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                             const struct ib_wc *in_wc,
  89                             const struct ib_grh *in_grh,
  90                             const struct ib_mad_hdr *in_mad,
  91                             size_t in_mad_size,
  92                             struct ib_mad_hdr *out_mad,
  93                             size_t *out_mad_size,
  94                             u16 *out_mad_pkey_index)
  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,
 146                                     const struct ib_cq_init_attr *attr,
 147                                     struct ib_ucontext *ib_context,
 148                                     struct ib_udata *udata)
 149 {
 150         int entries = attr->cqe;
 151         struct iwch_dev *rhp;
 152         struct iwch_cq *chp;
 153         struct iwch_create_cq_resp uresp;
 154         struct iwch_create_cq_req ureq;
 155         struct iwch_ucontext *ucontext = NULL;
 156         static int warned;
 157         size_t resplen;
 158
 159         PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
 160         if (attr->flags)
 161                 return ERR_PTR(-EINVAL);
 162
 163         rhp = to_iwch_dev(ibdev);
 164         chp = kzalloc(sizeof(*chp), GFP_KERNEL);
 165         if (!chp)
 166                 return ERR_PTR(-ENOMEM);
 167
 168         if (ib_context) {
 169                 ucontext = to_iwch_ucontext(ib_context);
 170                 if (!t3a_device(rhp)) {
 171                         if (ib_copy_from_udata(&ureq, udata, sizeof (ureq))) {
 172                                 kfree(chp);
 173                                 return ERR_PTR(-EFAULT);
 174                         }
 175                         chp->user_rptr_addr = (u32 __user *)(unsigned long)ureq.user_rptr_addr;
 176                 }
 177         }
 178
 179         if (t3a_device(rhp)) {
 180
 181                 /*
 182                  * T3A: Add some fluff to handle extra CQEs inserted
 183                  * for various errors.
 184                  * Additional CQE possibilities:
 185                  *      TERMINATE,
 186                  *      incoming RDMA WRITE Failures
 187                  *      incoming RDMA READ REQUEST FAILUREs
 188                  * NOTE: We cannot ensure the CQ won't overflow.
 189                  */
 190                 entries += 16;
 191         }
 192         entries = roundup_pow_of_two(entries);
 193         chp->cq.size_log2 = ilog2(entries);
 194
 195         if (cxio_create_cq(&rhp->rdev, &chp->cq, !ucontext)) {
 196                 kfree(chp);
 197                 return ERR_PTR(-ENOMEM);
 198         }
 199         chp->rhp = rhp;
 200         chp->ibcq.cqe = 1 << chp->cq.size_log2;
 201         spin_lock_init(&chp->lock);
 202         spin_lock_init(&chp->comp_handler_lock);
 203         atomic_set(&chp->refcnt, 1);
 204         init_waitqueue_head(&chp->wait);
 205         if (insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid)) {
 206                 cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
 207                 kfree(chp);
 208                 return ERR_PTR(-ENOMEM);
 209         }
 210
 211         if (ucontext) {
 212                 struct iwch_mm_entry *mm;
 213
 214                 mm = kmalloc(sizeof *mm, GFP_KERNEL);
 215                 if (!mm) {
 216                         iwch_destroy_cq(&chp->ibcq);
 217                         return ERR_PTR(-ENOMEM);
 218                 }
 219                 uresp.cqid = chp->cq.cqid;
 220                 uresp.size_log2 = chp->cq.size_log2;
 221                 spin_lock(&ucontext->mmap_lock);
 222                 uresp.key = ucontext->key;
 223                 ucontext->key += PAGE_SIZE;
 224                 spin_unlock(&ucontext->mmap_lock);
 225                 mm->key = uresp.key;
 226                 mm->addr = virt_to_phys(chp->cq.queue);
 227                 if (udata->outlen < sizeof uresp) {
 228                         if (!warned++)
 229                                 printk(KERN_WARNING MOD "Warning - "
 230                                        "downlevel libcxgb3 (non-fatal).\n");
 231                         mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
 232                                              sizeof(struct t3_cqe));
 233                         resplen = sizeof(struct iwch_create_cq_resp_v0);
 234                 } else {
 235                         mm->len = PAGE_ALIGN(((1UL << uresp.size_log2) + 1) *
 236                                              sizeof(struct t3_cqe));
 237                         uresp.memsize = mm->len;
 238                         uresp.reserved = 0;
 239                         resplen = sizeof uresp;
 240                 }
 241                 if (ib_copy_to_udata(udata, &uresp, resplen)) {
 242                         kfree(mm);
 243                         iwch_destroy_cq(&chp->ibcq);
 244                         return ERR_PTR(-EFAULT);
 245                 }
 246                 insert_mmap(ucontext, mm);
 247         }
 248         PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
 249              chp->cq.cqid, chp, (1 << chp->cq.size_log2),
 250              (unsigned long long) chp->cq.dma_addr);
 251         return &chp->ibcq;
 252 }
 253
 254 static int iwch_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
 255 {
 256 #ifdef notyet
 257         struct iwch_cq *chp = to_iwch_cq(cq);
 258         struct t3_cq oldcq, newcq;
 259         int ret;
 260
 261         PDBG("%s ib_cq %p cqe %d\n", __func__, cq, cqe);
 262
 263         /* We don't downsize... */
 264         if (cqe <= cq->cqe)
 265                 return 0;
 266
 267         /* create new t3_cq with new size */
 268         cqe = roundup_pow_of_two(cqe+1);
 269         newcq.size_log2 = ilog2(cqe);
 270
 271         /* Dont allow resize to less than the current wce count */
 272         if (cqe < Q_COUNT(chp->cq.rptr, chp->cq.wptr)) {
 273                 return -ENOMEM;
 274         }
 275
 276         /* Quiesce all QPs using this CQ */
 277         ret = iwch_quiesce_qps(chp);
 278         if (ret) {
 279                 return ret;
 280         }
 281
 282         ret = cxio_create_cq(&chp->rhp->rdev, &newcq);
 283         if (ret) {
 284                 return ret;
 285         }
 286
 287         /* copy CQEs */
 288         memcpy(newcq.queue, chp->cq.queue, (1 << chp->cq.size_log2) *
 289                                         sizeof(struct t3_cqe));
 290
 291         /* old iwch_qp gets new t3_cq but keeps old cqid */
 292         oldcq = chp->cq;
 293         chp->cq = newcq;
 294         chp->cq.cqid = oldcq.cqid;
 295
 296         /* resize new t3_cq to update the HW context */
 297         ret = cxio_resize_cq(&chp->rhp->rdev, &chp->cq);
 298         if (ret) {
 299                 chp->cq = oldcq;
 300                 return ret;
 301         }
 302         chp->ibcq.cqe = (1<<chp->cq.size_log2) - 1;
 303
 304         /* destroy old t3_cq */
 305         oldcq.cqid = newcq.cqid;
 306         ret = cxio_destroy_cq(&chp->rhp->rdev, &oldcq);
 307         if (ret) {
 308                 printk(KERN_ERR MOD "%s - cxio_destroy_cq failed %d\n",
 309                         __func__, ret);
 310         }
 311
 312         /* add user hooks here */
 313
 314         /* resume qps */
 315         ret = iwch_resume_qps(chp);
