iser_verbs.c revision e657571b76faf96a1de1aaf40b2111adcf76c673
1/* 2 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved. 3 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved. 4 * Copyright (c) 2013 Mellanox Technologies. All rights reserved. 5 * 6 * This software is available to you under a choice of one of two 7 * licenses. You may choose to be licensed under the terms of the GNU 8 * General Public License (GPL) Version 2, available from the file 9 * COPYING in the main directory of this source tree, or the 10 * OpenIB.org BSD license below: 11 * 12 * Redistribution and use in source and binary forms, with or 13 * without modification, are permitted provided that the following 14 * conditions are met: 15 * 16 * - Redistributions of source code must retain the above 17 * copyright notice, this list of conditions and the following 18 * disclaimer. 19 * 20 * - Redistributions in binary form must reproduce the above 21 * copyright notice, this list of conditions and the following 22 * disclaimer in the documentation and/or other materials 23 * provided with the distribution. 24 * 25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 32 * SOFTWARE. 33 */ 34#include <linux/kernel.h> 35#include <linux/module.h> 36#include <linux/slab.h> 37#include <linux/delay.h> 38 39#include "iscsi_iser.h" 40 41#define ISCSI_ISER_MAX_CONN 8 42#define ISER_MAX_RX_CQ_LEN (ISER_QP_MAX_RECV_DTOS * ISCSI_ISER_MAX_CONN) 43#define ISER_MAX_TX_CQ_LEN (ISER_QP_MAX_REQ_DTOS * ISCSI_ISER_MAX_CONN) 44 45static void iser_cq_tasklet_fn(unsigned long data); 46static void iser_cq_callback(struct ib_cq *cq, void *cq_context); 47 48static void iser_cq_event_callback(struct ib_event *cause, void *context) 49{ 50 iser_err("got cq event %d \n", cause->event); 51} 52 53static void iser_qp_event_callback(struct ib_event *cause, void *context) 54{ 55 iser_err("got qp event %d\n",cause->event); 56} 57 58static void iser_event_handler(struct ib_event_handler *handler, 59 struct ib_event *event) 60{ 61 iser_err("async event %d on device %s port %d\n", event->event, 62 event->device->name, event->element.port_num); 63} 64 65/** 66 * iser_create_device_ib_res - creates Protection Domain (PD), Completion 67 * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with 68 * the adapator. 69 * 70 * returns 0 on success, -1 on failure 71 */ 72static int iser_create_device_ib_res(struct iser_device *device) 73{ 74 int i, j; 75 struct iser_cq_desc *cq_desc; 76 77 /* Assign function handles */ 78 device->iser_alloc_rdma_reg_res = iser_create_fmr_pool; 79 device->iser_free_rdma_reg_res = iser_free_fmr_pool; 80 device->iser_reg_rdma_mem = iser_reg_rdma_mem_fmr; 81 device->iser_unreg_rdma_mem = iser_unreg_mem_fmr; 82 83 device->cqs_used = min(ISER_MAX_CQ, device->ib_device->num_comp_vectors); 84 iser_info("using %d CQs, device %s supports %d vectors\n", 85 device->cqs_used, device->ib_device->name, 86 device->ib_device->num_comp_vectors); 87 88 device->cq_desc = kmalloc(sizeof(struct iser_cq_desc) * device->cqs_used, 89 GFP_KERNEL); 90 if (device->cq_desc == NULL) 91 goto cq_desc_err; 92 cq_desc = device->cq_desc; 93 94 device->pd = ib_alloc_pd(device->ib_device); 95 if (IS_ERR(device->pd)) 96 goto pd_err; 97 98 for (i = 0; i < device->cqs_used; i++) { 99 cq_desc[i].device = device; 100 cq_desc[i].cq_index = i; 101 102 device->rx_cq[i] = ib_create_cq(device->ib_device, 103 iser_cq_callback, 104 iser_cq_event_callback, 105 (void *)&cq_desc[i], 106 