xref: /linux/drivers/infiniband/hw/bnxt_re/ib_verbs.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
1 /*
2  * Broadcom NetXtreme-E RoCE driver.
3  *
4  * Copyright (c) 2016 - 2017, Broadcom. All rights reserved.  The term
5  * Broadcom refers to Broadcom Limited and/or its subsidiaries.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * BSD license below:
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  *
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in
21  *    the documentation and/or other materials provided with the
22  *    distribution.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS''
25  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
26  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS
28  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
32  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
33  * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
34  * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Description: IB Verbs interpreter
37  */
38 
39 #include <linux/interrupt.h>
40 #include <linux/types.h>
41 #include <linux/pci.h>
42 #include <linux/netdevice.h>
43 #include <linux/if_ether.h>
44 
45 #include <rdma/ib_verbs.h>
46 #include <rdma/ib_user_verbs.h>
47 #include <rdma/ib_umem.h>
48 #include <rdma/ib_addr.h>
49 #include <rdma/ib_mad.h>
50 #include <rdma/ib_cache.h>
51 
52 #include "bnxt_ulp.h"
53 
54 #include "roce_hsi.h"
55 #include "qplib_res.h"
56 #include "qplib_sp.h"
57 #include "qplib_fp.h"
58 #include "qplib_rcfw.h"
59 
60 #include "bnxt_re.h"
61 #include "ib_verbs.h"
62 #include <rdma/bnxt_re-abi.h>
63 
64 static int bnxt_re_build_sgl(struct ib_sge *ib_sg_list,
65 			     struct bnxt_qplib_sge *sg_list, int num)
66 {
67 	int i, total = 0;
68 
69 	for (i = 0; i < num; i++) {
70 		sg_list[i].addr = ib_sg_list[i].addr;
71 		sg_list[i].lkey = ib_sg_list[i].lkey;
72 		sg_list[i].size = ib_sg_list[i].length;
73 		total += sg_list[i].size;
74 	}
75 	return total;
76 }
77 
78 /* Device */
79 struct net_device *bnxt_re_get_netdev(struct ib_device *ibdev, u8 port_num)
80 {
81 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
82 	struct net_device *netdev = NULL;
83 
84 	rcu_read_lock();
85 	if (rdev)
86 		netdev = rdev->netdev;
87 	if (netdev)
88 		dev_hold(netdev);
89 
90 	rcu_read_unlock();
91 	return netdev;
92 }
93 
94 int bnxt_re_query_device(struct ib_device *ibdev,
95 			 struct ib_device_attr *ib_attr,
96 			 struct ib_udata *udata)
97 {
98 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
99 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
100 
101 	memset(ib_attr, 0, sizeof(*ib_attr));
102 
103 	ib_attr->fw_ver = (u64)(unsigned long)(dev_attr->fw_ver);
104 	bnxt_qplib_get_guid(rdev->netdev->dev_addr,
105 			    (u8 *)&ib_attr->sys_image_guid);
106 	ib_attr->max_mr_size = ~0ull;
107 	ib_attr->page_size_cap = BNXT_RE_PAGE_SIZE_4K | BNXT_RE_PAGE_SIZE_8K |
108 				 BNXT_RE_PAGE_SIZE_64K | BNXT_RE_PAGE_SIZE_2M |
109 				 BNXT_RE_PAGE_SIZE_8M | BNXT_RE_PAGE_SIZE_1G;
110 
111 	ib_attr->vendor_id = rdev->en_dev->pdev->vendor;
112 	ib_attr->vendor_part_id = rdev->en_dev->pdev->device;
113 	ib_attr->hw_ver = rdev->en_dev->pdev->subsystem_device;
114 	ib_attr->max_qp = dev_attr->max_qp;
115 	ib_attr->max_qp_wr = dev_attr->max_qp_wqes;
116 	ib_attr->device_cap_flags =
117 				    IB_DEVICE_CURR_QP_STATE_MOD
118 				    | IB_DEVICE_RC_RNR_NAK_GEN
119 				    | IB_DEVICE_SHUTDOWN_PORT
120 				    | IB_DEVICE_SYS_IMAGE_GUID
121 				    | IB_DEVICE_LOCAL_DMA_LKEY
122 				    | IB_DEVICE_RESIZE_MAX_WR
123 				    | IB_DEVICE_PORT_ACTIVE_EVENT
124 				    | IB_DEVICE_N_NOTIFY_CQ
125 				    | IB_DEVICE_MEM_WINDOW
126 				    | IB_DEVICE_MEM_WINDOW_TYPE_2B
127 				    | IB_DEVICE_MEM_MGT_EXTENSIONS;
128 	ib_attr->max_sge = dev_attr->max_qp_sges;
129 	ib_attr->max_sge_rd = dev_attr->max_qp_sges;
130 	ib_attr->max_cq = dev_attr->max_cq;
131 	ib_attr->max_cqe = dev_attr->max_cq_wqes;
132 	ib_attr->max_mr = dev_attr->max_mr;
133 	ib_attr->max_pd = dev_attr->max_pd;
134 	ib_attr->max_qp_rd_atom = dev_attr->max_qp_rd_atom;
135 	ib_attr->max_qp_init_rd_atom = dev_attr->max_qp_rd_atom;
136 	ib_attr->atomic_cap = IB_ATOMIC_HCA;
137 	ib_attr->masked_atomic_cap = IB_ATOMIC_HCA;
138 
139 	ib_attr->max_ee_rd_atom = 0;
140 	ib_attr->max_res_rd_atom = 0;
141 	ib_attr->max_ee_init_rd_atom = 0;
142 	ib_attr->max_ee = 0;
143 	ib_attr->max_rdd = 0;
144 	ib_attr->max_mw = dev_attr->max_mw;
145 	ib_attr->max_raw_ipv6_qp = 0;
146 	ib_attr->max_raw_ethy_qp = dev_attr->max_raw_ethy_qp;
147 	ib_attr->max_mcast_grp = 0;
148 	ib_attr->max_mcast_qp_attach = 0;
149 	ib_attr->max_total_mcast_qp_attach = 0;
150 	ib_attr->max_ah = dev_attr->max_ah;
151 
152 	ib_attr->max_fmr = dev_attr->max_fmr;
153 	ib_attr->max_map_per_fmr = 1;	/* ? */
154 
155 	ib_attr->max_srq = dev_attr->max_srq;
156 	ib_attr->max_srq_wr = dev_attr->max_srq_wqes;
157 	ib_attr->max_srq_sge = dev_attr->max_srq_sges;
158 
159 	ib_attr->max_fast_reg_page_list_len = MAX_PBL_LVL_1_PGS;
160 
161 	ib_attr->max_pkeys = 1;
162 	ib_attr->local_ca_ack_delay = 0;
163 	return 0;
164 }
165 
166 int bnxt_re_modify_device(struct ib_device *ibdev,
167 			  int device_modify_mask,
168 			  struct ib_device_modify *device_modify)
169 {
170 	switch (device_modify_mask) {
171 	case IB_DEVICE_MODIFY_SYS_IMAGE_GUID:
172 		/* Modify the GUID requires the modification of the GID table */
173 		/* GUID should be made as READ-ONLY */
174 		break;
175 	case IB_DEVICE_MODIFY_NODE_DESC:
176 		/* Node Desc should be made as READ-ONLY */
177 		break;
178 	default:
179 		break;
180 	}
181 	return 0;
182 }
183 
184 static void __to_ib_speed_width(struct net_device *netdev, u8 *speed, u8 *width)
185 {
186 	struct ethtool_link_ksettings lksettings;
187 	u32 espeed;
188 
189 	if (netdev->ethtool_ops && netdev->ethtool_ops->get_link_ksettings) {
190 		memset(&lksettings, 0, sizeof(lksettings));
191 		rtnl_lock();
192 		netdev->ethtool_ops->get_link_ksettings(netdev, &lksettings);
193 		rtnl_unlock();
194 		espeed = lksettings.base.speed;
195 	} else {
196 		espeed = SPEED_UNKNOWN;
197 	}
198 	switch (espeed) {
199 	case SPEED_1000:
200 		*speed = IB_SPEED_SDR;
201 		*width = IB_WIDTH_1X;
202 		break;
203 	case SPEED_10000:
204 		*speed = IB_SPEED_QDR;
205 		*width = IB_WIDTH_1X;
206 		break;
207 	case SPEED_20000:
208 		*speed = IB_SPEED_DDR;
209 		*width = IB_WIDTH_4X;
210 		break;
211 	case SPEED_25000:
212 		*speed = IB_SPEED_EDR;
213 		*width = IB_WIDTH_1X;
214 		break;
215 	case SPEED_40000:
216 		*speed = IB_SPEED_QDR;
217 		*width = IB_WIDTH_4X;
218 		break;
219 	case SPEED_50000:
220 		break;
221 	default:
222 		*speed = IB_SPEED_SDR;
223 		*width = IB_WIDTH_1X;
224 		break;
225 	}
226 }
227 
228 /* Port */
229 int bnxt_re_query_port(struct ib_device *ibdev, u8 port_num,
230 		       struct ib_port_attr *port_attr)
231 {
232 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
233 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
234 
235 	memset(port_attr, 0, sizeof(*port_attr));
236 
237 	if (netif_running(rdev->netdev) && netif_carrier_ok(rdev->netdev)) {
238 		port_attr->state = IB_PORT_ACTIVE;
239 		port_attr->phys_state = 5;
240 	} else {
241 		port_attr->state = IB_PORT_DOWN;
242 		port_attr->phys_state = 3;
243 	}
244 	port_attr->max_mtu = IB_MTU_4096;
245 	port_attr->active_mtu = iboe_get_mtu(rdev->netdev->mtu);
246 	port_attr->gid_tbl_len = dev_attr->max_sgid;
247 	port_attr->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_REINIT_SUP |
248 				    IB_PORT_DEVICE_MGMT_SUP |
249 				    IB_PORT_VENDOR_CLASS_SUP |
250 				    IB_PORT_IP_BASED_GIDS;
251 
252 	/* Max MSG size set to 2G for now */
253 	port_attr->max_msg_sz = 0x80000000;
254 	port_attr->bad_pkey_cntr = 0;
255 	port_attr->qkey_viol_cntr = 0;
256 	port_attr->pkey_tbl_len = dev_attr->max_pkey;
257 	port_attr->lid = 0;
258 	port_attr->sm_lid = 0;
259 	port_attr->lmc = 0;
260 	port_attr->max_vl_num = 4;
261 	port_attr->sm_sl = 0;
262 	port_attr->subnet_timeout = 0;
263 	port_attr->init_type_reply = 0;
264 	/* call the underlying netdev's ethtool hooks to query speed settings
265 	 * for which we acquire rtnl_lock _only_ if it's registered with
266 	 * IB stack to avoid race in the NETDEV_UNREG path
267 	 */
268 	if (test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
269 		__to_ib_speed_width(rdev->netdev, &port_attr->active_speed,
270 				    &port_attr->active_width);
271 	return 0;
272 }
273 
274 int bnxt_re_modify_port(struct ib_device *ibdev, u8 port_num,
275 			int port_modify_mask,
276 			struct ib_port_modify *port_modify)
277 {
278 	switch (port_modify_mask) {
279 	case IB_PORT_SHUTDOWN:
280 		break;
281 	case IB_PORT_INIT_TYPE:
282 		break;
283 	case IB_PORT_RESET_QKEY_CNTR:
284 		break;
285 	default:
286 		break;
287 	}
288 	return 0;
289 }
290 
291 int bnxt_re_get_port_immutable(struct ib_device *ibdev, u8 port_num,
292 			       struct ib_port_immutable *immutable)
293 {
294 	struct ib_port_attr port_attr;
295 
296 	if (bnxt_re_query_port(ibdev, port_num, &port_attr))
297 		return -EINVAL;
298 
299 	immutable->pkey_tbl_len = port_attr.pkey_tbl_len;
300 	immutable->gid_tbl_len = port_attr.gid_tbl_len;
301 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
302 	immutable->core_cap_flags |= RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
303 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
304 	return 0;
305 }
306 
307 int bnxt_re_query_pkey(struct ib_device *ibdev, u8 port_num,
308 		       u16 index, u16 *pkey)
309 {
310 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
311 
312 	/* Ignore port_num */
313 
314 	memset(pkey, 0, sizeof(*pkey));
315 	return bnxt_qplib_get_pkey(&rdev->qplib_res,
316 				   &rdev->qplib_res.pkey_tbl, index, pkey);
317 }
318 
319 int bnxt_re_query_gid(struct ib_device *ibdev, u8 port_num,
320 		      int index, union ib_gid *gid)
321 {
322 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
323 	int rc = 0;
324 
325 	/* Ignore port_num */
326 	memset(gid, 0, sizeof(*gid));
327 	rc = bnxt_qplib_get_sgid(&rdev->qplib_res,
328 				 &rdev->qplib_res.sgid_tbl, index,
329 				 (struct bnxt_qplib_gid *)gid);
330 	return rc;
331 }
332 
333 int bnxt_re_del_gid(struct ib_device *ibdev, u8 port_num,
334 		    unsigned int index, void **context)
335 {
336 	int rc = 0;
337 	struct bnxt_re_gid_ctx *ctx, **ctx_tbl;
338 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
339 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
340 
341 	/* Delete the entry from the hardware */
342 	ctx = *context;
343 	if (!ctx)
344 		return -EINVAL;
345 
346 	if (sgid_tbl && sgid_tbl->active) {
347 		if (ctx->idx >= sgid_tbl->max)
348 			return -EINVAL;
349 		ctx->refcnt--;
350 		if (!ctx->refcnt) {
351 			rc = bnxt_qplib_del_sgid
352 					(sgid_tbl,
353 					 &sgid_tbl->tbl[ctx->idx], true);
354 			if (rc)
355 				dev_err(rdev_to_dev(rdev),
356 					"Failed to remove GID: %#x", rc);
357 			ctx_tbl = sgid_tbl->ctx;
358 			ctx_tbl[ctx->idx] = NULL;
359 			kfree(ctx);
360 		}
361 	} else {
362 		return -EINVAL;
363 	}
364 	return rc;
365 }
366 
367 int bnxt_re_add_gid(struct ib_device *ibdev, u8 port_num,
368 		    unsigned int index, const union ib_gid *gid,
369 		    const struct ib_gid_attr *attr, void **context)
370 {
371 	int rc;
372 	u32 tbl_idx = 0;
373 	u16 vlan_id = 0xFFFF;
374 	struct bnxt_re_gid_ctx *ctx, **ctx_tbl;
375 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
376 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
377 
378 	if ((attr->ndev) && is_vlan_dev(attr->ndev))
379 		vlan_id = vlan_dev_vlan_id(attr->ndev);
380 
381 	rc = bnxt_qplib_add_sgid(sgid_tbl, (struct bnxt_qplib_gid *)gid,
382 				 rdev->qplib_res.netdev->dev_addr,
383 				 vlan_id, true, &tbl_idx);
384 	if (rc == -EALREADY) {
385 		ctx_tbl = sgid_tbl->ctx;
386 		ctx_tbl[tbl_idx]->refcnt++;
387 		*context = ctx_tbl[tbl_idx];
388 		return 0;
389 	}
390 
391 	if (rc < 0) {
392 		dev_err(rdev_to_dev(rdev), "Failed to add GID: %#x", rc);
393 		return rc;
394 	}
395 
396 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
397 	if (!ctx)
398 		return -ENOMEM;
399 	ctx_tbl = sgid_tbl->ctx;
400 	ctx->idx = tbl_idx;
401 	ctx->refcnt = 1;
402 	ctx_tbl[tbl_idx] = ctx;
403 
404 	return rc;
405 }
406 
407 enum rdma_link_layer bnxt_re_get_link_layer(struct ib_device *ibdev,
408 					    u8 port_num)
409 {
410 	return IB_LINK_LAYER_ETHERNET;
411 }
412 
413 /* Protection Domains */
414 int bnxt_re_dealloc_pd(struct ib_pd *ib_pd)
415 {
416 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
417 	struct bnxt_re_dev *rdev = pd->rdev;
418 	int rc;
419 
420 	if (ib_pd->uobject && pd->dpi.dbr) {
421 		struct ib_ucontext *ib_uctx = ib_pd->uobject->context;
422 		struct bnxt_re_ucontext *ucntx;
423 
424 		/* Free DPI only if this is the first PD allocated by the
425 		 * application and mark the context dpi as NULL
426 		 */
427 		ucntx = container_of(ib_uctx, struct bnxt_re_ucontext, ib_uctx);
428 
429 		rc = bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
430 					    &rdev->qplib_res.dpi_tbl,
431 					    &pd->dpi);
432 		if (rc)
433 			dev_err(rdev_to_dev(rdev), "Failed to deallocate HW DPI");
434 			/* Don't fail, continue*/
435 		ucntx->dpi = NULL;
436 	}
437 
438 	rc = bnxt_qplib_dealloc_pd(&rdev->qplib_res,
439 				   &rdev->qplib_res.pd_tbl,
440 				   &pd->qplib_pd);
441 	if (rc) {
442 		dev_err(rdev_to_dev(rdev), "Failed to deallocate HW PD");
443 		return rc;
444 	}
445 
446 	kfree(pd);
447 	return 0;
448 }
449 
450 struct ib_pd *bnxt_re_alloc_pd(struct ib_device *ibdev,
451 			       struct ib_ucontext *ucontext,
452 			       struct ib_udata *udata)
453 {
454 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
455 	struct bnxt_re_ucontext *ucntx = container_of(ucontext,
456 						      struct bnxt_re_ucontext,
457 						      ib_uctx);
458 	struct bnxt_re_pd *pd;
459 	int rc;
460 
461 	pd = kzalloc(sizeof(*pd), GFP_KERNEL);
462 	if (!pd)
463 		return ERR_PTR(-ENOMEM);
464 
465 	pd->rdev = rdev;
466 	if (bnxt_qplib_alloc_pd(&rdev->qplib_res.pd_tbl, &pd->qplib_pd)) {
467 		dev_err(rdev_to_dev(rdev), "Failed to allocate HW PD");
468 		rc = -ENOMEM;
469 		goto fail;
470 	}
471 
472 	if (udata) {
473 		struct bnxt_re_pd_resp resp;
474 
475 		if (!ucntx->dpi) {
476 			/* Allocate DPI in alloc_pd to avoid failing of
477 			 * ibv_devinfo and family of application when DPIs
478 			 * are depleted.