 316         return ret;
 317 #else
 318         return -ENOSYS;
 319 #endif
 320 }
 321
 322 static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
 323 {
 324         struct iwch_dev *rhp;
 325         struct iwch_cq *chp;
 326         enum t3_cq_opcode cq_op;
 327         int err;
 328         unsigned long flag;
 329         u32 rptr;
 330
 331         chp = to_iwch_cq(ibcq);
 332         rhp = chp->rhp;
 333         if ((flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED)
 334                 cq_op = CQ_ARM_SE;
 335         else
 336                 cq_op = CQ_ARM_AN;
 337         if (chp->user_rptr_addr) {
 338                 if (get_user(rptr, chp->user_rptr_addr))
 339                         return -EFAULT;
 340                 spin_lock_irqsave(&chp->lock, flag);
 341                 chp->cq.rptr = rptr;
 342         } else
 343                 spin_lock_irqsave(&chp->lock, flag);
 344         PDBG("%s rptr 0x%x\n", __func__, chp->cq.rptr);
 345         err = cxio_hal_cq_op(&rhp->rdev, &chp->cq, cq_op, 0);
 346         spin_unlock_irqrestore(&chp->lock, flag);
 347         if (err < 0)
 348                 printk(KERN_ERR MOD "Error %d rearming CQID 0x%x\n", err,
 349                        chp->cq.cqid);
 350         if (err > 0 && !(flags & IB_CQ_REPORT_MISSED_EVENTS))
 351                 err = 0;
 352         return err;
 353 }
 354
 355 static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 356 {
 357         int len = vma->vm_end - vma->vm_start;
 358         u32 key = vma->vm_pgoff << PAGE_SHIFT;
 359         struct cxio_rdev *rdev_p;
 360         int ret = 0;
 361         struct iwch_mm_entry *mm;
 362         struct iwch_ucontext *ucontext;
 363         u64 addr;
 364
 365         PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __func__, vma->vm_pgoff,
 366              key, len);
 367
 368         if (vma->vm_start & (PAGE_SIZE-1)) {
 369                 return -EINVAL;
 370         }
 371
 372         rdev_p = &(to_iwch_dev(context->device)->rdev);
 373         ucontext = to_iwch_ucontext(context);
 374
 375         mm = remove_mmap(ucontext, key, len);
 376         if (!mm)
 377                 return -EINVAL;
 378         addr = mm->addr;
 379         kfree(mm);
 380
 381         if ((addr >= rdev_p->rnic_info.udbell_physbase) &&
 382             (addr < (rdev_p->rnic_info.udbell_physbase +
 383                        rdev_p->rnic_info.udbell_len))) {
 384
 385                 /*
 386                  * Map T3 DB register.
 387                  */
 388                 if (vma->vm_flags & VM_READ) {
 389                         return -EPERM;
 390                 }
 391
 392                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 393                 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
 394                 vma->vm_flags &= ~VM_MAYREAD;
 395                 ret = io_remap_pfn_range(vma, vma->vm_start,
 396                                          addr >> PAGE_SHIFT,
 397                                          len, vma->vm_page_prot);
 398         } else {
 399
 400                 /*
 401                  * Map WQ or CQ contig dma memory...
 402                  */
 403                 ret = remap_pfn_range(vma, vma->vm_start,
 404                                       addr >> PAGE_SHIFT,
 405                                       len, vma->vm_page_prot);
 406         }
 407
 408         return ret;
 409 }
 410
 411 static int iwch_deallocate_pd(struct ib_pd *pd)
 412 {
 413         struct iwch_dev *rhp;
 414         struct iwch_pd *php;
 415
 416         php = to_iwch_pd(pd);
 417         rhp = php->rhp;
 418         PDBG("%s ibpd %p pdid 0x%x\n", __func__, pd, php->pdid);
 419         cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid);
 420         kfree(php);
 421         return 0;
 422 }
 423
 424 static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev,
 425                                struct ib_ucontext *context,
 426                                struct ib_udata *udata)
 427 {
 428         struct iwch_pd *php;
 429         u32 pdid;
 430         struct iwch_dev *rhp;
 431
 432         PDBG("%s ibdev %p\n", __func__, ibdev);
 433         rhp = (struct iwch_dev *) ibdev;
 434         pdid = cxio_hal_get_pdid(rhp->rdev.rscp);
 435         if (!pdid)
 436                 return ERR_PTR(-EINVAL);
 437         php = kzalloc(sizeof(*php), GFP_KERNEL);
 438         if (!php) {
 439                 cxio_hal_put_pdid(rhp->rdev.rscp, pdid);
 440                 return ERR_PTR(-ENOMEM);
 441         }
 442         php->pdid = pdid;
 443         php->rhp = rhp;
 444         if (context) {
 445                 if (ib_copy_to_udata(udata, &php->pdid, sizeof (__u32))) {
 446                         iwch_deallocate_pd(&php->ibpd);
 447                         return ERR_PTR(-EFAULT);
 448                 }
 449         }
 450         PDBG("%s pdid 0x%0x ptr 0x%p\n", __func__, pdid, php);
 451         return &php->ibpd;
 452 }
 453
 454 static int iwch_dereg_mr(struct ib_mr *ib_mr)
 455 {
 456         struct iwch_dev *rhp;
 457         struct iwch_mr *mhp;
 458         u32 mmid;
 459
 460         PDBG("%s ib_mr %p\n", __func__, ib_mr);
 461         /* There can be no memory windows */
 462         if (atomic_read(&ib_mr->usecnt))
 463                 return -EINVAL;
 464
 465         mhp = to_iwch_mr(ib_mr);
 466         rhp = mhp->rhp;
 467         mmid = mhp->attr.stag >> 8;
 468         cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
 469                        mhp->attr.pbl_addr);
 470         iwch_free_pbl(mhp);
 471         remove_handle(rhp, &rhp->mmidr, mmid);
 472         if (mhp->kva)
 473                 kfree((void *) (unsigned long) mhp->kva);
 474         if (mhp->umem)
 475                 ib_umem_release(mhp->umem);
 476         PDBG("%s mmid 0x%x ptr %p\n", __func__, mmid, mhp);
 477         kfree(mhp);
 478         return 0;
 479 }
 480
 481 static struct ib_mr *iwch_register_phys_mem(struct ib_pd *pd,
 482                                         struct ib_phys_buf *buffer_list,
 483                                         int num_phys_buf,
 484                                         int acc,
 485                                         u64 *iova_start)
 486 {
 487         __be64 *page_list;
 488         int shift;
 489         u64 total_size;
 490         int npages;
 491         struct iwch_dev *rhp;
 492         struct iwch_pd *php;
 493         struct iwch_mr *mhp;
 494         int ret;
 495
 496         PDBG("%s ib_pd %p\n", __func__, pd);
 497         php = to_iwch_pd(pd);
 498         rhp = php->rhp;
 499
 500         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 501         if (!mhp)
 502                 return ERR_PTR(-ENOMEM);
 503
 504         mhp->rhp = rhp;
 505
 506         /* First check that we have enough alignment */
 507         if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
 508                 ret = -EINVAL;
 509                 goto err;
 510         }
 511
 512         if (num_phys_buf > 1 &&
 513             ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
 514                 ret = -EINVAL;
 515                 goto err;
 516         }
 517
 518         ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
 519                                    &total_size, &npages, &shift, &page_list);
 520         if (ret)
 521                 goto err;
 522
