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  */
pvrdma_query_device(struct ib_device * ibdev,struct ib_device_attr * props,struct ib_udata * uhw)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_send_sge = dev->dsr->caps.max_sge;
86 	props->max_recv_sge = dev->dsr->caps.max_sge;
87 	props->max_sge_rd = PVRDMA_GET_CAP(dev, dev->dsr->caps.max_sge,
88 					   dev->dsr->caps.max_sge_rd);
89 	props->max_srq = dev->dsr->caps.max_srq;
90 	props->max_srq_wr = dev->dsr->caps.max_srq_wr;
91 	props->max_srq_sge = dev->dsr->caps.max_srq_sge;
92 	props->max_cq = dev->dsr->caps.max_cq;
93 	props->max_cqe = dev->dsr->caps.max_cqe;
94 	props->max_mr = dev->dsr->caps.max_mr;
95 	props->max_pd = dev->dsr->caps.max_pd;
96 	props->max_qp_rd_atom = dev->dsr->caps.max_qp_rd_atom;
97 	props->max_qp_init_rd_atom = dev->dsr->caps.max_qp_init_rd_atom;
98 	props->atomic_cap =
99 		dev->dsr->caps.atomic_ops &
100 		(PVRDMA_ATOMIC_OP_COMP_SWAP | PVRDMA_ATOMIC_OP_FETCH_ADD) ?
101 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
102 	props->masked_atomic_cap = props->atomic_cap;
103 	props->max_ah = dev->dsr->caps.max_ah;
104 	props->max_pkeys = dev->dsr->caps.max_pkeys;
105 	props->local_ca_ack_delay = dev->dsr->caps.local_ca_ack_delay;
106 	if ((dev->dsr->caps.bmme_flags & PVRDMA_BMME_FLAG_LOCAL_INV) &&
107 	    (dev->dsr->caps.bmme_flags & PVRDMA_BMME_FLAG_REMOTE_INV) &&
108 	    (dev->dsr->caps.bmme_flags & PVRDMA_BMME_FLAG_FAST_REG_WR)) {
109 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
110 		props->max_fast_reg_page_list_len = PVRDMA_GET_CAP(dev,
111 				PVRDMA_MAX_FAST_REG_PAGES,
112 				dev->dsr->caps.max_fast_reg_page_list_len);
113 	}
114 
115 	props->device_cap_flags |= IB_DEVICE_PORT_ACTIVE_EVENT |
116 				   IB_DEVICE_RC_RNR_NAK_GEN;
117 
118 	return 0;
119 }
120 
121 /**
122  * pvrdma_query_port - query device port attributes
123  * @ibdev: the device to query
124  * @port: the port number
125  * @props: the device properties
126  *
127  * @return: 0 on success, otherwise negative errno
128  */
pvrdma_query_port(struct ib_device * ibdev,u8 port,struct ib_port_attr * props)129 int pvrdma_query_port(struct ib_device *ibdev, u8 port,
130 		      struct ib_port_attr *props)
131 {
132 	struct pvrdma_dev *dev = to_vdev(ibdev);
133 	union pvrdma_cmd_req req;
134 	union pvrdma_cmd_resp rsp;
135 	struct pvrdma_cmd_query_port *cmd = &req.query_port;
136 	struct pvrdma_cmd_query_port_resp *resp = &rsp.query_port_resp;
137 	int err;
138 
139 	memset(cmd, 0, sizeof(*cmd));
140 	cmd->hdr.cmd = PVRDMA_CMD_QUERY_PORT;
141 	cmd->port_num = port;
142 
143 	err = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_QUERY_PORT_RESP);
144 	if (err < 0) {
145 		dev_warn(&dev->pdev->dev,
146 			 "could not query port, error: %d\n", err);
147 		return err;
148 	}
149 
150 	/* props being zeroed by the caller, avoid zeroing it here */
151 
152 	props->state = pvrdma_port_state_to_ib(resp->attrs.state);
153 	props->max_mtu = pvrdma_mtu_to_ib(resp->attrs.max_mtu);
154 	props->active_mtu = pvrdma_mtu_to_ib(resp->attrs.active_mtu);
155 	props->gid_tbl_len = resp->attrs.gid_tbl_len;
156 	props->port_cap_flags =
157 		pvrdma_port_cap_flags_to_ib(resp->attrs.port_cap_flags);
158 	props->port_cap_flags |= IB_PORT_CM_SUP;
159 	props->ip_gids = true;
160 	props->max_msg_sz = resp->attrs.max_msg_sz;
161 	props->bad_pkey_cntr = resp->attrs.bad_pkey_cntr;
