1 // SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause 2 /* 3 * Copyright(c) 2016 Intel Corporation. 4 */ 5 6 #include <linux/slab.h> 7 #include <linux/vmalloc.h> 8 #include <rdma/ib_umem.h> 9 #include <rdma/rdma_vt.h> 10 #include "vt.h" 11 #include "mr.h" 12 #include "trace.h" 13 14 /** 15 * rvt_driver_mr_init - Init MR resources per driver 16 * @rdi: rvt dev struct 17 * 18 * Do any intilization needed when a driver registers with rdmavt. 19 * 20 * Return: 0 on success or errno on failure 21 */ 22 int rvt_driver_mr_init(struct rvt_dev_info *rdi) 23 { 24 unsigned int lkey_table_size = rdi->dparms.lkey_table_size; 25 unsigned lk_tab_size; 26 int i; 27 28 /* 29 * The top hfi1_lkey_table_size bits are used to index the 30 * table. The lower 8 bits can be owned by the user (copied from 31 * the LKEY). The remaining bits act as a generation number or tag. 32 */ 33 if (!lkey_table_size) 34 return -EINVAL; 35 36 spin_lock_init(&rdi->lkey_table.lock); 37 38 /* ensure generation is at least 4 bits */ 39 if (lkey_table_size > RVT_MAX_LKEY_TABLE_BITS) { 40 rvt_pr_warn(rdi, "lkey bits %u too large, reduced to %u\n", 41 lkey_table_size, RVT_MAX_LKEY_TABLE_BITS); 42 rdi->dparms.lkey_table_size = RVT_MAX_LKEY_TABLE_BITS; 43 lkey_table_size = rdi->dparms.lkey_table_size; 44 } 45 rdi->lkey_table.max = 1 << lkey_table_size; 46 rdi->lkey_table.shift = 32 - lkey_table_size; 47 lk_tab_size = rdi->lkey_table.max * sizeof(*rdi->lkey_table.table); 48 rdi->lkey_table.table = (struct rvt_mregion __rcu **) 49 vmalloc_node(lk_tab_size, rdi->dparms.node); 50 if (!rdi->lkey_table.table) 51 return -ENOMEM; 52 53 RCU_INIT_POINTER(rdi->dma_mr, NULL); 54 for (i = 0; i < rdi->lkey_table.max; i++) 55 RCU_INIT_POINTER(rdi->lkey_table.table[i], NULL); 56 57 rdi->dparms.props.max_mr = rdi->lkey_table.max; 58 return 0; 59 } 60 61 /** 62 * rvt_mr_exit - clean up MR 63 * @rdi: rvt dev structure 64 * 65 * called when drivers have unregistered or perhaps failed to register with us 66 */ 67 void rvt_mr_exit(struct rvt_dev_info *rdi) 68 { 69 if (rdi->dma_mr) 70 rvt_pr_err(rdi, "DMA MR not null!\n"); 71 72 vfree(rdi->lkey_table.table); 73 } 74 75 static void rvt_deinit_mregion(struct rvt_mregion *mr) 76 { 77 int i = mr->mapsz; 78 79 mr->mapsz = 0; 80 while (i) 81 kfree(mr->map[--i]); 82 percpu_ref_exit(&mr->refcount); 83 } 84 85 static void __rvt_mregion_complete(struct percpu_ref *ref) 86 { 87 struct rvt_mregion *mr = container_of(ref, struct rvt_mregion, 88 refcount); 89 90 complete(&mr->comp); 91 } 92 93 static int rvt_init_mregion(struct rvt_mregion *mr, struct ib_pd *pd, 94 int count, unsigned int percpu_flags) 95 { 96 int m, i = 0; 97 struct rvt_dev_info *dev = ib_to_rvt(pd->device); 98 99 mr->mapsz = 0; 100 m = (count + RVT_SEGSZ - 1) / RVT_SEGSZ; 101 for (; i < m; i++) { 102 mr->map[i] = kzalloc_node(sizeof(*mr->map[0]), GFP_KERNEL, 103 dev->dparms.node); 104 if (!mr->map[i]) 105 goto bail; 106 mr->mapsz++; 107 } 108 init_completion(&mr->comp); 109 /* count returning the ptr to user */ 110 if (percpu_ref_init(&mr->refcount, &__rvt_mregion_complete, 111 percpu_flags, GFP_KERNEL)) 112 goto bail; 113 114 atomic_set(&mr->lkey_invalid, 0); 115 mr->pd = pd; 116 mr->max_segs = count; 117 return 0; 118 bail: 119 rvt_deinit_mregion(mr); 120 return -ENOMEM; 121 } 122 123 /** 124 * rvt_alloc_lkey - allocate an lkey 125 * @mr: memory region that this lkey protects 126 * @dma_region: 0->normal key, 1->restricted DMA key 127 * 128 * Returns 0 if successful, otherwise returns -errno. 129 * 130 * Increments mr reference count as required. 131 * 132 * Sets the lkey field mr for non-dma regions. 133 * 134 */ 135 static int rvt_alloc_lkey(struct rvt_mregion *mr, int dma_region) 136 { 137 unsigned long flags; 138 u32 r; 139 u32 n; 140 int ret = 0; 141 struct rvt_dev_info *dev = ib_to_rvt(mr->pd->device); 142 struct rvt_lkey_table *rkt = &dev->lkey_table; 143 144 rvt_get_mr(mr); 145 spin_lock_irqsave(&rkt->lock, flags); 146 147 /* special case for dma_mr lkey == 0 */ 148 if (dma_region) { 149 struct rvt_mregion *tmr; 150 151 tmr = rcu_access_pointer(dev->dma_mr); 152 if (!tmr) { 153 mr->lkey_published = 1; 154 /* Insure published written first */ 155 rcu_assign_pointer(dev->dma_mr, mr); 156 rvt_get_mr(mr); 157 } 158 goto success; 159 } 160 161 /* Find the next available LKEY */ 162 r = rkt->next; 163 n = r; 164 for (;;) { 165 if (!rcu_access_pointer(rkt->table[r])) 166 break; 167 r = (r + 1) & (rkt->max - 1); 168 if (r == n) 169 goto bail; 170 } 171 rkt->next = (r + 1) & (rkt->max - 1); 172 /* 173 * Make sure lkey is never zero which is reserved to indicate an 174 * unrestricted LKEY. 175 */ 176 rkt->gen++; 177 /* 178 * bits are capped to ensure enough bits for generation number 179 */ 180 mr->lkey = (r << (32 - dev->dparms.lkey_table_size)) | 181 ((((1 << (24 - dev->dparms.lkey_table_size)) - 1) & rkt->gen) 182 << 8); 183 if (mr->lkey == 0) { 184 mr->lkey |= 1 << 8; 185 rkt->gen++; 186 } 187 mr->lkey_published = 1; 188 /* Insure published written first */ 189 rcu_assign_pointer(rkt->table[r], mr); 190 success: 191 spin_unlock_irqrestore(&rkt->lock, flags); 192 out: 193 return ret; 194 bail: 195 rvt_put_mr(mr); 196 spin_unlock_irqrestore(&rkt->lock, flags); 197 ret = -ENOMEM; 198 goto out; 199 } 200 201 /** 202 * rvt_free_lkey - free an lkey 203 * @mr: mr to free from tables 204 */ 205 static void rvt_free_lkey(struct rvt_mregion *mr) 206 { 207 unsigned long flags; 208 u32 lkey = mr->lkey; 209 u32 r; 210 struct rvt_dev_info *dev = ib_to_rvt(mr->pd->device); 211 struct rvt_lkey_table *rkt = &dev->lkey_table; 212 int freed = 0; 213 214 spin_lock_irqsave(&rkt->lock, flags); 215 if (!lkey) { 216 if (mr->lkey_published) { 217 mr->lkey_published = 0; 218 /* insure published is written before pointer */ 219 rcu_assign_pointer(dev->dma_mr, NULL); 220 rvt_put_mr(mr); 221 } 222 } else { 223 if (!mr->lkey_published) 224 goto out; 225 r = lkey >> (32 - dev->dparms.lkey_table_size); 226 