ISER_MAX_RX_CQ_LEN, i); 107 if (IS_ERR(device->rx_cq[i])) 108 goto cq_err; 109 110 device->tx_cq[i] = ib_create_cq(device->ib_device, 111 NULL, iser_cq_event_callback, 112 (void *)&cq_desc[i], 113 ISER_MAX_TX_CQ_LEN, i); 114 115 if (IS_ERR(device->tx_cq[i])) 116 goto cq_err; 117 118 if (ib_req_notify_cq(device->rx_cq[i], IB_CQ_NEXT_COMP)) 119 goto cq_err; 120 121 tasklet_init(&device->cq_tasklet[i], 122 iser_cq_tasklet_fn, 123 (unsigned long)&cq_desc[i]); 124 } 125 126 device->mr = ib_get_dma_mr(device->pd, IB_ACCESS_LOCAL_WRITE | 127 IB_ACCESS_REMOTE_WRITE | 128 IB_ACCESS_REMOTE_READ); 129 if (IS_ERR(device->mr)) 130 goto dma_mr_err; 131 132 INIT_IB_EVENT_HANDLER(&device->event_handler, device->ib_device, 133 iser_event_handler); 134 if (ib_register_event_handler(&device->event_handler)) 135 goto handler_err; 136 137 return 0; 138 139handler_err: 140 ib_dereg_mr(device->mr); 141dma_mr_err: 142 for (j = 0; j < device->cqs_used; j++) 143 tasklet_kill(&device->cq_tasklet[j]); 144cq_err: 145 for (j = 0; j < i; j++) { 146 if (device->tx_cq[j]) 147 ib_destroy_cq(device->tx_cq[j]); 148 if (device->rx_cq[j]) 149 ib_destroy_cq(device->rx_cq[j]); 150 } 151 ib_dealloc_pd(device->pd); 152pd_err: 153 kfree(device->cq_desc); 154cq_desc_err: 155 iser_err("failed to allocate an IB resource\n"); 156 return -1; 157} 158 159/** 160 * iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR, 161 * CQ and PD created with the device associated with the adapator. 162 */ 163static void iser_free_device_ib_res(struct iser_device *device) 164{ 165 int i; 166 BUG_ON(device->mr == NULL); 167 168 for (i = 0; i < device->cqs_used; i++) { 169 tasklet_kill(&device->cq_tasklet[i]); 170 (void)ib_destroy_cq(device->tx_cq[i]); 171 (void)ib_destroy_cq(device->rx_cq[i]); 172 device->tx_cq[i] = NULL; 173 device->rx_cq[i] = NULL; 174 } 175 176 (void)ib_unregister_event_handler(&device->event_handler); 177 (void)ib_dereg_mr(device->mr); 178 (void)ib_dealloc_pd(device->pd); 179 180 kfree(device->cq_desc); 181 182 device->mr = NULL; 183 device->pd = NULL; 184} 185 186/** 187 * iser_create_fmr_pool - Creates FMR pool and page_vector 188 * 189 * returns 0 on success, or errno code on failure 190 */ 191int iser_create_fmr_pool(struct iser_conn *ib_conn, unsigned cmds_max) 192{ 193 struct iser_device *device = ib_conn->device; 194 struct ib_fmr_pool_param params; 195 int ret = -ENOMEM; 196 197 ib_conn->fastreg.fmr.page_vec = kmalloc(sizeof(struct iser_page_vec) + 198 (sizeof(u64)*(ISCSI_ISER_SG_TABLESIZE + 1)), 199 GFP_KERNEL); 200 if (!ib_conn->fastreg.fmr.page_vec) 201 return ret; 202 203 ib_conn->fastreg.fmr.page_vec->pages = (u64 *)(ib_conn->fastreg.fmr.page_vec + 1); 204 205 params.page_shift = SHIFT_4K; 206 /* when the first/last SG element are not start/end * 207 * page aligned, the map whould be of N+1 pages */ 208 params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1; 209 /* make the pool size twice the max number of SCSI commands * 210 * the ML is expected to queue, watermark for unmap at 50% */ 211 params.pool_size = cmds_max * 2; 212 params.dirty_watermark = cmds_max; 213 params.cache = 0; 214 params.flush_function = NULL; 215 params.access = (IB_ACCESS_LOCAL_WRITE | 216 IB_ACCESS_REMOTE_WRITE | 217 IB_ACCESS_REMOTE_READ); 218 219 ib_conn->fastreg.fmr.pool = ib_create_fmr_pool(device->pd, ¶ms); 220 if (!IS_ERR(ib_conn->fastreg.fmr.pool)) 221 return 0; 222 223 /* no FMR => no need for page_vec */ 224 kfree(ib_conn->fastreg.fmr.page_vec); 