162 	props->qkey_viol_cntr = resp->attrs.qkey_viol_cntr;
163 	props->pkey_tbl_len = resp->attrs.pkey_tbl_len;
164 	props->lid = resp->attrs.lid;
165 	props->sm_lid = resp->attrs.sm_lid;
166 	props->lmc = resp->attrs.lmc;
167 	props->max_vl_num = resp->attrs.max_vl_num;
168 	props->sm_sl = resp->attrs.sm_sl;
169 	props->subnet_timeout = resp->attrs.subnet_timeout;
170 	props->init_type_reply = resp->attrs.init_type_reply;
171 	props->active_width = pvrdma_port_width_to_ib(resp->attrs.active_width);
172 	props->active_speed = pvrdma_port_speed_to_ib(resp->attrs.active_speed);
173 	props->phys_state = resp->attrs.phys_state;
174 
175 	return 0;
176 }
177 
178 /**
179  * pvrdma_query_gid - query device gid
180  * @ibdev: the device to query
181  * @port: the port number
182  * @index: the index
183  * @gid: the device gid value
184  *
185  * @return: 0 on success, otherwise negative errno
186  */
pvrdma_query_gid(struct ib_device * ibdev,u8 port,int index,union ib_gid * gid)187 int pvrdma_query_gid(struct ib_device *ibdev, u8 port, int index,
188 		     union ib_gid *gid)
189 {
190 	struct pvrdma_dev *dev = to_vdev(ibdev);
191 
192 	if (index >= dev->dsr->caps.gid_tbl_len)
193 		return -EINVAL;
194 
195 	memcpy(gid, &dev->sgid_tbl[index], sizeof(union ib_gid));
196 
197 	return 0;
198 }
199 
200 /**
201  * pvrdma_query_pkey - query device port's P_Key table
202  * @ibdev: the device to query
203  * @port: the port number
204  * @index: the index
205  * @pkey: the device P_Key value
206  *
207  * @return: 0 on success, otherwise negative errno
208  */
pvrdma_query_pkey(struct ib_device * ibdev,u8 port,u16 index,u16 * pkey)209 int pvrdma_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
210 		      u16 *pkey)
211 {
212 	int err = 0;
213 	union pvrdma_cmd_req req;
214 	union pvrdma_cmd_resp rsp;
215 	struct pvrdma_cmd_query_pkey *cmd = &req.query_pkey;
216 
217 	memset(cmd, 0, sizeof(*cmd));
218 	cmd->hdr.cmd = PVRDMA_CMD_QUERY_PKEY;
219 	cmd->port_num = port;
220 	cmd->index = index;
221 
222 	err = pvrdma_cmd_post(to_vdev(ibdev), &req, &rsp,
223 			      PVRDMA_CMD_QUERY_PKEY_RESP);
224 	if (err < 0) {
225 		dev_warn(&to_vdev(ibdev)->pdev->dev,
226 			 "could not query pkey, error: %d\n", err);
227 		return err;
228 	}
229 
230 	*pkey = rsp.query_pkey_resp.pkey;
231 
232 	return 0;
233 }
234 
pvrdma_port_link_layer(struct ib_device * ibdev,u8 port)235 enum rdma_link_layer pvrdma_port_link_layer(struct ib_device *ibdev,
236 					    u8 port)
237 {
238 	return IB_LINK_LAYER_ETHERNET;
239 }
240 
pvrdma_modify_device(struct ib_device * ibdev,int mask,struct ib_device_modify * props)241 int pvrdma_modify_device(struct ib_device *ibdev, int mask,
242 			 struct ib_device_modify *props)
243 {
244 	unsigned long flags;
245 
246 	if (mask & ~(IB_DEVICE_MODIFY_SYS_IMAGE_GUID |
247 		     IB_DEVICE_MODIFY_NODE_DESC)) {
248 		dev_warn(&to_vdev(ibdev)->pdev->dev,
249 			 "unsupported device modify mask %#x\n", mask);
250 		return -EOPNOTSUPP;
251 	}
252 
253 	if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
254 		spin_lock_irqsave(&to_vdev(ibdev)->desc_lock, flags);
255 		memcpy(ibdev->node_desc, props->node_desc, 64);
256 		spin_unlock_irqrestore(&to_vdev(ibdev)->desc_lock, flags);
257 	}
258 
259 	if (mask & IB_DEVICE_MODIFY_SYS_IMAGE_GUID) {
260 		mutex_lock(&to_vdev(ibdev)->port_mutex);
261 		to_vdev(ibdev)->sys_image_guid =
