Linux 4.14.13
[linux-stable.git] / drivers / infiniband / hw / vmw_pvrdma / pvrdma_verbs.c
blob48776f5ffb0e105cdb0a5fb6e52a7ddd663e04e9
1 /*
2 * Copyright (c) 2012-2016 VMware, Inc. All rights reserved.
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.
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.
17 * The BSD 2-Clause License
19 * Redistribution and use in source and binary forms, with or
20 * without modification, are permitted provided that the following
21 * conditions are met:
23 * - Redistributions of source code must retain the above
24 * copyright notice, this list of conditions and the following
25 * disclaimer.
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.
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.
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>
54 #include "pvrdma.h"
56 /**
57 * pvrdma_query_device - query device
58 * @ibdev: the device to query
59 * @props: the device properties
60 * @uhw: user data
62 * @return: 0 on success, otherwise negative errno
64 int pvrdma_query_device(struct ib_device *ibdev,
65 struct ib_device_attr *props,
66 struct ib_udata *uhw)
68 struct pvrdma_dev *dev = to_vdev(ibdev);
70 if (uhw->inlen || uhw->outlen)
71 return -EINVAL;
73 memset(props, 0, sizeof(*props));
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_sge_rd = PVRDMA_GET_CAP(dev, dev->dsr->caps.max_sge,
87 dev->dsr->caps.max_sge_rd);
88 props->max_cq = dev->dsr->caps.max_cq;
89 props->max_cqe = dev->dsr->caps.max_cqe;
90 props->max_mr = dev->dsr->caps.max_mr;
91 props->max_pd = dev->dsr->caps.max_pd;
92 props->max_qp_rd_atom = dev->dsr->caps.max_qp_rd_atom;
93 props->max_qp_init_rd_atom = dev->dsr->caps.max_qp_init_rd_atom;
94 props->atomic_cap =
95 dev->dsr->caps.atomic_ops &
96 (PVRDMA_ATOMIC_OP_COMP_SWAP | PVRDMA_ATOMIC_OP_FETCH_ADD) ?
97 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
98 props->masked_atomic_cap = props->atomic_cap;
99 props->max_ah = dev->dsr->caps.max_ah;
100 props->max_pkeys = dev->dsr->caps.max_pkeys;
101 props->local_ca_ack_delay = dev->dsr->caps.local_ca_ack_delay;
102 if ((dev->dsr->caps.bmme_flags & PVRDMA_BMME_FLAG_LOCAL_INV) &&
103 (dev->dsr->caps.bmme_flags & PVRDMA_BMME_FLAG_REMOTE_INV) &&
104 (dev->dsr->caps.bmme_flags & PVRDMA_BMME_FLAG_FAST_REG_WR)) {
105 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
106 props->max_fast_reg_page_list_len = PVRDMA_GET_CAP(dev,
107 PVRDMA_MAX_FAST_REG_PAGES,
108 dev->dsr->caps.max_fast_reg_page_list_len);
111 props->device_cap_flags |= IB_DEVICE_PORT_ACTIVE_EVENT |
112 IB_DEVICE_RC_RNR_NAK_GEN;
114 return 0;
118 * pvrdma_query_port - query device port attributes
119 * @ibdev: the device to query
120 * @port: the port number
121 * @props: the device properties
123 * @return: 0 on success, otherwise negative errno
125 int pvrdma_query_port(struct ib_device *ibdev, u8 port,
126 struct ib_port_attr *props)
128 struct pvrdma_dev *dev = to_vdev(ibdev);
129 union pvrdma_cmd_req req;
130 union pvrdma_cmd_resp rsp;
131 struct pvrdma_cmd_query_port *cmd = &req.query_port;
132 struct pvrdma_cmd_query_port_resp *resp = &rsp.query_port_resp;
133 int err;
135 memset(cmd, 0, sizeof(*cmd));
136 cmd->hdr.cmd = PVRDMA_CMD_QUERY_PORT;
137 cmd->port_num = port;
