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
  46 #include <asm/io.h>
  47 #include <asm/irq.h>
  48 #include <asm/byteorder.h>
  49
  50 #include <rdma/iw_cm.h>
  51 #include <rdma/ib_verbs.h>
  52 #include <rdma/ib_smi.h>
  53 #include <rdma/ib_umem.h>
  54 #include <rdma/ib_user_verbs.h>
  55
  56 #include "cxio_hal.h"
  57 #include "iwch.h"
  58 #include "iwch_provider.h"
  59 #include "iwch_cm.h"
  60 #include "iwch_user.h"
  61 #include "common.h"
  62
  63 static int iwch_modify_port(struct ib_device *ibdev,
  64                             u8 port, int port_modify_mask,
  65                             struct ib_port_modify *props)
  66 {
  67         return -ENOSYS;
  68 }
  69
  70 static struct ib_ah *iwch_ah_create(struct ib_pd *pd,
  71                                     struct ib_ah_attr *ah_attr)
  72 {
  73         return ERR_PTR(-ENOSYS);
  74 }
  75
  76 static int iwch_ah_destroy(struct ib_ah *ah)
  77 {
  78         return -ENOSYS;
  79 }
  80
  81 static int iwch_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
  82 {
  83         return -ENOSYS;
  84 }
  85
  86 static int iwch_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
  87 {
  88         return -ENOSYS;
  89 }
  90
  91 static int iwch_process_mad(struct ib_device *ibdev,
  92                             int mad_flags,
  93                             u8 port_num,
  94                             struct ib_wc *in_wc,
  95                             struct ib_grh *in_grh,
  96                             struct ib_mad *in_mad, struct ib_mad *out_mad)
  97 {
  98         return -ENOSYS;
  99 }
 100
 101 static int iwch_dealloc_ucontext(struct ib_ucontext *context)
 102 {
 103         struct iwch_dev *rhp = to_iwch_dev(context->device);
 104         struct iwch_ucontext *ucontext = to_iwch_ucontext(context);
 105         struct iwch_mm_entry *mm, *tmp;
 106
 107         PDBG("%s context %p\n", __func__, context);
 108         list_for_each_entry_safe(mm, tmp, &ucontext->mmaps, entry)
 109                 kfree(mm);
 110         cxio_release_ucontext(&rhp->rdev, &ucontext->uctx);
 111         kfree(ucontext);
 112         return 0;
 113 }
 114
 115 static struct ib_ucontext *iwch_alloc_ucontext(struct ib_device *ibdev,
 116                                         struct ib_udata *udata)
 117 {
 118         struct iwch_ucontext *context;
 119         struct iwch_dev *rhp = to_iwch_dev(ibdev);
 120
 121         PDBG("%s ibdev %p\n", __func__, ibdev);
 122         context = kzalloc(sizeof(*context), GFP_KERNEL);
 123         if (!context)
 124                 return ERR_PTR(-ENOMEM);
 125         cxio_init_ucontext(&rhp->rdev, &context->uctx);
 126         INIT_LIST_HEAD(&context->mmaps);
 127         spin_lock_init(&context->mmap_lock);
 128         return &context->ibucontext;
 129 }
 130
 131 static int iwch_destroy_cq(struct ib_cq *ib_cq)
 132 {
 133         struct iwch_cq *chp;
 134
 135         PDBG("%s ib_cq %p\n", __func__, ib_cq);
 136         chp = to_iwch_cq(ib_cq);
 137
 138         remove_handle(chp->rhp, &chp->rhp->cqidr, chp->cq.cqid);
 139         atomic_dec(&chp->refcnt);
 140         wait_event(chp->wait, !atomic_read(&chp->refcnt));
 141
 142         cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
 143         kfree(chp);
 144         return 0;
 145 }
 146
 147 static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries, int vector,
 148                              struct ib_ucontext *ib_context,
 149                              struct ib_udata *udata)
 150 {
 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
 157         PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
 158         rhp = to_iwch_dev(ibdev);
 159         chp = kzalloc(sizeof(*chp), GFP_KERNEL);
 160         if (!chp)
 161                 return ERR_PTR(-ENOMEM);
 162
 163         if (ib_context) {
 164                 ucontext = to_iwch_ucontext(ib_context);
 165                 if (!t3a_device(rhp)) {
 166                         if (ib_copy_from_udata(&ureq, udata, sizeof (ureq))) {
 167                                 kfree(chp);
 168                                 return ERR_PTR(-EFAULT);
 169                         }
 170                         chp->user_rptr_addr = (u32 __user *)(unsigned long)ureq.user_rptr_addr;
 171                 }
 172         }
 173
 174         if (t3a_device(rhp)) {
 175
 176                 /*
 177                  * T3A: Add some fluff to handle extra CQEs inserted
 178                  * for various errors.
 179                  * Additional CQE possibilities:
 180                  *      TERMINATE,
 181                  *      incoming RDMA WRITE Failures
 182                  *      incoming RDMA READ REQUEST FAILUREs
 183                  * NOTE: We cannot ensure the CQ won't overflow.