262 			cpu_to_be64(props->sys_image_guid);
263 		mutex_unlock(&to_vdev(ibdev)->port_mutex);
264 	}
265 
266 	return 0;
267 }
268 
269 /**
270  * pvrdma_modify_port - modify device port attributes
271  * @ibdev: the device to modify
272  * @port: the port number
273  * @mask: attributes to modify
274  * @props: the device properties
275  *
276  * @return: 0 on success, otherwise negative errno
277  */
pvrdma_modify_port(struct ib_device * ibdev,u8 port,int mask,struct ib_port_modify * props)278 int pvrdma_modify_port(struct ib_device *ibdev, u8 port, int mask,
279 		       struct ib_port_modify *props)
280 {
281 	struct ib_port_attr attr;
282 	struct pvrdma_dev *vdev = to_vdev(ibdev);
283 	int ret;
284 
285 	if (mask & ~IB_PORT_SHUTDOWN) {
286 		dev_warn(&vdev->pdev->dev,
287 			 "unsupported port modify mask %#x\n", mask);
288 		return -EOPNOTSUPP;
289 	}
290 
291 	mutex_lock(&vdev->port_mutex);
292 	ret = ib_query_port(ibdev, port, &attr);
293 	if (ret)
294 		goto out;
295 
296 	vdev->port_cap_mask |= props->set_port_cap_mask;
297 	vdev->port_cap_mask &= ~props->clr_port_cap_mask;
298 
299 	if (mask & IB_PORT_SHUTDOWN)
300 		vdev->ib_active = false;
301 
302 out:
303 	mutex_unlock(&vdev->port_mutex);
304 	return ret;
305 }
306 
307 /**
308  * pvrdma_alloc_ucontext - allocate ucontext
309  * @ibdev: the IB device
310  * @udata: user data
311  *
312  * @return: the ib_ucontext pointer on success, otherwise errno.
313  */
pvrdma_alloc_ucontext(struct ib_device * ibdev,struct ib_udata * udata)314 struct ib_ucontext *pvrdma_alloc_ucontext(struct ib_device *ibdev,
315 					  struct ib_udata *udata)
316 {
317 	struct pvrdma_dev *vdev = to_vdev(ibdev);
318 	struct pvrdma_ucontext *context;
319 	union pvrdma_cmd_req req;
320 	union pvrdma_cmd_resp rsp;
321 	struct pvrdma_cmd_create_uc *cmd = &req.create_uc;
322 	struct pvrdma_cmd_create_uc_resp *resp = &rsp.create_uc_resp;
323 	struct pvrdma_alloc_ucontext_resp uresp = {0};
324 	int ret;
325 	void *ptr;
326 
327 	if (!vdev->ib_active)
328 		return ERR_PTR(-EAGAIN);
329 
330 	context = kmalloc(sizeof(*context), GFP_KERNEL);
331 	if (!context)
332 		return ERR_PTR(-ENOMEM);
333 
334 	context->dev = vdev;
335 	ret = pvrdma_uar_alloc(vdev, &context->uar);
336 	if (ret) {
337 		kfree(context);
338 		return ERR_PTR(-ENOMEM);
339 	}
340 
341 	/* get ctx_handle from host */
342 	memset(cmd, 0, sizeof(*cmd));
343 	cmd->pfn = context->uar.pfn;
344 	cmd->hdr.cmd = PVRDMA_CMD_CREATE_UC;
345 	ret = pvrdma_cmd_post(vdev, &req, &rsp, PVRDMA_CMD_CREATE_UC_RESP);
346 	if (ret < 0) {
347 		dev_warn(&vdev->pdev->dev,
348 			 "could not create ucontext, error: %d\n", ret);
349 		ptr = ERR_PTR(ret);
350 		goto err;
351 	}
352 
353 	context->ctx_handle = resp->ctx_handle;
354 
355 	/* copy back to user */
356 	uresp.qp_tab_size = vdev->dsr->caps.max_qp;
357 	ret = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
358 	if (ret) {
359 		pvrdma_uar_free(vdev, &context->uar);
360 		context->ibucontext.device = ibdev;
361 		pvrdma_dealloc_ucontext(&context->ibucontext);
362 		return ERR_PTR(-EFAULT);
363 	}
364 
365 	return &context->ibucontext;
366 
367 err:
368 	pvrdma_uar_free(vdev, &context->uar);
369 	kfree(context);
370 	return ptr;
371 }
372 
373 /**
374  * pvrdma_dealloc_ucontext - deallocate ucontext
375  * @ibcontext: the ucontext
376  *
377  * @return: 0 on success, otherwise errno.
