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 static 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 int mthca_write_mtt_size(struct mthca_dev *dev) 299 { 300 if (dev->mr_table.fmr_mtt_buddy != &dev->mr_table.mtt_buddy || 301 !(dev->mthca_flags & MTHCA_FLAG_FMR)) 302 /* 303 * Be friendly to WRITE_MTT command 304 * and leave two empty slots for the 305 * index and reserved fields of the 306 * mailbox. 307 */ 308 return PAGE_SIZE / sizeof (u64) - 2; 309 310 /* For Arbel, all MTTs must fit in the same page. */ 311 return mthca_is_memfree(dev) ? (PAGE_SIZE / sizeof (u64)) : 0x7ffffff; 312 } 313 314 static void mthca_tavor_write_mtt_seg(struct mthca_dev *dev, 315 struct mthca_mtt *mtt, int start_index, 316 u64 *buffer_list, int list_len) 317 { 318 u64 __iomem *mtts; 319 int i; 320 321 mtts = dev->mr_table.tavor_fmr.mtt_base + mtt->first_seg * MTHCA_MTT_SEG_SIZE + 322 start_index * sizeof (u64); 323 for (i = 0; i < list_len; ++i) 324 mthca_write64_raw(cpu_to_be64(buffer_list[i] | MTHCA_MTT_FLAG_PRESENT), 325 mtts + i); 326 } 327 328 static void mthca_arbel_write_mtt_seg(struct mthca_dev *dev, 329 struct mthca_mtt *mtt, int start_index, 330 u64 *buffer_list, int list_len) 331 { 332 __be64 *mtts; 333 dma_addr_t dma_handle; 334 int i; 335 int s = start_index * sizeof (u64); 336 337 /* For Arbel, all MTTs must fit in the same page. */ 338 BUG_ON(s / PAGE_SIZE != (s + list_len * sizeof(u64) - 1) / PAGE_SIZE); 339 /* Require full segments */ 340 BUG_ON(s % MTHCA_MTT_SEG_SIZE); 341 342 mtts = mthca_table_find(dev->mr_table.mtt_table, mtt->first_seg + 343 s / MTHCA_MTT_SEG_SIZE, &dma_handle); 344 345 BUG_ON(!mtts); 346 347 for (i = 0; i < list_len; ++i) 348 mtts[i] = cpu_to_be64(buffer_list[i] | MTHCA_MTT_FLAG_PRESENT); 349 350 dma_sync_single(&dev->pdev->dev, dma_handle, list_len * sizeof (u64), DMA_TO_DEVICE); 351 } 352 353 int mthca_write_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt, 354 int start_index, u64 *buffer_list, int list_len) 355 { 356 int size = mthca_write_mtt_size(dev); 357 int chunk; 358 359 if (dev->mr_table.fmr_mtt_buddy != &dev->mr_table.mtt_buddy || 360 !(dev->mthca_flags & MTHCA_FLAG_FMR)) 361 return __mthca_write_mtt(dev, mtt, start_index, buffer_list, list_len); 362 363 while (list_len > 0) { 364 chunk = min(size, list_len); 365 if (mthca_is_memfree(dev)) 366 mthca_arbel_write_mtt_seg(dev, mtt, start_index, 367 buffer_list, chunk); 368 else 369 mthca_tavor_write_mtt_seg(dev, mtt, start_index, 370 buffer_list, chunk); 371 372 list_len -= chunk; 373 start_index += chunk; 374 buffer_list += chunk; 375 } 376 377 return 0; 378 } 379 380 static inline u32 tavor_hw_index_to_key(u32 ind) 381 { 382 return ind; 383 } 384 385 static inline u32 