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                         uresp.reserved = 0;
 230                         resplen = sizeof uresp;
 231                 }
 232                 if (ib_copy_to_udata(udata, &uresp, resplen)) {
 233                         kfree(mm);
 234                         iwch_destroy_cq(&chp->ibcq);
 235                         return ERR_PTR(-EFAULT);
 236                 }
 237                 insert_mmap(ucontext, mm);
 238         }
 239         PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
 240              chp->cq.cqid, chp, (1 << chp->cq.size_log2),
 241              (unsigned long long) chp->cq.dma_addr);
 242         return &chp->ibcq;
 243 }
 244
 245 static int iwch_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
 246 {
 247 #ifdef notyet
 248         struct iwch_cq *chp = to_iwch_cq(cq);
 249         struct t3_cq oldcq, newcq;
 250         int ret;
 251
 252         PDBG("%s ib_cq %p cqe %d\n", __func__, cq, cqe);
 253
 254         /* We don't downsize... */
 255         if (cqe <= cq->cqe)
 256                 return 0;
 257
 258         /* create new t3_cq with new size */
 259         cqe = roundup_pow_of_two(cqe+1);
 260         newcq.size_log2 = ilog2(cqe);
 261
 262         /* Dont allow resize to less than the current wce count */
 263         if (cqe < Q_COUNT(chp->cq.rptr, chp->cq.wptr)) {
 264                 return -ENOMEM;
 265         }
 266
 267         /* Quiesce all QPs using this CQ */
 268         ret = iwch_quiesce_qps(chp);
 269         if (ret) {
 270                 return ret;
 271         }
 272
 273         ret = cxio_create_cq(&chp->rhp->rdev, &newcq);
 274         if (ret) {
 275                 return ret;
 276         }
 277
 278         /* copy CQEs */
 279         memcpy(newcq.queue, chp->cq.queue, (1 << chp->cq.size_log2) *
 280                                         sizeof(struct t3_cqe));
 281
 282         /* old iwch_qp gets new t3_cq but keeps old cqid */
 283         oldcq = chp->cq;
 284         chp->cq = newcq;
 285         chp->cq.cqid = oldcq.cqid;
 286
 287         /* resize new t3_cq to update the HW context */
 288         ret = cxio_resize_cq(&chp->rhp->rdev, &chp->cq);
 289         if (ret) {
 290                 chp->cq = oldcq;
 291                 return ret;
 292         }
 293         chp->ibcq.cqe = (1<<chp->cq.size_log2) - 1;
 294
 295         /* destroy old t3_cq */
 296         oldcq.cqid = newcq.cqid;
 297         ret = cxio_destroy_cq(&chp->rhp->rdev, &oldcq);
 298         if (ret) {
 299                 printk(KERN_ERR MOD "%s - cxio_destroy_cq failed %d\n",
 300                         __func__, ret);
 301         }
 302
 303         /* add user hooks here */
 304
 305         /* resume qps */
 306         ret = iwch_resume_qps(chp);
 307         return ret;
 308 #else
 309         return -ENOSYS;
 310 #endif
 311 }
 312
 313 static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
 314 {
 315         struct iwch_dev *rhp;
 316         struct iwch_cq *chp;
 317         enum t3_cq_opcode cq_op;
 318         int err;
 319         unsigned long flag;
 320         u32 rptr;
 321
 322         chp = to_iwch_cq(ibcq);
 323         rhp = chp->rhp;
 324         if ((flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED)
 325                 cq_op = CQ_ARM_SE;
 326         else
 327                 cq_op = CQ_ARM_AN;
 328         if (chp->user_rptr_addr) {
 329                 if (get_user(rptr, chp->user_rptr_addr))
 330                         return -EFAULT;
 331                 spin_lock_irqsave(&chp->lock, flag);
 332                 chp->cq.rptr = rptr;
 333         } else
 334                 spin_lock_irqsave(&chp->lock, flag);
 335         PDBG("%s rptr 0x%x\n", __func__, chp->cq.rptr);
 336         err = cxio_hal_cq_op(&rhp->rdev, &chp->cq, cq_op, 0);
 337         spin_unlock_irqrestore(&chp->lock, flag);
 338         if (err < 0)
 339                 printk(KERN_ERR MOD "Error %d rearming CQID 0x%x\n", err,
 340                        chp->cq.cqid);
 341         if (err > 0 && !(flags & IB_CQ_REPORT_MISSED_EVENTS))
 342                 err = 0;
 343         return err;
 344 }
 345
 346 static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 347 {
 348         int len = vma->vm_end - vma->vm_start;
 349         u32 key = vma->vm_pgoff << PAGE_SHIFT;
 350         struct cxio_rdev *rdev_p;
 351         int ret = 0;
 352         struct iwch_mm_entry *mm;
 353         struct iwch_ucontext *ucontext;
 354         u64 addr;
 355
 356         PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __func__, vma->vm_pgoff,
 357              key, len);
 358
 359         if (vma->vm_start & (PAGE_SIZE-1)) {
 360                 return -EINVAL;
 361         }
 362
 363         rdev_p = &(to_iwch_dev(context->device)->rdev);
 364         ucontext = to_iwch_ucontext(context);
 365
 366         mm = remove_mmap(ucontext, key, len);
 367         if (!mm)
 368                 return -EINVAL;
 369         addr = mm->addr;
 370         kfree(mm);
 371
 372         if ((addr >= rdev_p->rnic_info.udbell_physbase) &&
 373             (addr < (rdev_p->rnic_info.udbell_physbase +
 374                        rdev_p->rnic_info.udbell_len))) {
 375
 376                 /*
 377                  * Map T3 DB register.
 378                  */
 379                 if (vma->vm_flags & VM_READ) {
 380                         return -EPERM;
 381                 }
 382
 383                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 384                 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
 385                 vma->vm_flags &= ~VM_MAYREAD;
 386                 ret = io_remap_pfn_range(vma, vma->vm_start,
 387                                          addr >> PAGE_SHIFT,
 388                                          len, vma->vm_page_prot);
 389         } else {
 390
 391                 /*
 392                  * Map WQ or CQ contig dma memory...