479 			 */
480 			if (bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
481 						 &pd->dpi, ucntx)) {
482 				rc = -ENOMEM;
483 				goto dbfail;
484 			}
485 			ucntx->dpi = &pd->dpi;
486 		}
487 
488 		resp.pdid = pd->qplib_pd.id;
489 		/* Still allow mapping this DBR to the new user PD. */
490 		resp.dpi = ucntx->dpi->dpi;
491 		resp.dbr = (u64)ucntx->dpi->umdbr;
492 
493 		rc = ib_copy_to_udata(udata, &resp, sizeof(resp));
494 		if (rc) {
495 			dev_err(rdev_to_dev(rdev),
496 				"Failed to copy user response\n");
497 			goto dbfail;
498 		}
499 	}
500 
501 	return &pd->ib_pd;
502 dbfail:
503 	(void)bnxt_qplib_dealloc_pd(&rdev->qplib_res, &rdev->qplib_res.pd_tbl,
504 				    &pd->qplib_pd);
505 fail:
506 	kfree(pd);
507 	return ERR_PTR(rc);
508 }
509 
510 /* Address Handles */
511 int bnxt_re_destroy_ah(struct ib_ah *ib_ah)
512 {
513 	struct bnxt_re_ah *ah = container_of(ib_ah, struct bnxt_re_ah, ib_ah);
514 	struct bnxt_re_dev *rdev = ah->rdev;
515 	int rc;
516 
517 	rc = bnxt_qplib_destroy_ah(&rdev->qplib_res, &ah->qplib_ah);
518 	if (rc) {
519 		dev_err(rdev_to_dev(rdev), "Failed to destroy HW AH");
520 		return rc;
521 	}
522 	kfree(ah);
523 	return 0;
524 }
525 
526 struct ib_ah *bnxt_re_create_ah(struct ib_pd *ib_pd,
527 				struct ib_ah_attr *ah_attr,
528 				struct ib_udata *udata)
529 {
530 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
531 	struct bnxt_re_dev *rdev = pd->rdev;
532 	struct bnxt_re_ah *ah;
533 	int rc;
534 	u16 vlan_tag;
535 	u8 nw_type;
536 
537 	struct ib_gid_attr sgid_attr;
538 
539 	if (!(ah_attr->ah_flags & IB_AH_GRH)) {
540 		dev_err(rdev_to_dev(rdev), "Failed to alloc AH: GRH not set");
541 		return ERR_PTR(-EINVAL);
542 	}
543 	ah = kzalloc(sizeof(*ah), GFP_ATOMIC);
544 	if (!ah)
545 		return ERR_PTR(-ENOMEM);
546 
547 	ah->rdev = rdev;
548 	ah->qplib_ah.pd = &pd->qplib_pd;
549 
550 	/* Supply the configuration for the HW */
551 	memcpy(ah->qplib_ah.dgid.data, ah_attr->grh.dgid.raw,
552 	       sizeof(union ib_gid));
553 	/*
554 	 * If RoCE V2 is enabled, stack will have two entries for
555 	 * each GID entry. Avoiding this duplicte entry in HW. Dividing
556 	 * the GID index by 2 for RoCE V2
557 	 */
558 	ah->qplib_ah.sgid_index = ah_attr->grh.sgid_index / 2;
559 	ah->qplib_ah.host_sgid_index = ah_attr->grh.sgid_index;
560 	ah->qplib_ah.traffic_class = ah_attr->grh.traffic_class;
561 	ah->qplib_ah.flow_label = ah_attr->grh.flow_label;
562 	ah->qplib_ah.hop_limit = ah_attr->grh.hop_limit;
563 	ah->qplib_ah.sl = ah_attr->sl;
564 	if (ib_pd->uobject &&
565 	    !rdma_is_multicast_addr((struct in6_addr *)
566 				    ah_attr->grh.dgid.raw) &&
567 	    !rdma_link_local_addr((struct in6_addr *)
568 				  ah_attr->grh.dgid.raw)) {
569 		union ib_gid sgid;
570 
571 		rc = ib_get_cached_gid(&rdev->ibdev, 1,
572 				       ah_attr->grh.sgid_index, &sgid,
573 				       &sgid_attr);
574 		if (rc) {
575 			dev_err(rdev_to_dev(rdev),
576 				"Failed to query gid at index %d",
577 				ah_attr->grh.sgid_index);
578 			goto fail;
579 		}
580 		if (sgid_attr.ndev) {
581 			if (is_vlan_dev(sgid_attr.ndev))
582 				vlan_tag = vlan_dev_vlan_id(sgid_attr.ndev);
583 			dev_put(sgid_attr.ndev);
584 		}
585 		/* Get network header type for this GID */
586 		nw_type = ib_gid_to_network_type(sgid_attr.gid_type, &sgid);
587 		switch (nw_type) {
588 		case RDMA_NETWORK_IPV4:
589 			ah->qplib_ah.nw_type = CMDQ_CREATE_AH_TYPE_V2IPV4;
590 			break;
591 		case RDMA_NETWORK_IPV6:
592 			ah->qplib_ah.nw_type = CMDQ_CREATE_AH_TYPE_V2IPV6;
593 			break;
594 		default:
595 			ah->qplib_ah.nw_type = CMDQ_CREATE_AH_TYPE_V1;
596 			break;
597 		}
598 		rc = rdma_addr_find_l2_eth_by_grh(&sgid, &ah_attr->grh.dgid,
599 						  ah_attr->dmac, &vlan_tag,
600 						  &sgid_attr.ndev->ifindex,
601 						  NULL);
602 		if (rc) {
603 			dev_err(rdev_to_dev(rdev), "Failed to get dmac\n");
604 			goto fail;
605 		}
606 	}
607 
608 	memcpy(ah->qplib_ah.dmac, ah_attr->dmac, ETH_ALEN);
609 	rc = bnxt_qplib_create_ah(&rdev->qplib_res, &ah->qplib_ah);
610 	if (rc) {
611 		dev_err(rdev_to_dev(rdev), "Failed to allocate HW AH");
612 		goto fail;
613 	}
614 
615 	/* Write AVID to shared page. */
616 	if (ib_pd->uobject) {
617 		struct ib_ucontext *ib_uctx = ib_pd->uobject->context;
618 		struct bnxt_re_ucontext *uctx;
619 		unsigned long flag;
620 		u32 *wrptr;
621 
622 		uctx = container_of(ib_uctx, struct bnxt_re_ucontext, ib_uctx);
623 		spin_lock_irqsave(&uctx->sh_lock, flag);
624 		wrptr = (u32 *)(uctx->shpg + BNXT_RE_AVID_OFFT);
625 		*wrptr = ah->qplib_ah.id;
626 		wmb(); /* make sure cache is updated. */
627 		spin_unlock_irqrestore(&uctx->sh_lock, flag);
628 	}
629 
630 	return &ah->ib_ah;
631 
632 fail:
633 	kfree(ah);
634 	return ERR_PTR(rc);
635 }
636 
637 int bnxt_re_modify_ah(struct ib_ah *ib_ah, struct ib_ah_attr *ah_attr)
638 {
639 	return 0;
640 }
641 
642 int bnxt_re_query_ah(struct ib_ah *ib_ah, struct ib_ah_attr *ah_attr)
643 {
644 	struct bnxt_re_ah *ah = container_of(ib_ah, struct bnxt_re_ah, ib_ah);
645 
646 	memcpy(ah_attr->grh.dgid.raw, ah->qplib_ah.dgid.data,
647 	       sizeof(union ib_gid));
648 	ah_attr->grh.sgid_index = ah->qplib_ah.host_sgid_index;
649 	ah_attr->grh.traffic_class = ah->qplib_ah.traffic_class;
650 	ah_attr->sl = ah->qplib_ah.sl;
651 	memcpy(ah_attr->dmac, ah->qplib_ah.dmac, ETH_ALEN);
652 	ah_attr->ah_flags = IB_AH_GRH;
653 	ah_attr->port_num = 1;
654 	ah_attr->static_rate = 0;
655 	return 0;
656 }
657 
658 /* Queue Pairs */
659 int bnxt_re_destroy_qp(struct ib_qp *ib_qp)
660 {
661 	struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
662 	struct bnxt_re_dev *rdev = qp->rdev;
663 	int rc;
664 
665 	rc = bnxt_qplib_destroy_qp(&rdev->qplib_res, &qp->qplib_qp);
666 	if (rc) {
667 		dev_err(rdev_to_dev(rdev), "Failed to destroy HW QP");
668 		return rc;
669 	}
670 	if (ib_qp->qp_type == IB_QPT_GSI && rdev->qp1_sqp) {
671 		rc = bnxt_qplib_destroy_ah(&rdev->qplib_res,
672 					   &rdev->sqp_ah->qplib_ah);
673 		if (rc) {
674 			dev_err(rdev_to_dev(rdev),
675 				"Failed to destroy HW AH for shadow QP");
676 			return rc;
677 		}
678 
679 		rc = bnxt_qplib_destroy_qp(&rdev->qplib_res,
680 					   &rdev->qp1_sqp->qplib_qp);
681 		if (rc) {
682 			dev_err(rdev_to_dev(rdev),
683 				"Failed to destroy Shadow QP");
684 			return rc;
685 		}
686 		mutex_lock(&rdev->qp_lock);
687 		list_del(&rdev->qp1_sqp->list);
688 		atomic_dec(&rdev->qp_count);
689 		mutex_unlock(&rdev->qp_lock);
690 
691 		kfree(rdev->sqp_ah);
692 		kfree(rdev->qp1_sqp);
693 	}
694 
695 	if (qp->rumem && !IS_ERR(qp->rumem))
696 		ib_umem_release(qp->rumem);
697 	if (qp->sumem && !IS_ERR(qp->sumem))
698 		ib_umem_release(qp->sumem);
699 
700 	mutex_lock(&rdev->qp_lock);
701 	list_del(&qp->list);
702 	atomic_dec(&rdev->qp_count);
703 	mutex_unlock(&rdev->qp_lock);
704 	kfree(qp);
705 	return 0;
706 }
707 
708 static u8 __from_ib_qp_type(enum ib_qp_type type)
709 {
710 	switch (type) {
711 	case IB_QPT_GSI:
712 		return CMDQ_CREATE_QP1_TYPE_GSI;
713 	case IB_QPT_RC:
714 		return CMDQ_CREATE_QP_TYPE_RC;
715 	case IB_QPT_UD:
716 		return CMDQ_CREATE_QP_TYPE_UD;
717 	default:
718 		return IB_QPT_MAX;
719 	}
720 }
721 
722 static int bnxt_re_init_user_qp(struct bnxt_re_dev *rdev, struct bnxt_re_pd *pd,
723 				struct bnxt_re_qp *qp, struct ib_udata *udata)
724 {
725 	struct bnxt_re_qp_req ureq;
726 	struct bnxt_qplib_qp *qplib_qp = &qp->qplib_qp;
727 	struct ib_umem *umem;
728 	int bytes = 0;
729 	struct ib_ucontext *context = pd->ib_pd.uobject->context;
730 	struct bnxt_re_ucontext *cntx = container_of(context,
731 						     struct bnxt_re_ucontext,
732 						     ib_uctx);
733 	if (ib_copy_from_udata(&ureq, udata, sizeof(ureq)))
734 		return -EFAULT;
735 
736 	bytes = (qplib_qp->sq.max_wqe * BNXT_QPLIB_MAX_SQE_ENTRY_SIZE);
737 	/* Consider mapping PSN search memory only for RC QPs. */
738 	if (qplib_qp->type == CMDQ_CREATE_QP_TYPE_RC)
739 		bytes += (qplib_qp->sq.max_wqe * sizeof(struct sq_psn_search));
740 	bytes = PAGE_ALIGN(bytes);
741 	umem = ib_umem_get(context, ureq.qpsva, bytes,
742 			   IB_ACCESS_LOCAL_WRITE, 1);
743 	if (IS_ERR(umem))
744 		return PTR_ERR(umem);
745 
746 	qp->sumem = umem;
747 	qplib_qp->sq.sglist = umem->sg_head.sgl;
748 	qplib_qp->sq.nmap = umem->nmap;
749 	qplib_qp->qp_handle = ureq.qp_handle;
750 
751 	if (!qp->qplib_qp.srq) {
752 		bytes = (qplib_qp->rq.max_wqe * BNXT_QPLIB_MAX_RQE_ENTRY_SIZE);
753 		bytes = PAGE_ALIGN(bytes);
754 		umem = ib_umem_get(context, ureq.qprva, bytes,
755 				   IB_ACCESS_LOCAL_WRITE, 1);
756 		if (IS_ERR(umem))
757 			goto rqfail;
758 		qp->rumem = umem;
759 		qplib_qp->rq.sglist = umem->sg_head.sgl;
760 		qplib_qp->rq.nmap = umem->nmap;
761 	}
762 
763 	qplib_qp->dpi = cntx->dpi;
764 	return 0;
765 rqfail:
766 	ib_umem_release(qp->sumem);
767 	qp->sumem = NULL;
768 	qplib_qp->sq.sglist = NULL;
769 	qplib_qp->sq.nmap = 0;
770 
771 	return PTR_ERR(umem);
772 }
773 
774 static struct bnxt_re_ah *bnxt_re_create_shadow_qp_ah
775 				(struct bnxt_re_pd *pd,
776 				 struct bnxt_qplib_res *qp1_res,
777 				 struct bnxt_qplib_qp *qp1_qp)
778 {
779 	struct bnxt_re_dev *rdev = pd->rdev;
780 	struct bnxt_re_ah *ah;
781 	union ib_gid sgid;
782 	int rc;
783 
784 	ah = kzalloc(sizeof(*ah), GFP_KERNEL);
785 	if (!ah)
786 		return NULL;
787 
788 	memset(ah, 0, sizeof(*ah));
789 	ah->rdev = rdev;
790 	ah->qplib_ah.pd = &pd->qplib_pd;
791 
792 	rc = bnxt_re_query_gid(&rdev->ibdev, 1, 0, &sgid);
793 	if (rc)
794 		goto fail;
795 
796 	/* supply the dgid data same as sgid */
797 	memcpy(ah->qplib_ah.dgid.data, &sgid.raw,
798 	       sizeof(union ib_gid));
799 	ah->qplib_ah.sgid_index = 0;
800 
