2  * Copyright (c) 2006 Chelsio, Inc. All rights reserved.
 
   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:
 
  10  *     Redistribution and use in source and binary forms, with or
 
  11  *     without modification, are permitted provided that the following
 
  14  *      - Redistributions of source code must retain the above
 
  15  *        copyright notice, this list of conditions and the following
 
  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.
 
  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
 
  32 #include <linux/module.h>
 
  33 #include <linux/moduleparam.h>
 
  34 #include <linux/device.h>
 
  35 #include <linux/netdevice.h>
 
  36 #include <linux/etherdevice.h>
 
  37 #include <linux/delay.h>
 
  38 #include <linux/errno.h>
 
  39 #include <linux/list.h>
 
  40 #include <linux/spinlock.h>
 
  41 #include <linux/ethtool.h>
 
  42 #include <linux/rtnetlink.h>
 
  46 #include <asm/byteorder.h>
 
  48 #include <rdma/iw_cm.h>
 
  49 #include <rdma/ib_verbs.h>
 
  50 #include <rdma/ib_smi.h>
 
  51 #include <rdma/ib_umem.h>
 
  52 #include <rdma/ib_user_verbs.h>
 
  56 #include "iwch_provider.h"
 
  58 #include "iwch_user.h"
 
  60 static int iwch_modify_port(struct ib_device *ibdev,
 
  61                             u8 port, int port_modify_mask,
 
  62                             struct ib_port_modify *props)
 
  67 static struct ib_ah *iwch_ah_create(struct ib_pd *pd,
 
  68                                     struct ib_ah_attr *ah_attr)
 
  70         return ERR_PTR(-ENOSYS);
 
  73 static int iwch_ah_destroy(struct ib_ah *ah)
 
  78 static int iwch_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 
  83 static int iwch_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 
  88 static int iwch_process_mad(struct ib_device *ibdev,
 
  92                             struct ib_grh *in_grh,
 
  93                             struct ib_mad *in_mad, struct ib_mad *out_mad)
 
  98 static int iwch_dealloc_ucontext(struct ib_ucontext *context)
 
 100         struct iwch_dev *rhp = to_iwch_dev(context->device);
 
 101         struct iwch_ucontext *ucontext = to_iwch_ucontext(context);
 
 102         struct iwch_mm_entry *mm, *tmp;
 
 104         PDBG("%s context %p\n", __FUNCTION__, context);
 
 105         list_for_each_entry_safe(mm, tmp, &ucontext->mmaps, entry)
 
 107         cxio_release_ucontext(&rhp->rdev, &ucontext->uctx);
 
 112 static struct ib_ucontext *iwch_alloc_ucontext(struct ib_device *ibdev,
 
 113                                         struct ib_udata *udata)
 
 115         struct iwch_ucontext *context;
 
 116         struct iwch_dev *rhp = to_iwch_dev(ibdev);
 
 118         PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
 
 119         context = kzalloc(sizeof(*context), GFP_KERNEL);
 
 121                 return ERR_PTR(-ENOMEM);
 
 122         cxio_init_ucontext(&rhp->rdev, &context->uctx);
 
 123         INIT_LIST_HEAD(&context->mmaps);
 
 124         spin_lock_init(&context->mmap_lock);
 
 125         return &context->ibucontext;
 
 128 static int iwch_destroy_cq(struct ib_cq *ib_cq)
 
 132         PDBG("%s ib_cq %p\n", __FUNCTION__, ib_cq);
 
 133         chp = to_iwch_cq(ib_cq);
 
 135         remove_handle(chp->rhp, &chp->rhp->cqidr, chp->cq.cqid);
 
 136         atomic_dec(&chp->refcnt);
 
 137         wait_event(chp->wait, !atomic_read(&chp->refcnt));
 
 139         cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
 
 144 static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries, int vector,
 
 145                              struct ib_ucontext *ib_context,
 
 146                              struct ib_udata *udata)
 
 148         struct iwch_dev *rhp;
 
 150         struct iwch_create_cq_resp uresp;
 
 151         struct iwch_create_cq_req ureq;
 
 152         struct iwch_ucontext *ucontext = NULL;
 
 154         PDBG("%s ib_dev %p entries %d\n", __FUNCTION__, ibdev, entries);
 
 155         rhp = to_iwch_dev(ibdev);
 
 156         chp = kzalloc(sizeof(*chp), GFP_KERNEL);
 
 158                 return ERR_PTR(-ENOMEM);
 
 161                 ucontext = to_iwch_ucontext(ib_context);
 
 162                 if (!t3a_device(rhp)) {
 
 163                         if (ib_copy_from_udata(&ureq, udata, sizeof (ureq))) {
 
 165                                 return ERR_PTR(-EFAULT);
 
 167                         chp->user_rptr_addr = (u32 __user *)(unsigned long)ureq.user_rptr_addr;
 
 171         if (t3a_device(rhp)) {
 
 174                  * T3A: Add some fluff to handle extra CQEs inserted
 
 175                  * for various errors.
 
