drivers/infiniband/hw/ehca/ehca_mrmw.c

   1 /*
   2  *  IBM eServer eHCA Infiniband device driver for Linux on POWER
   3  *
   4  *  MR/MW functions
   5  *
   6  *  Authors: Dietmar Decker <ddecker@de.ibm.com>
   7  *           Christoph Raisch <raisch@de.ibm.com>
   8  *           Hoang-Nam Nguyen <hnguyen@de.ibm.com>
   9  *
  10  *  Copyright (c) 2005 IBM Corporation
  11  *
  12  *  All rights reserved.
  13  *
  14  *  This source code is distributed under a dual license of GPL v2.0 and OpenIB
  15  *  BSD.
  16  *
  17  * OpenIB BSD License
  18  *
  19  * Redistribution and use in source and binary forms, with or without
  20  * modification, are permitted provided that the following conditions are met:
  21  *
  22  * Redistributions of source code must retain the above copyright notice, this
  23  * list of conditions and the following disclaimer.
  24  *
  25  * Redistributions in binary form must reproduce the above copyright notice,
  26  * this list of conditions and the following disclaimer in the documentation
  27  * and/or other materials
  28  * provided with the distribution.
  29  *
  30  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  31  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  33  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
  34  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  35  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  36  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  37  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER
  38  * IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  39  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  40  * POSSIBILITY OF SUCH DAMAGE.
  41  */
  42
  43 #include <asm/current.h>
  44
  45 #include <rdma/ib_umem.h>
  46
  47 #include "ehca_iverbs.h"
  48 #include "ehca_mrmw.h"
  49 #include "hcp_if.h"
  50 #include "hipz_hw.h"
  51
  52 #define NUM_CHUNKS(length, chunk_size) \
  53         (((length) + (chunk_size - 1)) / (chunk_size))
  54 /* max number of rpages (per hcall register_rpages) */
  55 #define MAX_RPAGES 512
  56
  57 static struct kmem_cache *mr_cache;
  58 static struct kmem_cache *mw_cache;
  59
  60 enum ehca_mr_pgsize {
  61         EHCA_MR_PGSIZE4K  = 0x1000L,
  62         EHCA_MR_PGSIZE64K = 0x10000L,
  63         EHCA_MR_PGSIZE1M  = 0x100000L,
  64         EHCA_MR_PGSIZE16M = 0x1000000L
  65 };
  66
  67 static u32 ehca_encode_hwpage_size(u32 pgsize)
  68 {
  69         u32 idx = 0;
  70         pgsize >>= 12;
  71         /*
  72          * map mr page size into hw code:
  73          * 0, 1, 2, 3 for 4K, 64K, 1M, 64M
  74          */
  75         while (!(pgsize & 1)) {
  76                 idx++;
  77                 pgsize >>= 4;
  78         }
  79         return idx;
  80 }
  81
  82 static u64 ehca_get_max_hwpage_size(struct ehca_shca *shca)
  83 {
  84         if (shca->hca_cap_mr_pgsize & HCA_CAP_MR_PGSIZE_16M)
  85                 return EHCA_MR_PGSIZE16M;
  86         return EHCA_MR_PGSIZE4K;
  87 }
  88
  89 static struct ehca_mr *ehca_mr_new(void)
  90 {
  91         struct ehca_mr *me;
  92
  93         me = kmem_cache_zalloc(mr_cache, GFP_KERNEL);
  94         if (me)
  95                 spin_lock_init(&me->mrlock);
  96         else
  97                 ehca_gen_err("alloc failed");
  98
  99         return me;
 100 }
 101
 102 static void ehca_mr_delete(struct ehca_mr *me)
 103 {
 104         kmem_cache_free(mr_cache, me);
 105 }
 106
 107 static struct ehca_mw *ehca_mw_new(void)
 108 {
 109         struct ehca_mw *me;
 110
 111         me = kmem_cache_zalloc(mw_cache, GFP_KERNEL);
 112         if (me)
 113                 spin_lock_init(&me->mwlock);
 114         else
 115                 ehca_gen_err("alloc failed");
 116
 117         return me;
 118 }
 119
 120 static void ehca_mw_delete(struct ehca_mw *me)
 121 {
 122         kmem_cache_free(mw_cache, me);
 123 }
 124
 125 /*----------------------------------------------------------------------*/
 126
 127 struct ib_mr *ehca_get_dma_mr(struct ib_pd *pd, int mr_access_flags)
 128 {
 129         struct ib_mr *ib_mr;
 130         int ret;
 131         struct ehca_mr *e_maxmr;
 132         struct ehca_pd *e_pd = container_of(pd, struct ehca_pd, ib_pd);
 133         struct ehca_shca *shca =
 134                 container_of(pd->device, struct ehca_shca, ib_device);
 135
 136         if (shca->maxmr) {
 137                 e_maxmr = ehca_mr_new();
 138                 if (!e_maxmr) {
 139                         ehca_err(&shca->ib_device, "out of memory");
 140                         ib_mr = ERR_PTR(-ENOMEM);
 141                         goto get_dma_mr_exit0;
 142                 }
 143
 144                 ret = ehca_reg_maxmr(shca, e_maxmr, (u64 *)KERNELBASE,
 145                                      mr_access_flags, e_pd,
 146                                      &e_maxmr->ib.ib_mr.lkey,
 147                                      &e_maxmr->ib.ib_mr.rkey);
 148                 if (ret) {
 149                         ehca_mr_delete(e_maxmr);
 150                         ib_mr = ERR_PTR(ret);
 151                         goto get_dma_mr_exit0;
 152                 }
 153                 ib_mr = &e_maxmr->ib.ib_mr;
 154         } else {
 155                 ehca_err(&shca->ib_device, "no internal max-MR exist!");
 156                 ib_mr = ERR_PTR(-EINVAL);
 157                 goto get_dma_mr_exit0;
 158         }
 159
 160 get_dma_mr_exit0:
 161         if (IS_ERR(ib_mr))
 162                 ehca_err(&shca->ib_device, "rc=%lx pd=%p mr_access_flags=%x ",
 163                          PTR_ERR(ib_mr), pd, mr_access_flags);
 164         return ib_mr;
 165 } /* end ehca_get_dma_mr() */
 166
 167 /*----------------------------------------------------------------------*/
 168
 169 struct ib_mr *ehca_reg_phys_mr(struct ib_pd *pd,
 170                                struct ib_phys_buf *phys_buf_array,
 171                                int num_phys_buf,
 172                                int mr_access_flags,
 173                                u64 *iova_start)
 174 {
 175         struct ib_mr *ib_mr;
 176         int ret;
 177         struct ehca_mr *e_mr;
 178         struct ehca_shca *shca =
 179                 container_of(pd->device, struct ehca_shca, ib_device);
 180         struct ehca_pd *e_pd = container_of(pd, struct ehca_pd, ib_pd);
 181
 182         u64 size;
 183
 184         if ((num_phys_buf <= 0) || !phys_buf_array) {
 185                 ehca_err(pd->device, "bad input values: num_phys_buf=%x "
 186                          "phys_buf_array=%p", num_phys_buf, phys_buf_array);
 187                 ib_mr = ERR_PTR(-EINVAL);
 188                 goto reg_phys_mr_exit0;
 189         }
 190         if (((mr_access_flags & IB_ACCESS_REMOTE_WRITE) &&
 191              !(mr_access_flags & IB_ACCESS_LOCAL_WRITE)) ||
 192             ((mr_access_flags & IB_ACCESS_REMOTE_ATOMIC) &&
 193              !(mr_access_flags & IB_ACCESS_LOCAL_WRITE))) {
 194                 /*
 195                  * Remote Write Access requires Local Write Access
 196                  * Remote Atomic Access requires Local Write Access
 197                  */
 198                 ehca_err(pd->device, "bad input values: mr_access_flags=%x",
 199                          mr_access_flags);
 200                 ib_mr = ERR_PTR(-EINVAL);
 201                 goto reg_phys_mr_exit0;
 202         }
 203
 204         /* check physical buffer list and calculate size */
 205         ret = ehca_mr_chk_buf_and_calc_size(phys_buf_array, num_phys_buf,
 206                                             iova_start, &size);
 207         if (ret) {
 208                 ib_mr = ERR_PTR(ret);
 209                 goto reg_phys_mr_exit0;
 210         }
 211         if ((size == 0) ||
 212             (((u64)iova_start + size) < (u64)iova_start)) {
 213                 ehca_err(pd->device, "bad input values: size=%lx iova_start=%p",
 214                          size, iova_start);
 215                 ib_mr = ERR_PTR(-EINVAL);
 216                 goto reg_phys_mr_exit0;
 217         }
 218
 219         e_mr = ehca_mr_new();
 220         if (!e_mr) {
 221                 ehca_err(pd->device, "out of memory");
 222                 ib_mr = ERR_PTR(-ENOMEM);
 223                 goto reg_phys_mr_exit0;
 224         }
 225
 226         /* register MR on HCA */
 227         if (ehca_mr_is_maxmr(size, iova_start)) {
 228                 e_mr->flags |= EHCA_MR_FLAG_MAXMR;
 229                 ret = ehca_reg_maxmr(shca, e_mr, iova_start, mr_access_flags,
 230                                      e_pd, &e_mr->ib.ib_mr.lkey,
 231                                      &e_mr->ib.ib_mr.rkey);
 232                 if (ret) {
 233                         ib_mr = ERR_PTR(ret);
 234                         goto reg_phys_mr_exit1;
 235                 }
 236         } else {
 237                 struct ehca_mr_pginfo pginfo;
 238                 u32 num_kpages;
 239                 u32 num_hwpages;
 240                 u64 hw_pgsize;
 241
 242                 num_kpages = NUM_CHUNKS(((u64)iova_start % PAGE_SIZE) + size,
 243                                         PAGE_SIZE);
 244                 /* for kernel space we try most possible pgsize */
 245                 hw_pgsize = ehca_get_max_hwpage_size(shca);
 246                 num_hwpages = NUM_CHUNKS(((u64)iova_start % hw_pgsize) + size,
 247                                          hw_pgsize);
 248                 memset(&pginfo, 0, sizeof(pginfo));
 249                 pginfo.type = EHCA_MR_PGI_PHYS;
 250                 pginfo.num_kpages = num_kpages;
 251                 pginfo.hwpage_size = hw_pgsize;
 252                 pginfo.num_hwpages = num_hwpages;
 253                 pginfo.u.phy.num_phys_buf = num_phys_buf;
 254                 pginfo.u.phy.phys_buf_array = phys_buf_array;
 255                 pginfo.next_hwpage =
 256                         ((u64)iova_start & ~(hw_pgsize - 1)) / hw_pgsize;
 257
 258                 ret = ehca_reg_mr(shca, e_mr, iova_start, size, mr_access_flags,
 259                                   e_pd, &pginfo, &e_mr->ib.ib_mr.lkey,
 260                                   &e_mr->ib.ib_mr.rkey);
 261                 if (ret) {
 262                         ib_mr = ERR_PTR(ret);
 263                         goto reg_phys_mr_exit1;
 264                 }
 265         }
 266
 267         /* successful registration of all pages */
 268         return &e_mr->ib.ib_mr;
 269
 270 reg_phys_mr_exit1:
 271         ehca_mr_delete(e_mr);
 272 reg_phys_mr_exit0:
 273         if (IS_ERR(ib_mr))
 274                 ehca_err(pd->device, "rc=%lx pd=%p phys_buf_array=%p "
 275                          "num_phys_buf=%x mr_access_flags=%x iova_start=%p",
 276                          PTR_ERR(ib_mr), pd, phys_buf_array,
 277                          num_phys_buf, mr_access_flags, iova_start);
 278         return ib_mr;
 279 } /* end ehca_reg_phys_mr() */
 280
 281 /*----------------------------------------------------------------------*/
 282
 283 struct ib_mr *ehca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 284                                u64 virt, int mr_access_flags,
 285                                struct ib_udata *udata)
 286 {
 287         struct ib_mr *ib_mr;
 288         struct ehca_mr *e_mr;
 289         struct ehca_shca *shca =
 290                 container_of(pd->device, struct ehca_shca, ib_device);
 291         struct ehca_pd *e_pd = container_of(pd, struct ehca_pd, ib_pd);
 292         struct ehca_mr_pginfo pginfo;
 293         int ret;
 294         u32 num_kpages;
 295         u32 num_hwpages;
 296         u64 hwpage_size;
 297
 298         if (!pd) {
 299                 ehca_gen_err("bad pd=%p", pd);
 300                 return ERR_PTR(-EFAULT);
 301         }
 302
 303         if (((mr_access_flags & IB_ACCESS_REMOTE_WRITE) &&