139 err = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_QUERY_PORT_RESP);
140 if (err < 0) {
141 dev_warn(&dev->pdev->dev,
142 "could not query port, error: %d\n", err);
143 return err;
146 /* props being zeroed by the caller, avoid zeroing it here */
148 props->state = pvrdma_port_state_to_ib(resp->attrs.state);
149 props->max_mtu = pvrdma_mtu_to_ib(resp->attrs.max_mtu);
150 props->active_mtu = pvrdma_mtu_to_ib(resp->attrs.active_mtu);
151 props->gid_tbl_len = resp->attrs.gid_tbl_len;
152 props->port_cap_flags =
153 pvrdma_port_cap_flags_to_ib(resp->attrs.port_cap_flags);
154 props->port_cap_flags |= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
155 props->max_msg_sz = resp->attrs.max_msg_sz;
156 props->bad_pkey_cntr = resp->attrs.bad_pkey_cntr;
157 props->qkey_viol_cntr = resp->attrs.qkey_viol_cntr;
158 props->pkey_tbl_len = resp->attrs.pkey_tbl_len;
159 props->lid = resp->attrs.lid;
160 props->sm_lid = resp->attrs.sm_lid;
161 props->lmc = resp->attrs.lmc;
162 props->max_vl_num = resp->attrs.max_vl_num;
163 props->sm_sl = resp->attrs.sm_sl;
164 props->subnet_timeout = resp->attrs.subnet_timeout;
165 props->init_type_reply = resp->attrs.init_type_reply;
166 props->active_width = pvrdma_port_width_to_ib(resp->attrs.active_width);
167 props->active_speed = pvrdma_port_speed_to_ib(resp->attrs.active_speed);
168 props->phys_state = resp->attrs.phys_state;
170 return 0;
174 * pvrdma_query_gid - query device gid
175 * @ibdev: the device to query
176 * @port: the port number
177 * @index: the index
178 * @gid: the device gid value
180 * @return: 0 on success, otherwise negative errno
182 int pvrdma_query_gid(struct ib_device *ibdev, u8 port, int index,
183 union ib_gid *gid)
185 struct pvrdma_dev *dev = to_vdev(ibdev);
187 if (index >= dev->dsr->caps.gid_tbl_len)
188 return -EINVAL;
190 memcpy(gid, &dev->sgid_tbl[index], sizeof(union ib_gid));
192 return 0;
196 * pvrdma_query_pkey - query device port's P_Key table
197 * @ibdev: the device to query
198 * @port: the port number
199 * @index: the index
200 * @pkey: the device P_Key value
202 * @return: 0 on success, otherwise negative errno
204 int pvrdma_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
205 u16 *pkey)
207 int err = 0;
208 union pvrdma_cmd_req req;
209 union pvrdma_cmd_resp rsp;
210 struct pvrdma_cmd_query_pkey *cmd = &req.query_pkey;
212 memset(cmd, 0, sizeof(*cmd));
213 cmd->hdr.cmd = PVRDMA_CMD_QUERY_PKEY;
214 cmd->port_num = port;
215 cmd->index = index;
217 err = pvrdma_cmd_post(to_vdev(ibdev), &req, &rsp,
218 PVRDMA_CMD_QUERY_PKEY_RESP);
219 if (err < 0) {
220 dev_warn(&to_vdev(ibdev)->pdev->dev,
221 "could not query pkey, error: %d\n", err);
222 return err;
225 *pkey = rsp.query_pkey_resp.pkey;
227 return 0;
230 enum rdma_link_layer pvrdma_port_link_layer(struct ib_device *ibdev,
231 u8 port)
233 return IB_LINK_LAYER_ETHERNET;
236 int pvrdma_modify_device(struct ib_device *ibdev, int mask,
237 struct ib_device_modify *props)
239 unsigned long flags;
241 if (mask & ~(IB_DEVICE_MODIFY_SYS_IMAGE_GUID |
242 IB_DEVICE_MODIFY_NODE_DESC)) {
243 dev_warn(&to_vdev(ibdev)->pdev->dev,
244 "unsupported device modify mask %#x\n", mask);
245 return -EOPNOTSUPP;
