
1004                 uresp.size_log2 = qhp->wq.size_log2;
1005                 uresp.sq_size_log2 = qhp->wq.sq_size_log2;
1006                 uresp.rq_size_log2 = qhp->wq.rq_size_log2;
1007                 spin_lock(&ucontext->mmap_lock);
1008                 uresp.key = ucontext->key;
1009                 ucontext->key += PAGE_SIZE;
1010                 uresp.db_key = ucontext->key;
1011                 ucontext->key += PAGE_SIZE;
1012                 spin_unlock(&ucontext->mmap_lock);
1013                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
1014                         kfree(mm1);
1015                         kfree(mm2);
1016                         iwch_destroy_qp(&qhp->ibqp);
1017                         return ERR_PTR(-EFAULT);
1018                 }
1019                 mm1->key = uresp.key;
1020                 mm1->addr = virt_to_phys(qhp->wq.queue);
1021                 mm1->len = PAGE_ALIGN(wqsize * sizeof (union t3_wr));
1022                 insert_mmap(ucontext, mm1);
1023                 mm2->key = uresp.db_key;
1024                 mm2->addr = qhp->wq.udb & PAGE_MASK;
1025                 mm2->len = PAGE_SIZE;
1026                 insert_mmap(ucontext, mm2);
1027         }
1028         qhp->ibqp.qp_num = qhp->wq.qpid;
1029         init_timer(&(qhp->timer));
1030         PDBG("%s sq_num_entries %d, rq_num_entries %d "
1031              "qpid 0x%0x qhp %p dma_addr 0x%llx size %d rq_addr 0x%x\n",
1032              __func__, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
1033              qhp->wq.qpid, qhp, (unsigned long long) qhp->wq.dma_addr,
1034              1 << qhp->wq.size_log2, qhp->wq.rq_addr);
1035         return &qhp->ibqp;
1036 }
1037
1038 static int iwch_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1039                       int attr_mask, struct ib_udata *udata)
1040 {
1041         struct iwch_dev *rhp;
1042         struct iwch_qp *qhp;
1043         enum iwch_qp_attr_mask mask = 0;
1044         struct iwch_qp_attributes attrs;
1045
1046         PDBG("%s ib_qp %p\n", __func__, ibqp);
1047
1048         /* iwarp does not support the RTR state */
1049         if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
1050                 attr_mask &= ~IB_QP_STATE;
1051
1052         /* Make sure we still have something left to do */
1053         if (!attr_mask)
1054                 return 0;
1055
1056         memset(&attrs, 0, sizeof attrs);
1057         qhp = to_iwch_qp(ibqp);
1058         rhp = qhp->rhp;
1059
1060         attrs.next_state = iwch_convert_state(attr->qp_state);
1061         attrs.enable_rdma_read = (attr->qp_access_flags &
1062                                IB_ACCESS_REMOTE_READ) ?  1 : 0;
1063         attrs.enable_rdma_write = (attr->qp_access_flags &
1064                                 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
1065         attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
1066
1067
1068         mask |= (attr_mask & IB_QP_STATE) ? IWCH_QP_ATTR_NEXT_STATE : 0;
1069         mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
1070                         (IWCH_QP_ATTR_ENABLE_RDMA_READ |
1071                          IWCH_QP_ATTR_ENABLE_RDMA_WRITE |
1072                          IWCH_QP_ATTR_ENABLE_RDMA_BIND) : 0;
1073
1074         return iwch_modify_qp(rhp, qhp, mask, &attrs, 0);
1075 }
1076
1077 void iwch_qp_add_ref(struct ib_qp *qp)
1078 {
1079         PDBG("%s ib_qp %p\n", __func__, qp);
1080         atomic_inc(&(to_iwch_qp(qp)->refcnt));
1081 }
1082
1083 void iwch_qp_rem_ref(struct ib_qp *qp)
1084 {
1085         PDBG("%s ib_qp %p\n", __func__, qp);
1086         if (atomic_dec_and_test(&(to_iwch_qp(qp)->refcnt)))
1087                 wake_up(&(to_iwch_qp(qp)->wait));
1088 }
1089
1090 static struct ib_qp *iwch_get_qp(struct ib_device *dev, int qpn)
1091 {
1092         PDBG("%s ib_dev %p qpn 0x%x\n", __func__, dev, qpn);
1093         return (struct ib_qp *)get_qhp(to_iwch_dev(dev), qpn);
1094 }
1095
1096
1097 static int iwch_query_pkey(struct ib_device *ibdev,