 393                  */
 394                 ret = remap_pfn_range(vma, vma->vm_start,
 395                                       addr >> PAGE_SHIFT,
 396                                       len, vma->vm_page_prot);
 397         }
 398
 399         return ret;
 400 }
 401
 402 static int iwch_deallocate_pd(struct ib_pd *pd)
 403 {
 404         struct iwch_dev *rhp;
 405         struct iwch_pd *php;
 406
 407         php = to_iwch_pd(pd);
 408         rhp = php->rhp;
 409         PDBG("%s ibpd %p pdid 0x%x\n", __func__, pd, php->pdid);
 410         cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid);
 411         kfree(php);
 412         return 0;
 413 }
 414
 415 static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev,
 416                                struct ib_ucontext *context,
 417                                struct ib_udata *udata)
 418 {
 419         struct iwch_pd *php;
 420         u32 pdid;
 421         struct iwch_dev *rhp;
 422
 423         PDBG("%s ibdev %p\n", __func__, ibdev);
 424         rhp = (struct iwch_dev *) ibdev;
 425         pdid = cxio_hal_get_pdid(rhp->rdev.rscp);
 426         if (!pdid)
 427                 return ERR_PTR(-EINVAL);
 428         php = kzalloc(sizeof(*php), GFP_KERNEL);
 429         if (!php) {
 430                 cxio_hal_put_pdid(rhp->rdev.rscp, pdid);
 431                 return ERR_PTR(-ENOMEM);
 432         }
 433         php->pdid = pdid;
 434         php->rhp = rhp;
 435         if (context) {
 436                 if (ib_copy_to_udata(udata, &php->pdid, sizeof (__u32))) {
 437                         iwch_deallocate_pd(&php->ibpd);
 438                         return ERR_PTR(-EFAULT);
 439                 }
 440         }
 441         PDBG("%s pdid 0x%0x ptr 0x%p\n", __func__, pdid, php);
 442         return &php->ibpd;
 443 }
 444
 445 static int iwch_dereg_mr(struct ib_mr *ib_mr)
 446 {
 447         struct iwch_dev *rhp;
 448         struct iwch_mr *mhp;
 449         u32 mmid;
 450
 451         PDBG("%s ib_mr %p\n", __func__, ib_mr);
 452         /* There can be no memory windows */
 453         if (atomic_read(&ib_mr->usecnt))
 454                 return -EINVAL;
 455
 456         mhp = to_iwch_mr(ib_mr);
 457         rhp = mhp->rhp;
 458         mmid = mhp->attr.stag >> 8;
 459         cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
 460                        mhp->attr.pbl_addr);
 461         iwch_free_pbl(mhp);
 462         remove_handle(rhp, &rhp->mmidr, mmid);
 463         if (mhp->kva)
 464                 kfree((void *) (unsigned long) mhp->kva);
 465         if (mhp->umem)
 466                 ib_umem_release(mhp->umem);
 467         PDBG("%s mmid 0x%x ptr %p\n", __func__, mmid, mhp);
 468         kfree(mhp);
 469         return 0;
 470 }
 471
 472 static struct ib_mr *iwch_register_phys_mem(struct ib_pd *pd,
 473                                         struct ib_phys_buf *buffer_list,
 474                                         int num_phys_buf,
 475                                         int acc,
 476                                         u64 *iova_start)
 477 {
 478         __be64 *page_list;
 479         int shift;
 480         u64 total_size;
 481         int npages;
 482         struct iwch_dev *rhp;
 483         struct iwch_pd *php;
 484         struct iwch_mr *mhp;
 485         int ret;
 486
 487         PDBG("%s ib_pd %p\n", __func__, pd);
 488         php = to_iwch_pd(pd);
 489         rhp = php->rhp;
 490
 491         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 492         if (!mhp)
 493                 return ERR_PTR(-ENOMEM);
 494
 495         mhp->rhp = rhp;
 496
 497         /* First check that we have enough alignment */
 498         if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
 499                 ret = -EINVAL;
 500                 goto err;
 501         }
 502
 503         if (num_phys_buf > 1 &&
 504             ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
 505                 ret = -EINVAL;
 506                 goto err;
 507         }
 508
 509         ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
 510                                    &total_size, &npages, &shift, &page_list);
 511         if (ret)
 512                 goto err;
 513
 514         ret = iwch_alloc_pbl(mhp, npages);
 515         if (ret) {
 516                 kfree(page_list);
 517                 goto err_pbl;
 518         }
 519
 520         ret = iwch_write_pbl(mhp, page_list, npages, 0);
 521         kfree(page_list);
 522         if (ret)
 523                 goto err_pbl;
 524
 525         mhp->attr.pdid = php->pdid;
 526         mhp->attr.zbva = 0;
 527
 528         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 529         mhp->attr.va_fbo = *iova_start;
 530         mhp->attr.page_size = shift - 12;
 531
 532         mhp->attr.len = (u32) total_size;
 533         mhp->attr.pbl_size = npages;
 534         ret = iwch_register_mem(rhp, php, mhp, shift);
 535         if (ret)
 536                 goto err_pbl;
 537
 538         return &mhp->ibmr;
 539
 540 err_pbl:
 541         iwch_free_pbl(mhp);
 542
 543 err:
 544         kfree(mhp);
 545         return ERR_PTR(ret);
 546
 547 }
 548
 549 static int iwch_reregister_phys_mem(struct ib_mr *mr,
 550                                      int mr_rereg_mask,
 551                                      struct ib_pd *pd,
 552                                      struct ib_phys_buf *buffer_list,
 553                                      int num_phys_buf,
 554                                      int acc, u64 * iova_start)
 555 {
 556
 557         struct iwch_mr mh, *mhp;
 558         struct iwch_pd *php;
 559         struct iwch_dev *rhp;
 560         __be64 *page_list = NULL;
 561         int shift = 0;
 562         u64 total_size;
 563         int npages = 0;
 564         int ret;
 565
 566         PDBG("%s ib_mr %p ib_pd %p\n", __func__, mr, pd);
 567
 568         /* There can be no memory windows */