225 ib_conn->fastreg.fmr.page_vec = NULL; 226 227 ret = PTR_ERR(ib_conn->fastreg.fmr.pool); 228 ib_conn->fastreg.fmr.pool = NULL; 229 if (ret != -ENOSYS) { 230 iser_err("FMR allocation failed, err %d\n", ret); 231 return ret; 232 } else { 233 iser_warn("FMRs are not supported, using unaligned mode\n"); 234 return 0; 235 } 236} 237 238/** 239 * iser_free_fmr_pool - releases the FMR pool and page vec 240 */ 241void iser_free_fmr_pool(struct iser_conn *ib_conn) 242{ 243 iser_info("freeing conn %p fmr pool %p\n", 244 ib_conn, ib_conn->fastreg.fmr.pool); 245 246 if (ib_conn->fastreg.fmr.pool != NULL) 247 ib_destroy_fmr_pool(ib_conn->fastreg.fmr.pool); 248 249 ib_conn->fastreg.fmr.pool = NULL; 250 251 kfree(ib_conn->fastreg.fmr.page_vec); 252 ib_conn->fastreg.fmr.page_vec = NULL; 253} 254 255/** 256 * iser_create_ib_conn_res - Queue-Pair (QP) 257 * 258 * returns 0 on success, -1 on failure 259 */ 260static int iser_create_ib_conn_res(struct iser_conn *ib_conn) 261{ 262 struct iser_device *device; 263 struct ib_qp_init_attr init_attr; 264 int ret = -ENOMEM; 265 int index, min_index = 0; 266 267 BUG_ON(ib_conn->device == NULL); 268 269 device = ib_conn->device; 270 271 memset(&init_attr, 0, sizeof init_attr); 272 273 mutex_lock(&ig.connlist_mutex); 274 /* select the CQ with the minimal number of usages */ 275 for (index = 0; index < device->cqs_used; index++) 276 if (device->cq_active_qps[index] < 277 device->cq_active_qps[min_index]) 278 min_index = index; 279 device->cq_active_qps[min_index]++; 280 mutex_unlock(&ig.connlist_mutex); 281 iser_info("cq index %d used for ib_conn %p\n", min_index, ib_conn); 282 283 init_attr.event_handler = iser_qp_event_callback; 284 init_attr.qp_context = (void *)ib_conn; 285 init_attr.send_cq = device->tx_cq[min_index]; 286 init_attr.recv_cq = device->rx_cq[min_index]; 287 init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS; 288 init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS; 289 init_attr.cap.max_send_sge = 2; 290 init_attr.cap.max_recv_sge = 1; 291 init_attr.sq_sig_type = IB_SIGNAL_REQ_WR; 292 init_attr.qp_type = IB_QPT_RC; 293 294 ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr); 295 if (ret) 296 goto out_err; 297 298 ib_conn->qp = ib_conn->cma_id->qp; 299 iser_info("setting conn %p cma_id %p qp %p\n", 300 ib_conn, ib_conn->cma_id, 301 ib_conn->cma_id->qp); 302 return ret; 303 304out_err: 305 iser_err("unable to alloc mem or create resource, err %d\n", ret); 306 return ret; 307} 308 309/** 310 * releases the QP objects, returns 0 on success, 311 * -1 on failure 312 */ 313static int iser_free_ib_conn_res(struct iser_conn *ib_conn) 314{ 315 int cq_index; 316 BUG_ON(ib_conn == NULL); 317 318 iser_info("freeing conn %p cma_id %p qp %p\n", 319 ib_conn, ib_conn->cma_id, 320 ib_conn->qp); 321 322 /* qp is created only once both addr & route are resolved */ 323 324 if (ib_conn->qp != NULL) { 325 cq_index = ((struct iser_cq_desc *)ib_conn->qp->recv_cq->cq_context)->cq_index; 326 ib_conn->device->cq_active_qps[cq_index]--; 327 328 rdma_destroy_qp(ib_conn->cma_id); 329 } 330 331 ib_conn->qp = NULL; 332 333 return 0; 334} 335 336/** 337 * based on the resolved device node GUID see if there already allocated 338 * device for this device. If there's no such, create one. 339 */ 340static 341struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id) 342{ 343 struct iser_device *device; 344 345 mutex_lock(&ig.device_list_mutex); 346 347 