 304              !(mr_access_flags & IB_ACCESS_LOCAL_WRITE)) ||
 305             ((mr_access_flags & IB_ACCESS_REMOTE_ATOMIC) &&
 306              !(mr_access_flags & IB_ACCESS_LOCAL_WRITE))) {
 307                 /*
 308                  * Remote Write Access requires Local Write Access
 309                  * Remote Atomic Access requires Local Write Access
 310                  */
 311                 ehca_err(pd->device, "bad input values: mr_access_flags=%x",
 312                          mr_access_flags);
 313                 ib_mr = ERR_PTR(-EINVAL);
 314                 goto reg_user_mr_exit0;
 315         }
 316
 317         if (length == 0 || virt + length < virt) {
 318                 ehca_err(pd->device, "bad input values: length=%lx "
 319                          "virt_base=%lx", length, virt);
 320                 ib_mr = ERR_PTR(-EINVAL);
 321                 goto reg_user_mr_exit0;
 322         }
 323
 324         e_mr = ehca_mr_new();
 325         if (!e_mr) {
 326                 ehca_err(pd->device, "out of memory");
 327                 ib_mr = ERR_PTR(-ENOMEM);
 328                 goto reg_user_mr_exit0;
 329         }
 330
 331         e_mr->umem = ib_umem_get(pd->uobject->context, start, length,
 332                                  mr_access_flags);
 333         if (IS_ERR(e_mr->umem)) {
 334                 ib_mr = (void *)e_mr->umem;
 335                 goto reg_user_mr_exit1;
 336         }
 337
 338         if (e_mr->umem->page_size != PAGE_SIZE) {
 339                 ehca_err(pd->device, "page size not supported, "
 340                          "e_mr->umem->page_size=%x", e_mr->umem->page_size);
 341                 ib_mr = ERR_PTR(-EINVAL);
 342                 goto reg_user_mr_exit2;
 343         }
 344
 345         /* determine number of MR pages */
 346         num_kpages = NUM_CHUNKS((virt % PAGE_SIZE) + length, PAGE_SIZE);
 347         /* select proper hw_pgsize */
 348         if (ehca_mr_largepage &&
 349             (shca->hca_cap_mr_pgsize & HCA_CAP_MR_PGSIZE_16M)) {
 350                 if (length <= EHCA_MR_PGSIZE4K
 351                     && PAGE_SIZE == EHCA_MR_PGSIZE4K)
 352                         hwpage_size = EHCA_MR_PGSIZE4K;
 353                 else if (length <= EHCA_MR_PGSIZE64K)
 354                         hwpage_size = EHCA_MR_PGSIZE64K;
 355                 else if (length <= EHCA_MR_PGSIZE1M)
 356                         hwpage_size = EHCA_MR_PGSIZE1M;
 357                 else
 358                         hwpage_size = EHCA_MR_PGSIZE16M;
 359         } else
 360                 hwpage_size = EHCA_MR_PGSIZE4K;
 361         ehca_dbg(pd->device, "hwpage_size=%lx", hwpage_size);
 362
 363 reg_user_mr_fallback:
 364         num_hwpages = NUM_CHUNKS((virt % hwpage_size) + length, hwpage_size);
 365         /* register MR on HCA */
 366         memset(&pginfo, 0, sizeof(pginfo));
 367         pginfo.type = EHCA_MR_PGI_USER;
 368         pginfo.hwpage_size = hwpage_size;
 369         pginfo.num_kpages = num_kpages;
 370         pginfo.num_hwpages = num_hwpages;
 371         pginfo.u.usr.region = e_mr->umem;
 372         pginfo.next_hwpage = e_mr->umem->offset / hwpage_size;
 373         pginfo.u.usr.next_chunk = list_prepare_entry(pginfo.u.usr.next_chunk,
 374                                                      (&e_mr->umem->chunk_list),
 375                                                      list);
 376
 377         ret = ehca_reg_mr(shca, e_mr, (u64 *)virt, length, mr_access_flags,
 378                           e_pd, &pginfo, &e_mr->ib.ib_mr.lkey,
 379                           &e_mr->ib.ib_mr.rkey);
 380         if (ret == -EINVAL && pginfo.hwpage_size > PAGE_SIZE) {
 381                 ehca_warn(pd->device, "failed to register mr "
 382                           "with hwpage_size=%lx", hwpage_size);
 383                 ehca_info(pd->device, "try to register mr with "
 384                           "kpage_size=%lx", PAGE_SIZE);
 385                 /*
 386                  * this means kpages are not contiguous for a hw page
 387                  * try kernel page size as fallback solution
 388                  */
 389                 hwpage_size = PAGE_SIZE;
 390                 goto reg_user_mr_fallback;
 391         }
 392         if (ret) {
 393                 ib_mr = ERR_PTR(ret);
 394                 goto reg_user_mr_exit2;
 395         }
 396
 397         /* successful registration of all pages */
 398         return &e_mr->ib.ib_mr;
 399
 400 reg_user_mr_exit2:
 401         ib_umem_release(e_mr->umem);
 402 reg_user_mr_exit1:
 403         ehca_mr_delete(e_mr);
 404 reg_user_mr_exit0:
 405         if (IS_ERR(ib_mr))
 406                 ehca_err(pd->device, "rc=%lx pd=%p mr_access_flags=%x"
 407                          " udata=%p",
 408                          PTR_ERR(ib_mr), pd, mr_access_flags, udata);
 409         return ib_mr;
 410 } /* end ehca_reg_user_mr() */
 411
 412 /*----------------------------------------------------------------------*/
 413
 414 int ehca_rereg_phys_mr(struct ib_mr *mr,
 415                        int mr_rereg_mask,
 416                        struct ib_pd *pd,
 417                        struct ib_phys_buf *phys_buf_array,
 418                        int num_phys_buf,
 419                        int mr_access_flags,
 420                        u64 *iova_start)
 421 {
 422         int ret;
 423
 424         struct ehca_shca *shca =
 425                 container_of(mr->device, struct ehca_shca, ib_device);
 426         struct ehca_mr *e_mr = container_of(mr, struct ehca_mr, ib.ib_mr);
 427         struct ehca_pd *my_pd = container_of(mr->pd, struct ehca_pd, ib_pd);
 428         u64 new_size;
 429         u64 *new_start;
 430         u32 new_acl;
 431         struct ehca_pd *new_pd;
 432         u32 tmp_lkey, tmp_rkey;
 433         unsigned long sl_flags;
 434         u32 num_kpages = 0;
 435         u32 num_hwpages = 0;
 436         struct ehca_mr_pginfo pginfo;
 437         u32 cur_pid = current->tgid;
 438
 439         if (my_pd->ib_pd.uobject && my_pd->ib_pd.uobject->context &&
 440             (my_pd->ownpid != cur_pid)) {
 441                 ehca_err(mr->device, "Invalid caller pid=%x ownpid=%x",
 442                          cur_pid, my_pd->ownpid);
 443                 ret = -EINVAL;
 444                 goto rereg_phys_mr_exit0;
 445         }
 446
 447         if (!(mr_rereg_mask & IB_MR_REREG_TRANS)) {
 448                 /* TODO not supported, because PHYP rereg hCall needs pages */
 449                 ehca_err(mr->device, "rereg without IB_MR_REREG_TRANS not "
 450                          "supported yet, mr_rereg_mask=%x", mr_rereg_mask);
 451                 ret = -EINVAL;
 452                 goto rereg_phys_mr_exit0;
 453         }
 454
 455         if (mr_rereg_mask & IB_MR_REREG_PD) {
 456                 if (!pd) {
 457                         ehca_err(mr->device, "rereg with bad pd, pd=%p "
 458                                  "mr_rereg_mask=%x", pd, mr_rereg_mask);
 459                         ret = -EINVAL;
 460                         goto rereg_phys_mr_exit0;
 461                 }
 462         }
 463
 464         if ((mr_rereg_mask &
 465              ~(IB_MR_REREG_TRANS | IB_MR_REREG_PD | IB_MR_REREG_ACCESS)) ||
 466             (mr_rereg_mask == 0)) {
 467                 ret = -EINVAL;
 468                 goto rereg_phys_mr_exit0;
 469         }
 470
 471         /* check other parameters */
 472         if (e_mr == shca->maxmr) {
 473                 /* should be impossible, however reject to be sure */
 474                 ehca_err(mr->device, "rereg internal max-MR impossible, mr=%p "
 475                          "shca->maxmr=%p mr->lkey=%x",
 476                          mr, shca->maxmr, mr->lkey);
 477                 ret = -EINVAL;
 478                 goto rereg_phys_mr_exit0;
 479         }
 480         if (mr_rereg_mask & IB_MR_REREG_TRANS) { /* transl., i.e. addr/size */
 481                 if (e_mr->flags & EHCA_MR_FLAG_FMR) {
 482                         ehca_err(mr->device, "not supported for FMR, mr=%p "
 483                                  "flags=%x", mr, e_mr->flags);
 484                         ret = -EINVAL;
 485                         goto rereg_phys_mr_exit0;
 486                 }
 487                 if (!phys_buf_array || num_phys_buf <= 0) {
 488                         ehca_err(mr->device, "bad input values mr_rereg_mask=%x"
 489                                  " phys_buf_array=%p num_phys_buf=%x",
 490                                  mr_rereg_mask, phys_buf_array, num_phys_buf);
 491                         ret = -EINVAL;
 492                         goto rereg_phys_mr_exit0;
 493                 }
 494         }
 495         if ((mr_rereg_mask & IB_MR_REREG_ACCESS) &&     /* change ACL */
 496             (((mr_access_flags & IB_ACCESS_REMOTE_WRITE) &&
 497               !(mr_access_flags & IB_ACCESS_LOCAL_WRITE)) ||
 498              ((mr_access_flags & IB_ACCESS_REMOTE_ATOMIC) &&
 499               !(mr_access_flags & IB_ACCESS_LOCAL_WRITE)))) {
 500                 /*
 501                  * Remote Write Access requires Local Write Access
 502                  * Remote Atomic Access requires Local Write Access
 503                  */
 504                 ehca_err(mr->device, "bad input values: mr_rereg_mask=%x "
 505                          "mr_access_flags=%x", mr_rereg_mask, mr_access_flags);
 506                 ret = -EINVAL;
 507                 goto rereg_phys_mr_exit0;
 508         }
 509
 510         /* set requested values dependent on rereg request */
 511         spin_lock_irqsave(&e_mr->mrlock, sl_flags);
 512         new_start = e_mr->start;
 513         new_size = e_mr->size;
 514         new_acl = e_mr->acl;
 515         new_pd = container_of(mr->pd, struct ehca_pd, ib_pd);
 516
 517         if (mr_rereg_mask & IB_MR_REREG_TRANS) {
 518                 u64 hw_pgsize = ehca_get_max_hwpage_size(shca);
 519
 520                 new_start = iova_start; /* change address */
 521                 /* check physical buffer list and calculate size */
 522                 ret = ehca_mr_chk_buf_and_calc_size(phys_buf_array,
 523                                                     num_phys_buf, iova_start,
 524                                                     &new_size);
 525                 if (ret)
 526                         goto rereg_phys_mr_exit1;
 527                 if ((new_size == 0) ||
 528                     (((u64)iova_start + new_size) < (u64)iova_start)) {
 529                         ehca_err(mr->device, "bad input values: new_size=%lx "
 530                                  "iova_start=%p", new_size, iova_start);
 531                         ret = -EINVAL;
 532                         goto rereg_phys_mr_exit1;
 533                 }
 534                 num_kpages = NUM_CHUNKS(((u64)new_start % PAGE_SIZE) +
 535                                         new_size, PAGE_SIZE);
 536                 num_hwpages = NUM_CHUNKS(((u64)new_start % hw_pgsize) +
 537                                          new_size, hw_pgsize);
 538                 memset(&pginfo, 0, sizeof(pginfo));
 539                 pginfo.type = EHCA_MR_PGI_PHYS;
 540                 pginfo.num_kpages = num_kpages;
 541                 pginfo.hwpage_size = hw_pgsize;
 542                 pginfo.num_hwpages = num_hwpages;
 543                 pginfo.u.phy.num_phys_buf = num_phys_buf;
 544                 pginfo.u.phy.phys_buf_array = phys_buf_array;
 545                 pginfo.next_hwpage =
 546                         ((u64)iova_start & ~(hw_pgsize - 1)) / hw_pgsize;
 547         }
 548         if (mr_rereg_mask & IB_MR_REREG_ACCESS)
 549                 new_acl = mr_access_flags;
 550         if (mr_rereg_mask & IB_MR_REREG_PD)
 551                 new_pd = container_of(pd, struct ehca_pd, ib_pd);
 552
 553         ret = ehca_rereg_mr(shca, e_mr, new_start, new_size, new_acl,