 569         if (atomic_read(&mr->usecnt))
 570                 return -EINVAL;
 571
 572         mhp = to_iwch_mr(mr);
 573         rhp = mhp->rhp;
 574         php = to_iwch_pd(mr->pd);
 575
 576         /* make sure we are on the same adapter */
 577         if (rhp != php->rhp)
 578                 return -EINVAL;
 579
 580         memcpy(&mh, mhp, sizeof *mhp);
 581
 582         if (mr_rereg_mask & IB_MR_REREG_PD)
 583                 php = to_iwch_pd(pd);
 584         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 585                 mh.attr.perms = iwch_ib_to_tpt_access(acc);
 586         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 587                 ret = build_phys_page_list(buffer_list, num_phys_buf,
 588                                            iova_start,
 589                                            &total_size, &npages,
 590                                            &shift, &page_list);
 591                 if (ret)
 592                         return ret;
 593         }
 594
 595         ret = iwch_reregister_mem(rhp, php, &mh, shift, npages);
 596         kfree(page_list);
 597         if (ret) {
 598                 return ret;
 599         }
 600         if (mr_rereg_mask & IB_MR_REREG_PD)
 601                 mhp->attr.pdid = php->pdid;
 602         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 603                 mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 604         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 605                 mhp->attr.zbva = 0;
 606                 mhp->attr.va_fbo = *iova_start;
 607                 mhp->attr.page_size = shift - 12;
 608                 mhp->attr.len = (u32) total_size;
 609                 mhp->attr.pbl_size = npages;
 610         }
 611
 612         return 0;
 613 }
 614
 615
 616 static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 617                                       u64 virt, int acc, struct ib_udata *udata)
 618 {
 619         __be64 *pages;
 620         int shift, n, len;
 621         int i, j, k;
 622         int err = 0;
 623         struct ib_umem_chunk *chunk;
 624         struct iwch_dev *rhp;
 625         struct iwch_pd *php;
 626         struct iwch_mr *mhp;
 627         struct iwch_reg_user_mr_resp uresp;
 628
 629         PDBG("%s ib_pd %p\n", __func__, pd);
 630
 631         php = to_iwch_pd(pd);
 632         rhp = php->rhp;
 633         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 634         if (!mhp)
 635                 return ERR_PTR(-ENOMEM);
 636
 637         mhp->rhp = rhp;
 638
 639         mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
 640         if (IS_ERR(mhp->umem)) {
 641                 err = PTR_ERR(mhp->umem);
 642                 kfree(mhp);
 643                 return ERR_PTR(err);
 644         }
 645
 646         shift = ffs(mhp->umem->page_size) - 1;
 647
 648         n = 0;
 649         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 650                 n += chunk->nents;
 651
 652         err = iwch_alloc_pbl(mhp, n);
 653         if (err)
 654                 goto err;
 655
 656         pages = (__be64 *) __get_free_page(GFP_KERNEL);
 657         if (!pages) {
 658                 err = -ENOMEM;
 659                 goto err_pbl;
 660         }
 661
 662         i = n = 0;
 663
 664         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 665                 for (j = 0; j < chunk->nmap; ++j) {
 666                         len = sg_dma_len(&chunk->page_list[j]) >> shift;
 667                         for (k = 0; k < len; ++k) {
 668                                 pages[i++] = cpu_to_be64(sg_dma_address(
 669                                         &chunk->page_list[j]) +
 670                                         mhp->umem->page_size * k);
 671                                 if (i == PAGE_SIZE / sizeof *pages) {
 672                                         err = iwch_write_pbl(mhp, pages, i, n);
 673                                         if (err)
 674                                                 goto pbl_done;
 675                                         n += i;
 676                                         i = 0;
 677                                 }
 678                         }
 679                 }
 680
 681         if (i)
 682                 err = iwch_write_pbl(mhp, pages, i, n);
 683
 684 pbl_done:
 685         free_page((unsigned long) pages);
 686         if (err)
 687                 goto err_pbl;
 688
 689         mhp->attr.pdid = php->pdid;
 690         mhp->attr.zbva = 0;
 691         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 692         mhp->attr.va_fbo = virt;
 693         mhp->attr.page_size = shift - 12;
 694         mhp->attr.len = (u32) length;
 695
 696         err = iwch_register_mem(rhp, php, mhp, shift);
 697         if (err)
 698                 goto err_pbl;
 699
 700         if (udata && !t3a_device(rhp)) {
 701                 uresp.pbl_addr = (mhp->attr.pbl_addr -
 702                                  rhp->rdev.rnic_info.pbl_base) >> 3;
 703                 PDBG("%s user resp pbl_addr 0x%x\n", __func__,
 704                      uresp.pbl_addr);
 705
 706                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 707                         iwch_dereg_mr(&mhp->ibmr);
 708                         err = -EFAULT;
 709                         goto err;
 710                 }
 711         }
 712
 713         return &mhp->ibmr;
 714
 715 err_pbl:
 716         iwch_free_pbl(mhp);
 717
 718 err:
 719         ib_umem_release(mhp->umem);
 720         kfree(mhp);
 721         return ERR_PTR(err);
 722 }
 723
 724 static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
 725 {
 726         struct ib_phys_buf bl;
 727         u64 kva;
 728         struct ib_mr *ibmr;
 729
 730         PDBG("%s ib_pd %p\n", __func__, pd);
 731
 732         /*
 733          * T3 only supports 32 bits of size.
