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 int iwch_modify_port(struct ib_device *ibdev,
  65                             u8 port, int port_modify_mask,
  66                             struct ib_port_modify *props)
  67 {
  68         return -ENOSYS;
  69 }
  70
  71 static struct ib_ah *iwch_ah_create(struct ib_pd *pd,
  72                                     struct ib_ah_attr *ah_attr)
  73 {
  74         return ERR_PTR(-ENOSYS);
  75 }
  76
  77 static int iwch_ah_destroy(struct ib_ah *ah)
  78 {
  79         return -ENOSYS;
  80 }
  81
  82 static int iwch_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
  83 {
  84         return -ENOSYS;
  85 }
  86
  87 static int iwch_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
  88 {
  89         return -ENOSYS;
  90 }
  91
  92 static int iwch_process_mad(struct ib_device *ibdev,
  93                             int mad_flags,
  94                             u8 port_num,
  95                             struct ib_wc *in_wc,
  96                             struct ib_grh *in_grh,
  97                             struct ib_mad *in_mad, struct ib_mad *out_mad)
  98 {
  99         return -ENOSYS;
 100 }
 101
 102 static int iwch_dealloc_ucontext(struct ib_ucontext *context)
 103 {
 104         struct iwch_dev *rhp = to_iwch_dev(context->device);
 105         struct iwch_ucontext *ucontext = to_iwch_ucontext(context);
 106         struct iwch_mm_entry *mm, *tmp;
 107
 108         PDBG("%s context %p\n", __func__, context);
 109         list_for_each_entry_safe(mm, tmp, &ucontext->mmaps, entry)
 110                 kfree(mm);
 111         cxio_release_ucontext(&rhp->rdev, &ucontext->uctx);
 112         kfree(ucontext);
 113         return 0;
 114 }
 115
 116 static struct ib_ucontext *iwch_alloc_ucontext(struct ib_device *ibdev,
 117                                         struct ib_udata *udata)
 118 {
 119         struct iwch_ucontext *context;
 120         struct iwch_dev *rhp = to_iwch_dev(ibdev);
 121
 122         PDBG("%s ibdev %p\n", __func__, ibdev);
 123         context = kzalloc(sizeof(*context), GFP_KERNEL);
 124         if (!context)
 125                 return ERR_PTR(-ENOMEM);
 126         cxio_init_ucontext(&rhp->rdev, &context->uctx);
 127         INIT_LIST_HEAD(&context->mmaps);
 128         spin_lock_init(&context->mmap_lock);
 129         return &context->ibucontext;
 130 }
 131
 132 static int iwch_destroy_cq(struct ib_cq *ib_cq)
 133 {
 134         struct iwch_cq *chp;
 135
 136         PDBG("%s ib_cq %p\n", __func__, ib_cq);
 137         chp = to_iwch_cq(ib_cq);
 138
 139         remove_handle(chp->rhp, &chp->rhp->cqidr, chp->cq.cqid);
 140         atomic_dec(&chp->refcnt);
 141         wait_event(chp->wait, !atomic_read(&chp->refcnt));
 142
 143         cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
 144         kfree(chp);
 145         return 0;
 146 }
 147
 148 static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries, int vector,
 149                              struct ib_ucontext *ib_context,
 150                              struct ib_udata *udata)
 151 {
 152         struct iwch_dev *rhp;
 153         struct iwch_cq *chp;
 154         struct iwch_create_cq_resp uresp;
 155         struct iwch_create_cq_req ureq;
 156         struct iwch_ucontext *ucontext = NULL;
 157
 158         PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
 159         rhp = to_iwch_dev(ibdev);
 160         chp = kzalloc(sizeof(*chp), GFP_KERNEL);
 161         if (!chp)
 162                 return ERR_PTR(-ENOMEM);
 163
 164         if (ib_context) {
 165                 ucontext = to_iwch_ucontext(ib_context);
 166                 if (!t3a_device(rhp)) {
 167                         if (ib_copy_from_udata(&ureq, udata, sizeof (ureq))) {
 168                                 kfree(chp);
 169                                 return ERR_PTR(-EFAULT);
 170                         }
 171                         chp->user_rptr_addr = (u32 __user *)(unsigned long)ureq.user_rptr_addr;
 172                 }
 173         }
 174
 175         if (t3a_device(rhp)) {
 176
 177                 /*
 178                  * T3A: Add some fluff to handle extra CQEs inserted
 179                  * for various errors.
 180                  * Additional CQE possibilities:
 181                  *      TERMINATE,
 182                  *      incoming RDMA WRITE Failures
 183                  *      incoming RDMA READ REQUEST FAILUREs
 184                  * NOTE: We cannot ensure the CQ won't overflow.
 185                  */
 186                 entries += 16;
 187         }
 188         entries = roundup_pow_of_two(entries);
 189         chp->cq.size_log2 = ilog2(entries);
 190
 191         if (cxio_create_cq(&rhp->rdev, &chp->cq, !ucontext)) {
 192                 kfree(chp);
 193                 return ERR_PTR(-ENOMEM);
 194         }
 195         chp->rhp = rhp;
 196         chp->ibcq.cqe = 1 << chp->cq.size_log2;
 197         spin_lock_init(&chp->lock);
 198         atomic_set(&chp->refcnt, 1);
 199         init_waitqueue_head(&chp->wait);
 200         if (insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid)) {
 201                 cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
 202                 kfree(chp);
 203                 return ERR_PTR(-ENOMEM);
 204         }
 205
 206         if (ucontext) {
 207                 struct iwch_mm_entry *mm;
 208
 209                 mm = kmalloc(sizeof *mm, GFP_KERNEL);
 210                 if (!mm) {
 211                         iwch_destroy_cq(&chp->ibcq);
 212                         return ERR_PTR(-ENOMEM);
 213                 }
 214                 uresp.cqid = chp->cq.cqid;
 215                 uresp.size_log2 = chp->cq.size_log2;
 216                 spin_lock(&ucontext->mmap_lock);
 217                 uresp.key = ucontext->key;
 218                 ucontext->key += PAGE_SIZE;
 219                 spin_unlock(&ucontext->mmap_lock);