 184                  */
 185                 entries += 16;
 186         }
 187         entries = roundup_pow_of_two(entries);
 188         chp->cq.size_log2 = ilog2(entries);
 189
 190         if (cxio_create_cq(&rhp->rdev, &chp->cq, !ucontext)) {
 191                 kfree(chp);
 192                 return ERR_PTR(-ENOMEM);
 193         }
 194         chp->rhp = rhp;
 195         chp->ibcq.cqe = 1 << chp->cq.size_log2;
 196         spin_lock_init(&chp->lock);
 197         atomic_set(&chp->refcnt, 1);
 198         init_waitqueue_head(&chp->wait);
 199         if (insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid)) {
 200                 cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
 201                 kfree(chp);
 202                 return ERR_PTR(-ENOMEM);
 203         }
 204
 205         if (ucontext) {
 206                 struct iwch_mm_entry *mm;
 207
 208                 mm = kmalloc(sizeof *mm, GFP_KERNEL);
 209                 if (!mm) {
 210                         iwch_destroy_cq(&chp->ibcq);
 211                         return ERR_PTR(-ENOMEM);
 212                 }
 213                 uresp.cqid = chp->cq.cqid;
 214                 uresp.size_log2 = chp->cq.size_log2;
 215                 spin_lock(&ucontext->mmap_lock);
 216                 uresp.key = ucontext->key;
 217                 ucontext->key += PAGE_SIZE;
 218                 spin_unlock(&ucontext->mmap_lock);
 219                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 220                         kfree(mm);
 221                         iwch_destroy_cq(&chp->ibcq);
 222                         return ERR_PTR(-EFAULT);
 223                 }
 224                 mm->key = uresp.key;
 225                 mm->addr = virt_to_phys(chp->cq.queue);
 226                 mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
 227                                              sizeof (struct t3_cqe));
 228                 insert_mmap(ucontext, mm);
 229         }
 230         PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
 231              chp->cq.cqid, chp, (1 << chp->cq.size_log2),
 232              (unsigned long long) chp->cq.dma_addr);
 233         return &chp->ibcq;
 234 }
 235
 236 static int iwch_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
 237 {
 238 #ifdef notyet
 239         struct iwch_cq *chp = to_iwch_cq(cq);
 240         struct t3_cq oldcq, newcq;
 241         int ret;
 242
 243         PDBG("%s ib_cq %p cqe %d\n", __func__, cq, cqe);
 244
 245         /* We don't downsize... */
 246         if (cqe <= cq->cqe)
 247                 return 0;
 248
 249         /* create new t3_cq with new size */
 250         cqe = roundup_pow_of_two(cqe+1);
 251         newcq.size_log2 = ilog2(cqe);
 252
 253         /* Dont allow resize to less than the current wce count */
 254         if (cqe < Q_COUNT(chp->cq.rptr, chp->cq.wptr)) {
 255                 return -ENOMEM;
 256         }
 257
 258         /* Quiesce all QPs using this CQ */
 259         ret = iwch_quiesce_qps(chp);
 260         if (ret) {
 261                 return ret;
 262         }
 263
 264         ret = cxio_create_cq(&chp->rhp->rdev, &newcq);
 265         if (ret) {
 266                 return ret;
 267         }
 268
 269         /* copy CQEs */
 270         memcpy(newcq.queue, chp->cq.queue, (1 << chp->cq.size_log2) *
 271                                         sizeof(struct t3_cqe));
 272
 273         /* old iwch_qp gets new t3_cq but keeps old cqid */
 274         oldcq = chp->cq;
 275         chp->cq = newcq;
 276         chp->cq.cqid = oldcq.cqid;
 277
 278         /* resize new t3_cq to update the HW context */
 279         ret = cxio_resize_cq(&chp->rhp->rdev, &chp->cq);
 280         if (ret) {
 281                 chp->cq = oldcq;
 282                 return ret;
 283         }
 284         chp->ibcq.cqe = (1<<chp->cq.size_log2) - 1;
 285
 286         /* destroy old t3_cq */
 287         oldcq.cqid = newcq.cqid;
 288         ret = cxio_destroy_cq(&chp->rhp->rdev, &oldcq);
 289         if (ret) {
 290                 printk(KERN_ERR MOD "%s - cxio_destroy_cq failed %d\n",
 291                         __func__, ret);
 292         }
 293
 294         /* add user hooks here */
 295
 296         /* resume qps */
 297         ret = iwch_resume_qps(chp);
 298         return ret;
 299 #else
 300         return -ENOSYS;
 301 #endif
 302 }
 303
 304 static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
 305 {
 306         struct iwch_dev *rhp;
 307         struct iwch_cq *chp;
 308         enum t3_cq_opcode cq_op;
 309         int err;
 310         unsigned long flag;
 311         u32 rptr;
 312
 313         chp = to_iwch_cq(ibcq);
 314         rhp = chp->rhp;
 315         if ((flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED)
 316                 cq_op = CQ_ARM_SE;
 317         else
 318                 cq_op = CQ_ARM_AN;
 319         if (chp->user_rptr_addr) {
 320                 if (get_user(rptr, chp->user_rptr_addr))
 321                         return -EFAULT;
 322                 spin_lock_irqsave(&chp->lock, flag);
 323                 chp->cq.rptr = rptr;
 324         } else
 325                 spin_lock_irqsave(&chp->lock, flag);
 326         PDBG("%s rptr 0x%x\n", __func__, chp->cq.rptr);
 327         err = cxio_hal_cq_op(&rhp->rdev, &chp->cq, cq_op, 0);
 328         spin_unlock_irqrestore(&chp->lock, flag);
 329         if (err < 0)
 330                 printk(KERN_ERR MOD "Error %d rearming CQID 0x%x\n", err,
 331                        chp->cq.cqid);
 332         if (err > 0 && !(flags & IB_CQ_REPORT_MISSED_EVENTS))
 333                 err = 0;
 334         return err;
 335 }
 336
 337 static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 338 {
 339         int len = vma->vm_end - vma->vm_start;
 340         u32 key = vma->vm_pgoff << PAGE_SHIFT;
 341         struct cxio_rdev *rdev_p;
 342         int ret = 0;
 343         struct iwch_mm_entry *mm;
 344         struct iwch_ucontext *ucontext;
 345         u64 addr;
 346
 347         PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __func__, vma->vm_pgoff,
 348              key, len);
 349
 350         if (vma->vm_start & (PAGE_SIZE-1)) {
 351                 return -EINVAL;
 352         }
 353
 354         rdev_p = &(to_iwch_dev(context->device)->rdev);
 355         ucontext = to_iwch_ucontext(context);
 356
 357         mm = remove_mmap(ucontext, key, len);
 358         if (!mm)
 359                 return -EINVAL;
 360         addr = mm->addr;
 361         kfree(mm);
 362
 363         if ((addr >= rdev_p->rnic_info.udbell_physbase) &&
 364             (addr < (rdev_p->rnic_info.udbell_physbase +
 365                        rdev_p->rnic_info.udbell_len))) {
 366
 367                 /*
 368                  * Map T3 DB register.
 369                  */
 370                 if (vma->vm_flags & VM_READ) {
 371                         return -EPERM;
 372                 }
 373
 374                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 375                 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
 376                 vma->vm_flags &= ~VM_MAYREAD;
 377                 ret = io_remap_pfn_range(vma, vma->vm_start,
 378                                          addr >> PAGE_SHIFT,
 379                                          len, vma->vm_page_prot);
 380         } else {
 381
 382                 /*
 383                  * Map WQ or CQ contig dma memory...