mr->lkey_published = 0; 227 /* insure published is written before pointer */ 228 rcu_assign_pointer(rkt->table[r], NULL); 229 } 230 freed++; 231 out: 232 spin_unlock_irqrestore(&rkt->lock, flags); 233 if (freed) 234 percpu_ref_kill(&mr->refcount); 235 } 236 237 static struct rvt_mr *__rvt_alloc_mr(int count, struct ib_pd *pd) 238 { 239 struct rvt_mr *mr; 240 int rval = -ENOMEM; 241 int m; 242 243 /* Allocate struct plus pointers to first level page tables. */ 244 m = (count + RVT_SEGSZ - 1) / RVT_SEGSZ; 245 mr = kzalloc(struct_size(mr, mr.map, m), GFP_KERNEL); 246 if (!mr) 247 goto bail; 248 249 rval = rvt_init_mregion(&mr->mr, pd, count, 0); 250 if (rval) 251 goto bail; 252 /* 253 * ib_reg_phys_mr() will initialize mr->ibmr except for 254 * lkey and rkey. 255 */ 256 rval = rvt_alloc_lkey(&mr->mr, 0); 257 if (rval) 258 goto bail_mregion; 259 mr->ibmr.lkey = mr->mr.lkey; 260 mr->ibmr.rkey = mr->mr.lkey; 261 done: 262 return mr; 263 264 bail_mregion: 265 rvt_deinit_mregion(&mr->mr); 266 bail: 267 kfree(mr); 268 mr = ERR_PTR(rval); 269 goto done; 270 } 271 272 static void __rvt_free_mr(struct rvt_mr *mr) 273 { 274 rvt_free_lkey(&mr->mr); 275 rvt_deinit_mregion(&mr->mr); 276 kfree(mr); 277 } 278 279 /** 280 * rvt_get_dma_mr - get a DMA memory region 281 * @pd: protection domain for this memory region 282 * @acc: access flags 283 * 284 * Return: the memory region on success, otherwise returns an errno. 285 */ 286 struct ib_mr *rvt_get_dma_mr(struct ib_pd *pd, int acc) 287 { 288 struct rvt_mr *mr; 289 struct ib_mr *ret; 290 int rval; 291 292 if (ibpd_to_rvtpd(pd)->user) 293 return ERR_PTR(-EPERM); 294 295 mr = kzalloc(sizeof(*mr), GFP_KERNEL); 296 if (!mr) { 297 ret = ERR_PTR(-ENOMEM); 298 goto bail; 299 } 300 301 rval = rvt_init_mregion(&mr->mr, pd, 0, 0); 302 if (rval) { 303 ret = ERR_PTR(rval); 304 goto bail; 305 } 306 307 rval = rvt_alloc_lkey(&mr->mr, 1); 308 if (rval) { 309 ret = ERR_PTR(rval); 310 goto bail_mregion; 311 } 312 313 mr->mr.access_flags = acc; 314 ret = &mr->ibmr; 315 done: 316 return ret; 317 318 bail_mregion: 319 rvt_deinit_mregion(&mr->mr); 320 bail: 321 kfree(mr); 322 goto done; 323 } 324 325 /** 326 * rvt_reg_user_mr - register a userspace memory region 327 * @pd: protection domain for this memory region 328 * @start: starting userspace address 329 * @length: length of region to register 330 * @virt_addr: associated virtual address 331 * @mr_access_flags: access flags for this memory region 332 * @udata: unused by the driver 333 * 334 * Return: the memory region on success, otherwise returns an errno. 