 220                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 221                         kfree(mm);
 222                         iwch_destroy_cq(&chp->ibcq);
 223                         return ERR_PTR(-EFAULT);
 224                 }
 225                 mm->key = uresp.key;
 226                 mm->addr = virt_to_phys(chp->cq.queue);
 227                 mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
 228                                              sizeof (struct t3_cqe));
 229                 insert_mmap(ucontext, mm);
 230         }
 231         PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
 232              chp->cq.cqid, chp, (1 << chp->cq.size_log2),
 233              (unsigned long long) chp->cq.dma_addr);
 234         return &chp->ibcq;
 235 }
 236
 237 static int iwch_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
 238 {
 239 #ifdef notyet
 240         struct iwch_cq *chp = to_iwch_cq(cq);
 241         struct t3_cq oldcq, newcq;
 242         int ret;
 243
 244         PDBG("%s ib_cq %p cqe %d\n", __func__, cq, cqe);
 245
 246         /* We don't downsize... */
 247         if (cqe <= cq->cqe)
 248                 return 0;
 249
 250         /* create new t3_cq with new size */
 251         cqe = roundup_pow_of_two(cqe+1);
 252         newcq.size_log2 = ilog2(cqe);
 253
 254         /* Dont allow resize to less than the current wce count */
 255         if (cqe < Q_COUNT(chp->cq.rptr, chp->cq.wptr)) {
 256                 return -ENOMEM;
 257         }
 258
 259         /* Quiesce all QPs using this CQ */
 260         ret = iwch_quiesce_qps(chp);
 261         if (ret) {
 262                 return ret;
 263         }
 264
 265         ret = cxio_create_cq(&chp->rhp->rdev, &newcq);
 266         if (ret) {
 267                 return ret;
 268         }
 269
 270         /* copy CQEs */
 271         memcpy(newcq.queue, chp->cq.queue, (1 << chp->cq.size_log2) *
 272                                         sizeof(struct t3_cqe));
 273
 274         /* old iwch_qp gets new t3_cq but keeps old cqid */
 275         oldcq = chp->cq;
 276         chp->cq = newcq;
 277         chp->cq.cqid = oldcq.cqid;
 278
 279         /* resize new t3_cq to update the HW context */
 280         ret = cxio_resize_cq(&chp->rhp->rdev, &chp->cq);
 281         if (ret) {
 282                 chp->cq = oldcq;
 283                 return ret;
 284         }
 285         chp->ibcq.cqe = (1<<chp->cq.size_log2) - 1;
 286
 287         /* destroy old t3_cq */
 288         oldcq.cqid = newcq.cqid;
 289         ret = cxio_destroy_cq(&chp->rhp->rdev, &oldcq);
 290         if (ret) {
 291                 printk(KERN_ERR MOD "%s - cxio_destroy_cq failed %d\n",
 292                         __func__, ret);
 293         }
 294
 295         /* add user hooks here */
 296
 297         /* resume qps */
 298         ret = iwch_resume_qps(chp);
 299         return ret;
 300 #else
 301         return -ENOSYS;
 302 #endif
 303 }
 304
 305 static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
 306 {
 307         struct iwch_dev *rhp;
 308         struct iwch_cq *chp;
 309         enum t3_cq_opcode cq_op;
 310         int err;
 311         unsigned long flag;
 312         u32 rptr;
 313
 314         chp = to_iwch_cq(ibcq);
 315         rhp = chp->rhp;
 316         if ((flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED)
 317                 cq_op = CQ_ARM_SE;
 318         else
 319                 cq_op = CQ_ARM_AN;
 320         if (chp->user_rptr_addr) {
 321                 if (get_user(rptr, chp->user_rptr_addr))
 322                         return -EFAULT;
 323                 spin_lock_irqsave(&chp->lock, flag);
 324                 chp->cq.rptr = rptr;
 325         } else
 326                 spin_lock_irqsave(&chp->lock, flag);
 327         PDBG("%s rptr 0x%x\n", __func__, chp->cq.rptr);
 328         err = cxio_hal_cq_op(&rhp->rdev, &chp->cq, cq_op, 0);
 329         spin_unlock_irqrestore(&chp->lock, flag);
 330         if (err < 0)
 331                 printk(KERN_ERR MOD "Error %d rearming CQID 0x%x\n", err,
 332                        chp->cq.cqid);
 333         if (err > 0 && !(flags & IB_CQ_REPORT_MISSED_EVENTS))
 334                 err = 0;
 335         return err;
 336 }
 337
 338 static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 339 {
 340         int len = vma->vm_end - vma->vm_start;
 341         u32 key = vma->vm_pgoff << PAGE_SHIFT;
 342         struct cxio_rdev *rdev_p;
 343         int ret = 0;
 344         struct iwch_mm_entry *mm;
 345         struct iwch_ucontext *ucontext;
 346         u64 addr;
 347
 348         PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __func__, vma->vm_pgoff,
 349              key, len);
 350
 351         if (vma->vm_start & (PAGE_SIZE-1)) {
 352                 return -EINVAL;
 353         }
 354
 355         rdev_p = &(to_iwch_dev(context->device)->rdev);
 356         ucontext = to_iwch_ucontext(context);
 357
 358         mm = remove_mmap(ucontext, key, len);
 359         if (!mm)
 360                 return -EINVAL;
 361         addr = mm->addr;
 362         kfree(mm);
 363
 364         if ((addr >= rdev_p->rnic_info.udbell_physbase) &&
 365             (addr < (rdev_p->rnic_info.udbell_physbase +
 366                        rdev_p->rnic_info.udbell_len))) {
 367
 368                 /*
 369                  * Map T3 DB register.
 370                  */
 371                 if (vma->vm_flags & VM_READ) {
 372                         return -EPERM;
 373                 }
 374
 375                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 376                 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
 377                 vma->vm_flags &= ~VM_MAYREAD;
 378                 ret = io_remap_pfn_range(vma, vma->vm_start,
 379                                          addr >> PAGE_SHIFT,
 380                                          len, vma->vm_page_prot);
 381         } else {
 382
 383                 /*
 384                  * Map WQ or CQ contig dma memory...
