1/* 2 * Copyright (c) 2012-2016 VMware, Inc. All rights reserved. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of EITHER the GNU General Public License 6 * version 2 as published by the Free Software Foundation or the BSD 7 * 2-Clause License. This program is distributed in the hope that it 8 * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED 9 * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. 10 * See the GNU General Public License version 2 for more details at 11 * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html. 12 * 13 * You should have received a copy of the GNU General Public License 14 * along with this program available in the file COPYING in the main 15 * directory of this source tree. 16 * 17 * The BSD 2-Clause License 18 * 19 * Redistribution and use in source and binary forms, with or 20 * without modification, are permitted provided that the following 21 * conditions are met: 22 * 23 * - Redistributions of source code must retain the above 24 * copyright notice, this list of conditions and the following 25 * disclaimer. 26 * 27 * - Redistributions in binary form must reproduce the above 28 * copyright notice, this list of conditions and the following 29 * disclaimer in the documentation and/or other materials 30 * provided with the distribution. 31 * 32 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 33 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 34 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 35 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 36 * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, 37 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 38 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 39 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 40 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 41 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 42 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED 43 * OF THE POSSIBILITY OF SUCH DAMAGE. 44 */ 45 46#include <asm/page.h> 47#include <linux/io.h> 48#include <linux/wait.h> 49#include <rdma/ib_addr.h> 50#include <rdma/ib_smi.h> 51#include <rdma/ib_user_verbs.h> 52 53#include "pvrdma.h" 54 55static void __pvrdma_destroy_qp(struct pvrdma_dev *dev, 56 struct pvrdma_qp *qp); 57 58static inline void get_cqs(struct pvrdma_qp *qp, struct pvrdma_cq **send_cq, 59 struct pvrdma_cq **recv_cq) 60{ 61 *send_cq = to_vcq(qp->ibqp.send_cq); 62 *recv_cq = to_vcq(qp->ibqp.recv_cq); 63} 64 65static void pvrdma_lock_cqs(struct pvrdma_cq *scq, struct pvrdma_cq *rcq, 66 unsigned long *scq_flags, 67 unsigned long *rcq_flags) 68 __acquires(scq->cq_lock) __acquires(rcq->cq_lock) 69{ 70 if (scq == rcq) { 71 spin_lock_irqsave(&scq->cq_lock, *scq_flags); 72 __acquire(rcq->cq_lock); 73 } else if (scq->cq_handle < rcq->cq_handle) { 74 spin_lock_irqsave(&scq->cq_lock, *scq_flags); 75 spin_lock_irqsave_nested(&rcq->cq_lock, *rcq_flags, 76 SINGLE_DEPTH_NESTING); 77 } else { 78 spin_lock_irqsave(&rcq->cq_lock, *rcq_flags); 79 spin_lock_irqsave_nested(&scq->cq_lock, *scq_flags, 80 SINGLE_DEPTH_NESTING); 81 } 82} 83 84static void pvrdma_unlock_cqs(struct pvrdma_cq *scq, struct pvrdma_cq *rcq, 85 unsigned long *scq_flags, 86 unsigned long *rcq_flags) 87 __releases(scq->cq_lock) __releases(rcq->cq_lock) 88{ 89 if (scq == rcq) { 90 __release(rcq->cq_lock); 91 spin_unlock_irqrestore(&scq->cq_lock, *scq_flags); 92 } else if (scq->cq_handle < rcq->cq_handle) { 93 spin_unlock_irqrestore(&rcq->cq_lock, *rcq_flags); 94 spin_unlock_irqrestore(&scq->cq_lock, *scq_flags); 95 } else { 96 spin_unlock_irqrestore(&scq->cq_lock, *scq_flags); 97 spin_unlock_irqrestore(&rcq->cq_lock, *rcq_flags); 98 } 99} 100 101static void pvrdma_reset_qp(struct pvrdma_qp *qp) 102{ 103 struct pvrdma_cq *scq, *rcq; 104 unsigned long scq_flags, rcq_flags; 105 106 /* Clean up cqes */ 107 get_cqs(qp, &scq, &rcq); 108 pvrdma_lock_cqs(scq, rcq, &scq_flags, &rcq_flags); 109 110 _pvrdma_flush_cqe(qp, scq); 111 if (scq != rcq) 112 _pvrdma_flush_cqe(qp, rcq); 113 114 pvrdma_unlock_cqs(scq, rcq, &scq_flags, &rcq_flags); 115 116 /* 117 * Reset queuepair. The checks are because usermode queuepairs won't 118 * have kernel ringstates. 