335 */ 336 struct ib_mr *rvt_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, 337 u64 virt_addr, int mr_access_flags, 338 struct ib_udata *udata) 339 { 340 struct rvt_mr *mr; 341 struct ib_umem *umem; 342 struct sg_page_iter sg_iter; 343 int n, m; 344 struct ib_mr *ret; 345 346 if (length == 0) 347 return ERR_PTR(-EINVAL); 348 349 umem = ib_umem_get(pd->device, start, length, mr_access_flags); 350 if (IS_ERR(umem)) 351 return (void *)umem; 352 353 n = ib_umem_num_pages(umem); 354 355 mr = __rvt_alloc_mr(n, pd); 356 if (IS_ERR(mr)) { 357 ret = (struct ib_mr *)mr; 358 goto bail_umem; 359 } 360 361 mr->mr.user_base = start; 362 mr->mr.iova = virt_addr; 363 mr->mr.length = length; 364 mr->mr.offset = ib_umem_offset(umem); 365 mr->mr.access_flags = mr_access_flags; 366 mr->umem = umem; 367 368 mr->mr.page_shift = PAGE_SHIFT; 369 m = 0; 370 n = 0; 371 for_each_sgtable_page (&umem->sgt_append.sgt, &sg_iter, 0) { 372 void *vaddr; 373 374 vaddr = page_address(sg_page_iter_page(&sg_iter)); 375 if (!vaddr) { 376 ret = ERR_PTR(-EINVAL); 377 goto bail_inval; 378 } 379 mr->mr.map[m]->segs[n].vaddr = vaddr; 380 mr->mr.map[m]->segs[n].length = PAGE_SIZE; 381 trace_rvt_mr_user_seg(&mr->mr, m, n, vaddr, PAGE_SIZE); 382 if (++n == RVT_SEGSZ) { 383 m++; 384 n = 0; 385 } 386 } 387 return &mr->ibmr; 388 389 bail_inval: 390 __rvt_free_mr(mr); 391 392 bail_umem: 393 ib_umem_release(umem); 394 395 return ret; 396 } 397 398 /** 399 * rvt_dereg_clean_qp_cb - callback from iterator 400 * @qp: the qp 401 * @v: the mregion (as u64) 402 * 403 * This routine fields the callback for all QPs and 404 * for QPs in the same PD as the MR will call the 405 * rvt_qp_mr_clean() to potentially cleanup references. 406 */ 407 static void rvt_dereg_clean_qp_cb(struct rvt_qp *qp, u64 v) 408 { 409 struct rvt_mregion *mr = (struct rvt_mregion *)v; 410 411 /* skip PDs that are not ours */ 412 if (mr->pd != qp->ibqp.pd) 413 return; 414 rvt_qp_mr_clean(qp, mr->lkey); 415 } 416 417 /** 418 * rvt_dereg_clean_qps - find QPs for reference cleanup 419 * @mr: the MR that is being deregistered 420 * 421 * This routine iterates RC QPs looking for references 422 * to the lkey noted in mr. 423 */ 424 static void rvt_dereg_clean_qps(struct rvt_mregion *mr) 425 { 426 struct rvt_dev_info *rdi = ib_to_rvt(mr->pd->device); 427 428 rvt_qp_iter(rdi, (u64)mr, rvt_dereg_clean_qp_cb); 429 } 430 431 /** 432 * rvt_check_refs - check references 433 * @mr: the megion 434 * @t: the caller identification 435 * 436 * This routine checks MRs holding a reference during 437 * when being de-registered. 438 * 439 * If the count is non-zero, the code calls a clean routine then 440 * waits for the timeout for the count to zero. 441 */ 442 static int rvt_check_refs(struct rvt_mregion *mr, const char *t) 443 { 444 unsigned long timeout; 445 struct rvt_dev_info *rdi = ib_to_rvt(mr->pd->device); 446 447 if (mr->lkey) { 448 /* avoid dma mr */ 449 rvt_dereg_clean_qps(mr); 450 /* @mr was indexed on rcu protected @lkey_table */ 451 synchronize_rcu(); 452 } 453 454 timeout = wait_for_completion_timeout(&mr->comp, 5 * HZ); 455 if (!timeout) { 456 rvt_pr_err(rdi, 457 "%s timeout mr %p pd %p lkey %x refcount %ld\n", 458 t, mr, mr->pd, mr->lkey, 459 atomic_long_read(&mr->refcount.data->count)); 460 rvt_get_mr(mr); 461 return -EBUSY; 462 } 463 return 0; 464 } 465 466 /** 467 * rvt_mr_has_lkey - is MR 468 * @mr: the mregion 469 * @lkey: the lkey 470 */ 471 bool rvt_mr_has_lkey(struct rvt_mregion *mr, u32 lkey) 472 { 473 return mr && lkey == mr->lkey; 474 } 475 476 /** 477 * rvt_ss_has_lkey - is mr in sge tests 478 * @ss: the sge state 479 * @lkey: the lkey 480 * 481 * This code tests for an MR in the indicated 482 * sge state. 