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