1098                            u8 port, u16 index, u16 * pkey)
1099 {
1100         PDBG("%s ibdev %p\n", __func__, ibdev);
1101         *pkey = 0;
1102         return 0;
1103 }
1104
1105 static int iwch_query_gid(struct ib_device *ibdev, u8 port,
1106                           int index, union ib_gid *gid)
1107 {
1108         struct iwch_dev *dev;
1109
1110         PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
1111                __func__, ibdev, port, index, gid);
1112         dev = to_iwch_dev(ibdev);
1113         BUG_ON(port == 0 || port > 2);
1114         memset(&(gid->raw[0]), 0, sizeof(gid->raw));
1115         memcpy(&(gid->raw[0]), dev->rdev.port_info.lldevs[port-1]->dev_addr, 6);
1116         return 0;
1117 }
1118
1119 static u64 fw_vers_string_to_u64(struct iwch_dev *iwch_dev)
1120 {
1121         struct ethtool_drvinfo info;
1122         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1123         char *cp, *next;
1124         unsigned fw_maj, fw_min, fw_mic;
1125
1126         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1127
1128         next = info.fw_version + 1;
1129         cp = strsep(&next, ".");
1130         sscanf(cp, "%i", &fw_maj);
1131         cp = strsep(&next, ".");
1132         sscanf(cp, "%i", &fw_min);
1133         cp = strsep(&next, ".");
1134         sscanf(cp, "%i", &fw_mic);
1135
1136         return (((u64)fw_maj & 0xffff) << 32) | ((fw_min & 0xffff) << 16) |
1137                (fw_mic & 0xffff);
1138 }
1139
1140 static int iwch_query_device(struct ib_device *ibdev,
1141                              struct ib_device_attr *props)
1142 {
1143
1144         struct iwch_dev *dev;
1145         PDBG("%s ibdev %p\n", __func__, ibdev);
1146
1147         dev = to_iwch_dev(ibdev);
1148         memset(props, 0, sizeof *props);
1149         memcpy(&props->sys_image_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1150         props->hw_ver = dev->rdev.t3cdev_p->type;
1151         props->fw_ver = fw_vers_string_to_u64(dev);
1152         props->device_cap_flags = dev->device_cap_flags;
1153         props->page_size_cap = dev->attr.mem_pgsizes_bitmask;
1154         props->vendor_id = (u32)dev->rdev.rnic_info.pdev->vendor;
1155         props->vendor_part_id = (u32)dev->rdev.rnic_info.pdev->device;
1156         props->max_mr_size = dev->attr.max_mr_size;
1157         props->max_qp = dev->attr.max_qps;
1158         props->max_qp_wr = dev->attr.max_wrs;
1159         props->max_sge = dev->attr.max_sge_per_wr;
1160         props->max_sge_rd = 1;
1161         props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
1162         props->max_qp_init_rd_atom = dev->attr.max_rdma_reads_per_qp;
1163         props->max_cq = dev->attr.max_cqs;
1164         props->max_cqe = dev->attr.max_cqes_per_cq;
1165         props->max_mr = dev->attr.max_mem_regs;
1166         props->max_pd = dev->attr.max_pds;
1167         props->local_ca_ack_delay = 0;
1168         props->max_fast_reg_page_list_len = T3_MAX_FASTREG_DEPTH;
1169
1170         return 0;
1171 }
1172
1173 static int iwch_query_port(struct ib_device *ibdev,
1174                            u8 port, struct ib_port_attr *props)
1175 {
1176         struct iwch_dev *dev;
1177         struct net_device *netdev;
1178         struct in_device *inetdev;
1179
1180         PDBG("%s ibdev %p\n", __func__, ibdev);
1181
1182         dev = to_iwch_dev(ibdev);
1183         netdev = dev->rdev.port_info.lldevs[port-1];
1184
1185         memset(props, 0, sizeof(struct ib_port_attr));
1186         props->max_mtu = IB_MTU_4096;
1187         if (netdev->mtu >= 4096)
1188                 props->active_mtu = IB_MTU_4096;
1189         else if (netdev->mtu >= 2048)
1190                 props->active_mtu = IB_MTU_2048;
1191         else if (netdev->mtu >= 1024)
1192                 props->active_mtu = IB_MTU_1024;
1193         else if (netdev->mtu >= 512)
1194                 props->active_mtu = IB_MTU_512;
1195         else
1196                 props->active_mtu = IB_MTU_256;
1197
1198         if (!netif_carrier_ok(netdev))