tavor_key_to_hw_index(u32 key) 386 { 387 return key; 388 } 389 390 static inline u32 arbel_hw_index_to_key(u32 ind) 391 { 392 return (ind >> 24) | (ind << 8); 393 } 394 395 static inline u32 arbel_key_to_hw_index(u32 key) 396 { 397 return (key << 24) | (key >> 8); 398 } 399 400 static inline u32 hw_index_to_key(struct mthca_dev *dev, u32 ind) 401 { 402 if (mthca_is_memfree(dev)) 403 return arbel_hw_index_to_key(ind); 404 else 405 return tavor_hw_index_to_key(ind); 406 } 407 408 static inline u32 key_to_hw_index(struct mthca_dev *dev, u32 key) 409 { 410 if (mthca_is_memfree(dev)) 411 return arbel_key_to_hw_index(key); 412 else 413 return tavor_key_to_hw_index(key); 414 } 415 416 static inline u32 adjust_key(struct mthca_dev *dev, u32 key) 417 { 418 if (dev->mthca_flags & MTHCA_FLAG_SINAI_OPT) 419 return ((key << 20) & 0x800000) | (key & 0x7fffff); 420 else 421 return key; 422 } 423 424 int mthca_mr_alloc(struct mthca_dev *dev, u32 pd, int buffer_size_shift, 425 u64 iova, u64 total_size, u32 access, struct mthca_mr *mr) 426 { 427 struct mthca_mailbox *mailbox; 428 struct mthca_mpt_entry *mpt_entry; 429 u32 key; 430 int i; 431 int err; 432 u8 status; 433 434 WARN_ON(buffer_size_shift >= 32); 435 436 key = mthca_alloc(&dev->mr_table.mpt_alloc); 437 if (key == -1) 438 return -ENOMEM; 439 key = adjust_key(dev, key); 440 mr->ibmr.rkey = mr->ibmr.lkey = hw_index_to_key(dev, key); 441 442 if (mthca_is_memfree(dev)) { 443 err = mthca_table_get(dev, dev->mr_table.mpt_table, key); 444 if (err) 445 goto err_out_mpt_free; 446 } 447 448 mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL); 449 if (IS_ERR(mailbox)) { 450 err = PTR_ERR(mailbox); 451 goto err_out_table; 452 } 453 mpt_entry = mailbox->buf; 454 455 mpt_entry->flags = cpu_to_be32(MTHCA_MPT_FLAG_SW_OWNS | 456 MTHCA_MPT_FLAG_MIO | 457 MTHCA_MPT_FLAG_REGION | 458 access); 459 if (!mr->mtt) 460 mpt_entry->flags |= cpu_to_be32(MTHCA_MPT_FLAG_PHYSICAL); 461 462 mpt_entry->page_size = cpu_to_be32(buffer_size_shift - 12); 463 mpt_entry->key = cpu_to_be32(key); 464 mpt_entry->pd = cpu_to_be32(pd); 465 mpt_entry->start = cpu_to_be64(iova); 466 mpt_entry->length = cpu_to_be64(total_size); 467 468 memset(&mpt_entry->lkey, 0, 469 sizeof *mpt_entry - offsetof(struct mthca_mpt_entry, lkey)); 470 471 if (mr->mtt) 472 mpt_entry->mtt_seg = 473 cpu_to_be64(dev->mr_table.mtt_base + 474 mr->mtt->first_seg * MTHCA_MTT_SEG_SIZE); 475 476 if (0) { 477 mthca_dbg(dev, "Dumping MPT entry %08x:\n", mr->ibmr.lkey); 478 for (i = 0; i < sizeof (struct mthca_mpt_entry) / 4; ++i) { 479 if (i % 4 == 0) 480 printk("[%02x] ", i * 4); 481 printk(" %08x", be32_to_cpu(((__be32 *) mpt_entry)[i])); 482 if ((i + 1) % 4 == 0) 483 printk("\n"); 484 } 485 } 486 487 err = mthca_SW2HW_MPT(dev, mailbox, 488 key & (dev->limits.num_mpts - 1), 489 &status); 490 if (err) { 491 mthca_warn(dev, "SW2HW_MPT failed (%d)\n", err); 492 goto err_out_mailbox; 493 } else if (status) { 494 mthca_warn(dev, "SW2HW_MPT returned status 0x%02x\n", 495 status); 496 err = -EINVAL; 497 goto err_out_mailbox; 498 } 499 500 mthca_free_mailbox(dev, mailbox); 501 return err; 502 503 err_out_mailbox: 504 mthca_free_mailbox(dev, mailbox); 505 506 err_out_table: 507 mthca_table_put(dev, dev->mr_table.mpt_table, key); 508 509 err_out_mpt_free: 510 mthca_free(&dev->mr_table.mpt_alloc, key); 511 return err; 512 } 513 514 int mthca_mr_alloc_notrans(struct mthca_dev *dev, u32 pd, 515 u32 access, struct mthca_mr *mr) 516 { 517 mr->mtt = NULL; 518 return mthca_mr_alloc(dev, pd, 12, 0, ~0ULL, access, mr); 519 } 520 521 int mthca_mr_alloc_phys(struct mthca_dev *dev, u32 pd, 522 u64 *buffer_list, int buffer_size_shift, 523 int list_len, u64 iova, u64 total_size, 524 u32 access, struct mthca_mr *mr) 525 { 526 int err; 527 528 mr->mtt = mthca_alloc_mtt(dev, list_len); 529 if (IS_ERR(mr->mtt)) 530 return PTR_ERR(mr->mtt); 531 532 err = mthca_write_mtt(dev, mr->mtt, 0, buffer_list, list_len); 533 if (err) { 534 mthca_free_mtt(dev, mr->mtt); 535 return err; 536 } 537 538 err = mthca_mr_alloc(dev, pd, buffer_size_shift, iova, 539 total_size, access, mr); 540 if (err) 541 mthca_free_mtt(dev, mr->mtt); 542 543 return err; 544 } 545 546 /* Free mr or fmr */ 547 static void mthca_free_region(struct mthca_dev *dev, u32 lkey) 548 { 549 mthca_table_put(dev, dev->mr_table.mpt_table, 550 key_to_hw_index(dev, lkey)); 551 552 mthca_free(&dev->mr_table.mpt_alloc, key_to_hw_index(dev, lkey)); 553 } 554 555 void mthca_free_mr(struct mthca_dev *dev, struct mthca_mr *mr) 556 { 557 int err; 558 u8 status; 559 560 err = mthca_HW2SW_MPT(dev, NULL, 561 key_to_hw_index(dev, mr->ibmr.lkey) & 562 (dev->limits.num_mpts - 1), 563 &status); 564 if (err) 565 mthca_warn(dev, "HW2SW_MPT failed (%d)\n", err); 566 else if (status) 567 mthca_warn(dev, "HW2SW_MPT returned status 0x%02x\n", 568 status); 569 570 mthca_free_region(dev, mr->ibmr.lkey); 571 mthca_free_mtt(dev, mr->mtt); 572 } 573 574 int mthca_fmr_alloc(struct mthca_dev *dev, u32 pd, 575 u32 access, struct mthca_fmr *mr) 576 { 577 struct mthca_mpt_entry *mpt_entry; 578 struct mthca_mailbox *mailbox; 579 u64 mtt_seg; 580 u32 key, idx; 581 u8 status; 582 int list_len = mr->attr.max_pages; 583 int err = -ENOMEM; 584 int i; 585 586 if (mr->attr.page_shift < 12 || mr->attr.page_shift >= 32) 587 return -EINVAL; 588 589 /* For Arbel, all MTTs must fit in the same page. */ 590 if (mthca_is_memfree(dev) && 591 mr->attr.max_pages * sizeof *mr->mem.arbel.mtts > PAGE_SIZE) 