 554                             new_pd, &pginfo, &tmp_lkey, &tmp_rkey);
 555         if (ret)
 556                 goto rereg_phys_mr_exit1;
 557
 558         /* successful reregistration */
 559         if (mr_rereg_mask & IB_MR_REREG_PD)
 560                 mr->pd = pd;
 561         mr->lkey = tmp_lkey;
 562         mr->rkey = tmp_rkey;
 563
 564 rereg_phys_mr_exit1:
 565         spin_unlock_irqrestore(&e_mr->mrlock, sl_flags);
 566 rereg_phys_mr_exit0:
 567         if (ret)
 568                 ehca_err(mr->device, "ret=%x mr=%p mr_rereg_mask=%x pd=%p "
 569                          "phys_buf_array=%p num_phys_buf=%x mr_access_flags=%x "
 570                          "iova_start=%p",
 571                          ret, mr, mr_rereg_mask, pd, phys_buf_array,
 572                          num_phys_buf, mr_access_flags, iova_start);
 573         return ret;
 574 } /* end ehca_rereg_phys_mr() */
 575
 576 /*----------------------------------------------------------------------*/
 577
 578 int ehca_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr)
 579 {
 580         int ret = 0;
 581         u64 h_ret;
 582         struct ehca_shca *shca =
 583                 container_of(mr->device, struct ehca_shca, ib_device);
 584         struct ehca_mr *e_mr = container_of(mr, struct ehca_mr, ib.ib_mr);
 585         struct ehca_pd *my_pd = container_of(mr->pd, struct ehca_pd, ib_pd);
 586         u32 cur_pid = current->tgid;
 587         unsigned long sl_flags;
 588         struct ehca_mr_hipzout_parms hipzout;
 589
 590         if (my_pd->ib_pd.uobject && my_pd->ib_pd.uobject->context &&
 591             (my_pd->ownpid != cur_pid)) {
 592                 ehca_err(mr->device, "Invalid caller pid=%x ownpid=%x",
 593                          cur_pid, my_pd->ownpid);
 594                 ret = -EINVAL;
 595                 goto query_mr_exit0;
 596         }
 597
 598         if ((e_mr->flags & EHCA_MR_FLAG_FMR)) {
 599                 ehca_err(mr->device, "not supported for FMR, mr=%p e_mr=%p "
 600                          "e_mr->flags=%x", mr, e_mr, e_mr->flags);
 601                 ret = -EINVAL;
 602                 goto query_mr_exit0;
 603         }
 604
 605         memset(mr_attr, 0, sizeof(struct ib_mr_attr));
 606         spin_lock_irqsave(&e_mr->mrlock, sl_flags);
 607
 608         h_ret = hipz_h_query_mr(shca->ipz_hca_handle, e_mr, &hipzout);
 609         if (h_ret != H_SUCCESS) {
 610                 ehca_err(mr->device, "hipz_mr_query failed, h_ret=%lx mr=%p "
 611                          "hca_hndl=%lx mr_hndl=%lx lkey=%x",
 612                          h_ret, mr, shca->ipz_hca_handle.handle,
 613                          e_mr->ipz_mr_handle.handle, mr->lkey);
 614                 ret = ehca2ib_return_code(h_ret);
 615                 goto query_mr_exit1;
 616         }
 617         mr_attr->pd = mr->pd;
 618         mr_attr->device_virt_addr = hipzout.vaddr;
 619         mr_attr->size = hipzout.len;
 620         mr_attr->lkey = hipzout.lkey;
 621         mr_attr->rkey = hipzout.rkey;
 622         ehca_mrmw_reverse_map_acl(&hipzout.acl, &mr_attr->mr_access_flags);
 623
 624 query_mr_exit1:
 625         spin_unlock_irqrestore(&e_mr->mrlock, sl_flags);
 626 query_mr_exit0:
 627         if (ret)
 628                 ehca_err(mr->device, "ret=%x mr=%p mr_attr=%p",
 629                          ret, mr, mr_attr);
 630         return ret;
 631 } /* end ehca_query_mr() */
 632
 633 /*----------------------------------------------------------------------*/
 634
 635 int ehca_dereg_mr(struct ib_mr *mr)
 636 {
 637         int ret = 0;
 638         u64 h_ret;
 639         struct ehca_shca *shca =
 640                 container_of(mr->device, struct ehca_shca, ib_device);
 641         struct ehca_mr *e_mr = container_of(mr, struct ehca_mr, ib.ib_mr);
 642         struct ehca_pd *my_pd = container_of(mr->pd, struct ehca_pd, ib_pd);
 643         u32 cur_pid = current->tgid;
 644
 645         if (my_pd->ib_pd.uobject && my_pd->ib_pd.uobject->context &&
 646             (my_pd->ownpid != cur_pid)) {
 647                 ehca_err(mr->device, "Invalid caller pid=%x ownpid=%x",
 648                          cur_pid, my_pd->ownpid);
 649                 ret = -EINVAL;
 650                 goto dereg_mr_exit0;
 651         }
 652
 653         if ((e_mr->flags & EHCA_MR_FLAG_FMR)) {
 654                 ehca_err(mr->device, "not supported for FMR, mr=%p e_mr=%p "
 655                          "e_mr->flags=%x", mr, e_mr, e_mr->flags);
 656                 ret = -EINVAL;
 657                 goto dereg_mr_exit0;
 658         } else if (e_mr == shca->maxmr) {
 659                 /* should be impossible, however reject to be sure */
 660                 ehca_err(mr->device, "dereg internal max-MR impossible, mr=%p "
 661                          "shca->maxmr=%p mr->lkey=%x",
 662                          mr, shca->maxmr, mr->lkey);
 663                 ret = -EINVAL;
 664                 goto dereg_mr_exit0;
 665         }
 666
 667         /* TODO: BUSY: MR still has bound window(s) */
 668         h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_mr);
 669         if (h_ret != H_SUCCESS) {
 670                 ehca_err(mr->device, "hipz_free_mr failed, h_ret=%lx shca=%p "
 671                          "e_mr=%p hca_hndl=%lx mr_hndl=%lx mr->lkey=%x",
 672                          h_ret, shca, e_mr, shca->ipz_hca_handle.handle,
 673                          e_mr->ipz_mr_handle.handle, mr->lkey);
 674                 ret = ehca2ib_return_code(h_ret);
 675                 goto dereg_mr_exit0;
 676         }
 677
 678         if (e_mr->umem)
 679                 ib_umem_release(e_mr->umem);
 680
 681         /* successful deregistration */
 682         ehca_mr_delete(e_mr);
 683
 684 dereg_mr_exit0:
 685         if (ret)
 686                 ehca_err(mr->device, "ret=%x mr=%p", ret, mr);
 687         return ret;
 688 } /* end ehca_dereg_mr() */
 689
 690 /*----------------------------------------------------------------------*/
 691
 692 struct ib_mw *ehca_alloc_mw(struct ib_pd *pd)
 693 {
 694         struct ib_mw *ib_mw;
 695         u64 h_ret;
 696         struct ehca_mw *e_mw;
 697         struct ehca_pd *e_pd = container_of(pd, struct ehca_pd, ib_pd);
 698         struct ehca_shca *shca =
 699                 container_of(pd->device, struct ehca_shca, ib_device);
 700         struct ehca_mw_hipzout_parms hipzout;
 701
 702         e_mw = ehca_mw_new();
 703         if (!e_mw) {
 704                 ib_mw = ERR_PTR(-ENOMEM);
 705                 goto alloc_mw_exit0;
 706         }
 707
 708         h_ret = hipz_h_alloc_resource_mw(shca->ipz_hca_handle, e_mw,
 709                                          e_pd->fw_pd, &hipzout);
 710         if (h_ret != H_SUCCESS) {
 711                 ehca_err(pd->device, "hipz_mw_allocate failed, h_ret=%lx "
 712                          "shca=%p hca_hndl=%lx mw=%p",
 713                          h_ret, shca, shca->ipz_hca_handle.handle, e_mw);
 714                 ib_mw = ERR_PTR(ehca2ib_return_code(h_ret));
 715                 goto alloc_mw_exit1;
 716         }
 717         /* successful MW allocation */
 718         e_mw->ipz_mw_handle = hipzout.handle;
 719         e_mw->ib_mw.rkey    = hipzout.rkey;
 720         return &e_mw->ib_mw;
 721
 722 alloc_mw_exit1:
 723         ehca_mw_delete(e_mw);
 724 alloc_mw_exit0:
 725         if (IS_ERR(ib_mw))
 726                 ehca_err(pd->device, "rc=%lx pd=%p", PTR_ERR(ib_mw), pd);
 727         return ib_mw;
 728 } /* end ehca_alloc_mw() */
 729
 730 /*----------------------------------------------------------------------*/
 731
 732 int ehca_bind_mw(struct ib_qp *qp,
 733                  struct ib_mw *mw,
 734                  struct ib_mw_bind *mw_bind)
 735 {
 736         /* TODO: not supported up to now */
 737         ehca_gen_err("bind MW currently not supported by HCAD");
 738
 739         return -EPERM;
 740 } /* end ehca_bind_mw() */
 741
 742 /*----------------------------------------------------------------------*/
 743
 744 int ehca_dealloc_mw(struct ib_mw *mw)
 745 {
 746         u64 h_ret;
 747         struct ehca_shca *shca =
 748                 container_of(mw->device, struct ehca_shca, ib_device);
 749         struct ehca_mw *e_mw = container_of(mw, struct ehca_mw, ib_mw);
 750
 751         h_ret = hipz_h_free_resource_mw(shca->ipz_hca_handle, e_mw);
 752         if (h_ret != H_SUCCESS) {
 753                 ehca_err(mw->device, "hipz_free_mw failed, h_ret=%lx shca=%p "
 754                          "mw=%p rkey=%x hca_hndl=%lx mw_hndl=%lx",
 755                          h_ret, shca, mw, mw->rkey, shca->ipz_hca_handle.handle,
 756                          e_mw->ipz_mw_handle.handle);
 757                 return ehca2ib_return_code(h_ret);
 758         }
 759         /* successful deallocation */
 760         ehca_mw_delete(e_mw);
 761         return 0;
 762 } /* end ehca_dealloc_mw() */
 763
 764 /*----------------------------------------------------------------------*/
 765
 766 struct ib_fmr *ehca_alloc_fmr(struct ib_pd *pd,
 767                               int mr_access_flags,
 768                               struct ib_fmr_attr *fmr_attr)
 769 {
 770         struct ib_fmr *ib_fmr;
 771         struct ehca_shca *shca =
 772                 container_of(pd->device, struct ehca_shca, ib_device);
 773         struct ehca_pd *e_pd = container_of(pd, struct ehca_pd, ib_pd);
 774         struct ehca_mr *e_fmr;
 775         int ret;
 776         u32 tmp_lkey, tmp_rkey;
 777         struct ehca_mr_pginfo pginfo;
 778         u64 hw_pgsize;
 779
 780         /* check other parameters */
 781         if (((mr_access_flags & IB_ACCESS_REMOTE_WRITE) &&
 782              !(mr_access_flags & IB_ACCESS_LOCAL_WRITE)) ||
 783             ((mr_access_flags & IB_ACCESS_REMOTE_ATOMIC) &&
 784              !(mr_access_flags & IB_ACCESS_LOCAL_WRITE))) {
 785                 /*
 786                  * Remote Write Access requires Local Write Access
 787                  * Remote Atomic Access requires Local Write Access
 788                  */
 789                 ehca_err(pd->device, "bad input values: mr_access_flags=%x",
 790                          mr_access_flags);
 791                 ib_fmr = ERR_PTR(-EINVAL);
 792                 goto alloc_fmr_exit0;
 793         }
 794         if (mr_access_flags & IB_ACCESS_MW_BIND) {
 795                 ehca_err(pd->device, "bad input values: mr_access_flags=%x",
 796                          mr_access_flags);
 797                 ib_fmr = ERR_PTR(-EINVAL);
 798                 goto alloc_fmr_exit0;
 799         }
 800         if ((fmr_attr->max_pages == 0) || (fmr_attr->max_maps == 0)) {
 801                 ehca_err(pd->device, "bad input values: fmr_attr->max_pages=%x "
 802                          "fmr_attr->max_maps=%x fmr_attr->page_shift=%x",
 803                          fmr_attr->max_pages, fmr_attr->max_maps,
 804                          fmr_attr->page_shift);
 805                 ib_fmr = ERR_PTR(-EINVAL);
 806                 goto alloc_fmr_exit0;
 807         }
 808         hw_pgsize = ehca_get_max_hwpage_size(shca);
 809         if ((1 << fmr_attr->page_shift) != hw_pgsize) {
 810                 ehca_err(pd->device, "unsupported fmr_attr->page_shift=%x",
 811                          fmr_attr->page_shift);
 812                 ib_fmr = ERR_PTR(-EINVAL);
 813                 goto alloc_fmr_exit0;
 814         }
 815
 816         e_fmr = ehca_mr_new();
 817         if (!e_fmr) {
 818                 ib_fmr = ERR_PTR(-ENOMEM);
 819                 goto alloc_fmr_exit0;
 820         }
 821         e_fmr->flags |= EHCA_MR_FLAG_FMR;
 822
 823         /* register MR on HCA */
 824         memset(&pginfo, 0, sizeof(pginfo));
 825         /*
 826          * pginfo.num_hwpages==0, ie register_rpages() will not be called
 827          * but deferred to map_phys_fmr()
 828          */