 384                  */
 385                 ret = remap_pfn_range(vma, vma->vm_start,
 386                                       addr >> PAGE_SHIFT,
 387                                       len, vma->vm_page_prot);
 388         }
 389
 390         return ret;
 391 }
 392
 393 static int iwch_deallocate_pd(struct ib_pd *pd)
 394 {
 395         struct iwch_dev *rhp;
 396         struct iwch_pd *php;
 397
 398         php = to_iwch_pd(pd);
 399         rhp = php->rhp;
 400         PDBG("%s ibpd %p pdid 0x%x\n", __func__, pd, php->pdid);
 401         cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid);
 402         kfree(php);
 403         return 0;
 404 }
 405
 406 static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev,
 407                                struct ib_ucontext *context,
 408                                struct ib_udata *udata)
 409 {
 410         struct iwch_pd *php;
 411         u32 pdid;
 412         struct iwch_dev *rhp;
 413
 414         PDBG("%s ibdev %p\n", __func__, ibdev);
 415         rhp = (struct iwch_dev *) ibdev;
 416         pdid = cxio_hal_get_pdid(rhp->rdev.rscp);
 417         if (!pdid)
 418                 return ERR_PTR(-EINVAL);
 419         php = kzalloc(sizeof(*php), GFP_KERNEL);
 420         if (!php) {
 421                 cxio_hal_put_pdid(rhp->rdev.rscp, pdid);
 422                 return ERR_PTR(-ENOMEM);
 423         }
 424         php->pdid = pdid;
 425         php->rhp = rhp;
 426         if (context) {
 427                 if (ib_copy_to_udata(udata, &php->pdid, sizeof (__u32))) {
 428                         iwch_deallocate_pd(&php->ibpd);
 429                         return ERR_PTR(-EFAULT);
 430                 }
 431         }
 432         PDBG("%s pdid 0x%0x ptr 0x%p\n", __func__, pdid, php);
 433         return &php->ibpd;
 434 }
 435
 436 static int iwch_dereg_mr(struct ib_mr *ib_mr)
 437 {
 438         struct iwch_dev *rhp;
 439         struct iwch_mr *mhp;
 440         u32 mmid;
 441
 442         PDBG("%s ib_mr %p\n", __func__, ib_mr);
 443         /* There can be no memory windows */
 444         if (atomic_read(&ib_mr->usecnt))
 445                 return -EINVAL;
 446
 447         mhp = to_iwch_mr(ib_mr);
 448         rhp = mhp->rhp;
 449         mmid = mhp->attr.stag >> 8;
 450         cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
 451                        mhp->attr.pbl_addr);
 452         iwch_free_pbl(mhp);
 453         remove_handle(rhp, &rhp->mmidr, mmid);
 454         if (mhp->kva)
 455                 kfree((void *) (unsigned long) mhp->kva);
 456         if (mhp->umem)
 457                 ib_umem_release(mhp->umem);
 458         PDBG("%s mmid 0x%x ptr %p\n", __func__, mmid, mhp);
 459         kfree(mhp);
 460         return 0;
 461 }
 462
 463 static struct ib_mr *iwch_register_phys_mem(struct ib_pd *pd,
 464                                         struct ib_phys_buf *buffer_list,
 465                                         int num_phys_buf,
 466                                         int acc,
 467                                         u64 *iova_start)
 468 {
 469         __be64 *page_list;
 470         int shift;
 471         u64 total_size;
 472         int npages;
 473         struct iwch_dev *rhp;
 474         struct iwch_pd *php;
 475         struct iwch_mr *mhp;
 476         int ret;
 477
 478         PDBG("%s ib_pd %p\n", __func__, pd);
 479         php = to_iwch_pd(pd);
 480         rhp = php->rhp;
 481
 482         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 483         if (!mhp)
 484                 return ERR_PTR(-ENOMEM);
 485
 486         mhp->rhp = rhp;
 487
 488         /* First check that we have enough alignment */
 489         if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
 490                 ret = -EINVAL;
 491                 goto err;
 492         }
 493
 494         if (num_phys_buf > 1 &&
 495             ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
 496                 ret = -EINVAL;
 497                 goto err;
 498         }
 499
 500         ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
 501                                    &total_size, &npages, &shift, &page_list);
 502         if (ret)
 503                 goto err;
 504
 505         ret = iwch_alloc_pbl(mhp, npages);
 506         if (ret) {
 507                 kfree(page_list);
 508                 goto err_pbl;
 509         }
 510
 511         ret = iwch_write_pbl(mhp, page_list, npages, 0);
 512         kfree(page_list);
 513         if (ret)
 514                 goto err_pbl;
 515
 516         mhp->attr.pdid = php->pdid;
 517         mhp->attr.zbva = 0;
 518
 519         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 520         mhp->attr.va_fbo = *iova_start;
 521         mhp->attr.page_size = shift - 12;
 522
 523         mhp->attr.len = (u32) total_size;
 524         mhp->attr.pbl_size = npages;
 525         ret = iwch_register_mem(rhp, php, mhp, shift);
 526         if (ret)
 527                 goto err_pbl;
 528
 529         return &mhp->ibmr;
 530
 531 err_pbl:
 532         iwch_free_pbl(mhp);
 533
 534 err:
 535         kfree(mhp);
 536         return ERR_PTR(ret);
 537
 538 }
 539
 540 static int iwch_reregister_phys_mem(struct ib_mr *mr,
 541                                      int mr_rereg_mask,
 542                                      struct ib_pd *pd,
 543                                      struct ib_phys_buf *buffer_list,
 544                                      int num_phys_buf,
 545                                      int acc, u64 * iova_start)
 546 {
 547
 548         struct iwch_mr mh, *mhp;
 549         struct iwch_pd *php;
 550         struct iwch_dev *rhp;
 551         __be64 *page_list = NULL;
 552         int shift = 0;
 553         u64 total_size;
 554         int npages;
 555         int ret;
 556
 557         PDBG("%s ib_mr %p ib_pd %p\n", __func__, mr, pd);
 558
 559         /* There can be no memory windows */
 560         if (atomic_read(&mr->usecnt))
 561                 return -EINVAL;
 562
 563         mhp = to_iwch_mr(mr);
 564         rhp = mhp->rhp;
 565         php = to_iwch_pd(mr->pd);
 566
 567         /* make sure we are on the same adapter */
 568         if (rhp != php->rhp)
 569                 return -EINVAL;
 570
 571         memcpy(&mh, mhp, sizeof *mhp);
 572
 573         if (mr_rereg_mask & IB_MR_REREG_PD)
 574                 php = to_iwch_pd(pd);
 575         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 576                 mh.attr.perms = iwch_ib_to_tpt_access(acc);
 577         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 578                 ret = build_phys_page_list(buffer_list, num_phys_buf,
 579                                            iova_start,
 580                                            &total_size, &npages,
 581                                            &shift, &page_list);
 582                 if (ret)
 583                         return ret;
 584         }
 585
 586         ret = iwch_reregister_mem(rhp, php, &mh, shift, npages);