592 return -EINVAL; 593 594 mr->maps = 0; 595 596 key = mthca_alloc(&dev->mr_table.mpt_alloc); 597 if (key == -1) 598 return -ENOMEM; 599 key = adjust_key(dev, key); 600 601 idx = key & (dev->limits.num_mpts - 1); 602 mr->ibmr.rkey = mr->ibmr.lkey = hw_index_to_key(dev, key); 603 604 if (mthca_is_memfree(dev)) { 605 err = mthca_table_get(dev, dev->mr_table.mpt_table, key); 606 if (err) 607 goto err_out_mpt_free; 608 609 mr->mem.arbel.mpt = mthca_table_find(dev->mr_table.mpt_table, key, NULL); 610 BUG_ON(!mr->mem.arbel.mpt); 611 } else 612 mr->mem.tavor.mpt = dev->mr_table.tavor_fmr.mpt_base + 613 sizeof *(mr->mem.tavor.mpt) * idx; 614 615 mr->mtt = __mthca_alloc_mtt(dev, list_len, dev->mr_table.fmr_mtt_buddy); 616 if (IS_ERR(mr->mtt)) { 617 err = PTR_ERR(mr->mtt); 618 goto err_out_table; 619 } 620 621 mtt_seg = mr->mtt->first_seg * MTHCA_MTT_SEG_SIZE; 622 623 if (mthca_is_memfree(dev)) { 624 mr->mem.arbel.mtts = mthca_table_find(dev->mr_table.mtt_table, 625 mr->mtt->first_seg, 626 &mr->mem.arbel.dma_handle); 627 BUG_ON(!mr->mem.arbel.mtts); 628 } else 629 mr->mem.tavor.mtts = dev->mr_table.tavor_fmr.mtt_base + mtt_seg; 630 631 mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL); 632 if (IS_ERR(mailbox)) { 633 err = PTR_ERR(mailbox); 634 goto err_out_free_mtt; 635 } 636 637 mpt_entry = mailbox->buf; 638 639 mpt_entry->flags = cpu_to_be32(MTHCA_MPT_FLAG_SW_OWNS | 640 MTHCA_MPT_FLAG_MIO | 641 MTHCA_MPT_FLAG_REGION | 642 access); 643 644 mpt_entry->page_size = cpu_to_be32(mr->attr.page_shift - 12); 645 mpt_entry->key = cpu_to_be32(key); 646 mpt_entry->pd = cpu_to_be32(pd); 647 memset(&mpt_entry->start, 0, 648 sizeof *mpt_entry - offsetof(struct mthca_mpt_entry, start)); 649 mpt_entry->mtt_seg = cpu_to_be64(dev->mr_table.mtt_base + mtt_seg); 650 651 if (0) { 652 mthca_dbg(dev, "Dumping MPT entry %08x:\n", mr->ibmr.lkey); 653 for (i = 0; i < sizeof (struct mthca_mpt_entry) / 4; ++i) { 654 if (i % 4 == 0) 655 printk("[%02x] ", i * 4); 656 printk(" %08x", be32_to_cpu(((__be32 *) mpt_entry)[i])); 657 if ((i + 1) % 4 == 0) 658 printk("\n"); 659 } 660 } 661 662 err = mthca_SW2HW_MPT(dev, mailbox, 663 key & (dev->limits.num_mpts - 1), 664 &status); 665 if (err) { 666 mthca_warn(dev, "SW2HW_MPT failed (%d)\n", err); 667 goto err_out_mailbox_free; 668 } 669 if (status) { 670 mthca_warn(dev, "SW2HW_MPT returned status 0x%02x\n", 671 status); 672 err = -EINVAL; 673 goto err_out_mailbox_free; 674 } 675 676 mthca_free_mailbox(dev, mailbox); 677 return 0; 678 679 err_out_mailbox_free: 680 mthca_free_mailbox(dev, mailbox); 681 682 err_out_free_mtt: 683 mthca_free_mtt(dev, mr->mtt); 684 685 err_out_table: 686 mthca_table_put(dev, dev->mr_table.mpt_table, key); 687 688 err_out_mpt_free: 