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/inet.h> 48 #include <linux/io.h> 49 #include <rdma/ib_addr.h> 50 #include <rdma/ib_smi.h> 51 #include <rdma/ib_user_verbs.h> 52 #include <rdma/vmw_pvrdma-abi.h> 53 54 #include "pvrdma.h" 55 56 /** 57 * pvrdma_query_device - query device 58 * @ibdev: the device to query 59 * @props: the device properties 60 * @uhw: user data 61 * 62 * @return: 0 on success, otherwise negative errno 63 */ 64 int pvrdma_query_device(struct ib_device *ibdev, 65 struct ib_device_attr *props, 66 struct ib_udata *uhw) 67 { 68 struct pvrdma_dev *dev = to_vdev(ibdev); 69 70 if (uhw->inlen || uhw->outlen) 71 return -EINVAL; 72 73 memset(props, 0, sizeof(*props)); 74 75 props->fw_ver = dev->dsr->caps.fw_ver; 76 props->sys_image_guid = dev->dsr->caps.sys_image_guid; 77 props->max_mr_size = dev->dsr->caps.max_mr_size; 78 props->page_size_cap = dev->dsr->caps.page_size_cap; 79 props->vendor_id = dev->dsr->caps.vendor_id; 80 props->vendor_part_id = dev->pdev->device; 81 props->hw_ver = dev->dsr->caps.hw_ver; 82 props->max_qp = dev->dsr->caps.max_qp; 83 props->max_qp_wr = dev->dsr->caps.max_qp_wr; 84 props->device_cap_flags = dev->dsr->caps.device_cap_flags; 85 props->max_sge = dev->dsr->caps.max_sge; 86 props->max_cq = dev->dsr->caps.max_cq; 87 props->max_cqe = dev->dsr->caps.max_cqe; 88 props->max_mr = dev->dsr->caps.max_mr; 89 props->max_pd = dev->dsr->caps.max_pd; 90 props->max_qp_rd_atom = dev->dsr->caps.max_qp_rd_atom; 91 props->max_qp_init_rd_atom = dev->dsr->caps.max_qp_init_rd_atom; 92 props->atomic_cap = 93 dev->dsr->caps.atomic_ops & 94 (PVRDMA_ATOMIC_OP_COMP_SWAP | PVRDMA_ATOMIC_OP_FETCH_ADD) ? 95 IB_ATOMIC_HCA : IB_ATOMIC_NONE; 96 props->masked_atomic_cap = props->atomic_cap; 97 props->max_ah = dev->dsr->caps.max_ah; 98 props->max_pkeys = dev->dsr->caps.max_pkeys; 99 props->local_ca_ack_delay = dev->dsr->caps.local_ca_ack_delay; 100 if ((dev->dsr->caps.bmme_flags & PVRDMA_BMME_FLAG_LOCAL_INV) && 101 (dev->dsr->caps.bmme_flags & PVRDMA_BMME_FLAG_REMOTE_INV) && 102 (dev->dsr->caps.bmme_flags & PVRDMA_BMME_FLAG_FAST_REG_WR)) { 103 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS; 104 } 105 106 return 0; 107 } 108 109 /** 110 * pvrdma_query_port - query device port attributes 111 * @ibdev: the device to query 112 * @port: the port number 113 * @props: the device properties 114 * 115 * @return: 0 on success, otherwise negative errno 116 */ 117 int pvrdma_query_port(struct ib_device *ibdev, u8 port, 118 struct ib_port_attr *props) 119 { 120 struct pvrdma_dev *dev = to_vdev(ibdev); 121 union pvrdma_cmd_req req; 122 union pvrdma_cmd_resp rsp; 123 struct pvrdma_cmd_query_port *cmd = &req.query_port; 124 struct pvrdma_cmd_query_port_resp *resp = &rsp.query_port_resp; 125 int err; 126 127 memset(cmd, 0, sizeof(*cmd)); 128 cmd->hdr.cmd = PVRDMA_CMD_QUERY_PORT; 129 cmd->port_num = port; 130 131 err = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_QUERY_PORT_RESP); 132 if (err < 0) { 133 dev_warn(&dev->pdev->dev, 134 "could not query port, error: %d\n", err); 135 return err; 136 } 137 138 /* props being zeroed by the caller, avoid zeroing it here */ 139 140 props->state = pvrdma_port_state_to_ib(resp->attrs.state); 141 props->max_mtu = pvrdma_mtu_to_ib(resp->attrs.max_mtu); 142 props->active_mtu = pvrdma_mtu_to_ib(resp->attrs.active_mtu); 143 props->gid_tbl_len = resp->attrs.gid_tbl_len; 144 props->port_cap_flags = 145 pvrdma_port_cap_flags_to_ib(resp->attrs.port_cap_flags); 146 props->max_msg_sz = resp->attrs.max_msg_sz; 147 props->bad_pkey_cntr = resp->attrs.bad_pkey_cntr; 148 props->qkey_viol_cntr = resp->attrs.qkey_viol_cntr; 149 props->pkey_tbl_len = resp->attrs.pkey_tbl_len; 150 props->lid = resp->attrs.lid; 151 props->sm_lid = resp->attrs.sm_lid; 152 props->lmc = resp->attrs.lmc; 153 props->max_vl_num = resp->attrs.max_vl_num; 154 props->sm_sl = resp->attrs.sm_sl; 155 props->subnet_timeout = resp->attrs.subnet_timeout; 156 props->init_type_reply = resp->attrs.init_type_reply; 157 props->active_width = pvrdma_port_width_to_ib(resp->attrs.active_width); 158 props->active_speed = pvrdma_port_speed_to_ib(resp->attrs.active_speed); 159 props->phys_state = resp->attrs.phys_state; 160 161 return 0; 162 } 163 164 /** 165 * pvrdma_query_gid - query device gid 166 * @ibdev: the device to query 167 * @port: the port number 168 * @index: the index 169 * @gid: the device gid value 170 * 171 * @return: 0 on success, otherwise negative errno 172 */ 173 int pvrdma_query_gid(struct ib_device *ibdev, u8 port, int index, 174 union ib_gid *gid) 175 { 176 struct pvrdma_dev *dev = to_vdev(ibdev); 177 178 if (index >= dev->dsr->caps.gid_tbl_len) 179 return -EINVAL; 180 181 memcpy(gid, &dev->sgid_tbl[index], sizeof(union ib_gid)); 182 183 return 0; 184 } 185 186 /** 187 * pvrdma_query_pkey - query device port's P_Key table 188 * @ibdev: the device to query 189 * @port: the port number 190 * @index: the index 191 * @pkey: the device P_Key value 192 * 193 * @return: 0 on success, otherwise negative errno 194 */ 195 int pvrdma_query_pkey(struct ib_device *ibdev, u8 port, u16 index, 196 u16 *pkey) 197 { 198 int err = 0; 199 union pvrdma_cmd_req req; 200 union pvrdma_cmd_resp rsp; 201 struct pvrdma_cmd_query_pkey *cmd = &req.query_pkey; 202 203 memset(cmd, 0, sizeof(*cmd)); 204 cmd->hdr.cmd = PVRDMA_CMD_QUERY_PKEY; 205 cmd->port_num = port; 206 cmd->index = index; 207 208 err = pvrdma_cmd_post(to_vdev(ibdev), &req, &rsp, 209 PVRDMA_CMD_QUERY_PKEY_RESP); 210 if (err < 0) { 211 dev_warn(&to_vdev(ibdev)->pdev->dev, 212 "could not query pkey, error: %d\n", err); 213 return err; 214 } 215 216 *pkey = rsp.query_pkey_resp.pkey; 217 218 return 0; 219 } 220 221 enum rdma_link_layer pvrdma_port_link_layer(struct ib_device *ibdev, 222 u8 port) 223 { 224 return IB_LINK_LAYER_ETHERNET; 225 } 226 227 int pvrdma_modify_device(struct ib_device *ibdev, int mask, 228 struct ib_device_modify *props) 229 { 230 unsigned long flags; 231 232 if (mask & ~(IB_DEVICE_MODIFY_SYS_IMAGE_GUID | 233 IB_DEVICE_MODIFY_NODE_DESC)) { 234 dev_warn(&to_vdev(ibdev)->pdev->dev, 235 "unsupported device modify mask %#x\n", mask); 236 return -EOPNOTSUPP; 237 } 238 239 if (mask & IB_DEVICE_MODIFY_NODE_DESC) { 240 spin_lock_irqsave(&to_vdev(ibdev)->desc_lock, flags); 241 