 523         ret = iwch_alloc_pbl(mhp, npages);
 524         if (ret) {
 525                 kfree(page_list);
 526                 goto err_pbl;
 527         }
 528
 529         ret = iwch_write_pbl(mhp, page_list, npages, 0);
 530         kfree(page_list);
 531         if (ret)
 532                 goto err_pbl;
 533
 534         mhp->attr.pdid = php->pdid;
 535         mhp->attr.zbva = 0;
 536
 537         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 538         mhp->attr.va_fbo = *iova_start;
 539         mhp->attr.page_size = shift - 12;
 540
 541         mhp->attr.len = (u32) total_size;
 542         mhp->attr.pbl_size = npages;
 543         ret = iwch_register_mem(rhp, php, mhp, shift);
 544         if (ret)
 545                 goto err_pbl;
 546
 547         return &mhp->ibmr;
 548
 549 err_pbl:
 550         iwch_free_pbl(mhp);
 551
 552 err:
 553         kfree(mhp);
 554         return ERR_PTR(ret);
 555
 556 }
 557
 558 static int iwch_reregister_phys_mem(struct ib_mr *mr,
 559                                      int mr_rereg_mask,
 560                                      struct ib_pd *pd,
 561                                      struct ib_phys_buf *buffer_list,
 562                                      int num_phys_buf,
 563                                      int acc, u64 * iova_start)
 564 {
 565
 566         struct iwch_mr mh, *mhp;
 567         struct iwch_pd *php;
 568         struct iwch_dev *rhp;
 569         __be64 *page_list = NULL;
 570         int shift = 0;
 571         u64 total_size;
 572         int npages = 0;
 573         int ret;
 574
 575         PDBG("%s ib_mr %p ib_pd %p\n", __func__, mr, pd);
 576
 577         /* There can be no memory windows */
 578         if (atomic_read(&mr->usecnt))
 579                 return -EINVAL;
 580
 581         mhp = to_iwch_mr(mr);
 582         rhp = mhp->rhp;
 583         php = to_iwch_pd(mr->pd);
 584
 585         /* make sure we are on the same adapter */
 586         if (rhp != php->rhp)
 587                 return -EINVAL;
 588
 589         memcpy(&mh, mhp, sizeof *mhp);
 590
 591         if (mr_rereg_mask & IB_MR_REREG_PD)
 592                 php = to_iwch_pd(pd);
 593         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 594                 mh.attr.perms = iwch_ib_to_tpt_access(acc);
 595         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 596                 ret = build_phys_page_list(buffer_list, num_phys_buf,
 597                                            iova_start,
 598                                            &total_size, &npages,
 599                                            &shift, &page_list);
 600                 if (ret)
 601                         return ret;
 602         }
 603
 604         ret = iwch_reregister_mem(rhp, php, &mh, shift, npages);
 605         kfree(page_list);
 606         if (ret) {
 607                 return ret;
 608         }
 609         if (mr_rereg_mask & IB_MR_REREG_PD)
 610                 mhp->attr.pdid = php->pdid;
 611         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 612                 mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 613         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 614                 mhp->attr.zbva = 0;
 615                 mhp->attr.va_fbo = *iova_start;
 616                 mhp->attr.page_size = shift - 12;
 617                 mhp->attr.len = (u32) total_size;
 618                 mhp->attr.pbl_size = npages;
 619         }
 620
 621         return 0;
 622 }
 623
 624
 625 static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 626                                       u64 virt, int acc, struct ib_udata *udata)
 627 {
 628         __be64 *pages;
 629         int shift, n, len;
 630         int i, k, entry;
 631         int err = 0;
 632         struct iwch_dev *rhp;
 633         struct iwch_pd *php;
 634         struct iwch_mr *mhp;
 635         struct iwch_reg_user_mr_resp uresp;
 636         struct scatterlist *sg;
 637         PDBG("%s ib_pd %p\n", __func__, pd);
 638
 639         php = to_iwch_pd(pd);
 640         rhp = php->rhp;
 641         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 642         if (!mhp)
 643                 return ERR_PTR(-ENOMEM);
 644
 645         mhp->rhp = rhp;
 646
 647         mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
 648         if (IS_ERR(mhp->umem)) {
 649                 err = PTR_ERR(mhp->umem);
 650                 kfree(mhp);
 651                 return ERR_PTR(err);
 652         }
 653
 654         shift = ffs(mhp->umem->page_size) - 1;
 655
 656         n = mhp->umem->nmap;
 657
 658         err = iwch_alloc_pbl(mhp, n);
 659         if (err)
 660                 goto err;
 661
 662         pages = (__be64 *) __get_free_page(GFP_KERNEL);
 663         if (!pages) {
 664                 err = -ENOMEM;
 665                 goto err_pbl;
 666         }
 667
 668         i = n = 0;
 669
 670         for_each_sg(mhp->umem->sg_head.sgl, sg, mhp->umem->nmap, entry) {
 671                         len = sg_dma_len(sg) >> shift;
 672                         for (k = 0; k < len; ++k) {
 673                                 pages[i++] = cpu_to_be64(sg_dma_address(sg) +
 674                                         mhp->umem->page_size * k);
 675                                 if (i == PAGE_SIZE / sizeof *pages) {
 676                                         err = iwch_write_pbl(mhp, pages, i, n);
 677                                         if (err)
 678                                                 goto pbl_done;
 679                                         n += i;
 680                                         i = 0;
 681                                 }
 682                         }
 683         }
 684
 685         if (i)
 686                 err = iwch_write_pbl(mhp, pages, i, n);
 687
 688 pbl_done:
 689         free_page((unsigned long) pages);
 690         if (err)
 691                 goto err_pbl;
 692
 693         mhp->attr.pdid = php->pdid;
 694         mhp->attr.zbva = 0;
 695         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 696         mhp->attr.va_fbo = virt;
 697         mhp->attr.page_size = shift - 12;
 698         mhp->attr.len = (u32) length;
 699
 700         err = iwch_register_mem(rhp, php, mhp, shift);
 701         if (err)
 702                 goto err_pbl;
 703
 704         if (udata && !t3a_device(rhp)) {
 705                 uresp.pbl_addr = (mhp->attr.pbl_addr -
 706                                  rhp->rdev.rnic_info.pbl_base) >> 3;
 707                 PDBG("%s user resp pbl_addr 0x%x\n", __func__,
 708                      uresp.pbl_addr);
 709
 710                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 711                         iwch_dereg_mr(&mhp->ibmr);
 712                         err = -EFAULT;
 713                         goto err;
 714                 }
 715         }
 716
 717         return &mhp->ibmr;
 718
 719 err_pbl:
 720         iwch_free_pbl(mhp);
 721
 722 err:
 723         ib_umem_release(mhp->umem);
 724         kfree(mhp);
 725         return ERR_PTR(err);
 726 }
 727
 728 static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
 729 {
 730         struct ib_phys_buf bl;
 731         u64 kva;
 732         struct ib_mr *ibmr;
 733
 734         PDBG("%s ib_pd %p\n", __func__, pd);
 735
 736         /*
 737          * T3 only supports 32 bits of size.