689 mthca_free(&dev->mr_table.mpt_alloc, key); 690 return err; 691 } 692 693 int mthca_free_fmr(struct mthca_dev *dev, struct mthca_fmr *fmr) 694 { 695 if (fmr->maps) 696 return -EBUSY; 697 698 mthca_free_region(dev, fmr->ibmr.lkey); 699 mthca_free_mtt(dev, fmr->mtt); 700 701 return 0; 702 } 703 704 static inline int mthca_check_fmr(struct mthca_fmr *fmr, u64 *page_list, 705 int list_len, u64 iova) 706 { 707 int i, page_mask; 708 709 if (list_len > fmr->attr.max_pages) 710 return -EINVAL; 711 712 page_mask = (1 << fmr->attr.page_shift) - 1; 713 714 /* We are getting page lists, so va must be page aligned. */ 715 if (iova & page_mask) 716 return -EINVAL; 717 718 /* Trust the user not to pass misaligned data in page_list */ 719 if (0) 720 for (i = 0; i < list_len; ++i) { 721 if (page_list[i] & ~page_mask) 722 return -EINVAL; 723 } 724 725 if (fmr->maps >= fmr->attr.max_maps) 726 return -EINVAL; 727 728 return 0; 729 } 730 731 732 int mthca_tavor_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list, 733 int list_len, u64 iova) 734 { 735 struct mthca_fmr *fmr = to_mfmr(ibfmr); 736 struct mthca_dev *dev = to_mdev(ibfmr->device); 737 struct mthca_mpt_entry mpt_entry; 738 u32 key; 739 int i, err; 740 741 err = mthca_check_fmr(fmr, page_list, list_len, iova); 742 if (err) 743 return err; 744 745 ++fmr->maps; 746 747 key = tavor_key_to_hw_index(fmr->ibmr.lkey); 748 key += dev->limits.num_mpts; 749 fmr->ibmr.lkey = fmr->ibmr.rkey = tavor_hw_index_to_key(key); 750 751 writeb(MTHCA_MPT_STATUS_SW, fmr->mem.tavor.mpt); 752 753 for (i = 0; i < list_len; ++i) { 754 __be64 mtt_entry = cpu_to_be64(page_list[i] | 755 MTHCA_MTT_FLAG_PRESENT); 756 mthca_write64_raw(mtt_entry, fmr->mem.tavor.mtts + i); 757 } 758 759 mpt_entry.lkey = cpu_to_be32(key); 760 mpt_entry.length = cpu_to_be64(list_len * (1ull << fmr->attr.page_shift)); 761 mpt_entry.start = cpu_to_be64(iova); 762 763 __raw_writel((__force u32) mpt_entry.lkey, &fmr->mem.tavor.mpt->key); 764 memcpy_toio(&fmr->mem.tavor.mpt->start, &mpt_entry.start, 765 offsetof(struct mthca_mpt_entry, window_count) - 766 offsetof(struct mthca_mpt_entry, start)); 767 768 writeb(MTHCA_MPT_STATUS_HW, fmr->mem.tavor.mpt); 769 770 return 0; 771 } 772 773 int mthca_arbel_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list, 774 int list_len, u64 iova) 775 { 776 struct mthca_fmr *fmr = to_mfmr(ibfmr); 777 struct mthca_dev *dev = to_mdev(ibfmr->device); 778 u32 key; 779 int i, err; 780 781 err = mthca_check_fmr(fmr, page_list, list_len, iova); 782 if (err) 783 return err; 784 785 ++fmr->maps; 786 787 key = arbel_key_to_hw_index(fmr->ibmr.lkey); 788 if (dev->mthca_flags & MTHCA_FLAG_SINAI_OPT) 789 key += SINAI_FMR_KEY_INC; 790 else 791 key += dev->limits.num_mpts; 792 fmr->ibmr.lkey = fmr->ibmr.rkey = arbel_hw_index_to_key(key); 793 794 *(u8 *) fmr->mem.arbel.mpt = MTHCA_MPT_STATUS_SW; 795 796 wmb(); 797 798 for (i = 0; i < list_len; ++i) 799 fmr->mem.arbel.mtts[i] = cpu_to_be64(page_list[i] | 800 MTHCA_MTT_FLAG_PRESENT); 801 802 dma_sync_single(&dev->pdev->dev, fmr->mem.arbel.dma_handle, 803 list_len * sizeof(u64), DMA_TO_DEVICE); 804 805 fmr->mem.arbel.mpt->key = cpu_to_be32(key); 806 fmr->mem.arbel.mpt->lkey = cpu_to_be32(key); 807 fmr->mem.arbel.mpt->length = cpu_to_be64(list_len * (1ull << fmr->attr.page_shift)); 808 fmr->mem.arbel.mpt->start = cpu_to_be64(iova); 809 810 wmb(); 811 812 *(u8 *) fmr->mem.arbel.mpt = MTHCA_MPT_STATUS_HW; 813 814 wmb(); 815 816 return 0; 817 } 818 819 void mthca_tavor_fmr_unmap(struct mthca_dev *dev, struct mthca_fmr *fmr) 820 { 821 u32 key; 822 823 if (!fmr->maps) 824 return; 825 826 key = tavor_key_to_hw_index(fmr->ibmr.lkey); 827 key &= dev->limits.num_mpts - 1; 828 fmr->ibmr.lkey = fmr->ibmr.rkey = tavor_hw_index_to_key(key); 829 830 fmr->maps = 0; 831 832 writeb(MTHCA_MPT_STATUS_SW, fmr->mem.tavor.mpt); 833 } 834 835 void mthca_arbel_fmr_unmap(struct mthca_dev *dev, struct mthca_fmr *fmr) 836 { 837 u32 key; 838 839 if (!fmr->maps) 840 return; 841 842 key = arbel_key_to_hw_index(fmr->ibmr.lkey); 843 key &= dev->limits.num_mpts - 1; 844 key = adjust_key(dev, key); 845 fmr->ibmr.lkey = fmr->ibmr.rkey = arbel_hw_index_to_key(key); 846 847 fmr->maps = 0; 848 849 *(u8 *) fmr->mem.arbel.mpt = MTHCA_MPT_STATUS_SW; 850 } 851 852 int mthca_init_mr_table(struct mthca_dev *dev) 853 { 854 unsigned long addr; 855 int mpts, mtts, err, i; 856 857 err = mthca_alloc_init(&dev->mr_table.mpt_alloc, 858 dev->limits.num_mpts, 859 ~0, dev->limits.reserved_mrws); 860 if (err) 861 return err; 862 863 if (!mthca_is_memfree(dev) && 864 (dev->mthca_flags & MTHCA_FLAG_DDR_HIDDEN)) 865 dev->limits.fmr_reserved_mtts = 0; 866 else 867 dev->mthca_flags |= MTHCA_FLAG_FMR; 868 869 if (dev->mthca_flags & MTHCA_FLAG_SINAI_OPT) 870 mthca_dbg(dev, "Memory key throughput optimization activated.\n"); 871 872 err = mthca_buddy_init(&dev->mr_table.mtt_buddy, 873 fls(dev->limits.num_mtt_segs - 1)); 874 875 if (err) 876 goto err_mtt_buddy; 877 878 dev->mr_table.tavor_fmr.mpt_base = NULL; 879 dev->mr_table.tavor_fmr.mtt_base = NULL; 880 881 if (dev->limits.fmr_reserved_mtts) { 882 i = fls(dev->limits.fmr_reserved_mtts - 1); 883 884 if (i >= 31) { 885 mthca_warn(dev, "Unable to reserve 2^31 FMR MTTs.