 829         ret = ehca_reg_mr(shca, e_fmr, NULL,
 830                           fmr_attr->max_pages * (1 << fmr_attr->page_shift),
 831                           mr_access_flags, e_pd, &pginfo,
 832                           &tmp_lkey, &tmp_rkey);
 833         if (ret) {
 834                 ib_fmr = ERR_PTR(ret);
 835                 goto alloc_fmr_exit1;
 836         }
 837
 838         /* successful */
 839         e_fmr->hwpage_size = hw_pgsize;
 840         e_fmr->fmr_page_size = 1 << fmr_attr->page_shift;
 841         e_fmr->fmr_max_pages = fmr_attr->max_pages;
 842         e_fmr->fmr_max_maps = fmr_attr->max_maps;
 843         e_fmr->fmr_map_cnt = 0;
 844         return &e_fmr->ib.ib_fmr;
 845
 846 alloc_fmr_exit1:
 847         ehca_mr_delete(e_fmr);
 848 alloc_fmr_exit0:
 849         return ib_fmr;
 850 } /* end ehca_alloc_fmr() */
 851
 852 /*----------------------------------------------------------------------*/
 853
 854 int ehca_map_phys_fmr(struct ib_fmr *fmr,
 855                       u64 *page_list,
 856                       int list_len,
 857                       u64 iova)
 858 {
 859         int ret;
 860         struct ehca_shca *shca =
 861                 container_of(fmr->device, struct ehca_shca, ib_device);
 862         struct ehca_mr *e_fmr = container_of(fmr, struct ehca_mr, ib.ib_fmr);
 863         struct ehca_pd *e_pd = container_of(fmr->pd, struct ehca_pd, ib_pd);
 864         struct ehca_mr_pginfo pginfo;
 865         u32 tmp_lkey, tmp_rkey;
 866
 867         if (!(e_fmr->flags & EHCA_MR_FLAG_FMR)) {
 868                 ehca_err(fmr->device, "not a FMR, e_fmr=%p e_fmr->flags=%x",
 869                          e_fmr, e_fmr->flags);
 870                 ret = -EINVAL;
 871                 goto map_phys_fmr_exit0;
 872         }
 873         ret = ehca_fmr_check_page_list(e_fmr, page_list, list_len);
 874         if (ret)
 875                 goto map_phys_fmr_exit0;
 876         if (iova % e_fmr->fmr_page_size) {
 877                 /* only whole-numbered pages */
 878                 ehca_err(fmr->device, "bad iova, iova=%lx fmr_page_size=%x",
 879                          iova, e_fmr->fmr_page_size);
 880                 ret = -EINVAL;
 881                 goto map_phys_fmr_exit0;
 882         }
 883         if (e_fmr->fmr_map_cnt >= e_fmr->fmr_max_maps) {
 884                 /* HCAD does not limit the maps, however trace this anyway */
 885                 ehca_info(fmr->device, "map limit exceeded, fmr=%p "
 886                           "e_fmr->fmr_map_cnt=%x e_fmr->fmr_max_maps=%x",
 887                           fmr, e_fmr->fmr_map_cnt, e_fmr->fmr_max_maps);
 888         }
 889
 890         memset(&pginfo, 0, sizeof(pginfo));
 891         pginfo.type = EHCA_MR_PGI_FMR;
 892         pginfo.num_kpages = list_len;
 893         pginfo.hwpage_size = e_fmr->hwpage_size;
 894         pginfo.num_hwpages =
 895                 list_len * e_fmr->fmr_page_size / pginfo.hwpage_size;
 896         pginfo.u.fmr.page_list = page_list;
 897         pginfo.next_hwpage =
 898                 (iova & (e_fmr->fmr_page_size-1)) / pginfo.hwpage_size;
 899         pginfo.u.fmr.fmr_pgsize = e_fmr->fmr_page_size;
 900
 901         ret = ehca_rereg_mr(shca, e_fmr, (u64 *)iova,
 902                             list_len * e_fmr->fmr_page_size,
 903                             e_fmr->acl, e_pd, &pginfo, &tmp_lkey, &tmp_rkey);
 904         if (ret)
 905                 goto map_phys_fmr_exit0;
 906
 907         /* successful reregistration */
 908         e_fmr->fmr_map_cnt++;
 909         e_fmr->ib.ib_fmr.lkey = tmp_lkey;
 910         e_fmr->ib.ib_fmr.rkey = tmp_rkey;
 911         return 0;
 912
 913 map_phys_fmr_exit0:
 914         if (ret)
 915                 ehca_err(fmr->device, "ret=%x fmr=%p page_list=%p list_len=%x "
 916                          "iova=%lx", ret, fmr, page_list, list_len, iova);
 917         return ret;
 918 } /* end ehca_map_phys_fmr() */
 919
 920 /*----------------------------------------------------------------------*/
 921
 922 int ehca_unmap_fmr(struct list_head *fmr_list)
 923 {
 924         int ret = 0;
 925         struct ib_fmr *ib_fmr;
 926         struct ehca_shca *shca = NULL;
 927         struct ehca_shca *prev_shca;
 928         struct ehca_mr *e_fmr;
 929         u32 num_fmr = 0;
 930         u32 unmap_fmr_cnt = 0;
 931
 932         /* check all FMR belong to same SHCA, and check internal flag */
 933         list_for_each_entry(ib_fmr, fmr_list, list) {
 934                 prev_shca = shca;
 935                 if (!ib_fmr) {
 936                         ehca_gen_err("bad fmr=%p in list", ib_fmr);
 937                         ret = -EINVAL;
 938                         goto unmap_fmr_exit0;
 939                 }
 940                 shca = container_of(ib_fmr->device, struct ehca_shca,
 941                                     ib_device);
 942                 e_fmr = container_of(ib_fmr, struct ehca_mr, ib.ib_fmr);
 943                 if ((shca != prev_shca) && prev_shca) {
 944                         ehca_err(&shca->ib_device, "SHCA mismatch, shca=%p "
 945                                  "prev_shca=%p e_fmr=%p",
 946                                  shca, prev_shca, e_fmr);
 947                         ret = -EINVAL;
 948                         goto unmap_fmr_exit0;
 949                 }
 950                 if (!(e_fmr->flags & EHCA_MR_FLAG_FMR)) {
 951                         ehca_err(&shca->ib_device, "not a FMR, e_fmr=%p "
 952                                  "e_fmr->flags=%x", e_fmr, e_fmr->flags);
 953                         ret = -EINVAL;
 954                         goto unmap_fmr_exit0;
 955                 }
 956                 num_fmr++;
 957         }
 958
 959         /* loop over all FMRs to unmap */
 960         list_for_each_entry(ib_fmr, fmr_list, list) {
 961                 unmap_fmr_cnt++;
 962                 e_fmr = container_of(ib_fmr, struct ehca_mr, ib.ib_fmr);
 963                 shca = container_of(ib_fmr->device, struct ehca_shca,
 964                                     ib_device);
 965                 ret = ehca_unmap_one_fmr(shca, e_fmr);
 966                 if (ret) {
 967                         /* unmap failed, stop unmapping of rest of FMRs */
 968                         ehca_err(&shca->ib_device, "unmap of one FMR failed, "
 969                                  "stop rest, e_fmr=%p num_fmr=%x "
 970                                  "unmap_fmr_cnt=%x lkey=%x", e_fmr, num_fmr,
 971                                  unmap_fmr_cnt, e_fmr->ib.ib_fmr.lkey);
 972                         goto unmap_fmr_exit0;
 973                 }
 974         }
 975
 976 unmap_fmr_exit0:
 977         if (ret)
 978                 ehca_gen_err("ret=%x fmr_list=%p num_fmr=%x unmap_fmr_cnt=%x",
 979                              ret, fmr_list, num_fmr, unmap_fmr_cnt);
 980         return ret;
 981 } /* end ehca_unmap_fmr() */
 982
 983 /*----------------------------------------------------------------------*/
 984
 985 int ehca_dealloc_fmr(struct ib_fmr *fmr)
 986 {
 987         int ret;
 988         u64 h_ret;
 989         struct ehca_shca *shca =
 990                 container_of(fmr->device, struct ehca_shca, ib_device);
 991         struct ehca_mr *e_fmr = container_of(fmr, struct ehca_mr, ib.ib_fmr);
 992
 993         if (!(e_fmr->flags & EHCA_MR_FLAG_FMR)) {
 994                 ehca_err(fmr->device, "not a FMR, e_fmr=%p e_fmr->flags=%x",
 995                          e_fmr, e_fmr->flags);
 996                 ret = -EINVAL;
 997                 goto free_fmr_exit0;
 998         }
 999
1000         h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_fmr);
1001         if (h_ret != H_SUCCESS) {
1002                 ehca_err(fmr->device, "hipz_free_mr failed, h_ret=%lx e_fmr=%p "
1003                          "hca_hndl=%lx fmr_hndl=%lx fmr->lkey=%x",
1004                          h_ret, e_fmr, shca->ipz_hca_handle.handle,
1005                          e_fmr->ipz_mr_handle.handle, fmr->lkey);
1006                 ret = ehca2ib_return_code(h_ret);
1007                 goto free_fmr_exit0;
1008         }
1009         /* successful deregistration */
1010         ehca_mr_delete(e_fmr);
1011         return 0;
1012
1013 free_fmr_exit0:
1014         if (ret)
1015                 ehca_err(&shca->ib_device, "ret=%x fmr=%p", ret, fmr);
1016         return ret;
1017 } /* end ehca_dealloc_fmr() */
1018
1019 /*----------------------------------------------------------------------*/
1020
1021 int ehca_reg_mr(struct ehca_shca *shca,
1022                 struct ehca_mr *e_mr,
1023                 u64 *iova_start,
1024                 u64 size,
1025                 int acl,
1026                 struct ehca_pd *e_pd,
1027                 struct ehca_mr_pginfo *pginfo,
1028                 u32 *lkey, /*OUT*/
1029                 u32 *rkey) /*OUT*/
1030 {
1031         int ret;
1032         u64 h_ret;
1033         u32 hipz_acl;
1034         struct ehca_mr_hipzout_parms hipzout;
1035
1036         ehca_mrmw_map_acl(acl, &hipz_acl);
1037         ehca_mrmw_set_pgsize_hipz_acl(pginfo->hwpage_size, &hipz_acl);
1038         if (ehca_use_hp_mr == 1)
1039                 hipz_acl |= 0x00000001;
1040
1041         h_ret = hipz_h_alloc_resource_mr(shca->ipz_hca_handle, e_mr,
1042                                          (u64)iova_start, size, hipz_acl,
1043                                          e_pd->fw_pd, &hipzout);
1044         if (h_ret != H_SUCCESS) {
1045                 ehca_err(&shca->ib_device, "hipz_alloc_mr failed, h_ret=%lx "
1046                          "hca_hndl=%lx", h_ret, shca->ipz_hca_handle.handle);
1047                 ret = ehca2ib_return_code(h_ret);
1048                 goto ehca_reg_mr_exit0;
1049         }
1050
1051         e_mr->ipz_mr_handle = hipzout.handle;
1052
1053         ret = ehca_reg_mr_rpages(shca, e_mr, pginfo);
1054         if (ret)
1055                 goto ehca_reg_mr_exit1;
1056
1057         /* successful registration */
1058         e_mr->num_kpages = pginfo->num_kpages;
1059         e_mr->num_hwpages = pginfo->num_hwpages;
1060         e_mr->hwpage_size = pginfo->hwpage_size;
1061         e_mr->start = iova_start;
1062         e_mr->size = size;
1063         e_mr->acl = acl;
1064         *lkey = hipzout.lkey;
1065         *rkey = hipzout.rkey;
1066         return 0;
1067
1068 ehca_reg_mr_exit1:
1069         h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_mr);
1070         if (h_ret != H_SUCCESS) {
1071                 ehca_err(&shca->ib_device, "h_ret=%lx shca=%p e_mr=%p "
1072                          "iova_start=%p size=%lx acl=%x e_pd=%p lkey=%x "
1073                          "pginfo=%p num_kpages=%lx num_hwpages=%lx ret=%x",
1074                          h_ret, shca, e_mr, iova_start, size, acl, e_pd,
1075                          hipzout.lkey, pginfo, pginfo->num_kpages,
1076                          pginfo->num_hwpages, ret);
1077                 ehca_err(&shca->ib_device, "internal error in ehca_reg_mr, "
1078                          "not recoverable");
1079         }
1080 ehca_reg_mr_exit0:
1081         if (ret)
1082                 ehca_err(&shca->ib_device, "ret=%x shca=%p e_mr=%p "
1083                          "iova_start=%p size=%lx acl=%x e_pd=%p pginfo=%p "
1084                          "num_kpages=%lx num_hwpages=%lx",
1085                          ret, shca, e_mr, iova_start, size, acl, e_pd, pginfo,
1086                          pginfo->num_kpages, pginfo->num_hwpages);
1087         return ret;
1088 } /* end ehca_reg_mr() */
1089
1090 /*----------------------------------------------------------------------*/
1091
1092 int ehca_reg_mr_rpages(struct ehca_shca *shca,
1093                        struct ehca_mr *e_mr,
1094                        struct ehca_mr_pginfo *pginfo)
1095 {
1096         int ret = 0;
1097         u64 h_ret;
1098         u32 rnum;
1099         u64 rpage;
1100         u32 i;