 587         kfree(page_list);
 588         if (ret) {
 589                 return ret;
 590         }
 591         if (mr_rereg_mask & IB_MR_REREG_PD)
 592                 mhp->attr.pdid = php->pdid;
 593         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 594                 mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 595         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 596                 mhp->attr.zbva = 0;
 597                 mhp->attr.va_fbo = *iova_start;
 598                 mhp->attr.page_size = shift - 12;
 599                 mhp->attr.len = (u32) total_size;
 600                 mhp->attr.pbl_size = npages;
 601         }
 602
 603         return 0;
 604 }
 605
 606
 607 static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 608                                       u64 virt, int acc, struct ib_udata *udata)
 609 {
 610         __be64 *pages;
 611         int shift, n, len;
 612         int i, j, k;
 613         int err = 0;
 614         struct ib_umem_chunk *chunk;
 615         struct iwch_dev *rhp;
 616         struct iwch_pd *php;
 617         struct iwch_mr *mhp;
 618         struct iwch_reg_user_mr_resp uresp;
 619
 620         PDBG("%s ib_pd %p\n", __func__, pd);
 621
 622         php = to_iwch_pd(pd);
 623         rhp = php->rhp;
 624         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 625         if (!mhp)
 626                 return ERR_PTR(-ENOMEM);
 627
 628         mhp->rhp = rhp;
 629
 630         mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
 631         if (IS_ERR(mhp->umem)) {
 632                 err = PTR_ERR(mhp->umem);
 633                 kfree(mhp);
 634                 return ERR_PTR(err);
 635         }
 636
 637         shift = ffs(mhp->umem->page_size) - 1;
 638
 639         n = 0;
 640         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 641                 n += chunk->nents;
 642
 643         err = iwch_alloc_pbl(mhp, n);
 644         if (err)
 645                 goto err;
 646
 647         pages = (__be64 *) __get_free_page(GFP_KERNEL);
 648         if (!pages) {
 649                 err = -ENOMEM;
 650                 goto err_pbl;
 651         }
 652
 653         i = n = 0;
 654
 655         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 656                 for (j = 0; j < chunk->nmap; ++j) {
 657                         len = sg_dma_len(&chunk->page_list[j]) >> shift;
 658                         for (k = 0; k < len; ++k) {
 659                                 pages[i++] = cpu_to_be64(sg_dma_address(
 660                                         &chunk->page_list[j]) +
 661                                         mhp->umem->page_size * k);
 662                                 if (i == PAGE_SIZE / sizeof *pages) {
 663                                         err = iwch_write_pbl(mhp, pages, i, n);
 664                                         if (err)
 665                                                 goto pbl_done;
 666                                         n += i;
 667                                         i = 0;
 668                                 }
 669                         }
 670                 }
 671
 672         if (i)
 673                 err = iwch_write_pbl(mhp, pages, i, n);
 674
 675 pbl_done:
 676         free_page((unsigned long) pages);
 677         if (err)
 678                 goto err_pbl;
 679
 680         mhp->attr.pdid = php->pdid;
 681         mhp->attr.zbva = 0;
 682         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 683         mhp->attr.va_fbo = virt;
 684         mhp->attr.page_size = shift - 12;
 685         mhp->attr.len = (u32) length;
 686
 687         err = iwch_register_mem(rhp, php, mhp, shift);
 688         if (err)
 689                 goto err_pbl;
 690
 691         if (udata && !t3a_device(rhp)) {
 692                 uresp.pbl_addr = (mhp->attr.pbl_addr -
 693                                  rhp->rdev.rnic_info.pbl_base) >> 3;
 694                 PDBG("%s user resp pbl_addr 0x%x\n", __func__,
 695                      uresp.pbl_addr);
 696
 697                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 698                         iwch_dereg_mr(&mhp->ibmr);
 699                         err = -EFAULT;
 700                         goto err;
 701                 }
 702         }
 703
 704         return &mhp->ibmr;
 705
 706 err_pbl:
 707         iwch_free_pbl(mhp);
 708
 709 err:
 710         ib_umem_release(mhp->umem);
 711         kfree(mhp);
 712         return ERR_PTR(err);
 713 }
 714
 715 static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
 716 {
 717         struct ib_phys_buf bl;
 718         u64 kva;
 719         struct ib_mr *ibmr;
 720
 721         PDBG("%s ib_pd %p\n", __func__, pd);
 722
 723         /*
 724          * T3 only supports 32 bits of size.
 725          */
 726         bl.size = 0xffffffff;
 727         bl.addr = 0;
 728         kva = 0;
 729         ibmr = iwch_register_phys_mem(pd, &bl, 1, acc, &kva);
 730         return ibmr;
 731 }
 732
 733 static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
 734 {
 735         struct iwch_dev *rhp;
 736         struct iwch_pd *php;
 737         struct iwch_mw *mhp;
 738         u32 mmid;
 739         u32 stag = 0;
 740         int ret;
 741
 742         php = to_iwch_pd(pd);
 743         rhp = php->rhp;
 744         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 745         if (!mhp)
 746                 return ERR_PTR(-ENOMEM);
 747         ret = cxio_allocate_window(&rhp->rdev, &stag, php->pdid);
 748         if (ret) {
 749                 kfree(mhp);
 750                 return ERR_PTR(ret);
 751         }
 752         mhp->rhp = rhp;
 753         mhp->attr.pdid = php->pdid;
 754         mhp->attr.type = TPT_MW;
 755         mhp->attr.stag = stag;
 756         mmid = (stag) >> 8;
 757         mhp->ibmw.rkey = stag;
 758         if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) {
 759                 cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 760                 kfree(mhp);
 761                 return ERR_PTR(-ENOMEM);
 762         }
 763         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 764         return &(mhp->ibmw);
 765 }
 766
 767 static int iwch_dealloc_mw(struct ib_mw *mw)
 768 {
 769         struct iwch_dev *rhp;
 770         struct iwch_mw *mhp;
 771         u32 mmid;
 772
 773         mhp = to_iwch_mw(mw);
 774         rhp = mhp->rhp;
 775         mmid = (mw->rkey) >> 8;
 776         cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 777         remove_handle(rhp, &rhp->mmidr, mmid);
 778         kfree(mhp);
 779         PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
 780         return 0;
 781 }
 782
 783 static struct ib_mr *iwch_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth)
 784 {
 785         struct iwch_dev *rhp;
 786         struct iwch_pd *php;
 787         struct iwch_mr *mhp;
 788         u32 mmid;
 789         u32 stag = 0;
 790         int ret = 0;
 791
 792         php = to_iwch_pd(pd);
