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/sched.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/rtnetlink.h>
  44 #include <linux/inetdevice.h>
  45 #include <linux/slab.h>
  46
  47 #include <asm/io.h>
  48 #include <asm/irq.h>
  49 #include <asm/byteorder.h>
  50
  51 #include <rdma/iw_cm.h>
  52 #include <rdma/ib_verbs.h>
  53 #include <rdma/ib_smi.h>
  54 #include <rdma/ib_umem.h>
  55 #include <rdma/ib_user_verbs.h>
  56
  57 #include "cxio_hal.h"
  58 #include "iwch.h"
  59 #include "iwch_provider.h"
  60 #include "iwch_cm.h"
  61 #include "iwch_user.h"
  62 #include "common.h"
  63
  64 static int iwch_modify_port(struct ib_device *ibdev,
  65                             u8 port, int port_modify_mask,
  66                             struct ib_port_modify *props)
  67 {
  68         return -ENOSYS;
  69 }
  70
  71 static struct ib_ah *iwch_ah_create(struct ib_pd *pd,
  72                                     struct ib_ah_attr *ah_attr)
  73 {
  74         return ERR_PTR(-ENOSYS);
  75 }
  76
  77 static int iwch_ah_destroy(struct ib_ah *ah)
  78 {
  79         return -ENOSYS;
  80 }
  81
  82 static int iwch_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
  83 {
  84         return -ENOSYS;
  85 }
  86
  87 static int iwch_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
  88 {
  89         return -ENOSYS;
  90 }
  91
  92 static int iwch_process_mad(struct ib_device *ibdev,
  93                             int mad_flags,
  94                             u8 port_num,
  95                             struct ib_wc *in_wc,
  96                             struct ib_grh *in_grh,
  97                             struct ib_mad *in_mad, struct ib_mad *out_mad)
  98 {
  99         return -ENOSYS;
 100 }
 101
 102 static int iwch_dealloc_ucontext(struct ib_ucontext *context)
 103 {
 104         struct iwch_dev *rhp = to_iwch_dev(context->device);
 105         struct iwch_ucontext *ucontext = to_iwch_ucontext(context);
 106         struct iwch_mm_entry *mm, *tmp;
 107
 108         PDBG("%s context %p\n", __func__, context);
 109         list_for_each_entry_safe(mm, tmp, &ucontext->mmaps, entry)
 110                 kfree(mm);
 111         cxio_release_ucontext(&rhp->rdev, &ucontext->uctx);
 112         kfree(ucontext);
 113         return 0;
 114 }
 115
 116 static struct ib_ucontext *iwch_alloc_ucontext(struct ib_device *ibdev,
 117                                         struct ib_udata *udata)
 118 {
 119         struct iwch_ucontext *context;
 120         struct iwch_dev *rhp = to_iwch_dev(ibdev);
 121
 122         PDBG("%s ibdev %p\n", __func__, ibdev);
 123         context = kzalloc(sizeof(*context), GFP_KERNEL);
 124         if (!context)
 125                 return ERR_PTR(-ENOMEM);
 126         cxio_init_ucontext(&rhp->rdev, &context->uctx);
 127         INIT_LIST_HEAD(&context->mmaps);
 128         spin_lock_init(&context->mmap_lock);
 129         return &context->ibucontext;
 130 }
 131
 132 static int iwch_destroy_cq(struct ib_cq *ib_cq)
 133 {
 134         struct iwch_cq *chp;
 135
 136         PDBG("%s ib_cq %p\n", __func__, ib_cq);
 137         chp = to_iwch_cq(ib_cq);
 138
 139         remove_handle(chp->rhp, &chp->rhp->cqidr, chp->cq.cqid);
 140         atomic_dec(&chp->refcnt);
 141         wait_event(chp->wait, !atomic_read(&chp->refcnt));
 142
 143         cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
 144         kfree(chp);
 145         return 0;
 146 }
 147
 148 static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries, int vector,
 149                              struct ib_ucontext *ib_context,
 150                              struct ib_udata *udata)
 151 {
 152         struct iwch_dev *rhp;
 153         struct iwch_cq *chp;
 154         struct iwch_create_cq_resp uresp;
 155         struct iwch_create_cq_req ureq;
 156         struct iwch_ucontext *ucontext = NULL;
 157         static int warned;
 158         size_t resplen;
 159
 160         PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
 161         rhp = to_iwch_dev(ibdev);
 162         chp = kzalloc(sizeof(*chp), GFP_KERNEL);
 163         if (!chp)
 164                 return ERR_PTR(-ENOMEM);
 165
 166         if (ib_context) {
 167                 ucontext = to_iwch_ucontext(ib_context);
 168                 if (!t3a_device(rhp)) {
 169                         if (ib_copy_from_udata(&ureq, udata, sizeof (ureq))) {
 170                                 kfree(chp);
 171                                 return ERR_PTR(-EFAULT);
 172                         }
 173                         chp->user_rptr_addr = (u32 __user *)(unsigned long)ureq.user_rptr_addr;
 174                 }
 175         }
 176
 177         if (t3a_device(rhp)) {
 178
 179                 /*
 180                  * T3A: Add some fluff to handle extra CQEs inserted
 181                  * for various errors.
 182                  * Additional CQE possibilities:
 183                  *      TERMINATE,
 184                  *      incoming RDMA WRITE Failures
 185                  *      incoming RDMA READ REQUEST FAILUREs
 186                  * NOTE: We cannot ensure the CQ won't overflow.