 734          */
 735         bl.size = 0xffffffff;
 736         bl.addr = 0;
 737         kva = 0;
 738         ibmr = iwch_register_phys_mem(pd, &bl, 1, acc, &kva);
 739         return ibmr;
 740 }
 741
 742 static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
 743 {
 744         struct iwch_dev *rhp;
 745         struct iwch_pd *php;
 746         struct iwch_mw *mhp;
 747         u32 mmid;
 748         u32 stag = 0;
 749         int ret;
 750
 751         if (type != IB_MW_TYPE_1)
 752                 return ERR_PTR(-EINVAL);
 753
 754         php = to_iwch_pd(pd);
 755         rhp = php->rhp;
 756         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 757         if (!mhp)
 758                 return ERR_PTR(-ENOMEM);
 759         ret = cxio_allocate_window(&rhp->rdev, &stag, php->pdid);
 760         if (ret) {
 761                 kfree(mhp);
 762                 return ERR_PTR(ret);
 763         }
 764         mhp->rhp = rhp;
 765         mhp->attr.pdid = php->pdid;
 766         mhp->attr.type = TPT_MW;
 767         mhp->attr.stag = stag;
 768         mmid = (stag) >> 8;
 769         mhp->ibmw.rkey = stag;
 770         if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) {
 771                 cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 772                 kfree(mhp);
 773                 return ERR_PTR(-ENOMEM);
 774         }
 775         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 776         return &(mhp->ibmw);
 777 }
 778
 779 static int iwch_dealloc_mw(struct ib_mw *mw)
 780 {
 781         struct iwch_dev *rhp;
 782         struct iwch_mw *mhp;
 783         u32 mmid;
 784
 785         mhp = to_iwch_mw(mw);
 786         rhp = mhp->rhp;
 787         mmid = (mw->rkey) >> 8;
 788         cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 789         remove_handle(rhp, &rhp->mmidr, mmid);
 790         PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
 791         kfree(mhp);
 792         return 0;
 793 }
 794
 795 static struct ib_mr *iwch_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth)
 796 {
 797         struct iwch_dev *rhp;
 798         struct iwch_pd *php;
 799         struct iwch_mr *mhp;
 800         u32 mmid;
 801         u32 stag = 0;
 802         int ret = 0;
 803
 804         php = to_iwch_pd(pd);
 805         rhp = php->rhp;
 806         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 807         if (!mhp)
 808                 goto err;
 809
 810         mhp->rhp = rhp;
 811         ret = iwch_alloc_pbl(mhp, pbl_depth);
 812         if (ret)
 813                 goto err1;
 814         mhp->attr.pbl_size = pbl_depth;
 815         ret = cxio_allocate_stag(&rhp->rdev, &stag, php->pdid,
 816                                  mhp->attr.pbl_size, mhp->attr.pbl_addr);
 817         if (ret)
 818                 goto err2;
 819         mhp->attr.pdid = php->pdid;
 820         mhp->attr.type = TPT_NON_SHARED_MR;
 821         mhp->attr.stag = stag;
 822         mhp->attr.state = 1;
 823         mmid = (stag) >> 8;
 824         mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
 825         if (insert_handle(rhp, &rhp->mmidr, mhp, mmid))
 826                 goto err3;
 827
 828         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 829         return &(mhp->ibmr);
 830 err3:
 831         cxio_dereg_mem(&rhp->rdev, stag, mhp->attr.pbl_size,
 832                        mhp->attr.pbl_addr);
 833 err2:
 834         iwch_free_pbl(mhp);
 835 err1:
 836         kfree(mhp);
 837 err:
 838         return ERR_PTR(ret);
 839 }
 840
 841 static struct ib_fast_reg_page_list *iwch_alloc_fastreg_pbl(
 842                                         struct ib_device *device,
 843                                         int page_list_len)
 844 {
 845         struct ib_fast_reg_page_list *page_list;
 846
 847         page_list = kmalloc(sizeof *page_list + page_list_len * sizeof(u64),
 848                             GFP_KERNEL);
 849         if (!page_list)
 850                 return ERR_PTR(-ENOMEM);
 851
 852         page_list->page_list = (u64 *)(page_list + 1);
 853         page_list->max_page_list_len = page_list_len;
 854
 855         return page_list;
 856 }
 857
 858 static void iwch_free_fastreg_pbl(struct ib_fast_reg_page_list *page_list)
 859 {
 860         kfree(page_list);
 861 }
 862
 863 static int iwch_destroy_qp(struct ib_qp *ib_qp)
 864 {
 865         struct iwch_dev *rhp;
 866         struct iwch_qp *qhp;
 867         struct iwch_qp_attributes attrs;
 868         struct iwch_ucontext *ucontext;
 869
 870         qhp = to_iwch_qp(ib_qp);
 871         rhp = qhp->rhp;
 872
 873         attrs.next_state = IWCH_QP_STATE_ERROR;
 874         iwch_modify_qp(rhp, qhp, IWCH_QP_ATTR_NEXT_STATE, &attrs, 0);
 875         wait_event(qhp->wait, !qhp->ep);
 876
 877         remove_handle(rhp, &rhp->qpidr, qhp->wq.qpid);
 878
 879         atomic_dec(&qhp->refcnt);
 880         wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
 881
 882         ucontext = ib_qp->uobject ? to_iwch_ucontext(ib_qp->uobject->context)
 883                                   : NULL;
 884         cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 885                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 886
 887         PDBG("%s ib_qp %p qpid 0x%0x qhp %p\n", __func__,
 888              ib_qp, qhp->wq.qpid, qhp);
 889         kfree(qhp);
 890         return 0;
 891 }
 892
 893 static struct ib_qp *iwch_create_qp(struct ib_pd *pd,
 894                              struct ib_qp_init_attr *attrs,
 895                              struct ib_udata *udata)
 896 {
 897         struct iwch_dev *rhp;
 898         struct iwch_qp *qhp;
 899         struct iwch_pd *php;
 900         struct iwch_cq *schp;
 901         struct iwch_cq *rchp;
 902         struct iwch_create_qp_resp uresp;
 903         int wqsize, sqsize, rqsize;
 904         struct iwch_ucontext *ucontext;
 905
 906         PDBG("%s ib_pd %p\n", __func__, pd);
 907         if (attrs->qp_type != IB_QPT_RC)
 908                 return ERR_PTR(-EINVAL);
 909         php = to_iwch_pd(pd);
 910         rhp = php->rhp;
 911         schp = get_chp(rhp, ((struct iwch_cq *) attrs->send_cq)->cq.cqid);
 912         rchp = get_chp(rhp, ((struct iwch_cq *) attrs->recv_cq)->cq.cqid);
 913         if (!schp || !rchp)
 914                 return ERR_PTR(-EINVAL);
 915
 916         /* The RQT size must be # of entries + 1 rounded up to a power of two */
 917         rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr);
 918         if (rqsize == attrs->cap.max_recv_wr)
 919                 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr+1);
 920
 921         /* T3 doesn't support RQT depth < 16 */
 922         if (rqsize < 16)
 923                 rqsize = 16;
 924
 925         if (rqsize > T3_MAX_RQ_SIZE)
 926                 return ERR_PTR(-EINVAL);
 927
 928         if (attrs->cap.max_inline_data > T3_MAX_INLINE)
 929                 return ERR_PTR(-EINVAL);
 930
 931         /*
 932          * NOTE: The SQ and total WQ sizes don't need to be
 933          * a power of two.  However, all the code assumes
 934          * they are. EG: Q_FREECNT() and friends.