 385                  */
 386                 ret = remap_pfn_range(vma, vma->vm_start,
 387                                       addr >> PAGE_SHIFT,
 388                                       len, vma->vm_page_prot);
 389         }
 390
 391         return ret;
 392 }
 393
 394 static int iwch_deallocate_pd(struct ib_pd *pd)
 395 {
 396         struct iwch_dev *rhp;
 397         struct iwch_pd *php;
 398
 399         php = to_iwch_pd(pd);
 400         rhp = php->rhp;
 401         PDBG("%s ibpd %p pdid 0x%x\n", __func__, pd, php->pdid);
 402         cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid);
 403         kfree(php);
 404         return 0;
 405 }
 406
 407 static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev,
 408                                struct ib_ucontext *context,
 409                                struct ib_udata *udata)
 410 {
 411         struct iwch_pd *php;
 412         u32 pdid;
 413         struct iwch_dev *rhp;
 414
 415         PDBG("%s ibdev %p\n", __func__, ibdev);
 416         rhp = (struct iwch_dev *) ibdev;
 417         pdid = cxio_hal_get_pdid(rhp->rdev.rscp);
 418         if (!pdid)
 419                 return ERR_PTR(-EINVAL);
 420         php = kzalloc(sizeof(*php), GFP_KERNEL);
 421         if (!php) {
 422                 cxio_hal_put_pdid(rhp->rdev.rscp, pdid);
 423                 return ERR_PTR(-ENOMEM);
 424         }
 425         php->pdid = pdid;
 426         php->rhp = rhp;
 427         if (context) {
 428                 if (ib_copy_to_udata(udata, &php->pdid, sizeof (__u32))) {
 429                         iwch_deallocate_pd(&php->ibpd);
 430                         return ERR_PTR(-EFAULT);
 431                 }
 432         }
 433         PDBG("%s pdid 0x%0x ptr 0x%p\n", __func__, pdid, php);
 434         return &php->ibpd;
 435 }
 436
 437 static int iwch_dereg_mr(struct ib_mr *ib_mr)
 438 {
 439         struct iwch_dev *rhp;
 440         struct iwch_mr *mhp;
 441         u32 mmid;
 442
 443         PDBG("%s ib_mr %p\n", __func__, ib_mr);
 444         /* There can be no memory windows */
 445         if (atomic_read(&ib_mr->usecnt))
 446                 return -EINVAL;
 447
 448         mhp = to_iwch_mr(ib_mr);
 449         rhp = mhp->rhp;
 450         mmid = mhp->attr.stag >> 8;
 451         cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
 452                        mhp->attr.pbl_addr);
 453         iwch_free_pbl(mhp);
 454         remove_handle(rhp, &rhp->mmidr, mmid);
 455         if (mhp->kva)
 456                 kfree((void *) (unsigned long) mhp->kva);
 457         if (mhp->umem)
 458                 ib_umem_release(mhp->umem);
 459         PDBG("%s mmid 0x%x ptr %p\n", __func__, mmid, mhp);
 460         kfree(mhp);
 461         return 0;
 462 }
 463
 464 static struct ib_mr *iwch_register_phys_mem(struct ib_pd *pd,
 465                                         struct ib_phys_buf *buffer_list,
 466                                         int num_phys_buf,
 467                                         int acc,
 468                                         u64 *iova_start)
 469 {
 470         __be64 *page_list;
 471         int shift;
 472         u64 total_size;
 473         int npages;
 474         struct iwch_dev *rhp;
 475         struct iwch_pd *php;
 476         struct iwch_mr *mhp;
 477         int ret;
 478
 479         PDBG("%s ib_pd %p\n", __func__, pd);
 480         php = to_iwch_pd(pd);
 481         rhp = php->rhp;
 482
 483         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 484         if (!mhp)
 485                 return ERR_PTR(-ENOMEM);
 486
 487         mhp->rhp = rhp;
 488
 489         /* First check that we have enough alignment */
 490         if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
 491                 ret = -EINVAL;
 492                 goto err;
 493         }
 494
 495         if (num_phys_buf > 1 &&
 496             ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
 497                 ret = -EINVAL;
 498                 goto err;
 499         }
 500
 501         ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
 502                                    &total_size, &npages, &shift, &page_list);
 503         if (ret)
 504                 goto err;
 505
 506         ret = iwch_alloc_pbl(mhp, npages);
 507         if (ret) {
 508                 kfree(page_list);
 509                 goto err_pbl;
 510         }
 511
 512         ret = iwch_write_pbl(mhp, page_list, npages, 0);
 513         kfree(page_list);
 514         if (ret)
 515                 goto err_pbl;
 516
 517         mhp->attr.pdid = php->pdid;
 518         mhp->attr.zbva = 0;
 519
 520         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 521         mhp->attr.va_fbo = *iova_start;
 522         mhp->attr.page_size = shift - 12;
 523
 524         mhp->attr.len = (u32) total_size;
 525         mhp->attr.pbl_size = npages;
 526         ret = iwch_register_mem(rhp, php, mhp, shift);
 527         if (ret)
 528                 goto err_pbl;
 529
 530         return &mhp->ibmr;
 531
 532 err_pbl:
 533         iwch_free_pbl(mhp);
 534
 535 err:
 536         kfree(mhp);
 537         return ERR_PTR(ret);
 538
 539 }
 540
 541 static int iwch_reregister_phys_mem(struct ib_mr *mr,
 542                                      int mr_rereg_mask,
 543                                      struct ib_pd *pd,
 544                                      struct ib_phys_buf *buffer_list,
 545                                      int num_phys_buf,
 546                                      int acc, u64 * iova_start)
 547 {
 548
 549         struct iwch_mr mh, *mhp;
 550         struct iwch_pd *php;
 551         struct iwch_dev *rhp;
 552         __be64 *page_list = NULL;
 553         int shift = 0;
 554         u64 total_size;
 555         int npages;
 556         int ret;
 557
 558         PDBG("%s ib_mr %p ib_pd %p\n", __func__, mr, pd);
 559
 560         /* There can be no memory windows */
 561         if (atomic_read(&mr->usecnt))
 562                 return -EINVAL;
 563
 564         mhp = to_iwch_mr(mr);
 565         rhp = mhp->rhp;
 566         php = to_iwch_pd(mr->pd);
 567
 568         /* make sure we are on the same adapter */
 569         if (rhp != php->rhp)
 570                 return -EINVAL;
 571
 572         memcpy(&mh, mhp, sizeof *mhp);
 573
 574         if (mr_rereg_mask & IB_MR_REREG_PD)
 575                 php = to_iwch_pd(pd);
 576         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 577                 mh.attr.perms = iwch_ib_to_tpt_access(acc);
 578         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 579                 ret = build_phys_page_list(buffer_list, num_phys_buf,
 580                                            iova_start,
 581                                            &total_size, &npages,
 582                                            &shift, &page_list);
 583                 if (ret)
 584                         return ret;
 585         }
 586
 587         ret = iwch_reregister_mem(rhp, php, &mh, shift, npages);