 738          */
 739         if (sizeof(phys_addr_t) > 4) {
 740                 pr_warn_once(MOD "Cannot support dma_mrs on this platform.\n");
 741                 return ERR_PTR(-ENOTSUPP);
 742         }
 743         bl.size = 0xffffffff;
 744         bl.addr = 0;
 745         kva = 0;
 746         ibmr = iwch_register_phys_mem(pd, &bl, 1, acc, &kva);
 747         return ibmr;
 748 }
 749
 750 static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
 751 {
 752         struct iwch_dev *rhp;
 753         struct iwch_pd *php;
 754         struct iwch_mw *mhp;
 755         u32 mmid;
 756         u32 stag = 0;
 757         int ret;
 758
 759         if (type != IB_MW_TYPE_1)
 760                 return ERR_PTR(-EINVAL);
 761
 762         php = to_iwch_pd(pd);
 763         rhp = php->rhp;
 764         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 765         if (!mhp)
 766                 return ERR_PTR(-ENOMEM);
 767         ret = cxio_allocate_window(&rhp->rdev, &stag, php->pdid);
 768         if (ret) {
 769                 kfree(mhp);
 770                 return ERR_PTR(ret);
 771         }
 772         mhp->rhp = rhp;
 773         mhp->attr.pdid = php->pdid;
 774         mhp->attr.type = TPT_MW;
 775         mhp->attr.stag = stag;
 776         mmid = (stag) >> 8;
 777         mhp->ibmw.rkey = stag;
 778         if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) {
 779                 cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 780                 kfree(mhp);
 781                 return ERR_PTR(-ENOMEM);
 782         }
 783         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 784         return &(mhp->ibmw);
 785 }
 786
 787 static int iwch_dealloc_mw(struct ib_mw *mw)
 788 {
 789         struct iwch_dev *rhp;
 790         struct iwch_mw *mhp;
 791         u32 mmid;
 792
 793         mhp = to_iwch_mw(mw);
 794         rhp = mhp->rhp;
 795         mmid = (mw->rkey) >> 8;
 796         cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 797         remove_handle(rhp, &rhp->mmidr, mmid);
 798         PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
 799         kfree(mhp);
 800         return 0;
 801 }
 802
 803 static struct ib_mr *iwch_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth)
 804 {
 805         struct iwch_dev *rhp;
 806         struct iwch_pd *php;
 807         struct iwch_mr *mhp;
 808         u32 mmid;
 809         u32 stag = 0;
 810         int ret = 0;
 811
 812         php = to_iwch_pd(pd);
 813         rhp = php->rhp;
 814         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 815         if (!mhp)
 816                 goto err;
 817
 818         mhp->rhp = rhp;
 819         ret = iwch_alloc_pbl(mhp, pbl_depth);
 820         if (ret)
 821                 goto err1;
 822         mhp->attr.pbl_size = pbl_depth;
 823         ret = cxio_allocate_stag(&rhp->rdev, &stag, php->pdid,
 824                                  mhp->attr.pbl_size, mhp->attr.pbl_addr);
 825         if (ret)
 826                 goto err2;
 827         mhp->attr.pdid = php->pdid;
 828         mhp->attr.type = TPT_NON_SHARED_MR;
 829         mhp->attr.stag = stag;
 830         mhp->attr.state = 1;
 831         mmid = (stag) >> 8;
 832         mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
 833         if (insert_handle(rhp, &rhp->mmidr, mhp, mmid))
 834                 goto err3;
 835
 836         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 837         return &(mhp->ibmr);
 838 err3:
 839         cxio_dereg_mem(&rhp->rdev, stag, mhp->attr.pbl_size,
 840                        mhp->attr.pbl_addr);
 841 err2:
 842         iwch_free_pbl(mhp);
 843 err1:
 844         kfree(mhp);
 845 err:
 846         return ERR_PTR(ret);
 847 }
 848
 849 static struct ib_fast_reg_page_list *iwch_alloc_fastreg_pbl(
 850                                         struct ib_device *device,
 851                                         int page_list_len)
 852 {
 853         struct ib_fast_reg_page_list *page_list;
 854
 855         page_list = kmalloc(sizeof *page_list + page_list_len * sizeof(u64),
 856                             GFP_KERNEL);
 857         if (!page_list)
 858                 return ERR_PTR(-ENOMEM);
 859
 860         page_list->page_list = (u64 *)(page_list + 1);
 861         page_list->max_page_list_len = page_list_len;
 862
 863         return page_list;
 864 }
 865
 866 static void iwch_free_fastreg_pbl(struct ib_fast_reg_page_list *page_list)
 867 {
 868         kfree(page_list);
 869 }
 870
 871 static int iwch_destroy_qp(struct ib_qp *ib_qp)
 872 {
 873         struct iwch_dev *rhp;
 874         struct iwch_qp *qhp;
 875         struct iwch_qp_attributes attrs;
 876         struct iwch_ucontext *ucontext;
 877
 878         qhp = to_iwch_qp(ib_qp);
 879         rhp = qhp->rhp;
 880
 881         attrs.next_state = IWCH_QP_STATE_ERROR;
 882         iwch_modify_qp(rhp, qhp, IWCH_QP_ATTR_NEXT_STATE, &attrs, 0);
 883         wait_event(qhp->wait, !qhp->ep);
 884
 885         remove_handle(rhp, &rhp->qpidr, qhp->wq.qpid);
 886
 887         atomic_dec(&qhp->refcnt);
 888         wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
 889
 890         ucontext = ib_qp->uobject ? to_iwch_ucontext(ib_qp->uobject->context)
 891                                   : NULL;
 892         cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 893                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 894
 895         PDBG("%s ib_qp %p qpid 0x%0x qhp %p\n", __func__,
 896              ib_qp, qhp->wq.qpid, qhp);
 897         kfree(qhp);
 898         return 0;
 899 }
 900
 901 static struct ib_qp *iwch_create_qp(struct ib_pd *pd,
 902                              struct ib_qp_init_attr *attrs,
 903                              struct ib_udata *udata)
 904 {
 905         struct iwch_dev *rhp;
 906         struct iwch_qp *qhp;
 907         struct iwch_pd *php;
 908         struct iwch_cq *schp;
 909         struct iwch_cq *rchp;
 910         struct iwch_create_qp_resp uresp;
 911         int wqsize, sqsize, rqsize;
 912         struct iwch_ucontext *ucontext;
 913
 914         PDBG("%s ib_pd %p\n", __func__, pd);
 915         if (attrs->qp_type != IB_QPT_RC)
 916                 return ERR_PTR(-EINVAL);
 917         php = to_iwch_pd(pd);
 918         rhp = php->rhp;
 919         schp = get_chp(rhp, ((struct iwch_cq *) attrs->send_cq)->cq.cqid);
 920         rchp = get_chp(rhp, ((struct iwch_cq *) attrs->recv_cq)->cq.cqid);
 921         if (!schp || !rchp)
 922                 return ERR_PTR(-EINVAL);
 923
 924         /* The RQT size must be # of entries + 1 rounded up to a power of two */
 925         rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr);
 926         if (rqsize == attrs->cap.max_recv_wr)
 927                 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr+1);
 928
 929         /* T3 doesn't support RQT depth < 16 */
 930         if (rqsize < 16)
 931                 rqsize = 16;
 932
 933         if (rqsize > T3_MAX_RQ_SIZE)
 934                 return ERR_PTR(-EINVAL);
 935
 936         if (attrs->cap.max_inline_data > T3_MAX_INLINE)
 937                 return ERR_PTR(-EINVAL);
 938
 939         /*
 940          * NOTE: The SQ and total WQ sizes don't need to be
 941          * a power of two.  However, all the code assumes
 942          * they are. EG: Q_FREECNT() and friends.