248 if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
249 spin_lock_irqsave(&to_vdev(ibdev)->desc_lock, flags);
250 memcpy(ibdev->node_desc, props->node_desc, 64);
251 spin_unlock_irqrestore(&to_vdev(ibdev)->desc_lock, flags);
254 if (mask & IB_DEVICE_MODIFY_SYS_IMAGE_GUID) {
255 mutex_lock(&to_vdev(ibdev)->port_mutex);
256 to_vdev(ibdev)->sys_image_guid =
257 cpu_to_be64(props->sys_image_guid);
258 mutex_unlock(&to_vdev(ibdev)->port_mutex);
261 return 0;
265 * pvrdma_modify_port - modify device port attributes
266 * @ibdev: the device to modify
267 * @port: the port number
268 * @mask: attributes to modify
269 * @props: the device properties
271 * @return: 0 on success, otherwise negative errno
273 int pvrdma_modify_port(struct ib_device *ibdev, u8 port, int mask,
274 struct ib_port_modify *props)
276 struct ib_port_attr attr;
277 struct pvrdma_dev *vdev = to_vdev(ibdev);
278 int ret;
280 if (mask & ~IB_PORT_SHUTDOWN) {
281 dev_warn(&vdev->pdev->dev,
282 "unsupported port modify mask %#x\n", mask);
283 return -EOPNOTSUPP;
286 mutex_lock(&vdev->port_mutex);
287 ret = ib_query_port(ibdev, port, &attr);
288 if (ret)
289 goto out;
291 vdev->port_cap_mask |= props->set_port_cap_mask;
292 vdev->port_cap_mask &= ~props->clr_port_cap_mask;
294 if (mask & IB_PORT_SHUTDOWN)
295 vdev->ib_active = false;
297 out:
298 mutex_unlock(&vdev->port_mutex);
299 return ret;
303 * pvrdma_alloc_ucontext - allocate ucontext
304 * @ibdev: the IB device
305 * @udata: user data
307 * @return: the ib_ucontext pointer on success, otherwise errno.
309 struct ib_ucontext *pvrdma_alloc_ucontext(struct ib_device *ibdev,
310 struct ib_udata *udata)
312 struct pvrdma_dev *vdev = to_vdev(ibdev);
313 struct pvrdma_ucontext *context;
314 union pvrdma_cmd_req req;
315 union pvrdma_cmd_resp rsp;
316 struct pvrdma_cmd_create_uc *cmd = &req.create_uc;
317 struct pvrdma_cmd_create_uc_resp *resp = &rsp.create_uc_resp;
318 struct pvrdma_alloc_ucontext_resp uresp = {0};
319 int ret;
320 void *ptr;
322 if (!vdev->ib_active)
323 return ERR_PTR(-EAGAIN);
325 context = kmalloc(sizeof(*context), GFP_KERNEL);
326 if (!context)
327 return ERR_PTR(-ENOMEM);
329 context->dev = vdev;
330 ret = pvrdma_uar_alloc(vdev, &context->uar);
331 if (ret) {
332 kfree(context);
333 return ERR_PTR(-ENOMEM);
336 /* get ctx_handle from host */
337 memset(cmd, 0, sizeof(*cmd));
338 cmd->pfn = context->uar.pfn;
339 cmd->hdr.cmd = PVRDMA_CMD_CREATE_UC;
340 ret = pvrdma_cmd_post(vdev, &req, &rsp, PVRDMA_CMD_CREATE_UC_RESP);
341 if (ret < 0) {
342 dev_warn(&vdev->pdev->dev,
343 "could not create ucontext, error: %d\n", ret);
344 ptr = ERR_PTR(ret);
345 goto err;
348 context->ctx_handle = resp->ctx_handle;
350 /* copy back to user */
351 uresp.qp_tab_size = vdev->dsr->caps.max_qp;
352 ret = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
353 if (ret) {
354 pvrdma_uar_free(vdev, &context->uar);
355 context->ibucontext.device = ibdev;
356 pvrdma_dealloc_ucontext(&context->ibucontext);
357 return ERR_PTR(-EFAULT);
360 return &context->ibucontext;
362 err:
363 pvrdma_uar_free(vdev, &context->uar);
364 kfree(context);
365 return ptr;
369 * pvrdma_dealloc_ucontext - deallocate ucontext
370 * @ibcontext: the ucontext
372 * @return: 0 on success, otherwise errno.