483 */ 484 bool rvt_ss_has_lkey(struct rvt_sge_state *ss, u32 lkey) 485 { 486 int i; 487 bool rval = false; 488 489 if (!ss->num_sge) 490 return rval; 491 /* first one */ 492 rval = rvt_mr_has_lkey(ss->sge.mr, lkey); 493 /* any others */ 494 for (i = 0; !rval && i < ss->num_sge - 1; i++) 495 rval = rvt_mr_has_lkey(ss->sg_list[i].mr, lkey); 496 return rval; 497 } 498 499 /** 500 * rvt_dereg_mr - unregister and free a memory region 501 * @ibmr: the memory region to free 502 * @udata: unused by the driver 503 * 504 * Note that this is called to free MRs created by rvt_get_dma_mr() 505 * or rvt_reg_user_mr(). 506 * 507 * Returns 0 on success. 508 */ 509 int rvt_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) 510 { 511 struct rvt_mr *mr = to_imr(ibmr); 512 int ret; 513 514 rvt_free_lkey(&mr->mr); 515 516 rvt_put_mr(&mr->mr); /* will set completion if last */ 517 ret = rvt_check_refs(&mr->mr, __func__); 518 if (ret) 519 goto out; 520 rvt_deinit_mregion(&mr->mr); 521 ib_umem_release(mr->umem); 522 kfree(mr); 523 out: 524 return ret; 525 } 526 527 /** 528 * rvt_alloc_mr - Allocate a memory region usable with the 529 * @pd: protection domain for this memory region 530 * @mr_type: mem region type 531 * @max_num_sg: Max number of segments allowed 532 * 533 * Return: the memory region on success, otherwise return an errno. 534 */ 535 struct ib_mr *rvt_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type, 536 u32 max_num_sg) 537 { 538 struct rvt_mr *mr; 539 540 if (mr_type != IB_MR_TYPE_MEM_REG) 541 return ERR_PTR(-EINVAL); 542 543 mr = __rvt_alloc_mr(max_num_sg, pd); 544 if (IS_ERR(mr)) 545 return (struct ib_mr *)mr; 546 547 return &mr->ibmr; 548 } 549 550 /** 551 * rvt_set_page - page assignment function called by ib_sg_to_pages 552 * @ibmr: memory region 553 * @addr: dma address of mapped page 554 * 555 * Return: 0 on success 556 */ 557 static int rvt_set_page(struct ib_mr *ibmr, u64 addr) 558 { 559 struct rvt_mr *mr = to_imr(ibmr); 560 u32 ps = 1 << mr->mr.page_shift; 561 u32 mapped_segs = mr->mr.length >> mr->mr.page_shift; 562 int m, n; 563 564 if (unlikely(mapped_segs == mr->mr.max_segs)) 565 return -ENOMEM; 566 567 m = mapped_segs / RVT_SEGSZ; 568 n = mapped_segs % RVT_SEGSZ; 569 mr->mr.map[m]->segs[n].vaddr = (void *)addr; 570 mr->mr.map[m]->segs[n].length = ps; 571 mr->mr.length += ps; 572 trace_rvt_mr_page_seg(&mr->mr, m, n, (void *)addr, ps); 573 574 return 0; 575 } 576 577 /** 578 * rvt_map_mr_sg - map sg list and set it the memory region 579 * @ibmr: memory region 580 * @sg: dma mapped scatterlist 581 * @sg_nents: number of entries in sg 582 * @sg_offset: offset in bytes into sg 583 * 584 * Overwrite rvt_mr length with mr length calculated by ib_sg_to_pages. 