 588         kfree(page_list);
 589         if (ret) {
 590                 return ret;
 591         }
 592         if (mr_rereg_mask & IB_MR_REREG_PD)
 593                 mhp->attr.pdid = php->pdid;
 594         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 595                 mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 596         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 597                 mhp->attr.zbva = 0;
 598                 mhp->attr.va_fbo = *iova_start;
 599                 mhp->attr.page_size = shift - 12;
 600                 mhp->attr.len = (u32) total_size;
 601                 mhp->attr.pbl_size = npages;
 602         }
 603
 604         return 0;
 605 }
 606
 607
 608 static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 609                                       u64 virt, int acc, struct ib_udata *udata)
 610 {
 611         __be64 *pages;
 612         int shift, n, len;
 613         int i, j, k;
 614         int err = 0;
 615         struct ib_umem_chunk *chunk;
 616         struct iwch_dev *rhp;
 617         struct iwch_pd *php;
 618         struct iwch_mr *mhp;
 619         struct iwch_reg_user_mr_resp uresp;
 620
 621         PDBG("%s ib_pd %p\n", __func__, pd);
 622
 623         php = to_iwch_pd(pd);
 624         rhp = php->rhp;
 625         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 626         if (!mhp)
 627                 return ERR_PTR(-ENOMEM);
 628
 629         mhp->rhp = rhp;
 630
 631         mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
 632         if (IS_ERR(mhp->umem)) {
 633                 err = PTR_ERR(mhp->umem);
 634                 kfree(mhp);
 635                 return ERR_PTR(err);
 636         }
 637
 638         shift = ffs(mhp->umem->page_size) - 1;
 639
 640         n = 0;
 641         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 642                 n += chunk->nents;
 643
 644         err = iwch_alloc_pbl(mhp, n);
 645         if (err)
 646                 goto err;
 647
 648         pages = (__be64 *) __get_free_page(GFP_KERNEL);
 649         if (!pages) {
 650                 err = -ENOMEM;
 651                 goto err_pbl;
 652         }
 653
 654         i = n = 0;
 655
 656         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 657                 for (j = 0; j < chunk->nmap; ++j) {
 658                         len = sg_dma_len(&chunk->page_list[j]) >> shift;
 659                         for (k = 0; k < len; ++k) {
 660                                 pages[i++] = cpu_to_be64(sg_dma_address(
 661                                         &chunk->page_list[j]) +
 662                                         mhp->umem->page_size * k);
 663                                 if (i == PAGE_SIZE / sizeof *pages) {
 664                                         err = iwch_write_pbl(mhp, pages, i, n);
 665                                         if (err)
 666                                                 goto pbl_done;
 667                                         n += i;
 668                                         i = 0;
 669                                 }
 670                         }
 671                 }
 672
 673         if (i)
 674                 err = iwch_write_pbl(mhp, pages, i, n);
 675
 676 pbl_done:
 677         free_page((unsigned long) pages);
 678         if (err)
 679                 goto err_pbl;
 680
 681         mhp->attr.pdid = php->pdid;
 682         mhp->attr.zbva = 0;
 683         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 684         mhp->attr.va_fbo = virt;
 685         mhp->attr.page_size = shift - 12;
 686         mhp->attr.len = (u32) length;
 687
 688         err = iwch_register_mem(rhp, php, mhp, shift);
 689         if (err)
 690                 goto err_pbl;
 691
 692         if (udata && !t3a_device(rhp)) {
 693                 uresp.pbl_addr = (mhp->attr.pbl_addr -
 694                                  rhp->rdev.rnic_info.pbl_base) >> 3;
 695                 PDBG("%s user resp pbl_addr 0x%x\n", __func__,
 696                      uresp.pbl_addr);
 697
 698                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 699                         iwch_dereg_mr(&mhp->ibmr);
 700                         err = -EFAULT;
 701                         goto err;
 702                 }
 703         }
 704
 705         return &mhp->ibmr;
 706
 707 err_pbl:
 708         iwch_free_pbl(mhp);
 709
 710 err:
 711         ib_umem_release(mhp->umem);
 712         kfree(mhp);
 713         return ERR_PTR(err);
 714 }
 715
 716 static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
 717 {
 718         struct ib_phys_buf bl;
 719         u64 kva;
 720         struct ib_mr *ibmr;
 721
 722         PDBG("%s ib_pd %p\n", __func__, pd);
 723
 724         /*
 725          * T3 only supports 32 bits of size.
 726          */
 727         bl.size = 0xffffffff;
 728         bl.addr = 0;
 729         kva = 0;
 730         ibmr = iwch_register_phys_mem(pd, &bl, 1, acc, &kva);
 731         return ibmr;
 732 }
 733
 734 static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
 735 {
 736         struct iwch_dev *rhp;
 737         struct iwch_pd *php;
 738         struct iwch_mw *mhp;
 739         u32 mmid;
 740         u32 stag = 0;
 741         int ret;
 742
 743         php = to_iwch_pd(pd);
 744         rhp = php->rhp;
 745         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 746         if (!mhp)
 747                 return ERR_PTR(-ENOMEM);
 748         ret = cxio_allocate_window(&rhp->rdev, &stag, php->pdid);
 749         if (ret) {
 750                 kfree(mhp);
 751                 return ERR_PTR(ret);
 752         }
 753         mhp->rhp = rhp;
 754         mhp->attr.pdid = php->pdid;
 755         mhp->attr.type = TPT_MW;
 756         mhp->attr.stag = stag;
 757         mmid = (stag) >> 8;
 758         mhp->ibmw.rkey = stag;
 759         if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) {
 760                 cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 761                 kfree(mhp);
 762                 return ERR_PTR(-ENOMEM);
 763         }
 764         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 765         return &(mhp->ibmw);
 766 }
 767
 768 static int iwch_dealloc_mw(struct ib_mw *mw)
 769 {
 770         struct iwch_dev *rhp;
 771         struct iwch_mw *mhp;
 772         u32 mmid;
 773
 774         mhp = to_iwch_mw(mw);
 775         rhp = mhp->rhp;
 776         mmid = (mw->rkey) >> 8;
 777         cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 778         remove_handle(rhp, &rhp->mmidr, mmid);
 779         kfree(mhp);
 780         PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
 781         return 0;
 782 }
 783
 784 static struct ib_mr *iwch_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth)
 785 {
 786         struct iwch_dev *rhp;
 787         struct iwch_pd *php;
 788         struct iwch_mr *mhp;
 789         u32 mmid;
 790         u32 stag = 0;
 791         int ret = 0;
 792
 793         php = to_iwch_pd(pd);