1101         u64 *kpage;
1102
1103         if (!pginfo->num_hwpages) /* in case of fmr */
1104                 return 0;
1105
1106         kpage = ehca_alloc_fw_ctrlblock(GFP_KERNEL);
1107         if (!kpage) {
1108                 ehca_err(&shca->ib_device, "kpage alloc failed");
1109                 ret = -ENOMEM;
1110                 goto ehca_reg_mr_rpages_exit0;
1111         }
1112
1113         /* max MAX_RPAGES ehca mr pages per register call */
1114         for (i = 0; i < NUM_CHUNKS(pginfo->num_hwpages, MAX_RPAGES); i++) {
1115
1116                 if (i == NUM_CHUNKS(pginfo->num_hwpages, MAX_RPAGES) - 1) {
1117                         rnum = pginfo->num_hwpages % MAX_RPAGES; /* last shot */
1118                         if (rnum == 0)
1119                                 rnum = MAX_RPAGES;      /* last shot is full */
1120                 } else
1121                         rnum = MAX_RPAGES;
1122
1123                 ret = ehca_set_pagebuf(pginfo, rnum, kpage);
1124                 if (ret) {
1125                         ehca_err(&shca->ib_device, "ehca_set_pagebuf "
1126                                  "bad rc, ret=%x rnum=%x kpage=%p",
1127                                  ret, rnum, kpage);
1128                         goto ehca_reg_mr_rpages_exit1;
1129                 }
1130
1131                 if (rnum > 1) {
1132                         rpage = virt_to_abs(kpage);
1133                         if (!rpage) {
1134                                 ehca_err(&shca->ib_device, "kpage=%p i=%x",
1135                                          kpage, i);
1136                                 ret = -EFAULT;
1137                                 goto ehca_reg_mr_rpages_exit1;
1138                         }
1139                 } else
1140                         rpage = *kpage;
1141
1142                 h_ret = hipz_h_register_rpage_mr(
1143                         shca->ipz_hca_handle, e_mr,
1144                         ehca_encode_hwpage_size(pginfo->hwpage_size),
1145                         0, rpage, rnum);
1146
1147                 if (i == NUM_CHUNKS(pginfo->num_hwpages, MAX_RPAGES) - 1) {
1148                         /*
1149                          * check for 'registration complete'==H_SUCCESS
1150                          * and for 'page registered'==H_PAGE_REGISTERED
1151                          */
1152                         if (h_ret != H_SUCCESS) {
1153                                 ehca_err(&shca->ib_device, "last "
1154                                          "hipz_reg_rpage_mr failed, h_ret=%lx "
1155                                          "e_mr=%p i=%x hca_hndl=%lx mr_hndl=%lx"
1156                                          " lkey=%x", h_ret, e_mr, i,
1157                                          shca->ipz_hca_handle.handle,
1158                                          e_mr->ipz_mr_handle.handle,
1159                                          e_mr->ib.ib_mr.lkey);
1160                                 ret = ehca2ib_return_code(h_ret);
1161                                 break;
1162                         } else
1163                                 ret = 0;
1164                 } else if (h_ret != H_PAGE_REGISTERED) {
1165                         ehca_err(&shca->ib_device, "hipz_reg_rpage_mr failed, "
1166                                  "h_ret=%lx e_mr=%p i=%x lkey=%x hca_hndl=%lx "
1167                                  "mr_hndl=%lx", h_ret, e_mr, i,
1168                                  e_mr->ib.ib_mr.lkey,
1169                                  shca->ipz_hca_handle.handle,
1170                                  e_mr->ipz_mr_handle.handle);
1171                         ret = ehca2ib_return_code(h_ret);
1172                         break;
1173                 } else
1174                         ret = 0;
1175         } /* end for(i) */
1176
1177
1178 ehca_reg_mr_rpages_exit1:
1179         ehca_free_fw_ctrlblock(kpage);
1180 ehca_reg_mr_rpages_exit0:
1181         if (ret)
1182                 ehca_err(&shca->ib_device, "ret=%x shca=%p e_mr=%p pginfo=%p "
1183                          "num_kpages=%lx num_hwpages=%lx", ret, shca, e_mr,
1184                          pginfo, pginfo->num_kpages, pginfo->num_hwpages);
1185         return ret;
1186 } /* end ehca_reg_mr_rpages() */
1187
1188 /*----------------------------------------------------------------------*/
1189
1190 inline int ehca_rereg_mr_rereg1(struct ehca_shca *shca,
1191                                 struct ehca_mr *e_mr,
1192                                 u64 *iova_start,
1193                                 u64 size,
1194                                 u32 acl,
1195                                 struct ehca_pd *e_pd,
1196                                 struct ehca_mr_pginfo *pginfo,
1197                                 u32 *lkey, /*OUT*/
1198                                 u32 *rkey) /*OUT*/
1199 {
1200         int ret;
1201         u64 h_ret;
1202         u32 hipz_acl;
1203         u64 *kpage;
1204         u64 rpage;
1205         struct ehca_mr_pginfo pginfo_save;
1206         struct ehca_mr_hipzout_parms hipzout;
1207
1208         ehca_mrmw_map_acl(acl, &hipz_acl);
1209         ehca_mrmw_set_pgsize_hipz_acl(pginfo->hwpage_size, &hipz_acl);
1210
1211         kpage = ehca_alloc_fw_ctrlblock(GFP_KERNEL);
1212         if (!kpage) {
1213                 ehca_err(&shca->ib_device, "kpage alloc failed");
1214                 ret = -ENOMEM;
1215                 goto ehca_rereg_mr_rereg1_exit0;
1216         }
1217
1218         pginfo_save = *pginfo;
1219         ret = ehca_set_pagebuf(pginfo, pginfo->num_hwpages, kpage);
1220         if (ret) {
1221                 ehca_err(&shca->ib_device, "set pagebuf failed, e_mr=%p "
1222                          "pginfo=%p type=%x num_kpages=%lx num_hwpages=%lx "
1223                          "kpage=%p", e_mr, pginfo, pginfo->type,
1224                          pginfo->num_kpages, pginfo->num_hwpages, kpage);
1225                 goto ehca_rereg_mr_rereg1_exit1;
1226         }
1227         rpage = virt_to_abs(kpage);
1228         if (!rpage) {
1229                 ehca_err(&shca->ib_device, "kpage=%p", kpage);
1230                 ret = -EFAULT;
1231                 goto ehca_rereg_mr_rereg1_exit1;
1232         }
1233         h_ret = hipz_h_reregister_pmr(shca->ipz_hca_handle, e_mr,
1234                                       (u64)iova_start, size, hipz_acl,
1235                                       e_pd->fw_pd, rpage, &hipzout);
1236         if (h_ret != H_SUCCESS) {
1237                 /*
1238                  * reregistration unsuccessful, try it again with the 3 hCalls,
1239                  * e.g. this is required in case H_MR_CONDITION
1240                  * (MW bound or MR is shared)
1241                  */
1242                 ehca_warn(&shca->ib_device, "hipz_h_reregister_pmr failed "
1243                           "(Rereg1), h_ret=%lx e_mr=%p", h_ret, e_mr);
1244                 *pginfo = pginfo_save;
1245                 ret = -EAGAIN;
1246         } else if ((u64 *)hipzout.vaddr != iova_start) {
1247                 ehca_err(&shca->ib_device, "PHYP changed iova_start in "
1248                          "rereg_pmr, iova_start=%p iova_start_out=%lx e_mr=%p "
1249                          "mr_handle=%lx lkey=%x lkey_out=%x", iova_start,
1250                          hipzout.vaddr, e_mr, e_mr->ipz_mr_handle.handle,
1251                          e_mr->ib.ib_mr.lkey, hipzout.lkey);
1252                 ret = -EFAULT;
1253         } else {
1254                 /*
1255                  * successful reregistration
1256                  * note: start and start_out are identical for eServer HCAs
1257                  */
1258                 e_mr->num_kpages = pginfo->num_kpages;
1259                 e_mr->num_hwpages = pginfo->num_hwpages;
1260                 e_mr->hwpage_size = pginfo->hwpage_size;
1261                 e_mr->start = iova_start;
1262                 e_mr->size = size;
1263                 e_mr->acl = acl;
1264                 *lkey = hipzout.lkey;
1265                 *rkey = hipzout.rkey;
1266         }
1267
1268 ehca_rereg_mr_rereg1_exit1:
1269         ehca_free_fw_ctrlblock(kpage);
1270 ehca_rereg_mr_rereg1_exit0:
1271         if ( ret && (ret != -EAGAIN) )
1272                 ehca_err(&shca->ib_device, "ret=%x lkey=%x rkey=%x "
1273                          "pginfo=%p num_kpages=%lx num_hwpages=%lx",
1274                          ret, *lkey, *rkey, pginfo, pginfo->num_kpages,
1275                          pginfo->num_hwpages);
1276         return ret;
1277 } /* end ehca_rereg_mr_rereg1() */
1278
1279 /*----------------------------------------------------------------------*/
1280
1281 int ehca_rereg_mr(struct ehca_shca *shca,
1282                   struct ehca_mr *e_mr,
1283                   u64 *iova_start,
1284                   u64 size,
1285                   int acl,
1286                   struct ehca_pd *e_pd,
1287                   struct ehca_mr_pginfo *pginfo,
1288                   u32 *lkey,
1289                   u32 *rkey)
1290 {
1291         int ret = 0;
1292         u64 h_ret;
1293         int rereg_1_hcall = 1; /* 1: use hipz_h_reregister_pmr directly */
1294         int rereg_3_hcall = 0; /* 1: use 3 hipz calls for reregistration */
1295
1296         /* first determine reregistration hCall(s) */
1297         if ((pginfo->num_hwpages > MAX_RPAGES) ||
1298             (e_mr->num_hwpages > MAX_RPAGES) ||
1299             (pginfo->num_hwpages > e_mr->num_hwpages)) {
1300                 ehca_dbg(&shca->ib_device, "Rereg3 case, "
1301                          "pginfo->num_hwpages=%lx e_mr->num_hwpages=%x",
1302                          pginfo->num_hwpages, e_mr->num_hwpages);
1303                 rereg_1_hcall = 0;
1304                 rereg_3_hcall = 1;
1305         }
1306
1307         if (e_mr->flags & EHCA_MR_FLAG_MAXMR) { /* check for max-MR */
1308                 rereg_1_hcall = 0;
1309                 rereg_3_hcall = 1;
1310                 e_mr->flags &= ~EHCA_MR_FLAG_MAXMR;
1311                 ehca_err(&shca->ib_device, "Rereg MR for max-MR! e_mr=%p",
1312                          e_mr);
1313         }
1314
1315         if (rereg_1_hcall) {
1316                 ret = ehca_rereg_mr_rereg1(shca, e_mr, iova_start, size,
1317                                            acl, e_pd, pginfo, lkey, rkey);
1318                 if (ret) {
1319                         if (ret == -EAGAIN)
1320                                 rereg_3_hcall = 1;
1321                         else
1322                                 goto ehca_rereg_mr_exit0;
1323                 }
1324         }
1325
1326         if (rereg_3_hcall) {
1327                 struct ehca_mr save_mr;
1328
1329                 /* first deregister old MR */
1330                 h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_mr);
1331                 if (h_ret != H_SUCCESS) {
1332                         ehca_err(&shca->ib_device, "hipz_free_mr failed, "
1333                                  "h_ret=%lx e_mr=%p hca_hndl=%lx mr_hndl=%lx "
1334                                  "mr->lkey=%x",
1335                                  h_ret, e_mr, shca->ipz_hca_handle.handle,
1336                                  e_mr->ipz_mr_handle.handle,
1337                                  e_mr->ib.ib_mr.lkey);
1338                         ret = ehca2ib_return_code(h_ret);
1339                         goto ehca_rereg_mr_exit0;
1340                 }
1341                 /* clean ehca_mr_t, without changing struct ib_mr and lock */
1342                 save_mr = *e_mr;
1343                 ehca_mr_deletenew(e_mr);
1344
1345                 /* set some MR values */
1346                 e_mr->flags = save_mr.flags;
1347                 e_mr->hwpage_size = save_mr.hwpage_size;
1348                 e_mr->fmr_page_size = save_mr.fmr_page_size;
1349                 e_mr->fmr_max_pages = save_mr.fmr_max_pages;
