drivers/infiniband/hw/mthca/mthca_mr.c

   1 /*
   2  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
   3  * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
   4  *
   5  * This software is available to you under a choice of one of two
   6  * licenses.  You may choose to be licensed under the terms of the GNU
   7  * General Public License (GPL) Version 2, available from the file
   8  * COPYING in the main directory of this source tree, or the
   9  * OpenIB.org BSD license below:
  10  *
  11  *     Redistribution and use in source and binary forms, with or
  12  *     without modification, are permitted provided that the following
  13  *     conditions are met:
  14  *
  15  *      - Redistributions of source code must retain the above
  16  *        copyright notice, this list of conditions and the following
  17  *        disclaimer.
  18  *
  19  *      - Redistributions in binary form must reproduce the above
  20  *        copyright notice, this list of conditions and the following
  21  *        disclaimer in the documentation and/or other materials
  22  *        provided with the distribution.
  23  *
  24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  31  * SOFTWARE.
  32  *
  33  * $Id: mthca_mr.c 1349 2004-12-16 21:09:43Z roland $
  34  */
  35
  36 #include <linux/slab.h>
  37 #include <linux/errno.h>
  38
  39 #include "mthca_dev.h"
  40 #include "mthca_cmd.h"
  41 #include "mthca_memfree.h"
  42
  43 struct mthca_mtt {
  44         struct mthca_buddy *buddy;
  45         int                 order;
  46         u32                 first_seg;
  47 };
  48
  49 /*
  50  * Must be packed because mtt_seg is 64 bits but only aligned to 32 bits.
  51  */
  52 struct mthca_mpt_entry {
  53         __be32 flags;
  54         __be32 page_size;
  55         __be32 key;
  56         __be32 pd;
  57         __be64 start;
  58         __be64 length;
  59         __be32 lkey;
  60         __be32 window_count;
  61         __be32 window_count_limit;
  62         __be64 mtt_seg;
  63         __be32 mtt_sz;          /* Arbel only */
  64         u32    reserved[2];
  65 } __attribute__((packed));
  66
  67 #define MTHCA_MPT_FLAG_SW_OWNS       (0xfUL << 28)
  68 #define MTHCA_MPT_FLAG_MIO           (1 << 17)
  69 #define MTHCA_MPT_FLAG_BIND_ENABLE   (1 << 15)
  70 #define MTHCA_MPT_FLAG_PHYSICAL      (1 <<  9)
  71 #define MTHCA_MPT_FLAG_REGION        (1 <<  8)
  72
  73 #define MTHCA_MTT_FLAG_PRESENT       1
  74
  75 #define MTHCA_MPT_STATUS_SW 0xF0
  76 #define MTHCA_MPT_STATUS_HW 0x00
  77
  78 #define SINAI_FMR_KEY_INC 0x1000000
  79
  80 /*
  81  * Buddy allocator for MTT segments (currently not very efficient
  82  * since it doesn't keep a free list and just searches linearly
  83  * through the bitmaps)
  84  */
  85
  86 static u32 mthca_buddy_alloc(struct mthca_buddy *buddy, int order)
  87 {
  88         int o;
  89         int m;
  90         u32 seg;
  91
  92         spin_lock(&buddy->lock);
  93
  94         for (o = order; o <= buddy->max_order; ++o) {
  95                 m = 1 << (buddy->max_order - o);
  96                 seg = find_first_bit(buddy->bits[o], m);
  97                 if (seg < m)
  98                         goto found;
  99         }
 100
 101         spin_unlock(&buddy->lock);
 102         return -1;
 103
 104  found:
 105         clear_bit(seg, buddy->bits[o]);
 106
 107         while (o > order) {
 108                 --o;
 109                 seg <<= 1;
 110                 set_bit(seg ^ 1, buddy->bits[o]);
 111         }
 112
 113         spin_unlock(&buddy->lock);
 114
 115         seg <<= order;
 116
 117         return seg;
 118 }
 119
 120 static void mthca_buddy_free(struct mthca_buddy *buddy, u32 seg, int order)
 121 {
 122         seg >>= order;
 123
 124         spin_lock(&buddy->lock);
 125
 126         while (test_bit(seg ^ 1, buddy->bits[order])) {
 127                 clear_bit(seg ^ 1, buddy->bits[order]);
 128                 seg >>= 1;
 129                 ++order;
 130         }
 131
 132         set_bit(seg, buddy->bits[order]);
 133
 134         spin_unlock(&buddy->lock);
 135 }
 136
 137 static int mthca_buddy_init(struct mthca_buddy *buddy, int max_order)
 138 {
 139         int i, s;
 140
 141         buddy->max_order = max_order;
 142         spin_lock_init(&buddy->lock);
 143
 144         buddy->bits = kzalloc((buddy->max_order + 1) * sizeof (long *),