 176                  * Additional CQE possibilities:
 
 178                  *      incoming RDMA WRITE Failures
 
 179                  *      incoming RDMA READ REQUEST FAILUREs
 
 180                  * NOTE: We cannot ensure the CQ won't overflow.
 
 184         entries = roundup_pow_of_two(entries);
 
 185         chp->cq.size_log2 = ilog2(entries);
 
 187         if (cxio_create_cq(&rhp->rdev, &chp->cq)) {
 
 189                 return ERR_PTR(-ENOMEM);
 
 192         chp->ibcq.cqe = 1 << chp->cq.size_log2;
 
 193         spin_lock_init(&chp->lock);
 
 194         atomic_set(&chp->refcnt, 1);
 
 195         init_waitqueue_head(&chp->wait);
 
 196         insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid);
 
 199                 struct iwch_mm_entry *mm;
 
 201                 mm = kmalloc(sizeof *mm, GFP_KERNEL);
 
 203                         iwch_destroy_cq(&chp->ibcq);
 
 204                         return ERR_PTR(-ENOMEM);
 
 206                 uresp.cqid = chp->cq.cqid;
 
 207                 uresp.size_log2 = chp->cq.size_log2;
 
 208                 spin_lock(&ucontext->mmap_lock);
 
 209                 uresp.key = ucontext->key;
 
 210                 ucontext->key += PAGE_SIZE;
 
 211                 spin_unlock(&ucontext->mmap_lock);
 
 212                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 
 214                         iwch_destroy_cq(&chp->ibcq);
 
 215                         return ERR_PTR(-EFAULT);
 
 218                 mm->addr = virt_to_phys(chp->cq.queue);
 
 219                 mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
 
 220                                              sizeof (struct t3_cqe));
 
 221                 insert_mmap(ucontext, mm);
 
 223         PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
 
 224              chp->cq.cqid, chp, (1 << chp->cq.size_log2),
 
 225              (unsigned long long) chp->cq.dma_addr);
 
 229 static int iwch_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
 
 232         struct iwch_cq *chp = to_iwch_cq(cq);
 
 233         struct t3_cq oldcq, newcq;
 
 236         PDBG("%s ib_cq %p cqe %d\n", __FUNCTION__, cq, cqe);
 
 238         /* We don't downsize... */
 
 242         /* create new t3_cq with new size */
 
 243         cqe = roundup_pow_of_two(cqe+1);
 
 244         newcq.size_log2 = ilog2(cqe);
 
 246         /* Dont allow resize to less than the current wce count */
 
 247         if (cqe < Q_COUNT(chp->cq.rptr, chp->cq.wptr)) {
 
 251         /* Quiesce all QPs using this CQ */
 
 252         ret = iwch_quiesce_qps(chp);
 
 257         ret = cxio_create_cq(&chp->rhp->rdev, &newcq);
 
 263         memcpy(newcq.queue, chp->cq.queue, (1 << chp->cq.size_log2) *
 
 264                                         sizeof(struct t3_cqe));
 
 266         /* old iwch_qp gets new t3_cq but keeps old cqid */
 
 269         chp->cq.cqid = oldcq.cqid;
 
 271         /* resize new t3_cq to update the HW context */
 
 272         ret = cxio_resize_cq(&chp->rhp->rdev, &chp->cq);
 
 277         chp->ibcq.cqe = (1<<chp->cq.size_log2) - 1;
 
 279         /* destroy old t3_cq */
 
 280         oldcq.cqid = newcq.cqid;
 
 281         ret = cxio_destroy_cq(&chp->rhp->rdev, &oldcq);
 
 283                 printk(KERN_ERR MOD "%s - cxio_destroy_cq failed %d\n",
 
 287         /* add user hooks here */
 
 290         ret = iwch_resume_qps(chp);
 
 297 static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
 
 299         struct iwch_dev *rhp;
 
 301         enum t3_cq_opcode cq_op;
 
 306         chp = to_iwch_cq(ibcq);
 
 308         if ((flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED)
 
 312         if (chp->user_rptr_addr) {
 
 313                 if (get_user(rptr, chp->user_rptr_addr))
 
 315                 spin_lock_irqsave(&chp->lock, flag);
 
 318                 spin_lock_irqsave(&chp->lock, flag);
 
 319         PDBG("%s rptr 0x%x\n", __FUNCTION__, chp->cq.rptr);
 
 320         err = cxio_hal_cq_op(&rhp->rdev, &chp->cq, cq_op, 0);
 
 321         spin_unlock_irqrestore(&chp->lock, flag);
 
 323                 printk(KERN_ERR MOD "Error %d rearming CQID 0x%x\n", err,
 
 325         if (err > 0 && !(flags & IB_CQ_REPORT_MISSED_EVENTS))
 
 330 static int iwch_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 
 332         int len = vma->vm_end - vma->vm_start;
 
 333         u32 key = vma->vm_pgoff << PAGE_SHIFT;
 
 334         struct cxio_rdev *rdev_p;
 
 336         struct iwch_mm_entry *mm;
 
 337         struct iwch_ucontext *ucontext;
 
 340         PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __FUNCTION__, vma->vm_pgoff,
 
 343         if (vma->vm_start & (PAGE_SIZE-1)) {
 
 347         rdev_p = &(to_iwch_dev(context->device)->rdev);
 
 348         ucontext = to_iwch_ucontext(context);
 
 350         mm = remove_mmap(ucontext, key, len);
 
 356         if ((addr >= rdev_p->rnic_info.udbell_physbase) &&
 
 357             (addr < (rdev_p->rnic_info.udbell_physbase +
 
 358                        rdev_p->rnic_info.udbell_len))) {
 
 361                  * Map T3 DB register.
 