1350                 e_mr->fmr_max_maps = save_mr.fmr_max_maps;
1351                 e_mr->fmr_map_cnt = save_mr.fmr_map_cnt;
1352
1353                 ret = ehca_reg_mr(shca, e_mr, iova_start, size, acl,
1354                                   e_pd, pginfo, lkey, rkey);
1355                 if (ret) {
1356                         u32 offset = (u64)(&e_mr->flags) - (u64)e_mr;
1357                         memcpy(&e_mr->flags, &(save_mr.flags),
1358                                sizeof(struct ehca_mr) - offset);
1359                         goto ehca_rereg_mr_exit0;
1360                 }
1361         }
1362
1363 ehca_rereg_mr_exit0:
1364         if (ret)
1365                 ehca_err(&shca->ib_device, "ret=%x shca=%p e_mr=%p "
1366                          "iova_start=%p size=%lx acl=%x e_pd=%p pginfo=%p "
1367                          "num_kpages=%lx lkey=%x rkey=%x rereg_1_hcall=%x "
1368                          "rereg_3_hcall=%x", ret, shca, e_mr, iova_start, size,
1369                          acl, e_pd, pginfo, pginfo->num_kpages, *lkey, *rkey,
1370                          rereg_1_hcall, rereg_3_hcall);
1371         return ret;
1372 } /* end ehca_rereg_mr() */
1373
1374 /*----------------------------------------------------------------------*/
1375
1376 int ehca_unmap_one_fmr(struct ehca_shca *shca,
1377                        struct ehca_mr *e_fmr)
1378 {
1379         int ret = 0;
1380         u64 h_ret;
1381         struct ehca_pd *e_pd =
1382                 container_of(e_fmr->ib.ib_fmr.pd, struct ehca_pd, ib_pd);
1383         struct ehca_mr save_fmr;
1384         u32 tmp_lkey, tmp_rkey;
1385         struct ehca_mr_pginfo pginfo;
1386         struct ehca_mr_hipzout_parms hipzout;
1387         struct ehca_mr save_mr;
1388
1389         if (e_fmr->fmr_max_pages <= MAX_RPAGES) {
1390                 /*
1391                  * note: after using rereg hcall with len=0,
1392                  * rereg hcall must be used again for registering pages
1393                  */
1394                 h_ret = hipz_h_reregister_pmr(shca->ipz_hca_handle, e_fmr, 0,
1395                                               0, 0, e_pd->fw_pd, 0, &hipzout);
1396                 if (h_ret == H_SUCCESS) {
1397                         /* successful reregistration */
1398                         e_fmr->start = NULL;
1399                         e_fmr->size = 0;
1400                         tmp_lkey = hipzout.lkey;
1401                         tmp_rkey = hipzout.rkey;
1402                         return 0;
1403                 }
1404                 /*
1405                  * should not happen, because length checked above,
1406                  * FMRs are not shared and no MW bound to FMRs
1407                  */
1408                 ehca_err(&shca->ib_device, "hipz_reregister_pmr failed "
1409                          "(Rereg1), h_ret=%lx e_fmr=%p hca_hndl=%lx "
1410                          "mr_hndl=%lx lkey=%x lkey_out=%x",
1411                          h_ret, e_fmr, shca->ipz_hca_handle.handle,
1412                          e_fmr->ipz_mr_handle.handle,
1413                          e_fmr->ib.ib_fmr.lkey, hipzout.lkey);
1414                 /* try free and rereg */
1415         }
1416
1417         /* first free old FMR */
1418         h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_fmr);
1419         if (h_ret != H_SUCCESS) {
1420                 ehca_err(&shca->ib_device, "hipz_free_mr failed, "
1421                          "h_ret=%lx e_fmr=%p hca_hndl=%lx mr_hndl=%lx "
1422                          "lkey=%x",
1423                          h_ret, e_fmr, shca->ipz_hca_handle.handle,
1424                          e_fmr->ipz_mr_handle.handle,
1425                          e_fmr->ib.ib_fmr.lkey);
1426                 ret = ehca2ib_return_code(h_ret);
1427                 goto ehca_unmap_one_fmr_exit0;
1428         }
1429         /* clean ehca_mr_t, without changing lock */
1430         save_fmr = *e_fmr;
1431         ehca_mr_deletenew(e_fmr);
1432
1433         /* set some MR values */
1434         e_fmr->flags = save_fmr.flags;
1435         e_fmr->hwpage_size = save_fmr.hwpage_size;
1436         e_fmr->fmr_page_size = save_fmr.fmr_page_size;
1437         e_fmr->fmr_max_pages = save_fmr.fmr_max_pages;
1438         e_fmr->fmr_max_maps = save_fmr.fmr_max_maps;
1439         e_fmr->fmr_map_cnt = save_fmr.fmr_map_cnt;
1440         e_fmr->acl = save_fmr.acl;
1441
1442         memset(&pginfo, 0, sizeof(pginfo));
1443         pginfo.type = EHCA_MR_PGI_FMR;
1444         ret = ehca_reg_mr(shca, e_fmr, NULL,
1445                           (e_fmr->fmr_max_pages * e_fmr->fmr_page_size),
1446                           e_fmr->acl, e_pd, &pginfo, &tmp_lkey,
1447                           &tmp_rkey);
1448         if (ret) {
1449                 u32 offset = (u64)(&e_fmr->flags) - (u64)e_fmr;
1450                 memcpy(&e_fmr->flags, &(save_mr.flags),
1451                        sizeof(struct ehca_mr) - offset);
1452         }
1453
1454 ehca_unmap_one_fmr_exit0:
1455         if (ret)
1456                 ehca_err(&shca->ib_device, "ret=%x tmp_lkey=%x tmp_rkey=%x "
1457                          "fmr_max_pages=%x",
1458                          ret, tmp_lkey, tmp_rkey, e_fmr->fmr_max_pages);
1459         return ret;
1460 } /* end ehca_unmap_one_fmr() */
1461
1462 /*----------------------------------------------------------------------*/
1463
1464 int ehca_reg_smr(struct ehca_shca *shca,
1465                  struct ehca_mr *e_origmr,
1466                  struct ehca_mr *e_newmr,
1467                  u64 *iova_start,
1468                  int acl,
1469                  struct ehca_pd *e_pd,
1470                  u32 *lkey, /*OUT*/
1471                  u32 *rkey) /*OUT*/
1472 {
1473         int ret = 0;
1474         u64 h_ret;
1475         u32 hipz_acl;
1476         struct ehca_mr_hipzout_parms hipzout;
1477
1478         ehca_mrmw_map_acl(acl, &hipz_acl);
1479         ehca_mrmw_set_pgsize_hipz_acl(e_origmr->hwpage_size, &hipz_acl);
1480
1481         h_ret = hipz_h_register_smr(shca->ipz_hca_handle, e_newmr, e_origmr,
1482                                     (u64)iova_start, hipz_acl, e_pd->fw_pd,
1483                                     &hipzout);
1484         if (h_ret != H_SUCCESS) {
1485                 ehca_err(&shca->ib_device, "hipz_reg_smr failed, h_ret=%lx "
1486                          "shca=%p e_origmr=%p e_newmr=%p iova_start=%p acl=%x "
1487                          "e_pd=%p hca_hndl=%lx mr_hndl=%lx lkey=%x",
1488                          h_ret, shca, e_origmr, e_newmr, iova_start, acl, e_pd,
1489                          shca->ipz_hca_handle.handle,
1490                          e_origmr->ipz_mr_handle.handle,
1491                          e_origmr->ib.ib_mr.lkey);
1492                 ret = ehca2ib_return_code(h_ret);
1493                 goto ehca_reg_smr_exit0;
1494         }
1495         /* successful registration */
1496         e_newmr->num_kpages = e_origmr->num_kpages;
1497         e_newmr->num_hwpages = e_origmr->num_hwpages;
1498         e_newmr->hwpage_size   = e_origmr->hwpage_size;
1499         e_newmr->start = iova_start;
1500         e_newmr->size = e_origmr->size;
1501         e_newmr->acl = acl;
1502         e_newmr->ipz_mr_handle = hipzout.handle;
1503         *lkey = hipzout.lkey;
1504         *rkey = hipzout.rkey;
1505         return 0;
1506
1507 ehca_reg_smr_exit0:
1508         if (ret)
1509                 ehca_err(&shca->ib_device, "ret=%x shca=%p e_origmr=%p "
1510                          "e_newmr=%p iova_start=%p acl=%x e_pd=%p",
1511                          ret, shca, e_origmr, e_newmr, iova_start, acl, e_pd);
1512         return ret;
1513 } /* end ehca_reg_smr() */
1514
1515 /*----------------------------------------------------------------------*/
1516
1517 /* register internal max-MR to internal SHCA */
1518 int ehca_reg_internal_maxmr(
1519         struct ehca_shca *shca,
1520         struct ehca_pd *e_pd,
1521         struct ehca_mr **e_maxmr)  /*OUT*/
1522 {
1523         int ret;
1524         struct ehca_mr *e_mr;
1525         u64 *iova_start;
1526         u64 size_maxmr;
1527         struct ehca_mr_pginfo pginfo;
1528         struct ib_phys_buf ib_pbuf;
1529         u32 num_kpages;
1530         u32 num_hwpages;
1531         u64 hw_pgsize;
1532
1533         e_mr = ehca_mr_new();
1534         if (!e_mr) {
1535                 ehca_err(&shca->ib_device, "out of memory");
1536                 ret = -ENOMEM;
1537                 goto ehca_reg_internal_maxmr_exit0;
1538         }
1539         e_mr->flags |= EHCA_MR_FLAG_MAXMR;
1540
1541         /* register internal max-MR on HCA */
1542         size_maxmr = (u64)high_memory - PAGE_OFFSET;
1543         iova_start = (u64 *)KERNELBASE;
1544         ib_pbuf.addr = 0;
1545         ib_pbuf.size = size_maxmr;
1546         num_kpages = NUM_CHUNKS(((u64)iova_start % PAGE_SIZE) + size_maxmr,
1547                                 PAGE_SIZE);
1548         hw_pgsize = ehca_get_max_hwpage_size(shca);
1549         num_hwpages = NUM_CHUNKS(((u64)iova_start % hw_pgsize) + size_maxmr,
1550                                  hw_pgsize);
1551
1552         memset(&pginfo, 0, sizeof(pginfo));
1553         pginfo.type = EHCA_MR_PGI_PHYS;
1554         pginfo.num_kpages = num_kpages;
1555         pginfo.num_hwpages = num_hwpages;
1556         pginfo.hwpage_size = hw_pgsize;
1557         pginfo.u.phy.num_phys_buf = 1;
1558         pginfo.u.phy.phys_buf_array = &ib_pbuf;
1559
1560         ret = ehca_reg_mr(shca, e_mr, iova_start, size_maxmr, 0, e_pd,
1561                           &pginfo, &e_mr->ib.ib_mr.lkey,
1562                           &e_mr->ib.ib_mr.rkey);
1563         if (ret) {
1564                 ehca_err(&shca->ib_device, "reg of internal max MR failed, "
1565                          "e_mr=%p iova_start=%p size_maxmr=%lx num_kpages=%x "
1566                          "num_hwpages=%x", e_mr, iova_start, size_maxmr,
1567                          num_kpages, num_hwpages);
1568                 goto ehca_reg_internal_maxmr_exit1;
1569         }
1570
1571         /* successful registration of all pages */
1572         e_mr->ib.ib_mr.device = e_pd->ib_pd.device;
1573         e_mr->ib.ib_mr.pd = &e_pd->ib_pd;
1574         e_mr->ib.ib_mr.uobject = NULL;
1575         atomic_inc(&(e_pd->ib_pd.usecnt));
1576         atomic_set(&(e_mr->ib.ib_mr.usecnt), 0);
1577         *e_maxmr = e_mr;
1578         return 0;
1579
1580 ehca_reg_internal_maxmr_exit1:
1581         ehca_mr_delete(e_mr);
1582 ehca_reg_internal_maxmr_exit0:
1583         if (ret)
1584                 ehca_err(&shca->ib_device, "ret=%x shca=%p e_pd=%p e_maxmr=%p",
1585                          ret, shca, e_pd, e_maxmr);
1586         return ret;
1587 } /* end ehca_reg_internal_maxmr() */
1588
1589 /*----------------------------------------------------------------------*/
1590
1591 int ehca_reg_maxmr(struct ehca_shca *shca,
1592                    struct ehca_mr *e_newmr,
1593                    u64 *iova_start,
1594                    int acl,
1595                    struct ehca_pd *e_pd,
1596                    u32 *lkey,
1597                    u32 *rkey)
1598 {
1599         u64 h_ret;
1600         struct ehca_mr *e_origmr = shca->maxmr;
1601         u32 hipz_acl;
1602         struct ehca_mr_hipzout_parms hipzout;
1603
1604         ehca_mrmw_map_acl(acl, &hipz_acl);
1605         ehca_mrmw_set_pgsize_hipz_acl(e_origmr->hwpage_size, &hipz_acl);
1606
1607         h_ret = hipz_h_register_smr(shca->ipz_hca_handle, e_newmr, e_origmr,
1608                                     (u64)iova_start, hipz_acl, e_pd->fw_pd,
1609                                     &hipzout);
1610         if (h_ret != H_SUCCESS) {
1611                 ehca_err(&shca->ib_device, "hipz_reg_smr failed, h_ret=%lx "
1612                          "e_origmr=%p hca_hndl=%lx mr_hndl=%lx lkey=%x",
1613                          h_ret, e_origmr, shca->ipz_hca_handle.handle,