 794         rhp = php->rhp;
 795         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 796         if (!mhp)
 797                 goto err;
 798
 799         mhp->rhp = rhp;
 800         ret = iwch_alloc_pbl(mhp, pbl_depth);
 801         if (ret)
 802                 goto err1;
 803         mhp->attr.pbl_size = pbl_depth;
 804         ret = cxio_allocate_stag(&rhp->rdev, &stag, php->pdid,
 805                                  mhp->attr.pbl_size, mhp->attr.pbl_addr);
 806         if (ret)
 807                 goto err2;
 808         mhp->attr.pdid = php->pdid;
 809         mhp->attr.type = TPT_NON_SHARED_MR;
 810         mhp->attr.stag = stag;
 811         mhp->attr.state = 1;
 812         mmid = (stag) >> 8;
 813         mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
 814         if (insert_handle(rhp, &rhp->mmidr, mhp, mmid))
 815                 goto err3;
 816
 817         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 818         return &(mhp->ibmr);
 819 err3:
 820         cxio_dereg_mem(&rhp->rdev, stag, mhp->attr.pbl_size,
 821                        mhp->attr.pbl_addr);
 822 err2:
 823         iwch_free_pbl(mhp);
 824 err1:
 825         kfree(mhp);
 826 err:
 827         return ERR_PTR(ret);
 828 }
 829
 830 static struct ib_fast_reg_page_list *iwch_alloc_fastreg_pbl(
 831                                         struct ib_device *device,
 832                                         int page_list_len)
 833 {
 834         struct ib_fast_reg_page_list *page_list;
 835
 836         page_list = kmalloc(sizeof *page_list + page_list_len * sizeof(u64),
 837                             GFP_KERNEL);
 838         if (!page_list)
 839                 return ERR_PTR(-ENOMEM);
 840
 841         page_list->page_list = (u64 *)(page_list + 1);
 842         page_list->max_page_list_len = page_list_len;
 843
 844         return page_list;
 845 }
 846
 847 static void iwch_free_fastreg_pbl(struct ib_fast_reg_page_list *page_list)
 848 {
 849         kfree(page_list);
 850 }
 851
 852 static int iwch_destroy_qp(struct ib_qp *ib_qp)
 853 {
 854         struct iwch_dev *rhp;
 855         struct iwch_qp *qhp;
 856         struct iwch_qp_attributes attrs;
 857         struct iwch_ucontext *ucontext;
 858
 859         qhp = to_iwch_qp(ib_qp);
 860         rhp = qhp->rhp;
 861
 862         attrs.next_state = IWCH_QP_STATE_ERROR;
 863         iwch_modify_qp(rhp, qhp, IWCH_QP_ATTR_NEXT_STATE, &attrs, 0);
 864         wait_event(qhp->wait, !qhp->ep);
 865
 866         remove_handle(rhp, &rhp->qpidr, qhp->wq.qpid);
 867
 868         atomic_dec(&qhp->refcnt);
 869         wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
 870
 871         ucontext = ib_qp->uobject ? to_iwch_ucontext(ib_qp->uobject->context)
 872                                   : NULL;
 873         cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 874                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 875
 876         PDBG("%s ib_qp %p qpid 0x%0x qhp %p\n", __func__,
 877              ib_qp, qhp->wq.qpid, qhp);
 878         kfree(qhp);
 879         return 0;
 880 }
 881
 882 static struct ib_qp *iwch_create_qp(struct ib_pd *pd,
 883                              struct ib_qp_init_attr *attrs,
 884                              struct ib_udata *udata)
 885 {
 886         struct iwch_dev *rhp;
 887         struct iwch_qp *qhp;
 888         struct iwch_pd *php;
 889         struct iwch_cq *schp;
 890         struct iwch_cq *rchp;
 891         struct iwch_create_qp_resp uresp;
 892         int wqsize, sqsize, rqsize;
 893         struct iwch_ucontext *ucontext;
 894
 895         PDBG("%s ib_pd %p\n", __func__, pd);
 896         if (attrs->qp_type != IB_QPT_RC)
 897                 return ERR_PTR(-EINVAL);
 898         php = to_iwch_pd(pd);
 899         rhp = php->rhp;
 900         schp = get_chp(rhp, ((struct iwch_cq *) attrs->send_cq)->cq.cqid);
 901         rchp = get_chp(rhp, ((struct iwch_cq *) attrs->recv_cq)->cq.cqid);
 902         if (!schp || !rchp)
 903                 return ERR_PTR(-EINVAL);
 904
 905         /* The RQT size must be # of entries + 1 rounded up to a power of two */
 906         rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr);
 907         if (rqsize == attrs->cap.max_recv_wr)
 908                 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr+1);
 909
 910         /* T3 doesn't support RQT depth < 16 */
 911         if (rqsize < 16)
 912                 rqsize = 16;
 913
 914         if (rqsize > T3_MAX_RQ_SIZE)
 915                 return ERR_PTR(-EINVAL);
 916
 917         if (attrs->cap.max_inline_data > T3_MAX_INLINE)
 918                 return ERR_PTR(-EINVAL);
 919
 920         /*
 921          * NOTE: The SQ and total WQ sizes don't need to be
 922          * a power of two.  However, all the code assumes
 923          * they are. EG: Q_FREECNT() and friends.
 924          */
 925         sqsize = roundup_pow_of_two(attrs->cap.max_send_wr);
 926         wqsize = roundup_pow_of_two(rqsize + sqsize);
 927
 928         /*
 929          * Kernel users need more wq space for fastreg WRs which can take
 930          * 2 WR fragments.
 931          */
 932         ucontext = pd->uobject ? to_iwch_ucontext(pd->uobject->context) : NULL;
 933         if (!ucontext && wqsize < (rqsize + (2 * sqsize)))
 934                 wqsize = roundup_pow_of_two(rqsize +
 935                                 roundup_pow_of_two(attrs->cap.max_send_wr * 2));
 936         PDBG("%s wqsize %d sqsize %d rqsize %d\n", __func__,
 937              wqsize, sqsize, rqsize);
 938         qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
 939         if (!qhp)
 940                 return ERR_PTR(-ENOMEM);
 941         qhp->wq.size_log2 = ilog2(wqsize);
 942         qhp->wq.rq_size_log2 = ilog2(rqsize);
 943         qhp->wq.sq_size_log2 = ilog2(sqsize);
 944         if (cxio_create_qp(&rhp->rdev, !udata, &qhp->wq,
 945                            ucontext ? &ucontext->uctx : &rhp->rdev.uctx)) {
 946                 kfree(qhp);
 947                 return ERR_PTR(-ENOMEM);
 948         }
 949
 950         attrs->cap.max_recv_wr = rqsize - 1;
 951         attrs->cap.max_send_wr = sqsize;
 952         attrs->cap.max_inline_data = T3_MAX_INLINE;
 953
 954         qhp->rhp = rhp;
 955         qhp->attr.pd = php->pdid;
 956         qhp->attr.scq = ((struct iwch_cq *) attrs->send_cq)->cq.cqid;
 957         qhp->attr.rcq = ((struct iwch_cq *) attrs->recv_cq)->cq.cqid;
 958         qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
 959         qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
 960         qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
 961         qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
 962         qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
 963         qhp->attr.state = IWCH_QP_STATE_IDLE;
 964         qhp->attr.next_state = IWCH_QP_STATE_IDLE;
 965
 966         /*
 967          * XXX - These don't get passed in from the openib user
 968          * at create time.  The CM sets them via a QP modify.
