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