list_for_each_entry(device, &ig.device_list, ig_list) 348 /* find if there's a match using the node GUID */ 349 if (device->ib_device->node_guid == cma_id->device->node_guid) 350 goto inc_refcnt; 351 352 device = kzalloc(sizeof *device, GFP_KERNEL); 353 if (device == NULL) 354 goto out; 355 356 /* assign this device to the device */ 357 device->ib_device = cma_id->device; 358 /* init the device and link it into ig device list */ 359 if (iser_create_device_ib_res(device)) { 360 kfree(device); 361 device = NULL; 362 goto out; 363 } 364 list_add(&device->ig_list, &ig.device_list); 365 366inc_refcnt: 367 device->refcount++; 368out: 369 mutex_unlock(&ig.device_list_mutex); 370 return device; 371} 372 373/* if there's no demand for this device, release it */ 374static void iser_device_try_release(struct iser_device *device) 375{ 376 mutex_lock(&ig.device_list_mutex); 377 device->refcount--; 378 iser_info("device %p refcount %d\n", device, device->refcount); 379 if (!device->refcount) { 380 iser_free_device_ib_res(device); 381 list_del(&device->ig_list); 382 kfree(device); 383 } 384 mutex_unlock(&ig.device_list_mutex); 385} 386 387static int iser_conn_state_comp_exch(struct iser_conn *ib_conn, 388 enum iser_ib_conn_state comp, 389 enum iser_ib_conn_state exch) 390{ 391 int ret; 392 393 spin_lock_bh(&ib_conn->lock); 394 if ((ret = (ib_conn->state == comp))) 395 ib_conn->state = exch; 396 spin_unlock_bh(&ib_conn->lock); 397 return ret; 398} 399 400/** 401 * Frees all conn objects and deallocs conn descriptor 402 */ 403static void iser_conn_release(struct iser_conn *ib_conn, int can_destroy_id) 404{ 405 struct iser_device *device = ib_conn->device; 406 407 BUG_ON(ib_conn->state != ISER_CONN_DOWN); 408 409 mutex_lock(&ig.connlist_mutex); 410 list_del(&ib_conn->conn_list); 411 mutex_unlock(&ig.connlist_mutex); 412 iser_free_rx_descriptors(ib_conn); 413 iser_free_ib_conn_res(ib_conn); 414 ib_conn->device = NULL; 415 /* on EVENT_ADDR_ERROR there's no device yet for this conn */ 416 if (device != NULL) 417 iser_device_try_release(device); 418 /* if cma handler context, the caller actually destroy the id */ 419 if (ib_conn->cma_id != NULL && can_destroy_id) { 420 rdma_destroy_id(ib_conn->cma_id); 421 ib_conn->cma_id = NULL; 422 } 423 iscsi_destroy_endpoint(ib_conn->ep); 424} 425 426void iser_conn_get(struct iser_conn *ib_conn) 427{ 428 atomic_inc(&ib_conn->refcount); 429} 430 431int iser_conn_put(struct iser_conn *ib_conn, int can_destroy_id) 432{ 433 if (atomic_dec_and_test(&ib_conn->refcount)) { 434 iser_conn_release(ib_conn, can_destroy_id); 435 return 1; 436 } 437 return 0; 438} 439 440/** 441 * triggers start of the disconnect procedures and wait for them to be done 442 */ 443void iser_conn_terminate(struct iser_conn *ib_conn) 444{ 445 int err = 0; 446 447 /* change the ib conn state only if the conn is UP, however always call 448 * rdma_disconnect since this is the only way to cause the CMA to change 449 * the QP state to ERROR 450 */ 451 452 iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP, ISER_CONN_TERMINATING); 453 err = rdma_disconnect(ib_conn->cma_id); 454 if (err) 455 iser_err("Failed to disconnect, conn: 0x%p err %d\n", 456 ib_conn,err); 457 458 wait_event_interruptible(ib_conn->wait, 459 ib_conn->state == ISER_CONN_DOWN); 460 461 iser_conn_put(ib_conn, 1); /* deref ib conn deallocate */ 462} 463 464static int iser_connect_error(struct rdma_cm_id *cma_id) 465{ 466 struct iser_conn *ib_conn; 467 ib_conn = (struct