801 	ah->qplib_ah.traffic_class = 0;
802 	ah->qplib_ah.flow_label = 0;
803 	ah->qplib_ah.hop_limit = 1;
804 	ah->qplib_ah.sl = 0;
805 	/* Have DMAC same as SMAC */
806 	ether_addr_copy(ah->qplib_ah.dmac, rdev->netdev->dev_addr);
807 
808 	rc = bnxt_qplib_create_ah(&rdev->qplib_res, &ah->qplib_ah);
809 	if (rc) {
810 		dev_err(rdev_to_dev(rdev),
811 			"Failed to allocate HW AH for Shadow QP");
812 		goto fail;
813 	}
814 
815 	return ah;
816 
817 fail:
818 	kfree(ah);
819 	return NULL;
820 }
821 
822 static struct bnxt_re_qp *bnxt_re_create_shadow_qp
823 				(struct bnxt_re_pd *pd,
824 				 struct bnxt_qplib_res *qp1_res,
825 				 struct bnxt_qplib_qp *qp1_qp)
826 {
827 	struct bnxt_re_dev *rdev = pd->rdev;
828 	struct bnxt_re_qp *qp;
829 	int rc;
830 
831 	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
832 	if (!qp)
833 		return NULL;
834 
835 	memset(qp, 0, sizeof(*qp));
836 	qp->rdev = rdev;
837 
838 	/* Initialize the shadow QP structure from the QP1 values */
839 	ether_addr_copy(qp->qplib_qp.smac, rdev->netdev->dev_addr);
840 
841 	qp->qplib_qp.pd = &pd->qplib_pd;
842 	qp->qplib_qp.qp_handle = (u64)(unsigned long)(&qp->qplib_qp);
843 	qp->qplib_qp.type = IB_QPT_UD;
844 
845 	qp->qplib_qp.max_inline_data = 0;
846 	qp->qplib_qp.sig_type = true;
847 
848 	/* Shadow QP SQ depth should be same as QP1 RQ depth */
849 	qp->qplib_qp.sq.max_wqe = qp1_qp->rq.max_wqe;
850 	qp->qplib_qp.sq.max_sge = 2;
851 
852 	qp->qplib_qp.scq = qp1_qp->scq;
853 	qp->qplib_qp.rcq = qp1_qp->rcq;
854 
855 	qp->qplib_qp.rq.max_wqe = qp1_qp->rq.max_wqe;
856 	qp->qplib_qp.rq.max_sge = qp1_qp->rq.max_sge;
857 
858 	qp->qplib_qp.mtu = qp1_qp->mtu;
859 
860 	qp->qplib_qp.sq_hdr_buf_size = 0;
861 	qp->qplib_qp.rq_hdr_buf_size = BNXT_QPLIB_MAX_GRH_HDR_SIZE_IPV6;
862 	qp->qplib_qp.dpi = &rdev->dpi_privileged;
863 
864 	rc = bnxt_qplib_create_qp(qp1_res, &qp->qplib_qp);
865 	if (rc)
866 		goto fail;
867 
868 	rdev->sqp_id = qp->qplib_qp.id;
869 
870 	spin_lock_init(&qp->sq_lock);
871 	INIT_LIST_HEAD(&qp->list);
872 	mutex_lock(&rdev->qp_lock);
873 	list_add_tail(&qp->list, &rdev->qp_list);
874 	atomic_inc(&rdev->qp_count);
875 	mutex_unlock(&rdev->qp_lock);
876 	return qp;
877 fail:
878 	kfree(qp);
879 	return NULL;
880 }
881 
882 struct ib_qp *bnxt_re_create_qp(struct ib_pd *ib_pd,
883 				struct ib_qp_init_attr *qp_init_attr,
884 				struct ib_udata *udata)
885 {
886 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
887 	struct bnxt_re_dev *rdev = pd->rdev;
888 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
889 	struct bnxt_re_qp *qp;
890 	struct bnxt_re_cq *cq;
891 	int rc, entries;
892 
893 	if ((qp_init_attr->cap.max_send_wr > dev_attr->max_qp_wqes) ||
894 	    (qp_init_attr->cap.max_recv_wr > dev_attr->max_qp_wqes) ||
895 	    (qp_init_attr->cap.max_send_sge > dev_attr->max_qp_sges) ||
896 	    (qp_init_attr->cap.max_recv_sge > dev_attr->max_qp_sges) ||
897 	    (qp_init_attr->cap.max_inline_data > dev_attr->max_inline_data))
898 		return ERR_PTR(-EINVAL);
899 
900 	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
901 	if (!qp)
902 		return ERR_PTR(-ENOMEM);
903 
904 	qp->rdev = rdev;
905 	ether_addr_copy(qp->qplib_qp.smac, rdev->netdev->dev_addr);
906 	qp->qplib_qp.pd = &pd->qplib_pd;
907 	qp->qplib_qp.qp_handle = (u64)(unsigned long)(&qp->qplib_qp);
908 	qp->qplib_qp.type = __from_ib_qp_type(qp_init_attr->qp_type);
909 	if (qp->qplib_qp.type == IB_QPT_MAX) {
910 		dev_err(rdev_to_dev(rdev), "QP type 0x%x not supported",
911 			qp->qplib_qp.type);
912 		rc = -EINVAL;
913 		goto fail;
914 	}
915 	qp->qplib_qp.max_inline_data = qp_init_attr->cap.max_inline_data;
916 	qp->qplib_qp.sig_type = ((qp_init_attr->sq_sig_type ==
917 				  IB_SIGNAL_ALL_WR) ? true : false);
918 
919 	entries = roundup_pow_of_two(qp_init_attr->cap.max_send_wr + 1);
920 	qp->qplib_qp.sq.max_wqe = min_t(u32, entries,
921 					dev_attr->max_qp_wqes + 1);
922 
923 	qp->qplib_qp.sq.max_sge = qp_init_attr->cap.max_send_sge;
924 	if (qp->qplib_qp.sq.max_sge > dev_attr->max_qp_sges)
925 		qp->qplib_qp.sq.max_sge = dev_attr->max_qp_sges;
926 
927 	if (qp_init_attr->send_cq) {
928 		cq = container_of(qp_init_attr->send_cq, struct bnxt_re_cq,
929 				  ib_cq);
930 		if (!cq) {
931 			dev_err(rdev_to_dev(rdev), "Send CQ not found");
932 			rc = -EINVAL;
933 			goto fail;
934 		}
935 		qp->qplib_qp.scq = &cq->qplib_cq;
936 	}
937 
938 	if (qp_init_attr->recv_cq) {
939 		cq = container_of(qp_init_attr->recv_cq, struct bnxt_re_cq,
940 				  ib_cq);
941 		if (!cq) {
942 			dev_err(rdev_to_dev(rdev), "Receive CQ not found");
943 			rc = -EINVAL;
944 			goto fail;
945 		}
946 		qp->qplib_qp.rcq = &cq->qplib_cq;
947 	}
948 
949 	if (qp_init_attr->srq) {
950 		dev_err(rdev_to_dev(rdev), "SRQ not supported");
951 		rc = -ENOTSUPP;
952 		goto fail;
953 	} else {
954 		/* Allocate 1 more than what's provided so posting max doesn't
955 		 * mean empty
956 		 */
957 		entries = roundup_pow_of_two(qp_init_attr->cap.max_recv_wr + 1);
958 		qp->qplib_qp.rq.max_wqe = min_t(u32, entries,
959 						dev_attr->max_qp_wqes + 1);
960 
961 		qp->qplib_qp.rq.max_sge = qp_init_attr->cap.max_recv_sge;
962 		if (qp->qplib_qp.rq.max_sge > dev_attr->max_qp_sges)
963 			qp->qplib_qp.rq.max_sge = dev_attr->max_qp_sges;
964 	}
965 
966 	qp->qplib_qp.mtu = ib_mtu_enum_to_int(iboe_get_mtu(rdev->netdev->mtu));
967 
968 	if (qp_init_attr->qp_type == IB_QPT_GSI) {
969 		qp->qplib_qp.rq.max_sge = dev_attr->max_qp_sges;
970 		if (qp->qplib_qp.rq.max_sge > dev_attr->max_qp_sges)
971 			qp->qplib_qp.rq.max_sge = dev_attr->max_qp_sges;
972 		qp->qplib_qp.sq.max_sge++;
973 		if (qp->qplib_qp.sq.max_sge > dev_attr->max_qp_sges)
974 			qp->qplib_qp.sq.max_sge = dev_attr->max_qp_sges;
975 
976 		qp->qplib_qp.rq_hdr_buf_size =
977 					BNXT_QPLIB_MAX_QP1_RQ_HDR_SIZE_V2;
978 
979 		qp->qplib_qp.sq_hdr_buf_size =
980 					BNXT_QPLIB_MAX_QP1_SQ_HDR_SIZE_V2;
981 		qp->qplib_qp.dpi = &rdev->dpi_privileged;
982 		rc = bnxt_qplib_create_qp1(&rdev->qplib_res, &qp->qplib_qp);
983 		if (rc) {
984 			dev_err(rdev_to_dev(rdev), "Failed to create HW QP1");
985 			goto fail;
986 		}
987 		/* Create a shadow QP to handle the QP1 traffic */
988 		rdev->qp1_sqp = bnxt_re_create_shadow_qp(pd, &rdev->qplib_res,
989 							 &qp->qplib_qp);
990 		if (!rdev->qp1_sqp) {
991 			rc = -EINVAL;
992 			dev_err(rdev_to_dev(rdev),
993 				"Failed to create Shadow QP for QP1");
994 			goto qp_destroy;
995 		}
996 		rdev->sqp_ah = bnxt_re_create_shadow_qp_ah(pd, &rdev->qplib_res,
997 							   &qp->qplib_qp);
998 		if (!rdev->sqp_ah) {
999 			bnxt_qplib_destroy_qp(&rdev->qplib_res,
1000 					      &rdev->qp1_sqp->qplib_qp);
1001 			rc = -EINVAL;
1002 			dev_err(rdev_to_dev(rdev),
1003 				"Failed to create AH entry for ShadowQP");
1004 			goto qp_destroy;
1005 		}
1006 
1007 	} else {
1008 		qp->qplib_qp.max_rd_atomic = dev_attr->max_qp_rd_atom;
1009 		qp->qplib_qp.max_dest_rd_atomic = dev_attr->max_qp_init_rd_atom;
1010 		if (udata) {
1011 			rc = bnxt_re_init_user_qp(rdev, pd, qp, udata);
1012 			if (rc)
1013 				goto fail;
1014 		} else {
1015 			qp->qplib_qp.dpi = &rdev->dpi_privileged;
1016 		}
1017 
1018 		rc = bnxt_qplib_create_qp(&rdev->qplib_res, &qp->qplib_qp);
1019 		if (rc) {
1020 			dev_err(rdev_to_dev(rdev), "Failed to create HW QP");
1021 			goto fail;
1022 		}
1023 	}
1024 
1025 	qp->ib_qp.qp_num = qp->qplib_qp.id;
1026 	spin_lock_init(&qp->sq_lock);
1027 
1028 	if (udata) {
1029 		struct bnxt_re_qp_resp resp;
1030 
1031 		resp.qpid = qp->ib_qp.qp_num;
1032 		resp.rsvd = 0;
1033 		rc = ib_copy_to_udata(udata, &resp, sizeof(resp));
1034 		if (rc) {
1035 			dev_err(rdev_to_dev(rdev), "Failed to copy QP udata");
1036 			goto qp_destroy;
1037 		}
1038 	}
1039 	INIT_LIST_HEAD(&qp->list);
1040 	mutex_lock(&rdev->qp_lock);
1041 	list_add_tail(&qp->list, &rdev->qp_list);
1042 	atomic_inc(&rdev->qp_count);
1043 	mutex_unlock(&rdev->qp_lock);
1044 
1045 	return &qp->ib_qp;
1046 qp_destroy:
1047 	bnxt_qplib_destroy_qp(&rdev->qplib_res, &qp->qplib_qp);
1048 fail:
1049 	kfree(qp);
1050 	return ERR_PTR(rc);
1051 }
1052 
1053 static u8 __from_ib_qp_state(enum ib_qp_state state)
1054 {
1055 	switch (state) {
1056 	case IB_QPS_RESET:
1057 		return CMDQ_MODIFY_QP_NEW_STATE_RESET;
1058 	case IB_QPS_INIT:
1059 		return CMDQ_MODIFY_QP_NEW_STATE_INIT;
1060 	case IB_QPS_RTR:
1061 		return CMDQ_MODIFY_QP_NEW_STATE_RTR;
1062 	case IB_QPS_RTS:
1063 		return CMDQ_MODIFY_QP_NEW_STATE_RTS;
1064 	case IB_QPS_SQD:
1065 		return CMDQ_MODIFY_QP_NEW_STATE_SQD;
1066 	case IB_QPS_SQE:
1067 		return CMDQ_MODIFY_QP_NEW_STATE_SQE;
1068 	case IB_QPS_ERR:
1069 	default:
1070 		return CMDQ_MODIFY_QP_NEW_STATE_ERR;
1071 	}
1072 }
1073 
1074 static enum ib_qp_state __to_ib_qp_state(u8 state)
1075 {
1076 	switch (state) {
1077 	case CMDQ_MODIFY_QP_NEW_STATE_RESET:
1078 		return IB_QPS_RESET;
1079 	case CMDQ_MODIFY_QP_NEW_STATE_INIT:
1080 		return IB_QPS_INIT;
1081 	case CMDQ_MODIFY_QP_NEW_STATE_RTR:
1082 		return IB_QPS_RTR;
1083 	case CMDQ_MODIFY_QP_NEW_STATE_RTS:
1084 		return IB_QPS_RTS;
1085 	case CMDQ_MODIFY_QP_NEW_STATE_SQD:
1086 		return IB_QPS_SQD;
1087 	case CMDQ_MODIFY_QP_NEW_STATE_SQE:
1088 		return IB_QPS_SQE;
1089 	case CMDQ_MODIFY_QP_NEW_STATE_ERR:
1090 	default:
1091 		return IB_QPS_ERR;
1092 	}
1093 }
1094 
1095 static u32 __from_ib_mtu(enum ib_mtu mtu)
1096 {
1097 	switch (mtu) {
1098 	case IB_MTU_256:
1099 		return CMDQ_MODIFY_QP_PATH_MTU_MTU_256;
1100 	case IB_MTU_512:
1101 		return CMDQ_MODIFY_QP_PATH_MTU_MTU_512;
1102 	case IB_MTU_1024:
1103 		return CMDQ_MODIFY_QP_PATH_MTU_MTU_1024;
1104 	case IB_MTU_2048:
1105 		return CMDQ_MODIFY_QP_PATH_MTU_MTU_2048;
1106 	case IB_MTU_4096:
1107 		return CMDQ_MODIFY_QP_PATH_MTU_MTU_4096;
1108 	default:
1109 		return CMDQ_MODIFY_QP_PATH_MTU_MTU_2048;
1110 	}
1111 }
1112 
1113 static enum ib_mtu __to_ib_mtu(u32 mtu)
1114 {
1115 	switch (mtu & CREQ_QUERY_QP_RESP_SB_PATH_MTU_MASK) {
1116 	case CMDQ_MODIFY_QP_PATH_MTU_MTU_256:
1117 		return IB_MTU_256;
1118 	case CMDQ_MODIFY_QP_PATH_MTU_MTU_512:
1119 		return IB_MTU_512;
1120 	case CMDQ_MODIFY_QP_PATH_MTU_MTU_1024:
1121 		return IB_MTU_1024;
1122 	case CMDQ_MODIFY_QP_PATH_MTU_MTU_2048:
1123 		return IB_MTU_2048;
1124 	case CMDQ_MODIFY_QP_PATH_MTU_MTU_4096:
1125 		return IB_MTU_4096;
1126 	default:
1127 		return IB_MTU_2048;
1128 	}
1129 }
1130 
1131 static int __from_ib_access_flags(int iflags)
1132 {
1133 	int qflags = 0;
1134 
1135 	if (iflags & IB_ACCESS_LOCAL_WRITE)
1136 		qflags |= BNXT_QPLIB_ACCESS_LOCAL_WRITE;
1137 	if (iflags & IB_ACCESS_REMOTE_READ)
1138 		qflags |= BNXT_QPLIB_ACCESS_REMOTE_READ;
1139 	if (iflags & IB_ACCESS_REMOTE_WRITE)
1140 		qflags |= BNXT_QPLIB_ACCESS_REMOTE_WRITE;
1141 	if (iflags & IB_ACCESS_REMOTE_ATOMIC)
1142 		qflags |= BNXT_QPLIB_ACCESS_REMOTE_ATOMIC;
1143 	if (iflags & IB_ACCESS_MW_BIND)
1144 		qflags |= BNXT_QPLIB_ACCESS_MW_BIND;
1145 	if (iflags & IB_ZERO_BASED)
1146 		qflags |= BNXT_QPLIB_ACCESS_ZERO_BASED;
1147 	if (iflags & IB_ACCESS_ON_DEMAND)
1148 		qflags |= BNXT_QPLIB_ACCESS_ON_DEMAND;
1149 	return qflags;
1150 };
1151 
1152 static enum ib_access_flags __to_ib_access_flags(int qflags)
1153 {
1154 	enum ib_access_flags iflags = 0;
1155 
1156 	if (qflags & BNXT_QPLIB_ACCESS_LOCAL_WRITE)
1157 		iflags |= IB_ACCESS_LOCAL_WRITE;
1158 	if (qflags & BNXT_QPLIB_ACCESS_REMOTE_WRITE)
1159 		iflags |= IB_ACCESS_REMOTE_WRITE;
1160 	if (qflags & BNXT_QPLIB_ACCESS_REMOTE_READ)
1161 		iflags |= IB_ACCESS_REMOTE_READ;
1162 	if (qflags & BNXT_QPLIB_ACCESS_REMOTE_ATOMIC)
1163 		iflags |= IB_ACCESS_REMOTE_ATOMIC;
1164 	if (qflags & BNXT_QPLIB_ACCESS_MW_BIND)
1165 		iflags |= IB_ACCESS_MW_BIND;
1166 	if (qflags & BNXT_QPLIB_ACCESS_ZERO_BASED)
1167 		iflags |= IB_ZERO_BASED;
1168 	if (qflags & BNXT_QPLIB_ACCESS_ON_DEMAND)
1169 		iflags |= IB_ACCESS_ON_DEMAND;
1170 	return iflags;
1171 };
1172 
1173 static int bnxt_re_modify_shadow_qp(struct bnxt_re_dev *rdev,
1174 				    struct bnxt_re_qp *qp1_qp,
1175 				    int qp_attr_mask)
1176 {
1177 	struct bnxt_re_qp *qp = rdev->qp1_sqp;
1178 	int rc = 0;
1179 
1180 	if (qp_attr_mask & IB_QP_STATE) {
1181 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_STATE;
1182 		qp->qplib_qp.state = qp1_qp->qplib_qp.state;
1183 	}
1184 	if (qp_attr_mask & IB_QP_PKEY_INDEX) {
1185 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_PKEY;
1186 		qp->qplib_qp.pkey_index = qp1_qp->qplib_qp.pkey_index;
1187 	}
1188 
1189 	if (qp_attr_mask & IB_QP_QKEY) {
1190 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_QKEY;
1191 		/* Using a Random  QKEY */
1192 		qp->qplib_qp.qkey = 0x81818181;
1193 	}
1194 	if (qp_attr_mask & IB_QP_SQ_PSN) {
1195 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_SQ_PSN;
1196 		qp->qplib_qp.sq.psn = qp1_qp->qplib_qp.sq.psn;
1197 	}
1198 
1199 	rc = bnxt_qplib_modify_qp(&rdev->qplib_res, &qp->qplib_qp);
1200 	if (rc)
1201 		dev_err(rdev_to_dev(rdev),
1202 			"Failed to modify Shadow QP for QP1");
1203 	return rc;
1204 }
1205 
1206 int bnxt_re_modify_qp(struct ib_qp *ib_qp, struct ib_qp_attr *qp_attr,
1207 		      int qp_attr_mask, struct ib_udata *udata)
1208 {
1209 	struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
1210 	struct bnxt_re_dev *rdev = qp->rdev;
1211 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
1212 	enum ib_qp_state curr_qp_state, new_qp_state;
1213 	int rc, entries;
1214 	int status;
1215 	union ib_gid sgid;
1216 	struct ib_gid_attr sgid_attr;
1217 	u8 nw_type;
1218 
1219 	qp->qplib_qp.modify_flags = 0;
1220 	if (qp_attr_mask & IB_QP_STATE) {
1221 		curr_qp_state = __to_ib_qp_state(qp->qplib_qp.cur_qp_state);
1222 		new_qp_state = qp_attr->qp_state;
1223 		if (!ib_modify_qp_is_ok(curr_qp_state, new_qp_state,
1224 					ib_qp->qp_type, qp_attr_mask,
1225 					IB_LINK_LAYER_ETHERNET)) {
1226 			dev_err(rdev_to_dev(rdev),
1227 				"Invalid attribute mask: %#x specified ",
1228 				qp_attr_mask);
1229 			dev_err(rdev_to_dev(rdev),
1230 				"for qpn: %#x type: %#x",
1231 				ib_qp->qp_num, ib_qp->qp_type);
1232 			dev_err(rdev_to_dev(rdev),
1233 				"curr_qp_state=0x%x, new_qp_state=0x%x\n",
1234 				curr_qp_state, new_qp_state);
1235 			return -EINVAL;
1236 		}
1237 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_STATE;
1238 		qp->qplib_qp.state = __from_ib_qp_state(qp_attr->qp_state);
1239 	}
1240 	if (qp_attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY) {
1241 		qp->qplib_qp.modify_flags |=
1242 				CMDQ_MODIFY_QP_MODIFY_MASK_EN_SQD_ASYNC_NOTIFY;
1243 		qp->qplib_qp.en_sqd_async_notify = true;
1244 	}
1245 	if (qp_attr_mask & IB_QP_ACCESS_FLAGS) {
1246 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_ACCESS;
1247 		qp->qplib_qp.access =
1248 			__from_ib_access_flags(qp_attr->qp_access_flags);
1249 		/* LOCAL_WRITE access must be set to allow RC receive */
1250 		qp->qplib_qp.access |= BNXT_QPLIB_ACCESS_LOCAL_WRITE;
1251 	}
1252 	if (qp_attr_mask & IB_QP_PKEY_INDEX) {
1253 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_PKEY;
1254 		qp->qplib_qp.pkey_index = qp_attr->pkey_index;
1255 	}
1256 	if (qp_attr_mask & IB_QP_QKEY) {
1257 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_QKEY;
1258 		qp->qplib_qp.qkey = qp_attr->qkey;
1259 	}
1260 	if (qp_attr_mask & IB_QP_AV) {
1261 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_DGID |
1262 				     CMDQ_MODIFY_QP_MODIFY_MASK_FLOW_LABEL |
1263 				     CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX |
1264 				     CMDQ_MODIFY_QP_MODIFY_MASK_HOP_LIMIT |
1265 				     CMDQ_MODIFY_QP_MODIFY_MASK_TRAFFIC_CLASS |
1266 				     CMDQ_MODIFY_QP_MODIFY_MASK_DEST_MAC |
1267 				     CMDQ_MODIFY_QP_MODIFY_MASK_VLAN_ID;
1268 		memcpy(qp->qplib_qp.ah.dgid.data, qp_attr->ah_attr.grh.dgid.raw,
1269 		       sizeof(qp->qplib_qp.ah.dgid.data));
1270 		qp->qplib_qp.ah.flow_label = qp_attr->ah_attr.grh.flow_label;
1271 		/* If RoCE V2 is enabled, stack will have two entries for
1272 		 * each GID entry. Avoiding this duplicte entry in HW. Dividing
1273 		 * the GID index by 2 for RoCE V2
1274 		 */
1275 		qp->qplib_qp.ah.sgid_index =
1276 					qp_attr->ah_attr.grh.sgid_index / 2;
1277 		qp->qplib_qp.ah.host_sgid_index =
1278 					qp_attr->ah_attr.grh.sgid_index;
1279 		qp->qplib_qp.ah.hop_limit = qp_attr->ah_attr.grh.hop_limit;
1280 		qp->qplib_qp.ah.traffic_class =
1281 					qp_attr->ah_attr.grh.traffic_class;
1282 		qp->qplib_qp.ah.sl = qp_attr->ah_attr.sl;
1283 		ether_addr_copy(qp->qplib_qp.ah.dmac, qp_attr->ah_attr.dmac);
1284 
1285 		status = ib_get_cached_gid(&rdev->ibdev, 1,
1286 					   qp_attr->ah_attr.grh.sgid_index,
1287 					   &sgid, &sgid_attr);
1288 		if (!status && sgid_attr.ndev) {
1289 			memcpy(qp->qplib_qp.smac, sgid_attr.ndev->dev_addr,
1290 			       ETH_ALEN);
1291 			dev_put(sgid_attr.ndev);
1292 			nw_type = ib_gid_to_network_type(sgid_attr.gid_type,
1293 							 &sgid);
1294 			switch (nw_type) {
1295 			case RDMA_NETWORK_IPV4:
1296 				qp->qplib_qp.nw_type =
1297 					CMDQ_MODIFY_QP_NETWORK_TYPE_ROCEV2_IPV4;
1298 				break;
1299 			case RDMA_NETWORK_IPV6:
1300 				qp->qplib_qp.nw_type =
1301 					CMDQ_MODIFY_QP_NETWORK_TYPE_ROCEV2_IPV6;
1302 				break;
1303 			default:
1304 				qp->qplib_qp.nw_type =
1305 					CMDQ_MODIFY_QP_NETWORK_TYPE_ROCEV1;
1306 				break;
1307 			}
1308 		}
1309 	}
1310 
1311 	if (qp_attr_mask & IB_QP_PATH_MTU) {
1312 		qp->qplib_qp.modify_flags |=
1313 				CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU;
1314 		qp->qplib_qp.path_mtu = __from_ib_mtu(qp_attr->path_mtu);
1315 	} else if (qp_attr->qp_state == IB_QPS_RTR) {
1316 		qp->qplib_qp.modify_flags |=
1317 			CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU;
1318 		qp->qplib_qp.path_mtu =
1319 			__from_ib_mtu(iboe_get_mtu(rdev->netdev->mtu));
1320 	}
1321 
1322 	if (qp_attr_mask & IB_QP_TIMEOUT) {
1323 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_TIMEOUT;
1324 		qp->qplib_qp.timeout = qp_attr->timeout;
1325 	}
1326 	if (qp_attr_mask & IB_QP_RETRY_CNT) {
1327 		qp->qplib_qp.modify_flags |=
1328 				CMDQ_MODIFY_QP_MODIFY_MASK_RETRY_CNT;
1329 		qp->qplib_qp.retry_cnt = qp_attr->retry_cnt;
1330 	}
1331 	if (qp_attr_mask & IB_QP_RNR_RETRY) {
1332 		qp->qplib_qp.modify_flags |=
1333 				CMDQ_MODIFY_QP_MODIFY_MASK_RNR_RETRY;
1334 		qp->qplib_qp.rnr_retry = qp_attr->rnr_retry;
1335 	}
1336 	if (qp_attr_mask & IB_QP_MIN_RNR_TIMER) {
1337 		qp->qplib_qp.modify_flags |=
1338 				CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER;
1339 		qp->qplib_qp.min_rnr_timer = qp_attr->min_rnr_timer;
1340 	}
1341 	if (qp_attr_mask & IB_QP_RQ_PSN) {
1342 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_RQ_PSN;
1343 		qp->qplib_qp.rq.psn = qp_attr->rq_psn;
1344 	}
1345 	if (qp_attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
1346 		qp->qplib_qp.modify_flags |=
1347 				CMDQ_MODIFY_QP_MODIFY_MASK_MAX_RD_ATOMIC;
1348 		qp->qplib_qp.max_rd_atomic = qp_attr->max_rd_atomic;
1349 	}
1350 	if (qp_attr_mask & IB_QP_SQ_PSN) {
1351 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_SQ_PSN;
1352 		qp->qplib_qp.sq.psn = qp_attr->sq_psn;
1353 	}
1354 	if (qp_attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
1355 		qp->qplib_qp.modify_flags |=
1356 				CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC;
1357 		qp->qplib_qp.max_dest_rd_atomic = qp_attr->max_dest_rd_atomic;
1358 	}
1359 	if (qp_attr_mask & IB_QP_CAP) {
1360 		qp->qplib_qp.modify_flags |=