 793         rhp = php->rhp;
 794         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 795         if (!mhp)
 796                 goto err;
 797
 798         mhp->rhp = rhp;
 799         ret = iwch_alloc_pbl(mhp, pbl_depth);
 800         if (ret)
 801                 goto err1;
 802         mhp->attr.pbl_size = pbl_depth;
 803         ret = cxio_allocate_stag(&rhp->rdev, &stag, php->pdid,
 804                                  mhp->attr.pbl_size, mhp->attr.pbl_addr);
 805         if (ret)
 806                 goto err2;
 807         mhp->attr.pdid = php->pdid;
 808         mhp->attr.type = TPT_NON_SHARED_MR;
 809         mhp->attr.stag = stag;
 810         mhp->attr.state = 1;
 811         mmid = (stag) >> 8;
 812         mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
 813         if (insert_handle(rhp, &rhp->mmidr, mhp, mmid))
 814                 goto err3;
 815
 816         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 817         return &(mhp->ibmr);
 818 err3:
 819         cxio_dereg_mem(&rhp->rdev, stag, mhp->attr.pbl_size,
 820                        mhp->attr.pbl_addr);
 821 err2:
 822         iwch_free_pbl(mhp);
 823 err1:
 824         kfree(mhp);
 825 err:
 826         return ERR_PTR(ret);
 827 }
 828
 829 static struct ib_fast_reg_page_list *iwch_alloc_fastreg_pbl(
 830                                         struct ib_device *device,
 831                                         int page_list_len)
 832 {
 833         struct ib_fast_reg_page_list *page_list;
 834
 835         page_list = kmalloc(sizeof *page_list + page_list_len * sizeof(u64),
 836                             GFP_KERNEL);
 837         if (!page_list)
 838                 return ERR_PTR(-ENOMEM);
 839
 840         page_list->page_list = (u64 *)(page_list + 1);
 841         page_list->max_page_list_len = page_list_len;
 842
 843         return page_list;
 844 }
 845
 846 static void iwch_free_fastreg_pbl(struct ib_fast_reg_page_list *page_list)
 847 {
 848         kfree(page_list);
 849 }
 850
 851 static int iwch_destroy_qp(struct ib_qp *ib_qp)
 852 {
 853         struct iwch_dev *rhp;
 854         struct iwch_qp *qhp;
 855         struct iwch_qp_attributes attrs;
 856         struct iwch_ucontext *ucontext;
 857
 858         qhp = to_iwch_qp(ib_qp);
 859         rhp = qhp->rhp;
 860
 861         attrs.next_state = IWCH_QP_STATE_ERROR;
 862         iwch_modify_qp(rhp, qhp, IWCH_QP_ATTR_NEXT_STATE, &attrs, 0);
 863         wait_event(qhp->wait, !qhp->ep);
 864
 865         remove_handle(rhp, &rhp->qpidr, qhp->wq.qpid);
 866
 867         atomic_dec(&qhp->refcnt);
 868         wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
 869
 870         ucontext = ib_qp->uobject ? to_iwch_ucontext(ib_qp->uobject->context)
 871                                   : NULL;
 872         cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 873                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 874
 875         PDBG("%s ib_qp %p qpid 0x%0x qhp %p\n", __func__,
 876              ib_qp, qhp->wq.qpid, qhp);
 877         kfree(qhp);
 878         return 0;
 879 }
 880
 881 static struct ib_qp *iwch_create_qp(struct ib_pd *pd,
 882                              struct ib_qp_init_attr *attrs,
 883                              struct ib_udata *udata)
 884 {
 885         struct iwch_dev *rhp;
 886         struct iwch_qp *qhp;
 887         struct iwch_pd *php;
 888         struct iwch_cq *schp;
 889         struct iwch_cq *rchp;
 890         struct iwch_create_qp_resp uresp;
 891         int wqsize, sqsize, rqsize;
 892         struct iwch_ucontext *ucontext;
 893
 894         PDBG("%s ib_pd %p\n", __func__, pd);
 895         if (attrs->qp_type != IB_QPT_RC)
 896                 return ERR_PTR(-EINVAL);
 897         php = to_iwch_pd(pd);
 898         rhp = php->rhp;
 899         schp = get_chp(rhp, ((struct iwch_cq *) attrs->send_cq)->cq.cqid);
 900         rchp = get_chp(rhp, ((struct iwch_cq *) attrs->recv_cq)->cq.cqid);
 901         if (!schp || !rchp)
 902                 return ERR_PTR(-EINVAL);
 903
 904         /* The RQT size must be # of entries + 1 rounded up to a power of two */
 905         rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr);
 906         if (rqsize == attrs->cap.max_recv_wr)
 907                 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr+1);
 908
 909         /* T3 doesn't support RQT depth < 16 */
 910         if (rqsize < 16)
 911                 rqsize = 16;
 912
 913         if (rqsize > T3_MAX_RQ_SIZE)
 914                 return ERR_PTR(-EINVAL);
 915
 916         if (attrs->cap.max_inline_data > T3_MAX_INLINE)
 917                 return ERR_PTR(-EINVAL);
 918
 919         /*
 920          * NOTE: The SQ and total WQ sizes don't need to be
 921          * a power of two.  However, all the code assumes
 922          * they are. EG: Q_FREECNT() and friends.
 923          */
 924         sqsize = roundup_pow_of_two(attrs->cap.max_send_wr);
 925         wqsize = roundup_pow_of_two(rqsize + sqsize);
 926
 927         /*
 928          * Kernel users need more wq space for fastreg WRs which can take
 929          * 2 WR fragments.
 930          */
 931         ucontext = pd->uobject ? to_iwch_ucontext(pd->uobject->context) : NULL;
 932         if (!ucontext && wqsize < (rqsize + (2 * sqsize)))
 933                 wqsize = roundup_pow_of_two(rqsize +
 934                                 roundup_pow_of_two(attrs->cap.max_send_wr * 2));
 935         PDBG("%s wqsize %d sqsize %d rqsize %d\n", __func__,
 936              wqsize, sqsize, rqsize);
 937         qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
 938         if (!qhp)
 939                 return ERR_PTR(-ENOMEM);
 940         qhp->wq.size_log2 = ilog2(wqsize);
 941         qhp->wq.rq_size_log2 = ilog2(rqsize);
 942         qhp->wq.sq_size_log2 = ilog2(sqsize);
 943         if (cxio_create_qp(&rhp->rdev, !udata, &qhp->wq,
 944                            ucontext ? &ucontext->uctx : &rhp->rdev.uctx)) {
 945                 kfree(qhp);
 946                 return ERR_PTR(-ENOMEM);
 947         }
 948
 949         attrs->cap.max_recv_wr = rqsize - 1;
 950         attrs->cap.max_send_wr = sqsize;
 951         attrs->cap.max_inline_data = T3_MAX_INLINE;
 952
 953         qhp->rhp = rhp;
 954         qhp->attr.pd = php->pdid;
 955         qhp->attr.scq = ((struct iwch_cq *) attrs->send_cq)->cq.cqid;
 956         qhp->attr.rcq = ((struct iwch_cq *) attrs->recv_cq)->cq.cqid;
 957         qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
 958         qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
 959         qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
 960         qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
 961         qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
 962         qhp->attr.state = IWCH_QP_STATE_IDLE;
 963         qhp->attr.next_state = IWCH_QP_STATE_IDLE;
 964
 965         /*
 966          * XXX - These don't get passed in from the openib user
 967          * at create time.  The CM sets them via a QP modify.