iser_conn *)cma_id->context; 468 469 ib_conn->state = ISER_CONN_DOWN; 470 wake_up_interruptible(&ib_conn->wait); 471 return iser_conn_put(ib_conn, 0); /* deref ib conn's cma id */ 472} 473 474static int iser_addr_handler(struct rdma_cm_id *cma_id) 475{ 476 struct iser_device *device; 477 struct iser_conn *ib_conn; 478 int ret; 479 480 device = iser_device_find_by_ib_device(cma_id); 481 if (!device) { 482 iser_err("device lookup/creation failed\n"); 483 return iser_connect_error(cma_id); 484 } 485 486 ib_conn = (struct iser_conn *)cma_id->context; 487 ib_conn->device = device; 488 489 ret = rdma_resolve_route(cma_id, 1000); 490 if (ret) { 491 iser_err("resolve route failed: %d\n", ret); 492 return iser_connect_error(cma_id); 493 } 494 495 return 0; 496} 497 498static int iser_route_handler(struct rdma_cm_id *cma_id) 499{ 500 struct rdma_conn_param conn_param; 501 int ret; 502 struct iser_cm_hdr req_hdr; 503 504 ret = iser_create_ib_conn_res((struct iser_conn *)cma_id->context); 505 if (ret) 506 goto failure; 507 508 memset(&conn_param, 0, sizeof conn_param); 509 conn_param.responder_resources = 4; 510 conn_param.initiator_depth = 1; 511 conn_param.retry_count = 7; 512 conn_param.rnr_retry_count = 6; 513 514 memset(&req_hdr, 0, sizeof(req_hdr)); 515 req_hdr.flags = (ISER_ZBVA_NOT_SUPPORTED | 516 ISER_SEND_W_INV_NOT_SUPPORTED); 517 conn_param.private_data = (void *)&req_hdr; 518 conn_param.private_data_len = sizeof(struct iser_cm_hdr); 519 520 ret = rdma_connect(cma_id, &conn_param); 521 if (ret) { 522 iser_err("failure connecting: %d\n", ret); 523 goto failure; 524 } 525 526 return 0; 527failure: 528 return iser_connect_error(cma_id); 529} 530 531static void iser_connected_handler(struct rdma_cm_id *cma_id) 532{ 533 struct iser_conn *ib_conn; 534 535 ib_conn = (struct iser_conn *)cma_id->context; 536 ib_conn->state = ISER_CONN_UP; 537 wake_up_interruptible(&ib_conn->wait); 538} 539 540static int iser_disconnected_handler(struct rdma_cm_id *cma_id) 541{ 542 struct iser_conn *ib_conn; 543 int ret; 544 545 ib_conn = (struct iser_conn *)cma_id->context; 546 547 /* getting here when the state is UP means that the conn is being * 548 * terminated asynchronously from the iSCSI layer's perspective. */ 549 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP, 550 ISER_CONN_TERMINATING)) 551 iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn, 552 ISCSI_ERR_CONN_FAILED); 553 554 /* Complete the termination process if no posts are pending */ 555 if (ib_conn->post_recv_buf_count == 0 && 556 (atomic_read(&ib_conn->post_send_buf_count) == 0)) { 557 ib_conn->state = ISER_CONN_DOWN; 558 wake_up_interruptible(&ib_conn->wait); 559 } 560 561 ret = iser_conn_put(ib_conn, 0); /* deref ib conn's cma id */ 562 return ret; 563} 564 565static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event) 566{ 567 int ret = 0; 568 569 iser_info("event %d status %d conn %p id %p\n", 570 event->event, event->status, cma_id->context, cma_id); 571 572 switch (event->event) { 573 case RDMA_CM_EVENT_ADDR_RESOLVED: 574 ret = iser_addr_handler(cma_id); 575 break; 576 case RDMA_CM_EVENT_ROUTE_RESOLVED: 577 ret = iser_route_handler(cma_id); 578 break; 579 case RDMA_CM_EVENT_ESTABLISHED: 580 iser_connected_handler(cma_id); 581 break; 582 case RDMA_CM_EVENT_ADDR_ERROR: 583 case RDMA_CM_EVENT_ROUTE_ERROR: 584 case RDMA_CM_EVENT_CONNECT_ERROR: 585 case RDMA_CM_EVENT_UNREACHABLE: 586 case RDMA_CM_EVENT_REJECTED: 587 ret = iser_connect_error(cma_id); 