1199                 props->state = IB_PORT_DOWN;
1200         else {
1201                 inetdev = in_dev_get(netdev);
1202                 if (inetdev->ifa_list)
1203                         props->state = IB_PORT_ACTIVE;
1204                 else
1205                         props->state = IB_PORT_INIT;
1206                 in_dev_put(inetdev);
1207         }
1208
1209         props->port_cap_flags =
1210             IB_PORT_CM_SUP |
1211             IB_PORT_SNMP_TUNNEL_SUP |
1212             IB_PORT_REINIT_SUP |
1213             IB_PORT_DEVICE_MGMT_SUP |
1214             IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
1215         props->gid_tbl_len = 1;
1216         props->pkey_tbl_len = 1;
1217         props->active_width = 2;
1218         props->active_speed = 2;
1219         props->max_msg_sz = -1;
1220
1221         return 0;
1222 }
1223
1224 static ssize_t show_rev(struct device *dev, struct device_attribute *attr,
1225                         char *buf)
1226 {
1227         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1228                                                  ibdev.dev);
1229         PDBG("%s dev 0x%p\n", __func__, dev);
1230         return sprintf(buf, "%d\n", iwch_dev->rdev.t3cdev_p->type);
1231 }
1232
1233 static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, char *buf)
1234 {
1235         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1236                                                  ibdev.dev);
1237         struct ethtool_drvinfo info;
1238         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1239
1240         PDBG("%s dev 0x%p\n", __func__, dev);
1241         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1242         return sprintf(buf, "%s\n", info.fw_version);
1243 }
1244
1245 static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
1246                         char *buf)
1247 {
1248         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1249                                                  ibdev.dev);
1250         struct ethtool_drvinfo info;
1251         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1252
1253         PDBG("%s dev 0x%p\n", __func__, dev);
1254         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1255         return sprintf(buf, "%s\n", info.driver);
1256 }
1257
1258 static ssize_t show_board(struct device *dev, struct device_attribute *attr,
1259                           char *buf)
1260 {
1261         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1262                                                  ibdev.dev);
1263         PDBG("%s dev 0x%p\n", __func__, dev);
1264         return sprintf(buf, "%x.%x\n", iwch_dev->rdev.rnic_info.pdev->vendor,
1265                        iwch_dev->rdev.rnic_info.pdev->device);
1266 }
1267
1268 static int iwch_get_mib(struct ib_device *ibdev,
1269                         union rdma_protocol_stats *stats)
1270 {
1271         struct iwch_dev *dev;
1272         struct tp_mib_stats m;
1273         int ret;
1274
1275         PDBG("%s ibdev %p\n", __func__, ibdev);
1276         dev = to_iwch_dev(ibdev);
1277         ret = dev->rdev.t3cdev_p->ctl(dev->rdev.t3cdev_p, RDMA_GET_MIB, &m);
1278         if (ret)
1279                 return -ENOSYS;
1280
1281         memset(stats, 0, sizeof *stats);
1282         stats->iw.ipInReceives = ((u64) m.ipInReceive_hi << 32) +
1283                                 m.ipInReceive_lo;
1284         stats->iw.ipInHdrErrors = ((u64) m.ipInHdrErrors_hi << 32) +
1285                                   m.ipInHdrErrors_lo;
1286         stats->iw.ipInAddrErrors = ((u64) m.ipInAddrErrors_hi << 32) +
1287                                    m.ipInAddrErrors_lo;
1288         stats->iw.ipInUnknownProtos = ((u64) m.ipInUnknownProtos_hi << 32) +
1289                                       m.ipInUnknownProtos_lo;
1290         stats->iw.ipInDiscards = ((u64) m.ipInDiscards_hi << 32) +
1291                                  m.ipInDiscards_lo;
1292         stats->iw.ipInDelivers = ((u64) m.ipInDelivers_hi << 32) +
1293                                  m.ipInDelivers_lo;
