drivers/infiniband/hw/cxgb3/iwch_provider.c

   1 /*
   2  * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32 #include <linux/module.h>
  33 #include <linux/moduleparam.h>
  34 #include <linux/device.h>
  35 #include <linux/netdevice.h>
  36 #include <linux/etherdevice.h>
  37 #include <linux/delay.h>
  38 #include <linux/errno.h>
  39 #include <linux/list.h>
  40 #include <linux/spinlock.h>
  41 #include <linux/ethtool.h>
  42 #include <linux/rtnetlink.h>
  43
  44 #include <asm/io.h>
  45 #include <asm/irq.h>
  46 #include <asm/byteorder.h>
  47
  48 #include <rdma/iw_cm.h>
  49 #include <rdma/ib_verbs.h>
  50 #include <rdma/ib_smi.h>
  51 #include <rdma/ib_umem.h>
  52 #include <rdma/ib_user_verbs.h>
  53
  54 #include "cxio_hal.h"
  55 #include "iwch.h"
  56 #include "iwch_provider.h"
  57 #include "iwch_cm.h"
  58 #include "iwch_user.h"
  59 #include "common.h"
  60
  61 static int iwch_modify_port(struct ib_device *ibdev,
  62                             u8 port, int port_modify_mask,
  63                             struct ib_port_modify *props)
  64 {
  65         return -ENOSYS;
  66 }
  67
  68 static struct ib_ah *iwch_ah_create(struct ib_pd *pd,
  69                                     struct ib_ah_attr *ah_attr)
  70 {
  71         return ERR_PTR(-ENOSYS);
  72 }
  73
  74 static int iwch_ah_destroy(struct ib_ah *ah)
  75 {
  76         return -ENOSYS;
  77 }
  78
  79 static int iwch_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
  80 {
  81         return -ENOSYS;
  82 }
  83
  84 static int iwch_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
  85 {
  86         return -ENOSYS;
  87 }
  88
  89 static int iwch_process_mad(struct ib_device *ibdev,
  90                             int mad_flags,
  91                             u8 port_num,
  92                             struct ib_wc *in_wc,
  93                             struct ib_grh *in_grh,
  94                             struct ib_mad *in_mad, struct ib_mad *out_mad)
  95 {
  96         return -ENOSYS;
  97 }
  98
  99 static int iwch_dealloc_ucontext(struct ib_ucontext *context)
 100 {
 101         struct iwch_dev *rhp = to_iwch_dev(context->device);
 102         struct iwch_ucontext *ucontext = to_iwch_ucontext(context);
 103         struct iwch_mm_entry *mm, *tmp;
 104
 105         PDBG("%s context %p\n", __func__, context);
 106         list_for_each_entry_safe(mm, tmp, &ucontext->mmaps, entry)
 107                 kfree(mm);
 108         cxio_release_ucontext(&rhp->rdev, &ucontext->uctx);
 109         kfree(ucontext);
 110         return 0;
 111 }
 112
 113 static struct ib_ucontext *iwch_alloc_ucontext(struct ib_device *ibdev,
 114                                         struct ib_udata *udata)
 115 {
 116         struct iwch_ucontext *context;
 117         struct iwch_dev *rhp = to_iwch_dev(ibdev);
 118
 119         PDBG("%s ibdev %p\n", __func__, ibdev);
 120         context = kzalloc(sizeof(*context), GFP_KERNEL);
 121         if (!context)
 122                 return ERR_PTR(-ENOMEM);
 123         cxio_init_ucontext(&rhp->rdev, &context->uctx);
 124         INIT_LIST_HEAD(&context->mmaps);
 125         spin_lock_init(&context->mmap_lock);
 126         return &context->ibucontext;
 127 }
 128
 129 static int iwch_destroy_cq(struct ib_cq *ib_cq)
 130 {
 131         struct iwch_cq *chp;
 132
 133         PDBG("%s ib_cq %p\n", __func__, ib_cq);
 134         chp = to_iwch_cq(ib_cq);
 135
 136         remove_handle(chp->rhp, &chp->rhp->cqidr, chp->cq.cqid);
 137         atomic_dec(&chp->refcnt);
 138         wait_event(chp->wait, !atomic_read(&chp->refcnt));
 139
 140         cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
 141         kfree(chp);
 142         return 0;
 143 }
 144
 145 static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries, int vector,
 146                              struct ib_ucontext *ib_context,
 147                              struct ib_udata *udata)
 148 {
 149         struct iwch_dev *rhp;
 150         struct iwch_cq *chp;
 151         struct iwch_create_cq_resp uresp;
 152         struct iwch_create_cq_req ureq;
 153         struct iwch_ucontext *ucontext = NULL;
 154
 155         PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
 156         rhp = to_iwch_dev(ibdev);
 157         chp = kzalloc(sizeof(*chp), GFP_KERNEL);
 158         if (!chp)
 159                 return ERR_PTR(-ENOMEM);
 160
 161         if (ib_context) {
 162                 ucontext = to_iwch_ucontext(ib_context);
 163                 if (!t3a_device(rhp)) {
 164                         if (ib_copy_from_udata(&ureq, udata, sizeof (ureq))) {
 165                                 kfree(chp);
 166                                 return ERR_PTR(-EFAULT);
 167                         }
 168                         chp->user_rptr_addr = (u32 __user *)(unsigned long)ureq.user_rptr_addr;
 169                 }
 170         }
 171
 172         if (t3a_device(rhp)) {
 173
 174                 /*
 175                  * T3A: Add some fluff to handle extra CQEs inserted
 176                  * for various errors.
 177                  * Additional CQE possibilities:
 178                  *      TERMINATE,
 179                  *      incoming RDMA WRITE Failures
 180                  *      incoming RDMA READ REQUEST FAILUREs
 181                  * NOTE: We cannot ensure the CQ won't overflow.