 969          * Need to fix...  I think the CM should
 970          */
 971         qhp->attr.enable_rdma_read = 1;
 972         qhp->attr.enable_rdma_write = 1;
 973         qhp->attr.enable_bind = 1;
 974         qhp->attr.max_ord = 1;
 975         qhp->attr.max_ird = 1;
 976
 977         spin_lock_init(&qhp->lock);
 978         init_waitqueue_head(&qhp->wait);
 979         atomic_set(&qhp->refcnt, 1);
 980
 981         if (insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid)) {
 982                 cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 983                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 984                 kfree(qhp);
 985                 return ERR_PTR(-ENOMEM);
 986         }
 987
 988         if (udata) {
 989
 990                 struct iwch_mm_entry *mm1, *mm2;
 991
 992                 mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
 993                 if (!mm1) {
 994                         iwch_destroy_qp(&qhp->ibqp);
 995                         return ERR_PTR(-ENOMEM);
 996                 }
 997
 998                 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
 999                 if (!mm2) {
1000                         kfree(mm1);
1001                         iwch_destroy_qp(&qhp->ibqp);
1002                         return ERR_PTR(-ENOMEM);
1003                 }
1004
1005                 uresp.qpid = qhp->wq.qpid;
1006                 uresp.size_log2 = qhp->wq.size_log2;
1007                 uresp.sq_size_log2 = qhp->wq.sq_size_log2;
1008                 uresp.rq_size_log2 = qhp->wq.rq_size_log2;
1009                 spin_lock(&ucontext->mmap_lock);
1010                 uresp.key = ucontext->key;
1011                 ucontext->key += PAGE_SIZE;
1012                 uresp.db_key = ucontext->key;
1013                 ucontext->key += PAGE_SIZE;
1014                 spin_unlock(&ucontext->mmap_lock);
1015                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
1016                         kfree(mm1);
1017                         kfree(mm2);
1018                         iwch_destroy_qp(&qhp->ibqp);
1019                         return ERR_PTR(-EFAULT);
1020                 }
1021                 mm1->key = uresp.key;
1022                 mm1->addr = virt_to_phys(qhp->wq.queue);
1023                 mm1->len = PAGE_ALIGN(wqsize * sizeof (union t3_wr));
1024                 insert_mmap(ucontext, mm1);
1025                 mm2->key = uresp.db_key;
1026                 mm2->addr = qhp->wq.udb & PAGE_MASK;
1027                 mm2->len = PAGE_SIZE;
1028                 insert_mmap(ucontext, mm2);
1029         }
1030         qhp->ibqp.qp_num = qhp->wq.qpid;
1031         init_timer(&(qhp->timer));
1032         PDBG("%s sq_num_entries %d, rq_num_entries %d "
1033              "qpid 0x%0x qhp %p dma_addr 0x%llx size %d rq_addr 0x%x\n",
1034              __func__, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
1035              qhp->wq.qpid, qhp, (unsigned long long) qhp->wq.dma_addr,
1036              1 << qhp->wq.size_log2, qhp->wq.rq_addr);
1037         return &qhp->ibqp;
1038 }
1039
1040 static int iwch_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1041                       int attr_mask, struct ib_udata *udata)
1042 {
1043         struct iwch_dev *rhp;
1044         struct iwch_qp *qhp;
1045         enum iwch_qp_attr_mask mask = 0;
1046         struct iwch_qp_attributes attrs;
1047
1048         PDBG("%s ib_qp %p\n", __func__, ibqp);
1049
1050         /* iwarp does not support the RTR state */
1051         if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
1052                 attr_mask &= ~IB_QP_STATE;
1053
1054         /* Make sure we still have something left to do */
1055         if (!attr_mask)
1056                 return 0;
1057
1058         memset(&attrs, 0, sizeof attrs);
1059         qhp = to_iwch_qp(ibqp);
1060         rhp = qhp->rhp;
1061
1062         attrs.next_state = iwch_convert_state(attr->qp_state);
1063         attrs.enable_rdma_read = (attr->qp_access_flags &
1064                                IB_ACCESS_REMOTE_READ) ?  1 : 0;
1065         attrs.enable_rdma_write = (attr->qp_access_flags &
1066                                 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
1067         attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
1068
1069
1070         mask |= (attr_mask & IB_QP_STATE) ? IWCH_QP_ATTR_NEXT_STATE : 0;
1071         mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
1072                         (IWCH_QP_ATTR_ENABLE_RDMA_READ |
1073                          IWCH_QP_ATTR_ENABLE_RDMA_WRITE |
1074                          IWCH_QP_ATTR_ENABLE_RDMA_BIND) : 0;
1075
1076         return iwch_modify_qp(rhp, qhp, mask, &attrs, 0);
1077 }
1078
1079 void iwch_qp_add_ref(struct ib_qp *qp)
1080 {
1081         PDBG("%s ib_qp %p\n", __func__, qp);
1082         atomic_inc(&(to_iwch_qp(qp)->refcnt));
1083 }
1084
1085 void iwch_qp_rem_ref(struct ib_qp *qp)
1086 {
1087         PDBG("%s ib_qp %p\n", __func__, qp);
1088         if (atomic_dec_and_test(&(to_iwch_qp(qp)->refcnt)))
1089                 wake_up(&(to_iwch_qp(qp)->wait));
1090 }
1091
1092 static struct ib_qp *iwch_get_qp(struct ib_device *dev, int qpn)
1093 {
1094         PDBG("%s ib_dev %p qpn 0x%x\n", __func__, dev, qpn);
1095         return (struct ib_qp *)get_qhp(to_iwch_dev(dev), qpn);
1096 }
1097
1098
1099 static int iwch_query_pkey(struct ib_device *ibdev,
1100                            u8 port, u16 index, u16 * pkey)
1101 {
1102         PDBG("%s ibdev %p\n", __func__, ibdev);
1103         *pkey = 0;
1104         return 0;
1105 }
1106
1107 static int iwch_query_gid(struct ib_device *ibdev, u8 port,
1108                           int index, union ib_gid *gid)
1109 {
1110         struct iwch_dev *dev;
1111
1112         PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
1113                __func__, ibdev, port, index, gid);
1114         dev = to_iwch_dev(ibdev);
1115         BUG_ON(port == 0 || port > 2);
1116         memset(&(gid->raw[0]), 0, sizeof(gid->raw));
1117         memcpy(&(gid->raw[0]), dev->rdev.port_info.lldevs[port-1]->dev_addr, 6);
1118         return 0;
1119 }
1120
1121 static u64 fw_vers_string_to_u64(struct iwch_dev *iwch_dev)
1122 {
1123         struct ethtool_drvinfo info;
1124         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1125         char *cp, *next;
1126         unsigned fw_maj, fw_min, fw_mic;
1127
1128         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1129
1130         next = info.fw_version + 1;
1131         cp = strsep(&next, ".");
1132         sscanf(cp, "%i", &fw_maj);
1133         cp = strsep(&next, ".");
1134         sscanf(cp, "%i", &fw_min);
1135         cp = strsep(&next, ".");
1136         sscanf(cp, "%i", &fw_mic);
1137
1138         return (((u64)fw_maj & 0xffff) << 32) | ((fw_min & 0xffff) << 16) |
1139                (fw_mic & 0xffff);
1140 }
1141
1142 static int iwch_query_device(struct ib_device *ibdev,
1143                              struct ib_device_attr *props)
1144 {
1145
1146         struct iwch_dev *dev;
1147         PDBG("%s ibdev %p\n", __func__, ibdev);
1148
1149         dev = to_iwch_dev(ibdev);
1150         memset(props, 0, sizeof *props);