 943          */
 944         sqsize = roundup_pow_of_two(attrs->cap.max_send_wr);
 945         wqsize = roundup_pow_of_two(rqsize + sqsize);
 946
 947         /*
 948          * Kernel users need more wq space for fastreg WRs which can take
 949          * 2 WR fragments.
 950          */
 951         ucontext = pd->uobject ? to_iwch_ucontext(pd->uobject->context) : NULL;
 952         if (!ucontext && wqsize < (rqsize + (2 * sqsize)))
 953                 wqsize = roundup_pow_of_two(rqsize +
 954                                 roundup_pow_of_two(attrs->cap.max_send_wr * 2));
 955         PDBG("%s wqsize %d sqsize %d rqsize %d\n", __func__,
 956              wqsize, sqsize, rqsize);
 957         qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
 958         if (!qhp)
 959                 return ERR_PTR(-ENOMEM);
 960         qhp->wq.size_log2 = ilog2(wqsize);
 961         qhp->wq.rq_size_log2 = ilog2(rqsize);
 962         qhp->wq.sq_size_log2 = ilog2(sqsize);
 963         if (cxio_create_qp(&rhp->rdev, !udata, &qhp->wq,
 964                            ucontext ? &ucontext->uctx : &rhp->rdev.uctx)) {
 965                 kfree(qhp);
 966                 return ERR_PTR(-ENOMEM);
 967         }
 968
 969         attrs->cap.max_recv_wr = rqsize - 1;
 970         attrs->cap.max_send_wr = sqsize;
 971         attrs->cap.max_inline_data = T3_MAX_INLINE;
 972
 973         qhp->rhp = rhp;
 974         qhp->attr.pd = php->pdid;
 975         qhp->attr.scq = ((struct iwch_cq *) attrs->send_cq)->cq.cqid;
 976         qhp->attr.rcq = ((struct iwch_cq *) attrs->recv_cq)->cq.cqid;
 977         qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
 978         qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
 979         qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
 980         qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
 981         qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
 982         qhp->attr.state = IWCH_QP_STATE_IDLE;
 983         qhp->attr.next_state = IWCH_QP_STATE_IDLE;
 984
 985         /*
 986          * XXX - These don't get passed in from the openib user
 987          * at create time.  The CM sets them via a QP modify.
 988          * Need to fix...  I think the CM should
 989          */
 990         qhp->attr.enable_rdma_read = 1;
 991         qhp->attr.enable_rdma_write = 1;
 992         qhp->attr.enable_bind = 1;
 993         qhp->attr.max_ord = 1;
 994         qhp->attr.max_ird = 1;
 995
 996         spin_lock_init(&qhp->lock);
 997         init_waitqueue_head(&qhp->wait);
 998         atomic_set(&qhp->refcnt, 1);
 999
1000         if (insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid)) {
1001                 cxio_destroy_qp(&rhp->rdev, &qhp->wq,
1002                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
1003                 kfree(qhp);
1004                 return ERR_PTR(-ENOMEM);
1005         }
1006
1007         if (udata) {
1008
1009                 struct iwch_mm_entry *mm1, *mm2;
1010
1011                 mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
1012                 if (!mm1) {
1013                         iwch_destroy_qp(&qhp->ibqp);
1014                         return ERR_PTR(-ENOMEM);
1015                 }
1016
1017                 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
1018                 if (!mm2) {
1019                         kfree(mm1);
1020                         iwch_destroy_qp(&qhp->ibqp);
1021                         return ERR_PTR(-ENOMEM);
1022                 }
1023
1024                 uresp.qpid = qhp->wq.qpid;
1025                 uresp.size_log2 = qhp->wq.size_log2;
1026                 uresp.sq_size_log2 = qhp->wq.sq_size_log2;
1027                 uresp.rq_size_log2 = qhp->wq.rq_size_log2;
1028                 spin_lock(&ucontext->mmap_lock);
1029                 uresp.key = ucontext->key;
1030                 ucontext->key += PAGE_SIZE;
1031                 uresp.db_key = ucontext->key;
1032                 ucontext->key += PAGE_SIZE;
1033                 spin_unlock(&ucontext->mmap_lock);
1034                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
1035                         kfree(mm1);
1036                         kfree(mm2);
1037                         iwch_destroy_qp(&qhp->ibqp);
1038                         return ERR_PTR(-EFAULT);
1039                 }
1040                 mm1->key = uresp.key;
1041                 mm1->addr = virt_to_phys(qhp->wq.queue);
1042                 mm1->len = PAGE_ALIGN(wqsize * sizeof (union t3_wr));
1043                 insert_mmap(ucontext, mm1);
1044                 mm2->key = uresp.db_key;
1045                 mm2->addr = qhp->wq.udb & PAGE_MASK;
1046                 mm2->len = PAGE_SIZE;
1047                 insert_mmap(ucontext, mm2);
1048         }
1049         qhp->ibqp.qp_num = qhp->wq.qpid;
1050         init_timer(&(qhp->timer));
1051         PDBG("%s sq_num_entries %d, rq_num_entries %d "
1052              "qpid 0x%0x qhp %p dma_addr 0x%llx size %d rq_addr 0x%x\n",
1053              __func__, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
1054              qhp->wq.qpid, qhp, (unsigned long long) qhp->wq.dma_addr,
1055              1 << qhp->wq.size_log2, qhp->wq.rq_addr);
1056         return &qhp->ibqp;
1057 }
1058
1059 static int iwch_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1060                       int attr_mask, struct ib_udata *udata)
1061 {
1062         struct iwch_dev *rhp;
1063         struct iwch_qp *qhp;
1064         enum iwch_qp_attr_mask mask = 0;
1065         struct iwch_qp_attributes attrs;
1066
1067         PDBG("%s ib_qp %p\n", __func__, ibqp);
1068
1069         /* iwarp does not support the RTR state */
1070         if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
1071                 attr_mask &= ~IB_QP_STATE;
1072
1073         /* Make sure we still have something left to do */
1074         if (!attr_mask)
1075                 return 0;
1076
1077         memset(&attrs, 0, sizeof attrs);
1078         qhp = to_iwch_qp(ibqp);
1079         rhp = qhp->rhp;
1080
1081         attrs.next_state = iwch_convert_state(attr->qp_state);
