1 /* 2 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved. 3 * Copyright (c) 2013-2014 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 #include <linux/module.h> 34 #include <linux/kernel.h> 35 #include <linux/slab.h> 36 #include <linux/mm.h> 37 #include <linux/highmem.h> 38 #include <linux/scatterlist.h> 39 40 #include "iscsi_iser.h" 41 static 42 int iser_fast_reg_fmr(struct iscsi_iser_task *iser_task, 43 struct iser_data_buf *mem, 44 struct iser_reg_resources *rsc, 45 struct iser_mem_reg *mem_reg); 46 static 47 int iser_fast_reg_mr(struct iscsi_iser_task *iser_task, 48 struct iser_data_buf *mem, 49 struct iser_reg_resources *rsc, 50 struct iser_mem_reg *mem_reg); 51 52 static const struct iser_reg_ops fastreg_ops = { 53 .alloc_reg_res = iser_alloc_fastreg_pool, 54 .free_reg_res = iser_free_fastreg_pool, 55 .reg_mem = iser_fast_reg_mr, 56 .unreg_mem = iser_unreg_mem_fastreg, 57 .reg_desc_get = iser_reg_desc_get_fr, 58 .reg_desc_put = iser_reg_desc_put_fr, 59 }; 60 61 static const struct iser_reg_ops fmr_ops = { 62 .alloc_reg_res = iser_alloc_fmr_pool, 63 .free_reg_res = iser_free_fmr_pool, 64 .reg_mem = iser_fast_reg_fmr, 65 .unreg_mem = iser_unreg_mem_fmr, 66 .reg_desc_get = iser_reg_desc_get_fmr, 67 .reg_desc_put = iser_reg_desc_put_fmr, 68 }; 69 70 void iser_reg_comp(struct ib_cq *cq, struct ib_wc *wc) 71 { 72 iser_err_comp(wc, "memreg"); 73 } 74 75 int iser_assign_reg_ops(struct iser_device *device) 76 { 77 struct ib_device *ib_dev = device->ib_device; 78 79 /* Assign function handles - based on FMR support */ 80 if (ib_dev->alloc_fmr && ib_dev->dealloc_fmr && 81 ib_dev->map_phys_fmr && ib_dev->unmap_fmr) { 82 iser_info("FMR supported, using FMR for registration\n"); 83 device->reg_ops = &fmr_ops; 84 } else if (ib_dev->attrs.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) { 85 iser_info("FastReg supported, using FastReg for registration\n"); 86 device->reg_ops = &fastreg_ops; 87 device->remote_inv_sup = iser_always_reg; 88 } else { 89 iser_err("IB device does not support FMRs nor FastRegs, can't register memory\n"); 90 return -1; 91 } 92 93 return 0; 94 } 95 96 struct iser_fr_desc * 97 iser_reg_desc_get_fr(struct ib_conn *ib_conn) 98 { 99 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool; 100 struct iser_fr_desc *desc; 101 unsigned long flags; 102 103 spin_lock_irqsave(&fr_pool->lock, flags); 104 desc = list_first_entry(&fr_pool->list, 105 struct iser_fr_desc, list); 106 list_del(&desc->list); 107 spin_unlock_irqrestore(&fr_pool->lock, flags); 108 109 return desc; 110 } 111 112 void 113 iser_reg_desc_put_fr(struct ib_conn *ib_conn, 114 struct iser_fr_desc *desc) 115 { 116 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool; 117 unsigned long flags; 118 119 spin_lock_irqsave(&fr_pool->lock, flags); 120 list_add(&desc->list, &fr_pool->list); 121 spin_unlock_irqrestore(&fr_pool->lock, flags); 122 } 123 124 struct iser_fr_desc * 125 iser_reg_desc_get_fmr(struct ib_conn *ib_conn) 126 { 127 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool; 128 129 return list_first_entry(&fr_pool->list, 130 struct iser_fr_desc, list); 131 } 132 133 void 134 iser_reg_desc_put_fmr(struct ib_conn *ib_conn, 135 struct iser_fr_desc *desc) 136 { 137 } 138 139 static void iser_data_buf_dump(struct iser_data_buf *data, 140 struct ib_device *ibdev) 141 { 142 struct scatterlist *sg; 143 int i; 144 145 for_each_sg(data->sg, sg, data->dma_nents, i) 146 iser_dbg("sg[%d] dma_addr:0x%lX page:0x%p " 147 "off:0x%x sz:0x%x dma_len:0x%x\n", 148 i, (unsigned long)ib_sg_dma_address(ibdev, sg), 149 sg_page(sg), sg->offset, 150 sg->length, ib_sg_dma_len(ibdev, sg)); 151 } 152 153 static void iser_dump_page_vec(struct iser_page_vec *page_vec) 154 { 155 int i; 156 157 iser_err("page vec npages %d data length %lld\n", 158 page_vec->npages, page_vec->fake_mr.length); 159 for (i = 0; i < page_vec->npages; i++) 160 iser_err("vec[%d]: %llx\n", i, page_vec->pages[i]); 161 } 162 163 int iser_dma_map_task_data(struct iscsi_iser_task *iser_task, 164 struct iser_data_buf *data, 165 enum iser_data_dir iser_dir, 166 enum dma_data_direction dma_dir) 167 { 168 struct ib_device *dev; 169 170 iser_task->dir[iser_dir] = 1; 171 dev = iser_task->iser_conn->ib_conn.device->ib_device; 172 173 data->dma_nents = ib_dma_map_sg(dev, data->sg, data->size, dma_dir); 174 if (data->dma_nents == 0) { 175 iser_err("dma_map_sg failed!!!\n"); 176 return -EINVAL; 177 } 178 return 0; 179 } 180 181 void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task, 182 struct iser_data_buf *data, 183 enum dma_data_direction dir) 184 { 185 struct ib_device *dev; 186 187 dev = iser_task->iser_conn->ib_conn.device->ib_device; 188 ib_dma_unmap_sg(dev, data->sg, data->size, dir); 189 } 190 191 static int 192 iser_reg_dma(struct iser_device *device, struct iser_data_buf *mem, 193 struct iser_mem_reg *reg) 194 { 195 struct scatterlist *sg = mem->sg; 196 197 reg->sge.lkey = device->pd->local_dma_lkey; 198 /* 199 * FIXME: rework the registration code path to differentiate 200 * rkey/lkey use cases 201 */ 202 203 if (device->pd->flags & IB_PD_UNSAFE_GLOBAL_RKEY) 204 reg->rkey = device->pd->unsafe_global_rkey; 205 else 206 reg->rkey = 0; 207 reg->sge.addr = ib_sg_dma_address(device->ib_device, &sg[0]); 208 reg->sge.length = ib_sg_dma_len(device->ib_device, &sg[0]); 209 210 iser_dbg("Single DMA entry: lkey=0x%x, rkey=0x%x, addr=0x%llx," 211 " length=0x%x\n", reg->sge.lkey, reg->rkey, 212 reg->sge.addr, reg->sge.length); 213 214 return 0; 215 } 216 217 static int iser_set_page(struct ib_mr *mr, u64 addr) 218 { 219 struct iser_page_vec *page_vec = 220 container_of(mr, struct iser_page_vec, fake_mr); 221 222 page_vec->pages[page_vec->npages++] = addr; 223 224 return 0; 225 } 226 227 static 228 int iser_fast_reg_fmr(struct iscsi_iser_task *iser_task, 229 struct iser_data_buf *mem, 230 struct iser_reg_resources *rsc, 231 struct iser_mem_reg *reg) 232 { 233 struct ib_conn *ib_conn = &iser_task->iser_conn->ib_conn; 234 struct iser_device *device = ib_conn->device; 235 struct iser_page_vec *page_vec = rsc->page_vec; 236 struct ib_fmr_pool *fmr_pool = rsc->fmr_pool; 237 struct ib_pool_fmr *fmr; 238 int ret, plen; 239 240 page_vec->npages = 0; 241 page_vec->fake_mr.page_size = SIZE_4K; 242 plen = ib_sg_to_pages(&page_vec->fake_mr, mem->sg, 243 mem->size, NULL, iser_set_page); 244 if (unlikely(plen < mem->size)) { 245 iser_err("page vec too short