 145                               GFP_KERNEL);
 146         if (!buddy->bits)
 147                 goto err_out;
 148
 149         for (i = 0; i <= buddy->max_order; ++i) {
 150                 s = BITS_TO_LONGS(1 << (buddy->max_order - i));
 151                 buddy->bits[i] = kmalloc(s * sizeof (long), GFP_KERNEL);
 152                 if (!buddy->bits[i])
 153                         goto err_out_free;
 154                 bitmap_zero(buddy->bits[i],
 155                             1 << (buddy->max_order - i));
 156         }
 157
 158         set_bit(0, buddy->bits[buddy->max_order]);
 159
 160         return 0;
 161
 162 err_out_free:
 163         for (i = 0; i <= buddy->max_order; ++i)
 164                 kfree(buddy->bits[i]);
 165
 166         kfree(buddy->bits);
 167
 168 err_out:
 169         return -ENOMEM;
 170 }
 171
 172 static void mthca_buddy_cleanup(struct mthca_buddy *buddy)
 173 {
 174         int i;
 175
 176         for (i = 0; i <= buddy->max_order; ++i)
 177                 kfree(buddy->bits[i]);
 178
 179         kfree(buddy->bits);
 180 }
 181
 182 static u32 mthca_alloc_mtt_range(struct mthca_dev *dev, int order,
 183                                  struct mthca_buddy *buddy)
 184 {
 185         u32 seg = mthca_buddy_alloc(buddy, order);
 186
 187         if (seg == -1)
 188                 return -1;
 189
 190         if (mthca_is_memfree(dev))
 191                 if (mthca_table_get_range(dev, dev->mr_table.mtt_table, seg,
 192                                           seg + (1 << order) - 1)) {
 193                         mthca_buddy_free(buddy, seg, order);
 194                         seg = -1;
 195                 }
 196
 197         return seg;
 198 }
 199
 200 static struct mthca_mtt *__mthca_alloc_mtt(struct mthca_dev *dev, int size,
 201                                            struct mthca_buddy *buddy)
 202 {
 203         struct mthca_mtt *mtt;
 204         int i;
 205
 206         if (size <= 0)
 207                 return ERR_PTR(-EINVAL);
 208
 209         mtt = kmalloc(sizeof *mtt, GFP_KERNEL);
 210         if (!mtt)
 211                 return ERR_PTR(-ENOMEM);
 212
 213         mtt->buddy = buddy;
 214         mtt->order = 0;
 215         for (i = MTHCA_MTT_SEG_SIZE / 8; i < size; i <<= 1)
 216                 ++mtt->order;
 217
 218         mtt->first_seg = mthca_alloc_mtt_range(dev, mtt->order, buddy);
 219         if (mtt->first_seg == -1) {
 220                 kfree(mtt);
 221                 return ERR_PTR(-ENOMEM);
 222         }
 223
 224         return mtt;
 225 }
 226
 227 struct mthca_mtt *mthca_alloc_mtt(struct mthca_dev *dev, int size)
 228 {
 229         return __mthca_alloc_mtt(dev, size, &dev->mr_table.mtt_buddy);
 230 }
 231
 232 void mthca_free_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt)
 233 {
 234         if (!mtt)
 235                 return;
 236
 237         mthca_buddy_free(mtt->buddy, mtt->first_seg, mtt->order);
 238
 239         mthca_table_put_range(dev, dev->mr_table.mtt_table,
 240                               mtt->first_seg,
 241                               mtt->first_seg + (1 << mtt->order) - 1);
 242
 243         kfree(mtt);
 244 }
 245
 246 int mthca_write_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt,
 247                     int start_index, u64 *buffer_list, int list_len)
 248 {
 249         struct mthca_mailbox *mailbox;
 250         __be64 *mtt_entry;
 251         int err = 0;
 252         u8 status;
 253         int i;
 254
 255         mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
 256         if (IS_ERR(mailbox))
 257                 return PTR_ERR(mailbox);
 258         mtt_entry = mailbox->buf;
 259
 260         while (list_len > 0) {
 261                 mtt_entry[0] = cpu_to_be64(dev->mr_table.mtt_base +
 262                                            mtt->first_seg * MTHCA_MTT_SEG_SIZE +
 263                                            start_index * 8);
 264                 mtt_entry[1] = 0;
 265                 for (i = 0; i < list_len && i < MTHCA_MAILBOX_SIZE / 8 - 2; ++i)
 266                         mtt_entry[i + 2] = cpu_to_be64(buffer_list[i] |
 267                                                        MTHCA_MTT_FLAG_PRESENT);
 268
 269                 /*
 270                  * If we have an odd number of entries to write, add
 271                  * one more dummy entry for firmware efficiency.