374 int pvrdma_dealloc_ucontext(struct ib_ucontext *ibcontext)
376 struct pvrdma_ucontext *context = to_vucontext(ibcontext);
377 union pvrdma_cmd_req req;
378 struct pvrdma_cmd_destroy_uc *cmd = &req.destroy_uc;
379 int ret;
381 memset(cmd, 0, sizeof(*cmd));
382 cmd->hdr.cmd = PVRDMA_CMD_DESTROY_UC;
383 cmd->ctx_handle = context->ctx_handle;
385 ret = pvrdma_cmd_post(context->dev, &req, NULL, 0);
386 if (ret < 0)
387 dev_warn(&context->dev->pdev->dev,
388 "destroy ucontext failed, error: %d\n", ret);
390 /* Free the UAR even if the device command failed */
391 pvrdma_uar_free(to_vdev(ibcontext->device), &context->uar);
392 kfree(context);
394 return ret;
398 * pvrdma_mmap - create mmap region
399 * @ibcontext: the user context
400 * @vma: the VMA
402 * @return: 0 on success, otherwise errno.
404 int pvrdma_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vma)
406 struct pvrdma_ucontext *context = to_vucontext(ibcontext);
407 unsigned long start = vma->vm_start;
408 unsigned long size = vma->vm_end - vma->vm_start;
409 unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
411 dev_dbg(&context->dev->pdev->dev, "create mmap region\n");
413 if ((size != PAGE_SIZE) || (offset & ~PAGE_MASK)) {
414 dev_warn(&context->dev->pdev->dev,
415 "invalid params for mmap region\n");
416 return -EINVAL;
419 /* Map UAR to kernel space, VM_LOCKED? */
420 vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
421 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
422 if (io_remap_pfn_range(vma, start, context->uar.pfn, size,
423 vma->vm_page_prot))
424 return -EAGAIN;
426 return 0;
430 * pvrdma_alloc_pd - allocate protection domain
431 * @ibdev: the IB device
432 * @context: user context
433 * @udata: user data
435 * @return: the ib_pd protection domain pointer on success, otherwise errno.
437 struct ib_pd *pvrdma_alloc_pd(struct ib_device *ibdev,
438 struct ib_ucontext *context,
439 struct ib_udata *udata)
441 struct pvrdma_pd *pd;
442 struct pvrdma_dev *dev = to_vdev(ibdev);
443 union pvrdma_cmd_req req;
444 union pvrdma_cmd_resp rsp;
445 struct pvrdma_cmd_create_pd *cmd = &req.create_pd;
446 struct pvrdma_cmd_create_pd_resp *resp = &rsp.create_pd_resp;
447 int ret;
448 void *ptr;
450 /* Check allowed max pds */
451 if (!atomic_add_unless(&dev->num_pds, 1, dev->dsr->caps.max_pd))
452 return ERR_PTR(-ENOMEM);
454 pd = kmalloc(sizeof(*pd), GFP_KERNEL);
455 if (!pd) {
456 ptr = ERR_PTR(-ENOMEM);
457 goto err;
460 memset(cmd, 0, sizeof(*cmd));
461 cmd->hdr.cmd = PVRDMA_CMD_CREATE_PD;
462 cmd->ctx_handle = (context) ? to_vucontext(context)->ctx_handle : 0;
463 ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_CREATE_PD_RESP);
464 if (ret < 0) {
465 dev_warn(&dev->pdev->dev,
466 "failed to allocate protection domain, error: %d\n",
467 ret);
468 ptr = ERR_PTR(ret);
469 goto freepd;
472 pd->privileged = !context;
473 pd->pd_handle = resp->pd_handle;
474 pd->pdn = resp->pd_handle;
476 if (context) {
477 if (ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) {
478 dev_warn(&dev->pdev->dev,
479 "failed to copy back protection domain\n");
480 pvrdma_dealloc_pd(&pd->ibpd);
481 return ERR_PTR(-EFAULT);
485 /* u32 pd handle */
486 return &pd->ibpd;
488 freepd:
489 kfree(pd);
490 err:
491 atomic_dec(&dev->num_pds);
492 return ptr;
496 * pvrdma_dealloc_pd - deallocate protection domain
497 * @pd: the protection domain to be released
499 * @return: 0 on success, otherwise errno.