 935          */
 936         sqsize = roundup_pow_of_two(attrs->cap.max_send_wr);
 937         wqsize = roundup_pow_of_two(rqsize + sqsize);
 938
 939         /*
 940          * Kernel users need more wq space for fastreg WRs which can take
 941          * 2 WR fragments.
 942          */
 943         ucontext = pd->uobject ? to_iwch_ucontext(pd->uobject->context) : NULL;
 944         if (!ucontext && wqsize < (rqsize + (2 * sqsize)))
 945                 wqsize = roundup_pow_of_two(rqsize +
 946                                 roundup_pow_of_two(attrs->cap.max_send_wr * 2));
 947         PDBG("%s wqsize %d sqsize %d rqsize %d\n", __func__,
 948              wqsize, sqsize, rqsize);
 949         qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
 950         if (!qhp)
 951                 return ERR_PTR(-ENOMEM);
 952         qhp->wq.size_log2 = ilog2(wqsize);
 953         qhp->wq.rq_size_log2 = ilog2(rqsize);
 954         qhp->wq.sq_size_log2 = ilog2(sqsize);
 955         if (cxio_create_qp(&rhp->rdev, !udata, &qhp->wq,
 956                            ucontext ? &ucontext->uctx : &rhp->rdev.uctx)) {
 957                 kfree(qhp);
 958                 return ERR_PTR(-ENOMEM);
 959         }
 960
 961         attrs->cap.max_recv_wr = rqsize - 1;
 962         attrs->cap.max_send_wr = sqsize;
 963         attrs->cap.max_inline_data = T3_MAX_INLINE;
 964
 965         qhp->rhp = rhp;
 966         qhp->attr.pd = php->pdid;
 967         qhp->attr.scq = ((struct iwch_cq *) attrs->send_cq)->cq.cqid;
 968         qhp->attr.rcq = ((struct iwch_cq *) attrs->recv_cq)->cq.cqid;
 969         qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
 970         qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
 971         qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
 972         qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
 973         qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
 974         qhp->attr.state = IWCH_QP_STATE_IDLE;
 975         qhp->attr.next_state = IWCH_QP_STATE_IDLE;
 976
 977         /*
 978          * XXX - These don't get passed in from the openib user
 979          * at create time.  The CM sets them via a QP modify.
 980          * Need to fix...  I think the CM should
 981          */
 982         qhp->attr.enable_rdma_read = 1;
 983         qhp->attr.enable_rdma_write = 1;
 984         qhp->attr.enable_bind = 1;
 985         qhp->attr.max_ord = 1;
 986         qhp->attr.max_ird = 1;
 987
 988         spin_lock_init(&qhp->lock);
 989         init_waitqueue_head(&qhp->wait);
 990         atomic_set(&qhp->refcnt, 1);
 991
 992         if (insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid)) {
 993                 cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 994                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 995                 kfree(qhp);
 996                 return ERR_PTR(-ENOMEM);
 997         }
 998
 999         if (udata) {
1000
1001                 struct iwch_mm_entry *mm1, *mm2;
1002
1003                 mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
1004                 if (!mm1) {
1005                         iwch_destroy_qp(&qhp->ibqp);
1006                         return ERR_PTR(-ENOMEM);
1007                 }
1008
1009                 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
1010                 if (!mm2) {
1011                         kfree(mm1);
1012                         iwch_destroy_qp(&qhp->ibqp);
1013                         return ERR_PTR(-ENOMEM);
1014                 }
1015
1016                 uresp.qpid = qhp->wq.qpid;
1017                 uresp.size_log2 = qhp->wq.size_log2;
1018                 uresp.sq_size_log2 = qhp->wq.sq_size_log2;
1019                 uresp.rq_size_log2 = qhp->wq.rq_size_log2;
1020                 spin_lock(&ucontext->mmap_lock);
1021                 uresp.key = ucontext->key;
1022                 ucontext->key += PAGE_SIZE;
1023                 uresp.db_key = ucontext->key;
1024                 ucontext->key += PAGE_SIZE;
1025                 spin_unlock(&ucontext->mmap_lock);
1026                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
1027                         kfree(mm1);
1028                         kfree(mm2);
1029                         iwch_destroy_qp(&qhp->ibqp);
1030                         return ERR_PTR(-EFAULT);
1031                 }
1032                 mm1->key = uresp.key;
1033                 mm1->addr = virt_to_phys(qhp->wq.queue);
1034                 mm1->len = PAGE_ALIGN(wqsize * sizeof (union t3_wr));
1035                 insert_mmap(ucontext, mm1);
1036                 mm2->key = uresp.db_key;
1037                 mm2->addr = qhp->wq.udb & PAGE_MASK;
1038                 mm2->len = PAGE_SIZE;
1039                 insert_mmap(ucontext, mm2);
1040         }
1041         qhp->ibqp.qp_num = qhp->wq.qpid;
1042         init_timer(&(qhp->timer));
1043         PDBG("%s sq_num_entries %d, rq_num_entries %d "
1044              "qpid 0x%0x qhp %p dma_addr 0x%llx size %d rq_addr 0x%x\n",
1045              __func__, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
1046              qhp->wq.qpid, qhp, (unsigned long long) qhp->wq.dma_addr,
1047              1 << qhp->wq.size_log2, qhp->wq.rq_addr);
1048         return &qhp->ibqp;
1049 }
1050
1051 static int iwch_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1052                       int attr_mask, struct ib_udata *udata)
1053 {
1054         struct iwch_dev *rhp;
1055         struct iwch_qp *qhp;
1056         enum iwch_qp_attr_mask mask = 0;
1057         struct iwch_qp_attributes attrs;
1058
1059         PDBG("%s ib_qp %p\n", __func__, ibqp);
1060
1061         /* iwarp does not support the RTR state */
1062         if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
1063                 attr_mask &= ~IB_QP_STATE;
1064
1065         /* Make sure we still have something left to do */
1066         if (!attr_mask)
1067                 return 0;
1068