1361 				CMDQ_MODIFY_QP_MODIFY_MASK_SQ_SIZE |
1362 				CMDQ_MODIFY_QP_MODIFY_MASK_RQ_SIZE |
1363 				CMDQ_MODIFY_QP_MODIFY_MASK_SQ_SGE |
1364 				CMDQ_MODIFY_QP_MODIFY_MASK_RQ_SGE |
1365 				CMDQ_MODIFY_QP_MODIFY_MASK_MAX_INLINE_DATA;
1366 		if ((qp_attr->cap.max_send_wr >= dev_attr->max_qp_wqes) ||
1367 		    (qp_attr->cap.max_recv_wr >= dev_attr->max_qp_wqes) ||
1368 		    (qp_attr->cap.max_send_sge >= dev_attr->max_qp_sges) ||
1369 		    (qp_attr->cap.max_recv_sge >= dev_attr->max_qp_sges) ||
1370 		    (qp_attr->cap.max_inline_data >=
1371 						dev_attr->max_inline_data)) {
1372 			dev_err(rdev_to_dev(rdev),
1373 				"Create QP failed - max exceeded");
1374 			return -EINVAL;
1375 		}
1376 		entries = roundup_pow_of_two(qp_attr->cap.max_send_wr);
1377 		qp->qplib_qp.sq.max_wqe = min_t(u32, entries,
1378 						dev_attr->max_qp_wqes + 1);
1379 		qp->qplib_qp.sq.max_sge = qp_attr->cap.max_send_sge;
1380 		if (qp->qplib_qp.rq.max_wqe) {
1381 			entries = roundup_pow_of_two(qp_attr->cap.max_recv_wr);
1382 			qp->qplib_qp.rq.max_wqe =
1383 				min_t(u32, entries, dev_attr->max_qp_wqes + 1);
1384 			qp->qplib_qp.rq.max_sge = qp_attr->cap.max_recv_sge;
1385 		} else {
1386 			/* SRQ was used prior, just ignore the RQ caps */
1387 		}
1388 	}
1389 	if (qp_attr_mask & IB_QP_DEST_QPN) {
1390 		qp->qplib_qp.modify_flags |=
1391 				CMDQ_MODIFY_QP_MODIFY_MASK_DEST_QP_ID;
1392 		qp->qplib_qp.dest_qpn = qp_attr->dest_qp_num;
1393 	}
1394 	rc = bnxt_qplib_modify_qp(&rdev->qplib_res, &qp->qplib_qp);
1395 	if (rc) {
1396 		dev_err(rdev_to_dev(rdev), "Failed to modify HW QP");
1397 		return rc;
1398 	}
1399 	if (ib_qp->qp_type == IB_QPT_GSI && rdev->qp1_sqp)
1400 		rc = bnxt_re_modify_shadow_qp(rdev, qp, qp_attr_mask);
1401 	return rc;
1402 }
1403 
1404 int bnxt_re_query_qp(struct ib_qp *ib_qp, struct ib_qp_attr *qp_attr,
1405 		     int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr)
1406 {
1407 	struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
1408 	struct bnxt_re_dev *rdev = qp->rdev;
1409 	struct bnxt_qplib_qp qplib_qp;
1410 	int rc;
1411 
1412 	memset(&qplib_qp, 0, sizeof(struct bnxt_qplib_qp));
1413 	qplib_qp.id = qp->qplib_qp.id;
1414 	qplib_qp.ah.host_sgid_index = qp->qplib_qp.ah.host_sgid_index;
1415 
1416 	rc = bnxt_qplib_query_qp(&rdev->qplib_res, &qplib_qp);
1417 	if (rc) {
1418 		dev_err(rdev_to_dev(rdev), "Failed to query HW QP");
1419 		return rc;
1420 	}
1421 	qp_attr->qp_state = __to_ib_qp_state(qplib_qp.state);
1422 	qp_attr->en_sqd_async_notify = qplib_qp.en_sqd_async_notify ? 1 : 0;
1423 	qp_attr->qp_access_flags = __to_ib_access_flags(qplib_qp.access);
1424 	qp_attr->pkey_index = qplib_qp.pkey_index;
1425 	qp_attr->qkey = qplib_qp.qkey;
1426 	memcpy(qp_attr->ah_attr.grh.dgid.raw, qplib_qp.ah.dgid.data,
1427 	       sizeof(qplib_qp.ah.dgid.data));
1428 	qp_attr->ah_attr.grh.flow_label = qplib_qp.ah.flow_label;
1429 	qp_attr->ah_attr.grh.sgid_index = qplib_qp.ah.host_sgid_index;
1430 	qp_attr->ah_attr.grh.hop_limit = qplib_qp.ah.hop_limit;
1431 	qp_attr->ah_attr.grh.traffic_class = qplib_qp.ah.traffic_class;
1432 	qp_attr->ah_attr.sl = qplib_qp.ah.sl;
1433 	ether_addr_copy(qp_attr->ah_attr.dmac, qplib_qp.ah.dmac);
1434 	qp_attr->path_mtu = __to_ib_mtu(qplib_qp.path_mtu);
1435 	qp_attr->timeout = qplib_qp.timeout;
1436 	qp_attr->retry_cnt = qplib_qp.retry_cnt;
1437 	qp_attr->rnr_retry = qplib_qp.rnr_retry;
1438 	qp_attr->min_rnr_timer = qplib_qp.min_rnr_timer;
1439 	qp_attr->rq_psn = qplib_qp.rq.psn;
1440 	qp_attr->max_rd_atomic = qplib_qp.max_rd_atomic;
1441 	qp_attr->sq_psn = qplib_qp.sq.psn;
1442 	qp_attr->max_dest_rd_atomic = qplib_qp.max_dest_rd_atomic;
1443 	qp_init_attr->sq_sig_type = qplib_qp.sig_type ? IB_SIGNAL_ALL_WR :
1444 							IB_SIGNAL_REQ_WR;
1445 	qp_attr->dest_qp_num = qplib_qp.dest_qpn;
1446 
1447 	qp_attr->cap.max_send_wr = qp->qplib_qp.sq.max_wqe;
1448 	qp_attr->cap.max_send_sge = qp->qplib_qp.sq.max_sge;
1449 	qp_attr->cap.max_recv_wr = qp->qplib_qp.rq.max_wqe;
1450 	qp_attr->cap.max_recv_sge = qp->qplib_qp.rq.max_sge;
1451 	qp_attr->cap.max_inline_data = qp->qplib_qp.max_inline_data;
1452 	qp_init_attr->cap = qp_attr->cap;
1453 
1454 	return 0;
1455 }
1456 
1457 /* Routine for sending QP1 packets for RoCE V1 an V2
1458  */
1459 static int bnxt_re_build_qp1_send_v2(struct bnxt_re_qp *qp,
1460 				     struct ib_send_wr *wr,
1461 				     struct bnxt_qplib_swqe *wqe,
1462 				     int payload_size)
1463 {
1464 	struct ib_device *ibdev = &qp->rdev->ibdev;
1465 	struct bnxt_re_ah *ah = container_of(ud_wr(wr)->ah, struct bnxt_re_ah,
1466 					     ib_ah);
1467 	struct bnxt_qplib_ah *qplib_ah = &ah->qplib_ah;
1468 	struct bnxt_qplib_sge sge;
1469 	union ib_gid sgid;
1470 	u8 nw_type;
1471 	u16 ether_type;
1472 	struct ib_gid_attr sgid_attr;
1473 	union ib_gid dgid;
1474 	bool is_eth = false;
1475 	bool is_vlan = false;
1476 	bool is_grh = false;
1477 	bool is_udp = false;
1478 	u8 ip_version = 0;
1479 	u16 vlan_id = 0xFFFF;
1480 	void *buf;
1481 	int i, rc = 0, size;
1482 
1483 	memset(&qp->qp1_hdr, 0, sizeof(qp->qp1_hdr));
1484 
1485 	rc = ib_get_cached_gid(ibdev, 1,
1486 			       qplib_ah->host_sgid_index, &sgid,
1487 			       &sgid_attr);
1488 	if (rc) {
1489 		dev_err(rdev_to_dev(qp->rdev),
1490 			"Failed to query gid at index %d",
1491 			qplib_ah->host_sgid_index);
1492 		return rc;
1493 	}
1494 	if (sgid_attr.ndev) {
1495 		if (is_vlan_dev(sgid_attr.ndev))
1496 			vlan_id = vlan_dev_vlan_id(sgid_attr.ndev);
1497 		dev_put(sgid_attr.ndev);
1498 	}
1499 	/* Get network header type for this GID */
1500 	nw_type = ib_gid_to_network_type(sgid_attr.gid_type, &sgid);
1501 	switch (nw_type) {
1502 	case RDMA_NETWORK_IPV4:
1503 		nw_type = BNXT_RE_ROCEV2_IPV4_PACKET;
1504 		break;
1505 	case RDMA_NETWORK_IPV6:
1506 		nw_type = BNXT_RE_ROCEV2_IPV6_PACKET;
1507 		break;
1508 	default:
1509 		nw_type = BNXT_RE_ROCE_V1_PACKET;
1510 		break;
1511 	}
1512 	memcpy(&dgid.raw, &qplib_ah->dgid, 16);
1513 	is_udp = sgid_attr.gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP;
1514 	if (is_udp) {
1515 		if (ipv6_addr_v4mapped((struct in6_addr *)&sgid)) {
1516 			ip_version = 4;
1517 			ether_type = ETH_P_IP;
1518 		} else {
1519 			ip_version = 6;
1520 			ether_type = ETH_P_IPV6;
1521 		}
1522 		is_grh = false;
1523 	} else {
1524 		ether_type = ETH_P_IBOE;
1525 		is_grh = true;
1526 	}
1527 
1528 	is_eth = true;
1529 	is_vlan = (vlan_id && (vlan_id < 0x1000)) ? true : false;
1530 
1531 	ib_ud_header_init(payload_size, !is_eth, is_eth, is_vlan, is_grh,
1532 			  ip_version, is_udp, 0, &qp->qp1_hdr);
1533 
1534 	/* ETH */
1535 	ether_addr_copy(qp->qp1_hdr.eth.dmac_h, ah->qplib_ah.dmac);
1536 	ether_addr_copy(qp->qp1_hdr.eth.smac_h, qp->qplib_qp.smac);
1537 
1538 	/* For vlan, check the sgid for vlan existence */
1539 
1540 	if (!is_vlan) {
1541 		qp->qp1_hdr.eth.type = cpu_to_be16(ether_type);
1542 	} else {
1543 		qp->qp1_hdr.vlan.type = cpu_to_be16(ether_type);
1544 		qp->qp1_hdr.vlan.tag = cpu_to_be16(vlan_id);
1545 	}
1546 
1547 	if (is_grh || (ip_version == 6)) {
1548 		memcpy(qp->qp1_hdr.grh.source_gid.raw, sgid.raw, sizeof(sgid));
1549 		memcpy(qp->qp1_hdr.grh.destination_gid.raw, qplib_ah->dgid.data,
1550 		       sizeof(sgid));
1551 		qp->qp1_hdr.grh.hop_limit     = qplib_ah->hop_limit;
1552 	}
1553 
1554 	if (ip_version == 4) {
1555 		qp->qp1_hdr.ip4.tos = 0;
1556 		qp->qp1_hdr.ip4.id = 0;
1557 		qp->qp1_hdr.ip4.frag_off = htons(IP_DF);
1558 		qp->qp1_hdr.ip4.ttl = qplib_ah->hop_limit;
1559 
1560 		memcpy(&qp->qp1_hdr.ip4.saddr, sgid.raw + 12, 4);
1561 		memcpy(&qp->qp1_hdr.ip4.daddr, qplib_ah->dgid.data + 12, 4);
1562 		qp->qp1_hdr.ip4.check = ib_ud_ip4_csum(&qp->qp1_hdr);
1563 	}
1564 
1565 	if (is_udp) {
1566 		qp->qp1_hdr.udp.dport = htons(ROCE_V2_UDP_DPORT);
1567 		qp->qp1_hdr.udp.sport = htons(0x8CD1);
1568 		qp->qp1_hdr.udp.csum = 0;
1569 	}
1570 
1571 	/* BTH */
1572 	if (wr->opcode == IB_WR_SEND_WITH_IMM) {
1573 		qp->qp1_hdr.bth.opcode = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE;
1574 		qp->qp1_hdr.immediate_present = 1;
1575 	} else {
1576 		qp->qp1_hdr.bth.opcode = IB_OPCODE_UD_SEND_ONLY;
1577 	}
1578 	if (wr->send_flags & IB_SEND_SOLICITED)
1579 		qp->qp1_hdr.bth.solicited_event = 1;
1580 	/* pad_count */
1581 	qp->qp1_hdr.bth.pad_count = (4 - payload_size) & 3;
1582 
1583 	/* P_key for QP1 is for all members */
1584 	qp->qp1_hdr.bth.pkey = cpu_to_be16(0xFFFF);
1585 	qp->qp1_hdr.bth.destination_qpn = IB_QP1;
1586 	qp->qp1_hdr.bth.ack_req = 0;
1587 	qp->send_psn++;
1588 	qp->send_psn &= BTH_PSN_MASK;
1589 	qp->qp1_hdr.bth.psn = cpu_to_be32(qp->send_psn);
1590 	/* DETH */
1591 	/* Use the priviledged Q_Key for QP1 */
1592 	qp->qp1_hdr.deth.qkey = cpu_to_be32(IB_QP1_QKEY);
1593 	qp->qp1_hdr.deth.source_qpn = IB_QP1;
1594 
1595 	/* Pack the QP1 to the transmit buffer */
1596 	buf = bnxt_qplib_get_qp1_sq_buf(&qp->qplib_qp, &sge);
1597 	if (buf) {
1598 		size = ib_ud_header_pack(&qp->qp1_hdr, buf);
1599 		for (i = wqe->num_sge; i; i--) {
1600 			wqe->sg_list[i].addr = wqe->sg_list[i - 1].addr;
1601 			wqe->sg_list[i].lkey = wqe->sg_list[i - 1].lkey;
1602 			wqe->sg_list[i].size = wqe->sg_list[i - 1].size;
1603 		}
1604 
1605 		/*
1606 		 * Max Header buf size for IPV6 RoCE V2 is 86,
1607 		 * which is same as the QP1 SQ header buffer.
1608 		 * Header buf size for IPV4 RoCE V2 can be 66.
1609 		 * ETH(14) + VLAN(4)+ IP(20) + UDP (8) + BTH(20).
1610 		 * Subtract 20 bytes from QP1 SQ header buf size
1611 		 */
1612 		if (is_udp && ip_version == 4)
1613 			sge.size -= 20;
1614 		/*
1615 		 * Max Header buf size for RoCE V1 is 78.
1616 		 * ETH(14) + VLAN(4) + GRH(40) + BTH(20).