119 */ 120 if (qp->rq.ring) { 121 atomic_set(&qp->rq.ring->cons_head, 0); 122 atomic_set(&qp->rq.ring->prod_tail, 0); 123 } 124 if (qp->sq.ring) { 125 atomic_set(&qp->sq.ring->cons_head, 0); 126 atomic_set(&qp->sq.ring->prod_tail, 0); 127 } 128} 129 130static int pvrdma_set_rq_size(struct pvrdma_dev *dev, 131 struct ib_qp_cap *req_cap, 132 struct pvrdma_qp *qp) 133{ 134 if (req_cap->max_recv_wr > dev->dsr->caps.max_qp_wr || 135 req_cap->max_recv_sge > dev->dsr->caps.max_sge) { 136 dev_warn(&dev->pdev->dev, "recv queue size invalid\n"); 137 return -EINVAL; 138 } 139 140 qp->rq.wqe_cnt = roundup_pow_of_two(max(1U, req_cap->max_recv_wr)); 141 qp->rq.max_sg = roundup_pow_of_two(max(1U, req_cap->max_recv_sge)); 142 143 /* Write back */ 144 req_cap->max_recv_wr = qp->rq.wqe_cnt; 145 req_cap->max_recv_sge = qp->rq.max_sg; 146 147 qp->rq.wqe_size = roundup_pow_of_two(sizeof(struct pvrdma_rq_wqe_hdr) + 148 sizeof(struct pvrdma_sge) * 149 qp->rq.max_sg); 150 qp->npages_recv = (qp->rq.wqe_cnt * qp->rq.wqe_size + PAGE_SIZE - 1) / 151 PAGE_SIZE; 152 153 return 0; 154} 155 156static int pvrdma_set_sq_size(struct pvrdma_dev *dev, struct ib_qp_cap *req_cap, 157 struct pvrdma_qp *qp) 158{ 159 if (req_cap->max_send_wr > dev->dsr->caps.max_qp_wr || 160 req_cap->max_send_sge > dev->dsr->caps.max_sge) { 161 dev_warn(&dev->pdev->dev, "send queue size invalid\n"); 162 return -EINVAL; 163 } 164 165 qp->sq.wqe_cnt = roundup_pow_of_two(max(1U, req_cap->max_send_wr)); 166 qp->sq.max_sg = roundup_pow_of_two(max(1U, req_cap->max_send_sge)); 167 168 /* Write back */ 169 req_cap->max_send_wr = qp->sq.wqe_cnt; 170 req_cap->max_send_sge = qp->sq.max_sg; 171 172 qp->sq.wqe_size = roundup_pow_of_two(sizeof(struct pvrdma_sq_wqe_hdr) + 173 sizeof(struct pvrdma_sge) * 174 qp->sq.max_sg); 175 /* Note: one extra page for the header. */ 176 qp->npages_send = PVRDMA_QP_NUM_HEADER_PAGES + 177 (qp->sq.wqe_cnt * qp->sq.wqe_size + PAGE_SIZE - 1) / 178 PAGE_SIZE; 179 180 return 0; 181} 182 183/** 184 * pvrdma_create_qp - create queue pair 185 * @ibqp: queue pair 186 * @init_attr: queue pair attributes 187 * @udata: user data 188 * 189 * @return: the 0 on success, otherwise returns an errno. 190 */ 191int pvrdma_create_qp(struct ib_qp *ibqp, struct ib_qp_init_attr *init_attr, 192 struct ib_udata *udata) 193{ 194 struct pvrdma_qp *qp = to_vqp(ibqp); 195 struct pvrdma_dev *dev = to_vdev(ibqp->device); 196 union pvrdma_cmd_req req; 197 union pvrdma_cmd_resp rsp; 198 struct pvrdma_cmd_create_qp *cmd = &req.create_qp; 199 struct pvrdma_cmd_create_qp_resp *resp = &rsp.create_qp_resp; 200 struct pvrdma_cmd_create_qp_resp_v2 *resp_v2 = &rsp.create_qp_resp_v2; 201 struct pvrdma_create_qp ucmd; 202 struct pvrdma_create_qp_resp qp_resp = {}; 203 unsigned long flags; 204 int ret; 205 bool is_srq = !!init_attr->srq; 206 207 if (init_attr->create_flags) { 208 dev_warn(&dev->pdev->dev, 209 "invalid create queuepair flags %#x\n", 210 init_attr->create_flags); 211 return -EOPNOTSUPP; 212 } 213 214 if (init_attr->qp_type != IB_QPT_RC && 215 init_attr->qp_type != IB_QPT_UD && 216 init_attr->qp_type != IB_QPT_GSI) { 217 dev_warn(&dev->pdev->dev, "queuepair type %d not supported\n", 218 init_attr->qp_type); 219 return -EOPNOTSUPP; 220 } 221 222 if (is_srq && !dev->dsr->caps.max_srq) { 223 dev_warn(&dev->pdev->dev, 224 "SRQs not supported by device\n"); 225 return -EINVAL; 226 } 227 228 if (!atomic_add_unless(&dev->num_qps, 1, dev->dsr->caps.max_qp)) 229 return -ENOMEM; 230 231 switch (init_attr->qp_type) { 232 case IB_QPT_GSI: 233 if (init_attr->port_num == 0 || 234 