1069         memset(&attrs, 0, sizeof attrs);
1070         qhp = to_iwch_qp(ibqp);
1071         rhp = qhp->rhp;
1072
1073         attrs.next_state = iwch_convert_state(attr->qp_state);
1074         attrs.enable_rdma_read = (attr->qp_access_flags &
1075                                IB_ACCESS_REMOTE_READ) ?  1 : 0;
1076         attrs.enable_rdma_write = (attr->qp_access_flags &
1077                                 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
1078         attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
1079
1080
1081         mask |= (attr_mask & IB_QP_STATE) ? IWCH_QP_ATTR_NEXT_STATE : 0;
1082         mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
1083                         (IWCH_QP_ATTR_ENABLE_RDMA_READ |
1084                          IWCH_QP_ATTR_ENABLE_RDMA_WRITE |
1085                          IWCH_QP_ATTR_ENABLE_RDMA_BIND) : 0;
1086
1087         return iwch_modify_qp(rhp, qhp, mask, &attrs, 0);
1088 }
1089
1090 void iwch_qp_add_ref(struct ib_qp *qp)
1091 {
1092         PDBG("%s ib_qp %p\n", __func__, qp);
1093         atomic_inc(&(to_iwch_qp(qp)->refcnt));
1094 }
1095
1096 void iwch_qp_rem_ref(struct ib_qp *qp)
1097 {
1098         PDBG("%s ib_qp %p\n", __func__, qp);
1099         if (atomic_dec_and_test(&(to_iwch_qp(qp)->refcnt)))
1100                 wake_up(&(to_iwch_qp(qp)->wait));
1101 }
1102
1103 static struct ib_qp *iwch_get_qp(struct ib_device *dev, int qpn)
1104 {
1105         PDBG("%s ib_dev %p qpn 0x%x\n", __func__, dev, qpn);
1106         return (struct ib_qp *)get_qhp(to_iwch_dev(dev), qpn);
1107 }
1108
1109
1110 static int iwch_query_pkey(struct ib_device *ibdev,
1111                            u8 port, u16 index, u16 * pkey)
1112 {
1113         PDBG("%s ibdev %p\n", __func__, ibdev);
1114         *pkey = 0;
1115         return 0;
1116 }
1117
1118 static int iwch_query_gid(struct ib_device *ibdev, u8 port,
1119                           int index, union ib_gid *gid)
1120 {
1121         struct iwch_dev *dev;
1122
1123         PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
1124                __func__, ibdev, port, index, gid);
1125         dev = to_iwch_dev(ibdev);
1126         BUG_ON(port == 0 || port > 2);
1127         memset(&(gid->raw[0]), 0, sizeof(gid->raw));
1128         memcpy(&(gid->raw[0]), dev->rdev.port_info.lldevs[port-1]->dev_addr, 6);
1129         return 0;
1130 }
1131
1132 static u64 fw_vers_string_to_u64(struct iwch_dev *iwch_dev)
1133 {
1134         struct ethtool_drvinfo info;
1135         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1136         char *cp, *next;
1137         unsigned fw_maj, fw_min, fw_mic;
1138
1139         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1140
1141         next = info.fw_version + 1;
1142         cp = strsep(&next, ".");
1143         sscanf(cp, "%i", &fw_maj);
1144         cp = strsep(&next, ".");
1145         sscanf(cp, "%i", &fw_min);
1146         cp = strsep(&next, ".");
1147         sscanf(cp, "%i", &fw_mic);
1148
1149         return (((u64)fw_maj & 0xffff) << 32) | ((fw_min & 0xffff) << 16) |
1150                (fw_mic & 0xffff);
1151 }
1152
1153 static int iwch_query_device(struct ib_device *ibdev,
1154                              struct ib_device_attr *props)
1155 {
1156
1157         struct iwch_dev *dev;
1158         PDBG("%s ibdev %p\n", __func__, ibdev);
1159
1160         dev = to_iwch_dev(ibdev);
1161         memset(props, 0, sizeof *props);
1162         memcpy(&props->sys_image_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1163         props->hw_ver = dev->rdev.t3cdev_p->type;
1164         props->fw_ver = fw_vers_string_to_u64(dev);
1165         props->device_cap_flags = dev->device_cap_flags;
1166         props->page_size_cap = dev->attr.mem_pgsizes_bitmask;
1167         props->vendor_id = (u32)dev->rdev.rnic_info.pdev->vendor;
1168         props->vendor_part_id = (u32)dev->rdev.rnic_info.pdev->device;
1169         props->max_mr_size = dev->attr.max_mr_size;
1170         props->max_qp = dev->attr.max_qps;
1171         props->max_qp_wr = dev->attr.max_wrs;
1172         props->max_sge = dev->attr.max_sge_per_wr;
1173         props->max_sge_rd = 1;
1174         props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
1175         props->max_qp_init_rd_atom = dev->attr.max_rdma_reads_per_qp;
1176         props->max_cq = dev->attr.max_cqs;
1177         props->max_cqe = dev->attr.max_cqes_per_cq;
1178         props->max_mr = dev->attr.max_mem_regs;
1179         props->max_pd = dev->attr.max_pds;
1180         props->local_ca_ack_delay = 0;
1181         props->max_fast_reg_page_list_len = T3_MAX_FASTREG_DEPTH;
1182
1183         return 0;
1184 }
1185
1186 static int iwch_query_port(struct ib_device *ibdev,
1187                            u8 port, struct ib_port_attr *props)
1188 {
1189         struct iwch_dev *dev;
1190         struct net_device *netdev;
1191         struct in_device *inetdev;
1192
1193         PDBG("%s ibdev %p\n", __func__, ibdev);
1194
1195         dev = to_iwch_dev(ibdev);
1196         netdev = dev->rdev.port_info.lldevs[port-1];
1197
1198         memset(props, 0, sizeof(struct ib_port_attr));
1199         props->max_mtu = IB_MTU_4096;
1200         if (netdev->mtu >= 4096)
1201                 props->active_mtu = IB_MTU_4096;
1202         else if (netdev->mtu >= 2048)
1203                 props->active_mtu = IB_MTU_2048;
1204         else if (netdev->mtu >= 1024)
1205                 props->active_mtu = IB_MTU_1024;
1206         else if (netdev->mtu >= 512)
1207                 props->active_mtu = IB_MTU_512;
1208         else
1209                 props->active_mtu = IB_MTU_256;
1210
1211         if (!netif_carrier_ok(netdev))
1212                 props->state = IB_PORT_DOWN;
1213         else {
1214                 inetdev = in_dev_get(netdev);
1215                 if (inetdev) {
1216                         if (inetdev->ifa_list)
1217                                 props->state = IB_PORT_ACTIVE;
1218                         else