memcpy(ibdev->node_desc, props->node_desc, 64); 242 spin_unlock_irqrestore(&to_vdev(ibdev)->desc_lock, flags); 243 } 244 245 if (mask & IB_DEVICE_MODIFY_SYS_IMAGE_GUID) { 246 mutex_lock(&to_vdev(ibdev)->port_mutex); 247 to_vdev(ibdev)->sys_image_guid = 248 cpu_to_be64(props->sys_image_guid); 249 mutex_unlock(&to_vdev(ibdev)->port_mutex); 250 } 251 252 return 0; 253 } 254 255 /** 256 * pvrdma_modify_port - modify device port attributes 257 * @ibdev: the device to modify 258 * @port: the port number 259 * @mask: attributes to modify 260 * @props: the device properties 261 * 262 * @return: 0 on success, otherwise negative errno 263 */ 264 int pvrdma_modify_port(struct ib_device *ibdev, u8 port, int mask, 265 struct ib_port_modify *props) 266 { 267 struct ib_port_attr attr; 268 struct pvrdma_dev *vdev = to_vdev(ibdev); 269 int ret; 270 271 if (mask & ~IB_PORT_SHUTDOWN) { 272 dev_warn(&vdev->pdev->dev, 273 "unsupported port modify mask %#x\n", mask); 274 return -EOPNOTSUPP; 275 } 276 277 mutex_lock(&vdev->port_mutex); 278 ret = ib_query_port(ibdev, port, &attr); 279 if (ret) 280 goto out; 281 282 vdev->port_cap_mask |= props->set_port_cap_mask; 283 vdev->port_cap_mask &= ~props->clr_port_cap_mask; 284 285 if (mask & IB_PORT_SHUTDOWN) 286 vdev->ib_active = false; 287 288 out: 289 mutex_unlock(&vdev->port_mutex); 290 return ret; 291 } 292 293 /** 294 * pvrdma_alloc_ucontext - allocate ucontext 295 * @ibdev: the IB device 296 * @udata: user data 297 * 298 * @return: the ib_ucontext pointer on success, otherwise errno. 299 */ 300 struct ib_ucontext *pvrdma_alloc_ucontext(struct ib_device *ibdev, 301 struct ib_udata *udata) 302 { 303 struct pvrdma_dev *vdev = to_vdev(ibdev); 304 struct pvrdma_ucontext *context; 305 union pvrdma_cmd_req req; 306 union pvrdma_cmd_resp rsp; 307 struct pvrdma_cmd_create_uc *cmd = &req.create_uc; 308 struct pvrdma_cmd_create_uc_resp *resp = &rsp.create_uc_resp; 309 struct pvrdma_alloc_ucontext_resp uresp = {0}; 310 int ret; 311 void *ptr; 312 313 if (!vdev->ib_active) 314 return ERR_PTR(-EAGAIN); 315 316 context = kmalloc(sizeof(*context), GFP_KERNEL); 317 if (!context) 318 return ERR_PTR(-ENOMEM); 319 320 context->dev = vdev; 321 ret = pvrdma_uar_alloc(vdev, &context->uar); 322 if (ret) { 323 kfree(context); 324 return ERR_PTR(-ENOMEM); 325 } 326 327 /* get ctx_handle from host */ 328 memset(cmd, 0, sizeof(*cmd)); 329 cmd->pfn = context->uar.pfn; 330 cmd->hdr.cmd = PVRDMA_CMD_CREATE_UC; 331 ret = pvrdma_cmd_post(vdev, &req, &rsp, PVRDMA_CMD_CREATE_UC_RESP); 332 if (ret < 0) { 333 dev_warn(&vdev->pdev->dev, 334 "could not create ucontext, error: %d\n", ret); 335 ptr = ERR_PTR(ret); 336 goto err; 337 } 338 339 context->ctx_handle = resp->ctx_handle; 340 341 /* copy back to user */ 342 uresp.qp_tab_size = vdev->dsr->caps.max_qp; 343 ret = ib_copy_to_udata(udata, &uresp, sizeof(uresp)); 344 if (ret) { 345 pvrdma_uar_free(vdev, &context->uar); 346 context->ibucontext.device = ibdev; 347 pvrdma_dealloc_ucontext(&context->ibucontext); 348 return ERR_PTR(-EFAULT); 349 } 350 351 return &context->ibucontext; 352 353 err: 354 pvrdma_uar_free(vdev, &context->uar); 355 kfree(context); 356 return ptr; 357 } 358 359 /** 360 * pvrdma_dealloc_ucontext - deallocate ucontext 361 * @ibcontext: the ucontext 362 * 363 * @return: 0 on success, otherwise errno. 