378  */
pvrdma_dealloc_ucontext(struct ib_ucontext * ibcontext)379 int pvrdma_dealloc_ucontext(struct ib_ucontext *ibcontext)
380 {
381 	struct pvrdma_ucontext *context = to_vucontext(ibcontext);
382 	union pvrdma_cmd_req req;
383 	struct pvrdma_cmd_destroy_uc *cmd = &req.destroy_uc;
384 	int ret;
385 
386 	memset(cmd, 0, sizeof(*cmd));
387 	cmd->hdr.cmd = PVRDMA_CMD_DESTROY_UC;
388 	cmd->ctx_handle = context->ctx_handle;
389 
390 	ret = pvrdma_cmd_post(context->dev, &req, NULL, 0);
391 	if (ret < 0)
392 		dev_warn(&context->dev->pdev->dev,
393 			 "destroy ucontext failed, error: %d\n", ret);
394 
395 	/* Free the UAR even if the device command failed */
396 	pvrdma_uar_free(to_vdev(ibcontext->device), &context->uar);
397 	kfree(context);
398 
399 	return ret;
400 }
401 
402 /**
403  * pvrdma_mmap - create mmap region
404  * @ibcontext: the user context
405  * @vma: the VMA
406  *
407  * @return: 0 on success, otherwise errno.
408  */
pvrdma_mmap(struct ib_ucontext * ibcontext,struct vm_area_struct * vma)409 int pvrdma_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vma)
410 {
411 	struct pvrdma_ucontext *context = to_vucontext(ibcontext);
412 	unsigned long start = vma->vm_start;
413 	unsigned long size = vma->vm_end - vma->vm_start;
414 	unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
415 
416 	dev_dbg(&context->dev->pdev->dev, "create mmap region\n");
417 
418 	if ((size != PAGE_SIZE) || (offset & ~PAGE_MASK)) {
419 		dev_warn(&context->dev->pdev->dev,
420 			 "invalid params for mmap region\n");
421 		return -EINVAL;
422 	}
423 
424 	/* Map UAR to kernel space, VM_LOCKED? */
425 	vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
426 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
427 	if (io_remap_pfn_range(vma, start, context->uar.pfn, size,
428 			       vma->vm_page_prot))
429 		return -EAGAIN;
430 
431 	return 0;
432 }
433 
434 /**
435  * pvrdma_alloc_pd - allocate protection domain
436  * @ibdev: the IB device
437  * @context: user context
438  * @udata: user data
439  *
440  * @return: the ib_pd protection domain pointer on success, otherwise errno.
441  */
pvrdma_alloc_pd(struct ib_device * ibdev,struct ib_ucontext * context,struct ib_udata * udata)442 struct ib_pd *pvrdma_alloc_pd(struct ib_device *ibdev,
443 			      struct ib_ucontext *context,
444 			      struct ib_udata *udata)
445 {
446 	struct pvrdma_pd *pd;
447 	struct pvrdma_dev *dev = to_vdev(ibdev);
448 	union pvrdma_cmd_req req;
449 	union pvrdma_cmd_resp rsp;
450 	struct pvrdma_cmd_create_pd *cmd = &req.create_pd;
451 	struct pvrdma_cmd_create_pd_resp *resp = &rsp.create_pd_resp;
452 	struct pvrdma_alloc_pd_resp pd_resp = {0};
453 	int ret;
454 	void *ptr;
455 
456 	/* Check allowed max pds */
457 	if (!atomic_add_unless(&dev->num_pds, 1, dev->dsr->caps.max_pd))
458 		return ERR_PTR(-ENOMEM);
459 
460 	pd = kmalloc(sizeof(*pd), GFP_KERNEL);
461 	if (!pd) {
462 		ptr = ERR_PTR(-ENOMEM);
463 		goto err;
464 	}
465 
466 	memset(cmd, 0, sizeof(*cmd));
467 	cmd->hdr.cmd = PVRDMA_CMD_CREATE_PD;
468 	cmd->ctx_handle = (context) ? to_vucontext(context)->ctx_handle : 0;
469 	ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_CREATE_PD_RESP);
470 	if (ret < 0) {
471 		dev_warn(&dev->pdev->dev,
472 			 "failed to allocate protection domain, error: %d\n",
473 			 ret);
474 		ptr = ERR_PTR(ret);
475 		goto freepd;
476 	}
477 
478 	pd->privileged = !context;
479 	pd->pd_handle = resp->pd_handle;
480 	pd->pdn = resp->pd_handle;
481 	pd_resp.pdn = resp->pd_handle;
482 
483 	if (context) {
484 		if (ib_copy_to_udata(udata, &pd_resp, sizeof(pd_resp))) {
485 			dev_warn(&dev->pdev->dev,
486 				 "failed to copy back protection domain\n");
487 			pvrdma_dealloc_pd(&pd->ibpd);
488 			return ERR_PTR(-EFAULT);
489 		}
490 	}
491 
492 	/* u32 pd handle */
493 	return &pd->ibpd;
494 
495 freepd:
496 	kfree(pd);
497 err:
498 	atomic_dec(&dev->num_pds);
499 	return ptr;
500 }
501 
502 /**
503  * pvrdma_dealloc_pd - deallocate protection domain
504  * @pd: the protection domain to be released
505  *
506  * @return: 0 on success, otherwise errno.