1294         stats->iw.ipOutRequests = ((u64) m.ipOutRequests_hi << 32) +
1295                                   m.ipOutRequests_lo;
1296         stats->iw.ipOutDiscards = ((u64) m.ipOutDiscards_hi << 32) +
1297                                   m.ipOutDiscards_lo;
1298         stats->iw.ipOutNoRoutes = ((u64) m.ipOutNoRoutes_hi << 32) +
1299                                   m.ipOutNoRoutes_lo;
1300         stats->iw.ipReasmTimeout = (u64) m.ipReasmTimeout;
1301         stats->iw.ipReasmReqds = (u64) m.ipReasmReqds;
1302         stats->iw.ipReasmOKs = (u64) m.ipReasmOKs;
1303         stats->iw.ipReasmFails = (u64) m.ipReasmFails;
1304         stats->iw.tcpActiveOpens = (u64) m.tcpActiveOpens;
1305         stats->iw.tcpPassiveOpens = (u64) m.tcpPassiveOpens;
1306         stats->iw.tcpAttemptFails = (u64) m.tcpAttemptFails;
1307         stats->iw.tcpEstabResets = (u64) m.tcpEstabResets;
1308         stats->iw.tcpOutRsts = (u64) m.tcpOutRsts;
1309         stats->iw.tcpCurrEstab = (u64) m.tcpCurrEstab;
1310         stats->iw.tcpInSegs = ((u64) m.tcpInSegs_hi << 32) +
1311                               m.tcpInSegs_lo;
1312         stats->iw.tcpOutSegs = ((u64) m.tcpOutSegs_hi << 32) +
1313                                m.tcpOutSegs_lo;
1314         stats->iw.tcpRetransSegs = ((u64) m.tcpRetransSeg_hi << 32) +
1315                                   m.tcpRetransSeg_lo;
1316         stats->iw.tcpInErrs = ((u64) m.tcpInErrs_hi << 32) +
1317                               m.tcpInErrs_lo;
1318         stats->iw.tcpRtoMin = (u64) m.tcpRtoMin;
1319         stats->iw.tcpRtoMax = (u64) m.tcpRtoMax;
1320         return 0;
1321 }
1322
1323 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1324 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1325 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1326 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1327
1328 static struct device_attribute *iwch_class_attributes[] = {
1329         &dev_attr_hw_rev,
1330         &dev_attr_fw_ver,
1331         &dev_attr_hca_type,
1332         &dev_attr_board_id,
1333 };
1334
1335 int iwch_register_device(struct iwch_dev *dev)
1336 {
1337         int ret;
1338         int i;
1339
1340         PDBG("%s iwch_dev %p\n", __func__, dev);
1341         strlcpy(dev->ibdev.name, "cxgb3_%d", IB_DEVICE_NAME_MAX);
1342         memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
1343         memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1344         dev->ibdev.owner = THIS_MODULE;
1345         dev->device_cap_flags = IB_DEVICE_LOCAL_DMA_LKEY |
1346                                 IB_DEVICE_MEM_WINDOW |
1347                                 IB_DEVICE_MEM_MGT_EXTENSIONS;
1348
1349         /* cxgb3 supports STag 0. */
1350         dev->ibdev.local_dma_lkey = 0;
1351
1352         dev->ibdev.uverbs_cmd_mask =
1353             (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1354             (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1355             (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1356             (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1357             (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1358             (1ull << IB_USER_VERBS_CMD_REG_MR) |
1359             (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1360             (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1361             (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1362             (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1363             (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1364             (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1365             (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1366             (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1367             (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1368             (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1369             (1ull << IB_USER_VERBS_CMD_POST_RECV);