1617 		 * Subtract 8 bytes from QP1 SQ header buf size
1618 		 */
1619 		if (!is_udp)
1620 			sge.size -= 8;
1621 
1622 		/* Subtract 4 bytes for non vlan packets */
1623 		if (!is_vlan)
1624 			sge.size -= 4;
1625 
1626 		wqe->sg_list[0].addr = sge.addr;
1627 		wqe->sg_list[0].lkey = sge.lkey;
1628 		wqe->sg_list[0].size = sge.size;
1629 		wqe->num_sge++;
1630 
1631 	} else {
1632 		dev_err(rdev_to_dev(qp->rdev), "QP1 buffer is empty!");
1633 		rc = -ENOMEM;
1634 	}
1635 	return rc;
1636 }
1637 
1638 /* For the MAD layer, it only provides the recv SGE the size of
1639  * ib_grh + MAD datagram.  No Ethernet headers, Ethertype, BTH, DETH,
1640  * nor RoCE iCRC.  The Cu+ solution must provide buffer for the entire
1641  * receive packet (334 bytes) with no VLAN and then copy the GRH
1642  * and the MAD datagram out to the provided SGE.
1643  */
1644 static int bnxt_re_build_qp1_shadow_qp_recv(struct bnxt_re_qp *qp,
1645 					    struct ib_recv_wr *wr,
1646 					    struct bnxt_qplib_swqe *wqe,
1647 					    int payload_size)
1648 {
1649 	struct bnxt_qplib_sge ref, sge;
1650 	u32 rq_prod_index;
1651 	struct bnxt_re_sqp_entries *sqp_entry;
1652 
1653 	rq_prod_index = bnxt_qplib_get_rq_prod_index(&qp->qplib_qp);
1654 
1655 	if (!bnxt_qplib_get_qp1_rq_buf(&qp->qplib_qp, &sge))
1656 		return -ENOMEM;
1657 
1658 	/* Create 1 SGE to receive the entire
1659 	 * ethernet packet
1660 	 */
1661 	/* Save the reference from ULP */
1662 	ref.addr = wqe->sg_list[0].addr;
1663 	ref.lkey = wqe->sg_list[0].lkey;
1664 	ref.size = wqe->sg_list[0].size;
1665 
1666 	sqp_entry = &qp->rdev->sqp_tbl[rq_prod_index];
1667 
1668 	/* SGE 1 */
1669 	wqe->sg_list[0].addr = sge.addr;
1670 	wqe->sg_list[0].lkey = sge.lkey;
1671 	wqe->sg_list[0].size = BNXT_QPLIB_MAX_QP1_RQ_HDR_SIZE_V2;
1672 	sge.size -= wqe->sg_list[0].size;
1673 
1674 	sqp_entry->sge.addr = ref.addr;
1675 	sqp_entry->sge.lkey = ref.lkey;
1676 	sqp_entry->sge.size = ref.size;
1677 	/* Store the wrid for reporting completion */
1678 	sqp_entry->wrid = wqe->wr_id;
1679 	/* change the wqe->wrid to table index */
1680 	wqe->wr_id = rq_prod_index;
1681 	return 0;
1682 }
1683 
1684 static int is_ud_qp(struct bnxt_re_qp *qp)
1685 {
1686 	return qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_UD;
1687 }
1688 
1689 static int bnxt_re_build_send_wqe(struct bnxt_re_qp *qp,
1690 				  struct ib_send_wr *wr,
1691 				  struct bnxt_qplib_swqe *wqe)
1692 {
1693 	struct bnxt_re_ah *ah = NULL;
1694 
1695 	if (is_ud_qp(qp)) {
1696 		ah = container_of(ud_wr(wr)->ah, struct bnxt_re_ah, ib_ah);
1697 		wqe->send.q_key = ud_wr(wr)->remote_qkey;
1698 		wqe->send.dst_qp = ud_wr(wr)->remote_qpn;
1699 		wqe->send.avid = ah->qplib_ah.id;
1700 	}
1701 	switch (wr->opcode) {
1702 	case IB_WR_SEND:
1703 		wqe->type = BNXT_QPLIB_SWQE_TYPE_SEND;
1704 		break;
1705 	case IB_WR_SEND_WITH_IMM:
1706 		wqe->type = BNXT_QPLIB_SWQE_TYPE_SEND_WITH_IMM;
1707 		wqe->send.imm_data = wr->ex.imm_data;
1708 		break;
1709 	case IB_WR_SEND_WITH_INV:
1710 		wqe->type = BNXT_QPLIB_SWQE_TYPE_SEND_WITH_INV;
1711 		wqe->send.inv_key = wr->ex.invalidate_rkey;
1712 		break;
1713 	default:
1714 		return -EINVAL;
1715 	}
1716 	if (wr->send_flags & IB_SEND_SIGNALED)
1717 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP;
1718 	if (wr->send_flags & IB_SEND_FENCE)
1719 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE;
1720 	if (wr->send_flags & IB_SEND_SOLICITED)
1721 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT;
1722 	if (wr->send_flags & IB_SEND_INLINE)
1723 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_INLINE;
1724 
1725 	return 0;
1726 }
1727 
1728 static int bnxt_re_build_rdma_wqe(struct ib_send_wr *wr,
1729 				  struct bnxt_qplib_swqe *wqe)
1730 {
1731 	switch (wr->opcode) {
1732 	case IB_WR_RDMA_WRITE:
1733 		wqe->type = BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE;
1734 		break;
1735 	case IB_WR_RDMA_WRITE_WITH_IMM:
1736 		wqe->type = BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE_WITH_IMM;
1737 		wqe->rdma.imm_data = wr->ex.imm_data;
1738 		break;
1739 	case IB_WR_RDMA_READ:
1740 		wqe->type = BNXT_QPLIB_SWQE_TYPE_RDMA_READ;
1741 		wqe->rdma.inv_key = wr->ex.invalidate_rkey;
1742 		break;
1743 	default:
1744 		return -EINVAL;
1745 	}
1746 	wqe->rdma.remote_va = rdma_wr(wr)->remote_addr;
1747 	wqe->rdma.r_key = rdma_wr(wr)->rkey;
1748 	if (wr->send_flags & IB_SEND_SIGNALED)
1749 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP;
1750 	if (wr->send_flags & IB_SEND_FENCE)
1751 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE;
1752 	if (wr->send_flags & IB_SEND_SOLICITED)
1753 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT;
1754 	if (wr->send_flags & IB_SEND_INLINE)
1755 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_INLINE;
1756 
1757 	return 0;
1758 }
1759 
1760 static int bnxt_re_build_atomic_wqe(struct ib_send_wr *wr,
1761 				    struct bnxt_qplib_swqe *wqe)
1762 {
1763 	switch (wr->opcode) {
1764 	case IB_WR_ATOMIC_CMP_AND_SWP:
1765 		wqe->type = BNXT_QPLIB_SWQE_TYPE_ATOMIC_CMP_AND_SWP;
1766 		wqe->atomic.swap_data = atomic_wr(wr)->swap;
1767 		break;
1768 	case IB_WR_ATOMIC_FETCH_AND_ADD:
1769 		wqe->type = BNXT_QPLIB_SWQE_TYPE_ATOMIC_FETCH_AND_ADD;
1770 		wqe->atomic.cmp_data = atomic_wr(wr)->compare_add;
1771 		break;
1772 	default:
1773 		return -EINVAL;
1774 	}
1775 	wqe->atomic.remote_va = atomic_wr(wr)->remote_addr;
1776 	wqe->atomic.r_key = atomic_wr(wr)->rkey;
1777 	if (wr->send_flags & IB_SEND_SIGNALED)
1778 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP;
1779 	if (wr->send_flags & IB_SEND_FENCE)
1780 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE;
1781 	if (wr->send_flags & IB_SEND_SOLICITED)
1782 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT;
1783 	return 0;
1784 }
1785 
1786 static int bnxt_re_build_inv_wqe(struct ib_send_wr *wr,
1787 				 struct bnxt_qplib_swqe *wqe)
1788 {
1789 	wqe->type = BNXT_QPLIB_SWQE_TYPE_LOCAL_INV;
1790 	wqe->local_inv.inv_l_key = wr->ex.invalidate_rkey;
1791 
1792 	if (wr->send_flags & IB_SEND_SIGNALED)
1793 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP;
1794 	if (wr->send_flags & IB_SEND_FENCE)
1795 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE;
1796 	if (wr->send_flags & IB_SEND_SOLICITED)
1797 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT;
1798 
1799 	return 0;
1800 }
1801 
1802 static int bnxt_re_build_reg_wqe(struct ib_reg_wr *wr,
1803 				 struct bnxt_qplib_swqe *wqe)
1804 {
1805 	struct bnxt_re_mr *mr = container_of(wr->mr, struct bnxt_re_mr, ib_mr);
1806 	struct bnxt_qplib_frpl *qplib_frpl = &mr->qplib_frpl;
1807 	int access = wr->access;
1808 
1809 	wqe->frmr.pbl_ptr = (__le64 *)qplib_frpl->hwq.pbl_ptr[0];
1810 	wqe->frmr.pbl_dma_ptr = qplib_frpl->hwq.pbl_dma_ptr[0];
1811 	wqe->frmr.page_list = mr->pages;
1812 	wqe->frmr.page_list_len = mr->npages;
1813 	wqe->frmr.levels = qplib_frpl->hwq.level + 1;
1814 	wqe->type = BNXT_QPLIB_SWQE_TYPE_REG_MR;
1815 
1816 	if (wr->wr.send_flags & IB_SEND_FENCE)
1817 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE;
1818 	if (wr->wr.send_flags & IB_SEND_SIGNALED)
1819 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP;
1820 
1821 	if (access & IB_ACCESS_LOCAL_WRITE)
1822 		wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_LOCAL_WRITE;
1823 	if (access & IB_ACCESS_REMOTE_READ)
1824 		wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_REMOTE_READ;
1825 	if (access & IB_ACCESS_REMOTE_WRITE)
1826 		wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_REMOTE_WRITE;
1827 	if (access & IB_ACCESS_REMOTE_ATOMIC)
1828 		wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_REMOTE_ATOMIC;
1829 	if (access & IB_ACCESS_MW_BIND)
1830 		wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_WINDOW_BIND;
1831 
1832 	wqe->frmr.l_key = wr->key;
1833 	wqe->frmr.length = wr->mr->length;
1834 	wqe->frmr.pbl_pg_sz_log = (wr->mr->page_size >> PAGE_SHIFT_4K) - 1;
1835 	wqe->frmr.va = wr->mr->iova;
1836 	return 0;
1837 }
1838 
1839 static int bnxt_re_copy_inline_data(struct bnxt_re_dev *rdev,
1840 				    struct ib_send_wr *wr,
1841 				    struct bnxt_qplib_swqe *wqe)
1842 {
1843 	/*  Copy the inline data to the data  field */
1844 	u8 *in_data;
1845 	u32 i, sge_len;
1846 	void *sge_addr;
1847 
1848 	in_data = wqe->inline_data;
1849 	for (i = 0; i < wr->num_sge; i++) {
1850 		sge_addr = (void *)(unsigned long)
1851 				wr->sg_list[i].addr;
1852 		sge_len = wr->sg_list[i].length;
1853 
1854 		if ((sge_len + wqe->inline_len) >
1855 		    BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH) {
1856 			dev_err(rdev_to_dev(rdev),
1857 				"Inline data size requested > supported value");
1858 			return -EINVAL;
1859 		}
1860 		sge_len = wr->sg_list[i].length;
1861 
1862 		memcpy(in_data, sge_addr, sge_len);
1863 		in_data += wr->sg_list[i].length;
1864 		wqe->inline_len += wr->sg_list[i].length;
1865 	}
1866 	return wqe->inline_len;
1867 }
1868 
1869 static int bnxt_re_copy_wr_payload(struct bnxt_re_dev *rdev,
1870 				   struct ib_send_wr *wr,
1871 				   struct bnxt_qplib_swqe *wqe)
1872 {
1873 	int payload_sz = 0;
1874 
1875 	if (wr->send_flags & IB_SEND_INLINE)
1876 		payload_sz = bnxt_re_copy_inline_data(rdev, wr, wqe);
1877 	else
1878 		payload_sz = bnxt_re_build_sgl(wr->sg_list, wqe->sg_list,
1879 					       wqe->num_sge);
1880 
1881 	return payload_sz;
1882 }
1883 
1884 static int bnxt_re_post_send_shadow_qp(struct bnxt_re_dev *rdev,
1885 				       struct bnxt_re_qp *qp,
1886 				struct ib_send_wr *wr)
1887 {
1888 	struct bnxt_qplib_swqe wqe;
1889 	int rc = 0, payload_sz = 0;
1890 	unsigned long flags;
1891 
1892 	spin_lock_irqsave(&qp->sq_lock, flags);
1893 	memset(&wqe, 0, sizeof(wqe));
1894 	while (wr) {
1895 		/* House keeping */
1896 		memset(&wqe, 0, sizeof(wqe));
1897 
1898 		/* Common */
1899 		wqe.num_sge = wr->num_sge;
1900 		if (wr->num_sge > qp->qplib_qp.sq.max_sge) {
1901 			dev_err(rdev_to_dev(rdev),
1902 				"Limit exceeded for Send SGEs");
1903 			rc = -EINVAL;
1904 			goto bad;
1905 		}
1906 
1907 		payload_sz = bnxt_re_copy_wr_payload(qp->rdev, wr, &wqe);
1908 		if (payload_sz < 0) {
1909 			rc = -EINVAL;
1910 			goto bad;
1911 		}
1912 		wqe.wr_id = wr->wr_id;
1913 
1914 		wqe.type = BNXT_QPLIB_SWQE_TYPE_SEND;
1915 
1916 		rc = bnxt_re_build_send_wqe(qp, wr, &wqe);
1917 		if (!rc)
1918 			rc = bnxt_qplib_post_send(&qp->qplib_qp, &wqe);
1919 bad:
1920 		if (rc) {
1921 			dev_err(rdev_to_dev(rdev),
1922 				"Post send failed opcode = %#x rc = %d",
1923 				wr->opcode, rc);
1924 			break;
1925 		}
1926 		wr = wr->next;
1927 	}
1928 	bnxt_qplib_post_send_db(&qp->qplib_qp);
1929 	spin_unlock_irqrestore(&qp->sq_lock, flags);
1930 	return rc;
1931 }
1932 
1933 int bnxt_re_post_send(struct ib_qp *ib_qp, struct ib_send_wr *wr,
1934 		      struct ib_send_wr **bad_wr)
1935 {
1936 	struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
1937 	struct bnxt_qplib_swqe wqe;
1938 	int rc = 0, payload_sz = 0;
1939 	unsigned long flags;
1940 
1941 	spin_lock_irqsave(&qp->sq_lock, flags);
1942 	while (wr) {
1943 		/* House keeping */
1944 		memset(&wqe, 0, sizeof(wqe));
1945 
1946 		/* Common */
1947 		wqe.num_sge = wr->num_sge;
1948 		if (wr->num_sge > qp->qplib_qp.sq.max_sge) {
1949 			dev_err(rdev_to_dev(qp->rdev),
1950 				"Limit exceeded for Send SGEs");
1951 			rc = -EINVAL;
1952 			goto bad;
1953 		}
1954 
1955 		payload_sz = bnxt_re_copy_wr_payload(qp->rdev, wr, &wqe);
1956 		if (payload_sz < 0) {
1957 			rc = -EINVAL;
1958 			goto bad;
1959 		}
1960 		wqe.wr_id = wr->wr_id;
1961 
1962 		switch (wr->opcode) {
1963 		case IB_WR_SEND:
1964 		case IB_WR_SEND_WITH_IMM:
1965 			if (ib_qp->qp_type == IB_QPT_GSI) {
1966 				rc = bnxt_re_build_qp1_send_v2(qp, wr, &wqe,
1967 							       payload_sz);
1968 				if (rc)
1969 					goto bad;
1970 				wqe.rawqp1.lflags |=
1971 					SQ_SEND_RAWETH_QP1_LFLAGS_ROCE_CRC;
1972 			}
1973 			switch (wr->send_flags) {
1974 			case IB_SEND_IP_CSUM:
1975 				wqe.rawqp1.lflags |=
1976 					SQ_SEND_RAWETH_QP1_LFLAGS_IP_CHKSUM;
1977 				break;
1978 			default:
1979 				break;
1980 			}
1981 			/* Fall thru to build the wqe */