 187                  */
 188                 entries += 16;
 189         }
 190         entries = roundup_pow_of_two(entries);
 191         chp->cq.size_log2 = ilog2(entries);
 192
 193         if (cxio_create_cq(&rhp->rdev, &chp->cq, !ucontext)) {
 194                 kfree(chp);
 195                 return ERR_PTR(-ENOMEM);
 196         }
 197         chp->rhp = rhp;
 198         chp->ibcq.cqe = 1 << chp->cq.size_log2;
 199         spin_lock_init(&chp->lock);
 200         atomic_set(&chp->refcnt, 1);
 201         init_waitqueue_head(&chp->wait);
 202         if (insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid)) {
 203                 cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
 204                 kfree(chp);
 205                 return ERR_PTR(-ENOMEM);
 206         }
 207
 208         if (ucontext) {
 209                 struct iwch_mm_entry *mm;
 210
 211                 mm = kmalloc(sizeof *mm, GFP_KERNEL);
 212                 if (!mm) {
 213                         iwch_destroy_cq(&chp->ibcq);
 214                         return ERR_PTR(-ENOMEM);
 215                 }
 216                 uresp.cqid = chp->cq.cqid;
 217                 uresp.size_log2 = chp->cq.size_log2;
 218                 spin_lock(&ucontext->mmap_lock);
 219                 uresp.key = ucontext->key;
 220                 ucontext->key += PAGE_SIZE;
 221                 spin_unlock(&ucontext->mmap_lock);
 222                 mm->key = uresp.key;
 223                 mm->addr = virt_to_phys(chp->cq.queue);
 224                 if (udata->outlen < sizeof uresp) {
 225                         if (!warned++)
 226                                 printk(KERN_WARNING MOD "Warning - "
 227                                        "downlevel libcxgb3 (non-fatal).\n");
 228                         mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
 229                                              sizeof(struct t3_cqe));
 230                         resplen = sizeof(struct iwch_create_cq_resp_v0);
 231                 } else {
 232                         mm->len = PAGE_ALIGN(((1UL << uresp.size_log2) + 1) *
 233                                              sizeof(struct t3_cqe));
 234                         uresp.memsize = mm->len;
 235                         resplen = sizeof uresp;
 236                 }
 237                 if (ib_copy_to_udata(udata, &uresp, resplen)) {
 238                         kfree(mm);
 239                         iwch_destroy_cq(&chp->ibcq);
 240                         return ERR_PTR(-EFAULT);
 241                 }
 242                 insert_mmap(ucontext, mm);
 243         }
 244         PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
 245              chp->cq.cqid, chp, (1 << chp->cq.size_log2),
 246              (unsigned long long) chp->cq.dma_addr);
 247         return &chp->ibcq;
 248 }
 249
 250 static int iwch_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
 251 {
 252 #ifdef notyet
 253         struct iwch_cq *chp = to_iwch_cq(cq);
 254         struct t3_cq oldcq, newcq;
 255         int ret;
 256
 257         PDBG("%s ib_cq %p cqe %d\n", __func__, cq, cqe);
 258
 259         /* We don't downsize... */
 260         if (cqe <= cq->cqe)
 261                 return 0;
 262
 263         /* create new t3_cq with new size */
 264         cqe = roundup_pow_of_two(cqe+1);
 265         newcq.size_log2 = ilog2(cqe);
 266
 267         /* Dont allow resize to less than the current wce count */
 268         if (cqe < Q_COUNT(chp->cq.rptr, chp->cq.wptr)) {
 269                 return -ENOMEM;
 270         }
 271
 272         /* Quiesce all QPs using this CQ */
 273         ret = iwch_quiesce_qps(chp);
 274         if (ret) {
 275                 return ret;
 276         }
 277
 278         ret = cxio_create_cq(&chp->rhp->rdev, &newcq);
 279         if (ret) {
 280                 return ret;
 281         }
 282
 283         /* copy CQEs */
 284         memcpy(newcq.queue, chp->cq.queue, (1 << chp->cq.size_log2) *
 285                                         sizeof(struct t3_cqe));
 286
 287         /* old iwch_qp gets new t3_cq but keeps old cqid */
 288         oldcq = chp->cq;
 289         chp->cq = newcq;
 290         chp->cq.cqid = oldcq.cqid;
 291
 292         /* resize new t3_cq to update the HW context */
 293         ret = cxio_resize_cq(&chp->rhp->rdev, &chp->cq);
 294         if (ret) {
 295                 chp->cq = oldcq;
 296                 return ret;
 297         }
 298         chp->ibcq.cqe = (1<<chp->cq.size_log2) - 1;
 299
 300         /* destroy old t3_cq */
 301         oldcq.cqid = newcq.cqid;
 302         ret = cxio_destroy_cq(&chp->rhp->rdev, &oldcq);
 303         if (ret) {
 304                 printk(KERN_ERR MOD "%s - cxio_destroy_cq failed %d\n",
 305                         __func__, ret);
 306         }
 307
 308         /* add user hooks here */
 309
 310         /* resume qps */
 311         ret = iwch_resume_qps(chp);
 312         return ret;
 313 #else
 314         return -ENOSYS;
 315 #endif
 316 }
 317
 318 static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
 319 {
 320         struct iwch_dev *rhp;
 321         struct iwch_cq *chp;
 322         enum t3_cq_opcode cq_op;
 323         int err;
 324         unsigned long flag;
 325         u32 rptr;
 326
 327         chp = to_iwch_cq(ibcq);
 328         rhp = chp->rhp;
 329         if ((flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED)
 330                 cq_op = CQ_ARM_SE;
 331         else
 332                 cq_op = CQ_ARM_AN;
 333         if (chp->user_rptr_addr) {
 334                 if (get_user(rptr, chp->user_rptr_addr))
 335                         return -EFAULT;
 336                 spin_lock_irqsave(&chp->lock, flag);
 337                 chp->cq.rptr = rptr;
 338         } else
 339                 spin_lock_irqsave(&chp->lock, flag);
 340         PDBG("%s rptr 0x%x\n", __func__, chp->cq.rptr);
 341         err = cxio_hal_cq_op(&rhp->rdev, &chp->cq, cq_op, 0);
 342         spin_unlock_irqrestore(&chp->lock, flag);
 343         if (err < 0)
 344                 printk(KERN_ERR MOD "Error %d rearming CQID 0x%x\n", err,
 345                        chp->cq.cqid);
 346         if (err > 0 && !(flags & IB_CQ_REPORT_MISSED_EVENTS))
 347                 err = 0;
 348         return err;
 349 }
 350
 351 static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 352 {
 353         int len = vma->vm_end - vma->vm_start;
 354         u32 key = vma->vm_pgoff << PAGE_SHIFT;
 355         struct cxio_rdev *rdev_p;
 356         int ret = 0;
 357         struct iwch_mm_entry *mm;
 358         struct iwch_ucontext *ucontext;
 359         u64 addr;
 360
 361         PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __func__, vma->vm_pgoff,
 362              key, len);
 363
 364         if (vma->vm_start & (PAGE_SIZE-1)) {
 365                 return -EINVAL;
 366         }
 367
 368         rdev_p = &(to_iwch_dev(context->device)->rdev);
 369         ucontext = to_iwch_ucontext(context);
 370
 371         mm = remove_mmap(ucontext, key, len);
 372         if (!mm)
 373                 return -EINVAL;
 374         addr = mm->addr;
 375         kfree(mm);
 376
 377         if ((addr >= rdev_p->rnic_info.udbell_physbase) &&
 378             (addr < (rdev_p->rnic_info.udbell_physbase +
 379                        rdev_p->rnic_info.udbell_len))) {
 380
 381                 /*
 382                  * Map T3 DB register.
 383                  */
 384                 if (vma->vm_flags & VM_READ) {
 385                         return -EPERM;
 386                 }
 387
 388                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 389                 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
 390                 vma->vm_flags &= ~VM_MAYREAD;
 391                 ret = io_remap_pfn_range(vma, vma->vm_start,
 392                                          addr >> PAGE_SHIFT,
 393                                          len, vma->vm_page_prot);
 394         } else {
 395
 396                 /*
 397                  * Map WQ or CQ contig dma memory...