to hold this SG\n"); 246 iser_data_buf_dump(mem, device->ib_device); 247 iser_dump_page_vec(page_vec); 248 return -EINVAL; 249 } 250 251 fmr = ib_fmr_pool_map_phys(fmr_pool, page_vec->pages, 252 page_vec->npages, page_vec->pages[0]); 253 if (IS_ERR(fmr)) { 254 ret = PTR_ERR(fmr); 255 iser_err("ib_fmr_pool_map_phys failed: %d\n", ret); 256 return ret; 257 } 258 259 reg->sge.lkey = fmr->fmr->lkey; 260 reg->rkey = fmr->fmr->rkey; 261 reg->sge.addr = page_vec->fake_mr.iova; 262 reg->sge.length = page_vec->fake_mr.length; 263 reg->mem_h = fmr; 264 265 iser_dbg("fmr reg: lkey=0x%x, rkey=0x%x, addr=0x%llx," 266 " length=0x%x\n", reg->sge.lkey, reg->rkey, 267 reg->sge.addr, reg->sge.length); 268 269 return 0; 270 } 271 272 /** 273 * Unregister (previosuly registered using FMR) memory. 274 * If memory is non-FMR does nothing. 275 */ 276 void iser_unreg_mem_fmr(struct iscsi_iser_task *iser_task, 277 enum iser_data_dir cmd_dir) 278 { 279 struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir]; 280 int ret; 281 282 if (!reg->mem_h) 283 return; 284 285 iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n", reg->mem_h); 286 287 ret = ib_fmr_pool_unmap((struct ib_pool_fmr *)reg->mem_h); 288 if (ret) 289 iser_err("ib_fmr_pool_unmap failed %d\n", ret); 290 291 reg->mem_h = NULL; 292 } 293 294 void iser_unreg_mem_fastreg(struct iscsi_iser_task *iser_task, 295 enum iser_data_dir cmd_dir) 296 { 297 struct iser_device *device = iser_task->iser_conn->ib_conn.device; 298 struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir]; 299 300 if (!reg->mem_h) 301 return; 302 303 device->reg_ops->reg_desc_put(&iser_task->iser_conn->ib_conn, 304 reg->mem_h); 305 reg->mem_h = NULL; 306 } 307 308 static void 309 iser_set_dif_domain(struct scsi_cmnd *sc, struct ib_sig_attrs *sig_attrs, 310 struct ib_sig_domain *domain) 311 { 312 domain->sig_type = IB_SIG_TYPE_T10_DIF; 313 domain->sig.dif.pi_interval = scsi_prot_interval(sc); 314 domain->sig.dif.ref_tag = t10_pi_ref_tag(sc->request); 315 /* 316 * At the moment we hard code those, but in the future 317 * we will take them from sc. 318 */ 319 domain->sig.dif.apptag_check_mask = 0xffff; 320 domain->sig.dif.app_escape = true; 321 domain->sig.dif.ref_escape = true; 322 if (sc->prot_flags & SCSI_PROT_REF_INCREMENT) 323 domain->sig.dif.ref_remap = true; 324 }; 325 326 static int 327 iser_set_sig_attrs(struct scsi_cmnd *sc, struct ib_sig_attrs *sig_attrs) 328 { 329 switch (scsi_get_prot_op(sc)) { 330 case SCSI_PROT_WRITE_INSERT: 331 case SCSI_PROT_READ_STRIP: 332 sig_attrs->mem.sig_type = IB_SIG_TYPE_NONE; 333 iser_set_dif_domain(sc, sig_attrs, &sig_attrs->wire); 334 sig_attrs->wire.sig.dif.bg_type = IB_T10DIF_CRC; 335 break; 336 case SCSI_PROT_READ_INSERT: 337 case SCSI_PROT_WRITE_STRIP: 338 sig_attrs->wire.sig_type = IB_SIG_TYPE_NONE; 339 iser_set_dif_domain(sc, sig_attrs, &sig_attrs->mem); 340 sig_attrs->mem.sig.dif.bg_type = sc->prot_flags & SCSI_PROT_IP_CHECKSUM ? 341 IB_T10DIF_CSUM : IB_T10DIF_CRC; 342 break; 343 case SCSI_PROT_READ_PASS: 344 case SCSI_PROT_WRITE_PASS: 345 iser_set_dif_domain(sc, sig_attrs, &sig_attrs->wire); 346 sig_attrs->wire.sig.dif.bg_type = IB_T10DIF_CRC; 347 iser_set_dif_domain(sc, sig_attrs, &sig_attrs->mem); 348 sig_attrs->mem.sig.dif.bg_type = sc->prot_flags & SCSI_PROT_IP_CHECKSUM ? 