1982 		case IB_WR_SEND_WITH_INV:
1983 			rc = bnxt_re_build_send_wqe(qp, wr, &wqe);
1984 			break;
1985 		case IB_WR_RDMA_WRITE:
1986 		case IB_WR_RDMA_WRITE_WITH_IMM:
1987 		case IB_WR_RDMA_READ:
1988 			rc = bnxt_re_build_rdma_wqe(wr, &wqe);
1989 			break;
1990 		case IB_WR_ATOMIC_CMP_AND_SWP:
1991 		case IB_WR_ATOMIC_FETCH_AND_ADD:
1992 			rc = bnxt_re_build_atomic_wqe(wr, &wqe);
1993 			break;
1994 		case IB_WR_RDMA_READ_WITH_INV:
1995 			dev_err(rdev_to_dev(qp->rdev),
1996 				"RDMA Read with Invalidate is not supported");
1997 			rc = -EINVAL;
1998 			goto bad;
1999 		case IB_WR_LOCAL_INV:
2000 			rc = bnxt_re_build_inv_wqe(wr, &wqe);
2001 			break;
2002 		case IB_WR_REG_MR:
2003 			rc = bnxt_re_build_reg_wqe(reg_wr(wr), &wqe);
2004 			break;
2005 		default:
2006 			/* Unsupported WRs */
2007 			dev_err(rdev_to_dev(qp->rdev),
2008 				"WR (%#x) is not supported", wr->opcode);
2009 			rc = -EINVAL;
2010 			goto bad;
2011 		}
2012 		if (!rc)
2013 			rc = bnxt_qplib_post_send(&qp->qplib_qp, &wqe);
2014 bad:
2015 		if (rc) {
2016 			dev_err(rdev_to_dev(qp->rdev),
2017 				"post_send failed op:%#x qps = %#x rc = %d\n",
2018 				wr->opcode, qp->qplib_qp.state, rc);
2019 			*bad_wr = wr;
2020 			break;
2021 		}
2022 		wr = wr->next;
2023 	}
2024 	bnxt_qplib_post_send_db(&qp->qplib_qp);
2025 	spin_unlock_irqrestore(&qp->sq_lock, flags);
2026 
2027 	return rc;
2028 }
2029 
2030 static int bnxt_re_post_recv_shadow_qp(struct bnxt_re_dev *rdev,
2031 				       struct bnxt_re_qp *qp,
2032 				       struct ib_recv_wr *wr)
2033 {
2034 	struct bnxt_qplib_swqe wqe;
2035 	int rc = 0, payload_sz = 0;
2036 
2037 	memset(&wqe, 0, sizeof(wqe));
2038 	while (wr) {
2039 		/* House keeping */
2040 		memset(&wqe, 0, sizeof(wqe));
2041 
2042 		/* Common */
2043 		wqe.num_sge = wr->num_sge;
2044 		if (wr->num_sge > qp->qplib_qp.rq.max_sge) {
2045 			dev_err(rdev_to_dev(rdev),
2046 				"Limit exceeded for Receive SGEs");
2047 			rc = -EINVAL;
2048 			break;
2049 		}
2050 		payload_sz = bnxt_re_build_sgl(wr->sg_list, wqe.sg_list,
2051 					       wr->num_sge);
2052 		wqe.wr_id = wr->wr_id;
2053 		wqe.type = BNXT_QPLIB_SWQE_TYPE_RECV;
2054 
2055 		rc = bnxt_qplib_post_recv(&qp->qplib_qp, &wqe);
2056 		if (rc)
2057 			break;
2058 
2059 		wr = wr->next;
2060 	}
2061 	if (!rc)
2062 		bnxt_qplib_post_recv_db(&qp->qplib_qp);
2063 	return rc;
2064 }
2065 
2066 int bnxt_re_post_recv(struct ib_qp *ib_qp, struct ib_recv_wr *wr,
2067 		      struct ib_recv_wr **bad_wr)
2068 {
2069 	struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
2070 	struct bnxt_qplib_swqe wqe;
2071 	int rc = 0, payload_sz = 0;
2072 
2073 	while (wr) {
2074 		/* House keeping */
2075 		memset(&wqe, 0, sizeof(wqe));
2076 
2077 		/* Common */
2078 		wqe.num_sge = wr->num_sge;
2079 		if (wr->num_sge > qp->qplib_qp.rq.max_sge) {
2080 			dev_err(rdev_to_dev(qp->rdev),
2081 				"Limit exceeded for Receive SGEs");
2082 			rc = -EINVAL;
2083 			*bad_wr = wr;
2084 			break;
2085 		}
2086 
2087 		payload_sz = bnxt_re_build_sgl(wr->sg_list, wqe.sg_list,
2088 					       wr->num_sge);
2089 		wqe.wr_id = wr->wr_id;
2090 		wqe.type = BNXT_QPLIB_SWQE_TYPE_RECV;
2091 
2092 		if (ib_qp->qp_type == IB_QPT_GSI)
2093 			rc = bnxt_re_build_qp1_shadow_qp_recv(qp, wr, &wqe,
2094 							      payload_sz);
2095 		if (!rc)
2096 			rc = bnxt_qplib_post_recv(&qp->qplib_qp, &wqe);
2097 		if (rc) {
2098 			*bad_wr = wr;
2099 			break;
2100 		}
2101 		wr = wr->next;
2102 	}
2103 	bnxt_qplib_post_recv_db(&qp->qplib_qp);
2104 	return rc;
2105 }
2106 
2107 /* Completion Queues */
2108 int bnxt_re_destroy_cq(struct ib_cq *ib_cq)
2109 {
2110 	struct bnxt_re_cq *cq = container_of(ib_cq, struct bnxt_re_cq, ib_cq);
2111 	struct bnxt_re_dev *rdev = cq->rdev;
2112 	int rc;
2113 
2114 	rc = bnxt_qplib_destroy_cq(&rdev->qplib_res, &cq->qplib_cq);
2115 	if (rc) {
2116 		dev_err(rdev_to_dev(rdev), "Failed to destroy HW CQ");
2117 		return rc;
2118 	}
2119 	if (cq->umem && !IS_ERR(cq->umem))
2120 		ib_umem_release(cq->umem);
2121 
2122 	if (cq) {
2123 		kfree(cq->cql);
2124 		kfree(cq);
2125 	}
2126 	atomic_dec(&rdev->cq_count);
2127 	rdev->nq.budget--;
2128 	return 0;
2129 }
2130 
2131 struct ib_cq *bnxt_re_create_cq(struct ib_device *ibdev,
2132 				const struct ib_cq_init_attr *attr,
2133 				struct ib_ucontext *context,
2134 				struct ib_udata *udata)
2135 {
2136 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
2137 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
2138 	struct bnxt_re_cq *cq = NULL;
2139 	int rc, entries;
2140 	int cqe = attr->cqe;
2141 
2142 	/* Validate CQ fields */
2143 	if (cqe < 1 || cqe > dev_attr->max_cq_wqes) {
2144 		dev_err(rdev_to_dev(rdev), "Failed to create CQ -max exceeded");
2145 		return ERR_PTR(-EINVAL);
2146 	}
2147 	cq = kzalloc(sizeof(*cq), GFP_KERNEL);
2148 	if (!cq)
2149 		return ERR_PTR(-ENOMEM);
2150 
2151 	cq->rdev = rdev;
2152 	cq->qplib_cq.cq_handle = (u64)(unsigned long)(&cq->qplib_cq);
2153 
2154 	entries = roundup_pow_of_two(cqe + 1);
2155 	if (entries > dev_attr->max_cq_wqes + 1)
2156 		entries = dev_attr->max_cq_wqes + 1;
2157 
2158 	if (context) {
2159 		struct bnxt_re_cq_req req;
2160 		struct bnxt_re_ucontext *uctx = container_of
2161 						(context,
2162 						 struct bnxt_re_ucontext,
2163 						 ib_uctx);
2164 		if (ib_copy_from_udata(&req, udata, sizeof(req))) {
2165 			rc = -EFAULT;
2166 			goto fail;
2167 		}
2168 
2169 		cq->umem = ib_umem_get(context, req.cq_va,
2170 				       entries * sizeof(struct cq_base),
2171 				       IB_ACCESS_LOCAL_WRITE, 1);
2172 		if (IS_ERR(cq->umem)) {
2173 			rc = PTR_ERR(cq->umem);
2174 			goto fail;
2175 		}
2176 		cq->qplib_cq.sghead = cq->umem->sg_head.sgl;
2177 		cq->qplib_cq.nmap = cq->umem->nmap;
2178 		cq->qplib_cq.dpi = uctx->dpi;
2179 	} else {
2180 		cq->max_cql = min_t(u32, entries, MAX_CQL_PER_POLL);
2181 		cq->cql = kcalloc(cq->max_cql, sizeof(struct bnxt_qplib_cqe),
2182 				  GFP_KERNEL);
2183 		if (!cq->cql) {
2184 			rc = -ENOMEM;
2185 			goto fail;
2186 		}
2187 
2188 		cq->qplib_cq.dpi = &rdev->dpi_privileged;
2189 		cq->qplib_cq.sghead = NULL;
2190 		cq->qplib_cq.nmap = 0;
2191 	}
2192 	cq->qplib_cq.max_wqe = entries;
2193 	cq->qplib_cq.cnq_hw_ring_id = rdev->nq.ring_id;
2194 
2195 	rc = bnxt_qplib_create_cq(&rdev->qplib_res, &cq->qplib_cq);
2196 	if (rc) {
2197 		dev_err(rdev_to_dev(rdev), "Failed to create HW CQ");
2198 		goto fail;
2199 	}
2200 
2201 	cq->ib_cq.cqe = entries;
2202 	cq->cq_period = cq->qplib_cq.period;
2203 	rdev->nq.budget++;
2204 
2205 	atomic_inc(&rdev->cq_count);
2206 
2207 	if (context) {
2208 		struct bnxt_re_cq_resp resp;
2209 
2210 		resp.cqid = cq->qplib_cq.id;
2211 		resp.tail = cq->qplib_cq.hwq.cons;
2212 		resp.phase = cq->qplib_cq.period;
2213 		resp.rsvd = 0;
2214 		rc = ib_copy_to_udata(udata, &resp, sizeof(resp));
2215 		if (rc) {
2216 			dev_err(rdev_to_dev(rdev), "Failed to copy CQ udata");
2217 			bnxt_qplib_destroy_cq(&rdev->qplib_res, &cq->qplib_cq);
2218 			goto c2fail;
2219 		}
2220 	}
2221 
2222 	return &cq->ib_cq;
2223 
2224 c2fail:
2225 	if (context)
2226 		ib_umem_release(cq->umem);
2227 fail:
2228 	kfree(cq->cql);
2229 	kfree(cq);
2230 	return ERR_PTR(rc);
2231 }
2232 
2233 static u8 __req_to_ib_wc_status(u8 qstatus)
2234 {
2235 	switch (qstatus) {
2236 	case CQ_REQ_STATUS_OK:
2237 		return IB_WC_SUCCESS;
2238 	case CQ_REQ_STATUS_BAD_RESPONSE_ERR:
2239 		return IB_WC_BAD_RESP_ERR;
2240 	case CQ_REQ_STATUS_LOCAL_LENGTH_ERR:
2241 		return IB_WC_LOC_LEN_ERR;
2242 	case CQ_REQ_STATUS_LOCAL_QP_OPERATION_ERR:
2243 		return IB_WC_LOC_QP_OP_ERR;
2244 	case CQ_REQ_STATUS_LOCAL_PROTECTION_ERR:
2245 		return IB_WC_LOC_PROT_ERR;
2246 	case CQ_REQ_STATUS_MEMORY_MGT_OPERATION_ERR:
2247 		return IB_WC_GENERAL_ERR;
2248 	case CQ_REQ_STATUS_REMOTE_INVALID_REQUEST_ERR:
2249 		return IB_WC_REM_INV_REQ_ERR;
2250 	case CQ_REQ_STATUS_REMOTE_ACCESS_ERR:
2251 		return IB_WC_REM_ACCESS_ERR;
2252 	case CQ_REQ_STATUS_REMOTE_OPERATION_ERR:
2253 		return IB_WC_REM_OP_ERR;
2254 	case CQ_REQ_STATUS_RNR_NAK_RETRY_CNT_ERR:
2255 		return IB_WC_RNR_RETRY_EXC_ERR;
2256 	case CQ_REQ_STATUS_TRANSPORT_RETRY_CNT_ERR:
2257 		return IB_WC_RETRY_EXC_ERR;
2258 	case CQ_REQ_STATUS_WORK_REQUEST_FLUSHED_ERR:
2259 		return IB_WC_WR_FLUSH_ERR;
2260 	default:
2261 		return IB_WC_GENERAL_ERR;
2262 	}
2263 	return 0;
2264 }
2265 
2266 static u8 __rawqp1_to_ib_wc_status(u8 qstatus)
2267 {
2268 	switch (qstatus) {
2269 	case CQ_RES_RAWETH_QP1_STATUS_OK:
2270 		return IB_WC_SUCCESS;
2271 	case CQ_RES_RAWETH_QP1_STATUS_LOCAL_ACCESS_ERROR:
2272 		return IB_WC_LOC_ACCESS_ERR;
2273 	case CQ_RES_RAWETH_QP1_STATUS_HW_LOCAL_LENGTH_ERR:
2274 		return IB_WC_LOC_LEN_ERR;
2275 	case CQ_RES_RAWETH_QP1_STATUS_LOCAL_PROTECTION_ERR:
2276 		return IB_WC_LOC_PROT_ERR;
2277 	case CQ_RES_RAWETH_QP1_STATUS_LOCAL_QP_OPERATION_ERR:
2278 		return IB_WC_LOC_QP_OP_ERR;
2279 	case CQ_RES_RAWETH_QP1_STATUS_MEMORY_MGT_OPERATION_ERR:
2280 		return IB_WC_GENERAL_ERR;
2281 	case CQ_RES_RAWETH_QP1_STATUS_WORK_REQUEST_FLUSHED_ERR:
2282 		return IB_WC_WR_FLUSH_ERR;
2283 	case CQ_RES_RAWETH_QP1_STATUS_HW_FLUSH_ERR:
2284 		return IB_WC_WR_FLUSH_ERR;
2285 	default:
2286 		return IB_WC_GENERAL_ERR;
2287 	}
2288 }
2289 
2290 static u8 __rc_to_ib_wc_status(u8 qstatus)
2291 {
2292 	switch (qstatus) {
2293 	case CQ_RES_RC_STATUS_OK:
2294 		return IB_WC_SUCCESS;
2295 	case CQ_RES_RC_STATUS_LOCAL_ACCESS_ERROR:
2296 		return IB_WC_LOC_ACCESS_ERR;
2297 	case CQ_RES_RC_STATUS_LOCAL_LENGTH_ERR:
2298 		return IB_WC_LOC_LEN_ERR;
2299 	case CQ_RES_RC_STATUS_LOCAL_PROTECTION_ERR:
2300 		return IB_WC_LOC_PROT_ERR;
2301 	case CQ_RES_RC_STATUS_LOCAL_QP_OPERATION_ERR:
2302 		return IB_WC_LOC_QP_OP_ERR;
2303 	case CQ_RES_RC_STATUS_MEMORY_MGT_OPERATION_ERR:
2304 		return IB_WC_GENERAL_ERR;
2305 	case CQ_RES_RC_STATUS_REMOTE_INVALID_REQUEST_ERR:
2306 		return IB_WC_REM_INV_REQ_ERR;
2307 	case CQ_RES_RC_STATUS_WORK_REQUEST_FLUSHED_ERR:
2308 		return IB_WC_WR_FLUSH_ERR;
2309 	case CQ_RES_RC_STATUS_HW_FLUSH_ERR:
2310 		return IB_WC_WR_FLUSH_ERR;
2311 	default:
2312 		return IB_WC_GENERAL_ERR;
2313 	}
2314 }
2315 
2316 static void bnxt_re_process_req_wc(struct ib_wc *wc, struct bnxt_qplib_cqe *cqe)
2317 {
2318 	switch (cqe->type) {
2319 	case BNXT_QPLIB_SWQE_TYPE_SEND:
2320 		wc->opcode = IB_WC_SEND;
2321 		break;
2322 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_IMM:
2323 		wc->opcode = IB_WC_SEND;
2324 		wc->wc_flags |= IB_WC_WITH_IMM;
2325 		break;
2326 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_INV:
2327 		wc->opcode = IB_WC_SEND;
2328 		wc->wc_flags |= IB_WC_WITH_INVALIDATE;
2329 		break;
2330 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE:
2331 		wc->opcode = IB_WC_RDMA_WRITE;
2332 		break;
2333 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE_WITH_IMM:
2334 		wc->opcode = IB_WC_RDMA_WRITE;
2335 		wc->wc_flags |= IB_WC_WITH_IMM;
2336 		break;
2337 	case BNXT_QPLIB_SWQE_TYPE_RDMA_READ:
2338 		wc->opcode = IB_WC_RDMA_READ;
2339 		break;
2340 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_CMP_AND_SWP:
2341 		wc->opcode = IB_WC_COMP_SWAP;
2342 		break;
2343 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_FETCH_AND_ADD:
2344 		wc->opcode = IB_WC_FETCH_ADD;
2345 		break;
2346 	case BNXT_QPLIB_SWQE_TYPE_LOCAL_INV:
2347 		wc->opcode = IB_WC_LOCAL_INV;
2348 		break;
2349 	case BNXT_QPLIB_SWQE_TYPE_REG_MR:
2350 		wc->opcode = IB_WC_REG_MR;
2351 		break;
2352 	default:
2353 		wc->opcode = IB_WC_SEND;
2354 		break;
2355 	}
2356 
2357 	wc->status = __req_to_ib_wc_status(cqe->status);
2358 }
2359 
2360 static int bnxt_re_check_packet_type(u16 raweth_qp1_flags,
2361 				     u16 raweth_qp1_flags2)
2362 {
2363 	bool is_udp = false, is_ipv6 = false, is_ipv4 = false;
2364 
2365 	/* raweth_qp1_flags Bit 9-6 indicates itype */
2366 	if ((raweth_qp1_flags & CQ_RES_RAWETH_QP1_RAWETH_QP1_FLAGS_ITYPE_ROCE)
2367 	    != CQ_RES_RAWETH_QP1_RAWETH_QP1_FLAGS_ITYPE_ROCE)
2368 		return -1;
2369 
2370 	if (raweth_qp1_flags2 &
2371 	    CQ_RES_RAWETH_QP1_RAWETH_QP1_FLAGS2_IP_CS_CALC &&
2372 	    raweth_qp1_flags2 &
2373 	    CQ_RES_RAWETH_QP1_RAWETH_QP1_FLAGS2_L4_CS_CALC) {
2374 		is_udp = true;