 182                  */
 183                 entries += 16;
 184         }
 185         entries = roundup_pow_of_two(entries);
 186         chp->cq.size_log2 = ilog2(entries);
 187
 188         if (cxio_create_cq(&rhp->rdev, &chp->cq)) {
 189                 kfree(chp);
 190                 return ERR_PTR(-ENOMEM);
 191         }
 192         chp->rhp = rhp;
 193         chp->ibcq.cqe = 1 << chp->cq.size_log2;
 194         spin_lock_init(&chp->lock);
 195         atomic_set(&chp->refcnt, 1);
 196         init_waitqueue_head(&chp->wait);
 197         insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid);
 198
 199         if (ucontext) {
 200                 struct iwch_mm_entry *mm;
 201
 202                 mm = kmalloc(sizeof *mm, GFP_KERNEL);
 203                 if (!mm) {
 204                         iwch_destroy_cq(&chp->ibcq);
 205                         return ERR_PTR(-ENOMEM);
 206                 }
 207                 uresp.cqid = chp->cq.cqid;
 208                 uresp.size_log2 = chp->cq.size_log2;
 209                 spin_lock(&ucontext->mmap_lock);
 210                 uresp.key = ucontext->key;
 211                 ucontext->key += PAGE_SIZE;
 212                 spin_unlock(&ucontext->mmap_lock);
 213                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 214                         kfree(mm);
 215                         iwch_destroy_cq(&chp->ibcq);
 216                         return ERR_PTR(-EFAULT);
 217                 }
 218                 mm->key = uresp.key;
 219                 mm->addr = virt_to_phys(chp->cq.queue);
 220                 mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
 221                                              sizeof (struct t3_cqe));
 222                 insert_mmap(ucontext, mm);
 223         }
 224         PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
 225              chp->cq.cqid, chp, (1 << chp->cq.size_log2),
 226              (unsigned long long) chp->cq.dma_addr);
 227         return &chp->ibcq;
 228 }
 229
 230 static int iwch_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
 231 {
 232 #ifdef notyet
 233         struct iwch_cq *chp = to_iwch_cq(cq);
 234         struct t3_cq oldcq, newcq;
 235         int ret;
 236
 237         PDBG("%s ib_cq %p cqe %d\n", __func__, cq, cqe);
 238
 239         /* We don't downsize... */
 240         if (cqe <= cq->cqe)
 241                 return 0;
 242
 243         /* create new t3_cq with new size */
 244         cqe = roundup_pow_of_two(cqe+1);
 245         newcq.size_log2 = ilog2(cqe);
 246
 247         /* Dont allow resize to less than the current wce count */
 248         if (cqe < Q_COUNT(chp->cq.rptr, chp->cq.wptr)) {
 249                 return -ENOMEM;
 250         }
 251
 252         /* Quiesce all QPs using this CQ */
 253         ret = iwch_quiesce_qps(chp);
 254         if (ret) {
 255                 return ret;
 256         }
 257
 258         ret = cxio_create_cq(&chp->rhp->rdev, &newcq);
 259         if (ret) {
 260                 return ret;
 261         }
 262
 263         /* copy CQEs */
 264         memcpy(newcq.queue, chp->cq.queue, (1 << chp->cq.size_log2) *
 265                                         sizeof(struct t3_cqe));
 266
 267         /* old iwch_qp gets new t3_cq but keeps old cqid */
 268         oldcq = chp->cq;
 269         chp->cq = newcq;
 270         chp->cq.cqid = oldcq.cqid;
 271
 272         /* resize new t3_cq to update the HW context */
 273         ret = cxio_resize_cq(&chp->rhp->rdev, &chp->cq);
 274         if (ret) {
 275                 chp->cq = oldcq;
 276                 return ret;
 277         }
 278         chp->ibcq.cqe = (1<<chp->cq.size_log2) - 1;
 279
 280         /* destroy old t3_cq */
 281         oldcq.cqid = newcq.cqid;
 282         ret = cxio_destroy_cq(&chp->rhp->rdev, &oldcq);
 283         if (ret) {
 284                 printk(KERN_ERR MOD "%s - cxio_destroy_cq failed %d\n",
 285                         __func__, ret);
 286         }
 287
 288         /* add user hooks here */
 289
 290         /* resume qps */
 291         ret = iwch_resume_qps(chp);
 292         return ret;
 293 #else
 294         return -ENOSYS;
 295 #endif
 296 }
 297
 298 static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
 299 {
 300         struct iwch_dev *rhp;
 301         struct iwch_cq *chp;
 302         enum t3_cq_opcode cq_op;
 303         int err;
 304         unsigned long flag;
 305         u32 rptr;
 306
 307         chp = to_iwch_cq(ibcq);
 308         rhp = chp->rhp;
 309         if ((flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED)
 310                 cq_op = CQ_ARM_SE;
 311         else
 312                 cq_op = CQ_ARM_AN;
 313         if (chp->user_rptr_addr) {
 314                 if (get_user(rptr, chp->user_rptr_addr))
 315                         return -EFAULT;
 316                 spin_lock_irqsave(&chp->lock, flag);
 317                 chp->cq.rptr = rptr;
 318         } else
 319                 spin_lock_irqsave(&chp->lock, flag);
 320         PDBG("%s rptr 0x%x\n", __func__, chp->cq.rptr);
 321         err = cxio_hal_cq_op(&rhp->rdev, &chp->cq, cq_op, 0);
 322         spin_unlock_irqrestore(&chp->lock, flag);
 323         if (err < 0)
 324                 printk(KERN_ERR MOD "Error %d rearming CQID 0x%x\n", err,
 325                        chp->cq.cqid);
 326         if (err > 0 && !(flags & IB_CQ_REPORT_MISSED_EVENTS))
 327                 err = 0;
 328         return err;
 329 }
 330
 331 static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 332 {
 333         int len = vma->vm_end - vma->vm_start;
 334         u32 key = vma->vm_pgoff << PAGE_SHIFT;
 335         struct cxio_rdev *rdev_p;
 336         int ret = 0;
 337         struct iwch_mm_entry *mm;
 338         struct iwch_ucontext *ucontext;
 339         u64 addr;
 340
 341         PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __func__, vma->vm_pgoff,
 342              key, len);
 343
 344         if (vma->vm_start & (PAGE_SIZE-1)) {
 345                 return -EINVAL;
 346         }
 347
 348         rdev_p = &(to_iwch_dev(context->device)->rdev);
 349         ucontext = to_iwch_ucontext(context);
 350
 351         mm = remove_mmap(ucontext, key, len);
 352         if (!mm)
 353                 return -EINVAL;
 354         addr = mm->addr;
 355         kfree(mm);
 356
 357         if ((addr >= rdev_p->rnic_info.udbell_physbase) &&
 358             (addr < (rdev_p->rnic_info.udbell_physbase +
 359                        rdev_p->rnic_info.udbell_len))) {
 360
 361                 /*
 362                  * Map T3 DB register.
 363                  */
 364                 if (vma->vm_flags & VM_READ) {
 365                         return -EPERM;
 366                 }
 367
 368                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 369                 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
 370                 vma->vm_flags &= ~VM_MAYREAD;
 371                 ret = io_remap_pfn_range(vma, vma->vm_start,
 372                                          addr >> PAGE_SHIFT,
 373                                          len, vma->vm_page_prot);
 374         } else {
 375
 376                 /*
 377                  * Map WQ or CQ contig dma memory...