init_attr->port_num > ibqp->device->phys_port_cnt) { 235 dev_warn(&dev->pdev->dev, "invalid queuepair attrs\n"); 236 ret = -EINVAL; 237 goto err_qp; 238 } 239 fallthrough; 240 case IB_QPT_RC: 241 case IB_QPT_UD: 242 spin_lock_init(&qp->sq.lock); 243 spin_lock_init(&qp->rq.lock); 244 mutex_init(&qp->mutex); 245 refcount_set(&qp->refcnt, 1); 246 init_completion(&qp->free); 247 248 qp->state = IB_QPS_RESET; 249 qp->is_kernel = !udata; 250 251 if (!qp->is_kernel) { 252 dev_dbg(&dev->pdev->dev, 253 "create queuepair from user space\n"); 254 255 if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) { 256 ret = -EFAULT; 257 goto err_qp; 258 } 259 260 /* Userspace supports qpn and qp handles? */ 261 if (dev->dsr_version >= PVRDMA_QPHANDLE_VERSION && 262 udata->outlen < sizeof(qp_resp)) { 263 dev_warn(&dev->pdev->dev, 264 "create queuepair not supported\n"); 265 ret = -EOPNOTSUPP; 266 goto err_qp; 267 } 268 269 if (!is_srq) { 270 /* set qp->sq.wqe_cnt, shift, buf_size.. */ 271 qp->rumem = ib_umem_get(ibqp->device, 272 ucmd.rbuf_addr, 273 ucmd.rbuf_size, 0); 274 if (IS_ERR(qp->rumem)) { 275 ret = PTR_ERR(qp->rumem); 276 goto err_qp; 277 } 278 qp->srq = NULL; 279 } else { 280 qp->rumem = NULL; 281 qp->srq = to_vsrq(init_attr->srq); 282 } 283 284 qp->sumem = ib_umem_get(ibqp->device, ucmd.sbuf_addr, 285 ucmd.sbuf_size, 0); 286 if (IS_ERR(qp->sumem)) { 287 if (!is_srq) 288 ib_umem_release(qp->rumem); 289 ret = PTR_ERR(qp->sumem); 290 goto err_qp; 291 } 292 293 qp->npages_send = 294 ib_umem_num_dma_blocks(qp->sumem, PAGE_SIZE); 295 if (!is_srq) 296 qp->npages_recv = ib_umem_num_dma_blocks( 297 qp->rumem, PAGE_SIZE); 298 else 299 qp->npages_recv = 0; 300 qp->npages = qp->npages_send + qp->npages_recv; 301 } else { 302 ret = pvrdma_set_sq_size(to_vdev(ibqp->device), 303 &init_attr->cap, qp); 304 if (ret) 305 goto err_qp; 306 307 ret = pvrdma_set_rq_size(to_vdev(ibqp->device), 308 &init_attr->cap, qp); 309 if (ret) 310 goto err_qp; 311 312 qp->npages = qp->npages_send + qp->npages_recv; 313 314 /* Skip header page. */ 315 qp->sq.offset = PVRDMA_QP_NUM_HEADER_PAGES * PAGE_SIZE; 316 317 /* Recv queue pages are after send pages. */ 318 qp->rq.offset = qp->npages_send * PAGE_SIZE; 319 } 320 321 if (qp->npages < 0 || qp->npages > PVRDMA_PAGE_DIR_MAX_PAGES) { 322 dev_warn(&dev->pdev->dev, 323 "overflow pages in queuepair\n"); 324 ret = -EINVAL; 325 goto err_umem; 326 } 327 328 ret = pvrdma_page_dir_init(dev, &qp->pdir, qp->npages, 329 qp->is_kernel); 330 if (ret) { 331 dev_warn(&dev->pdev->dev, 332 "could not allocate page directory\n"); 333 goto err_umem; 334 } 335 336 if (!qp->is_kernel) { 337 pvrdma_page_dir_insert_umem(&qp->pdir, qp->sumem, 0); 338 if (!is_srq) 339 pvrdma_page_dir_insert_umem(&qp->pdir, 340 qp->rumem, 341 qp->npages_send); 342 } else { 343 /* Ring state is always the first page. */ 344 qp->sq.ring = qp->pdir.pages[0]; 345 qp->rq.ring = is_srq ? NULL : &qp->sq.ring[1]; 346 } 347 break; 348 default: 349 ret = -EINVAL; 350 goto err_qp; 351 } 352 353 /* Not supported */ 354 init_attr->cap.max_inline_data = 0; 355 356 memset(cmd, 0, sizeof(*cmd)); 357 cmd->hdr.cmd = PVRDMA_CMD_CREATE_QP; 358 cmd->pd_handle = to_vpd(ibqp->pd)->pd_handle; 359 cmd->send_cq_handle = to_vcq(init_attr->send_cq)->cq_handle; 360 cmd->recv_cq_handle = to_vcq(init_attr->recv_cq)->cq_handle; 361 if (is_srq) 362 cmd->srq_handle = to_vsrq(init_attr->srq)->srq_handle; 363 else 364 cmd->srq_handle = 0; 365 cmd->max_send_wr = init_attr->cap.max_send_wr; 366 cmd->max_recv_wr = init_attr->cap.max_recv_wr; 367 cmd->max_send_sge = init_attr->cap.max_send_sge; 368 cmd->max_recv_sge = init_attr->cap.max_recv_sge; 369 cmd->max_inline_data = init_attr->cap.max_inline_data; 370 cmd->sq_sig_all = (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR) ? 