 363                 if (vma->vm_flags & VM_READ) {
 
 367                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 
 368                 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
 
 369                 vma->vm_flags &= ~VM_MAYREAD;
 
 370                 ret = io_remap_pfn_range(vma, vma->vm_start,
 
 372                                          len, vma->vm_page_prot);
 
 376                  * Map WQ or CQ contig dma memory...
 
 378                 ret = remap_pfn_range(vma, vma->vm_start,
 
 380                                       len, vma->vm_page_prot);
 
 386 static int iwch_deallocate_pd(struct ib_pd *pd)
 
 388         struct iwch_dev *rhp;
 
 391         php = to_iwch_pd(pd);
 
 393         PDBG("%s ibpd %p pdid 0x%x\n", __FUNCTION__, pd, php->pdid);
 
 394         cxio_hal_put_pdid(rhp->rdev.rscp, php->pdid);
 
 399 static struct ib_pd *iwch_allocate_pd(struct ib_device *ibdev,
 
 400                                struct ib_ucontext *context,
 
 401                                struct ib_udata *udata)
 
 405         struct iwch_dev *rhp;
 
 407         PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
 
 408         rhp = (struct iwch_dev *) ibdev;
 
 409         pdid = cxio_hal_get_pdid(rhp->rdev.rscp);
 
 411                 return ERR_PTR(-EINVAL);
 
 412         php = kzalloc(sizeof(*php), GFP_KERNEL);
 
 414                 cxio_hal_put_pdid(rhp->rdev.rscp, pdid);
 
 415                 return ERR_PTR(-ENOMEM);
 
 420                 if (ib_copy_to_udata(udata, &php->pdid, sizeof (__u32))) {
 
 421                         iwch_deallocate_pd(&php->ibpd);
 
 422                         return ERR_PTR(-EFAULT);
 
 425         PDBG("%s pdid 0x%0x ptr 0x%p\n", __FUNCTION__, pdid, php);
 
 429 static int iwch_dereg_mr(struct ib_mr *ib_mr)
 
 431         struct iwch_dev *rhp;
 
 435         PDBG("%s ib_mr %p\n", __FUNCTION__, ib_mr);
 
 436         /* There can be no memory windows */
 
 437         if (atomic_read(&ib_mr->usecnt))
 
 440         mhp = to_iwch_mr(ib_mr);
 
 442         mmid = mhp->attr.stag >> 8;
 
 443         cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
 
 445         remove_handle(rhp, &rhp->mmidr, mmid);
 
 447                 kfree((void *) (unsigned long) mhp->kva);
 
 449                 ib_umem_release(mhp->umem);
 
 450         PDBG("%s mmid 0x%x ptr %p\n", __FUNCTION__, mmid, mhp);
 
 455 static struct ib_mr *iwch_register_phys_mem(struct ib_pd *pd,
 
 456                                         struct ib_phys_buf *buffer_list,
 
 465         struct iwch_dev *rhp;
 
 470         PDBG("%s ib_pd %p\n", __FUNCTION__, pd);
 
 471         php = to_iwch_pd(pd);
 
 474         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 
 476                 return ERR_PTR(-ENOMEM);
 
 478         /* First check that we have enough alignment */
 
 479         if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
 
 484         if (num_phys_buf > 1 &&
 
 485             ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
 
 490         ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
 
 491                                    &total_size, &npages, &shift, &page_list);
 
 496         mhp->attr.pdid = php->pdid;
 
 499         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 
 500         mhp->attr.va_fbo = *iova_start;
 
 501         mhp->attr.page_size = shift - 12;
 
 503         mhp->attr.len = (u32) total_size;
 
 504         mhp->attr.pbl_size = npages;
 
 505         ret = iwch_register_mem(rhp, php, mhp, shift, page_list);
 
 517 static int iwch_reregister_phys_mem(struct ib_mr *mr,
 
 520                                      struct ib_phys_buf *buffer_list,
 
 522                                      int acc, u64 * iova_start)
 
 525         struct iwch_mr mh, *mhp;
 
 527         struct iwch_dev *rhp;
 
 528         __be64 *page_list = NULL;
 
 534         PDBG("%s ib_mr %p ib_pd %p\n", __FUNCTION__, mr, pd);
 
 536         /* There can be no memory windows */
 
 537         if (atomic_read(&mr->usecnt))
 
 540         mhp = to_iwch_mr(mr);
 
 542         php = to_iwch_pd(mr->pd);
 
 544         /* make sure we are on the same adapter */
 
 548         memcpy(&mh, mhp, sizeof *mhp);
 