364 */ 365 int pvrdma_dealloc_ucontext(struct ib_ucontext *ibcontext) 366 { 367 struct pvrdma_ucontext *context = to_vucontext(ibcontext); 368 union pvrdma_cmd_req req; 369 struct pvrdma_cmd_destroy_uc *cmd = &req.destroy_uc; 370 int ret; 371 372 memset(cmd, 0, sizeof(*cmd)); 373 cmd->hdr.cmd = PVRDMA_CMD_DESTROY_UC; 374 cmd->ctx_handle = context->ctx_handle; 375 376 ret = pvrdma_cmd_post(context->dev, &req, NULL, 0); 377 if (ret < 0) 378 dev_warn(&context->dev->pdev->dev, 379 "destroy ucontext failed, error: %d\n", ret); 380 381 /* Free the UAR even if the device command failed */ 382 pvrdma_uar_free(to_vdev(ibcontext->device), &context->uar); 383 kfree(context); 384 385 return ret; 386 } 387 388 /** 389 * pvrdma_mmap - create mmap region 390 * @ibcontext: the user context 391 * @vma: the VMA 392 * 393 * @return: 0 on success, otherwise errno. 394 */ 395 int pvrdma_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vma) 396 { 397 struct pvrdma_ucontext *context = to_vucontext(ibcontext); 398 unsigned long start = vma->vm_start; 399 unsigned long size = vma->vm_end - vma->vm_start; 400 unsigned long offset = vma->vm_pgoff << PAGE_SHIFT; 401 402 dev_dbg(&context->dev->pdev->dev, "create mmap region\n"); 403 404 if ((size != PAGE_SIZE) || (offset & ~PAGE_MASK)) { 405 dev_warn(&context->dev->pdev->dev, 406 "invalid params for mmap region\n"); 407 return -EINVAL; 408 } 409 410 /* Map UAR to kernel space, VM_LOCKED? */ 411 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND; 412 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); 413 if (io_remap_pfn_range(vma, start, context->uar.pfn, size, 414 vma->vm_page_prot)) 415 return -EAGAIN; 416 417 return 0; 418 } 419 420 /** 421 * pvrdma_alloc_pd - allocate protection domain 422 * @ibdev: the IB device 423 * @context: user context 424 * @udata: user data 425 * 426 * @return: the ib_pd protection domain pointer on success, otherwise errno. 427 */ 428 struct ib_pd *pvrdma_alloc_pd(struct ib_device *ibdev, 429 struct ib_ucontext *context, 430 struct ib_udata *udata) 431 { 432 struct pvrdma_pd *pd; 433 struct pvrdma_dev *dev = to_vdev(ibdev); 434 union pvrdma_cmd_req req; 435 union pvrdma_cmd_resp rsp; 436 struct pvrdma_cmd_create_pd *cmd = &req.create_pd; 437 struct pvrdma_cmd_create_pd_resp *resp = &rsp.create_pd_resp; 438 int ret; 439 void *ptr; 440 441 /* Check allowed max pds */ 442 if (!atomic_add_unless(&dev->num_pds, 1, dev->dsr->caps.max_pd)) 443 return ERR_PTR(-ENOMEM); 444 445 pd = kmalloc(sizeof(*pd), GFP_KERNEL); 446 if (!pd) { 447 ptr = ERR_PTR(-ENOMEM); 448 goto err; 449 } 450 451 memset(cmd, 0, sizeof(*cmd)); 452 cmd->hdr.cmd = PVRDMA_CMD_CREATE_PD; 453 cmd->ctx_handle = (context) ? to_vucontext(context)->ctx_handle : 0; 454 ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_CREATE_PD_RESP); 455 if (ret < 0) { 456 dev_warn(&dev->pdev->dev, 457 "failed to allocate protection domain, error: %d\n", 458 ret); 459 ptr = ERR_PTR(ret); 460 goto freepd; 461 } 