 378                  */
 379                 ret = remap_pfn_range(vma, vma->vm_start,
 380                                       addr >> PAGE_SHIFT,
 381                                       len, vma->vm_page_prot);
 382         }
 383
 384         return ret;
 385 }
 386
 387 static int iwch_deallocate_pd(struct ib_pd *pd)
 388 {
 389         struct iwch_dev *rhp;
 390         struct iwch_pd *php;
 391
 392         php = to_iwch_pd(pd);
 393         rhp = php->rhp;
 394         PDBG("%s ibpd %p pdid 0x%x\n", __func__, pd, php->pdid);
 395         cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid);
 396         kfree(php);
 397         return 0;
 398 }
 399
 400 static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev,
 401                                struct ib_ucontext *context,
 402                                struct ib_udata *udata)
 403 {
 404         struct iwch_pd *php;
 405         u32 pdid;
 406         struct iwch_dev *rhp;
 407
 408         PDBG("%s ibdev %p\n", __func__, ibdev);
 409         rhp = (struct iwch_dev *) ibdev;
 410         pdid = cxio_hal_get_pdid(rhp->rdev.rscp);
 411         if (!pdid)
 412                 return ERR_PTR(-EINVAL);
 413         php = kzalloc(sizeof(*php), GFP_KERNEL);
 414         if (!php) {
 415                 cxio_hal_put_pdid(rhp->rdev.rscp, pdid);
 416                 return ERR_PTR(-ENOMEM);
 417         }
 418         php->pdid = pdid;
 419         php->rhp = rhp;
 420         if (context) {
 421                 if (ib_copy_to_udata(udata, &php->pdid, sizeof (__u32))) {
 422                         iwch_deallocate_pd(&php->ibpd);
 423                         return ERR_PTR(-EFAULT);
 424                 }
 425         }
 426         PDBG("%s pdid 0x%0x ptr 0x%p\n", __func__, pdid, php);
 427         return &php->ibpd;
 428 }
 429
 430 static int iwch_dereg_mr(struct ib_mr *ib_mr)
 431 {
 432         struct iwch_dev *rhp;
 433         struct iwch_mr *mhp;
 434         u32 mmid;
 435
 436         PDBG("%s ib_mr %p\n", __func__, ib_mr);
 437         /* There can be no memory windows */
 438         if (atomic_read(&ib_mr->usecnt))
 439                 return -EINVAL;
 440
 441         mhp = to_iwch_mr(ib_mr);
 442         rhp = mhp->rhp;
 443         mmid = mhp->attr.stag >> 8;
 444         cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
 445                        mhp->attr.pbl_addr);
 446         iwch_free_pbl(mhp);
 447         remove_handle(rhp, &rhp->mmidr, mmid);
 448         if (mhp->kva)
 449                 kfree((void *) (unsigned long) mhp->kva);
 450         if (mhp->umem)
 451                 ib_umem_release(mhp->umem);
 452         PDBG("%s mmid 0x%x ptr %p\n", __func__, mmid, mhp);
 453         kfree(mhp);
 454         return 0;
 455 }
 456
 457 static struct ib_mr *iwch_register_phys_mem(struct ib_pd *pd,
 458                                         struct ib_phys_buf *buffer_list,
 459                                         int num_phys_buf,
 460                                         int acc,
 461                                         u64 *iova_start)
 462 {
 463         __be64 *page_list;
 464         int shift;
 465         u64 total_size;
 466         int npages;
 467         struct iwch_dev *rhp;
 468         struct iwch_pd *php;
 469         struct iwch_mr *mhp;
 470         int ret;
 471
 472         PDBG("%s ib_pd %p\n", __func__, pd);
 473         php = to_iwch_pd(pd);
 474         rhp = php->rhp;
 475
 476         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 477         if (!mhp)
 478                 return ERR_PTR(-ENOMEM);
 479
 480         mhp->rhp = rhp;
 481
 482         /* First check that we have enough alignment */
 483         if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
 484                 ret = -EINVAL;
 485                 goto err;
 486         }
 487
 488         if (num_phys_buf > 1 &&
 489             ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
 490                 ret = -EINVAL;
 491                 goto err;
 492         }
 493
 494         ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
 495                                    &total_size, &npages, &shift, &page_list);
 496         if (ret)
 497                 goto err;
 498
 499         ret = iwch_alloc_pbl(mhp, npages);
 500         if (ret) {
 501                 kfree(page_list);
 502                 goto err_pbl;
 503         }
 504
 505         ret = iwch_write_pbl(mhp, page_list, npages, 0);
 506         kfree(page_list);
 507         if (ret)
 508                 goto err_pbl;
 509
 510         mhp->attr.pdid = php->pdid;
 511         mhp->attr.zbva = 0;
 512
 513         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 514         mhp->attr.va_fbo = *iova_start;
 515         mhp->attr.page_size = shift - 12;
 516
 517         mhp->attr.len = (u32) total_size;
 518         mhp->attr.pbl_size = npages;
 519         ret = iwch_register_mem(rhp, php, mhp, shift);
 520         if (ret)
 521                 goto err_pbl;
 522
 523         return &mhp->ibmr;
 524
 525 err_pbl:
 526         iwch_free_pbl(mhp);
 527
 528 err:
 529         kfree(mhp);
 530         return ERR_PTR(ret);
 531
 532 }
 533
 534 static int iwch_reregister_phys_mem(struct ib_mr *mr,
 535                                      int mr_rereg_mask,
 536                                      struct ib_pd *pd,
 537                                      struct ib_phys_buf *buffer_list,
 538                                      int num_phys_buf,
 539                                      int acc, u64 * iova_start)
 540 {
 541
 542         struct iwch_mr mh, *mhp;
 543         struct iwch_pd *php;
 544         struct iwch_dev *rhp;
 545         __be64 *page_list = NULL;
 546         int shift = 0;
 547         u64 total_size;
 548         int npages;
 549         int ret;
 550
 551         PDBG("%s ib_mr %p ib_pd %p\n", __func__, mr, pd);
 552
 553         /* There can be no memory windows */
 554         if (atomic_read(&mr->usecnt))
 555                 return -EINVAL;
 556
 557         mhp = to_iwch_mr(mr);
 558         rhp = mhp->rhp;
 559         php = to_iwch_pd(mr->pd);
 560
 561         /* make sure we are on the same adapter */
 562         if (rhp != php->rhp)
 563                 return -EINVAL;
 564
 565         memcpy(&mh, mhp, sizeof *mhp);
 566
 567         if (mr_rereg_mask & IB_MR_REREG_PD)
 568                 php = to_iwch_pd(pd);
 569         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 570                 mh.attr.perms = iwch_ib_to_tpt_access(acc);
 571         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 572                 ret = build_phys_page_list(buffer_list, num_phys_buf,
 573                                            iova_start,
 574                                            &total_size, &npages,
 575                                            &shift, &page_list);
 576                 if (ret)