 550         if (mr_rereg_mask & IB_MR_REREG_PD)
 
 551                 php = to_iwch_pd(pd);
 
 552         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 
 553                 mh.attr.perms = iwch_ib_to_tpt_access(acc);
 
 554         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 
 555                 ret = build_phys_page_list(buffer_list, num_phys_buf,
 
 557                                            &total_size, &npages,
 
 563         ret = iwch_reregister_mem(rhp, php, &mh, shift, page_list, npages);
 
 568         if (mr_rereg_mask & IB_MR_REREG_PD)
 
 569                 mhp->attr.pdid = php->pdid;
 
 570         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 
 571                 mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 
 572         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 
 574                 mhp->attr.va_fbo = *iova_start;
 
 575                 mhp->attr.page_size = shift - 12;
 
 576                 mhp->attr.len = (u32) total_size;
 
 577                 mhp->attr.pbl_size = npages;
 
 584 static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 585                                       u64 virt, int acc, struct ib_udata *udata)
 
 591         struct ib_umem_chunk *chunk;
 
 592         struct iwch_dev *rhp;
 
 595         struct iwch_reg_user_mr_resp uresp;
 
 597         PDBG("%s ib_pd %p\n", __FUNCTION__, pd);
 
 599         php = to_iwch_pd(pd);
 
 601         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 
 603                 return ERR_PTR(-ENOMEM);
 
 605         mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc);
 
 606         if (IS_ERR(mhp->umem)) {
 
 607                 err = PTR_ERR(mhp->umem);
 
 612         shift = ffs(mhp->umem->page_size) - 1;
 
 615         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 
 618         pages = kmalloc(n * sizeof(u64), GFP_KERNEL);
 
 626         list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
 
 627                 for (j = 0; j < chunk->nmap; ++j) {
 
 628                         len = sg_dma_len(&chunk->page_list[j]) >> shift;
 
 629                         for (k = 0; k < len; ++k) {
 
 630                                 pages[i++] = cpu_to_be64(sg_dma_address(
 
 631                                         &chunk->page_list[j]) +
 
 632                                         mhp->umem->page_size * k);
 
 637         mhp->attr.pdid = php->pdid;
 
 639         mhp->attr.perms = iwch_ib_to_tpt_access(acc);
 
 640         mhp->attr.va_fbo = virt;
 
 641         mhp->attr.page_size = shift - 12;
 
 642         mhp->attr.len = (u32) length;
 
 643         mhp->attr.pbl_size = i;
 
 644         err = iwch_register_mem(rhp, php, mhp, shift, pages);
 
 649         if (udata && !t3a_device(rhp)) {
 
 650                 uresp.pbl_addr = (mhp->attr.pbl_addr -
 
 651                                  rhp->rdev.rnic_info.pbl_base) >> 3;
 
 652                 PDBG("%s user resp pbl_addr 0x%x\n", __FUNCTION__,
 
 655                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 
 656                         iwch_dereg_mr(&mhp->ibmr);
 
 665         ib_umem_release(mhp->umem);
 
 670 static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
 
 672         struct ib_phys_buf bl;
 
 676         PDBG("%s ib_pd %p\n", __FUNCTION__, pd);
 
 679          * T3 only supports 32 bits of size.
 
 681         bl.size = 0xffffffff;
 
 684         ibmr = iwch_register_phys_mem(pd, &bl, 1, acc, &kva);
 
 688 static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
 
 690         struct iwch_dev *rhp;
 
 697         php = to_iwch_pd(pd);
 
 699         mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 
 701                 return ERR_PTR(-ENOMEM);
 
 702         ret = cxio_allocate_window(&rhp->rdev, &stag, php->pdid);
 
 708         mhp->attr.pdid = php->pdid;
 
 709         mhp->attr.type = TPT_MW;
 
 710         mhp->attr.stag = stag;
 
 712         insert_handle(rhp, &rhp->mmidr, mhp, mmid);
 
 713         PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __FUNCTION__, mmid, mhp, stag);
 
 717 static int iwch_dealloc_mw(struct ib_mw *mw)
 
 719         struct iwch_dev *rhp;
 
 723         mhp = to_iwch_mw(mw);
 
 725         mmid = (mw->rkey) >> 8;
 
 726         cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 
 727         remove_handle(rhp, &rhp->mmidr, mmid);
 
 729         PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __FUNCTION__, mw, mmid, mhp);
 
 733 static int iwch_destroy_qp(struct ib_qp *ib_qp)
 
 735         struct iwch_dev *rhp;
 
 737         struct iwch_qp_attributes attrs;
 
 738         struct iwch_ucontext *ucontext;
 
 740         qhp = to_iwch_qp(ib_qp);
 
 743         attrs.next_state = IWCH_QP_STATE_ERROR;
 
 744         iwch_modify_qp(rhp, qhp, IWCH_QP_ATTR_NEXT_STATE, &attrs, 0);
 
 745         wait_event(qhp->wait, !qhp->ep);
 
 747         remove_handle(rhp, &rhp->qpidr, qhp->wq.qpid);
 