1219                                 props->state = IB_PORT_INIT;
1220                         in_dev_put(inetdev);
1221                 } else
1222                         props->state = IB_PORT_INIT;
1223         }
1224
1225         props->port_cap_flags =
1226             IB_PORT_CM_SUP |
1227             IB_PORT_SNMP_TUNNEL_SUP |
1228             IB_PORT_REINIT_SUP |
1229             IB_PORT_DEVICE_MGMT_SUP |
1230             IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
1231         props->gid_tbl_len = 1;
1232         props->pkey_tbl_len = 1;
1233         props->active_width = 2;
1234         props->active_speed = IB_SPEED_DDR;
1235         props->max_msg_sz = -1;
1236
1237         return 0;
1238 }
1239
1240 static ssize_t show_rev(struct device *dev, struct device_attribute *attr,
1241                         char *buf)
1242 {
1243         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1244                                                  ibdev.dev);
1245         PDBG("%s dev 0x%p\n", __func__, dev);
1246         return sprintf(buf, "%d\n", iwch_dev->rdev.t3cdev_p->type);
1247 }
1248
1249 static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, char *buf)
1250 {
1251         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1252                                                  ibdev.dev);
1253         struct ethtool_drvinfo info;
1254         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1255
1256         PDBG("%s dev 0x%p\n", __func__, dev);
1257         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1258         return sprintf(buf, "%s\n", info.fw_version);
1259 }
1260
1261 static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
1262                         char *buf)
1263 {
1264         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1265                                                  ibdev.dev);
1266         struct ethtool_drvinfo info;
1267         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1268
1269         PDBG("%s dev 0x%p\n", __func__, dev);
1270         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1271         return sprintf(buf, "%s\n", info.driver);
1272 }
1273
1274 static ssize_t show_board(struct device *dev, struct device_attribute *attr,
1275                           char *buf)
1276 {
1277         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1278                                                  ibdev.dev);
1279         PDBG("%s dev 0x%p\n", __func__, dev);
1280         return sprintf(buf, "%x.%x\n", iwch_dev->rdev.rnic_info.pdev->vendor,
1281                        iwch_dev->rdev.rnic_info.pdev->device);
1282 }
1283
1284 static int iwch_get_mib(struct ib_device *ibdev,
1285                         union rdma_protocol_stats *stats)
1286 {
1287         struct iwch_dev *dev;
1288         struct tp_mib_stats m;
1289         int ret;
1290
1291         PDBG("%s ibdev %p\n", __func__, ibdev);
1292         dev = to_iwch_dev(ibdev);
1293         ret = dev->rdev.t3cdev_p->ctl(dev->rdev.t3cdev_p, RDMA_GET_MIB, &m);
1294         if (ret)
1295                 return -ENOSYS;
1296
1297         memset(stats, 0, sizeof *stats);
1298         stats->iw.ipInReceives = ((u64) m.ipInReceive_hi << 32) +
1299                                 m.ipInReceive_lo;
1300         stats->iw.ipInHdrErrors = ((u64) m.ipInHdrErrors_hi << 32) +
1301                                   m.ipInHdrErrors_lo;
1302         stats->iw.ipInAddrErrors = ((u64) m.ipInAddrErrors_hi << 32) +
1303                                    m.ipInAddrErrors_lo;
1304         stats->iw.ipInUnknownProtos = ((u64) m.ipInUnknownProtos_hi << 32) +
1305                                       m.ipInUnknownProtos_lo;
1306         stats->iw.ipInDiscards = ((u64) m.ipInDiscards_hi << 32) +
1307                                  m.ipInDiscards_lo;
1308         stats->iw.ipInDelivers = ((u64) m.ipInDelivers_hi << 32) +
1309                                  m.ipInDelivers_lo;
1310         stats->iw.ipOutRequests = ((u64) m.ipOutRequests_hi << 32) +
1311                                   m.ipOutRequests_lo;
1312         stats->iw.ipOutDiscards = ((u64) m.ipOutDiscards_hi << 32) +
1313                                   m.ipOutDiscards_lo;
1314         stats->iw.ipOutNoRoutes = ((u64) m.ipOutNoRoutes_hi << 32) +
1315                                   m.ipOutNoRoutes_lo;
1316         stats->iw.ipReasmTimeout = (u64) m.ipReasmTimeout;
1317         stats->iw.ipReasmReqds = (u64) m.ipReasmReqds;
1318         stats->iw.ipReasmOKs = (u64) m.ipReasmOKs;
1319         stats->iw.ipReasmFails = (u64) m.ipReasmFails;
1320         stats->iw.tcpActiveOpens = (u64) m.tcpActiveOpens;
1321         stats->iw.tcpPassiveOpens = (u64) m.tcpPassiveOpens;
1322         stats->iw.tcpAttemptFails = (u64) m.tcpAttemptFails;
1323         stats->iw.tcpEstabResets = (u64) m.tcpEstabResets;
1324         stats->iw.tcpOutRsts = (u64) m.tcpOutRsts;
1325         stats->iw.tcpCurrEstab = (u64) m.tcpCurrEstab;
1326         stats->iw.tcpInSegs = ((u64) m.tcpInSegs_hi << 32) +
1327                               m.tcpInSegs_lo;
1328         stats->iw.tcpOutSegs = ((u64) m.tcpOutSegs_hi << 32) +
1329                                m.tcpOutSegs_lo;
1330         stats->iw.tcpRetransSegs = ((u64) m.tcpRetransSeg_hi << 32) +
1331                                   m.tcpRetransSeg_lo;
1332         stats->iw.tcpInErrs = ((u64) m.tcpInErrs_hi << 32) +
1333                               m.tcpInErrs_lo;
1334         stats->iw.tcpRtoMin = (u64) m.tcpRtoMin;
1335         stats->iw.tcpRtoMax = (u64) m.tcpRtoMax;
1336         return 0;
1337 }
1338
1339 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1340 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1341 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1342 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1343
1344 static struct device_attribute *iwch_class_attributes[] = {
1345         &dev_attr_hw_rev,
1346         &dev_attr_fw_ver,
1347         &dev_attr_hca_type,
1348         &dev_attr_board_id,
1349 };