 577                         return ret;
 578         }
 579
 580         ret = iwch_reregister_mem(rhp, php, &mh, shift, npages);
 581         kfree(page_list);
 582         if (ret) {
 583                 return ret;
 584         }
 585         if (mr_rereg_mask & IB_MR_REREG_PD)
 586                 mhp->attr.pdid = php->pdid;
 587         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 588                 mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 589         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 590                 mhp->attr.zbva = 0;
 591                 mhp->attr.va_fbo = *iova_start;
 592                 mhp->attr.page_size = shift - 12;
 593                 mhp->attr.len = (u32) total_size;
 594                 mhp->attr.pbl_size = npages;
 595         }
 596
 597         return 0;
 598 }
 599
 600
 601 static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 602                                       u64 virt, int acc, struct ib_udata *udata)
 603 {
 604         __be64 *pages;
 605         int shift, n, len;
 606         int i, j, k;
 607         int err = 0;
 608         struct ib_umem_chunk *chunk;
 609         struct iwch_dev *rhp;
 610         struct iwch_pd *php;
 611         struct iwch_mr *mhp;
 612         struct iwch_reg_user_mr_resp uresp;
 613
 614         PDBG("%s ib_pd %p\n", __func__, pd);
 615
 616         php = to_iwch_pd(pd);
 617         rhp = php->rhp;
 618         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 619         if (!mhp)
 620                 return ERR_PTR(-ENOMEM);
 621
 622         mhp->rhp = rhp;
 623
 624         mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
 625         if (IS_ERR(mhp->umem)) {
 626                 err = PTR_ERR(mhp->umem);
 627                 kfree(mhp);
 628                 return ERR_PTR(err);
 629         }
 630
 631         shift = ffs(mhp->umem->page_size) - 1;
 632
 633         n = 0;
 634         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 635                 n += chunk->nents;
 636
 637         err = iwch_alloc_pbl(mhp, n);
 638         if (err)
 639                 goto err;
 640
 641         pages = (__be64 *) __get_free_page(GFP_KERNEL);
 642         if (!pages) {
 643                 err = -ENOMEM;
 644                 goto err_pbl;
 645         }
 646
 647         i = n = 0;
 648
 649         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 650                 for (j = 0; j < chunk->nmap; ++j) {
 651                         len = sg_dma_len(&chunk->page_list[j]) >> shift;
 652                         for (k = 0; k < len; ++k) {
 653                                 pages[i++] = cpu_to_be64(sg_dma_address(
 654                                         &chunk->page_list[j]) +
 655                                         mhp->umem->page_size * k);
 656                                 if (i == PAGE_SIZE / sizeof *pages) {
 657                                         err = iwch_write_pbl(mhp, pages, i, n);
 658                                         if (err)
 659                                                 goto pbl_done;
 660                                         n += i;
 661                                         i = 0;
 662                                 }
 663                         }
 664                 }
 665
 666         if (i)
 667                 err = iwch_write_pbl(mhp, pages, i, n);
 668
 669 pbl_done:
 670         free_page((unsigned long) pages);
 671         if (err)
 672                 goto err_pbl;
 673
 674         mhp->attr.pdid = php->pdid;
 675         mhp->attr.zbva = 0;
 676         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 677         mhp->attr.va_fbo = virt;
 678         mhp->attr.page_size = shift - 12;
 679         mhp->attr.len = (u32) length;
 680
 681         err = iwch_register_mem(rhp, php, mhp, shift);
 682         if (err)
 683                 goto err_pbl;
 684
 685         if (udata && !t3a_device(rhp)) {
 686                 uresp.pbl_addr = (mhp->attr.pbl_addr -
 687                                  rhp->rdev.rnic_info.pbl_base) >> 3;
 688                 PDBG("%s user resp pbl_addr 0x%x\n", __func__,
 689                      uresp.pbl_addr);
 690
 691                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 692                         iwch_dereg_mr(&mhp->ibmr);
 693                         err = -EFAULT;
 694                         goto err;
 695                 }
 696         }
 697
 698         return &mhp->ibmr;
 699
 700 err_pbl:
 701         iwch_free_pbl(mhp);
 702
 703 err:
 704         ib_umem_release(mhp->umem);
 705         kfree(mhp);
 706         return ERR_PTR(err);
 707 }
 708
 709 static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
 710 {
 711         struct ib_phys_buf bl;
 712         u64 kva;
 713         struct ib_mr *ibmr;
 714
 715         PDBG("%s ib_pd %p\n", __func__, pd);
 716
 717         /*
 718          * T3 only supports 32 bits of size.
 719          */
 720         bl.size = 0xffffffff;
 721         bl.addr = 0;
 722         kva = 0;
 723         ibmr = iwch_register_phys_mem(pd, &bl, 1, acc, &kva);
 724         return ibmr;
 725 }
 726
 727 static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
 728 {
 729         struct iwch_dev *rhp;
 730         struct iwch_pd *php;
 731         struct iwch_mw *mhp;
 732         u32 mmid;
 733         u32 stag = 0;
 734         int ret;
 735
 736         php = to_iwch_pd(pd);
 737         rhp = php->rhp;
 738         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 739         if (!mhp)
 740                 return ERR_PTR(-ENOMEM);
 741         ret = cxio_allocate_window(&rhp->rdev, &stag, php->pdid);
 742         if (ret) {
 743                 kfree(mhp);
 744                 return ERR_PTR(ret);
 745         }
 746         mhp->rhp = rhp;
 747         mhp->attr.pdid = php->pdid;
 748         mhp->attr.type = TPT_MW;
 749         mhp->attr.stag = stag;
 750         mmid = (stag) >> 8;
 751         insert_handle(rhp, &rhp->mmidr, mhp, mmid);
 752         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 753         return &(mhp->ibmw);
 754 }
 755
 756 static int iwch_dealloc_mw(struct ib_mw *mw)
 757 {
 758         struct iwch_dev *rhp;
 759         struct iwch_mw *mhp;
 760         u32 mmid;
 761
 762         mhp = to_iwch_mw(mw);
 763         rhp = mhp->rhp;
 764         mmid = (mw->rkey) >> 8;
 765         cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 766         remove_handle(rhp, &rhp->mmidr, mmid);
 767         kfree(mhp);
 768         PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
 769         return 0;
 770 }
 771
 772 static struct ib_mr *iwch_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth)
 773 {
 774         struct iwch_dev *rhp;
 775         struct iwch_pd *php;
 776         struct iwch_mr *mhp;
 777         u32 mmid;
 778         u32 stag = 0;
 779         int ret;
 780
 781         php = to_iwch_pd(pd);