 749         atomic_dec(&qhp->refcnt);
 
 750         wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
 
 752         ucontext = ib_qp->uobject ? to_iwch_ucontext(ib_qp->uobject->context)
 
 754         cxio_destroy_qp(&rhp->rdev, &qhp->wq,
 
 755                         ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
 
 757         PDBG("%s ib_qp %p qpid 0x%0x qhp %p\n", __FUNCTION__,
 
 758              ib_qp, qhp->wq.qpid, qhp);
 
 763 static struct ib_qp *iwch_create_qp(struct ib_pd *pd,
 
 764                              struct ib_qp_init_attr *attrs,
 
 765                              struct ib_udata *udata)
 
 767         struct iwch_dev *rhp;
 
 770         struct iwch_cq *schp;
 
 771         struct iwch_cq *rchp;
 
 772         struct iwch_create_qp_resp uresp;
 
 773         int wqsize, sqsize, rqsize;
 
 774         struct iwch_ucontext *ucontext;
 
 776         PDBG("%s ib_pd %p\n", __FUNCTION__, pd);
 
 777         if (attrs->qp_type != IB_QPT_RC)
 
 778                 return ERR_PTR(-EINVAL);
 
 779         php = to_iwch_pd(pd);
 
 781         schp = get_chp(rhp, ((struct iwch_cq *) attrs->send_cq)->cq.cqid);
 
 782         rchp = get_chp(rhp, ((struct iwch_cq *) attrs->recv_cq)->cq.cqid);
 
 784                 return ERR_PTR(-EINVAL);
 
 786         /* The RQT size must be # of entries + 1 rounded up to a power of two */
 
 787         rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr);
 
 788         if (rqsize == attrs->cap.max_recv_wr)
 
 789                 rqsize = roundup_pow_of_two(attrs->cap.max_recv_wr+1);
 
 791         /* T3 doesn't support RQT depth < 16 */
 
 795         if (rqsize > T3_MAX_RQ_SIZE)
 
 796                 return ERR_PTR(-EINVAL);
 
 798         if (attrs->cap.max_inline_data > T3_MAX_INLINE)
 
 799                 return ERR_PTR(-EINVAL);
 
 802          * NOTE: The SQ and total WQ sizes don't need to be
 
 803          * a power of two.  However, all the code assumes
 
 804          * they are. EG: Q_FREECNT() and friends.
 
 806         sqsize = roundup_pow_of_two(attrs->cap.max_send_wr);
 
 807         wqsize = roundup_pow_of_two(rqsize + sqsize);
 
 808         PDBG("%s wqsize %d sqsize %d rqsize %d\n", __FUNCTION__,
 
 809              wqsize, sqsize, rqsize);
 
 810         qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
 
 812                 return ERR_PTR(-ENOMEM);
 
 813         qhp->wq.size_log2 = ilog2(wqsize);
 
 814         qhp->wq.rq_size_log2 = ilog2(rqsize);
 
 815         qhp->wq.sq_size_log2 = ilog2(sqsize);
 
 816         ucontext = pd->uobject ? to_iwch_ucontext(pd->uobject->context) : NULL;
 
 817         if (cxio_create_qp(&rhp->rdev, !udata, &qhp->wq,
 
 818                            ucontext ? &ucontext->uctx : &rhp->rdev.uctx)) {
 
 820                 return ERR_PTR(-ENOMEM);
 
 823         attrs->cap.max_recv_wr = rqsize - 1;
 
 824         attrs->cap.max_send_wr = sqsize;
 
 825         attrs->cap.max_inline_data = T3_MAX_INLINE;
 
 828         qhp->attr.pd = php->pdid;
 
 829         qhp->attr.scq = ((struct iwch_cq *) attrs->send_cq)->cq.cqid;
 
 830         qhp->attr.rcq = ((struct iwch_cq *) attrs->recv_cq)->cq.cqid;
 
 831         qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
 
 832         qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
 
 833         qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
 
 834         qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
 
 835         qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
 
 836         qhp->attr.state = IWCH_QP_STATE_IDLE;
 
 837         qhp->attr.next_state = IWCH_QP_STATE_IDLE;
 
 840          * XXX - These don't get passed in from the openib user
 
 841          * at create time.  The CM sets them via a QP modify.
 
 842          * Need to fix...  I think the CM should
 
 844         qhp->attr.enable_rdma_read = 1;
 
 845         qhp->attr.enable_rdma_write = 1;
 
 846         qhp->attr.enable_bind = 1;
 
 847         qhp->attr.max_ord = 1;
 
 848         qhp->attr.max_ird = 1;
 
 850         spin_lock_init(&qhp->lock);
 
 851         init_waitqueue_head(&qhp->wait);
 
 852         atomic_set(&qhp->refcnt, 1);
 
 853         insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid);
 
 857                 struct iwch_mm_entry *mm1, *mm2;
 
 859                 mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
 
 861                         iwch_destroy_qp(&qhp->ibqp);
 
 862                         return ERR_PTR(-ENOMEM);
 
 865                 mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
 
 868                         iwch_destroy_qp(&qhp->ibqp);
 