1350
1351 int iwch_register_device(struct iwch_dev *dev)
1352 {
1353         int ret;
1354         int i;
1355
1356         PDBG("%s iwch_dev %p\n", __func__, dev);
1357         strlcpy(dev->ibdev.name, "cxgb3_%d", IB_DEVICE_NAME_MAX);
1358         memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
1359         memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1360         dev->ibdev.owner = THIS_MODULE;
1361         dev->device_cap_flags = IB_DEVICE_LOCAL_DMA_LKEY |
1362                                 IB_DEVICE_MEM_WINDOW |
1363                                 IB_DEVICE_MEM_MGT_EXTENSIONS;
1364
1365         /* cxgb3 supports STag 0. */
1366         dev->ibdev.local_dma_lkey = 0;
1367
1368         dev->ibdev.uverbs_cmd_mask =
1369             (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1370             (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1371             (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1372             (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1373             (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1374             (1ull << IB_USER_VERBS_CMD_REG_MR) |
1375             (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1376             (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1377             (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1378             (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1379             (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1380             (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1381             (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1382             (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1383             (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1384             (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1385             (1ull << IB_USER_VERBS_CMD_POST_RECV);
1386         dev->ibdev.node_type = RDMA_NODE_RNIC;
1387         memcpy(dev->ibdev.node_desc, IWCH_NODE_DESC, sizeof(IWCH_NODE_DESC));
1388         dev->ibdev.phys_port_cnt = dev->rdev.port_info.nports;
1389         dev->ibdev.num_comp_vectors = 1;
1390         dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
1391         dev->ibdev.query_device = iwch_query_device;
1392         dev->ibdev.query_port = iwch_query_port;
1393         dev->ibdev.query_pkey = iwch_query_pkey;
1394         dev->ibdev.query_gid = iwch_query_gid;
1395         dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
1396         dev->ibdev.dealloc_ucontext = iwch_dealloc_ucontext;
1397         dev->ibdev.mmap = iwch_mmap;
1398         dev->ibdev.alloc_pd = iwch_allocate_pd;
1399         dev->ibdev.dealloc_pd = iwch_deallocate_pd;
1400         dev->ibdev.create_ah = iwch_ah_create;
1401         dev->ibdev.destroy_ah = iwch_ah_destroy;
1402         dev->ibdev.create_qp = iwch_create_qp;
1403         dev->ibdev.modify_qp = iwch_ib_modify_qp;
1404         dev->ibdev.destroy_qp = iwch_destroy_qp;
1405         dev->ibdev.create_cq = iwch_create_cq;
1406         dev->ibdev.destroy_cq = iwch_destroy_cq;
1407         dev->ibdev.resize_cq = iwch_resize_cq;
1408         dev->ibdev.poll_cq = iwch_poll_cq;
1409         dev->ibdev.get_dma_mr = iwch_get_dma_mr;
1410         dev->ibdev.reg_phys_mr = iwch_register_phys_mem;
1411         dev->ibdev.rereg_phys_mr = iwch_reregister_phys_mem;
1412         dev->ibdev.reg_user_mr = iwch_reg_user_mr;
1413         dev->ibdev.dereg_mr = iwch_dereg_mr;
1414         dev->ibdev.alloc_mw = iwch_alloc_mw;
1415         dev->ibdev.bind_mw = iwch_bind_mw;
1416         dev->ibdev.dealloc_mw = iwch_dealloc_mw;
1417         dev->ibdev.alloc_fast_reg_mr = iwch_alloc_fast_reg_mr;
1418         dev->ibdev.alloc_fast_reg_page_list = iwch_alloc_fastreg_pbl;
1419         dev->ibdev.free_fast_reg_page_list = iwch_free_fastreg_pbl;
1420         dev->ibdev.attach_mcast = iwch_multicast_attach;
1421         dev->ibdev.detach_mcast = iwch_multicast_detach;
1422         dev->ibdev.process_mad = iwch_process_mad;
1423         dev->ibdev.req_notify_cq = iwch_arm_cq;
1424         dev->ibdev.post_send = iwch_post_send;
1425         dev->ibdev.post_recv = iwch_post_receive;
1426         dev->ibdev.get_protocol_stats = iwch_get_mib;
1427         dev->ibdev.uverbs_abi_ver = IWCH_UVERBS_ABI_VERSION;
1428
1429         dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
1430         if (!dev->ibdev.iwcm)
1431                 return -ENOMEM;
1432
1433         dev->ibdev.iwcm->connect = iwch_connect;
1434         dev->ibdev.iwcm->accept = iwch_accept_cr;
1435         dev->ibdev.iwcm->reject = iwch_reject_cr;
1436         dev->ibdev.iwcm->create_listen = iwch_create_listen;
1437         dev->ibdev.iwcm->destroy_listen = iwch_destroy_listen;
1438         dev->ibdev.iwcm->add_ref = iwch_qp_add_ref;
1439         dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
1440         dev->ibdev.iwcm->get_qp = iwch_get_qp;
1441
1442         ret = ib_register_device(&dev->ibdev, NULL);
1443         if (ret)
1444                 goto bail1;
1445
1446         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i) {
1447                 ret = device_create_file(&dev->ibdev.dev,
1448                                          iwch_class_attributes[i]);
1449                 if (ret) {
1450                         goto bail2;
1451                 }
1452         }
1453         return 0;
1454 bail2:
1455         ib_unregister_device(&dev->ibdev);
1456 bail1:
1457         kfree(dev->ibdev.iwcm);
1458         return ret;
1459 }
1460
1461 void iwch_unregister_device(struct iwch_dev *dev)
1462 {
1463         int i;
1464
1465         PDBG("%s iwch_dev %p\n", __func__, dev);
1466         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i)
1467                 device_remove_file(&dev->ibdev.dev,
1468                                    iwch_class_attributes[i]);
1469         ib_unregister_device(&dev->ibdev);
1470         kfree(dev->ibdev.iwcm);
1471         return;
1472 }