 968          * Need to fix...  I think the CM should
 969          */
 970         qhp->attr.enable_rdma_read = 1;
 971         qhp->attr.enable_rdma_write = 1;
 972         qhp->attr.enable_bind = 1;
 973         qhp->attr.max_ord = 1;
 974         qhp->attr.max_ird = 1;
 975
 976         spin_lock_init(&qhp->lock);
 977         init_waitqueue_head(&qhp->wait);
 978         atomic_set(&qhp->refcnt, 1);
 979
 980         if (insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid)) {
 981                 cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 982                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 983                 kfree(qhp);
 984                 return ERR_PTR(-ENOMEM);
 985         }
 986
 987         if (udata) {
 988
 989                 struct iwch_mm_entry *mm1, *mm2;
 990
 991                 mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
 992                 if (!mm1) {
 993                         iwch_destroy_qp(&qhp->ibqp);
 994                         return ERR_PTR(-ENOMEM);
 995                 }
 996
 997                 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
 998                 if (!mm2) {
 999                         kfree(mm1);
1000                         iwch_destroy_qp(&qhp->ibqp);
1001                         return ERR_PTR(-ENOMEM);
1002                 }
1003
1004                 uresp.qpid = qhp->wq.qpid;
1005                 uresp.size_log2 = qhp->wq.size_log2;
1006                 uresp.sq_size_log2 = qhp->wq.sq_size_log2;
1007                 uresp.rq_size_log2 = qhp->wq.rq_size_log2;
1008                 spin_lock(&ucontext->mmap_lock);
1009                 uresp.key = ucontext->key;
1010                 ucontext->key += PAGE_SIZE;
1011                 uresp.db_key = ucontext->key;
1012                 ucontext->key += PAGE_SIZE;
1013                 spin_unlock(&ucontext->mmap_lock);
1014                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
1015                         kfree(mm1);
1016                         kfree(mm2);
1017                         iwch_destroy_qp(&qhp->ibqp);
1018                         return ERR_PTR(-EFAULT);
1019                 }
1020                 mm1->key = uresp.key;
1021                 mm1->addr = virt_to_phys(qhp->wq.queue);
1022                 mm1->len = PAGE_ALIGN(wqsize * sizeof (union t3_wr));
1023                 insert_mmap(ucontext, mm1);
1024                 mm2->key = uresp.db_key;
1025                 mm2->addr = qhp->wq.udb & PAGE_MASK;
1026                 mm2->len = PAGE_SIZE;
1027                 insert_mmap(ucontext, mm2);
1028         }
1029         qhp->ibqp.qp_num = qhp->wq.qpid;
1030         init_timer(&(qhp->timer));
1031         PDBG("%s sq_num_entries %d, rq_num_entries %d "
1032              "qpid 0x%0x qhp %p dma_addr 0x%llx size %d rq_addr 0x%x\n",
1033              __func__, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
1034              qhp->wq.qpid, qhp, (unsigned long long) qhp->wq.dma_addr,
1035              1 << qhp->wq.size_log2, qhp->wq.rq_addr);
1036         return &qhp->ibqp;
1037 }
1038
1039 static int iwch_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1040                       int attr_mask, struct ib_udata *udata)
1041 {
1042         struct iwch_dev *rhp;
1043         struct iwch_qp *qhp;
1044         enum iwch_qp_attr_mask mask = 0;
1045         struct iwch_qp_attributes attrs;
1046
1047         PDBG("%s ib_qp %p\n", __func__, ibqp);
1048
1049         /* iwarp does not support the RTR state */
1050         if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
1051                 attr_mask &= ~IB_QP_STATE;
1052
1053         /* Make sure we still have something left to do */
1054         if (!attr_mask)
1055                 return 0;
1056
1057         memset(&attrs, 0, sizeof attrs);
1058         qhp = to_iwch_qp(ibqp);
1059         rhp = qhp->rhp;
1060
1061         attrs.next_state = iwch_convert_state(attr->qp_state);
1062         attrs.enable_rdma_read = (attr->qp_access_flags &
1063                                IB_ACCESS_REMOTE_READ) ?  1 : 0;
1064         attrs.enable_rdma_write = (attr->qp_access_flags &
1065                                 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
1066         attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
1067
1068
1069         mask |= (attr_mask & IB_QP_STATE) ? IWCH_QP_ATTR_NEXT_STATE : 0;
1070         mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
1071                         (IWCH_QP_ATTR_ENABLE_RDMA_READ |
1072                          IWCH_QP_ATTR_ENABLE_RDMA_WRITE |
1073                          IWCH_QP_ATTR_ENABLE_RDMA_BIND) : 0;
1074
1075         return iwch_modify_qp(rhp, qhp, mask, &attrs, 0);
1076 }
1077
1078 void iwch_qp_add_ref(struct ib_qp *qp)
1079 {
1080         PDBG("%s ib_qp %p\n", __func__, qp);
1081         atomic_inc(&(to_iwch_qp(qp)->refcnt));
1082 }
1083
1084 void iwch_qp_rem_ref(struct ib_qp *qp)
1085 {
1086         PDBG("%s ib_qp %p\n", __func__, qp);
1087         if (atomic_dec_and_test(&(to_iwch_qp(qp)->refcnt)))
1088                 wake_up(&(to_iwch_qp(qp)->wait));
1089 }
1090
1091 static struct ib_qp *iwch_get_qp(struct ib_device *dev, int qpn)
1092 {
1093         PDBG("%s ib_dev %p qpn 0x%x\n", __func__, dev, qpn);
1094         return (struct ib_qp *)get_qhp(to_iwch_dev(dev), qpn);
1095 }
1096
1097
1098 static int iwch_query_pkey(struct ib_device *ibdev,
1099                            u8 port, u16 index, u16 * pkey)
1100 {
1101         PDBG("%s ibdev %p\n", __func__, ibdev);
1102         *pkey = 0;
1103         return 0;
1104 }
1105
1106 static int iwch_query_gid(struct ib_device *ibdev, u8 port,
1107                           int index, union ib_gid *gid)
1108 {
1109         struct iwch_dev *dev;
1110
1111         PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
1112                __func__, ibdev, port, index, gid);
1113         dev = to_iwch_dev(ibdev);
1114         BUG_ON(port == 0 || port > 2);
1115         memset(&(gid->raw[0]), 0, sizeof(gid->raw));
1116         memcpy(&(gid->raw[0]), dev->rdev.port_info.lldevs[port-1]->dev_addr, 6);
1117         return 0;
1118 }
1119
1120 static u64 fw_vers_string_to_u64(struct iwch_dev *iwch_dev)
1121 {
1122         struct ethtool_drvinfo info;
1123         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1124         char *cp, *next;
1125         unsigned fw_maj, fw_min, fw_mic;
1126
1127         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1128
1129         next = info.fw_version + 1;
1130         cp = strsep(&next, ".");
1131         sscanf(cp, "%i", &fw_maj);
1132         cp = strsep(&next, ".");
1133         sscanf(cp, "%i", &fw_min);
1134         cp = strsep(&next, ".");
1135         sscanf(cp, "%i", &fw_mic);
1136
1137         return (((u64)fw_maj & 0xffff) << 32) | ((fw_min & 0xffff) << 16) |
1138                (fw_mic & 0xffff);
1139 }
1140
1141 static int iwch_query_device(struct ib_device *ibdev,
1142                              struct ib_device_attr *props)
1143 {
1144
1145         struct iwch_dev *dev;
1146         PDBG("%s ibdev %p\n", __func__, ibdev);
1147
1148         dev = to_iwch_dev(ibdev);
1149         memset(props, 0, sizeof *props);