 782         rhp = php->rhp;
 783         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 784         if (!mhp)
 785                 return ERR_PTR(-ENOMEM);
 786
 787         mhp->rhp = rhp;
 788         ret = iwch_alloc_pbl(mhp, pbl_depth);
 789         if (ret) {
 790                 kfree(mhp);
 791                 return ERR_PTR(ret);
 792         }
 793         mhp->attr.pbl_size = pbl_depth;
 794         ret = cxio_allocate_stag(&rhp->rdev, &stag, php->pdid,
 795                                  mhp->attr.pbl_size, mhp->attr.pbl_addr);
 796         if (ret) {
 797                 iwch_free_pbl(mhp);
 798                 kfree(mhp);
 799                 return ERR_PTR(ret);
 800         }
 801         mhp->attr.pdid = php->pdid;
 802         mhp->attr.type = TPT_NON_SHARED_MR;
 803         mhp->attr.stag = stag;
 804         mhp->attr.state = 1;
 805         mmid = (stag) >> 8;
 806         mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
 807         insert_handle(rhp, &rhp->mmidr, mhp, mmid);
 808         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 809         return &(mhp->ibmr);
 810 }
 811
 812 static struct ib_fast_reg_page_list *iwch_alloc_fastreg_pbl(
 813                                         struct ib_device *device,
 814                                         int page_list_len)
 815 {
 816         struct ib_fast_reg_page_list *page_list;
 817
 818         page_list = kmalloc(sizeof *page_list + page_list_len * sizeof(u64),
 819                             GFP_KERNEL);
 820         if (!page_list)
 821                 return ERR_PTR(-ENOMEM);
 822
 823         page_list->page_list = (u64 *)(page_list + 1);
 824         page_list->max_page_list_len = page_list_len;
 825
 826         return page_list;
 827 }
 828
 829 static void iwch_free_fastreg_pbl(struct ib_fast_reg_page_list *page_list)
 830 {
 831         kfree(page_list);
 832 }
 833
 834 static int iwch_destroy_qp(struct ib_qp *ib_qp)
 835 {
 836         struct iwch_dev *rhp;
 837         struct iwch_qp *qhp;
 838         struct iwch_qp_attributes attrs;
 839         struct iwch_ucontext *ucontext;
 840
 841         qhp = to_iwch_qp(ib_qp);
 842         rhp = qhp->rhp;
 843
 844         attrs.next_state = IWCH_QP_STATE_ERROR;
 845         iwch_modify_qp(rhp, qhp, IWCH_QP_ATTR_NEXT_STATE, &attrs, 0);
 846         wait_event(qhp->wait, !qhp->ep);
 847
 848         remove_handle(rhp, &rhp->qpidr, qhp->wq.qpid);
 849
 850         atomic_dec(&qhp->refcnt);
 851         wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
 852
 853         ucontext = ib_qp->uobject ? to_iwch_ucontext(ib_qp->uobject->context)
 854                                   : NULL;
 855         cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 856                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 857
 858         PDBG("%s ib_qp %p qpid 0x%0x qhp %p\n", __func__,
 859              ib_qp, qhp->wq.qpid, qhp);
 860         kfree(qhp);
 861         return 0;
 862 }
 863
 864 static struct ib_qp *iwch_create_qp(struct ib_pd *pd,
 865                              struct ib_qp_init_attr *attrs,
 866                              struct ib_udata *udata)
 867 {
 868         struct iwch_dev *rhp;
 869         struct iwch_qp *qhp;
 870         struct iwch_pd *php;
 871         struct iwch_cq *schp;
 872         struct iwch_cq *rchp;
 873         struct iwch_create_qp_resp uresp;
 874         int wqsize, sqsize, rqsize;
 875         struct iwch_ucontext *ucontext;
 876
 877         PDBG("%s ib_pd %p\n", __func__, pd);
 878         if (attrs->qp_type != IB_QPT_RC)
 879                 return ERR_PTR(-EINVAL);
 880         php = to_iwch_pd(pd);
 881         rhp = php->rhp;
 882         schp = get_chp(rhp, ((struct iwch_cq *) attrs->send_cq)->cq.cqid);
 883         rchp = get_chp(rhp, ((struct iwch_cq *) attrs->recv_cq)->cq.cqid);
 884         if (!schp || !rchp)
 885                 return ERR_PTR(-EINVAL);
 886
 887         /* The RQT size must be # of entries + 1 rounded up to a power of two */
 888         rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr);
 889         if (rqsize == attrs->cap.max_recv_wr)
 890                 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr+1);
 891
 892         /* T3 doesn't support RQT depth < 16 */
 893         if (rqsize < 16)
 894                 rqsize = 16;
 895
 896         if (rqsize > T3_MAX_RQ_SIZE)
 897                 return ERR_PTR(-EINVAL);
 898
 899         if (attrs->cap.max_inline_data > T3_MAX_INLINE)
 900                 return ERR_PTR(-EINVAL);
 901
 902         /*
 903          * NOTE: The SQ and total WQ sizes don't need to be
 904          * a power of two.  However, all the code assumes
 905          * they are. EG: Q_FREECNT() and friends.
 906          */
 907         sqsize = roundup_pow_of_two(attrs->cap.max_send_wr);
 908         wqsize = roundup_pow_of_two(rqsize + sqsize);
 909
 910         /*
 911          * Kernel users need more wq space for fastreg WRs which can take
 912          * 2 WR fragments.
 913          */
 914         ucontext = pd->uobject ? to_iwch_ucontext(pd->uobject->context) : NULL;
 915         if (!ucontext && wqsize < (rqsize + (2 * sqsize)))
 916                 wqsize = roundup_pow_of_two(rqsize +
 917                                 roundup_pow_of_two(attrs->cap.max_send_wr * 2));
 918         PDBG("%s wqsize %d sqsize %d rqsize %d\n", __func__,
 919              wqsize, sqsize, rqsize);
 920         qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
 921         if (!qhp)
 922                 return ERR_PTR(-ENOMEM);
 923         qhp->wq.size_log2 = ilog2(wqsize);
 924         qhp->wq.rq_size_log2 = ilog2(rqsize);
 925         qhp->wq.sq_size_log2 = ilog2(sqsize);
 926         if (cxio_create_qp(&rhp->rdev, !udata, &qhp->wq,
 927                            ucontext ? &ucontext->uctx : &rhp->rdev.uctx)) {
 928                 kfree(qhp);
 929                 return ERR_PTR(-ENOMEM);
 930         }
 931
 932         attrs->cap.max_recv_wr = rqsize - 1;
 933         attrs->cap.max_send_wr = sqsize;
 934         attrs->cap.max_inline_data = T3_MAX_INLINE;
 935
 936         qhp->rhp = rhp;
 937         qhp->attr.pd = php->pdid;
 938         qhp->attr.scq = ((struct iwch_cq *) attrs->send_cq)->cq.cqid;
 939         qhp->attr.rcq = ((struct iwch_cq *) attrs->recv_cq)->cq.cqid;
 940         qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
 941         qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
 942         qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
 943         qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
 944         qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
 945         qhp->attr.state = IWCH_QP_STATE_IDLE;
 946         qhp->attr.next_state = IWCH_QP_STATE_IDLE;
 947
 948         /*
 949          * XXX - These don't get passed in from the openib user
 950          * at create time.  The CM sets them via a QP modify.