588 break; 589 case RDMA_CM_EVENT_DISCONNECTED: 590 case RDMA_CM_EVENT_DEVICE_REMOVAL: 591 case RDMA_CM_EVENT_ADDR_CHANGE: 592 ret = iser_disconnected_handler(cma_id); 593 break; 594 default: 595 iser_err("Unexpected RDMA CM event (%d)\n", event->event); 596 break; 597 } 598 return ret; 599} 600 601void iser_conn_init(struct iser_conn *ib_conn) 602{ 603 ib_conn->state = ISER_CONN_INIT; 604 init_waitqueue_head(&ib_conn->wait); 605 ib_conn->post_recv_buf_count = 0; 606 atomic_set(&ib_conn->post_send_buf_count, 0); 607 atomic_set(&ib_conn->refcount, 1); /* ref ib conn allocation */ 608 INIT_LIST_HEAD(&ib_conn->conn_list); 609 spin_lock_init(&ib_conn->lock); 610} 611 612 /** 613 * starts the process of connecting to the target 614 * sleeps until the connection is established or rejected 615 */ 616int iser_connect(struct iser_conn *ib_conn, 617 struct sockaddr_in *src_addr, 618 struct sockaddr_in *dst_addr, 619 int non_blocking) 620{ 621 struct sockaddr *src, *dst; 622 int err = 0; 623 624 sprintf(ib_conn->name, "%pI4:%d", 625 &dst_addr->sin_addr.s_addr, dst_addr->sin_port); 626 627 /* the device is known only --after-- address resolution */ 628 ib_conn->device = NULL; 629 630 iser_info("connecting to: %pI4, port 0x%x\n", 631 &dst_addr->sin_addr, dst_addr->sin_port); 632 633 ib_conn->state = ISER_CONN_PENDING; 634 635 iser_conn_get(ib_conn); /* ref ib conn's cma id */ 636 ib_conn->cma_id = rdma_create_id(iser_cma_handler, 637 (void *)ib_conn, 638 RDMA_PS_TCP, IB_QPT_RC); 639 if (IS_ERR(ib_conn->cma_id)) { 640 err = PTR_ERR(ib_conn->cma_id); 641 iser_err("rdma_create_id failed: %d\n", err); 642 goto id_failure; 643 } 644 645 src = (struct sockaddr *)src_addr; 646 dst = (struct sockaddr *)dst_addr; 647 err = rdma_resolve_addr(ib_conn->cma_id, src, dst, 1000); 648 if (err) { 649 iser_err("rdma_resolve_addr failed: %d\n", err); 650 goto addr_failure; 651 } 652 653 if (!non_blocking) { 654 wait_event_interruptible(ib_conn->wait, 655 (ib_conn->state != ISER_CONN_PENDING)); 656 657 if (ib_conn->state != ISER_CONN_UP) { 658 err = -EIO; 659 goto connect_failure; 660 } 661 } 662 663 mutex_lock(&ig.connlist_mutex); 664 list_add(&ib_conn->conn_list, &ig.connlist); 665 mutex_unlock(&ig.connlist_mutex); 666 return 0; 667 668id_failure: 669 ib_conn->cma_id = NULL; 670addr_failure: 671 ib_conn->state = ISER_CONN_DOWN; 672 iser_conn_put(ib_conn, 1); /* deref ib conn's cma id */ 673connect_failure: 674 iser_conn_put(ib_conn, 1); /* deref ib conn deallocate */ 675 return err; 676} 677 678/** 679 * iser_reg_page_vec - Register physical memory 680 * 681 * returns: 0 on success, errno code on failure 682 */ 683int iser_reg_page_vec(struct iser_conn *ib_conn, 684 struct iser_page_vec *page_vec, 685 struct iser_mem_reg *mem_reg) 686{ 687 struct ib_pool_fmr *mem; 688 u64 io_addr; 689 u64 *page_list; 690 int status; 691 692 page_list = page_vec->pages; 693 io_addr = page_list[0]; 694 695 mem = ib_fmr_pool_map_phys(ib_conn->fastreg.fmr.pool, 696 page_list, 697 page_vec->length, 698 io_addr); 699 700 if (IS_ERR(mem)) { 701 status = (int)PTR_ERR(mem); 702 iser_err("ib_fmr_pool_map_phys failed: %d\n", status); 703 return status; 704 } 705 706 mem_reg->lkey = mem->fmr->lkey; 707 mem_reg->rkey = mem->fmr->rkey; 708 mem_reg->len = page_vec->length * SIZE_4K; 709 mem_reg->va = io_addr; 710 mem_reg->is_fmr = 1; 711 mem_reg->mem_h = (void *)mem; 712 713 mem_reg->va += page_vec->offset; 714 mem_reg->len = page_vec->data_size; 715 716 iser_dbg("PHYSICAL Mem.register, [PHYS p_array: 0x%p, sz: %d, " 717 "entry[0]: (0x%08lx,%ld)] -> " 718 "[lkey: 0x%08X mem_h: 0x%p va: 0x%08lX sz: %ld]\n", 719 page_vec, page_vec->length, 720 (unsigned long)page_vec->pages[0], 721 (unsigned long)page_vec->data_size, 722 (unsigned int)mem_reg->lkey, mem_reg->mem_h, 723 (unsigned long)mem_reg->va, (unsigned long)mem_reg->len); 724 return 0; 725} 726 727/** 728 * Unregister (previosuly registered using FMR) memory. 