1082         attrs.enable_rdma_read = (attr->qp_access_flags &
1083                                IB_ACCESS_REMOTE_READ) ?  1 : 0;
1084         attrs.enable_rdma_write = (attr->qp_access_flags &
1085                                 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
1086         attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
1087
1088
1089         mask |= (attr_mask & IB_QP_STATE) ? IWCH_QP_ATTR_NEXT_STATE : 0;
1090         mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
1091                         (IWCH_QP_ATTR_ENABLE_RDMA_READ |
1092                          IWCH_QP_ATTR_ENABLE_RDMA_WRITE |
1093                          IWCH_QP_ATTR_ENABLE_RDMA_BIND) : 0;
1094
1095         return iwch_modify_qp(rhp, qhp, mask, &attrs, 0);
1096 }
1097
1098 void iwch_qp_add_ref(struct ib_qp *qp)
1099 {
1100         PDBG("%s ib_qp %p\n", __func__, qp);
1101         atomic_inc(&(to_iwch_qp(qp)->refcnt));
1102 }
1103
1104 void iwch_qp_rem_ref(struct ib_qp *qp)
1105 {
1106         PDBG("%s ib_qp %p\n", __func__, qp);
1107         if (atomic_dec_and_test(&(to_iwch_qp(qp)->refcnt)))
1108                 wake_up(&(to_iwch_qp(qp)->wait));
1109 }
1110
1111 static struct ib_qp *iwch_get_qp(struct ib_device *dev, int qpn)
1112 {
1113         PDBG("%s ib_dev %p qpn 0x%x\n", __func__, dev, qpn);
1114         return (struct ib_qp *)get_qhp(to_iwch_dev(dev), qpn);
1115 }
1116
1117
1118 static int iwch_query_pkey(struct ib_device *ibdev,
1119                            u8 port, u16 index, u16 * pkey)
1120 {
1121         PDBG("%s ibdev %p\n", __func__, ibdev);
1122         *pkey = 0;
1123         return 0;
1124 }
1125
1126 static int iwch_query_gid(struct ib_device *ibdev, u8 port,
1127                           int index, union ib_gid *gid)
1128 {
1129         struct iwch_dev *dev;
1130
1131         PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
1132                __func__, ibdev, port, index, gid);
1133         dev = to_iwch_dev(ibdev);
1134         BUG_ON(port == 0 || port > 2);
1135         memset(&(gid->raw[0]), 0, sizeof(gid->raw));
1136         memcpy(&(gid->raw[0]), dev->rdev.port_info.lldevs[port-1]->dev_addr, 6);
1137         return 0;
1138 }
1139
1140 static u64 fw_vers_string_to_u64(struct iwch_dev *iwch_dev)
1141 {
1142         struct ethtool_drvinfo info;
1143         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1144         char *cp, *next;
1145         unsigned fw_maj, fw_min, fw_mic;
1146
1147         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1148
1149         next = info.fw_version + 1;
1150         cp = strsep(&next, ".");
1151         sscanf(cp, "%i", &fw_maj);
1152         cp = strsep(&next, ".");
1153         sscanf(cp, "%i", &fw_min);
1154         cp = strsep(&next, ".");
1155         sscanf(cp, "%i", &fw_mic);
1156
1157         return (((u64)fw_maj & 0xffff) << 32) | ((fw_min & 0xffff) << 16) |
1158                (fw_mic & 0xffff);
1159 }
1160
1161 static int iwch_query_device(struct ib_device *ibdev, struct ib_device_attr *props,
1162                              struct ib_udata *uhw)
1163 {
1164
1165         struct iwch_dev *dev;
1166
1167         PDBG("%s ibdev %p\n", __func__, ibdev);
1168
1169         if (uhw->inlen || uhw->outlen)
1170                 return -EINVAL;
1171
1172         dev = to_iwch_dev(ibdev);
1173         memset(props, 0, sizeof *props);
1174         memcpy(&props->sys_image_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1175         props->hw_ver = dev->rdev.t3cdev_p->type;
1176         props->fw_ver = fw_vers_string_to_u64(dev);
1177         props->device_cap_flags = dev->device_cap_flags;
1178         props->page_size_cap = dev->attr.mem_pgsizes_bitmask;
1179         props->vendor_id = (u32)dev->rdev.rnic_info.pdev->vendor;
1180         props->vendor_part_id = (u32)dev->rdev.rnic_info.pdev->device;
1181         props->max_mr_size = dev->attr.max_mr_size;
1182         props->max_qp = dev->attr.max_qps;
1183         props->max_qp_wr = dev->attr.max_wrs;
1184         props->max_sge = dev->attr.max_sge_per_wr;
1185         props->max_sge_rd = 1;
1186         props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
1187         props->max_qp_init_rd_atom = dev->attr.max_rdma_reads_per_qp;
1188         props->max_cq = dev->attr.max_cqs;
1189         props->max_cqe = dev->attr.max_cqes_per_cq;
1190         props->max_mr = dev->attr.max_mem_regs;
1191         props->max_pd = dev->attr.max_pds;
1192         props->local_ca_ack_delay = 0;
1193         props->max_fast_reg_page_list_len = T3_MAX_FASTREG_DEPTH;
1194
1195         return 0;
1196 }
1197
1198 static int iwch_query_port(struct ib_device *ibdev,
1199                            u8 port, struct ib_port_attr *props)
1200 {
1201         struct iwch_dev *dev;
1202         struct net_device *netdev;
1203         struct in_device *inetdev;
1204
1205         PDBG("%s ibdev %p\n", __func__, ibdev);
1206
1207         dev = to_iwch_dev(ibdev);
1208         netdev = dev->rdev.port_info.lldevs[port-1];
1209
1210         memset(props, 0, sizeof(struct ib_port_attr));
1211         props->max_mtu = IB_MTU_4096;
1212         if (netdev->mtu >= 4096)
1213                 props->active_mtu = IB_MTU_4096;
1214         else if (netdev->mtu >= 2048)
1215                 props->active_mtu = IB_MTU_2048;
1216         else if (netdev->mtu >= 1024)
1217                 props->active_mtu = IB_MTU_1024;
1218         else if (netdev->mtu >= 512)
1219                 props->active_mtu = IB_MTU_512;
1220         else
1221                 props->active_mtu = IB_MTU_256;
1222
1223         if (!netif_carrier_ok(netdev))
1224                 props->state = IB_PORT_DOWN;
1225         else {
1226                 inetdev = in_dev_get(netdev);
1227                 if (inetdev) {
1228                         if (inetdev->ifa_list)
1229                                 props->state = IB_PORT_ACTIVE;
1230                         else
1231                                 props->state = IB_PORT_INIT;
1232                         in_dev_put(inetdev);
1233                 } else
1234                         props->state = IB_PORT_INIT;
1235         }
1236
1237         props->port_cap_flags =
1238             IB_PORT_CM_SUP |