 951          * Need to fix...  I think the CM should
 952          */
 953         qhp->attr.enable_rdma_read = 1;
 954         qhp->attr.enable_rdma_write = 1;
 955         qhp->attr.enable_bind = 1;
 956         qhp->attr.max_ord = 1;
 957         qhp->attr.max_ird = 1;
 958
 959         spin_lock_init(&qhp->lock);
 960         init_waitqueue_head(&qhp->wait);
 961         atomic_set(&qhp->refcnt, 1);
 962         insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid);
 963
 964         if (udata) {
 965
 966                 struct iwch_mm_entry *mm1, *mm2;
 967
 968                 mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
 969                 if (!mm1) {
 970                         iwch_destroy_qp(&qhp->ibqp);
 971                         return ERR_PTR(-ENOMEM);
 972                 }
 973
 974                 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
 975                 if (!mm2) {
 976                         kfree(mm1);
 977                         iwch_destroy_qp(&qhp->ibqp);
 978                         return ERR_PTR(-ENOMEM);
 979                 }
 980
 981                 uresp.qpid = qhp->wq.qpid;
 982                 uresp.size_log2 = qhp->wq.size_log2;
 983                 uresp.sq_size_log2 = qhp->wq.sq_size_log2;
 984                 uresp.rq_size_log2 = qhp->wq.rq_size_log2;
 985                 spin_lock(&ucontext->mmap_lock);
 986                 uresp.key = ucontext->key;
 987                 ucontext->key += PAGE_SIZE;
 988                 uresp.db_key = ucontext->key;
 989                 ucontext->key += PAGE_SIZE;
 990                 spin_unlock(&ucontext->mmap_lock);
 991                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 992                         kfree(mm1);
 993                         kfree(mm2);
 994                         iwch_destroy_qp(&qhp->ibqp);
 995                         return ERR_PTR(-EFAULT);
 996                 }
 997                 mm1->key = uresp.key;
 998                 mm1->addr = virt_to_phys(qhp->wq.queue);
 999                 mm1->len = PAGE_ALIGN(wqsize * sizeof (union t3_wr));
1000                 insert_mmap(ucontext, mm1);
1001                 mm2->key = uresp.db_key;
1002                 mm2->addr = qhp->wq.udb & PAGE_MASK;
1003                 mm2->len = PAGE_SIZE;
1004                 insert_mmap(ucontext, mm2);
1005         }
1006         qhp->ibqp.qp_num = qhp->wq.qpid;
1007         init_timer(&(qhp->timer));
1008         PDBG("%s sq_num_entries %d, rq_num_entries %d "
1009              "qpid 0x%0x qhp %p dma_addr 0x%llx size %d\n",
1010              __func__, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
1011              qhp->wq.qpid, qhp, (unsigned long long) qhp->wq.dma_addr,
1012              1 << qhp->wq.size_log2);
1013         return &qhp->ibqp;
1014 }
1015
1016 static int iwch_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1017                       int attr_mask, struct ib_udata *udata)
1018 {
1019         struct iwch_dev *rhp;
1020         struct iwch_qp *qhp;
1021         enum iwch_qp_attr_mask mask = 0;
1022         struct iwch_qp_attributes attrs;
1023
1024         PDBG("%s ib_qp %p\n", __func__, ibqp);
1025
1026         /* iwarp does not support the RTR state */
1027         if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
1028                 attr_mask &= ~IB_QP_STATE;
1029
1030         /* Make sure we still have something left to do */
1031         if (!attr_mask)
1032                 return 0;
1033
1034         memset(&attrs, 0, sizeof attrs);
1035         qhp = to_iwch_qp(ibqp);
1036         rhp = qhp->rhp;
1037
1038         attrs.next_state = iwch_convert_state(attr->qp_state);
1039         attrs.enable_rdma_read = (attr->qp_access_flags &
1040                                IB_ACCESS_REMOTE_READ) ?  1 : 0;
1041         attrs.enable_rdma_write = (attr->qp_access_flags &
1042                                 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
1043         attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
1044
1045
1046         mask |= (attr_mask & IB_QP_STATE) ? IWCH_QP_ATTR_NEXT_STATE : 0;
1047         mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
1048                         (IWCH_QP_ATTR_ENABLE_RDMA_READ |
1049                          IWCH_QP_ATTR_ENABLE_RDMA_WRITE |
1050                          IWCH_QP_ATTR_ENABLE_RDMA_BIND) : 0;
1051
1052         return iwch_modify_qp(rhp, qhp, mask, &attrs, 0);
1053 }
1054
1055 void iwch_qp_add_ref(struct ib_qp *qp)
1056 {
1057         PDBG("%s ib_qp %p\n", __func__, qp);
1058         atomic_inc(&(to_iwch_qp(qp)->refcnt));
1059 }
1060
1061 void iwch_qp_rem_ref(struct ib_qp *qp)
1062 {
1063         PDBG("%s ib_qp %p\n", __func__, qp);
1064         if (atomic_dec_and_test(&(to_iwch_qp(qp)->refcnt)))
1065                 wake_up(&(to_iwch_qp(qp)->wait));
1066 }
1067
1068 static struct ib_qp *iwch_get_qp(struct ib_device *dev, int qpn)
1069 {
1070         PDBG("%s ib_dev %p qpn 0x%x\n", __func__, dev, qpn);
1071         return (struct ib_qp *)get_qhp(to_iwch_dev(dev), qpn);
1072 }
1073
1074
1075 static int iwch_query_pkey(struct ib_device *ibdev,
1076                            u8 port, u16 index, u16 * pkey)
1077 {
1078         PDBG("%s ibdev %p\n", __func__, ibdev);
1079         *pkey = 0;
1080         return 0;
1081 }
1082
1083 static int iwch_query_gid(struct ib_device *ibdev, u8 port,
1084                           int index, union ib_gid *gid)
1085 {
1086         struct iwch_dev *dev;
1087
1088         PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
1089                __func__, ibdev, port, index, gid);
1090         dev = to_iwch_dev(ibdev);
1091         BUG_ON(port == 0 || port > 2);
1092         memset(&(gid->raw[0]), 0, sizeof(gid->raw));
1093         memcpy(&(gid->raw[0]), dev->rdev.port_info.lldevs[port-1]->dev_addr, 6);
1094         return 0;
1095 }
1096
1097 static u64 fw_vers_string_to_u64(struct iwch_dev *iwch_dev)
1098 {
1099         struct ethtool_drvinfo info;
1100         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1101         char *cp, *next;
1102         unsigned fw_maj, fw_min, fw_mic;
1103
1104         rtnl_lock();
1105         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1106         rtnl_unlock();
1107
1108         next = info.fw_version + 1;
1109         cp = strsep(&next, ".");
1110         sscanf(cp, "%i", &fw_maj);
1111         cp = strsep(&next, ".");
1112         sscanf(cp, "%i", &fw_min);
1113         cp = strsep(&next, ".");
1114         sscanf(cp, "%i", &fw_mic);
1115
1116         return (((u64)fw_maj & 0xffff) << 32) | ((fw_min & 0xffff) << 16) |
1117                (fw_mic & 0xffff);
1118 }
1119
1120 static int iwch_query_device(struct ib_device *ibdev,
1121                              struct ib_device_attr *props)
1122 {
1123
1124         struct iwch_dev *dev;
1125         PDBG("%s ibdev %p\n", __func__, ibdev);
1126
1127         dev = to_iwch_dev(ibdev);
1128         memset(props, 0, sizeof *props);