1150         memcpy(&props->sys_image_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1151         props->hw_ver = dev->rdev.t3cdev_p->type;
1152         props->fw_ver = fw_vers_string_to_u64(dev);
1153         props->device_cap_flags = dev->device_cap_flags;
1154         props->page_size_cap = dev->attr.mem_pgsizes_bitmask;
1155         props->vendor_id = (u32)dev->rdev.rnic_info.pdev->vendor;
1156         props->vendor_part_id = (u32)dev->rdev.rnic_info.pdev->device;
1157         props->max_mr_size = dev->attr.max_mr_size;
1158         props->max_qp = dev->attr.max_qps;
1159         props->max_qp_wr = dev->attr.max_wrs;
1160         props->max_sge = dev->attr.max_sge_per_wr;
1161         props->max_sge_rd = 1;
1162         props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
1163         props->max_qp_init_rd_atom = dev->attr.max_rdma_reads_per_qp;
1164         props->max_cq = dev->attr.max_cqs;
1165         props->max_cqe = dev->attr.max_cqes_per_cq;
1166         props->max_mr = dev->attr.max_mem_regs;
1167         props->max_pd = dev->attr.max_pds;
1168         props->local_ca_ack_delay = 0;
1169         props->max_fast_reg_page_list_len = T3_MAX_FASTREG_DEPTH;
1170
1171         return 0;
1172 }
1173
1174 static int iwch_query_port(struct ib_device *ibdev,
1175                            u8 port, struct ib_port_attr *props)
1176 {
1177         struct iwch_dev *dev;
1178         struct net_device *netdev;
1179         struct in_device *inetdev;
1180
1181         PDBG("%s ibdev %p\n", __func__, ibdev);
1182
1183         dev = to_iwch_dev(ibdev);
1184         netdev = dev->rdev.port_info.lldevs[port-1];
1185
1186         memset(props, 0, sizeof(struct ib_port_attr));
1187         props->max_mtu = IB_MTU_4096;
1188         if (netdev->mtu >= 4096)
1189                 props->active_mtu = IB_MTU_4096;
1190         else if (netdev->mtu >= 2048)
1191                 props->active_mtu = IB_MTU_2048;
1192         else if (netdev->mtu >= 1024)
1193                 props->active_mtu = IB_MTU_1024;
1194         else if (netdev->mtu >= 512)
1195                 props->active_mtu = IB_MTU_512;
1196         else
1197                 props->active_mtu = IB_MTU_256;
1198
1199         if (!netif_carrier_ok(netdev))
1200                 props->state = IB_PORT_DOWN;
1201         else {
1202                 inetdev = in_dev_get(netdev);
1203                 if (inetdev) {
1204                         if (inetdev->ifa_list)
1205                                 props->state = IB_PORT_ACTIVE;
1206                         else
1207                                 props->state = IB_PORT_INIT;
1208                         in_dev_put(inetdev);
1209                 } else
1210                         props->state = IB_PORT_INIT;
1211         }
1212
1213         props->port_cap_flags =
1214             IB_PORT_CM_SUP |
1215             IB_PORT_SNMP_TUNNEL_SUP |
1216             IB_PORT_REINIT_SUP |
1217             IB_PORT_DEVICE_MGMT_SUP |
1218             IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
1219         props->gid_tbl_len = 1;
1220         props->pkey_tbl_len = 1;
1221         props->active_width = 2;
1222         props->active_speed = 2;
1223         props->max_msg_sz = -1;
1224
1225         return 0;
1226 }
1227
1228 static ssize_t show_rev(struct device *dev, struct device_attribute *attr,
1229                         char *buf)
1230 {
1231         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1232                                                  ibdev.dev);
1233         PDBG("%s dev 0x%p\n", __func__, dev);
1234         return sprintf(buf, "%d\n", iwch_dev->rdev.t3cdev_p->type);
1235 }
1236
1237 static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, char *buf)
1238 {
1239         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1240                                                  ibdev.dev);
1241         struct ethtool_drvinfo info;
1242         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1243
1244         PDBG("%s dev 0x%p\n", __func__, dev);
1245         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1246         return sprintf(buf, "%s\n", info.fw_version);
1247 }
1248
1249 static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
1250                         char *buf)
1251 {
1252         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1253                                                  ibdev.dev);
1254         struct ethtool_drvinfo info;
1255         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1256
1257         PDBG("%s dev 0x%p\n", __func__, dev);
1258         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1259         return sprintf(buf, "%s\n", info.driver);
1260 }
1261
1262 static ssize_t show_board(struct device *dev, struct device_attribute *attr,
1263                           char *buf)
1264 {
1265         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1266                                                  ibdev.dev);
1267         PDBG("%s dev 0x%p\n", __func__, dev);
1268         return sprintf(buf, "%x.%x\n", iwch_dev->rdev.rnic_info.pdev->vendor,
1269                        iwch_dev->rdev.rnic_info.pdev->device);
1270 }
1271
1272 static int iwch_get_mib(struct ib_device *ibdev,
1273                         union rdma_protocol_stats *stats)
1274 {
1275         struct iwch_dev *dev;
1276         struct tp_mib_stats m;
1277         int ret;
1278
1279         PDBG("%s ibdev %p\n", __func__, ibdev);
1280         dev = to_iwch_dev(ibdev);
1281         ret = dev->rdev.t3cdev_p->ctl(dev->rdev.t3cdev_p, RDMA_GET_MIB, &m);
1282         if (ret)
1283                 return -ENOSYS;
1284
1285         memset(stats, 0, sizeof *stats);
1286         stats->iw.ipInReceives = ((u64) m.ipInReceive_hi << 32) +
1287                                 m.ipInReceive_lo;
1288         stats->iw.ipInHdrErrors = ((u64) m.ipInHdrErrors_hi << 32) +
1289                                   m.ipInHdrErrors_lo;
1290         stats->iw.ipInAddrErrors = ((u64) m.ipInAddrErrors_hi << 32) +
1291                                    m.ipInAddrErrors_lo;
1292         stats->iw.ipInUnknownProtos = ((u64) m.ipInUnknownProtos_hi << 32) +
1293                                       m.ipInUnknownProtos_lo;
1294         stats->iw.ipInDiscards = ((u64) m.ipInDiscards_hi << 32) +
1295                                  m.ipInDiscards_lo;
1296         stats->iw.ipInDelivers = ((u64) m.ipInDelivers_hi << 32) +
1297                                  m.ipInDelivers_lo;
1298         stats->iw.ipOutRequests = ((u64) m.ipOutRequests_hi << 32) +
1299                                   m.ipOutRequests_lo;
1300         stats->iw.ipOutDiscards = ((u64) m.ipOutDiscards_hi << 32) +
1301                                   m.ipOutDiscards_lo;
1302         stats->iw.ipOutNoRoutes = ((u64) m.ipOutNoRoutes_hi << 32) +
1303                                   m.ipOutNoRoutes_lo;
1304         stats->iw.ipReasmTimeout = (u64) m.ipReasmTimeout;
1305         stats->iw.ipReasmReqds = (u64) m.ipReasmReqds;
1306         stats->iw.ipReasmOKs = (u64) m.ipReasmOKs;
1307         stats->iw.ipReasmFails = (u64) m.ipReasmFails;
1308         stats->iw.tcpActiveOpens = (u64) m.tcpActiveOpens;
1309         stats->iw.tcpPassiveOpens = (u64) m.tcpPassiveOpens;