349 IB_T10DIF_CSUM : IB_T10DIF_CRC; 350 break; 351 default: 352 iser_err("Unsupported PI operation %d\n", 353 scsi_get_prot_op(sc)); 354 return -EINVAL; 355 } 356 357 return 0; 358 } 359 360 static inline void 361 iser_set_prot_checks(struct scsi_cmnd *sc, u8 *mask) 362 { 363 *mask = 0; 364 if (sc->prot_flags & SCSI_PROT_REF_CHECK) 365 *mask |= IB_SIG_CHECK_REFTAG; 366 if (sc->prot_flags & SCSI_PROT_GUARD_CHECK) 367 *mask |= IB_SIG_CHECK_GUARD; 368 } 369 370 static inline void 371 iser_inv_rkey(struct ib_send_wr *inv_wr, 372 struct ib_mr *mr, 373 struct ib_cqe *cqe) 374 { 375 inv_wr->opcode = IB_WR_LOCAL_INV; 376 inv_wr->wr_cqe = cqe; 377 inv_wr->ex.invalidate_rkey = mr->rkey; 378 inv_wr->send_flags = 0; 379 inv_wr->num_sge = 0; 380 } 381 382 static int 383 iser_reg_sig_mr(struct iscsi_iser_task *iser_task, 384 struct iser_pi_context *pi_ctx, 385 struct iser_mem_reg *data_reg, 386 struct iser_mem_reg *prot_reg, 387 struct iser_mem_reg *sig_reg) 388 { 389 struct iser_tx_desc *tx_desc = &iser_task->desc; 390 struct ib_sig_attrs *sig_attrs = &tx_desc->sig_attrs; 391 struct ib_cqe *cqe = &iser_task->iser_conn->ib_conn.reg_cqe; 392 struct ib_sig_handover_wr *wr; 393 struct ib_mr *mr = pi_ctx->sig_mr; 394 int ret; 395 396 memset(sig_attrs, 0, sizeof(*sig_attrs)); 397 ret = iser_set_sig_attrs(iser_task->sc, sig_attrs); 398 if (ret) 399 goto err; 400 401 iser_set_prot_checks(iser_task->sc, &sig_attrs->check_mask); 402 403 if (pi_ctx->sig_mr_valid) 404 iser_inv_rkey(iser_tx_next_wr(tx_desc), mr, cqe); 405 406 ib_update_fast_reg_key(mr, ib_inc_rkey(mr->rkey)); 407 408 wr = container_of(iser_tx_next_wr(tx_desc), struct ib_sig_handover_wr, 409 wr); 410 wr->wr.opcode = IB_WR_REG_SIG_MR; 411 wr->wr.wr_cqe = cqe; 412 wr->wr.sg_list = &data_reg->sge; 413 wr->wr.num_sge = 1; 414 wr->wr.send_flags = 0; 415 wr->sig_attrs = sig_attrs; 416 wr->sig_mr = mr; 417 if (scsi_prot_sg_count(iser_task->sc)) 418 wr->prot = &prot_reg->sge; 419 else 420 wr->prot = NULL; 421 wr->access_flags = IB_ACCESS_LOCAL_WRITE | 422 IB_ACCESS_REMOTE_READ | 423 IB_ACCESS_REMOTE_WRITE; 424 pi_ctx->sig_mr_valid = 1; 425 426 sig_reg->sge.lkey = mr->lkey; 427 sig_reg->rkey = mr->rkey; 428 sig_reg->sge.addr = 0; 429 sig_reg->sge.length = scsi_transfer_length(iser_task->sc); 430 431 iser_dbg("lkey=0x%x rkey=0x%x addr=0x%llx length=%u\n", 432 sig_reg->sge.lkey, sig_reg->rkey, sig_reg->sge.addr, 433 sig_reg->sge.length); 434 err: 435 return ret; 436 } 437 438 static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task, 439 struct iser_data_buf *mem, 440 struct iser_reg_resources *rsc, 441 struct iser_mem_reg *reg) 442 { 443 struct iser_tx_desc *tx_desc = &iser_task->desc; 444 struct ib_cqe *cqe = &iser_task->iser_conn->ib_conn.reg_cqe; 445 struct ib_mr *mr = rsc->mr; 446 struct ib_reg_wr *wr; 447 int n; 448 449 if (rsc->mr_valid) 450 iser_inv_rkey(iser_tx_next_wr(tx_desc), mr, cqe); 451 452 ib_update_fast_reg_key(mr, ib_inc_rkey(mr->rkey)); 453 454 n = ib_map_mr_sg(mr, mem->sg, mem->size, NULL, SIZE_4K); 455 if (unlikely(n != mem->size)) { 456 iser_err("failed to map sg (%d/%d)\n", 457 n, mem->size); 