729 * If memory is non-FMR does nothing. 730 */ 731void iser_unreg_mem_fmr(struct iscsi_iser_task *iser_task, 732 enum iser_data_dir cmd_dir) 733{ 734 struct iser_mem_reg *reg = &iser_task->rdma_regd[cmd_dir].reg; 735 int ret; 736 737 if (!reg->is_fmr) 738 return; 739 740 iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n",reg->mem_h); 741 742 ret = ib_fmr_pool_unmap((struct ib_pool_fmr *)reg->mem_h); 743 if (ret) 744 iser_err("ib_fmr_pool_unmap failed %d\n", ret); 745 746 reg->mem_h = NULL; 747} 748 749int iser_post_recvl(struct iser_conn *ib_conn) 750{ 751 struct ib_recv_wr rx_wr, *rx_wr_failed; 752 struct ib_sge sge; 753 int ib_ret; 754 755 sge.addr = ib_conn->login_resp_dma; 756 sge.length = ISER_RX_LOGIN_SIZE; 757 sge.lkey = ib_conn->device->mr->lkey; 758 759 rx_wr.wr_id = (unsigned long)ib_conn->login_resp_buf; 760 rx_wr.sg_list = &sge; 761 rx_wr.num_sge = 1; 762 rx_wr.next = NULL; 763 764 ib_conn->post_recv_buf_count++; 765 ib_ret = ib_post_recv(ib_conn->qp, &rx_wr, &rx_wr_failed); 766 if (ib_ret) { 767 iser_err("ib_post_recv failed ret=%d\n", ib_ret); 768 ib_conn->post_recv_buf_count--; 769 } 770 return ib_ret; 771} 772 773int iser_post_recvm(struct iser_conn *ib_conn, int count) 774{ 775 struct ib_recv_wr *rx_wr, *rx_wr_failed; 776 int i, ib_ret; 777 unsigned int my_rx_head = ib_conn->rx_desc_head; 778 struct iser_rx_desc *rx_desc; 779 780 for (rx_wr = ib_conn->rx_wr, i = 0; i < count; i++, rx_wr++) { 781 rx_desc = &ib_conn->rx_descs[my_rx_head]; 782 rx_wr->wr_id = (unsigned long)rx_desc; 783 rx_wr->sg_list = &rx_desc->rx_sg; 784 rx_wr->num_sge = 1; 785 rx_wr->next = rx_wr + 1; 786 my_rx_head = (my_rx_head + 1) & ib_conn->qp_max_recv_dtos_mask; 787 } 788 789 rx_wr--; 790 rx_wr->next = NULL; /* mark end of work requests list */ 791 792 ib_conn->post_recv_buf_count += count; 793 ib_ret = ib_post_recv(ib_conn->qp, ib_conn->rx_wr, &rx_wr_failed); 794 if (ib_ret) { 795 iser_err("ib_post_recv failed ret=%d\n", ib_ret); 796 ib_conn->post_recv_buf_count -= count; 797 } else 798 ib_conn->rx_desc_head = my_rx_head; 799 return ib_ret; 800} 801 802 803/** 804 * iser_start_send - Initiate a Send DTO operation 805 * 806 * returns 0 on success, -1 on failure 807 */ 808int iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc) 809{ 810 int ib_ret; 811 struct ib_send_wr send_wr, *send_wr_failed; 812 813 ib_dma_sync_single_for_device(ib_conn->device->ib_device, 814 tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE); 815 816 send_wr.next = NULL; 817 send_wr.wr_id = (unsigned long)tx_desc; 818 send_wr.sg_list = tx_desc->tx_sg; 819 send_wr.num_sge = tx_desc->num_sge; 820 send_wr.opcode = IB_WR_SEND; 821 send_wr.send_flags = IB_SEND_SIGNALED; 822 823 atomic_inc(&ib_conn->post_send_buf_count); 824 825 ib_ret = ib_post_send(ib_conn->qp, &send_wr, &send_wr_failed); 826 if (ib_ret) { 827 iser_err("ib_post_send failed, ret:%d\n", ib_ret); 828 atomic_dec(&ib_conn->post_send_buf_count); 829 } 830 return ib_ret; 831} 832 833static void iser_handle_comp_error(struct