 398                  */
 399                 ret = remap_pfn_range(vma, vma->vm_start,
 400                                       addr >> PAGE_SHIFT,
 401                                       len, vma->vm_page_prot);
 402         }
 403
 404         return ret;
 405 }
 406
 407 static int iwch_deallocate_pd(struct ib_pd *pd)
 408 {
 409         struct iwch_dev *rhp;
 410         struct iwch_pd *php;
 411
 412         php = to_iwch_pd(pd);
 413         rhp = php->rhp;
 414         PDBG("%s ibpd %p pdid 0x%x\n", __func__, pd, php->pdid);
 415         cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid);
 416         kfree(php);
 417         return 0;
 418 }
 419
 420 static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev,
 421                                struct ib_ucontext *context,
 422                                struct ib_udata *udata)
 423 {
 424         struct iwch_pd *php;
 425         u32 pdid;
 426         struct iwch_dev *rhp;
 427
 428         PDBG("%s ibdev %p\n", __func__, ibdev);
 429         rhp = (struct iwch_dev *) ibdev;
 430         pdid = cxio_hal_get_pdid(rhp->rdev.rscp);
 431         if (!pdid)
 432                 return ERR_PTR(-EINVAL);
 433         php = kzalloc(sizeof(*php), GFP_KERNEL);
 434         if (!php) {
 435                 cxio_hal_put_pdid(rhp->rdev.rscp, pdid);
 436                 return ERR_PTR(-ENOMEM);
 437         }
 438         php->pdid = pdid;
 439         php->rhp = rhp;
 440         if (context) {
 441                 if (ib_copy_to_udata(udata, &php->pdid, sizeof (__u32))) {
 442                         iwch_deallocate_pd(&php->ibpd);
 443                         return ERR_PTR(-EFAULT);
 444                 }
 445         }
 446         PDBG("%s pdid 0x%0x ptr 0x%p\n", __func__, pdid, php);
 447         return &php->ibpd;
 448 }
 449
 450 static int iwch_dereg_mr(struct ib_mr *ib_mr)
 451 {
 452         struct iwch_dev *rhp;
 453         struct iwch_mr *mhp;
 454         u32 mmid;
 455
 456         PDBG("%s ib_mr %p\n", __func__, ib_mr);
 457         /* There can be no memory windows */
 458         if (atomic_read(&ib_mr->usecnt))
 459                 return -EINVAL;
 460
 461         mhp = to_iwch_mr(ib_mr);
 462         rhp = mhp->rhp;
 463         mmid = mhp->attr.stag >> 8;
 464         cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
 465                        mhp->attr.pbl_addr);
 466         iwch_free_pbl(mhp);
 467         remove_handle(rhp, &rhp->mmidr, mmid);
 468         if (mhp->kva)
 469                 kfree((void *) (unsigned long) mhp->kva);
 470         if (mhp->umem)
 471                 ib_umem_release(mhp->umem);
 472         PDBG("%s mmid 0x%x ptr %p\n", __func__, mmid, mhp);
 473         kfree(mhp);
 474         return 0;
 475 }
 476
 477 static struct ib_mr *iwch_register_phys_mem(struct ib_pd *pd,
 478                                         struct ib_phys_buf *buffer_list,
 479                                         int num_phys_buf,
 480                                         int acc,
 481                                         u64 *iova_start)
 482 {
 483         __be64 *page_list;
 484         int shift;
 485         u64 total_size;
 486         int npages;
 487         struct iwch_dev *rhp;
 488         struct iwch_pd *php;
 489         struct iwch_mr *mhp;
 490         int ret;
 491
 492         PDBG("%s ib_pd %p\n", __func__, pd);
 493         php = to_iwch_pd(pd);
 494         rhp = php->rhp;
 495
 496         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 497         if (!mhp)
 498                 return ERR_PTR(-ENOMEM);
 499
 500         mhp->rhp = rhp;
 501
 502         /* First check that we have enough alignment */
 503         if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
 504                 ret = -EINVAL;
 505                 goto err;
 506         }
 507
 508         if (num_phys_buf > 1 &&
 509             ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
 510                 ret = -EINVAL;
 511                 goto err;
 512         }
 513
 514         ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
 515                                    &total_size, &npages, &shift, &page_list);
 516         if (ret)
 517                 goto err;
 518
 519         ret = iwch_alloc_pbl(mhp, npages);
 520         if (ret) {
 521                 kfree(page_list);
 522                 goto err_pbl;
 523         }
 524
 525         ret = iwch_write_pbl(mhp, page_list, npages, 0);
 526         kfree(page_list);
 527         if (ret)
 528                 goto err_pbl;
 529
 530         mhp->attr.pdid = php->pdid;
 531         mhp->attr.zbva = 0;
 532
 533         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 534         mhp->attr.va_fbo = *iova_start;
 535         mhp->attr.page_size = shift - 12;
 536
 537         mhp->attr.len = (u32) total_size;
 538         mhp->attr.pbl_size = npages;
 539         ret = iwch_register_mem(rhp, php, mhp, shift);
 540         if (ret)
 541                 goto err_pbl;
 542
 543         return &mhp->ibmr;
 544
 545 err_pbl:
 546         iwch_free_pbl(mhp);
 547
 548 err:
 549         kfree(mhp);
 550         return ERR_PTR(ret);
 551
 552 }
 553
 554 static int iwch_reregister_phys_mem(struct ib_mr *mr,
 555                                      int mr_rereg_mask,
 556                                      struct ib_pd *pd,
 557                                      struct ib_phys_buf *buffer_list,
 558                                      int num_phys_buf,
 559                                      int acc, u64 * iova_start)
 560 {
 561
 562         struct iwch_mr mh, *mhp;
 563         struct iwch_pd *php;
 564         struct iwch_dev *rhp;
 565         __be64 *page_list = NULL;
 566         int shift = 0;
 567         u64 total_size;
 568         int npages;
 569         int ret;
 570
 571         PDBG("%s ib_mr %p ib_pd %p\n", __func__, mr, pd);
 572
 573         /* There can be no memory windows */
 574         if (atomic_read(&mr->usecnt))
 575                 return -EINVAL;
 576
 577         mhp = to_iwch_mr(mr);
 578         rhp = mhp->rhp;
 579         php = to_iwch_pd(mr->pd);
 580
 581         /* make sure we are on the same adapter */
 582         if (rhp != php->rhp)
 583                 return -EINVAL;
 584
 585         memcpy(&mh, mhp, sizeof *mhp);
 586
 587         if (mr_rereg_mask & IB_MR_REREG_PD)
 588                 php = to_iwch_pd(pd);
 589         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 590                 mh.attr.perms = iwch_ib_to_tpt_access(acc);
 591         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 592                 ret = build_phys_page_list(buffer_list, num_phys_buf,
 593                                            iova_start,
 594                                            &total_size, &npages,
 595                                            &shift, &page_list);
 596                 if (ret)
 597                         return ret;
 598         }
 599
 600         ret = iwch_reregister_mem(rhp, php, &mh, shift, npages);