1370         dev->ibdev.node_type = RDMA_NODE_RNIC;
1371         memcpy(dev->ibdev.node_desc, IWCH_NODE_DESC, sizeof(IWCH_NODE_DESC));
1372         dev->ibdev.phys_port_cnt = dev->rdev.port_info.nports;
1373         dev->ibdev.num_comp_vectors = 1;
1374         dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
1375         dev->ibdev.query_device = iwch_query_device;
1376         dev->ibdev.query_port = iwch_query_port;
1377         dev->ibdev.modify_port = iwch_modify_port;
1378         dev->ibdev.query_pkey = iwch_query_pkey;
1379         dev->ibdev.query_gid = iwch_query_gid;
1380         dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
1381         dev->ibdev.dealloc_ucontext = iwch_dealloc_ucontext;
1382         dev->ibdev.mmap = iwch_mmap;
1383         dev->ibdev.alloc_pd = iwch_allocate_pd;
1384         dev->ibdev.dealloc_pd = iwch_deallocate_pd;
1385         dev->ibdev.create_ah = iwch_ah_create;
1386         dev->ibdev.destroy_ah = iwch_ah_destroy;
1387         dev->ibdev.create_qp = iwch_create_qp;
1388         dev->ibdev.modify_qp = iwch_ib_modify_qp;
1389         dev->ibdev.destroy_qp = iwch_destroy_qp;
1390         dev->ibdev.create_cq = iwch_create_cq;
1391         dev->ibdev.destroy_cq = iwch_destroy_cq;
1392         dev->ibdev.resize_cq = iwch_resize_cq;
1393         dev->ibdev.poll_cq = iwch_poll_cq;
1394         dev->ibdev.get_dma_mr = iwch_get_dma_mr;
1395         dev->ibdev.reg_phys_mr = iwch_register_phys_mem;
1396         dev->ibdev.rereg_phys_mr = iwch_reregister_phys_mem;
1397         dev->ibdev.reg_user_mr = iwch_reg_user_mr;
1398         dev->ibdev.dereg_mr = iwch_dereg_mr;
1399         dev->ibdev.alloc_mw = iwch_alloc_mw;
1400         dev->ibdev.bind_mw = iwch_bind_mw;
1401         dev->ibdev.dealloc_mw = iwch_dealloc_mw;
1402         dev->ibdev.alloc_fast_reg_mr = iwch_alloc_fast_reg_mr;
1403         dev->ibdev.alloc_fast_reg_page_list = iwch_alloc_fastreg_pbl;
1404         dev->ibdev.free_fast_reg_page_list = iwch_free_fastreg_pbl;
1405         dev->ibdev.attach_mcast = iwch_multicast_attach;
1406         dev->ibdev.detach_mcast = iwch_multicast_detach;
1407         dev->ibdev.process_mad = iwch_process_mad;
1408         dev->ibdev.req_notify_cq = iwch_arm_cq;
1409         dev->ibdev.post_send = iwch_post_send;
1410         dev->ibdev.post_recv = iwch_post_receive;
1411         dev->ibdev.get_protocol_stats = iwch_get_mib;
1412
1413         dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
1414         if (!dev->ibdev.iwcm)
1415                 return -ENOMEM;
1416
1417         dev->ibdev.iwcm->connect = iwch_connect;
1418         dev->ibdev.iwcm->accept = iwch_accept_cr;
1419         dev->ibdev.iwcm->reject = iwch_reject_cr;
1420         dev->ibdev.iwcm->create_listen = iwch_create_listen;
1421         dev->ibdev.iwcm->destroy_listen = iwch_destroy_listen;
1422         dev->ibdev.iwcm->add_ref = iwch_qp_add_ref;
1423         dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
1424         dev->ibdev.iwcm->get_qp = iwch_get_qp;
1425
1426         ret = ib_register_device(&dev->ibdev);
1427         if (ret)
1428                 goto bail1;
1429
1430         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i) {
1431                 ret = device_create_file(&dev->ibdev.dev,
1432                                          iwch_class_attributes[i]);
1433                 if (ret) {
1434                         goto bail2;
1435                 }
1436         }
1437         return 0;
1438 bail2:
1439         ib_unregister_device(&dev->ibdev);
1440 bail1:
1441         kfree(dev->ibdev.iwcm);
1442         return ret;
1443 }
1444
1445 void iwch_unregister_device(struct iwch_dev *dev)
1446 {
1447         int i;
1448
1449         PDBG("%s iwch_dev %p\n", __func__, dev);
1450         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i)
1451                 device_remove_file(&dev->ibdev.dev,
1452                                    iwch_class_attributes[i]);
1453         ib_unregister_device(&dev->ibdev);
1454         kfree(dev->ibdev.iwcm);
1455         return;
1456 }