iser_tx_desc *desc, 834 struct iser_conn *ib_conn) 835{ 836 if (desc && desc->type == ISCSI_TX_DATAOUT) 837 kmem_cache_free(ig.desc_cache, desc); 838 839 if (ib_conn->post_recv_buf_count == 0 && 840 atomic_read(&ib_conn->post_send_buf_count) == 0) { 841 /* getting here when the state is UP means that the conn is * 842 * being terminated asynchronously from the iSCSI layer's * 843 * perspective. */ 844 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP, 845 ISER_CONN_TERMINATING)) 846 iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn, 847 ISCSI_ERR_CONN_FAILED); 848 849 /* no more non completed posts to the QP, complete the 850 * termination process w.o worrying on disconnect event */ 851 ib_conn->state = ISER_CONN_DOWN; 852 wake_up_interruptible(&ib_conn->wait); 853 } 854} 855 856static int iser_drain_tx_cq(struct iser_device *device, int cq_index) 857{ 858 struct ib_cq *cq = device->tx_cq[cq_index]; 859 struct ib_wc wc; 860 struct iser_tx_desc *tx_desc; 861 struct iser_conn *ib_conn; 862 int completed_tx = 0; 863 864 while (ib_poll_cq(cq, 1, &wc) == 1) { 865 tx_desc = (struct iser_tx_desc *) (unsigned long) wc.wr_id; 866 ib_conn = wc.qp->qp_context; 867 if (wc.status == IB_WC_SUCCESS) { 868 if (wc.opcode == IB_WC_SEND) 869 iser_snd_completion(tx_desc, ib_conn); 870 else 871 iser_err("expected opcode %d got %d\n", 872 IB_WC_SEND, wc.opcode); 873 } else { 874 iser_err("tx id %llx status %d vend_err %x\n", 875 wc.wr_id, wc.status, wc.vendor_err); 876 atomic_dec(&ib_conn->post_send_buf_count); 877 iser_handle_comp_error(tx_desc, ib_conn); 878 } 879 completed_tx++; 880 } 881 return completed_tx; 882} 883 884 885static void iser_cq_tasklet_fn(unsigned long data) 886{ 887 struct iser_cq_desc *cq_desc = (struct iser_cq_desc *)data; 888 struct iser_device *device = cq_desc->device; 889 int cq_index = cq_desc->cq_index; 890 struct ib_cq *cq = device->rx_cq[cq_index]; 891 struct ib_wc wc; 892 struct iser_rx_desc *desc; 893 unsigned long xfer_len; 894 struct iser_conn *ib_conn; 895 int completed_tx, completed_rx; 896 completed_tx = completed_rx = 0; 897 898 while (ib_poll_cq(cq, 1, &wc) == 1) { 899 desc = (struct iser_rx_desc *) (unsigned long) wc.wr_id; 900 BUG_ON(desc == NULL); 901 ib_conn = wc.qp->qp_context; 902 if (wc.status == IB_WC_SUCCESS) { 903 if (wc.opcode == IB_WC_RECV) { 904 xfer_len = (unsigned long)wc.byte_len; 905 iser_rcv_completion(desc, xfer_len, ib_conn); 906 } else 907 iser_err("expected opcode %d got %d\n", 908 IB_WC_RECV, wc.opcode); 909 } else { 910 if (wc.status != IB_WC_WR_FLUSH_ERR) 911 iser_err("rx id %llx status %d vend_err %x\n", 912 wc.wr_id, wc.status, wc.vendor_err); 913 ib_conn->post_recv_buf_count--; 914 iser_handle_comp_error(NULL, ib_conn); 915 } 916 completed_rx++; 917 if (!(completed_rx & 63)) 918 completed_tx += iser_drain_tx_cq(device, cq_index); 919 } 920 /* #warning "it is assumed here that arming CQ only once its empty" * 921 * " would not cause interrupts to be missed" */ 922 ib_req_notify_cq(cq, IB_CQ_NEXT_COMP); 923 924 completed_tx += iser_drain_tx_cq(device, cq_index); 925 iser_dbg("got %d rx %d tx completions\n", completed_rx, completed_tx); 926} 927 928static void iser_cq_callback(struct ib_cq *cq, void *cq_context) 929{ 930 struct iser_cq_desc *cq_desc = (struct iser_cq_desc *)cq_context; 931 struct iser_device *device = cq_desc->device; 932 int cq_index = cq_desc->cq_index; 933 934 tasklet_schedule(&device->cq_tasklet[cq_index]); 935} 936