 272                  */
 273                 if (i & 1)
 274                         mtt_entry[i + 2] = 0;
 275
 276                 err = mthca_WRITE_MTT(dev, mailbox, (i + 1) & ~1, &status);
 277                 if (err) {
 278                         mthca_warn(dev, "WRITE_MTT failed (%d)\n", err);
 279                         goto out;
 280                 }
 281                 if (status) {
 282                         mthca_warn(dev, "WRITE_MTT returned status 0x%02x\n",
 283                                    status);
 284                         err = -EINVAL;
 285                         goto out;
 286                 }
 287
 288                 list_len    -= i;
 289                 start_index += i;
 290                 buffer_list += i;
 291         }
 292
 293 out:
 294         mthca_free_mailbox(dev, mailbox);
 295         return err;
 296 }
 297
 298 static inline u32 tavor_hw_index_to_key(u32 ind)
 299 {
 300         return ind;
 301 }
 302
 303 static inline u32 tavor_key_to_hw_index(u32 key)
 304 {
 305         return key;
 306 }
 307
 308 static inline u32 arbel_hw_index_to_key(u32 ind)
 309 {
 310         return (ind >> 24) | (ind << 8);
 311 }
 312
 313 static inline u32 arbel_key_to_hw_index(u32 key)
 314 {
 315         return (key << 24) | (key >> 8);
 316 }
 317
 318 static inline u32 hw_index_to_key(struct mthca_dev *dev, u32 ind)
 319 {
 320         if (mthca_is_memfree(dev))
 321                 return arbel_hw_index_to_key(ind);
 322         else
 323                 return tavor_hw_index_to_key(ind);
 324 }
 325
 326 static inline u32 key_to_hw_index(struct mthca_dev *dev, u32 key)
 327 {
 328         if (mthca_is_memfree(dev))
 329                 return arbel_key_to_hw_index(key);
 330         else
 331                 return tavor_key_to_hw_index(key);
 332 }
 333
 334 static inline u32 adjust_key(struct mthca_dev *dev, u32 key)
 335 {
 336         if (dev->mthca_flags & MTHCA_FLAG_SINAI_OPT)
 337                 return ((key << 20) & 0x800000) | (key & 0x7fffff);
 338         else
 339                 return key;
 340 }
 341
 342 int mthca_mr_alloc(struct mthca_dev *dev, u32 pd, int buffer_size_shift,
 343                    u64 iova, u64 total_size, u32 access, struct mthca_mr *mr)
 344 {
 345         struct mthca_mailbox *mailbox;
 346         struct mthca_mpt_entry *mpt_entry;
 347         u32 key;
 348         int i;
 349         int err;
 350         u8 status;
 351
 352         WARN_ON(buffer_size_shift >= 32);
 353
 354         key = mthca_alloc(&dev->mr_table.mpt_alloc);
 355         if (key == -1)
 356                 return -ENOMEM;
 357         key = adjust_key(dev, key);
 358         mr->ibmr.rkey = mr->ibmr.lkey = hw_index_to_key(dev, key);
 359
 360         if (mthca_is_memfree(dev)) {
 361                 err = mthca_table_get(dev, dev->mr_table.mpt_table, key);
 362                 if (err)
 363                         goto err_out_mpt_free;
 364         }
 365
 366         mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
 367         if (IS_ERR(mailbox)) {
 368                 err = PTR_ERR(mailbox);
 369                 goto err_out_table;
 370         }
 371         mpt_entry = mailbox->buf;
 372
 373         mpt_entry->flags = cpu_to_be32(MTHCA_MPT_FLAG_SW_OWNS     |
 374                                        MTHCA_MPT_FLAG_MIO         |
 375                                        MTHCA_MPT_FLAG_REGION      |
 376                                        access);
 377         if (!mr->mtt)
 378                 mpt_entry->flags |= cpu_to_be32(MTHCA_MPT_FLAG_PHYSICAL);
 379
 380         mpt_entry->page_size = cpu_to_be32(buffer_size_shift - 12);
 381         mpt_entry->key       = cpu_to_be32(key);
 382         mpt_entry->pd        = cpu_to_be32(pd);
 383         mpt_entry->start     = cpu_to_be64(iova);