1 : 0; 371 cmd->qp_type = ib_qp_type_to_pvrdma(init_attr->qp_type); 372 cmd->is_srq = is_srq; 373 cmd->lkey = 0; 374 cmd->access_flags = IB_ACCESS_LOCAL_WRITE; 375 cmd->total_chunks = qp->npages; 376 cmd->send_chunks = qp->npages_send - PVRDMA_QP_NUM_HEADER_PAGES; 377 cmd->pdir_dma = qp->pdir.dir_dma; 378 379 dev_dbg(&dev->pdev->dev, "create queuepair with %d, %d, %d, %d\n", 380 cmd->max_send_wr, cmd->max_recv_wr, cmd->max_send_sge, 381 cmd->max_recv_sge); 382 383 ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_CREATE_QP_RESP); 384 if (ret < 0) { 385 dev_warn(&dev->pdev->dev, 386 "could not create queuepair, error: %d\n", ret); 387 goto err_pdir; 388 } 389 390 /* max_send_wr/_recv_wr/_send_sge/_recv_sge/_inline_data */ 391 qp->port = init_attr->port_num; 392 393 if (dev->dsr_version >= PVRDMA_QPHANDLE_VERSION) { 394 qp->ibqp.qp_num = resp_v2->qpn; 395 qp->qp_handle = resp_v2->qp_handle; 396 } else { 397 qp->ibqp.qp_num = resp->qpn; 398 qp->qp_handle = resp->qpn; 399 } 400 401 spin_lock_irqsave(&dev->qp_tbl_lock, flags); 402 dev->qp_tbl[qp->qp_handle % dev->dsr->caps.max_qp] = qp; 403 spin_unlock_irqrestore(&dev->qp_tbl_lock, flags); 404 405 if (udata) { 406 qp_resp.qpn = qp->ibqp.qp_num; 407 qp_resp.qp_handle = qp->qp_handle; 408 409 if (ib_copy_to_udata(udata, &qp_resp, 410 min(udata->outlen, sizeof(qp_resp)))) { 411 dev_warn(&dev->pdev->dev, 412 "failed to copy back udata\n"); 413 __pvrdma_destroy_qp(dev, qp); 414 return -EINVAL; 415 } 416 } 417 418 return 0; 419 420err_pdir: 421 pvrdma_page_dir_cleanup(dev, &qp->pdir); 422err_umem: 423 ib_umem_release(qp->rumem); 424 ib_umem_release(qp->sumem); 425err_qp: 426 atomic_dec(&dev->num_qps); 427 return ret; 428} 429 430static void _pvrdma_free_qp(struct pvrdma_qp *qp) 431{ 432 unsigned long flags; 433 struct pvrdma_dev *dev = to_vdev(qp->ibqp.device); 434 435 spin_lock_irqsave(&dev->qp_tbl_lock, flags); 436 dev->qp_tbl[qp->qp_handle] = NULL; 437 spin_unlock_irqrestore(&dev->qp_tbl_lock, flags); 438 439 if (refcount_dec_and_test(&qp->refcnt)) 440 complete(&qp->free); 441 wait_for_completion(&qp->free); 442 443 ib_umem_release(qp->rumem); 444 ib_umem_release(qp->sumem); 445 446 pvrdma_page_dir_cleanup(dev, &qp->pdir); 447 448 atomic_dec(&dev->num_qps); 449} 450 451static void pvrdma_free_qp(struct pvrdma_qp *qp) 452{ 453 struct pvrdma_cq *scq; 454 struct pvrdma_cq *rcq; 455 unsigned long scq_flags, rcq_flags; 456 457 /* In case cq is polling */ 458 get_cqs(qp, &scq, &rcq); 459 pvrdma_lock_cqs(scq, rcq, &scq_flags, &rcq_flags); 460 461 _pvrdma_flush_cqe(qp, scq); 462 if (scq != rcq) 463 _pvrdma_flush_cqe(qp, rcq); 464 465 /* 466 * We're now unlocking the CQs before clearing out the qp handle this 467 * should still be safe. We have destroyed the backend QP and flushed 468 * the CQEs so there should be no other completions for this QP. 469 */ 470 pvrdma_unlock_cqs(scq, rcq, &scq_flags, &rcq_flags); 471 472 _pvrdma_free_qp(qp); 473} 474 475static inline void _pvrdma_destroy_qp_work(struct pvrdma_dev *dev, 476 u32 qp_handle) 477{ 478 union pvrdma_cmd_req req; 479 struct pvrdma_cmd_destroy_qp *cmd = &req.destroy_qp; 480 int ret; 481 482 memset(cmd, 0, sizeof(*cmd)); 483 cmd->hdr.cmd = PVRDMA_CMD_DESTROY_QP; 484 cmd->qp_handle = qp_handle; 485 486 ret = pvrdma_cmd_post(dev, &req, NULL, 0); 487 if (ret < 0) 488 dev_warn(&dev->pdev->dev, 489 "destroy queuepair failed, error: %d\n", ret); 490} 491 492/** 493 * pvrdma_destroy_qp - destroy a queue pair 494 * @qp: the queue pair to destroy 495 * @udata: user data or null for kernel object 496 * 497 * @return: always 0. 498 */ 499int pvrdma_destroy_qp(struct ib_qp *qp, struct ib_udata *udata) 500{ 501 struct pvrdma_qp *vqp = to_vqp(qp); 502 503 _pvrdma_destroy_qp_work(to_vdev(qp->device), vqp->qp_handle); 504 pvrdma_free_qp(vqp); 505 506 return 0; 507} 508 509static void __pvrdma_destroy_qp(struct pvrdma_dev *dev, 510 struct pvrdma_qp *qp) 511{ 512 _pvrdma_destroy_qp_work(dev, qp->qp_handle); 513 _pvrdma_free_qp(qp); 514} 515 516/** 517 * pvrdma_modify_qp - modify queue pair attributes 518 * @ibqp: the queue pair 519 * @attr: the new queue pair's attributes 520 * @attr_mask: attributes mask 521 * @udata: user data 522 * 523 * @returns 0 on success, otherwise returns an errno. 