\n"); 886 err = -EINVAL; 887 goto err_fmr_mpt; 888 } 889 mpts = mtts = 1 << i; 890 } else { 891 mtts = dev->limits.num_mtt_segs; 892 mpts = dev->limits.num_mpts; 893 } 894 895 if (!mthca_is_memfree(dev) && 896 (dev->mthca_flags & MTHCA_FLAG_FMR)) { 897 898 addr = pci_resource_start(dev->pdev, 4) + 899 ((pci_resource_len(dev->pdev, 4) - 1) & 900 dev->mr_table.mpt_base); 901 902 dev->mr_table.tavor_fmr.mpt_base = 903 ioremap(addr, mpts * sizeof(struct mthca_mpt_entry)); 904 905 if (!dev->mr_table.tavor_fmr.mpt_base) { 906 mthca_warn(dev, "MPT ioremap for FMR failed.\n"); 907 err = -ENOMEM; 908 goto err_fmr_mpt; 909 } 910 911 addr = pci_resource_start(dev->pdev, 4) + 912 ((pci_resource_len(dev->pdev, 4) - 1) & 913 dev->mr_table.mtt_base); 914 915 dev->mr_table.tavor_fmr.mtt_base = 916 ioremap(addr, mtts * MTHCA_MTT_SEG_SIZE); 917 if (!dev->mr_table.tavor_fmr.mtt_base) { 918 mthca_warn(dev, "MTT ioremap for FMR failed.\n"); 919 err = -ENOMEM; 920 goto err_fmr_mtt; 921 } 922 } 923 924 if (dev->limits.fmr_reserved_mtts) { 925 err = mthca_buddy_init(&dev->mr_table.tavor_fmr.mtt_buddy, fls(mtts - 1)); 926 if (err) 927 goto err_fmr_mtt_buddy; 928 929 /* Prevent regular MRs from using FMR keys */ 930 err = mthca_buddy_alloc(&dev->mr_table.mtt_buddy, fls(mtts - 1)); 931 if (err) 932 goto err_reserve_fmr; 933 934 dev->mr_table.fmr_mtt_buddy = 935 &dev->mr_table.tavor_fmr.mtt_buddy; 936 } else 937 dev->mr_table.fmr_mtt_buddy = &dev->mr_table.mtt_buddy; 938 939 /* FMR table is always the first, take reserved MTTs out of there */ 940 if (dev->limits.reserved_mtts) { 941 i = fls(dev->limits.reserved_mtts - 1); 942 943 if (mthca_alloc_mtt_range(dev, i, 944 dev->mr_table.fmr_mtt_buddy) == -1) { 945 mthca_warn(dev, "MTT table of order %d is too small.\n", 946 dev->mr_table.fmr_mtt_buddy->max_order); 947 err = -ENOMEM; 948 goto err_reserve_mtts; 949 } 950 } 951 952 return 0; 953 954 err_reserve_mtts: 955 err_reserve_fmr: 956 if (dev->limits.fmr_reserved_mtts) 957 mthca_buddy_cleanup(&dev->mr_table.tavor_fmr.mtt_buddy); 958 959 err_fmr_mtt_buddy: 960 if (dev->mr_table.tavor_fmr.mtt_base) 961 iounmap(dev->mr_table.tavor_fmr.mtt_base); 962 963 err_fmr_mtt: 964 if (dev->mr_table.tavor_fmr.mpt_base) 965 iounmap(dev->mr_table.tavor_fmr.mpt_base); 966 967 err_fmr_mpt: 968 mthca_buddy_cleanup(&dev->mr_table.mtt_buddy); 969 970 err_mtt_buddy: 971 mthca_alloc_cleanup(&dev->mr_table.mpt_alloc); 972 973 return err; 974 } 975 976 void mthca_cleanup_mr_table(struct mthca_dev *dev) 977 { 978 /* XXX check if any MRs are still allocated? */ 979 if (dev->limits.fmr_reserved_mtts) 980 mthca_buddy_cleanup(&dev->mr_table.tavor_fmr.mtt_buddy); 981 982 mthca_buddy_cleanup(&dev->mr_table.mtt_buddy); 983 984 if (dev->mr_table.tavor_fmr.mtt_base) 985 iounmap(dev->mr_table.tavor_fmr.mtt_base); 986 if (dev->mr_table.tavor_fmr.mpt_base) 987 iounmap(dev->mr_table.tavor_fmr.mpt_base); 988 989 mthca_alloc_cleanup(&dev->mr_table.mpt_alloc); 990 } 991