 384         mpt_entry->length    = cpu_to_be64(total_size);
 385
 386         memset(&mpt_entry->lkey, 0,
 387                sizeof *mpt_entry - offsetof(struct mthca_mpt_entry, lkey));
 388
 389         if (mr->mtt)
 390                 mpt_entry->mtt_seg =
 391                         cpu_to_be64(dev->mr_table.mtt_base +
 392                                     mr->mtt->first_seg * MTHCA_MTT_SEG_SIZE);
 393
 394         if (0) {
 395                 mthca_dbg(dev, "Dumping MPT entry %08x:\n", mr->ibmr.lkey);
 396                 for (i = 0; i < sizeof (struct mthca_mpt_entry) / 4; ++i) {
 397                         if (i % 4 == 0)
 398                                 printk("[%02x] ", i * 4);
 399                         printk(" %08x", be32_to_cpu(((__be32 *) mpt_entry)[i]));
 400                         if ((i + 1) % 4 == 0)
 401                                 printk("\n");
 402                 }
 403         }
 404
 405         err = mthca_SW2HW_MPT(dev, mailbox,
 406                               key & (dev->limits.num_mpts - 1),
 407                               &status);
 408         if (err) {
 409                 mthca_warn(dev, "SW2HW_MPT failed (%d)\n", err);
 410                 goto err_out_mailbox;
 411         } else if (status) {
 412                 mthca_warn(dev, "SW2HW_MPT returned status 0x%02x\n",
 413                            status);
 414                 err = -EINVAL;
 415                 goto err_out_mailbox;
 416         }
 417
 418         mthca_free_mailbox(dev, mailbox);
 419         return err;
 420
 421 err_out_mailbox:
 422         mthca_free_mailbox(dev, mailbox);
 423
 424 err_out_table:
 425         mthca_table_put(dev, dev->mr_table.mpt_table, key);
 426
 427 err_out_mpt_free:
 428         mthca_free(&dev->mr_table.mpt_alloc, key);
 429         return err;
 430 }
 431
 432 int mthca_mr_alloc_notrans(struct mthca_dev *dev, u32 pd,
 433                            u32 access, struct mthca_mr *mr)
 434 {
 435         mr->mtt = NULL;
 436         return mthca_mr_alloc(dev, pd, 12, 0, ~0ULL, access, mr);
 437 }
 438
 439 int mthca_mr_alloc_phys(struct mthca_dev *dev, u32 pd,
 440                         u64 *buffer_list, int buffer_size_shift,
 441                         int list_len, u64 iova, u64 total_size,
 442                         u32 access, struct mthca_mr *mr)
 443 {
 444         int err;
 445
 446         mr->mtt = mthca_alloc_mtt(dev, list_len);
 447         if (IS_ERR(mr->mtt))
 448                 return PTR_ERR(mr->mtt);
 449
 450         err = mthca_write_mtt(dev, mr->mtt, 0, buffer_list, list_len);
 451         if (err) {
 452                 mthca_free_mtt(dev, mr->mtt);
 453                 return err;
 454         }
 455
 456         err = mthca_mr_alloc(dev, pd, buffer_size_shift, iova,
 457                              total_size, access, mr);
 458         if (err)
 459                 mthca_free_mtt(dev, mr->mtt);
 460
 461         return err;
 462 }
 463
 464 /* Free mr or fmr */
 465 static void mthca_free_region(struct mthca_dev *dev, u32 lkey)
 466 {
 467         mthca_table_put(dev, dev->mr_table.mpt_table,
 468                         key_to_hw_index(dev, lkey));
 469
 470         mthca_free(&dev->mr_table.mpt_alloc, key_to_hw_index(dev, lkey));
 471 }
 472
 473 void mthca_free_mr(struct mthca_dev *dev, struct mthca_mr *mr)
 474 {
 475         int err;
 476         u8 status;
 477
 478         err = mthca_HW2SW_MPT(dev, NULL,
 479                               key_to_hw_index(dev, mr->ibmr.lkey) &
 480                               (dev->limits.num_mpts - 1),
 481                               &status);
 482         if (err)
 483                 mthca_warn(dev, "HW2SW_MPT failed (%d)\n", err);
 484         else if (status)
 485                 mthca_warn(dev, "HW2SW_MPT returned status 0x%02x\n",
 486                            status);