 601         kfree(page_list);
 602         if (ret) {
 603                 return ret;
 604         }
 605         if (mr_rereg_mask & IB_MR_REREG_PD)
 606                 mhp->attr.pdid = php->pdid;
 607         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 608                 mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 609         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 610                 mhp->attr.zbva = 0;
 611                 mhp->attr.va_fbo = *iova_start;
 612                 mhp->attr.page_size = shift - 12;
 613                 mhp->attr.len = (u32) total_size;
 614                 mhp->attr.pbl_size = npages;
 615         }
 616
 617         return 0;
 618 }
 619
 620
 621 static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 622                                       u64 virt, int acc, struct ib_udata *udata)
 623 {
 624         __be64 *pages;
 625         int shift, n, len;
 626         int i, j, k;
 627         int err = 0;
 628         struct ib_umem_chunk *chunk;
 629         struct iwch_dev *rhp;
 630         struct iwch_pd *php;
 631         struct iwch_mr *mhp;
 632         struct iwch_reg_user_mr_resp uresp;
 633
 634         PDBG("%s ib_pd %p\n", __func__, pd);
 635
 636         php = to_iwch_pd(pd);
 637         rhp = php->rhp;
 638         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 639         if (!mhp)
 640                 return ERR_PTR(-ENOMEM);
 641
 642         mhp->rhp = rhp;
 643
 644         mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
 645         if (IS_ERR(mhp->umem)) {
 646                 err = PTR_ERR(mhp->umem);
 647                 kfree(mhp);
 648                 return ERR_PTR(err);
 649         }
 650
 651         shift = ffs(mhp->umem->page_size) - 1;
 652
 653         n = 0;
 654         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 655                 n += chunk->nents;
 656
 657         err = iwch_alloc_pbl(mhp, n);
 658         if (err)
 659                 goto err;
 660
 661         pages = (__be64 *) __get_free_page(GFP_KERNEL);
 662         if (!pages) {
 663                 err = -ENOMEM;
 664                 goto err_pbl;
 665         }
 666
 667         i = n = 0;
 668
 669         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 670                 for (j = 0; j < chunk->nmap; ++j) {
 671                         len = sg_dma_len(&chunk->page_list[j]) >> shift;
 672                         for (k = 0; k < len; ++k) {
 673                                 pages[i++] = cpu_to_be64(sg_dma_address(
 674                                         &chunk->page_list[j]) +
 675                                         mhp->umem->page_size * k);
 676                                 if (i == PAGE_SIZE / sizeof *pages) {
 677                                         err = iwch_write_pbl(mhp, pages, i, n);
 678                                         if (err)
 679                                                 goto pbl_done;
 680                                         n += i;
 681                                         i = 0;
 682                                 }
 683                         }
 684                 }
 685
 686         if (i)
 687                 err = iwch_write_pbl(mhp, pages, i, n);
 688
 689 pbl_done:
 690         free_page((unsigned long) pages);
 691         if (err)
 692                 goto err_pbl;
 693
 694         mhp->attr.pdid = php->pdid;
 695         mhp->attr.zbva = 0;
 696         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 697         mhp->attr.va_fbo = virt;
 698         mhp->attr.page_size = shift - 12;
 699         mhp->attr.len = (u32) length;
 700
 701         err = iwch_register_mem(rhp, php, mhp, shift);
 702         if (err)
 703                 goto err_pbl;
 704
 705         if (udata && !t3a_device(rhp)) {
 706                 uresp.pbl_addr = (mhp->attr.pbl_addr -
 707                                  rhp->rdev.rnic_info.pbl_base) >> 3;
 708                 PDBG("%s user resp pbl_addr 0x%x\n", __func__,
 709                      uresp.pbl_addr);
 710
 711                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 712                         iwch_dereg_mr(&mhp->ibmr);
 713                         err = -EFAULT;
 714                         goto err;
 715                 }
 716         }
 717
 718         return &mhp->ibmr;
 719
 720 err_pbl:
 721         iwch_free_pbl(mhp);
 722
 723 err:
 724         ib_umem_release(mhp->umem);
 725         kfree(mhp);
 726         return ERR_PTR(err);
 727 }
 728
 729 static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
 730 {
 731         struct ib_phys_buf bl;
 732         u64 kva;
 733         struct ib_mr *ibmr;
 734
 735         PDBG("%s ib_pd %p\n", __func__, pd);
 736
 737         /*
 738          * T3 only supports 32 bits of size.
 739          */
 740         bl.size = 0xffffffff;
 741         bl.addr = 0;
 742         kva = 0;
 743         ibmr = iwch_register_phys_mem(pd, &bl, 1, acc, &kva);
 744         return ibmr;
 745 }
 746
 747 static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
 748 {
 749         struct iwch_dev *rhp;
 750         struct iwch_pd *php;
 751         struct iwch_mw *mhp;
 752         u32 mmid;
 753         u32 stag = 0;
 754         int ret;
 755
 756         php = to_iwch_pd(pd);
 757         rhp = php->rhp;
 758         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 759         if (!mhp)
 760                 return ERR_PTR(-ENOMEM);
 761         ret = cxio_allocate_window(&rhp->rdev, &stag, php->pdid);
 762         if (ret) {
 763                 kfree(mhp);
 764                 return ERR_PTR(ret);
 765         }
 766         mhp->rhp = rhp;
 767         mhp->attr.pdid = php->pdid;
 768         mhp->attr.type = TPT_MW;
 769         mhp->attr.stag = stag;
 770         mmid = (stag) >> 8;
 771         mhp->ibmw.rkey = stag;
 772         if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) {
 773                 cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 774                 kfree(mhp);
 775                 return ERR_PTR(-ENOMEM);
 776         }
 777         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 778         return &(mhp->ibmw);
 779 }
 780
 781 static int iwch_dealloc_mw(struct ib_mw *mw)
 782 {
 783         struct iwch_dev *rhp;
 784         struct iwch_mw *mhp;
 785         u32 mmid;
 786
 787         mhp = to_iwch_mw(mw);
 788         rhp = mhp->rhp;
 789         mmid = (mw->rkey) >> 8;
 790         cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 791         remove_handle(rhp, &rhp->mmidr, mmid);
 792         kfree(mhp);
 793         PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
 794         return 0;
 795 }
 796
 797 static struct ib_mr *iwch_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth)
 798 {
 799         struct iwch_dev *rhp;
 800         struct iwch_pd *php;
 801         struct iwch_mr *mhp;
 802         u32 mmid;
 803         u32 stag = 0;
 804         int ret = 0;
 805
 806         php = to_iwch_pd(pd);