458 return n < 0 ? n : -EINVAL; 459 } 460 461 wr = container_of(iser_tx_next_wr(tx_desc), struct ib_reg_wr, wr); 462 wr->wr.opcode = IB_WR_REG_MR; 463 wr->wr.wr_cqe = cqe; 464 wr->wr.send_flags = 0; 465 wr->wr.num_sge = 0; 466 wr->mr = mr; 467 wr->key = mr->rkey; 468 wr->access = IB_ACCESS_LOCAL_WRITE | 469 IB_ACCESS_REMOTE_WRITE | 470 IB_ACCESS_REMOTE_READ; 471 472 rsc->mr_valid = 1; 473 474 reg->sge.lkey = mr->lkey; 475 reg->rkey = mr->rkey; 476 reg->sge.addr = mr->iova; 477 reg->sge.length = mr->length; 478 479 iser_dbg("lkey=0x%x rkey=0x%x addr=0x%llx length=0x%x\n", 480 reg->sge.lkey, reg->rkey, reg->sge.addr, reg->sge.length); 481 482 return 0; 483 } 484 485 static int 486 iser_reg_prot_sg(struct iscsi_iser_task *task, 487 struct iser_data_buf *mem, 488 struct iser_fr_desc *desc, 489 bool use_dma_key, 490 struct iser_mem_reg *reg) 491 { 492 struct iser_device *device = task->iser_conn->ib_conn.device; 493 494 if (use_dma_key) 495 return iser_reg_dma(device, mem, reg); 496 497 return device->reg_ops->reg_mem(task, mem, &desc->pi_ctx->rsc, reg); 498 } 499 500 static int 501 iser_reg_data_sg(struct iscsi_iser_task *task, 502 struct iser_data_buf *mem, 503 struct iser_fr_desc *desc, 504 bool use_dma_key, 505 struct iser_mem_reg *reg) 506 { 507 struct iser_device *device = task->iser_conn->ib_conn.device; 508 509 if (use_dma_key) 510 return iser_reg_dma(device, mem, reg); 511 512 return device->reg_ops->reg_mem(task, mem, &desc->rsc, reg); 513 } 514 515 int iser_reg_rdma_mem(struct iscsi_iser_task *task, 516 enum iser_data_dir dir, 517 bool all_imm) 518 { 519 struct ib_conn *ib_conn = &task->iser_conn->ib_conn; 520 struct iser_device *device = ib_conn->device; 521 struct iser_data_buf *mem = &task->data[dir]; 522 struct iser_mem_reg *reg = &task->rdma_reg[dir]; 523 struct iser_mem_reg *data_reg; 524 struct iser_fr_desc *desc = NULL; 525 bool use_dma_key; 526 int err; 527 528 use_dma_key = mem->dma_nents == 1 && (all_imm || !iser_always_reg) && 529 scsi_get_prot_op(task->sc) == SCSI_PROT_NORMAL; 530 531 if (!use_dma_key) { 532 desc = device->reg_ops->reg_desc_get(ib_conn); 533 reg->mem_h = desc; 534 } 535 536 if (scsi_get_prot_op(task->sc) == SCSI_PROT_NORMAL) 537 data_reg = reg; 538 else 539 data_reg = &task->desc.data_reg; 540 541 err = iser_reg_data_sg(task, mem, desc, use_dma_key, data_reg); 542 if (unlikely(err)) 543 goto err_reg; 544 545 if (scsi_get_prot_op(task->sc) != SCSI_PROT_NORMAL) { 546 struct iser_mem_reg *prot_reg = &task->desc.prot_reg; 547 548 if (scsi_prot_sg_count(task->sc)) { 549 mem = &task->prot[dir]; 550 err = iser_reg_prot_sg(task, mem, desc, 551 use_dma_key, prot_reg); 552 if (unlikely(err)) 553 goto err_reg; 554 } 555 556 err = iser_reg_sig_mr(task, desc->pi_ctx, data_reg, 557 prot_reg, reg); 558 if (unlikely(err)) 559 goto err_reg; 560 561 desc->pi_ctx->sig_protected = 1; 562 } 563 564 return 0; 565 566 err_reg: 567 if (desc) 568 device->reg_ops->reg_desc_put(ib_conn, desc); 569 570 return err; 571 } 572 573 void iser_unreg_rdma_mem(struct iscsi_iser_task *task, 574 enum iser_data_dir dir) 575 { 576 struct iser_device *device = task->iser_conn->ib_conn.device; 577 578 device->reg_ops->unreg_mem(task, dir); 579 } 580