507  */
pvrdma_dealloc_pd(struct ib_pd * pd)508 int pvrdma_dealloc_pd(struct ib_pd *pd)
509 {
510 	struct pvrdma_dev *dev = to_vdev(pd->device);
511 	union pvrdma_cmd_req req;
512 	struct pvrdma_cmd_destroy_pd *cmd = &req.destroy_pd;
513 	int ret;
514 
515 	memset(cmd, 0, sizeof(*cmd));
516 	cmd->hdr.cmd = PVRDMA_CMD_DESTROY_PD;
517 	cmd->pd_handle = to_vpd(pd)->pd_handle;
518 
519 	ret = pvrdma_cmd_post(dev, &req, NULL, 0);
520 	if (ret)
521 		dev_warn(&dev->pdev->dev,
522 			 "could not dealloc protection domain, error: %d\n",
523 			 ret);
524 
525 	kfree(to_vpd(pd));
526 	atomic_dec(&dev->num_pds);
527 
528 	return 0;
529 }
530 
531 /**
532  * pvrdma_create_ah - create an address handle
533  * @pd: the protection domain
534  * @ah_attr: the attributes of the AH
535  * @udata: user data blob
536  *
537  * @return: the ib_ah pointer on success, otherwise errno.
538  */
pvrdma_create_ah(struct ib_pd * pd,struct rdma_ah_attr * ah_attr,struct ib_udata * udata)539 struct ib_ah *pvrdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
540 			       struct ib_udata *udata)
541 {
542 	struct pvrdma_dev *dev = to_vdev(pd->device);
543 	struct pvrdma_ah *ah;
544 	const struct ib_global_route *grh;
545 	u8 port_num = rdma_ah_get_port_num(ah_attr);
546 
547 	if (!(rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH))
548 		return ERR_PTR(-EINVAL);
549 
550 	grh = rdma_ah_read_grh(ah_attr);
551 	if ((ah_attr->type != RDMA_AH_ATTR_TYPE_ROCE)  ||
552 	    rdma_is_multicast_addr((struct in6_addr *)grh->dgid.raw))
553 		return ERR_PTR(-EINVAL);
554 
555 	if (!atomic_add_unless(&dev->num_ahs, 1, dev->dsr->caps.max_ah))
556 		return ERR_PTR(-ENOMEM);
557 
558 	ah = kzalloc(sizeof(*ah), GFP_KERNEL);
559 	if (!ah) {
560 		atomic_dec(&dev->num_ahs);
561 		return ERR_PTR(-ENOMEM);
562 	}
563 
564 	ah->av.port_pd = to_vpd(pd)->pd_handle | (port_num << 24);
565 	ah->av.src_path_bits = rdma_ah_get_path_bits(ah_attr);
566 	ah->av.src_path_bits |= 0x80;
567 	ah->av.gid_index = grh->sgid_index;
568 	ah->av.hop_limit = grh->hop_limit;
569 	ah->av.sl_tclass_flowlabel = (grh->traffic_class << 20) |
570 				      grh->flow_label;
571 	memcpy(ah->av.dgid, grh->dgid.raw, 16);
572 	memcpy(ah->av.dmac, ah_attr->roce.dmac, ETH_ALEN);
573 
574 	ah->ibah.device = pd->device;
575 	ah->ibah.pd = pd;
576 	ah->ibah.uobject = NULL;
577 
578 	return &ah->ibah;
579 }
580 
581 /**
582  * pvrdma_destroy_ah - destroy an address handle
583  * @ah: the address handle to destroyed
584  *
585  * @return: 0 on success.
586  */
pvrdma_destroy_ah(struct ib_ah * ah)587 int pvrdma_destroy_ah(struct ib_ah *ah)
588 {
589 	struct pvrdma_dev *dev = to_vdev(ah->device);
590 
591 	kfree(to_vah(ah));
592 	atomic_dec(&dev->num_ahs);
593 
594 	return 0;
595 }
596