 807         rhp = php->rhp;
 808         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 809         if (!mhp)
 810                 goto err;
 811
 812         mhp->rhp = rhp;
 813         ret = iwch_alloc_pbl(mhp, pbl_depth);
 814         if (ret)
 815                 goto err1;
 816         mhp->attr.pbl_size = pbl_depth;
 817         ret = cxio_allocate_stag(&rhp->rdev, &stag, php->pdid,
 818                                  mhp->attr.pbl_size, mhp->attr.pbl_addr);
 819         if (ret)
 820                 goto err2;
 821         mhp->attr.pdid = php->pdid;
 822         mhp->attr.type = TPT_NON_SHARED_MR;
 823         mhp->attr.stag = stag;
 824         mhp->attr.state = 1;
 825         mmid = (stag) >> 8;
 826         mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
 827         if (insert_handle(rhp, &rhp->mmidr, mhp, mmid))
 828                 goto err3;
 829
 830         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 831         return &(mhp->ibmr);
 832 err3:
 833         cxio_dereg_mem(&rhp->rdev, stag, mhp->attr.pbl_size,
 834                        mhp->attr.pbl_addr);
 835 err2:
 836         iwch_free_pbl(mhp);
 837 err1:
 838         kfree(mhp);
 839 err:
 840         return ERR_PTR(ret);
 841 }
 842
 843 static struct ib_fast_reg_page_list *iwch_alloc_fastreg_pbl(
 844                                         struct ib_device *device,
 845                                         int page_list_len)
 846 {
 847         struct ib_fast_reg_page_list *page_list;
 848
 849         page_list = kmalloc(sizeof *page_list + page_list_len * sizeof(u64),
 850                             GFP_KERNEL);
 851         if (!page_list)
 852                 return ERR_PTR(-ENOMEM);
 853
 854         page_list->page_list = (u64 *)(page_list + 1);
 855         page_list->max_page_list_len = page_list_len;
 856
 857         return page_list;
 858 }
 859
 860 static void iwch_free_fastreg_pbl(struct ib_fast_reg_page_list *page_list)
 861 {
 862         kfree(page_list);
 863 }
 864
 865 static int iwch_destroy_qp(struct ib_qp *ib_qp)
 866 {
 867         struct iwch_dev *rhp;
 868         struct iwch_qp *qhp;
 869         struct iwch_qp_attributes attrs;
 870         struct iwch_ucontext *ucontext;
 871
 872         qhp = to_iwch_qp(ib_qp);
 873         rhp = qhp->rhp;
 874
 875         attrs.next_state = IWCH_QP_STATE_ERROR;
 876         iwch_modify_qp(rhp, qhp, IWCH_QP_ATTR_NEXT_STATE, &attrs, 0);
 877         wait_event(qhp->wait, !qhp->ep);
 878
 879         remove_handle(rhp, &rhp->qpidr, qhp->wq.qpid);
 880
 881         atomic_dec(&qhp->refcnt);
 882         wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
 883
 884         ucontext = ib_qp->uobject ? to_iwch_ucontext(ib_qp->uobject->context)
 885                                   : NULL;
 886         cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 887                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 888
 889         PDBG("%s ib_qp %p qpid 0x%0x qhp %p\n", __func__,
 890              ib_qp, qhp->wq.qpid, qhp);
 891         kfree(qhp);
 892         return 0;
 893 }
 894
 895 static struct ib_qp *iwch_create_qp(struct ib_pd *pd,
 896                              struct ib_qp_init_attr *attrs,
 897                              struct ib_udata *udata)
 898 {
 899         struct iwch_dev *rhp;
 900         struct iwch_qp *qhp;
 901         struct iwch_pd *php;
 902         struct iwch_cq *schp;
 903         struct iwch_cq *rchp;
 904         struct iwch_create_qp_resp uresp;
 905         int wqsize, sqsize, rqsize;
 906         struct iwch_ucontext *ucontext;
 907
 908         PDBG("%s ib_pd %p\n", __func__, pd);
 909         if (attrs->qp_type != IB_QPT_RC)
 910                 return ERR_PTR(-EINVAL);
 911         php = to_iwch_pd(pd);
 912         rhp = php->rhp;
 913         schp = get_chp(rhp, ((struct iwch_cq *) attrs->send_cq)->cq.cqid);
 914         rchp = get_chp(rhp, ((struct iwch_cq *) attrs->recv_cq)->cq.cqid);
 915         if (!schp || !rchp)
 916                 return ERR_PTR(-EINVAL);
 917
 918         /* The RQT size must be # of entries + 1 rounded up to a power of two */
 919         rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr);
 920         if (rqsize == attrs->cap.max_recv_wr)
 921                 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr+1);
 922
 923         /* T3 doesn't support RQT depth < 16 */
 924         if (rqsize < 16)
 925                 rqsize = 16;
 926
 927         if (rqsize > T3_MAX_RQ_SIZE)
 928                 return ERR_PTR(-EINVAL);
 929
 930         if (attrs->cap.max_inline_data > T3_MAX_INLINE)
 931                 return ERR_PTR(-EINVAL);
 932
 933         /*
 934          * NOTE: The SQ and total WQ sizes don't need to be
 935          * a power of two.  However, all the code assumes
 936          * they are. EG: Q_FREECNT() and friends.
 937          */
 938         sqsize = roundup_pow_of_two(attrs->cap.max_send_wr);
 939         wqsize = roundup_pow_of_two(rqsize + sqsize);
 940
 941         /*
 942          * Kernel users need more wq space for fastreg WRs which can take
 943          * 2 WR fragments.
 944          */
 945         ucontext = pd->uobject ? to_iwch_ucontext(pd->uobject->context) : NULL;
 946         if (!ucontext && wqsize < (rqsize + (2 * sqsize)))
 947                 wqsize = roundup_pow_of_two(rqsize +
 948                                 roundup_pow_of_two(attrs->cap.max_send_wr * 2));
 949         PDBG("%s wqsize %d sqsize %d rqsize %d\n", __func__,
 950              wqsize, sqsize, rqsize);
 951         qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
 952         if (!qhp)
 953                 return ERR_PTR(-ENOMEM);
 954         qhp->wq.size_log2 = ilog2(wqsize);
 955         qhp->wq.rq_size_log2 = ilog2(rqsize);
 956         qhp->wq.sq_size_log2 = ilog2(sqsize);
 957         if (cxio_create_qp(&rhp->rdev, !udata, &qhp->wq,
 958                            ucontext ? &ucontext->uctx : &rhp->rdev.uctx)) {
 959                 kfree(qhp);
 960                 return ERR_PTR(-ENOMEM);
 961         }
 962
 963         attrs->cap.max_recv_wr = rqsize - 1;
 964         attrs->cap.max_send_wr = sqsize;
 965         attrs->cap.max_inline_data = T3_MAX_INLINE;
 966
 967         qhp->rhp = rhp;
 968         qhp->attr.pd = php->pdid;
 969         qhp->attr.scq = ((struct iwch_cq *) attrs->send_cq)->cq.cqid;
 970         qhp->attr.rcq = ((struct iwch_cq *) attrs->recv_cq)->cq.cqid;
 971         qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
 972         qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
 973         qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
 974         qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
 975         qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
 976         qhp->attr.state = IWCH_QP_STATE_IDLE;
 977         qhp->attr.next_state = IWCH_QP_STATE_IDLE;
 978
 979         /*
 980          * XXX - These don't get passed in from the openib user
 981          * at create time.  The CM sets them via a QP modify.