1310         stats->iw.tcpAttemptFails = (u64) m.tcpAttemptFails;
1311         stats->iw.tcpEstabResets = (u64) m.tcpEstabResets;
1312         stats->iw.tcpOutRsts = (u64) m.tcpOutRsts;
1313         stats->iw.tcpCurrEstab = (u64) m.tcpCurrEstab;
1314         stats->iw.tcpInSegs = ((u64) m.tcpInSegs_hi << 32) +
1315                               m.tcpInSegs_lo;
1316         stats->iw.tcpOutSegs = ((u64) m.tcpOutSegs_hi << 32) +
1317                                m.tcpOutSegs_lo;
1318         stats->iw.tcpRetransSegs = ((u64) m.tcpRetransSeg_hi << 32) +
1319                                   m.tcpRetransSeg_lo;
1320         stats->iw.tcpInErrs = ((u64) m.tcpInErrs_hi << 32) +
1321                               m.tcpInErrs_lo;
1322         stats->iw.tcpRtoMin = (u64) m.tcpRtoMin;
1323         stats->iw.tcpRtoMax = (u64) m.tcpRtoMax;
1324         return 0;
1325 }
1326
1327 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1328 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1329 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1330 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1331
1332 static struct device_attribute *iwch_class_attributes[] = {
1333         &dev_attr_hw_rev,
1334         &dev_attr_fw_ver,
1335         &dev_attr_hca_type,
1336         &dev_attr_board_id,
1337 };
1338
1339 int iwch_register_device(struct iwch_dev *dev)
1340 {
1341         int ret;
1342         int i;
1343
1344         PDBG("%s iwch_dev %p\n", __func__, dev);
1345         strlcpy(dev->ibdev.name, "cxgb3_%d", IB_DEVICE_NAME_MAX);
1346         memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
1347         memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1348         dev->ibdev.owner = THIS_MODULE;
1349         dev->device_cap_flags = IB_DEVICE_LOCAL_DMA_LKEY |
1350                                 IB_DEVICE_MEM_WINDOW |
1351                                 IB_DEVICE_MEM_MGT_EXTENSIONS;
1352
1353         /* cxgb3 supports STag 0. */
1354         dev->ibdev.local_dma_lkey = 0;
1355
1356         dev->ibdev.uverbs_cmd_mask =
1357             (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1358             (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1359             (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1360             (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1361             (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1362             (1ull << IB_USER_VERBS_CMD_REG_MR) |
1363             (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1364             (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1365             (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1366             (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1367             (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1368             (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1369             (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1370             (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1371             (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1372             (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1373             (1ull << IB_USER_VERBS_CMD_POST_RECV);
1374         dev->ibdev.node_type = RDMA_NODE_RNIC;
1375         memcpy(dev->ibdev.node_desc, IWCH_NODE_DESC, sizeof(IWCH_NODE_DESC));
1376         dev->ibdev.phys_port_cnt = dev->rdev.port_info.nports;
1377         dev->ibdev.num_comp_vectors = 1;
1378         dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
1379         dev->ibdev.query_device = iwch_query_device;
1380         dev->ibdev.query_port = iwch_query_port;
1381         dev->ibdev.modify_port = iwch_modify_port;
1382         dev->ibdev.query_pkey = iwch_query_pkey;
1383         dev->ibdev.query_gid = iwch_query_gid;
1384         dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
1385         dev->ibdev.dealloc_ucontext = iwch_dealloc_ucontext;
1386         dev->ibdev.mmap = iwch_mmap;
1387         dev->ibdev.alloc_pd = iwch_allocate_pd;
1388         dev->ibdev.dealloc_pd = iwch_deallocate_pd;
1389         dev->ibdev.create_ah = iwch_ah_create;
1390         dev->ibdev.destroy_ah = iwch_ah_destroy;
1391         dev->ibdev.create_qp = iwch_create_qp;
1392         dev->ibdev.modify_qp = iwch_ib_modify_qp;
1393         dev->ibdev.destroy_qp = iwch_destroy_qp;
1394         dev->ibdev.create_cq = iwch_create_cq;
1395         dev->ibdev.destroy_cq = iwch_destroy_cq;
1396         dev->ibdev.resize_cq = iwch_resize_cq;
1397         dev->ibdev.poll_cq = iwch_poll_cq;
1398         dev->ibdev.get_dma_mr = iwch_get_dma_mr;
1399         dev->ibdev.reg_phys_mr = iwch_register_phys_mem;
1400         dev->ibdev.rereg_phys_mr = iwch_reregister_phys_mem;
1401         dev->ibdev.reg_user_mr = iwch_reg_user_mr;
1402         dev->ibdev.dereg_mr = iwch_dereg_mr;
1403         dev->ibdev.alloc_mw = iwch_alloc_mw;
1404         dev->ibdev.bind_mw = iwch_bind_mw;
1405         dev->ibdev.dealloc_mw = iwch_dealloc_mw;
1406         dev->ibdev.alloc_fast_reg_mr = iwch_alloc_fast_reg_mr;
1407         dev->ibdev.alloc_fast_reg_page_list = iwch_alloc_fastreg_pbl;
1408         dev->ibdev.free_fast_reg_page_list = iwch_free_fastreg_pbl;
1409         dev->ibdev.attach_mcast = iwch_multicast_attach;
1410         dev->ibdev.detach_mcast = iwch_multicast_detach;
1411         dev->ibdev.process_mad = iwch_process_mad;
1412         dev->ibdev.req_notify_cq = iwch_arm_cq;
1413         dev->ibdev.post_send = iwch_post_send;
1414         dev->ibdev.post_recv = iwch_post_receive;
1415         dev->ibdev.get_protocol_stats = iwch_get_mib;
1416
1417         dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
1418         if (!dev->ibdev.iwcm)
1419                 return -ENOMEM;
1420
1421         dev->ibdev.iwcm->connect = iwch_connect;
1422         dev->ibdev.iwcm->accept = iwch_accept_cr;
1423         dev->ibdev.iwcm->reject = iwch_reject_cr;
1424         dev->ibdev.iwcm->create_listen = iwch_create_listen;
1425         dev->ibdev.iwcm->destroy_listen = iwch_destroy_listen;
1426         dev->ibdev.iwcm->add_ref = iwch_qp_add_ref;
1427         dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
1428         dev->ibdev.iwcm->get_qp = iwch_get_qp;
1429
1430         ret = ib_register_device(&dev->ibdev);
1431         if (ret)
1432                 goto bail1;
1433
1434         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i) {
1435                 ret = device_create_file(&dev->ibdev.dev,
1436                                          iwch_class_attributes[i]);
1437                 if (ret) {
1438                         goto bail2;
1439                 }
1440         }
1441         return 0;
1442 bail2:
1443         ib_unregister_device(&dev->ibdev);
1444 bail1:
1445         kfree(dev->ibdev.iwcm);
1446         return ret;
1447 }
1448
1449 void iwch_unregister_device(struct iwch_dev *dev)
1450 {
1451         int i;
1452
1453         PDBG("%s iwch_dev %p\n", __func__, dev);
1454         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i)
1455                 device_remove_file(&dev->ibdev.dev,
1456                                    iwch_class_attributes[i]);
1457         ib_unregister_device(&dev->ibdev);
1458         kfree(dev->ibdev.iwcm);
1459         return;
1460 }