524 */ 525int pvrdma_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, 526 int attr_mask, struct ib_udata *udata) 527{ 528 struct pvrdma_dev *dev = to_vdev(ibqp->device); 529 struct pvrdma_qp *qp = to_vqp(ibqp); 530 union pvrdma_cmd_req req; 531 union pvrdma_cmd_resp rsp; 532 struct pvrdma_cmd_modify_qp *cmd = &req.modify_qp; 533 enum ib_qp_state cur_state, next_state; 534 int ret; 535 536 if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS) 537 return -EOPNOTSUPP; 538 539 /* Sanity checking. Should need lock here */ 540 mutex_lock(&qp->mutex); 541 cur_state = (attr_mask & IB_QP_CUR_STATE) ? attr->cur_qp_state : 542 qp->state; 543 next_state = (attr_mask & IB_QP_STATE) ? attr->qp_state : cur_state; 544 545 if (!ib_modify_qp_is_ok(cur_state, next_state, ibqp->qp_type, 546 attr_mask)) { 547 ret = -EINVAL; 548 goto out; 549 } 550 551 if (attr_mask & IB_QP_PORT) { 552 if (attr->port_num == 0 || 553 attr->port_num > ibqp->device->phys_port_cnt) { 554 ret = -EINVAL; 555 goto out; 556 } 557 } 558 559 if (attr_mask & IB_QP_MIN_RNR_TIMER) { 560 if (attr->min_rnr_timer > 31) { 561 ret = -EINVAL; 562 goto out; 563 } 564 } 565 566 if (attr_mask & IB_QP_PKEY_INDEX) { 567 if (attr->pkey_index >= dev->dsr->caps.max_pkeys) { 568 ret = -EINVAL; 569 goto out; 570 } 571 } 572 573 if (attr_mask & IB_QP_QKEY) 574 qp->qkey = attr->qkey; 575 576 if (cur_state == next_state && cur_state == IB_QPS_RESET) { 577 ret = 0; 578 goto out; 579 } 580 581 qp->state = next_state; 582 memset(cmd, 0, sizeof(*cmd)); 583 cmd->hdr.cmd = PVRDMA_CMD_MODIFY_QP; 584 cmd->qp_handle = qp->qp_handle; 585 cmd->attr_mask = ib_qp_attr_mask_to_pvrdma(attr_mask); 586 cmd->attrs.qp_state = ib_qp_state_to_pvrdma(attr->qp_state); 587 cmd->attrs.cur_qp_state = 588 ib_qp_state_to_pvrdma(attr->cur_qp_state); 589 cmd->attrs.path_mtu = ib_mtu_to_pvrdma(attr->path_mtu); 590 cmd->attrs.path_mig_state = 591 ib_mig_state_to_pvrdma(attr->path_mig_state); 592 cmd->attrs.qkey = attr->qkey; 593 cmd->attrs.rq_psn = attr->rq_psn; 594 cmd->attrs.sq_psn = attr->sq_psn; 595 cmd->attrs.dest_qp_num = attr->dest_qp_num; 596 cmd->attrs.qp_access_flags = 597 ib_access_flags_to_pvrdma(attr->qp_access_flags); 598 cmd->attrs.pkey_index = attr->pkey_index; 599 cmd->attrs.alt_pkey_index = attr->alt_pkey_index; 600 cmd->attrs.en_sqd_async_notify = attr->en_sqd_async_notify; 601 cmd->attrs.sq_draining = attr->sq_draining; 602 cmd->attrs.max_rd_atomic = attr->max_rd_atomic; 603 cmd->attrs.max_dest_rd_atomic = attr->max_dest_rd_atomic; 604 cmd->attrs.min_rnr_timer = attr->min_rnr_timer; 605 cmd->attrs.port_num = attr->port_num; 606 cmd->attrs.timeout = attr->timeout; 607 cmd->attrs.retry_cnt = attr->retry_cnt; 608 cmd->attrs.rnr_retry = attr->rnr_retry; 609 cmd->attrs.alt_port_num = attr->alt_port_num; 610 cmd->attrs.alt_timeout = attr->alt_timeout; 611 ib_qp_cap_to_pvrdma(&cmd->attrs.cap, &attr->cap); 612 rdma_ah_attr_to_pvrdma(&cmd->attrs.ah_attr, &attr->ah_attr); 613 rdma_ah_attr_to_pvrdma(&cmd->attrs.alt_ah_attr, &attr->alt_ah_attr); 614 615 ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_MODIFY_QP_RESP); 616 if (ret < 0) { 617 dev_warn(&dev->pdev->dev, 618 "could not modify queuepair, error: %d\n", ret); 619 } else if (rsp.hdr.err > 0) { 620 dev_warn(&dev->pdev->dev, 621 "cannot modify queuepair, error: %d\n", rsp.hdr.err); 622 ret = -EINVAL; 623 } 624 625 if (ret == 0 && next_state == IB_QPS_RESET) 626 pvrdma_reset_qp(qp); 627 628out: 629 mutex_unlock(&qp->mutex); 630 631 return ret; 632} 633 634static inline void *get_sq_wqe(struct pvrdma_qp *qp, unsigned int n) 635{ 636 return pvrdma_page_dir_get_ptr(&qp->pdir, 