 982          * Need to fix...  I think the CM should
 983          */
 984         qhp->attr.enable_rdma_read = 1;
 985         qhp->attr.enable_rdma_write = 1;
 986         qhp->attr.enable_bind = 1;
 987         qhp->attr.max_ord = 1;
 988         qhp->attr.max_ird = 1;
 989
 990         spin_lock_init(&qhp->lock);
 991         init_waitqueue_head(&qhp->wait);
 992         atomic_set(&qhp->refcnt, 1);
 993
 994         if (insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid)) {
 995                 cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 996                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 997                 kfree(qhp);
 998                 return ERR_PTR(-ENOMEM);
 999         }
1000
1001         if (udata) {
1002
1003                 struct iwch_mm_entry *mm1, *mm2;
1004
1005                 mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
1006                 if (!mm1) {
1007                         iwch_destroy_qp(&qhp->ibqp);
1008                         return ERR_PTR(-ENOMEM);
1009                 }
1010
1011                 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
1012                 if (!mm2) {
1013                         kfree(mm1);
1014                         iwch_destroy_qp(&qhp->ibqp);
1015                         return ERR_PTR(-ENOMEM);
1016                 }
1017
1018                 uresp.qpid = qhp->wq.qpid;
1019                 uresp.size_log2 = qhp->wq.size_log2;
1020                 uresp.sq_size_log2 = qhp->wq.sq_size_log2;
1021                 uresp.rq_size_log2 = qhp->wq.rq_size_log2;
1022                 spin_lock(&ucontext->mmap_lock);
1023                 uresp.key = ucontext->key;
1024                 ucontext->key += PAGE_SIZE;
1025                 uresp.db_key = ucontext->key;
1026                 ucontext->key += PAGE_SIZE;
1027                 spin_unlock(&ucontext->mmap_lock);
1028                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
1029                         kfree(mm1);
1030                         kfree(mm2);
1031                         iwch_destroy_qp(&qhp->ibqp);
1032                         return ERR_PTR(-EFAULT);
1033                 }
1034                 mm1->key = uresp.key;
1035                 mm1->addr = virt_to_phys(qhp->wq.queue);
1036                 mm1->len = PAGE_ALIGN(wqsize * sizeof (union t3_wr));
1037                 insert_mmap(ucontext, mm1);
1038                 mm2->key = uresp.db_key;
1039                 mm2->addr = qhp->wq.udb & PAGE_MASK;
1040                 mm2->len = PAGE_SIZE;
1041                 insert_mmap(ucontext, mm2);
1042         }
1043         qhp->ibqp.qp_num = qhp->wq.qpid;
1044         init_timer(&(qhp->timer));
1045         PDBG("%s sq_num_entries %d, rq_num_entries %d "
1046              "qpid 0x%0x qhp %p dma_addr 0x%llx size %d rq_addr 0x%x\n",
1047              __func__, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
1048              qhp->wq.qpid, qhp, (unsigned long long) qhp->wq.dma_addr,
1049              1 << qhp->wq.size_log2, qhp->wq.rq_addr);
1050         return &qhp->ibqp;
1051 }
1052
1053 static int iwch_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1054                       int attr_mask, struct ib_udata *udata)
1055 {
1056         struct iwch_dev *rhp;
1057         struct iwch_qp *qhp;
1058         enum iwch_qp_attr_mask mask = 0;
1059         struct iwch_qp_attributes attrs;
1060
1061         PDBG("%s ib_qp %p\n", __func__, ibqp);
1062
1063         /* iwarp does not support the RTR state */
1064         if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
1065                 attr_mask &= ~IB_QP_STATE;
1066
1067         /* Make sure we still have something left to do */
1068         if (!attr_mask)
1069                 return 0;
1070
1071         memset(&attrs, 0, sizeof attrs);
1072         qhp = to_iwch_qp(ibqp);
1073         rhp = qhp->rhp;
1074
1075         attrs.next_state = iwch_convert_state(attr->qp_state);
1076         attrs.enable_rdma_read = (attr->qp_access_flags &
1077                                IB_ACCESS_REMOTE_READ) ?  1 : 0;
1078         attrs.enable_rdma_write = (attr->qp_access_flags &
1079                                 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
1080         attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
1081
1082
1083         mask |= (attr_mask & IB_QP_STATE) ? IWCH_QP_ATTR_NEXT_STATE : 0;
1084         mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
1085                         (IWCH_QP_ATTR_ENABLE_RDMA_READ |
1086                          IWCH_QP_ATTR_ENABLE_RDMA_WRITE |
1087                          IWCH_QP_ATTR_ENABLE_RDMA_BIND) : 0;
1088
1089         return iwch_modify_qp(rhp, qhp, mask, &attrs, 0);
1090 }
1091
1092 void iwch_qp_add_ref(struct ib_qp *qp)
1093 {
1094         PDBG("%s ib_qp %p\n", __func__, qp);
1095         atomic_inc(&(to_iwch_qp(qp)->refcnt));
1096 }
1097
1098 void iwch_qp_rem_ref(struct ib_qp *qp)
1099 {
1100         PDBG("%s ib_qp %p\n", __func__, qp);
1101         if (atomic_dec_and_test(&(to_iwch_qp(qp)->refcnt)))
1102                 wake_up(&(to_iwch_qp(qp)->wait));
1103 }
1104
1105 static struct ib_qp *iwch_get_qp(struct ib_device *dev, int qpn)
1106 {
1107         PDBG("%s ib_dev %p qpn 0x%x\n", __func__, dev, qpn);
1108         return (struct ib_qp *)get_qhp(to_iwch_dev(dev), qpn);
1109 }
1110
1111
1112 static int iwch_query_pkey(struct ib_device *ibdev,
1113                            u8 port, u16 index, u16 * pkey)
1114 {
1115         PDBG("%s ibdev %p\n", __func__, ibdev);
1116         *pkey = 0;
1117         return 0;
1118 }
1119
1120 static int iwch_query_gid(struct ib_device *ibdev, u8 port,
1121                           int index, union ib_gid *gid)
1122 {
1123         struct iwch_dev *dev;
1124
1125         PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
1126                __func__, ibdev, port, index, gid);
1127         dev = to_iwch_dev(ibdev);
1128         BUG_ON(port == 0 || port > 2);
1129         memset(&(gid->raw[0]), 0, sizeof(gid->raw));
1130         memcpy(&(gid->raw[0]), dev->rdev.port_info.lldevs[port-1]->dev_addr, 6);
1131         return 0;
1132 }
1133
1134 static u64 fw_vers_string_to_u64(struct iwch_dev *iwch_dev)
1135 {
1136         struct ethtool_drvinfo info;
1137         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1138         char *cp, *next;
1139         unsigned fw_maj, fw_min, fw_mic;
1140
1141         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1142
1143         next = info.fw_version + 1;
1144         cp = strsep(&next, ".");
1145         sscanf(cp, "%i", &fw_maj);
1146         cp = strsep(&next, ".");
1147         sscanf(cp, "%i", &fw_min);
1148         cp = strsep(&next, ".");
1149         sscanf(cp, "%i", &fw_mic);
1150
1151         return (((u64)fw_maj & 0xffff) << 32) | ((fw_min & 0xffff) << 16) |
1152                (fw_mic & 0xffff);
1153 }
1154
1155 static int iwch_query_device(struct ib_device *ibdev,
1156                              struct ib_device_attr *props)
1157 {
1158
1159         struct iwch_dev *dev;
1160         PDBG("%s ibdev %p\n", __func__, ibdev);
1161
1162         dev = to_iwch_dev(ibdev);
1163         memset(props, 0, sizeof *props);