585 * 586 * Return: number of sg elements mapped to the memory region 587 */ 588 int rvt_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, 589 int sg_nents, unsigned int *sg_offset) 590 { 591 struct rvt_mr *mr = to_imr(ibmr); 592 int ret; 593 594 mr->mr.length = 0; 595 mr->mr.page_shift = PAGE_SHIFT; 596 ret = ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, rvt_set_page); 597 mr->mr.user_base = ibmr->iova; 598 mr->mr.iova = ibmr->iova; 599 mr->mr.offset = ibmr->iova - (u64)mr->mr.map[0]->segs[0].vaddr; 600 mr->mr.length = (size_t)ibmr->length; 601 trace_rvt_map_mr_sg(ibmr, sg_nents, sg_offset); 602 return ret; 603 } 604 605 /** 606 * rvt_fast_reg_mr - fast register physical MR 607 * @qp: the queue pair where the work request comes from 608 * @ibmr: the memory region to be registered 609 * @key: updated key for this memory region 610 * @access: access flags for this memory region 611 * 612 * Returns 0 on success. 613 */ 614 int rvt_fast_reg_mr(struct rvt_qp *qp, struct ib_mr *ibmr, u32 key, 615 int access) 616 { 617 struct rvt_mr *mr = to_imr(ibmr); 618 619 if (qp->ibqp.pd != mr->mr.pd) 620 return -EACCES; 621 622 /* not applicable to dma MR or user MR */ 623 if (!mr->mr.lkey || mr->umem) 624 return -EINVAL; 625 626 if ((key & 0xFFFFFF00) != (mr->mr.lkey & 0xFFFFFF00)) 627 return -EINVAL; 628 629 ibmr->lkey = key; 630 ibmr->rkey = key; 631 mr->mr.lkey = key; 632 mr->mr.access_flags = access; 633 mr->mr.iova = ibmr->iova; 634 atomic_set(&mr->mr.lkey_invalid, 0); 635 636 return 0; 637 } 638 EXPORT_SYMBOL(rvt_fast_reg_mr); 639 640 /** 641 * rvt_invalidate_rkey - invalidate an MR rkey 642 * @qp: queue pair associated with the invalidate op 643 * @rkey: rkey to invalidate 644 * 645 * Returns 0 on success. 646 */ 647 int rvt_invalidate_rkey(struct rvt_qp *qp, u32 rkey) 648 { 649 struct rvt_dev_info *dev = ib_to_rvt(qp->ibqp.device); 650 struct rvt_lkey_table *rkt = &dev->lkey_table; 651 struct rvt_mregion *mr; 652 653 if (rkey == 0) 654 return -EINVAL; 655 656 rcu_read_lock(); 657 mr = rcu_dereference( 658 rkt->table[(rkey >> (32 - dev->dparms.lkey_table_size))]); 659 if (unlikely(!mr || mr->lkey != rkey || qp->ibqp.pd != mr->pd)) 660 goto bail; 661 662 atomic_set(&mr->lkey_invalid, 1); 663 rcu_read_unlock(); 664 return 0; 665 666 bail: 667 rcu_read_unlock(); 668 return -EINVAL; 669 } 670 EXPORT_SYMBOL(rvt_invalidate_rkey); 671 672 /** 673 * rvt_sge_adjacent - is isge compressible 674 * @last_sge: last outgoing SGE written 675 * @sge: SGE to check 676 * 677 * If adjacent will update last_sge to add length. 