501 int pvrdma_dealloc_pd(struct ib_pd *pd)
503 struct pvrdma_dev *dev = to_vdev(pd->device);
504 union pvrdma_cmd_req req;
505 struct pvrdma_cmd_destroy_pd *cmd = &req.destroy_pd;
506 int ret;
508 memset(cmd, 0, sizeof(*cmd));
509 cmd->hdr.cmd = PVRDMA_CMD_DESTROY_PD;
510 cmd->pd_handle = to_vpd(pd)->pd_handle;
512 ret = pvrdma_cmd_post(dev, &req, NULL, 0);
513 if (ret)
514 dev_warn(&dev->pdev->dev,
515 "could not dealloc protection domain, error: %d\n",
516 ret);
518 kfree(to_vpd(pd));
519 atomic_dec(&dev->num_pds);
521 return 0;
525 * pvrdma_create_ah - create an address handle
526 * @pd: the protection domain
527 * @ah_attr: the attributes of the AH
528 * @udata: user data blob
530 * @return: the ib_ah pointer on success, otherwise errno.
532 struct ib_ah *pvrdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
533 struct ib_udata *udata)
535 struct pvrdma_dev *dev = to_vdev(pd->device);
536 struct pvrdma_ah *ah;
537 const struct ib_global_route *grh;
538 u8 port_num = rdma_ah_get_port_num(ah_attr);
540 if (!(rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH))
541 return ERR_PTR(-EINVAL);
543 grh = rdma_ah_read_grh(ah_attr);
544 if ((ah_attr->type != RDMA_AH_ATTR_TYPE_ROCE) ||
545 rdma_is_multicast_addr((struct in6_addr *)grh->dgid.raw))
546 return ERR_PTR(-EINVAL);
548 if (!atomic_add_unless(&dev->num_ahs, 1, dev->dsr->caps.max_ah))
549 return ERR_PTR(-ENOMEM);
551 ah = kzalloc(sizeof(*ah), GFP_KERNEL);
552 if (!ah) {
553 atomic_dec(&dev->num_ahs);
554 return ERR_PTR(-ENOMEM);
557 ah->av.port_pd = to_vpd(pd)->pd_handle | (port_num << 24);
558 ah->av.src_path_bits = rdma_ah_get_path_bits(ah_attr);
559 ah->av.src_path_bits |= 0x80;
560 ah->av.gid_index = grh->sgid_index;
561 ah->av.hop_limit = grh->hop_limit;
562 ah->av.sl_tclass_flowlabel = (grh->traffic_class << 20) |
563 grh->flow_label;
564 memcpy(ah->av.dgid, grh->dgid.raw, 16);
565 memcpy(ah->av.dmac, ah_attr->roce.dmac, ETH_ALEN);
567 ah->ibah.device = pd->device;
568 ah->ibah.pd = pd;
569 ah->ibah.uobject = NULL;
571 return &ah->ibah;
575 * pvrdma_destroy_ah - destroy an address handle
576 * @ah: the address handle to destroyed
578 * @return: 0 on success.
580 int pvrdma_destroy_ah(struct ib_ah *ah)
582 struct pvrdma_dev *dev = to_vdev(ah->device);
584 kfree(to_vah(ah));
585 atomic_dec(&dev->num_ahs);
587 return 0;