1164         memcpy(&props->sys_image_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1165         props->hw_ver = dev->rdev.t3cdev_p->type;
1166         props->fw_ver = fw_vers_string_to_u64(dev);
1167         props->device_cap_flags = dev->device_cap_flags;
1168         props->page_size_cap = dev->attr.mem_pgsizes_bitmask;
1169         props->vendor_id = (u32)dev->rdev.rnic_info.pdev->vendor;
1170         props->vendor_part_id = (u32)dev->rdev.rnic_info.pdev->device;
1171         props->max_mr_size = dev->attr.max_mr_size;
1172         props->max_qp = dev->attr.max_qps;
1173         props->max_qp_wr = dev->attr.max_wrs;
1174         props->max_sge = dev->attr.max_sge_per_wr;
1175         props->max_sge_rd = 1;
1176         props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
1177         props->max_qp_init_rd_atom = dev->attr.max_rdma_reads_per_qp;
1178         props->max_cq = dev->attr.max_cqs;
1179         props->max_cqe = dev->attr.max_cqes_per_cq;
1180         props->max_mr = dev->attr.max_mem_regs;
1181         props->max_pd = dev->attr.max_pds;
1182         props->local_ca_ack_delay = 0;
1183         props->max_fast_reg_page_list_len = T3_MAX_FASTREG_DEPTH;
1184
1185         return 0;
1186 }
1187
1188 static int iwch_query_port(struct ib_device *ibdev,
1189                            u8 port, struct ib_port_attr *props)
1190 {
1191         struct iwch_dev *dev;
1192         struct net_device *netdev;
1193         struct in_device *inetdev;
1194
1195         PDBG("%s ibdev %p\n", __func__, ibdev);
1196
1197         dev = to_iwch_dev(ibdev);
1198         netdev = dev->rdev.port_info.lldevs[port-1];
1199
1200         memset(props, 0, sizeof(struct ib_port_attr));
1201         props->max_mtu = IB_MTU_4096;
1202         if (netdev->mtu >= 4096)
1203                 props->active_mtu = IB_MTU_4096;
1204         else if (netdev->mtu >= 2048)
1205                 props->active_mtu = IB_MTU_2048;
1206         else if (netdev->mtu >= 1024)
1207                 props->active_mtu = IB_MTU_1024;
1208         else if (netdev->mtu >= 512)
1209                 props->active_mtu = IB_MTU_512;
1210         else
1211                 props->active_mtu = IB_MTU_256;
1212
1213         if (!netif_carrier_ok(netdev))
1214                 props->state = IB_PORT_DOWN;
1215         else {
1216                 inetdev = in_dev_get(netdev);
1217                 if (inetdev) {
1218                         if (inetdev->ifa_list)
1219                                 props->state = IB_PORT_ACTIVE;
1220                         else
1221                                 props->state = IB_PORT_INIT;
1222                         in_dev_put(inetdev);
1223                 } else
1224                         props->state = IB_PORT_INIT;
1225         }
1226
1227         props->port_cap_flags =
1228             IB_PORT_CM_SUP |
1229             IB_PORT_SNMP_TUNNEL_SUP |
1230             IB_PORT_REINIT_SUP |
1231             IB_PORT_DEVICE_MGMT_SUP |
1232             IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
1233         props->gid_tbl_len = 1;
1234         props->pkey_tbl_len = 1;
1235         props->active_width = 2;
1236         props->active_speed = 2;
1237         props->max_msg_sz = -1;
1238
1239         return 0;
1240 }
1241
1242 static ssize_t show_rev(struct device *dev, struct device_attribute *attr,
1243                         char *buf)
1244 {
1245         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1246                                                  ibdev.dev);
1247         PDBG("%s dev 0x%p\n", __func__, dev);
1248         return sprintf(buf, "%d\n", iwch_dev->rdev.t3cdev_p->type);
1249 }
1250
1251 static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, char *buf)
1252 {
1253         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1254                                                  ibdev.dev);
1255         struct ethtool_drvinfo info;
1256         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1257
1258         PDBG("%s dev 0x%p\n", __func__, dev);
1259         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1260         return sprintf(buf, "%s\n", info.fw_version);
1261 }
1262
1263 static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
1264                         char *buf)
1265 {
1266         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1267                                                  ibdev.dev);
1268         struct ethtool_drvinfo info;
1269         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1270
1271         PDBG("%s dev 0x%p\n", __func__, dev);
1272         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1273         return sprintf(buf, "%s\n", info.driver);
1274 }
1275
1276 static ssize_t show_board(struct device *dev, struct device_attribute *attr,
1277                           char *buf)
1278 {
1279         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1280                                                  ibdev.dev);
1281         PDBG("%s dev 0x%p\n", __func__, dev);
1282         return sprintf(buf, "%x.%x\n", iwch_dev->rdev.rnic_info.pdev->vendor,
1283                        iwch_dev->rdev.rnic_info.pdev->device);
1284 }
1285
1286 static int iwch_get_mib(struct ib_device *ibdev,
1287                         union rdma_protocol_stats *stats)
1288 {
1289         struct iwch_dev *dev;
1290         struct tp_mib_stats m;
1291         int ret;
1292
1293         PDBG("%s ibdev %p\n", __func__, ibdev);
1294         dev = to_iwch_dev(ibdev);
1295         ret = dev->rdev.t3cdev_p->ctl(dev->rdev.t3cdev_p, RDMA_GET_MIB, &m);
1296         if (ret)
1297                 return -ENOSYS;
1298
1299         memset(stats, 0, sizeof *stats);
1300         stats->iw.ipInReceives = ((u64) m.ipInReceive_hi << 32) +
1301                                 m.ipInReceive_lo;
1302         stats->iw.ipInHdrErrors = ((u64) m.ipInHdrErrors_hi << 32) +
1303                                   m.ipInHdrErrors_lo;
1304         stats->iw.ipInAddrErrors = ((u64) m.ipInAddrErrors_hi << 32) +
1305                                    m.ipInAddrErrors_lo;
1306         stats->iw.ipInUnknownProtos = ((u64) m.ipInUnknownProtos_hi << 32) +
1307                                       m.ipInUnknownProtos_lo;
1308         stats->iw.ipInDiscards = ((u64) m.ipInDiscards_hi << 32) +
1309                                  m.ipInDiscards_lo;
1310         stats->iw.ipInDelivers = ((u64) m.ipInDelivers_hi << 32) +
1311                                  m.ipInDelivers_lo;
1312         stats->iw.ipOutRequests = ((u64) m.ipOutRequests_hi << 32) +
1313                                   m.ipOutRequests_lo;
1314         stats->iw.ipOutDiscards = ((u64) m.ipOutDiscards_hi << 32) +
1315                                   m.ipOutDiscards_lo;
1316         stats->iw.ipOutNoRoutes = ((u64) m.ipOutNoRoutes_hi << 32) +
1317                                   m.ipOutNoRoutes_lo;
1318         stats->iw.ipReasmTimeout = (u64) m.ipReasmTimeout;
1319         stats->iw.ipReasmReqds = (u64) m.ipReasmReqds;
1320         stats->iw.ipReasmOKs = (u64) m.ipReasmOKs;
1321         stats->iw.ipReasmFails = (u64) m.ipReasmFails;
1322         stats->iw.tcpActiveOpens = (u64) m.tcpActiveOpens;
1323         stats->iw.tcpPassiveOpens = (u64) m.tcpPassiveOpens;
1324         stats->iw.tcpAttemptFails = (u64) m.tcpAttemptFails;