 869                         return ERR_PTR(-ENOMEM);
 
 872                 uresp.qpid = qhp->wq.qpid;
 
 873                 uresp.size_log2 = qhp->wq.size_log2;
 
 874                 uresp.sq_size_log2 = qhp->wq.sq_size_log2;
 
 875                 uresp.rq_size_log2 = qhp->wq.rq_size_log2;
 
 876                 spin_lock(&ucontext->mmap_lock);
 
 877                 uresp.key = ucontext->key;
 
 878                 ucontext->key += PAGE_SIZE;
 
 879                 uresp.db_key = ucontext->key;
 
 880                 ucontext->key += PAGE_SIZE;
 
 881                 spin_unlock(&ucontext->mmap_lock);
 
 882                 if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
 
 885                         iwch_destroy_qp(&qhp->ibqp);
 
 886                         return ERR_PTR(-EFAULT);
 
 888                 mm1->key = uresp.key;
 
 889                 mm1->addr = virt_to_phys(qhp->wq.queue);
 
 890                 mm1->len = PAGE_ALIGN(wqsize * sizeof (union t3_wr));
 
 891                 insert_mmap(ucontext, mm1);
 
 892                 mm2->key = uresp.db_key;
 
 893                 mm2->addr = qhp->wq.udb & PAGE_MASK;
 
 894                 mm2->len = PAGE_SIZE;
 
 895                 insert_mmap(ucontext, mm2);
 
 897         qhp->ibqp.qp_num = qhp->wq.qpid;
 
 898         init_timer(&(qhp->timer));
 
 899         PDBG("%s sq_num_entries %d, rq_num_entries %d "
 
 900              "qpid 0x%0x qhp %p dma_addr 0x%llx size %d\n",
 
 901              __FUNCTION__, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
 
 902              qhp->wq.qpid, qhp, (unsigned long long) qhp->wq.dma_addr,
 
 903              1 << qhp->wq.size_log2);
 
 907 static int iwch_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 
 908                       int attr_mask, struct ib_udata *udata)
 
 910         struct iwch_dev *rhp;
 
 912         enum iwch_qp_attr_mask mask = 0;
 
 913         struct iwch_qp_attributes attrs;
 
 915         PDBG("%s ib_qp %p\n", __FUNCTION__, ibqp);
 
 917         /* iwarp does not support the RTR state */
 
 918         if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
 
 919                 attr_mask &= ~IB_QP_STATE;
 
 921         /* Make sure we still have something left to do */
 
 925         memset(&attrs, 0, sizeof attrs);
 
 926         qhp = to_iwch_qp(ibqp);
 
 929         attrs.next_state = iwch_convert_state(attr->qp_state);
 
 930         attrs.enable_rdma_read = (attr->qp_access_flags &
 
 931                                IB_ACCESS_REMOTE_READ) ?  1 : 0;
 
 932         attrs.enable_rdma_write = (attr->qp_access_flags &
 
 933                                 IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
 
 934         attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
 
 937         mask |= (attr_mask & IB_QP_STATE) ? IWCH_QP_ATTR_NEXT_STATE : 0;
 
 938         mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
 
 939                         (IWCH_QP_ATTR_ENABLE_RDMA_READ |
 
 940                          IWCH_QP_ATTR_ENABLE_RDMA_WRITE |
 
 941                          IWCH_QP_ATTR_ENABLE_RDMA_BIND) : 0;
 
 943         return iwch_modify_qp(rhp, qhp, mask, &attrs, 0);
 
 946 void iwch_qp_add_ref(struct ib_qp *qp)
 
 948         PDBG("%s ib_qp %p\n", __FUNCTION__, qp);
 
 949         atomic_inc(&(to_iwch_qp(qp)->refcnt));
 
 952 void iwch_qp_rem_ref(struct ib_qp *qp)
 
 954         PDBG("%s ib_qp %p\n", __FUNCTION__, qp);
 
 955         if (atomic_dec_and_test(&(to_iwch_qp(qp)->refcnt)))
 
 956                 wake_up(&(to_iwch_qp(qp)->wait));
 
 959 static struct ib_qp *iwch_get_qp(struct ib_device *dev, int qpn)
 
 961         PDBG("%s ib_dev %p qpn 0x%x\n", __FUNCTION__, dev, qpn);
 
 962         return (struct ib_qp *)get_qhp(to_iwch_dev(dev), qpn);
 
 966 static int iwch_query_pkey(struct ib_device *ibdev,
 
 967                            u8 port, u16 index, u16 * pkey)
 
 969         PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
 
 974 static int iwch_query_gid(struct ib_device *ibdev, u8 port,
 
 975                           int index, union ib_gid *gid)
 
 977         struct iwch_dev *dev;
 
 979         PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
 
 980                __FUNCTION__, ibdev, port, index, gid);
 
 981         dev = to_iwch_dev(ibdev);
 
 982         BUG_ON(port == 0 || port > 2);
 
 983         memset(&(gid->raw[0]), 0, sizeof(gid->raw));
 