 487
 488         mthca_free_region(dev, mr->ibmr.lkey);
 489         mthca_free_mtt(dev, mr->mtt);
 490 }
 491
 492 int mthca_fmr_alloc(struct mthca_dev *dev, u32 pd,
 493                     u32 access, struct mthca_fmr *mr)
 494 {
 495         struct mthca_mpt_entry *mpt_entry;
 496         struct mthca_mailbox *mailbox;
 497         u64 mtt_seg;
 498         u32 key, idx;
 499         u8 status;
 500         int list_len = mr->attr.max_pages;
 501         int err = -ENOMEM;
 502         int i;
 503
 504         if (mr->attr.page_shift < 12 || mr->attr.page_shift >= 32)
 505                 return -EINVAL;
 506
 507         /* For Arbel, all MTTs must fit in the same page. */
 508         if (mthca_is_memfree(dev) &&
 509             mr->attr.max_pages * sizeof *mr->mem.arbel.mtts > PAGE_SIZE)
 510                 return -EINVAL;
 511
 512         mr->maps = 0;
 513
 514         key = mthca_alloc(&dev->mr_table.mpt_alloc);
 515         if (key == -1)
 516                 return -ENOMEM;
 517         key = adjust_key(dev, key);
 518
 519         idx = key & (dev->limits.num_mpts - 1);
 520         mr->ibmr.rkey = mr->ibmr.lkey = hw_index_to_key(dev, key);
 521
 522         if (mthca_is_memfree(dev)) {
 523                 err = mthca_table_get(dev, dev->mr_table.mpt_table, key);
 524                 if (err)
 525                         goto err_out_mpt_free;
 526
 527                 mr->mem.arbel.mpt = mthca_table_find(dev->mr_table.mpt_table, key, NULL);
 528                 BUG_ON(!mr->mem.arbel.mpt);
 529         } else
 530                 mr->mem.tavor.mpt = dev->mr_table.tavor_fmr.mpt_base +
 531                         sizeof *(mr->mem.tavor.mpt) * idx;
 532
 533         mr->mtt = __mthca_alloc_mtt(dev, list_len, dev->mr_table.fmr_mtt_buddy);
 534         if (IS_ERR(mr->mtt))
 535                 goto err_out_table;
 536
 537         mtt_seg = mr->mtt->first_seg * MTHCA_MTT_SEG_SIZE;
 538
 539         if (mthca_is_memfree(dev)) {
 540                 mr->mem.arbel.mtts = mthca_table_find(dev->mr_table.mtt_table,
 541                                                       mr->mtt->first_seg,
 542                                                       &mr->mem.arbel.dma_handle);
 543                 BUG_ON(!mr->mem.arbel.mtts);
 544         } else
 545                 mr->mem.tavor.mtts = dev->mr_table.tavor_fmr.mtt_base + mtt_seg;
 546
 547         mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
 548         if (IS_ERR(mailbox))
 549                 goto err_out_free_mtt;
 550
 551         mpt_entry = mailbox->buf;
 552
 553         mpt_entry->flags = cpu_to_be32(MTHCA_MPT_FLAG_SW_OWNS     |
 554                                        MTHCA_MPT_FLAG_MIO         |
 555                                        MTHCA_MPT_FLAG_REGION      |
 556                                        access);
 557
 558         mpt_entry->page_size = cpu_to_be32(mr->attr.page_shift - 12);
 559         mpt_entry->key       = cpu_to_be32(key);
 560         mpt_entry->pd        = cpu_to_be32(pd);
 561         memset(&mpt_entry->start, 0,
 562                sizeof *mpt_entry - offsetof(struct mthca_mpt_entry, start));
 563         mpt_entry->mtt_seg   = cpu_to_be64(dev->mr_table.mtt_base + mtt_seg);
 564
 565         if (0) {
 566                 mthca_dbg(dev, "Dumping MPT entry %08x:\n", mr->ibmr.lkey);
 567                 for (i = 0; i < sizeof (struct mthca_mpt_entry) / 4; ++i) {
 568                         if (i % 4 == 0)
 569                                 printk("[%02x] ", i * 4);
 570                         printk(" %08x", be32_to_cpu(((__be32 *) mpt_entry)[i]));
 571                         if ((i + 1) % 4 == 0)
 572                                 printk("\n");
 573                 }
 574         }
 575
 576         err = mthca_SW2HW_MPT(dev, mailbox,