637 qp->sq.offset + n * qp->sq.wqe_size); 638} 639 640static inline void *get_rq_wqe(struct pvrdma_qp *qp, unsigned int n) 641{ 642 return pvrdma_page_dir_get_ptr(&qp->pdir, 643 qp->rq.offset + n * qp->rq.wqe_size); 644} 645 646static int set_reg_seg(struct pvrdma_sq_wqe_hdr *wqe_hdr, 647 const struct ib_reg_wr *wr) 648{ 649 struct pvrdma_user_mr *mr = to_vmr(wr->mr); 650 651 wqe_hdr->wr.fast_reg.iova_start = mr->ibmr.iova; 652 wqe_hdr->wr.fast_reg.pl_pdir_dma = mr->pdir.dir_dma; 653 wqe_hdr->wr.fast_reg.page_shift = mr->page_shift; 654 wqe_hdr->wr.fast_reg.page_list_len = mr->npages; 655 wqe_hdr->wr.fast_reg.length = mr->ibmr.length; 656 wqe_hdr->wr.fast_reg.access_flags = wr->access; 657 wqe_hdr->wr.fast_reg.rkey = wr->key; 658 659 return pvrdma_page_dir_insert_page_list(&mr->pdir, mr->pages, 660 mr->npages); 661} 662 663/** 664 * pvrdma_post_send - post send work request entries on a QP 665 * @ibqp: the QP 666 * @wr: work request list to post 667 * @bad_wr: the first bad WR returned 668 * 669 * @return: 0 on success, otherwise errno returned. 670 */ 671int pvrdma_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr, 672 const struct ib_send_wr **bad_wr) 673{ 674 struct pvrdma_qp *qp = to_vqp(ibqp); 675 struct pvrdma_dev *dev = to_vdev(ibqp->device); 676 unsigned long flags; 677 struct pvrdma_sq_wqe_hdr *wqe_hdr; 678 struct pvrdma_sge *sge; 679 int i, ret; 680 681 /* 682 * In states lower than RTS, we can fail immediately. In other states, 683 * just post and let the device figure it out. 684 */ 685 if (qp->state < IB_QPS_RTS) { 686 *bad_wr = wr; 687 return -EINVAL; 688 } 689 690 spin_lock_irqsave(&qp->sq.lock, flags); 691 692 while (wr) { 693 unsigned int tail = 0; 694 695 if (unlikely(!pvrdma_idx_ring_has_space( 696 qp->sq.ring, qp->sq.wqe_cnt, &tail))) { 697 dev_warn_ratelimited(&dev->pdev->dev, 698 "send queue is full\n"); 699 *bad_wr = wr; 700 ret = -ENOMEM; 701 goto out; 702 } 703 704 if (unlikely(wr->num_sge > qp->sq.max_sg || wr->num_sge < 0)) { 705 dev_warn_ratelimited(&dev->pdev->dev, 706 "send SGE overflow\n"); 707 *bad_wr = wr; 708 ret = -EINVAL; 709 goto out; 710 } 711 712 /* 713 * Only support UD, RC. 714 * Need to check opcode table for thorough checking. 715 * opcode _UD _UC _RC 716 * _SEND x x x 717 * _SEND_WITH_IMM x x x 718 * _RDMA_WRITE x x 719 * _RDMA_WRITE_WITH_IMM x x 720 * _LOCAL_INV x x 721 * _SEND_WITH_INV x x 722 * _RDMA_READ x 723 * _ATOMIC_CMP_AND_SWP x 724 * _ATOMIC_FETCH_AND_ADD x 725 * _MASK_ATOMIC_CMP_AND_SWP x 726 * _MASK_ATOMIC_FETCH_AND_ADD x 727 * _REG_MR x 728 * 729 */ 730 if (qp->ibqp.qp_type != IB_QPT_UD && 731 qp->ibqp.qp_type != IB_QPT_RC && 732 wr->opcode != IB_WR_SEND) { 733 dev_warn_ratelimited(&dev->pdev->dev, 734 "unsupported queuepair type\n"); 735 *bad_wr = wr; 736 ret = -EINVAL; 737 goto out; 738 } else if (qp->ibqp.qp_type == IB_QPT_UD || 739 qp->ibqp.qp_type == IB_QPT_GSI) { 740 if (wr->opcode != IB_WR_SEND && 741 wr->opcode != IB_WR_SEND_WITH_IMM) { 742 dev_warn_ratelimited(&dev->pdev->dev, 743 "invalid send opcode\n"); 744 *bad_wr = wr; 745 ret = -EINVAL; 746 goto out; 747 } 748 } 749 750 wqe_hdr = (struct pvrdma_sq_wqe_hdr *)get_sq_wqe(qp, tail); 751 memset(wqe_hdr, 0, sizeof(*wqe_hdr)); 752 wqe_hdr->wr_id = wr->wr_id; 753 wqe_hdr->num_sge = wr->num_sge; 754 wqe_hdr->opcode = ib_wr_opcode_to_pvrdma(wr->opcode); 755 wqe_hdr->send_flags = ib_send_flags_to_pvrdma(wr->send_flags); 756 if (wr->opcode == IB_WR_SEND_WITH_IMM || 757 wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) 758 wqe_hdr->ex.imm_data = wr->ex.imm_data; 759 760 if (unlikely(wqe_hdr->opcode == PVRDMA_WR_ERROR)) { 761 *bad_wr = wr; 762 ret = -EINVAL; 763 goto out; 764 } 765 766 switch (qp->ibqp.qp_type) { 767 case IB_QPT_GSI: 768 case IB_QPT_UD: 769 if (unlikely(!ud_wr(wr)->ah)) { 770 dev_warn_ratelimited(&dev->pdev->dev, 771 "invalid address handle\n"); 772 *bad_wr = wr; 773 ret = -EINVAL; 774 goto out; 775 } 776 777 /* 778 * Use qkey from qp context if high order bit set, 779 * otherwise from work request. 