1129         memcpy(&props->sys_image_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1130         props->hw_ver = dev->rdev.t3cdev_p->type;
1131         props->fw_ver = fw_vers_string_to_u64(dev);
1132         props->device_cap_flags = dev->device_cap_flags;
1133         props->page_size_cap = dev->attr.mem_pgsizes_bitmask;
1134         props->vendor_id = (u32)dev->rdev.rnic_info.pdev->vendor;
1135         props->vendor_part_id = (u32)dev->rdev.rnic_info.pdev->device;
1136         props->max_mr_size = dev->attr.max_mr_size;
1137         props->max_qp = dev->attr.max_qps;
1138         props->max_qp_wr = dev->attr.max_wrs;
1139         props->max_sge = dev->attr.max_sge_per_wr;
1140         props->max_sge_rd = 1;
1141         props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
1142         props->max_qp_init_rd_atom = dev->attr.max_rdma_reads_per_qp;
1143         props->max_cq = dev->attr.max_cqs;
1144         props->max_cqe = dev->attr.max_cqes_per_cq;
1145         props->max_mr = dev->attr.max_mem_regs;
1146         props->max_pd = dev->attr.max_pds;
1147         props->local_ca_ack_delay = 0;
1148         props->max_fast_reg_page_list_len = T3_MAX_FASTREG_DEPTH;
1149
1150         return 0;
1151 }
1152
1153 static int iwch_query_port(struct ib_device *ibdev,
1154                            u8 port, struct ib_port_attr *props)
1155 {
1156         PDBG("%s ibdev %p\n", __func__, ibdev);
1157         props->max_mtu = IB_MTU_4096;
1158         props->lid = 0;
1159         props->lmc = 0;
1160         props->sm_lid = 0;
1161         props->sm_sl = 0;
1162         props->state = IB_PORT_ACTIVE;
1163         props->phys_state = 0;
1164         props->port_cap_flags =
1165             IB_PORT_CM_SUP |
1166             IB_PORT_SNMP_TUNNEL_SUP |
1167             IB_PORT_REINIT_SUP |
1168             IB_PORT_DEVICE_MGMT_SUP |
1169             IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
1170         props->gid_tbl_len = 1;
1171         props->pkey_tbl_len = 1;
1172         props->qkey_viol_cntr = 0;
1173         props->active_width = 2;
1174         props->active_speed = 2;
1175         props->max_msg_sz = -1;
1176
1177         return 0;
1178 }
1179
1180 static ssize_t show_rev(struct device *dev, struct device_attribute *attr,
1181                         char *buf)
1182 {
1183         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1184                                                  ibdev.dev);
1185         PDBG("%s dev 0x%p\n", __func__, dev);
1186         return sprintf(buf, "%d\n", iwch_dev->rdev.t3cdev_p->type);
1187 }
1188
1189 static int fw_supports_fastreg(struct iwch_dev *iwch_dev)
1190 {
1191         struct ethtool_drvinfo info;
1192         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1193         char *cp, *next;
1194         unsigned fw_maj, fw_min;
1195
1196         rtnl_lock();
1197         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1198         rtnl_unlock();
1199
1200         next = info.fw_version+1;
1201         cp = strsep(&next, ".");
1202         sscanf(cp, "%i", &fw_maj);
1203         cp = strsep(&next, ".");
1204         sscanf(cp, "%i", &fw_min);
1205
1206         PDBG("%s maj %u min %u\n", __func__, fw_maj, fw_min);
1207
1208         return fw_maj > 6 || (fw_maj == 6 && fw_min > 0);
1209 }
1210
1211 static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, char *buf)
1212 {
1213         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1214                                                  ibdev.dev);
1215         struct ethtool_drvinfo info;
1216         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1217
1218         PDBG("%s dev 0x%p\n", __func__, dev);
1219         rtnl_lock();
1220         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1221         rtnl_unlock();
1222         return sprintf(buf, "%s\n", info.fw_version);
1223 }
1224
1225 static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
1226                         char *buf)
1227 {
1228         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1229                                                  ibdev.dev);
1230         struct ethtool_drvinfo info;
1231         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1232
1233         PDBG("%s dev 0x%p\n", __func__, dev);
1234         rtnl_lock();
1235         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1236         rtnl_unlock();
1237         return sprintf(buf, "%s\n", info.driver);
1238 }
1239
1240 static ssize_t show_board(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, "%x.%x\n", iwch_dev->rdev.rnic_info.pdev->vendor,
1247                        iwch_dev->rdev.rnic_info.pdev->device);
1248 }
1249
1250 static int iwch_get_mib(struct ib_device *ibdev,
1251                         union rdma_protocol_stats *stats)
1252 {
1253         struct iwch_dev *dev;
1254         struct tp_mib_stats m;
1255         int ret;
1256
1257         PDBG("%s ibdev %p\n", __func__, ibdev);
1258         dev = to_iwch_dev(ibdev);
1259         ret = dev->rdev.t3cdev_p->ctl(dev->rdev.t3cdev_p, RDMA_GET_MIB, &m);
1260         if (ret)
1261                 return -ENOSYS;
1262
1263         memset(stats, 0, sizeof *stats);
1264         stats->iw.ipInReceives = ((u64) m.ipInReceive_hi << 32) +
1265                                 m.ipInReceive_lo;
1266         stats->iw.ipInHdrErrors = ((u64) m.ipInHdrErrors_hi << 32) +
1267                                   m.ipInHdrErrors_lo;
1268         stats->iw.ipInAddrErrors = ((u64) m.ipInAddrErrors_hi << 32) +
1269                                    m.ipInAddrErrors_lo;
1270         stats->iw.ipInUnknownProtos = ((u64) m.ipInUnknownProtos_hi << 32) +
1271                                       m.ipInUnknownProtos_lo;
1272         stats->iw.ipInDiscards = ((u64) m.ipInDiscards_hi << 32) +
1273                                  m.ipInDiscards_lo;
1274         stats->iw.ipInDelivers = ((u64) m.ipInDelivers_hi << 32) +
1275                                  m.ipInDelivers_lo;
1276         stats->iw.ipOutRequests = ((u64) m.ipOutRequests_hi << 32) +
1277                                   m.ipOutRequests_lo;
1278         stats->iw.ipOutDiscards = ((u64) m.ipOutDiscards_hi << 32) +
1279                                   m.ipOutDiscards_lo;
1280         stats->iw.ipOutNoRoutes = ((u64) m.ipOutNoRoutes_hi << 32) +
1281                                   m.ipOutNoRoutes_lo;
1282         stats->iw.ipReasmTimeout = (u64) m.ipReasmTimeout;
1283         stats->iw.ipReasmReqds = (u64) m.ipReasmReqds;
1284         stats->iw.ipReasmOKs = (u64) m.ipReasmOKs;
1285         stats->iw.ipReasmFails = (u64) m.ipReasmFails;
1286         stats->iw.tcpActiveOpens = (u64) m.tcpActiveOpens;
1287         stats->iw.tcpPassiveOpens = (u64) m.tcpPassiveOpens;
1288         stats->iw.tcpAttemptFails = (u64) m.tcpAttemptFails;
1289         stats->iw.tcpEstabResets = (u64) m.tcpEstabResets;
1290         stats->iw.tcpOutRsts = (u64) m.tcpOutRsts;