 577                               key & (dev->limits.num_mpts - 1),
 578                               &status);
 579         if (err) {
 580                 mthca_warn(dev, "SW2HW_MPT failed (%d)\n", err);
 581                 goto err_out_mailbox_free;
 582         }
 583         if (status) {
 584                 mthca_warn(dev, "SW2HW_MPT returned status 0x%02x\n",
 585                            status);
 586                 err = -EINVAL;
 587                 goto err_out_mailbox_free;
 588         }
 589
 590         mthca_free_mailbox(dev, mailbox);
 591         return 0;
 592
 593 err_out_mailbox_free:
 594         mthca_free_mailbox(dev, mailbox);
 595
 596 err_out_free_mtt:
 597         mthca_free_mtt(dev, mr->mtt);
 598
 599 err_out_table:
 600         mthca_table_put(dev, dev->mr_table.mpt_table, key);
 601
 602 err_out_mpt_free:
 603         mthca_free(&dev->mr_table.mpt_alloc, mr->ibmr.lkey);
 604         return err;
 605 }
 606
 607 int mthca_free_fmr(struct mthca_dev *dev, struct mthca_fmr *fmr)
 608 {
 609         if (fmr->maps)
 610                 return -EBUSY;
 611
 612         mthca_free_region(dev, fmr->ibmr.lkey);
 613         mthca_free_mtt(dev, fmr->mtt);
 614
 615         return 0;
 616 }
 617
 618 static inline int mthca_check_fmr(struct mthca_fmr *fmr, u64 *page_list,
 619                                   int list_len, u64 iova)
 620 {
 621         int i, page_mask;
 622
 623         if (list_len > fmr->attr.max_pages)
 624                 return -EINVAL;
 625
 626         page_mask = (1 << fmr->attr.page_shift) - 1;
 627
 628         /* We are getting page lists, so va must be page aligned. */
 629         if (iova & page_mask)
 630                 return -EINVAL;
 631
 632         /* Trust the user not to pass misaligned data in page_list */
 633         if (0)
 634                 for (i = 0; i < list_len; ++i) {
 635                         if (page_list[i] & ~page_mask)
 636                                 return -EINVAL;
 637                 }
 638
 639         if (fmr->maps >= fmr->attr.max_maps)
 640                 return -EINVAL;
 641
 642         return 0;
 643 }
 644
 645
 646 int mthca_tavor_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
 647                              int list_len, u64 iova)
 648 {
 649         struct mthca_fmr *fmr = to_mfmr(ibfmr);
 650         struct mthca_dev *dev = to_mdev(ibfmr->device);
 651         struct mthca_mpt_entry mpt_entry;
 652         u32 key;
 653         int i, err;
 654
 655         err = mthca_check_fmr(fmr, page_list, list_len, iova);
 656         if (err)
 657                 return err;
 658
 659         ++fmr->maps;
 660
 661         key = tavor_key_to_hw_index(fmr->ibmr.lkey);
 662         key += dev->limits.num_mpts;
 663         fmr->ibmr.lkey = fmr->ibmr.rkey = tavor_hw_index_to_key(key);
 664
 665         writeb(MTHCA_MPT_STATUS_SW, fmr->mem.tavor.mpt);
 666
 667         for (i = 0; i < list_len; ++i) {
 668                 __be64 mtt_entry = cpu_to_be64(page_list[i] |
 669                                                MTHCA_MTT_FLAG_PRESENT);
 670                 mthca_write64_raw(mtt_entry, fmr->mem.tavor.mtts + i);
 671         }
 672
 673         mpt_entry.lkey   = cpu_to_be32(key);
 674         mpt_entry.length = cpu_to_be64(list_len * (1ull << fmr->attr.page_shift));
 675         mpt_entry.start  = cpu_to_be64(iova);
 676
 677         __raw_writel((__force u32) mpt_entry.lkey, &fmr->mem.tavor.mpt->key);
 678         memcpy_toio(&fmr->mem.tavor.mpt->start, &mpt_entry.start,
 679                     offsetof(struct mthca_mpt_entry, window_count) -
 680                     offsetof(struct mthca_mpt_entry, start));
 681
 682         writeb(MTHCA_MPT_STATUS_HW, fmr->mem.tavor.mpt);
 683
 684         return 0;
 685 }
 686
 687 int mthca_arbel_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