1151         memcpy(&props->sys_image_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1152         props->hw_ver = dev->rdev.t3cdev_p->type;
1153         props->fw_ver = fw_vers_string_to_u64(dev);
1154         props->device_cap_flags = dev->device_cap_flags;
1155         props->page_size_cap = dev->attr.mem_pgsizes_bitmask;
1156         props->vendor_id = (u32)dev->rdev.rnic_info.pdev->vendor;
1157         props->vendor_part_id = (u32)dev->rdev.rnic_info.pdev->device;
1158         props->max_mr_size = dev->attr.max_mr_size;
1159         props->max_qp = dev->attr.max_qps;
1160         props->max_qp_wr = dev->attr.max_wrs;
1161         props->max_sge = dev->attr.max_sge_per_wr;
1162         props->max_sge_rd = 1;
1163         props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
1164         props->max_qp_init_rd_atom = dev->attr.max_rdma_reads_per_qp;
1165         props->max_cq = dev->attr.max_cqs;
1166         props->max_cqe = dev->attr.max_cqes_per_cq;
1167         props->max_mr = dev->attr.max_mem_regs;
1168         props->max_pd = dev->attr.max_pds;
1169         props->local_ca_ack_delay = 0;
1170         props->max_fast_reg_page_list_len = T3_MAX_FASTREG_DEPTH;
1171
1172         return 0;
1173 }
1174
1175 static int iwch_query_port(struct ib_device *ibdev,
1176                            u8 port, struct ib_port_attr *props)
1177 {
1178         struct iwch_dev *dev;
1179         struct net_device *netdev;
1180         struct in_device *inetdev;
1181
1182         PDBG("%s ibdev %p\n", __func__, ibdev);
1183
1184         dev = to_iwch_dev(ibdev);
1185         netdev = dev->rdev.port_info.lldevs[port-1];
1186
1187         memset(props, 0, sizeof(struct ib_port_attr));
1188         props->max_mtu = IB_MTU_4096;
1189         if (netdev->mtu >= 4096)
1190                 props->active_mtu = IB_MTU_4096;
1191         else if (netdev->mtu >= 2048)
1192                 props->active_mtu = IB_MTU_2048;
1193         else if (netdev->mtu >= 1024)
1194                 props->active_mtu = IB_MTU_1024;
1195         else if (netdev->mtu >= 512)
1196                 props->active_mtu = IB_MTU_512;
1197         else
1198                 props->active_mtu = IB_MTU_256;
1199
1200         if (!netif_carrier_ok(netdev))
1201                 props->state = IB_PORT_DOWN;
1202         else {
1203                 inetdev = in_dev_get(netdev);
1204                 if (inetdev) {
1205                         if (inetdev->ifa_list)
1206                                 props->state = IB_PORT_ACTIVE;
1207                         else
1208                                 props->state = IB_PORT_INIT;
1209                         in_dev_put(inetdev);
1210                 } else
1211                         props->state = IB_PORT_INIT;
1212         }
1213
1214         props->port_cap_flags =
1215             IB_PORT_CM_SUP |
1216             IB_PORT_SNMP_TUNNEL_SUP |
1217             IB_PORT_REINIT_SUP |
1218             IB_PORT_DEVICE_MGMT_SUP |
1219             IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
1220         props->gid_tbl_len = 1;
1221         props->pkey_tbl_len = 1;
1222         props->active_width = 2;
1223         props->active_speed = 2;
1224         props->max_msg_sz = -1;
1225
1226         return 0;
1227 }
1228
1229 static ssize_t show_rev(struct device *dev, struct device_attribute *attr,
1230                         char *buf)
1231 {
1232         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1233                                                  ibdev.dev);
1234         PDBG("%s dev 0x%p\n", __func__, dev);
1235         return sprintf(buf, "%d\n", iwch_dev->rdev.t3cdev_p->type);
1236 }
1237
1238 static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, char *buf)
1239 {
1240         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1241                                                  ibdev.dev);
1242         struct ethtool_drvinfo info;
1243         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1244
1245         PDBG("%s dev 0x%p\n", __func__, dev);
1246         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1247         return sprintf(buf, "%s\n", info.fw_version);
1248 }
1249
1250 static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
1251                         char *buf)
1252 {
1253         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1254                                                  ibdev.dev);
1255         struct ethtool_drvinfo info;
1256         struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
1257
1258         PDBG("%s dev 0x%p\n", __func__, dev);
1259         lldev->ethtool_ops->get_drvinfo(lldev, &info);
1260         return sprintf(buf, "%s\n", info.driver);
1261 }
1262
1263 static ssize_t show_board(struct device *dev, struct device_attribute *attr,
1264                           char *buf)
1265 {
1266         struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
1267                                                  ibdev.dev);
1268         PDBG("%s dev 0x%p\n", __func__, dev);
1269         return sprintf(buf, "%x.%x\n", iwch_dev->rdev.rnic_info.pdev->vendor,
1270                        iwch_dev->rdev.rnic_info.pdev->device);
1271 }
1272
1273 static int iwch_get_mib(struct ib_device *ibdev,
1274                         union rdma_protocol_stats *stats)
1275 {
1276         struct iwch_dev *dev;
1277         struct tp_mib_stats m;
1278         int ret;
1279
1280         PDBG("%s ibdev %p\n", __func__, ibdev);
1281         dev = to_iwch_dev(ibdev);
1282         ret = dev->rdev.t3cdev_p->ctl(dev->rdev.t3cdev_p, RDMA_GET_MIB, &m);
1283         if (ret)
1284                 return -ENOSYS;
1285
1286         memset(stats, 0, sizeof *stats);
1287         stats->iw.ipInReceives = ((u64) m.ipInReceive_hi << 32) +
1288                                 m.ipInReceive_lo;
1289         stats->iw.ipInHdrErrors = ((u64) m.ipInHdrErrors_hi << 32) +
1290                                   m.ipInHdrErrors_lo;
1291         stats->iw.ipInAddrErrors = ((u64) m.ipInAddrErrors_hi << 32) +
1292                                    m.ipInAddrErrors_lo;
1293         stats->iw.ipInUnknownProtos = ((u64) m.ipInUnknownProtos_hi << 32) +
1294                                       m.ipInUnknownProtos_lo;
1295         stats->iw.ipInDiscards = ((u64) m.ipInDiscards_hi << 32) +
1296                                  m.ipInDiscards_lo;
1297         stats->iw.ipInDelivers = ((u64) m.ipInDelivers_hi << 32) +
1298                                  m.ipInDelivers_lo;
1299         stats->iw.ipOutRequests = ((u64) m.ipOutRequests_hi << 32) +
1300                                   m.ipOutRequests_lo;
1301         stats->iw.ipOutDiscards = ((u64) m.ipOutDiscards_hi << 32) +
1302                                   m.ipOutDiscards_lo;
1303         stats->iw.ipOutNoRoutes = ((u64) m.ipOutNoRoutes_hi << 32) +
1304                                   m.ipOutNoRoutes_lo;
1305         stats->iw.ipReasmTimeout = (u64) m.ipReasmTimeout;
1306         stats->iw.ipReasmReqds = (u64) m.ipReasmReqds;
1307         stats->iw.ipReasmOKs = (u64) m.ipReasmOKs;
1308         stats->iw.ipReasmFails = (u64) m.ipReasmFails;
1309         stats->iw.tcpActiveOpens = (u64) m.tcpActiveOpens;
1310         stats->iw.tcpPassiveOpens = (u64) m.tcpPassiveOpens;
1311         stats->iw.tcpAttemptFails = (u64) m.tcpAttemptFails;