 984         memcpy(&(gid->raw[0]), dev->rdev.port_info.lldevs[port-1]->dev_addr, 6);
 
 988 static int iwch_query_device(struct ib_device *ibdev,
 
 989                              struct ib_device_attr *props)
 
 992         struct iwch_dev *dev;
 
 993         PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
 
 995         dev = to_iwch_dev(ibdev);
 
 996         memset(props, 0, sizeof *props);
 
 997         memcpy(&props->sys_image_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
 
 998         props->device_cap_flags = dev->device_cap_flags;
 
 999         props->vendor_id = (u32)dev->rdev.rnic_info.pdev->vendor;
 
1000         props->vendor_part_id = (u32)dev->rdev.rnic_info.pdev->device;
 
1001         props->max_mr_size = ~0ull;
 
1002         props->max_qp = dev->attr.max_qps;
 
1003         props->max_qp_wr = dev->attr.max_wrs;
 
1004         props->max_sge = dev->attr.max_sge_per_wr;
 
1005         props->max_sge_rd = 1;
 
1006         props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp;
 
1007         props->max_qp_init_rd_atom = dev->attr.max_rdma_reads_per_qp;
 
1008         props->max_cq = dev->attr.max_cqs;
 
1009         props->max_cqe = dev->attr.max_cqes_per_cq;
 
1010         props->max_mr = dev->attr.max_mem_regs;
 
1011         props->max_pd = dev->attr.max_pds;
 
1012         props->local_ca_ack_delay = 0;
 
1017 static int iwch_query_port(struct ib_device *ibdev,
 
1018                            u8 port, struct ib_port_attr *props)
 
1020         PDBG("%s ibdev %p\n", __FUNCTION__, ibdev);
 
1021         props->max_mtu = IB_MTU_4096;
 
1026         props->state = IB_PORT_ACTIVE;
 
1027         props->phys_state = 0;
 
1028         props->port_cap_flags =
 
1030             IB_PORT_SNMP_TUNNEL_SUP |
 
1031             IB_PORT_REINIT_SUP |
 
1032             IB_PORT_DEVICE_MGMT_SUP |
 
1033             IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
 
1034         props->gid_tbl_len = 1;
 
1035         props->pkey_tbl_len = 1;
 
1036         props->qkey_viol_cntr = 0;
 
1037         props->active_width = 2;
 
1038         props->active_speed = 2;
 
1039         props->max_msg_sz = -1;
 
1044 static ssize_t show_rev(struct class_device *cdev, char *buf)
 
1046         struct iwch_dev *dev = container_of(cdev, struct iwch_dev,
 
1048         PDBG("%s class dev 0x%p\n", __FUNCTION__, cdev);
 
1049         return sprintf(buf, "%d\n", dev->rdev.t3cdev_p->type);
 
1052 static ssize_t show_fw_ver(struct class_device *cdev, char *buf)
 
1054         struct iwch_dev *dev = container_of(cdev, struct iwch_dev,
 
1056         struct ethtool_drvinfo info;
 
1057         struct net_device *lldev = dev->rdev.t3cdev_p->lldev;
 
1059         PDBG("%s class dev 0x%p\n", __FUNCTION__, cdev);
 
1061         lldev->ethtool_ops->get_drvinfo(lldev, &info);
 
1063         return sprintf(buf, "%s\n", info.fw_version);
 
1066 static ssize_t show_hca(struct class_device *cdev, char *buf)
 
1068         struct iwch_dev *dev = container_of(cdev, struct iwch_dev,
 
1070         struct ethtool_drvinfo info;
 
1071         struct net_device *lldev = dev->rdev.t3cdev_p->lldev;
 
1073         PDBG("%s class dev 0x%p\n", __FUNCTION__, cdev);
 
1075         lldev->ethtool_ops->get_drvinfo(lldev, &info);
 
1077         return sprintf(buf, "%s\n", info.driver);
 
1080 static ssize_t show_board(struct class_device *cdev, char *buf)
 
1082         struct iwch_dev *dev = container_of(cdev, struct iwch_dev,
 
1084         PDBG("%s class dev 0x%p\n", __FUNCTION__, dev);
 
1085         return sprintf(buf, "%x.%x\n", dev->rdev.rnic_info.pdev->vendor,
 
1086                                        dev->rdev.rnic_info.pdev->device);
 
1089 static CLASS_DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
 
1090 static CLASS_DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
 
1091 static CLASS_DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
 
1092 static CLASS_DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
 
1094 static struct class_device_attribute *iwch_class_attributes[] = {
 
1095         &class_device_attr_hw_rev,
 
1096         &class_device_attr_fw_ver,
 
1097         &class_device_attr_hca_type,
 
1098         &class_device_attr_board_id
 
1101 int iwch_register_device(struct iwch_dev *dev)
 
1106         PDBG("%s iwch_dev %p\n", __FUNCTION__, dev);
 
1107         strlcpy(dev->ibdev.name, "cxgb3_%d", IB_DEVICE_NAME_MAX);
 
1108         memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
 
1109         memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
 
1110         dev->ibdev.owner = THIS_MODULE;
 