1239             IB_PORT_SNMP_TUNNEL_SUP |
1240             IB_PORT_REINIT_SUP |
1241             IB_PORT_DEVICE_MGMT_SUP |
1242             IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
1243         props->gid_tbl_len = 1;
1244         props->pkey_tbl_len = 1;
1245         props->active_width = 2;
1246         props->active_speed = IB_SPEED_DDR;
1247         props->max_msg_sz = -1;
1248
1249         return 0;
1250 }
1251
1252 static ssize_t show_rev(struct device *dev, struct device_attribute *attr,
1253                         char *buf)
1254 {
1255         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1256                                                  ibdev.dev);
1257         PDBG("%s dev 0x%p\n", __func__, dev);
1258         return sprintf(buf, "%d\n", iwch_dev->rdev.t3cdev_p->type);
1259 }
1260
1261 static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, char *buf)
1262 {
1263         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1264                                                  ibdev.dev);
1265         struct ethtool_drvinfo info;
1266         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1267
1268         PDBG("%s dev 0x%p\n", __func__, dev);
1269         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1270         return sprintf(buf, "%s\n", info.fw_version);
1271 }
1272
1273 static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
1274                         char *buf)
1275 {
1276         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1277                                                  ibdev.dev);
1278         struct ethtool_drvinfo info;
1279         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1280
1281         PDBG("%s dev 0x%p\n", __func__, dev);
1282         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1283         return sprintf(buf, "%s\n", info.driver);
1284 }
1285
1286 static ssize_t show_board(struct device *dev, struct device_attribute *attr,
1287                           char *buf)
1288 {
1289         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1290                                                  ibdev.dev);
1291         PDBG("%s dev 0x%p\n", __func__, dev);
1292         return sprintf(buf, "%x.%x\n", iwch_dev->rdev.rnic_info.pdev->vendor,
1293                        iwch_dev->rdev.rnic_info.pdev->device);
1294 }
1295
1296 static int iwch_get_mib(struct ib_device *ibdev,
1297                         union rdma_protocol_stats *stats)
1298 {
1299         struct iwch_dev *dev;
1300         struct tp_mib_stats m;
1301         int ret;
1302
1303         PDBG("%s ibdev %p\n", __func__, ibdev);
1304         dev = to_iwch_dev(ibdev);
1305         ret = dev->rdev.t3cdev_p->ctl(dev->rdev.t3cdev_p, RDMA_GET_MIB, &m);
1306         if (ret)
1307                 return -ENOSYS;
1308
1309         memset(stats, 0, sizeof *stats);
1310         stats->iw.ipInReceives = ((u64) m.ipInReceive_hi << 32) +
1311                                 m.ipInReceive_lo;
1312         stats->iw.ipInHdrErrors = ((u64) m.ipInHdrErrors_hi << 32) +
1313                                   m.ipInHdrErrors_lo;
1314         stats->iw.ipInAddrErrors = ((u64) m.ipInAddrErrors_hi << 32) +
1315                                    m.ipInAddrErrors_lo;
1316         stats->iw.ipInUnknownProtos = ((u64) m.ipInUnknownProtos_hi << 32) +
1317                                       m.ipInUnknownProtos_lo;
1318         stats->iw.ipInDiscards = ((u64) m.ipInDiscards_hi << 32) +
1319                                  m.ipInDiscards_lo;
1320         stats->iw.ipInDelivers = ((u64) m.ipInDelivers_hi << 32) +
1321                                  m.ipInDelivers_lo;
1322         stats->iw.ipOutRequests = ((u64) m.ipOutRequests_hi << 32) +
1323                                   m.ipOutRequests_lo;
1324         stats->iw.ipOutDiscards = ((u64) m.ipOutDiscards_hi << 32) +
1325                                   m.ipOutDiscards_lo;
1326         stats->iw.ipOutNoRoutes = ((u64) m.ipOutNoRoutes_hi << 32) +
1327                                   m.ipOutNoRoutes_lo;
1328         stats->iw.ipReasmTimeout = (u64) m.ipReasmTimeout;
1329         stats->iw.ipReasmReqds = (u64) m.ipReasmReqds;
1330         stats->iw.ipReasmOKs = (u64) m.ipReasmOKs;
1331         stats->iw.ipReasmFails = (u64) m.ipReasmFails;
1332         stats->iw.tcpActiveOpens = (u64) m.tcpActiveOpens;
1333         stats->iw.tcpPassiveOpens = (u64) m.tcpPassiveOpens;
1334         stats->iw.tcpAttemptFails = (u64) m.tcpAttemptFails;
1335         stats->iw.tcpEstabResets = (u64) m.tcpEstabResets;
1336         stats->iw.tcpOutRsts = (u64) m.tcpOutRsts;
1337         stats->iw.tcpCurrEstab = (u64) m.tcpCurrEstab;
1338         stats->iw.tcpInSegs = ((u64) m.tcpInSegs_hi << 32) +
1339                               m.tcpInSegs_lo;
1340         stats->iw.tcpOutSegs = ((u64) m.tcpOutSegs_hi << 32) +
1341                                m.tcpOutSegs_lo;
1342         stats->iw.tcpRetransSegs = ((u64) m.tcpRetransSeg_hi << 32) +
1343                                   m.tcpRetransSeg_lo;
1344         stats->iw.tcpInErrs = ((u64) m.tcpInErrs_hi << 32) +
1345                               m.tcpInErrs_lo;
1346         stats->iw.tcpRtoMin = (u64) m.tcpRtoMin;
1347         stats->iw.tcpRtoMax = (u64) m.tcpRtoMax;
1348         return 0;
1349 }
1350
1351 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1352 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1353 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1354 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1355
1356 static struct device_attribute *iwch_class_attributes[] = {
1357         &dev_attr_hw_rev,
1358         &dev_attr_fw_ver,
1359         &dev_attr_hca_type,
1360         &dev_attr_board_id,
1361 };
1362
1363 static int iwch_port_immutable(struct ib_device *ibdev, u8 port_num,
1364                                struct ib_port_immutable *immutable)
1365 {
1366         struct ib_port_attr attr;
1367         int err;
1368
1369         err = iwch_query_port(ibdev, port_num, &attr);
1370         if (err)
1371                 return err;
1372
1373         immutable->pkey_tbl_len = attr.pkey_tbl_len;
1374         immutable->gid_tbl_len = attr.gid_tbl_len;