1312         stats->iw.tcpEstabResets = (u64) m.tcpEstabResets;
1313         stats->iw.tcpOutRsts = (u64) m.tcpOutRsts;
1314         stats->iw.tcpCurrEstab = (u64) m.tcpCurrEstab;
1315         stats->iw.tcpInSegs = ((u64) m.tcpInSegs_hi << 32) +
1316                               m.tcpInSegs_lo;
1317         stats->iw.tcpOutSegs = ((u64) m.tcpOutSegs_hi << 32) +
1318                                m.tcpOutSegs_lo;
1319         stats->iw.tcpRetransSegs = ((u64) m.tcpRetransSeg_hi << 32) +
1320                                   m.tcpRetransSeg_lo;
1321         stats->iw.tcpInErrs = ((u64) m.tcpInErrs_hi << 32) +
1322                               m.tcpInErrs_lo;
1323         stats->iw.tcpRtoMin = (u64) m.tcpRtoMin;
1324         stats->iw.tcpRtoMax = (u64) m.tcpRtoMax;
1325         return 0;
1326 }
1327
1328 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1329 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1330 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1331 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1332
1333 static struct device_attribute *iwch_class_attributes[] = {
1334         &dev_attr_hw_rev,
1335         &dev_attr_fw_ver,
1336         &dev_attr_hca_type,
1337         &dev_attr_board_id,
1338 };
1339
1340 int iwch_register_device(struct iwch_dev *dev)
1341 {
1342         int ret;
1343         int i;
1344
1345         PDBG("%s iwch_dev %p\n", __func__, dev);
1346         strlcpy(dev->ibdev.name, "cxgb3_%d", IB_DEVICE_NAME_MAX);
1347         memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
1348         memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
1349         dev->ibdev.owner = THIS_MODULE;
1350         dev->device_cap_flags = IB_DEVICE_LOCAL_DMA_LKEY |
1351                                 IB_DEVICE_MEM_WINDOW |
1352                                 IB_DEVICE_MEM_MGT_EXTENSIONS;
1353
1354         /* cxgb3 supports STag 0. */
1355         dev->ibdev.local_dma_lkey = 0;
1356
1357         dev->ibdev.uverbs_cmd_mask =
1358             (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1359             (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1360             (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1361             (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1362             (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1363             (1ull << IB_USER_VERBS_CMD_REG_MR) |
1364             (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1365             (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1366             (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1367             (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1368             (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1369             (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1370             (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1371             (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1372             (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1373             (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1374             (1ull << IB_USER_VERBS_CMD_POST_RECV);
1375         dev->ibdev.node_type = RDMA_NODE_RNIC;
1376         memcpy(dev->ibdev.node_desc, IWCH_NODE_DESC, sizeof(IWCH_NODE_DESC));
1377         dev->ibdev.phys_port_cnt = dev->rdev.port_info.nports;
1378         dev->ibdev.num_comp_vectors = 1;
1379         dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
1380         dev->ibdev.query_device = iwch_query_device;
1381         dev->ibdev.query_port = iwch_query_port;
1382         dev->ibdev.modify_port = iwch_modify_port;
1383         dev->ibdev.query_pkey = iwch_query_pkey;
1384         dev->ibdev.query_gid = iwch_query_gid;
1385         dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
1386         dev->ibdev.dealloc_ucontext = iwch_dealloc_ucontext;
1387         dev->ibdev.mmap = iwch_mmap;
1388         dev->ibdev.alloc_pd = iwch_allocate_pd;
1389         dev->ibdev.dealloc_pd = iwch_deallocate_pd;
1390         dev->ibdev.create_ah = iwch_ah_create;
1391         dev->ibdev.destroy_ah = iwch_ah_destroy;
1392         dev->ibdev.create_qp = iwch_create_qp;
1393         dev->ibdev.modify_qp = iwch_ib_modify_qp;
1394         dev->ibdev.destroy_qp = iwch_destroy_qp;
1395         dev->ibdev.create_cq = iwch_create_cq;
1396         dev->ibdev.destroy_cq = iwch_destroy_cq;
1397         dev->ibdev.resize_cq = iwch_resize_cq;
1398         dev->ibdev.poll_cq = iwch_poll_cq;
1399         dev->ibdev.get_dma_mr = iwch_get_dma_mr;
1400         dev->ibdev.reg_phys_mr = iwch_register_phys_mem;
1401         dev->ibdev.rereg_phys_mr = iwch_reregister_phys_mem;
1402         dev->ibdev.reg_user_mr = iwch_reg_user_mr;
1403         dev->ibdev.dereg_mr = iwch_dereg_mr;
1404         dev->ibdev.alloc_mw = iwch_alloc_mw;
1405         dev->ibdev.bind_mw = iwch_bind_mw;
1406         dev->ibdev.dealloc_mw = iwch_dealloc_mw;
1407         dev->ibdev.alloc_fast_reg_mr = iwch_alloc_fast_reg_mr;
1408         dev->ibdev.alloc_fast_reg_page_list = iwch_alloc_fastreg_pbl;
1409         dev->ibdev.free_fast_reg_page_list = iwch_free_fastreg_pbl;
1410         dev->ibdev.attach_mcast = iwch_multicast_attach;
1411         dev->ibdev.detach_mcast = iwch_multicast_detach;
1412         dev->ibdev.process_mad = iwch_process_mad;
1413         dev->ibdev.req_notify_cq = iwch_arm_cq;
1414         dev->ibdev.post_send = iwch_post_send;
1415         dev->ibdev.post_recv = iwch_post_receive;
1416         dev->ibdev.get_protocol_stats = iwch_get_mib;
1417
1418         dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
1419         if (!dev->ibdev.iwcm)
1420                 return -ENOMEM;
1421
1422         dev->ibdev.iwcm->connect = iwch_connect;
1423         dev->ibdev.iwcm->accept = iwch_accept_cr;
1424         dev->ibdev.iwcm->reject = iwch_reject_cr;
1425         dev->ibdev.iwcm->create_listen = iwch_create_listen;
1426         dev->ibdev.iwcm->destroy_listen = iwch_destroy_listen;
1427         dev->ibdev.iwcm->add_ref = iwch_qp_add_ref;
1428         dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
1429         dev->ibdev.iwcm->get_qp = iwch_get_qp;
1430
1431         ret = ib_register_device(&dev->ibdev, NULL);
1432         if (ret)
1433                 goto bail1;
1434
1435         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i) {
1436                 ret = device_create_file(&dev->ibdev.dev,
1437                                          iwch_class_attributes[i]);
1438                 if (ret) {
1439                         goto bail2;
1440                 }
1441         }
1442         return 0;
1443 bail2:
1444         ib_unregister_device(&dev->ibdev);
1445 bail1:
1446         kfree(dev->ibdev.iwcm);
1447         return ret;
1448 }
1449
1450 void iwch_unregister_device(struct iwch_dev *dev)
1451 {
1452         int i;
1453
1454         PDBG("%s iwch_dev %p\n", __func__, dev);
1455         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i)
1456                 device_remove_file(&dev->ibdev.dev,
1457                                    iwch_class_attributes[i]);
1458         ib_unregister_device(&dev->ibdev);
1459         kfree(dev->ibdev.iwcm);
1460         return;
1461 }