1614                          e_origmr->ipz_mr_handle.handle,
1615                          e_origmr->ib.ib_mr.lkey);
1616                 return ehca2ib_return_code(h_ret);
1617         }
1618         /* successful registration */
1619         e_newmr->num_kpages = e_origmr->num_kpages;
1620         e_newmr->num_hwpages = e_origmr->num_hwpages;
1621         e_newmr->hwpage_size = e_origmr->hwpage_size;
1622         e_newmr->start = iova_start;
1623         e_newmr->size = e_origmr->size;
1624         e_newmr->acl = acl;
1625         e_newmr->ipz_mr_handle = hipzout.handle;
1626         *lkey = hipzout.lkey;
1627         *rkey = hipzout.rkey;
1628         return 0;
1629 } /* end ehca_reg_maxmr() */
1630
1631 /*----------------------------------------------------------------------*/
1632
1633 int ehca_dereg_internal_maxmr(struct ehca_shca *shca)
1634 {
1635         int ret;
1636         struct ehca_mr *e_maxmr;
1637         struct ib_pd *ib_pd;
1638
1639         if (!shca->maxmr) {
1640                 ehca_err(&shca->ib_device, "bad call, shca=%p", shca);
1641                 ret = -EINVAL;
1642                 goto ehca_dereg_internal_maxmr_exit0;
1643         }
1644
1645         e_maxmr = shca->maxmr;
1646         ib_pd = e_maxmr->ib.ib_mr.pd;
1647         shca->maxmr = NULL; /* remove internal max-MR indication from SHCA */
1648
1649         ret = ehca_dereg_mr(&e_maxmr->ib.ib_mr);
1650         if (ret) {
1651                 ehca_err(&shca->ib_device, "dereg internal max-MR failed, "
1652                          "ret=%x e_maxmr=%p shca=%p lkey=%x",
1653                          ret, e_maxmr, shca, e_maxmr->ib.ib_mr.lkey);
1654                 shca->maxmr = e_maxmr;
1655                 goto ehca_dereg_internal_maxmr_exit0;
1656         }
1657
1658         atomic_dec(&ib_pd->usecnt);
1659
1660 ehca_dereg_internal_maxmr_exit0:
1661         if (ret)
1662                 ehca_err(&shca->ib_device, "ret=%x shca=%p shca->maxmr=%p",
1663                          ret, shca, shca->maxmr);
1664         return ret;
1665 } /* end ehca_dereg_internal_maxmr() */
1666
1667 /*----------------------------------------------------------------------*/
1668
1669 /*
1670  * check physical buffer array of MR verbs for validness and
1671  * calculates MR size
1672  */
1673 int ehca_mr_chk_buf_and_calc_size(struct ib_phys_buf *phys_buf_array,
1674                                   int num_phys_buf,
1675                                   u64 *iova_start,
1676                                   u64 *size)
1677 {
1678         struct ib_phys_buf *pbuf = phys_buf_array;
1679         u64 size_count = 0;
1680         u32 i;
1681
1682         if (num_phys_buf == 0) {
1683                 ehca_gen_err("bad phys buf array len, num_phys_buf=0");
1684                 return -EINVAL;
1685         }
1686         /* check first buffer */
1687         if (((u64)iova_start & ~PAGE_MASK) != (pbuf->addr & ~PAGE_MASK)) {
1688                 ehca_gen_err("iova_start/addr mismatch, iova_start=%p "
1689                              "pbuf->addr=%lx pbuf->size=%lx",
1690                              iova_start, pbuf->addr, pbuf->size);
1691                 return -EINVAL;
1692         }
1693         if (((pbuf->addr + pbuf->size) % PAGE_SIZE) &&
1694             (num_phys_buf > 1)) {
1695                 ehca_gen_err("addr/size mismatch in 1st buf, pbuf->addr=%lx "
1696                              "pbuf->size=%lx", pbuf->addr, pbuf->size);
1697                 return -EINVAL;
1698         }
1699
1700         for (i = 0; i < num_phys_buf; i++) {
1701                 if ((i > 0) && (pbuf->addr % PAGE_SIZE)) {
1702                         ehca_gen_err("bad address, i=%x pbuf->addr=%lx "
1703                                      "pbuf->size=%lx",
1704                                      i, pbuf->addr, pbuf->size);
1705                         return -EINVAL;
1706                 }
1707                 if (((i > 0) && /* not 1st */
1708                      (i < (num_phys_buf - 1)) &&        /* not last */
1709                      (pbuf->size % PAGE_SIZE)) || (pbuf->size == 0)) {
1710                         ehca_gen_err("bad size, i=%x pbuf->size=%lx",
1711                                      i, pbuf->size);
1712                         return -EINVAL;
1713                 }
1714                 size_count += pbuf->size;
1715                 pbuf++;
1716         }
1717
1718         *size = size_count;
1719         return 0;
1720 } /* end ehca_mr_chk_buf_and_calc_size() */
1721
1722 /*----------------------------------------------------------------------*/
1723
1724 /* check page list of map FMR verb for validness */
1725 int ehca_fmr_check_page_list(struct ehca_mr *e_fmr,
1726                              u64 *page_list,
1727                              int list_len)
1728 {
1729         u32 i;
1730         u64 *page;
1731
1732         if ((list_len == 0) || (list_len > e_fmr->fmr_max_pages)) {
1733                 ehca_gen_err("bad list_len, list_len=%x "
1734                              "e_fmr->fmr_max_pages=%x fmr=%p",
1735                              list_len, e_fmr->fmr_max_pages, e_fmr);
1736                 return -EINVAL;
1737         }
1738
1739         /* each page must be aligned */
1740         page = page_list;
1741         for (i = 0; i < list_len; i++) {
1742                 if (*page % e_fmr->fmr_page_size) {
1743                         ehca_gen_err("bad page, i=%x *page=%lx page=%p fmr=%p "
1744                                      "fmr_page_size=%x", i, *page, page, e_fmr,
1745                                      e_fmr->fmr_page_size);
1746                         return -EINVAL;
1747                 }
1748                 page++;
1749         }
1750
1751         return 0;
1752 } /* end ehca_fmr_check_page_list() */
1753
1754 /*----------------------------------------------------------------------*/
1755
1756 /* PAGE_SIZE >= pginfo->hwpage_size */
1757 static int ehca_set_pagebuf_user1(struct ehca_mr_pginfo *pginfo,
1758                                   u32 number,
1759                                   u64 *kpage)
1760 {
1761         int ret = 0;
1762         struct ib_umem_chunk *prev_chunk;
1763         struct ib_umem_chunk *chunk;
1764         u64 pgaddr;
1765         u32 i = 0;
1766         u32 j = 0;
1767         int hwpages_per_kpage = PAGE_SIZE / pginfo->hwpage_size;
1768
1769         /* loop over desired chunk entries */
1770         chunk      = pginfo->u.usr.next_chunk;
1771         prev_chunk = pginfo->u.usr.next_chunk;
1772         list_for_each_entry_continue(
1773                 chunk, (&(pginfo->u.usr.region->chunk_list)), list) {
1774                 for (i = pginfo->u.usr.next_nmap; i < chunk->nmap; ) {
1775                         pgaddr = page_to_pfn(chunk->page_list[i].page)
1776                                 << PAGE_SHIFT ;
1777                         *kpage = phys_to_abs(pgaddr +
1778                                              (pginfo->next_hwpage *
1779                                               pginfo->hwpage_size));
1780                         if ( !(*kpage) ) {
1781                                 ehca_gen_err("pgaddr=%lx "
1782                                              "chunk->page_list[i]=%lx "
1783                                              "i=%x next_hwpage=%lx",
1784                                              pgaddr, (u64)sg_dma_address(
1785                                                      &chunk->page_list[i]),
1786                                              i, pginfo->next_hwpage);
1787                                 return -EFAULT;
1788                         }
1789                         (pginfo->hwpage_cnt)++;
1790                         (pginfo->next_hwpage)++;
1791                         kpage++;
1792                         if (pginfo->next_hwpage % hwpages_per_kpage == 0) {
1793                                 (pginfo->kpage_cnt)++;
1794                                 (pginfo->u.usr.next_nmap)++;
1795                                 pginfo->next_hwpage = 0;
1796                                 i++;
1797                         }
1798                         j++;
1799                         if (j >= number) break;
1800                 }
1801                 if ((pginfo->u.usr.next_nmap >= chunk->nmap) &&
1802                     (j >= number)) {
1803                         pginfo->u.usr.next_nmap = 0;
1804                         prev_chunk = chunk;
1805                         break;
1806                 } else if (pginfo->u.usr.next_nmap >= chunk->nmap) {
1807                         pginfo->u.usr.next_nmap = 0;
1808                         prev_chunk = chunk;
1809                 } else if (j >= number)
1810                         break;
1811                 else
1812                         prev_chunk = chunk;
1813         }
1814         pginfo->u.usr.next_chunk =
1815                 list_prepare_entry(prev_chunk,
1816                                    (&(pginfo->u.usr.region->chunk_list)),
1817                                    list);
1818         return ret;
1819 }
1820
1821 /*
1822  * check given pages for contiguous layout
1823  * last page addr is returned in prev_pgaddr for further check
1824  */
1825 static int ehca_check_kpages_per_ate(struct scatterlist *page_list,
1826                                      int start_idx, int end_idx,
1827                                      u64 *prev_pgaddr)
1828 {
1829         int t;
1830         for (t = start_idx; t <= end_idx; t++) {
1831                 u64 pgaddr = page_to_pfn(page_list[t].page) << PAGE_SHIFT;
1832                 ehca_gen_dbg("chunk_page=%lx value=%016lx", pgaddr,
1833                              *(u64 *)abs_to_virt(phys_to_abs(pgaddr)));
1834                 if (pgaddr - PAGE_SIZE != *prev_pgaddr) {
1835                         ehca_gen_err("uncontiguous page found pgaddr=%lx "
1836                                      "prev_pgaddr=%lx page_list_i=%x",
1837                                      pgaddr, *prev_pgaddr, t);
1838                         return -EINVAL;
1839                 }
1840                 *prev_pgaddr = pgaddr;
1841         }
1842         return 0;
1843 }
1844
1845 /* PAGE_SIZE < pginfo->hwpage_size */
1846 static int ehca_set_pagebuf_user2(struct ehca_mr_pginfo *pginfo,
1847                                   u32 number,
1848                                   u64 *kpage)
1849 {
1850         int ret = 0;
1851         struct ib_umem_chunk *prev_chunk;
1852         struct ib_umem_chunk *chunk;
1853         u64 pgaddr, prev_pgaddr;
1854         u32 i = 0;
1855         u32 j = 0;
1856         int kpages_per_hwpage = pginfo->hwpage_size / PAGE_SIZE;
1857         int nr_kpages = kpages_per_hwpage;
1858
1859         /* loop over desired chunk entries */
1860         chunk      = pginfo->u.usr.next_chunk;
1861         prev_chunk = pginfo->u.usr.next_chunk;
1862         list_for_each_entry_continue(
1863                 chunk, (&(pginfo->u.usr.region->chunk_list)), list) {
1864                 for (i = pginfo->u.usr.next_nmap; i < chunk->nmap; ) {
1865                         if (nr_kpages == kpages_per_hwpage) {
1866                                 pgaddr = ( page_to_pfn(chunk->page_list[i].page)
1867                                            << PAGE_SHIFT );
1868                                 *kpage = phys_to_abs(pgaddr);
1869                                 if ( !(*kpage) ) {
1870                                         ehca_gen_err("pgaddr=%lx i=%x",
1871                                                      pgaddr, i);
1872                                         ret = -EFAULT;
1873                                         return ret;
1874                                 }
1875                                 /*
1876                                  * The first page in a hwpage must be aligned;
1877                                  * the first MR page is exempt from this rule.