462 463 pd->privileged = !context; 464 pd->pd_handle = resp->pd_handle; 465 pd->pdn = resp->pd_handle; 466 467 if (context) { 468 if (ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) { 469 dev_warn(&dev->pdev->dev, 470 "failed to copy back protection domain\n"); 471 pvrdma_dealloc_pd(&pd->ibpd); 472 return ERR_PTR(-EFAULT); 473 } 474 } 475 476 /* u32 pd handle */ 477 return &pd->ibpd; 478 479 freepd: 480 kfree(pd); 481 err: 482 atomic_dec(&dev->num_pds); 483 return ptr; 484 } 485 486 /** 487 * pvrdma_dealloc_pd - deallocate protection domain 488 * @pd: the protection domain to be released 489 * 490 * @return: 0 on success, otherwise errno. 491 */ 492 int pvrdma_dealloc_pd(struct ib_pd *pd) 493 { 494 struct pvrdma_dev *dev = to_vdev(pd->device); 495 union pvrdma_cmd_req req; 496 struct pvrdma_cmd_destroy_pd *cmd = &req.destroy_pd; 497 int ret; 498 499 memset(cmd, 0, sizeof(*cmd)); 500 cmd->hdr.cmd = PVRDMA_CMD_DESTROY_PD; 501 cmd->pd_handle = to_vpd(pd)->pd_handle; 502 503 ret = pvrdma_cmd_post(dev, &req, NULL, 0); 504 if (ret) 505 dev_warn(&dev->pdev->dev, 506 "could not dealloc protection domain, error: %d\n", 507 ret); 508 509 kfree(to_vpd(pd)); 510 atomic_dec(&dev->num_pds); 511 512 return 0; 513 } 514 515 /** 516 * pvrdma_create_ah - create an address handle 517 * @pd: the protection domain 518 * @ah_attr: the attributes of the AH 519 * @udata: user data blob 520 * 521 * @return: the ib_ah pointer on success, otherwise errno. 522 */ 523 struct ib_ah *pvrdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr, 524 struct ib_udata *udata) 525 { 526 struct pvrdma_dev *dev = to_vdev(pd->device); 527 struct pvrdma_ah *ah; 528 const struct ib_global_route *grh; 529 u8 port_num = rdma_ah_get_port_num(ah_attr); 530 531 if (!(rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH)) 532 return ERR_PTR(-EINVAL); 533 534 grh = rdma_ah_read_grh(ah_attr); 535 if ((ah_attr->type != RDMA_AH_ATTR_TYPE_ROCE) || 536 rdma_is_multicast_addr((struct in6_addr *)grh->dgid.raw)) 537 return ERR_PTR(-EINVAL); 538 539 if (!atomic_add_unless(&dev->num_ahs, 1, dev->dsr->caps.max_ah)) 540 return ERR_PTR(-ENOMEM); 541 542 ah = kzalloc(sizeof(*ah), GFP_KERNEL); 543 if (!ah) { 544 atomic_dec(&dev->num_ahs); 545 return ERR_PTR(-ENOMEM); 546 } 547 548 ah->av.port_pd = to_vpd(pd)->pd_handle | (port_num << 24); 549 ah->av.src_path_bits = rdma_ah_get_path_bits(ah_attr); 550 ah->av.src_path_bits |= 0x80; 551 ah->av.gid_index = grh->sgid_index; 552 ah->av.hop_limit = grh->hop_limit; 553 ah->av.sl_tclass_flowlabel = (grh->traffic_class << 20) | 554 grh->flow_label; 555 memcpy(ah->av.dgid, grh->dgid.raw, 16); 556 memcpy(ah->av.dmac, ah_attr->roce.dmac, ETH_ALEN); 557 558 ah->ibah.device = pd->device; 559 ah->ibah.pd = pd; 560 ah->ibah.uobject = NULL; 561 562 return &ah->ibah; 563 } 564 565 /** 566 * pvrdma_destroy_ah - destroy an address handle 567 * @ah: the address handle to destroyed 568 * 569 * @return: 0 on success. 570 */ 571 int pvrdma_destroy_ah(struct ib_ah *ah) 572 { 573 struct pvrdma_dev *dev = to_vdev(ah->device); 574 575 kfree(to_vah(ah)); 576 atomic_dec(&dev->num_ahs); 577 578 return 0; 579 } 580