780 */ 781 wqe_hdr->wr.ud.remote_qpn = ud_wr(wr)->remote_qpn; 782 wqe_hdr->wr.ud.remote_qkey = 783 ud_wr(wr)->remote_qkey & 0x80000000 ? 784 qp->qkey : ud_wr(wr)->remote_qkey; 785 wqe_hdr->wr.ud.av = to_vah(ud_wr(wr)->ah)->av; 786 787 break; 788 case IB_QPT_RC: 789 switch (wr->opcode) { 790 case IB_WR_RDMA_READ: 791 case IB_WR_RDMA_WRITE: 792 case IB_WR_RDMA_WRITE_WITH_IMM: 793 wqe_hdr->wr.rdma.remote_addr = 794 rdma_wr(wr)->remote_addr; 795 wqe_hdr->wr.rdma.rkey = rdma_wr(wr)->rkey; 796 break; 797 case IB_WR_LOCAL_INV: 798 case IB_WR_SEND_WITH_INV: 799 wqe_hdr->ex.invalidate_rkey = 800 wr->ex.invalidate_rkey; 801 break; 802 case IB_WR_ATOMIC_CMP_AND_SWP: 803 case IB_WR_ATOMIC_FETCH_AND_ADD: 804 wqe_hdr->wr.atomic.remote_addr = 805 atomic_wr(wr)->remote_addr; 806 wqe_hdr->wr.atomic.rkey = atomic_wr(wr)->rkey; 807 wqe_hdr->wr.atomic.compare_add = 808 atomic_wr(wr)->compare_add; 809 if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) 810 wqe_hdr->wr.atomic.swap = 811 atomic_wr(wr)->swap; 812 break; 813 case IB_WR_REG_MR: 814 ret = set_reg_seg(wqe_hdr, reg_wr(wr)); 815 if (ret < 0) { 816 dev_warn_ratelimited(&dev->pdev->dev, 817 "Failed to set fast register work request\n"); 818 *bad_wr = wr; 819 goto out; 820 } 821 break; 822 default: 823 break; 824 } 825 826 break; 827 default: 828 dev_warn_ratelimited(&dev->pdev->dev, 829 "invalid queuepair type\n"); 830 ret = -EINVAL; 831 *bad_wr = wr; 832 goto out; 833 } 834 835 sge = (struct pvrdma_sge *)(wqe_hdr + 1); 836 for (i = 0; i < wr->num_sge; i++) { 837 /* Need to check wqe_size 0 or max size */ 838 sge->addr = wr->sg_list[i].addr; 839 sge->length = wr->sg_list[i].length; 840 sge->lkey = wr->sg_list[i].lkey; 841 sge++; 842 } 843 844 /* Make sure wqe is written before index update */ 845 smp_wmb(); 846 847 /* Update shared sq ring */ 848 pvrdma_idx_ring_inc(&qp->sq.ring->prod_tail, 849 qp->sq.wqe_cnt); 850 851 wr = wr->next; 852 } 853 854 ret = 0; 855 856out: 857 spin_unlock_irqrestore(&qp->sq.lock, flags); 858 859 if (!ret) 860 pvrdma_write_uar_qp(dev, PVRDMA_UAR_QP_SEND | qp->qp_handle); 861 862 return ret; 863} 864 865/** 866 * pvrdma_post_recv - post receive work request entries on a QP 867 * @ibqp: the QP 868 * @wr: the work request list to post 869 * @bad_wr: the first bad WR returned 870 * 871 * @return: 0 on success, otherwise errno returned. 872 */ 873int pvrdma_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr, 874 const struct ib_recv_wr **bad_wr) 875{ 876 struct pvrdma_dev *dev = to_vdev(ibqp->device); 877 unsigned long flags; 878 struct pvrdma_qp *qp = to_vqp(ibqp); 879 struct pvrdma_rq_wqe_hdr *wqe_hdr; 880 struct pvrdma_sge *sge; 881 int ret = 0; 882 int i; 883 884 /* 885 * In the RESET state, we can fail immediately. For other states, 886 * just post and let the device figure it out. 887 */ 888 if (qp->state == IB_QPS_RESET) { 889 *bad_wr = wr; 890 return -EINVAL; 891 } 892 893 if (qp->srq) { 894 dev_warn(&dev->pdev->dev, "QP associated with SRQ\n"); 895 *bad_wr = wr; 896 return -EINVAL; 897 } 898 899 spin_lock_irqsave(&qp->rq.lock, flags); 900 901 while (wr) { 902 unsigned int tail = 0; 903 904 if (unlikely(wr->num_sge > qp->rq.max_sg || 905 wr->num_sge < 0)) { 906 ret = -EINVAL; 907 *bad_wr = wr; 908 dev_warn_ratelimited(&dev->pdev->dev, 909 "recv SGE overflow\n"); 910 goto out; 911 } 912 913 if (unlikely(!pvrdma_idx_ring_has_space( 914 qp->rq.ring, qp->rq.wqe_cnt, &tail))) { 915 ret = -ENOMEM; 916 *bad_wr = wr; 917 dev_warn_ratelimited(&dev->pdev->dev, 918 "recv queue full\n"); 919 goto out; 920 } 921 922 wqe_hdr = (struct pvrdma_rq_wqe_hdr *)get_rq_wqe(qp, tail); 923 wqe_hdr->wr_id = wr->wr_id; 924 wqe_hdr->num_sge = wr->num_sge; 925 wqe_hdr->total_len = 0; 926 927 sge = (struct pvrdma_sge *)(wqe_hdr + 1); 928 for (i = 0; i < wr->num_sge; i++) { 929 sge->addr = wr->sg_list[i].addr; 930 sge->length = wr->sg_list[i].length; 931 sge->lkey = wr->sg_list[i].lkey; 932 sge++; 933 } 934 935 /* Make sure wqe is written before index update */ 936 smp_wmb(); 937 938 /* Update shared rq ring */ 939 pvrdma_idx_ring_inc(&qp->rq.ring->prod_tail, 940 qp->rq.wqe_cnt); 941 942 wr = wr->next; 943 } 944 945 spin_unlock_irqrestore(&qp->rq.lock, flags); 946 947 pvrdma_write_uar_qp(dev, PVRDMA_UAR_QP_RECV | qp->qp_handle); 948 949 return ret; 950 951out: 952 spin_unlock_irqrestore(&qp->rq.lock, flags); 953 954 return ret; 955} 956 957/** 958 * pvrdma_query_qp - query a queue pair's attributes 959 * @ibqp: the queue pair to query 960 * @attr: the queue pair's attributes 961 * @attr_mask: attributes mask 962 * @init_attr: initial queue pair attributes 963 * 964 * @returns 0 on success, otherwise returns an errno. 965 */ 966int pvrdma_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, 967 int attr_mask, struct ib_qp_init_attr *init_attr) 968{ 969 struct pvrdma_dev *dev = to_vdev(ibqp->device); 970 struct pvrdma_qp *qp = to_vqp(ibqp); 971 union pvrdma_cmd_req req; 972 union pvrdma_cmd_resp rsp; 973 struct pvrdma_cmd_query_qp *cmd = &req.query_qp; 974 struct pvrdma_cmd_query_qp_resp *resp = &rsp.query_qp_resp; 975 int ret = 0; 976 977 mutex_lock(&qp->mutex); 978 979 if (qp->state == IB_QPS_RESET) { 980 attr->qp_state = IB_QPS_RESET; 981 goto out; 982 } 983 984 memset(cmd, 0, sizeof(*cmd)); 985 cmd->hdr.cmd = PVRDMA_CMD_QUERY_QP; 986 cmd->qp_handle = qp->qp_handle; 987 cmd->attr_mask = ib_qp_attr_mask_to_pvrdma(attr_mask); 988 989 ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_QUERY_QP_RESP); 990 if (ret < 0) { 991 dev_warn(&dev->pdev->dev, 992 "could not query queuepair, error: %d\n", ret); 993 goto out; 994 } 995 996 attr->qp_state = pvrdma_qp_state_to_ib(resp->attrs.qp_state); 997 attr->cur_qp_state = 998 pvrdma_qp_state_to_ib(resp->attrs.cur_qp_state); 999 attr->path_mtu = pvrdma_mtu_to_ib(resp->attrs.path_mtu); 1000 attr->path_mig_state = 1001 pvrdma_mig_state_to_ib(resp->attrs.path_mig_state); 1002 attr->qkey = resp->attrs.qkey; 1003 attr->rq_psn = resp->attrs.rq_psn; 1004 attr->sq_psn = resp->attrs.sq_psn; 1005 attr->dest_qp_num = resp->attrs.dest_qp_num; 1006 attr->qp_access_flags = 1007 pvrdma_access_flags_to_ib(resp->attrs.qp_access_flags); 1008 attr->pkey_index = resp->attrs.pkey_index; 1009 attr->alt_pkey_index = resp->attrs.alt_pkey_index; 1010 attr->en_sqd_async_notify = resp->attrs.en_sqd_async_notify; 1011 attr->sq_draining = resp->attrs.sq_draining; 1012 attr->max_rd_atomic = resp->attrs.max_rd_atomic; 1013 attr->max_dest_rd_atomic = resp->attrs.max_dest_rd_atomic; 1014 attr->min_rnr_timer = resp->attrs.min_rnr_timer; 1015 attr->port_num = resp->attrs.port_num; 1016 attr->timeout = resp->attrs.timeout; 1017 attr->retry_cnt = resp->attrs.retry_cnt; 1018 attr->rnr_retry = resp->attrs.rnr_retry; 1019 attr->alt_port_num = resp->attrs.alt_port_num; 1020 attr->alt_timeout = resp->attrs.alt_timeout; 1021 pvrdma_qp_cap_to_ib(&attr->cap, &resp->attrs.cap); 1022 pvrdma_ah_attr_to_rdma(&attr->ah_attr, &resp->attrs.ah_attr); 1023 pvrdma_ah_attr_to_rdma(&attr->alt_ah_attr, &resp->attrs.alt_ah_attr); 1024 1025 qp->state = attr->qp_state; 1026 1027 ret = 0; 1028 1029out: 1030 attr->cur_qp_state = attr->qp_state; 1031 1032 init_attr->event_handler = qp->ibqp.event_handler; 1033 init_attr->qp_context = qp->ibqp.qp_context; 1034 init_attr->send_cq = qp->ibqp.send_cq; 1035 init_attr->recv_cq = qp->ibqp.recv_cq; 1036 init_attr->srq = qp->ibqp.srq; 1037 init_attr->xrcd = NULL; 1038 init_attr->cap = attr->cap; 1039 init_attr->sq_sig_type = 0; 1040 init_attr->qp_type = qp->ibqp.qp_type; 1041 init_attr->create_flags = 0; 1042 init_attr->port_num = qp->port; 1043 1044 mutex_unlock(&qp->mutex); 1045 return ret; 1046} 1047