1291         stats->iw.tcpCurrEstab = (u64) m.tcpCurrEstab;
1292         stats->iw.tcpInSegs = ((u64) m.tcpInSegs_hi << 32) +
1293                               m.tcpInSegs_lo;
1294         stats->iw.tcpOutSegs = ((u64) m.tcpOutSegs_hi << 32) +
1295                                m.tcpOutSegs_lo;
1296         stats->iw.tcpRetransSegs = ((u64) m.tcpRetransSeg_hi << 32) +
1297                                   m.tcpRetransSeg_lo;
1298         stats->iw.tcpInErrs = ((u64) m.tcpInErrs_hi << 32) +
1299                               m.tcpInErrs_lo;
1300         stats->iw.tcpRtoMin = (u64) m.tcpRtoMin;
1301         stats->iw.tcpRtoMax = (u64) m.tcpRtoMax;
1302         return 0;
1303 }
1304
1305 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1306 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1307 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1308 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1309
1310 static struct device_attribute *iwch_class_attributes[] = {
1311         &dev_attr_hw_rev,
1312         &dev_attr_fw_ver,
1313         &dev_attr_hca_type,
1314         &dev_attr_board_id,
1315 };
1316
1317 int iwch_register_device(struct iwch_dev *dev)
1318 {
1319         int ret;
1320         int i;
1321
1322         PDBG("%s iwch_dev %p\n", __func__, dev);
1323         strlcpy(dev->ibdev.name, "cxgb3_%d", IB_DEVICE_NAME_MAX);
1324         memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
1325         memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1326         dev->ibdev.owner = THIS_MODULE;
1327         dev->device_cap_flags = IB_DEVICE_ZERO_STAG |
1328                                 IB_DEVICE_MEM_WINDOW;
1329         if (fw_supports_fastreg(dev))
1330                 dev->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
1331
1332         dev->ibdev.uverbs_cmd_mask =
1333             (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1334             (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1335             (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1336             (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1337             (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1338             (1ull << IB_USER_VERBS_CMD_REG_MR) |
1339             (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1340             (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1341             (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1342             (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1343             (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1344             (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1345             (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1346             (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1347             (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1348             (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1349             (1ull << IB_USER_VERBS_CMD_POST_RECV);
1350         dev->ibdev.node_type = RDMA_NODE_RNIC;
1351         memcpy(dev->ibdev.node_desc, IWCH_NODE_DESC, sizeof(IWCH_NODE_DESC));
1352         dev->ibdev.phys_port_cnt = dev->rdev.port_info.nports;
1353         dev->ibdev.num_comp_vectors = 1;
1354         dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
1355         dev->ibdev.query_device = iwch_query_device;
1356         dev->ibdev.query_port = iwch_query_port;
1357         dev->ibdev.modify_port = iwch_modify_port;
1358         dev->ibdev.query_pkey = iwch_query_pkey;
1359         dev->ibdev.query_gid = iwch_query_gid;
1360         dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
1361         dev->ibdev.dealloc_ucontext = iwch_dealloc_ucontext;
1362         dev->ibdev.mmap = iwch_mmap;
1363         dev->ibdev.alloc_pd = iwch_allocate_pd;
1364         dev->ibdev.dealloc_pd = iwch_deallocate_pd;
1365         dev->ibdev.create_ah = iwch_ah_create;
1366         dev->ibdev.destroy_ah = iwch_ah_destroy;
1367         dev->ibdev.create_qp = iwch_create_qp;
1368         dev->ibdev.modify_qp = iwch_ib_modify_qp;
1369         dev->ibdev.destroy_qp = iwch_destroy_qp;
1370         dev->ibdev.create_cq = iwch_create_cq;
1371         dev->ibdev.destroy_cq = iwch_destroy_cq;
1372         dev->ibdev.resize_cq = iwch_resize_cq;
1373         dev->ibdev.poll_cq = iwch_poll_cq;
1374         dev->ibdev.get_dma_mr = iwch_get_dma_mr;
1375         dev->ibdev.reg_phys_mr = iwch_register_phys_mem;
1376         dev->ibdev.rereg_phys_mr = iwch_reregister_phys_mem;
1377         dev->ibdev.reg_user_mr = iwch_reg_user_mr;
1378         dev->ibdev.dereg_mr = iwch_dereg_mr;
1379         dev->ibdev.alloc_mw = iwch_alloc_mw;
1380         dev->ibdev.bind_mw = iwch_bind_mw;
1381         dev->ibdev.dealloc_mw = iwch_dealloc_mw;
1382         dev->ibdev.alloc_fast_reg_mr = iwch_alloc_fast_reg_mr;
1383         dev->ibdev.alloc_fast_reg_page_list = iwch_alloc_fastreg_pbl;
1384         dev->ibdev.free_fast_reg_page_list = iwch_free_fastreg_pbl;
1385         dev->ibdev.attach_mcast = iwch_multicast_attach;
1386         dev->ibdev.detach_mcast = iwch_multicast_detach;
1387         dev->ibdev.process_mad = iwch_process_mad;
1388         dev->ibdev.req_notify_cq = iwch_arm_cq;
1389         dev->ibdev.post_send = iwch_post_send;
1390         dev->ibdev.post_recv = iwch_post_receive;
1391         dev->ibdev.get_protocol_stats = iwch_get_mib;
1392
1393         dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
1394         if (!dev->ibdev.iwcm)
1395                 return -ENOMEM;
1396
1397         dev->ibdev.iwcm->connect = iwch_connect;
1398         dev->ibdev.iwcm->accept = iwch_accept_cr;
1399         dev->ibdev.iwcm->reject = iwch_reject_cr;
1400         dev->ibdev.iwcm->create_listen = iwch_create_listen;
1401         dev->ibdev.iwcm->destroy_listen = iwch_destroy_listen;
1402         dev->ibdev.iwcm->add_ref = iwch_qp_add_ref;
1403         dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
1404         dev->ibdev.iwcm->get_qp = iwch_get_qp;
1405
1406         ret = ib_register_device(&dev->ibdev);
1407         if (ret)
1408                 goto bail1;
1409
1410         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i) {
1411                 ret = device_create_file(&dev->ibdev.dev,
1412                                          iwch_class_attributes[i]);
1413                 if (ret) {
1414                         goto bail2;
1415                 }
1416         }
1417         return 0;
1418 bail2:
1419         ib_unregister_device(&dev->ibdev);
1420 bail1:
1421         return ret;
1422 }
1423
1424 void iwch_unregister_device(struct iwch_dev *dev)
1425 {
1426         int i;
1427
1428         PDBG("%s iwch_dev %p\n", __func__, dev);
1429         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i)
1430                 device_remove_file(&dev->ibdev.dev,
1431                                    iwch_class_attributes[i]);
1432         ib_unregister_device(&dev->ibdev);
1433         return;
1434 }