1325         stats->iw.tcpEstabResets = (u64) m.tcpEstabResets;
1326         stats->iw.tcpOutRsts = (u64) m.tcpOutRsts;
1327         stats->iw.tcpCurrEstab = (u64) m.tcpCurrEstab;
1328         stats->iw.tcpInSegs = ((u64) m.tcpInSegs_hi << 32) +
1329                               m.tcpInSegs_lo;
1330         stats->iw.tcpOutSegs = ((u64) m.tcpOutSegs_hi << 32) +
1331                                m.tcpOutSegs_lo;
1332         stats->iw.tcpRetransSegs = ((u64) m.tcpRetransSeg_hi << 32) +
1333                                   m.tcpRetransSeg_lo;
1334         stats->iw.tcpInErrs = ((u64) m.tcpInErrs_hi << 32) +
1335                               m.tcpInErrs_lo;
1336         stats->iw.tcpRtoMin = (u64) m.tcpRtoMin;
1337         stats->iw.tcpRtoMax = (u64) m.tcpRtoMax;
1338         return 0;
1339 }
1340
1341 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1342 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1343 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1344 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1345
1346 static struct device_attribute *iwch_class_attributes[] = {
1347         &dev_attr_hw_rev,
1348         &dev_attr_fw_ver,
1349         &dev_attr_hca_type,
1350         &dev_attr_board_id,
1351 };
1352
1353 int iwch_register_device(struct iwch_dev *dev)
1354 {
1355         int ret;
1356         int i;
1357
1358         PDBG("%s iwch_dev %p\n", __func__, dev);
1359         strlcpy(dev->ibdev.name, "cxgb3_%d", IB_DEVICE_NAME_MAX);
1360         memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
1361         memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1362         dev->ibdev.owner = THIS_MODULE;
1363         dev->device_cap_flags = IB_DEVICE_LOCAL_DMA_LKEY |
1364                                 IB_DEVICE_MEM_WINDOW |
1365                                 IB_DEVICE_MEM_MGT_EXTENSIONS;
1366
1367         /* cxgb3 supports STag 0. */
1368         dev->ibdev.local_dma_lkey = 0;
1369
1370         dev->ibdev.uverbs_cmd_mask =
1371             (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1372             (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1373             (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1374             (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1375             (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1376             (1ull << IB_USER_VERBS_CMD_REG_MR) |
1377             (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1378             (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1379             (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1380             (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1381             (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1382             (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1383             (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1384             (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1385             (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1386             (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1387             (1ull << IB_USER_VERBS_CMD_POST_RECV);
1388         dev->ibdev.node_type = RDMA_NODE_RNIC;
1389         memcpy(dev->ibdev.node_desc, IWCH_NODE_DESC, sizeof(IWCH_NODE_DESC));
1390         dev->ibdev.phys_port_cnt = dev->rdev.port_info.nports;
1391         dev->ibdev.num_comp_vectors = 1;
1392         dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
1393         dev->ibdev.query_device = iwch_query_device;
1394         dev->ibdev.query_port = iwch_query_port;
1395         dev->ibdev.modify_port = iwch_modify_port;
1396         dev->ibdev.query_pkey = iwch_query_pkey;
1397         dev->ibdev.query_gid = iwch_query_gid;
1398         dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
1399         dev->ibdev.dealloc_ucontext = iwch_dealloc_ucontext;
1400         dev->ibdev.mmap = iwch_mmap;
1401         dev->ibdev.alloc_pd = iwch_allocate_pd;
1402         dev->ibdev.dealloc_pd = iwch_deallocate_pd;
1403         dev->ibdev.create_ah = iwch_ah_create;
1404         dev->ibdev.destroy_ah = iwch_ah_destroy;
1405         dev->ibdev.create_qp = iwch_create_qp;
1406         dev->ibdev.modify_qp = iwch_ib_modify_qp;
1407         dev->ibdev.destroy_qp = iwch_destroy_qp;
1408         dev->ibdev.create_cq = iwch_create_cq;
1409         dev->ibdev.destroy_cq = iwch_destroy_cq;
1410         dev->ibdev.resize_cq = iwch_resize_cq;
1411         dev->ibdev.poll_cq = iwch_poll_cq;
1412         dev->ibdev.get_dma_mr = iwch_get_dma_mr;
1413         dev->ibdev.reg_phys_mr = iwch_register_phys_mem;
1414         dev->ibdev.rereg_phys_mr = iwch_reregister_phys_mem;
1415         dev->ibdev.reg_user_mr = iwch_reg_user_mr;
1416         dev->ibdev.dereg_mr = iwch_dereg_mr;
1417         dev->ibdev.alloc_mw = iwch_alloc_mw;
1418         dev->ibdev.bind_mw = iwch_bind_mw;
1419         dev->ibdev.dealloc_mw = iwch_dealloc_mw;
1420         dev->ibdev.alloc_fast_reg_mr = iwch_alloc_fast_reg_mr;
1421         dev->ibdev.alloc_fast_reg_page_list = iwch_alloc_fastreg_pbl;
1422         dev->ibdev.free_fast_reg_page_list = iwch_free_fastreg_pbl;
1423         dev->ibdev.attach_mcast = iwch_multicast_attach;
1424         dev->ibdev.detach_mcast = iwch_multicast_detach;
1425         dev->ibdev.process_mad = iwch_process_mad;
1426         dev->ibdev.req_notify_cq = iwch_arm_cq;
1427         dev->ibdev.post_send = iwch_post_send;
1428         dev->ibdev.post_recv = iwch_post_receive;
1429         dev->ibdev.get_protocol_stats = iwch_get_mib;
1430         dev->ibdev.uverbs_abi_ver = IWCH_UVERBS_ABI_VERSION;
1431
1432         dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
1433         if (!dev->ibdev.iwcm)
1434                 return -ENOMEM;
1435
1436         dev->ibdev.iwcm->connect = iwch_connect;
1437         dev->ibdev.iwcm->accept = iwch_accept_cr;
1438         dev->ibdev.iwcm->reject = iwch_reject_cr;
1439         dev->ibdev.iwcm->create_listen = iwch_create_listen;
1440         dev->ibdev.iwcm->destroy_listen = iwch_destroy_listen;
1441         dev->ibdev.iwcm->add_ref = iwch_qp_add_ref;
1442         dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
1443         dev->ibdev.iwcm->get_qp = iwch_get_qp;
1444
1445         ret = ib_register_device(&dev->ibdev, NULL);
1446         if (ret)
1447                 goto bail1;
1448
1449         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i) {
1450                 ret = device_create_file(&dev->ibdev.dev,
1451                                          iwch_class_attributes[i]);
1452                 if (ret) {
1453                         goto bail2;
1454                 }
1455         }
1456         return 0;
1457 bail2:
1458         ib_unregister_device(&dev->ibdev);
1459 bail1:
1460         kfree(dev->ibdev.iwcm);
1461         return ret;
1462 }
1463
1464 void iwch_unregister_device(struct iwch_dev *dev)
1465 {
1466         int i;
1467
1468         PDBG("%s iwch_dev %p\n", __func__, dev);
1469         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i)
1470                 device_remove_file(&dev->ibdev.dev,
1471                                    iwch_class_attributes[i]);
1472         ib_unregister_device(&dev->ibdev);
1473         kfree(dev->ibdev.iwcm);
1474         return;
1475 }