1111         dev->device_cap_flags =
 
1112             (IB_DEVICE_ZERO_STAG |
 
1113              IB_DEVICE_SEND_W_INV | IB_DEVICE_MEM_WINDOW);
 
1115         dev->ibdev.uverbs_cmd_mask =
 
1116             (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
 
1117             (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
 
1118             (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
 
1119             (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
 
1120             (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
 
1121             (1ull << IB_USER_VERBS_CMD_REG_MR) |
 
1122             (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
 
1123             (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
 
1124             (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
 
1125             (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
 
1126             (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
 
1127             (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
 
1128             (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
 
1129             (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
 
1130             (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
 
1131             (1ull << IB_USER_VERBS_CMD_POST_SEND) |
 
1132             (1ull << IB_USER_VERBS_CMD_POST_RECV);
 
1133         dev->ibdev.node_type = RDMA_NODE_RNIC;
 
1134         memcpy(dev->ibdev.node_desc, IWCH_NODE_DESC, sizeof(IWCH_NODE_DESC));
 
1135         dev->ibdev.phys_port_cnt = dev->rdev.port_info.nports;
 
1136         dev->ibdev.num_comp_vectors = 1;
 
1137         dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
 
1138         dev->ibdev.query_device = iwch_query_device;
 
1139         dev->ibdev.query_port = iwch_query_port;
 
1140         dev->ibdev.modify_port = iwch_modify_port;
 
1141         dev->ibdev.query_pkey = iwch_query_pkey;
 
1142         dev->ibdev.query_gid = iwch_query_gid;
 
1143         dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
 
1144         dev->ibdev.dealloc_ucontext = iwch_dealloc_ucontext;
 
1145         dev->ibdev.mmap = iwch_mmap;
 
1146         dev->ibdev.alloc_pd = iwch_allocate_pd;
 
1147         dev->ibdev.dealloc_pd = iwch_deallocate_pd;
 
1148         dev->ibdev.create_ah = iwch_ah_create;
 
1149         dev->ibdev.destroy_ah = iwch_ah_destroy;
 
1150         dev->ibdev.create_qp = iwch_create_qp;
 
1151         dev->ibdev.modify_qp = iwch_ib_modify_qp;
 
1152         dev->ibdev.destroy_qp = iwch_destroy_qp;
 
1153         dev->ibdev.create_cq = iwch_create_cq;
 
1154         dev->ibdev.destroy_cq = iwch_destroy_cq;
 
1155         dev->ibdev.resize_cq = iwch_resize_cq;
 
1156         dev->ibdev.poll_cq = iwch_poll_cq;
 
1157         dev->ibdev.get_dma_mr = iwch_get_dma_mr;
 
1158         dev->ibdev.reg_phys_mr = iwch_register_phys_mem;
 
1159         dev->ibdev.rereg_phys_mr = iwch_reregister_phys_mem;
 
1160         dev->ibdev.reg_user_mr = iwch_reg_user_mr;
 
1161         dev->ibdev.dereg_mr = iwch_dereg_mr;
 
1162         dev->ibdev.alloc_mw = iwch_alloc_mw;
 
1163         dev->ibdev.bind_mw = iwch_bind_mw;
 
1164         dev->ibdev.dealloc_mw = iwch_dealloc_mw;
 
1166         dev->ibdev.attach_mcast = iwch_multicast_attach;
 
1167         dev->ibdev.detach_mcast = iwch_multicast_detach;
 
1168         dev->ibdev.process_mad = iwch_process_mad;
 
1170         dev->ibdev.req_notify_cq = iwch_arm_cq;
 
1171         dev->ibdev.post_send = iwch_post_send;
 
1172         dev->ibdev.post_recv = iwch_post_receive;
 
1175         dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
 
1176         if (!dev->ibdev.iwcm)
 
1179         dev->ibdev.iwcm->connect = iwch_connect;
 
1180         dev->ibdev.iwcm->accept = iwch_accept_cr;
 
1181         dev->ibdev.iwcm->reject = iwch_reject_cr;
 
1182         dev->ibdev.iwcm->create_listen = iwch_create_listen;
 
1183         dev->ibdev.iwcm->destroy_listen = iwch_destroy_listen;
 
1184         dev->ibdev.iwcm->add_ref = iwch_qp_add_ref;
 
1185         dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
 
1186         dev->ibdev.iwcm->get_qp = iwch_get_qp;
 
1188         ret = ib_register_device(&dev->ibdev);
 
1192         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i) {
 
1193                 ret = class_device_create_file(&dev->ibdev.class_dev,
 
1194                                                iwch_class_attributes[i]);
 
1201         ib_unregister_device(&dev->ibdev);
 
1206 void iwch_unregister_device(struct iwch_dev *dev)
 
1210         PDBG("%s iwch_dev %p\n", __FUNCTION__, dev);
 
1211         for (i = 0; i < ARRAY_SIZE(iwch_class_attributes); ++i)
 
1212                 class_device_remove_file(&dev->ibdev.class_dev,
 
1213                                          iwch_class_attributes[i]);
 
1214         ib_unregister_device(&dev->ibdev);