 688                              int list_len, u64 iova)
 689 {
 690         struct mthca_fmr *fmr = to_mfmr(ibfmr);
 691         struct mthca_dev *dev = to_mdev(ibfmr->device);
 692         u32 key;
 693         int i, err;
 694
 695         err = mthca_check_fmr(fmr, page_list, list_len, iova);
 696         if (err)
 697                 return err;
 698
 699         ++fmr->maps;
 700
 701         key = arbel_key_to_hw_index(fmr->ibmr.lkey);
 702         if (dev->mthca_flags & MTHCA_FLAG_SINAI_OPT)
 703                 key += SINAI_FMR_KEY_INC;
 704         else
 705                 key += dev->limits.num_mpts;
 706         fmr->ibmr.lkey = fmr->ibmr.rkey = arbel_hw_index_to_key(key);
 707
 708         *(u8 *) fmr->mem.arbel.mpt = MTHCA_MPT_STATUS_SW;
 709
 710         wmb();
 711
 712         for (i = 0; i < list_len; ++i)
 713                 fmr->mem.arbel.mtts[i] = cpu_to_be64(page_list[i] |
 714                                                      MTHCA_MTT_FLAG_PRESENT);
 715
 716         dma_sync_single(&dev->pdev->dev, fmr->mem.arbel.dma_handle,
 717                         list_len * sizeof(u64), DMA_TO_DEVICE);
 718
 719         fmr->mem.arbel.mpt->key    = cpu_to_be32(key);
 720         fmr->mem.arbel.mpt->lkey   = cpu_to_be32(key);
 721         fmr->mem.arbel.mpt->length = cpu_to_be64(list_len * (1ull << fmr->attr.page_shift));
 722         fmr->mem.arbel.mpt->start  = cpu_to_be64(iova);
 723
 724         wmb();
 725
 726         *(u8 *) fmr->mem.arbel.mpt = MTHCA_MPT_STATUS_HW;
 727
 728         wmb();
 729
 730         return 0;
 731 }
 732
 733 void mthca_tavor_fmr_unmap(struct mthca_dev *dev, struct mthca_fmr *fmr)
 734 {
 735         u32 key;
 736
 737         if (!fmr->maps)
 738                 return;
 739
 740         key = tavor_key_to_hw_index(fmr->ibmr.lkey);
 741         key &= dev->limits.num_mpts - 1;
 742         fmr->ibmr.lkey = fmr->ibmr.rkey = tavor_hw_index_to_key(key);
 743
 744         fmr->maps = 0;
 745
 746         writeb(MTHCA_MPT_STATUS_SW, fmr->mem.tavor.mpt);
 747 }
 748
 749 void mthca_arbel_fmr_unmap(struct mthca_dev *dev, struct mthca_fmr *fmr)
 750 {
 751         u32 key;
 752
 753         if (!fmr->maps)
 754                 return;
 755
 756         key = arbel_key_to_hw_index(fmr->ibmr.lkey);
 757         key &= dev->limits.num_mpts - 1;
 758         fmr->ibmr.lkey = fmr->ibmr.rkey = arbel_hw_index_to_key(key);
 759
 760         fmr->maps = 0;
 761
 762         *(u8 *) fmr->mem.arbel.mpt = MTHCA_MPT_STATUS_SW;
 763 }
 764
 765 int mthca_init_mr_table(struct mthca_dev *dev)
 766 {
 767         unsigned long addr;
 768         int err, i;
 769
 770         err = mthca_alloc_init(&dev->mr_table.mpt_alloc,
 771                                dev->limits.num_mpts,
 772                                ~0, dev->limits.reserved_mrws);
 773         if (err)
 774                 return err;
 775
 776         if (!mthca_is_memfree(dev) &&
 777             (dev->mthca_flags & MTHCA_FLAG_DDR_HIDDEN))
 778                 dev->limits.fmr_reserved_mtts = 0;
 779         else
 780                 dev->mthca_flags |= MTHCA_FLAG_FMR;
 781
 782         if (dev->mthca_flags & MTHCA_FLAG_SINAI_OPT)
 783                 mthca_dbg(dev, "Memory key throughput optimization activated.\n");
 784
 785         err = mthca_buddy_init(&dev->mr_table.mtt_buddy,
 786                                fls(dev->limits.num_mtt_segs - 1));
 787
 788         if (err)
 789                 goto err_mtt_buddy;
 790
 791         dev->mr_table.tavor_fmr.mpt_base = NULL;
 792         dev->mr_table.tavor_fmr.mtt_base = NULL;
 793
 794         if (dev->limits.fmr_reserved_mtts) {
 795                 i = fls(dev->limits.fmr_reserved_mtts - 1);
 796
 797                 if (i >= 31) {
 798                         mthca_warn(dev, "Unable to reserve 2^31 FMR MTTs.\n");