1878                                  */
1879                                 if (pgaddr & (pginfo->hwpage_size - 1)) {
1880                                         if (pginfo->hwpage_cnt) {
1881                                                 ehca_gen_err(
1882                                                         "invalid alignment "
1883                                                         "pgaddr=%lx i=%x "
1884                                                         "mr_pgsize=%lx",
1885                                                         pgaddr, i,
1886                                                         pginfo->hwpage_size);
1887                                                 ret = -EFAULT;
1888                                                 return ret;
1889                                         }
1890                                         /* first MR page */
1891                                         pginfo->kpage_cnt =
1892                                                 (pgaddr &
1893                                                  (pginfo->hwpage_size - 1)) >>
1894                                                 PAGE_SHIFT;
1895                                         nr_kpages -= pginfo->kpage_cnt;
1896                                         *kpage = phys_to_abs(
1897                                                 pgaddr &
1898                                                 ~(pginfo->hwpage_size - 1));
1899                                 }
1900                                 ehca_gen_dbg("kpage=%lx chunk_page=%lx "
1901                                              "value=%016lx", *kpage, pgaddr,
1902                                              *(u64 *)abs_to_virt(
1903                                                      phys_to_abs(pgaddr)));
1904                                 prev_pgaddr = pgaddr;
1905                                 i++;
1906                                 pginfo->kpage_cnt++;
1907                                 pginfo->u.usr.next_nmap++;
1908                                 nr_kpages--;
1909                                 if (!nr_kpages)
1910                                         goto next_kpage;
1911                                 continue;
1912                         }
1913                         if (i + nr_kpages > chunk->nmap) {
1914                                 ret = ehca_check_kpages_per_ate(
1915                                         chunk->page_list, i,
1916                                         chunk->nmap - 1, &prev_pgaddr);
1917                                 if (ret) return ret;
1918                                 pginfo->kpage_cnt += chunk->nmap - i;
1919                                 pginfo->u.usr.next_nmap += chunk->nmap - i;
1920                                 nr_kpages -= chunk->nmap - i;
1921                                 break;
1922                         }
1923
1924                         ret = ehca_check_kpages_per_ate(chunk->page_list, i,
1925                                                         i + nr_kpages - 1,
1926                                                         &prev_pgaddr);
1927                         if (ret) return ret;
1928                         i += nr_kpages;
1929                         pginfo->kpage_cnt += nr_kpages;
1930                         pginfo->u.usr.next_nmap += nr_kpages;
1931 next_kpage:
1932                         nr_kpages = kpages_per_hwpage;
1933                         (pginfo->hwpage_cnt)++;
1934                         kpage++;
1935                         j++;
1936                         if (j >= number) break;
1937                 }
1938                 if ((pginfo->u.usr.next_nmap >= chunk->nmap) &&
1939                     (j >= number)) {
1940                         pginfo->u.usr.next_nmap = 0;
1941                         prev_chunk = chunk;
1942                         break;
1943                 } else if (pginfo->u.usr.next_nmap >= chunk->nmap) {
1944                         pginfo->u.usr.next_nmap = 0;
1945                         prev_chunk = chunk;
1946                 } else if (j >= number)
1947                         break;
1948                 else
1949                         prev_chunk = chunk;
1950         }
1951         pginfo->u.usr.next_chunk =
1952                 list_prepare_entry(prev_chunk,
1953                                    (&(pginfo->u.usr.region->chunk_list)),
1954                                    list);
1955         return ret;
1956 }
1957
1958 int ehca_set_pagebuf_phys(struct ehca_mr_pginfo *pginfo,
1959                           u32 number,
1960                           u64 *kpage)
1961 {
1962         int ret = 0;
1963         struct ib_phys_buf *pbuf;
1964         u64 num_hw, offs_hw;
1965         u32 i = 0;
1966
1967         /* loop over desired phys_buf_array entries */
1968         while (i < number) {
1969                 pbuf   = pginfo->u.phy.phys_buf_array + pginfo->u.phy.next_buf;
1970                 num_hw  = NUM_CHUNKS((pbuf->addr % pginfo->hwpage_size) +
1971                                      pbuf->size, pginfo->hwpage_size);
1972                 offs_hw = (pbuf->addr & ~(pginfo->hwpage_size - 1)) /
1973                         pginfo->hwpage_size;
1974                 while (pginfo->next_hwpage < offs_hw + num_hw) {
1975                         /* sanity check */
1976                         if ((pginfo->kpage_cnt >= pginfo->num_kpages) ||
1977                             (pginfo->hwpage_cnt >= pginfo->num_hwpages)) {
1978                                 ehca_gen_err("kpage_cnt >= num_kpages, "
1979                                              "kpage_cnt=%lx num_kpages=%lx "
1980                                              "hwpage_cnt=%lx "
1981                                              "num_hwpages=%lx i=%x",
1982                                              pginfo->kpage_cnt,
1983                                              pginfo->num_kpages,
1984                                              pginfo->hwpage_cnt,
1985                                              pginfo->num_hwpages, i);
1986                                 return -EFAULT;
1987                         }
1988                         *kpage = phys_to_abs(
1989                                 (pbuf->addr & ~(pginfo->hwpage_size - 1)) +
1990                                 (pginfo->next_hwpage * pginfo->hwpage_size));
1991                         if ( !(*kpage) && pbuf->addr ) {
1992                                 ehca_gen_err("pbuf->addr=%lx pbuf->size=%lx "
1993                                              "next_hwpage=%lx", pbuf->addr,
1994                                              pbuf->size, pginfo->next_hwpage);
1995                                 return -EFAULT;
1996                         }
1997                         (pginfo->hwpage_cnt)++;
1998                         (pginfo->next_hwpage)++;
1999                         if (PAGE_SIZE >= pginfo->hwpage_size) {
2000                                 if (pginfo->next_hwpage %
2001                                     (PAGE_SIZE / pginfo->hwpage_size) == 0)
2002                                         (pginfo->kpage_cnt)++;
2003                         } else
2004                                 pginfo->kpage_cnt += pginfo->hwpage_size /
2005                                         PAGE_SIZE;
2006                         kpage++;
2007                         i++;
2008                         if (i >= number) break;
2009                 }
2010                 if (pginfo->next_hwpage >= offs_hw + num_hw) {
2011                         (pginfo->u.phy.next_buf)++;
2012                         pginfo->next_hwpage = 0;
2013                 }
2014         }
2015         return ret;
2016 }
2017
2018 int ehca_set_pagebuf_fmr(struct ehca_mr_pginfo *pginfo,
2019                          u32 number,
2020                          u64 *kpage)
2021 {
2022         int ret = 0;
2023         u64 *fmrlist;
2024         u32 i;
2025
2026         /* loop over desired page_list entries */
2027         fmrlist = pginfo->u.fmr.page_list + pginfo->u.fmr.next_listelem;
2028         for (i = 0; i < number; i++) {
2029                 *kpage = phys_to_abs((*fmrlist & ~(pginfo->hwpage_size - 1)) +
2030                                      pginfo->next_hwpage * pginfo->hwpage_size);
2031                 if ( !(*kpage) ) {
2032                         ehca_gen_err("*fmrlist=%lx fmrlist=%p "
2033                                      "next_listelem=%lx next_hwpage=%lx",
2034                                      *fmrlist, fmrlist,
2035                                      pginfo->u.fmr.next_listelem,
2036                                      pginfo->next_hwpage);
2037                         return -EFAULT;
2038                 }
2039                 (pginfo->hwpage_cnt)++;
2040                 if (pginfo->u.fmr.fmr_pgsize >= pginfo->hwpage_size) {
2041                         if (pginfo->next_hwpage %
2042                             (pginfo->u.fmr.fmr_pgsize /
2043                              pginfo->hwpage_size) == 0) {
2044                                 (pginfo->kpage_cnt)++;
2045                                 (pginfo->u.fmr.next_listelem)++;
2046                                 fmrlist++;
2047                                 pginfo->next_hwpage = 0;
2048                         } else
2049                                 (pginfo->next_hwpage)++;
2050                 } else {
2051                         unsigned int cnt_per_hwpage = pginfo->hwpage_size /
2052                                 pginfo->u.fmr.fmr_pgsize;
2053                         unsigned int j;
2054                         u64 prev = *kpage;
2055                         /* check if adrs are contiguous */
2056                         for (j = 1; j < cnt_per_hwpage; j++) {
2057                                 u64 p = phys_to_abs(fmrlist[j] &
2058                                                     ~(pginfo->hwpage_size - 1));
2059                                 if (prev + pginfo->u.fmr.fmr_pgsize != p) {
2060                                         ehca_gen_err("uncontiguous fmr pages "
2061                                                      "found prev=%lx p=%lx "
2062                                                      "idx=%x", prev, p, i + j);
2063                                         return -EINVAL;
2064                                 }
2065                                 prev = p;
2066                         }
2067                         pginfo->kpage_cnt += cnt_per_hwpage;
2068                         pginfo->u.fmr.next_listelem += cnt_per_hwpage;
2069                         fmrlist += cnt_per_hwpage;
2070                 }
2071                 kpage++;
2072         }
2073         return ret;
2074 }
2075
2076 /* setup page buffer from page info */
2077 int ehca_set_pagebuf(struct ehca_mr_pginfo *pginfo,
2078                      u32 number,
2079                      u64 *kpage)
2080 {
2081         int ret;
2082
2083         switch (pginfo->type) {
2084         case EHCA_MR_PGI_PHYS:
2085                 ret = ehca_set_pagebuf_phys(pginfo, number, kpage);
2086                 break;
2087         case EHCA_MR_PGI_USER:
2088                 ret = PAGE_SIZE >= pginfo->hwpage_size ?
2089                         ehca_set_pagebuf_user1(pginfo, number, kpage) :
2090                         ehca_set_pagebuf_user2(pginfo, number, kpage);
2091                 break;
2092         case EHCA_MR_PGI_FMR:
2093                 ret = ehca_set_pagebuf_fmr(pginfo, number, kpage);
2094                 break;
2095         default:
2096                 ehca_gen_err("bad pginfo->type=%x", pginfo->type);
2097                 ret = -EFAULT;
2098                 break;
2099         }
2100         return ret;
2101 } /* end ehca_set_pagebuf() */
2102
2103 /*----------------------------------------------------------------------*/
2104
2105 /*
2106  * check MR if it is a max-MR, i.e. uses whole memory
2107  * in case it's a max-MR 1 is returned, else 0
2108  */
2109 int ehca_mr_is_maxmr(u64 size,
2110                      u64 *iova_start)
2111 {
2112         /* a MR is treated as max-MR only if it fits following: */
2113         if ((size == ((u64)high_memory - PAGE_OFFSET)) &&
2114             (iova_start == (void *)KERNELBASE)) {
2115                 ehca_gen_dbg("this is a max-MR");
2116                 return 1;
2117         } else
2118                 return 0;
2119 } /* end ehca_mr_is_maxmr() */
2120
2121 /*----------------------------------------------------------------------*/
2122
2123 /* map access control for MR/MW. This routine is used for MR and MW. */
2124 void ehca_mrmw_map_acl(int ib_acl,
2125                        u32 *hipz_acl)
2126 {
2127         *hipz_acl = 0;
2128         if (ib_acl & IB_ACCESS_REMOTE_READ)
2129                 *hipz_acl |= HIPZ_ACCESSCTRL_R_READ;
2130         if (ib_acl & IB_ACCESS_REMOTE_WRITE)
2131                 *hipz_acl |= HIPZ_ACCESSCTRL_R_WRITE;
2132         if (ib_acl & IB_ACCESS_REMOTE_ATOMIC)
2133                 *hipz_acl |= HIPZ_ACCESSCTRL_R_ATOMIC;
2134         if (ib_acl & IB_ACCESS_LOCAL_WRITE)
2135                 *hipz_acl |= HIPZ_ACCESSCTRL_L_WRITE;
2136         if (ib_acl & IB_ACCESS_MW_BIND)
2137                 *hipz_acl |= HIPZ_ACCESSCTRL_MW_BIND;
2138 } /* end ehca_mrmw_map_acl() */
2139
2140 /*----------------------------------------------------------------------*/
2141
2142 /* sets page size in hipz access control for MR/MW. */
2143 void ehca_mrmw_set_pgsize_hipz_acl(u32 pgsize, u32 *hipz_acl) /*INOUT*/
2144 {
2145         *hipz_acl |= (ehca_encode_hwpage_size(pgsize) << 24);
2146 } /* end ehca_mrmw_set_pgsize_hipz_acl() */
2147
2148 /*----------------------------------------------------------------------*/
2149
2150 /*
2151  * reverse map access control for MR/MW.
2152  * This routine is used for MR and MW.
2153  */
2154 void ehca_mrmw_reverse_map_acl(const u32 *hipz_acl,
2155                                int *ib_acl) /*OUT*/
2156 {
2157         *ib_acl = 0;
2158         if (*hipz_acl & HIPZ_ACCESSCTRL_R_READ)
2159                 *ib_acl |= IB_ACCESS_REMOTE_READ;
2160         if (*hipz_acl & HIPZ_ACCESSCTRL_R_WRITE)
2161                 *ib_acl |= IB_ACCESS_REMOTE_WRITE;
2162         if (*hipz_acl & HIPZ_ACCESSCTRL_R_ATOMIC)
2163                 *ib_acl |= IB_ACCESS_REMOTE_ATOMIC;
2164         if (*hipz_acl & HIPZ_ACCESSCTRL_L_WRITE)
2165                 *ib_acl |= IB_ACCESS_LOCAL_WRITE;
2166         if (*hipz_acl & HIPZ_ACCESSCTRL_MW_BIND)
2167                 *ib_acl |= IB_ACCESS_MW_BIND;
2168 } /* end ehca_mrmw_reverse_map_acl() */
2169
2170
2171 /*----------------------------------------------------------------------*/
2172
2173 /*
2174  * MR destructor and constructor
2175  * used in Reregister MR verb, sets all fields in ehca_mr_t to 0,
2176  * except struct ib_mr and spinlock
2177  */
2178 void ehca_mr_deletenew(struct ehca_mr *mr)
2179 {
2180         mr->flags = 0;
2181         mr->num_kpages = 0;
2182         mr->num_hwpages = 0;
2183         mr->acl = 0;
2184         mr->start = NULL;
2185         mr->fmr_page_size = 0;
2186         mr->fmr_max_pages = 0;
2187         mr->fmr_max_maps = 0;
2188         mr->fmr_map_cnt = 0;
2189         memset(&mr->ipz_mr_handle, 0, sizeof(mr->ipz_mr_handle));
2190         memset(&mr->galpas, 0, sizeof(mr->galpas));
2191 } /* end ehca_mr_deletenew() */
2192
2193 int ehca_init_mrmw_cache(void)
2194 {
2195         mr_cache = kmem_cache_create("ehca_cache_mr",
2196                                      sizeof(struct ehca_mr), 0,
2197                                      SLAB_HWCACHE_ALIGN,
2198                                      NULL);
2199         if (!mr_cache)
2200                 return -ENOMEM;
2201         mw_cache = kmem_cache_create("ehca_cache_mw",
2202                                      sizeof(struct ehca_mw), 0,
2203                                      SLAB_HWCACHE_ALIGN,
2204                                      NULL);
2205         if (!mw_cache) {
2206                 kmem_cache_destroy(mr_cache);
2207                 mr_cache = NULL;
2208                 return -ENOMEM;
2209         }
2210         return 0;
2211 }
2212
2213 void ehca_cleanup_mrmw_cache(void)
2214 {
2215         if (mr_cache)
2216                 kmem_cache_destroy(mr_cache);
2217         if (mw_cache)
2218                 kmem_cache_destroy(mw_cache);
2219 }