1375         immutable->core_cap_flags = RDMA_CORE_PORT_IWARP;
1376
1377         return 0;
1378 }
1379
1380 int iwch_register_device(struct iwch_dev *dev)
1381 {
1382         int ret;
1383         int i;
1384
1385         PDBG("%s iwch_dev %p\n", __func__, dev);
1386         strlcpy(dev->ibdev.name, "cxgb3_%d", IB_DEVICE_NAME_MAX);
1387         memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
1388         memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1389         dev->ibdev.owner = THIS_MODULE;
1390         dev->device_cap_flags = IB_DEVICE_LOCAL_DMA_LKEY |
1391                                 IB_DEVICE_MEM_WINDOW |
1392                                 IB_DEVICE_MEM_MGT_EXTENSIONS;
1393
1394         /* cxgb3 supports STag 0. */
1395         dev->ibdev.local_dma_lkey = 0;
1396
1397         dev->ibdev.uverbs_cmd_mask =
1398             (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1399             (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1400             (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1401             (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1402             (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1403             (1ull << IB_USER_VERBS_CMD_REG_MR) |
1404             (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1405             (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1406             (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1407             (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1408             (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1409             (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1410             (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1411             (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1412             (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1413             (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1414             (1ull << IB_USER_VERBS_CMD_POST_RECV);
1415         dev->ibdev.node_type = RDMA_NODE_RNIC;
1416         memcpy(dev->ibdev.node_desc, IWCH_NODE_DESC, sizeof(IWCH_NODE_DESC));
1417         dev->ibdev.phys_port_cnt = dev->rdev.port_info.nports;
1418         dev->ibdev.num_comp_vectors = 1;
1419         dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
1420         dev->ibdev.query_device = iwch_query_device;
1421         dev->ibdev.query_port = iwch_query_port;
1422         dev->ibdev.query_pkey = iwch_query_pkey;
1423         dev->ibdev.query_gid = iwch_query_gid;
1424         dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
1425         dev->ibdev.dealloc_ucontext = iwch_dealloc_ucontext;
1426         dev->ibdev.mmap = iwch_mmap;
1427         dev->ibdev.alloc_pd = iwch_allocate_pd;
1428         dev->ibdev.dealloc_pd = iwch_deallocate_pd;
1429         dev->ibdev.create_ah = iwch_ah_create;
1430         dev->ibdev.destroy_ah = iwch_ah_destroy;
1431         dev->ibdev.create_qp = iwch_create_qp;
1432         dev->ibdev.modify_qp = iwch_ib_modify_qp;
1433         dev->ibdev.destroy_qp = iwch_destroy_qp;
1434         dev->ibdev.create_cq = iwch_create_cq;
1435         dev->ibdev.destroy_cq = iwch_destroy_cq;
1436         dev->ibdev.resize_cq = iwch_resize_cq;
1437         dev->ibdev.poll_cq = iwch_poll_cq;
1438         dev->ibdev.get_dma_mr = iwch_get_dma_mr;
1439         dev->ibdev.reg_phys_mr = iwch_register_phys_mem;
1440         dev->ibdev.rereg_phys_mr = iwch_reregister_phys_mem;
1441         dev->ibdev.reg_user_mr = iwch_reg_user_mr;
1442         dev->ibdev.dereg_mr = iwch_dereg_mr;
1443         dev->ibdev.alloc_mw = iwch_alloc_mw;
1444         dev->ibdev.bind_mw = iwch_bind_mw;
1445         dev->ibdev.dealloc_mw = iwch_dealloc_mw;
1446         dev->ibdev.alloc_fast_reg_mr = iwch_alloc_fast_reg_mr;
1447         dev->ibdev.alloc_fast_reg_page_list = iwch_alloc_fastreg_pbl;
1448         dev->ibdev.free_fast_reg_page_list = iwch_free_fastreg_pbl;
1449         dev->ibdev.attach_mcast = iwch_multicast_attach;
1450         dev->ibdev.detach_mcast = iwch_multicast_detach;
1451         dev->ibdev.process_mad = iwch_process_mad;
1452         dev->ibdev.req_notify_cq = iwch_arm_cq;
1453         dev->ibdev.post_send = iwch_post_send;
1454         dev->ibdev.post_recv = iwch_post_receive;
1455         dev->ibdev.get_protocol_stats = iwch_get_mib;
1456         dev->ibdev.uverbs_abi_ver = IWCH_UVERBS_ABI_VERSION;
1457         dev->ibdev.get_port_immutable = iwch_port_immutable;
1458
1459         dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
1460         if (!dev->ibdev.iwcm)
1461                 return -ENOMEM;
1462
1463         dev->ibdev.iwcm->connect = iwch_connect;
1464         dev->ibdev.iwcm->accept = iwch_accept_cr;
1465         dev->ibdev.iwcm->reject = iwch_reject_cr;
1466         dev->ibdev.iwcm->create_listen = iwch_create_listen;
1467         dev->ibdev.iwcm->destroy_listen = iwch_destroy_listen;
1468         dev->ibdev.iwcm->add_ref = iwch_qp_add_ref;
1469         dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
1470         dev->ibdev.iwcm->get_qp = iwch_get_qp;
1471
1472         ret = ib_register_device(&dev->ibdev, NULL);
1473         if (ret)
1474                 goto bail1;
1475
1476         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i) {
1477                 ret = device_create_file(&dev->ibdev.dev,
1478                                          iwch_class_attributes[i]);
1479                 if (ret) {
1480                         goto bail2;
1481                 }
1482         }
1483         return 0;
1484 bail2:
1485         ib_unregister_device(&dev->ibdev);
1486 bail1:
1487         kfree(dev->ibdev.iwcm);
1488         return ret;
1489 }
1490
1491 void iwch_unregister_device(struct iwch_dev *dev)
1492 {
1493         int i;
1494
1495         PDBG("%s iwch_dev %p\n", __func__, dev);
1496         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i)
1497                 device_remove_file(&dev->ibdev.dev,
1498                                    iwch_class_attributes[i]);
1499         ib_unregister_device(&dev->ibdev);
1500         kfree(dev->ibdev.iwcm);
1501         return;
1502 }