 799                         err = -EINVAL;
 800                         goto err_fmr_mpt;
 801                 }
 802
 803                 addr = pci_resource_start(dev->pdev, 4) +
 804                         ((pci_resource_len(dev->pdev, 4) - 1) &
 805                          dev->mr_table.mpt_base);
 806
 807                 dev->mr_table.tavor_fmr.mpt_base =
 808                         ioremap(addr, (1 << i) * sizeof(struct mthca_mpt_entry));
 809
 810                 if (!dev->mr_table.tavor_fmr.mpt_base) {
 811                         mthca_warn(dev, "MPT ioremap for FMR failed.\n");
 812                         err = -ENOMEM;
 813                         goto err_fmr_mpt;
 814                 }
 815
 816                 addr = pci_resource_start(dev->pdev, 4) +
 817                         ((pci_resource_len(dev->pdev, 4) - 1) &
 818                          dev->mr_table.mtt_base);
 819
 820                 dev->mr_table.tavor_fmr.mtt_base =
 821                         ioremap(addr, (1 << i) * MTHCA_MTT_SEG_SIZE);
 822                 if (!dev->mr_table.tavor_fmr.mtt_base) {
 823                         mthca_warn(dev, "MTT ioremap for FMR failed.\n");
 824                         err = -ENOMEM;
 825                         goto err_fmr_mtt;
 826                 }
 827
 828                 err = mthca_buddy_init(&dev->mr_table.tavor_fmr.mtt_buddy, i);
 829                 if (err)
 830                         goto err_fmr_mtt_buddy;
 831
 832                 /* Prevent regular MRs from using FMR keys */
 833                 err = mthca_buddy_alloc(&dev->mr_table.mtt_buddy, i);
 834                 if (err)
 835                         goto err_reserve_fmr;
 836
 837                 dev->mr_table.fmr_mtt_buddy =
 838                         &dev->mr_table.tavor_fmr.mtt_buddy;
 839         } else
 840                 dev->mr_table.fmr_mtt_buddy = &dev->mr_table.mtt_buddy;
 841
 842         /* FMR table is always the first, take reserved MTTs out of there */
 843         if (dev->limits.reserved_mtts) {
 844                 i = fls(dev->limits.reserved_mtts - 1);
 845
 846                 if (mthca_alloc_mtt_range(dev, i,
 847                                           dev->mr_table.fmr_mtt_buddy) == -1) {
 848                         mthca_warn(dev, "MTT table of order %d is too small.\n",
 849                                   dev->mr_table.fmr_mtt_buddy->max_order);
 850                         err = -ENOMEM;
 851                         goto err_reserve_mtts;
 852                 }
 853         }
 854
 855         return 0;
 856
 857 err_reserve_mtts:
 858 err_reserve_fmr:
 859         if (dev->limits.fmr_reserved_mtts)
 860                 mthca_buddy_cleanup(&dev->mr_table.tavor_fmr.mtt_buddy);
 861
 862 err_fmr_mtt_buddy:
 863         if (dev->mr_table.tavor_fmr.mtt_base)
 864                 iounmap(dev->mr_table.tavor_fmr.mtt_base);
 865
 866 err_fmr_mtt:
 867         if (dev->mr_table.tavor_fmr.mpt_base)
 868                 iounmap(dev->mr_table.tavor_fmr.mpt_base);
 869
 870 err_fmr_mpt:
 871         mthca_buddy_cleanup(&dev->mr_table.mtt_buddy);
 872
 873 err_mtt_buddy:
 874         mthca_alloc_cleanup(&dev->mr_table.mpt_alloc);
 875
 876         return err;
 877 }
 878
 879 void mthca_cleanup_mr_table(struct mthca_dev *dev)
 880 {
 881         /* XXX check if any MRs are still allocated? */
 882         if (dev->limits.fmr_reserved_mtts)
 883                 mthca_buddy_cleanup(&dev->mr_table.tavor_fmr.mtt_buddy);
 884
 885         mthca_buddy_cleanup(&dev->mr_table.mtt_buddy);
 886
 887         if (dev->mr_table.tavor_fmr.mtt_base)
 888                 iounmap(dev->mr_table.tavor_fmr.mtt_base);
 889         if (dev->mr_table.tavor_fmr.mpt_base)
 890                 iounmap(dev->mr_table.tavor_fmr.mpt_base);
 891
 892         mthca_alloc_cleanup(&dev->mr_table.mpt_alloc);
 893 }