678 * 679 * Return: true if isge is adjacent to last sge 680 */ 681 static inline bool rvt_sge_adjacent(struct rvt_sge *last_sge, 682 struct ib_sge *sge) 683 { 684 if (last_sge && sge->lkey == last_sge->mr->lkey && 685 ((uint64_t)(last_sge->vaddr + last_sge->length) == sge->addr)) { 686 if (sge->lkey) { 687 if (unlikely((sge->addr - last_sge->mr->user_base + 688 sge->length > last_sge->mr->length))) 689 return false; /* overrun, caller will catch */ 690 } else { 691 last_sge->length += sge->length; 692 } 693 last_sge->sge_length += sge->length; 694 trace_rvt_sge_adjacent(last_sge, sge); 695 return true; 696 } 697 return false; 698 } 699 700 /** 701 * rvt_lkey_ok - check IB SGE for validity and initialize 702 * @rkt: table containing lkey to check SGE against 703 * @pd: protection domain 704 * @isge: outgoing internal SGE 705 * @last_sge: last outgoing SGE written 706 * @sge: SGE to check 707 * @acc: access flags 708 * 709 * Check the IB SGE for validity and initialize our internal version 710 * of it. 711 * 712 * Increments the reference count when a new sge is stored. 713 * 714 * Return: 0 if compressed, 1 if added , otherwise returns -errno. 715 */ 716 int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd, 717 struct rvt_sge *isge, struct rvt_sge *last_sge, 718 struct ib_sge *sge, int acc) 719 { 720 struct rvt_mregion *mr; 721 unsigned n, m; 722 size_t off; 723 724 /* 725 * We use LKEY == zero for kernel virtual addresses 726 * (see rvt_get_dma_mr()). 727 */ 728 if (sge->lkey == 0) { 729 struct rvt_dev_info *dev = ib_to_rvt(pd->ibpd.device); 730 731 if (pd->user) 732 return -EINVAL; 733 if (rvt_sge_adjacent(last_sge, sge)) 734 return 0; 735 rcu_read_lock(); 736 mr = rcu_dereference(dev->dma_mr); 737 if (!mr) 738 goto bail; 739 rvt_get_mr(mr); 740 rcu_read_unlock(); 741 742 isge->mr = mr; 743 isge->vaddr = (void *)sge->addr; 744 isge->length = sge->length; 745 isge->sge_length = sge->length; 746 isge->m = 0; 747 isge->n = 0; 748 goto ok; 749 } 750 if (rvt_sge_adjacent(last_sge, sge)) 751 return 0; 752 rcu_read_lock(); 753 mr = rcu_dereference(rkt->table[sge->lkey >> rkt->shift]); 754 if (!mr) 755 goto bail; 756 rvt_get_mr(mr); 757 if (!READ_ONCE(mr->lkey_published)) 758 goto bail_unref; 759 760 if (unlikely(atomic_read(&mr->lkey_invalid) || 761 mr->lkey != sge->lkey || mr->pd != &pd->ibpd)) 762 goto bail_unref; 763 764 off = sge->addr - mr->user_base; 765 if (unlikely(sge->addr < mr->user_base || 766 off + sge->length > mr->length || 767 (mr->access_flags & acc) != acc)) 768 goto bail_unref; 769 rcu_read_unlock(); 770 771 off += mr->offset; 772 if (mr->page_shift) { 773 /* 774 * page sizes are uniform power of 2 so no loop is necessary 775 * entries_spanned_by_off is the number of times the loop below 776 * would have executed. 777 */ 778 size_t entries_spanned_by_off; 779 780 entries_spanned_by_off = off >> mr->page_shift; 781 off -= (entries_spanned_by_off << mr->page_shift); 782 m = entries_spanned_by_off / RVT_SEGSZ; 783 n = entries_spanned_by_off % RVT_SEGSZ; 784 } else { 785 m = 0; 786 n = 0; 787 while (off >= mr->map[m]->segs[n].length) { 788 off -= mr->map[m]->segs[n].length; 789 n++; 790 if (n >= RVT_SEGSZ) { 791 m++; 792 n = 0; 793 } 794 } 795 } 796 isge->mr = mr; 797 isge->vaddr = mr->map[m]->segs[n].vaddr + off; 798 isge->length = mr->map[m]->segs[n].length - off; 799 isge->sge_length = sge->length; 800 isge->m = m; 801 isge->n = n; 802 ok: 803 trace_rvt_sge_new(isge, sge); 804 return 1; 805 bail_unref: 806 rvt_put_mr(mr); 807 bail: 808 rcu_read_unlock(); 809 return -EINVAL; 810 } 811 EXPORT_SYMBOL(rvt_lkey_ok); 812 813 /** 814 * rvt_rkey_ok - check the IB virtual address, length, and RKEY 815 * @qp: qp for validation 816 * @sge: SGE state 817 * @len: length of data 818 * @vaddr: virtual address to place data 819 * @rkey: rkey to check 820 * @acc: access flags 821 * 822 * Return: 1 if successful, otherwise 0. 823 * 824 * increments the reference count upon success 825 */ 826 int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge, 827 u32 len, u64 vaddr, u32 rkey, int acc) 828 { 829 struct rvt_dev_info *dev = ib_to_rvt(qp->ibqp.device); 830 struct rvt_lkey_table *rkt = &dev->lkey_table; 831 struct rvt_mregion *mr; 832 unsigned n, m; 833 size_t off; 834 835 /* 836 * We use RKEY == zero for kernel virtual addresses 837 * (see rvt_get_dma_mr()). 838 */ 839 rcu_read_lock(); 840 if (rkey == 0) { 841 struct rvt_pd *pd = ibpd_to_rvtpd(qp->ibqp.pd); 842 struct rvt_dev_info *rdi = ib_to_rvt(pd->ibpd.device); 843 844 if (pd->user) 845 goto bail; 846 mr = rcu_dereference(rdi->dma_mr); 847 if (!mr) 848 goto bail; 849 rvt_get_mr(mr); 850 rcu_read_unlock(); 851 852 sge->mr = mr; 853 sge->vaddr = (void *)vaddr; 854 sge->length = len; 855 sge->sge_length = len; 856 sge->m = 0; 857 sge->n = 0; 858 goto ok; 859 } 860 861 mr = rcu_dereference(rkt->table[rkey >> rkt->shift]); 862 if (!mr) 863 goto bail; 864 rvt_get_mr(mr); 865 /* insure mr read is before test */ 866 if (!READ_ONCE(mr->lkey_published)) 867 goto bail_unref; 868 if (unlikely(atomic_read(&mr->lkey_invalid) || 869 mr->lkey != rkey || qp->ibqp.pd != mr->pd)) 870 goto bail_unref; 871 872 off = vaddr - mr->iova; 873 if (unlikely(vaddr < mr->iova || off + len > mr->length || 874 (mr->access_flags & acc) == 0)) 875 goto bail_unref; 876 rcu_read_unlock(); 877 878 off += mr->offset; 879 if (mr->page_shift) { 880 /* 881 * page sizes are uniform power of 2 so no loop is necessary 882 * entries_spanned_by_off is the number of times the loop below 883 * would have executed. 884 */ 885 size_t entries_spanned_by_off; 886 887 entries_spanned_by_off = off >> mr->page_shift; 888 off -= (entries_spanned_by_off << mr->page_shift); 889 m = entries_spanned_by_off / RVT_SEGSZ; 890 n = entries_spanned_by_off % RVT_SEGSZ; 891 } else { 892 m = 0; 893 n = 0; 894 while (off >= mr->map[m]->segs[n].length) { 895 off -= mr->map[m]->segs[n].length; 896 n++; 897 if (n >= RVT_SEGSZ) { 898 m++; 899 n = 0; 900 } 901 } 902 } 903 sge->mr = mr; 904 sge->vaddr = mr->map[m]->segs[n].vaddr + off; 905 sge->length = mr->map[m]->segs[n].length - off; 906 sge->sge_length = len; 907 sge->m = m; 908 sge->n = n; 909 ok: 910 return 1; 911 bail_unref: 912 rvt_put_mr(mr); 913 bail: 914 rcu_read_unlock(); 915 return 0; 916 } 917 EXPORT_SYMBOL(rvt_rkey_ok); 918