2375 		/* raweth_qp1_flags2 Bit 8 indicates ip_type. 0-v4 1 - v6 */
2376 		(raweth_qp1_flags2 &
2377 		 CQ_RES_RAWETH_QP1_RAWETH_QP1_FLAGS2_IP_TYPE) ?
2378 			(is_ipv6 = true) : (is_ipv4 = true);
2379 		return ((is_ipv6) ?
2380 			 BNXT_RE_ROCEV2_IPV6_PACKET :
2381 			 BNXT_RE_ROCEV2_IPV4_PACKET);
2382 	} else {
2383 		return BNXT_RE_ROCE_V1_PACKET;
2384 	}
2385 }
2386 
2387 static int bnxt_re_to_ib_nw_type(int nw_type)
2388 {
2389 	u8 nw_hdr_type = 0xFF;
2390 
2391 	switch (nw_type) {
2392 	case BNXT_RE_ROCE_V1_PACKET:
2393 		nw_hdr_type = RDMA_NETWORK_ROCE_V1;
2394 		break;
2395 	case BNXT_RE_ROCEV2_IPV4_PACKET:
2396 		nw_hdr_type = RDMA_NETWORK_IPV4;
2397 		break;
2398 	case BNXT_RE_ROCEV2_IPV6_PACKET:
2399 		nw_hdr_type = RDMA_NETWORK_IPV6;
2400 		break;
2401 	}
2402 	return nw_hdr_type;
2403 }
2404 
2405 static bool bnxt_re_is_loopback_packet(struct bnxt_re_dev *rdev,
2406 				       void *rq_hdr_buf)
2407 {
2408 	u8 *tmp_buf = NULL;
2409 	struct ethhdr *eth_hdr;
2410 	u16 eth_type;
2411 	bool rc = false;
2412 
2413 	tmp_buf = (u8 *)rq_hdr_buf;
2414 	/*
2415 	 * If dest mac is not same as I/F mac, this could be a
2416 	 * loopback address or multicast address, check whether
2417 	 * it is a loopback packet
2418 	 */
2419 	if (!ether_addr_equal(tmp_buf, rdev->netdev->dev_addr)) {
2420 		tmp_buf += 4;
2421 		/* Check the  ether type */
2422 		eth_hdr = (struct ethhdr *)tmp_buf;
2423 		eth_type = ntohs(eth_hdr->h_proto);
2424 		switch (eth_type) {
2425 		case ETH_P_IBOE:
2426 			rc = true;
2427 			break;
2428 		case ETH_P_IP:
2429 		case ETH_P_IPV6: {
2430 			u32 len;
2431 			struct udphdr *udp_hdr;
2432 
2433 			len = (eth_type == ETH_P_IP ? sizeof(struct iphdr) :
2434 						      sizeof(struct ipv6hdr));
2435 			tmp_buf += sizeof(struct ethhdr) + len;
2436 			udp_hdr = (struct udphdr *)tmp_buf;
2437 			if (ntohs(udp_hdr->dest) ==
2438 				    ROCE_V2_UDP_DPORT)
2439 				rc = true;
2440 			break;
2441 			}
2442 		default:
2443 			break;
2444 		}
2445 	}
2446 
2447 	return rc;
2448 }
2449 
2450 static int bnxt_re_process_raw_qp_pkt_rx(struct bnxt_re_qp *qp1_qp,
2451 					 struct bnxt_qplib_cqe *cqe)
2452 {
2453 	struct bnxt_re_dev *rdev = qp1_qp->rdev;
2454 	struct bnxt_re_sqp_entries *sqp_entry = NULL;
2455 	struct bnxt_re_qp *qp = rdev->qp1_sqp;
2456 	struct ib_send_wr *swr;
2457 	struct ib_ud_wr udwr;
2458 	struct ib_recv_wr rwr;
2459 	int pkt_type = 0;
2460 	u32 tbl_idx;
2461 	void *rq_hdr_buf;
2462 	dma_addr_t rq_hdr_buf_map;
2463 	dma_addr_t shrq_hdr_buf_map;
2464 	u32 offset = 0;
2465 	u32 skip_bytes = 0;
2466 	struct ib_sge s_sge[2];
2467 	struct ib_sge r_sge[2];
2468 	int rc;
2469 
2470 	memset(&udwr, 0, sizeof(udwr));
2471 	memset(&rwr, 0, sizeof(rwr));
2472 	memset(&s_sge, 0, sizeof(s_sge));
2473 	memset(&r_sge, 0, sizeof(r_sge));
2474 
2475 	swr = &udwr.wr;
2476 	tbl_idx = cqe->wr_id;
2477 
2478 	rq_hdr_buf = qp1_qp->qplib_qp.rq_hdr_buf +
2479 			(tbl_idx * qp1_qp->qplib_qp.rq_hdr_buf_size);
2480 	rq_hdr_buf_map = bnxt_qplib_get_qp_buf_from_index(&qp1_qp->qplib_qp,
2481 							  tbl_idx);
2482 
2483 	/* Shadow QP header buffer */
2484 	shrq_hdr_buf_map = bnxt_qplib_get_qp_buf_from_index(&qp->qplib_qp,
2485 							    tbl_idx);
2486 	sqp_entry = &rdev->sqp_tbl[tbl_idx];
2487 
2488 	/* Store this cqe */
2489 	memcpy(&sqp_entry->cqe, cqe, sizeof(struct bnxt_qplib_cqe));
2490 	sqp_entry->qp1_qp = qp1_qp;
2491 
2492 	/* Find packet type from the cqe */
2493 
2494 	pkt_type = bnxt_re_check_packet_type(cqe->raweth_qp1_flags,
2495 					     cqe->raweth_qp1_flags2);
2496 	if (pkt_type < 0) {
2497 		dev_err(rdev_to_dev(rdev), "Invalid packet\n");
2498 		return -EINVAL;
2499 	}
2500 
2501 	/* Adjust the offset for the user buffer and post in the rq */
2502 
2503 	if (pkt_type == BNXT_RE_ROCEV2_IPV4_PACKET)
2504 		offset = 20;
2505 
2506 	/*
2507 	 * QP1 loopback packet has 4 bytes of internal header before
2508 	 * ether header. Skip these four bytes.
2509 	 */
2510 	if (bnxt_re_is_loopback_packet(rdev, rq_hdr_buf))
2511 		skip_bytes = 4;
2512 
2513 	/* First send SGE . Skip the ether header*/
2514 	s_sge[0].addr = rq_hdr_buf_map + BNXT_QPLIB_MAX_QP1_RQ_ETH_HDR_SIZE
2515 			+ skip_bytes;
2516 	s_sge[0].lkey = 0xFFFFFFFF;
2517 	s_sge[0].length = offset ? BNXT_QPLIB_MAX_GRH_HDR_SIZE_IPV4 :
2518 				BNXT_QPLIB_MAX_GRH_HDR_SIZE_IPV6;
2519 
2520 	/* Second Send SGE */
2521 	s_sge[1].addr = s_sge[0].addr + s_sge[0].length +
2522 			BNXT_QPLIB_MAX_QP1_RQ_BDETH_HDR_SIZE;
2523 	if (pkt_type != BNXT_RE_ROCE_V1_PACKET)
2524 		s_sge[1].addr += 8;
2525 	s_sge[1].lkey = 0xFFFFFFFF;
2526 	s_sge[1].length = 256;
2527 
2528 	/* First recv SGE */
2529 
2530 	r_sge[0].addr = shrq_hdr_buf_map;
2531 	r_sge[0].lkey = 0xFFFFFFFF;
2532 	r_sge[0].length = 40;
2533 
2534 	r_sge[1].addr = sqp_entry->sge.addr + offset;
2535 	r_sge[1].lkey = sqp_entry->sge.lkey;
2536 	r_sge[1].length = BNXT_QPLIB_MAX_GRH_HDR_SIZE_IPV6 + 256 - offset;
2537 
2538 	/* Create receive work request */
2539 	rwr.num_sge = 2;
2540 	rwr.sg_list = r_sge;
2541 	rwr.wr_id = tbl_idx;
2542 	rwr.next = NULL;
2543 
2544 	rc = bnxt_re_post_recv_shadow_qp(rdev, qp, &rwr);
2545 	if (rc) {
2546 		dev_err(rdev_to_dev(rdev),
2547 			"Failed to post Rx buffers to shadow QP");
2548 		return -ENOMEM;
2549 	}
2550 
2551 	swr->num_sge = 2;
2552 	swr->sg_list = s_sge;
2553 	swr->wr_id = tbl_idx;
2554 	swr->opcode = IB_WR_SEND;
2555 	swr->next = NULL;
2556 
2557 	udwr.ah = &rdev->sqp_ah->ib_ah;
2558 	udwr.remote_qpn = rdev->qp1_sqp->qplib_qp.id;
2559 	udwr.remote_qkey = rdev->qp1_sqp->qplib_qp.qkey;
2560 
2561 	/* post data received  in the send queue */
2562 	rc = bnxt_re_post_send_shadow_qp(rdev, qp, swr);
2563 
2564 	return 0;
2565 }
2566 
2567 static void bnxt_re_process_res_rawqp1_wc(struct ib_wc *wc,
2568 					  struct bnxt_qplib_cqe *cqe)
2569 {
2570 	wc->opcode = IB_WC_RECV;
2571 	wc->status = __rawqp1_to_ib_wc_status(cqe->status);
2572 	wc->wc_flags |= IB_WC_GRH;
2573 }
2574 
2575 static void bnxt_re_process_res_rc_wc(struct ib_wc *wc,
2576 				      struct bnxt_qplib_cqe *cqe)
2577 {
2578 	wc->opcode = IB_WC_RECV;
2579 	wc->status = __rc_to_ib_wc_status(cqe->status);
2580 
2581 	if (cqe->flags & CQ_RES_RC_FLAGS_IMM)
2582 		wc->wc_flags |= IB_WC_WITH_IMM;
2583 	if (cqe->flags & CQ_RES_RC_FLAGS_INV)
2584 		wc->wc_flags |= IB_WC_WITH_INVALIDATE;
2585 	if ((cqe->flags & (CQ_RES_RC_FLAGS_RDMA | CQ_RES_RC_FLAGS_IMM)) ==
2586 	    (CQ_RES_RC_FLAGS_RDMA | CQ_RES_RC_FLAGS_IMM))
2587 		wc->opcode = IB_WC_RECV_RDMA_WITH_IMM;
2588 }
2589 
2590 static void bnxt_re_process_res_shadow_qp_wc(struct bnxt_re_qp *qp,
2591 					     struct ib_wc *wc,
2592 					     struct bnxt_qplib_cqe *cqe)
2593 {
2594 	u32 tbl_idx;
2595 	struct bnxt_re_dev *rdev = qp->rdev;
2596 	struct bnxt_re_qp *qp1_qp = NULL;
2597 	struct bnxt_qplib_cqe *orig_cqe = NULL;
2598 	struct bnxt_re_sqp_entries *sqp_entry = NULL;
2599 	int nw_type;
2600 
2601 	tbl_idx = cqe->wr_id;
2602 
2603 	sqp_entry = &rdev->sqp_tbl[tbl_idx];
2604 	qp1_qp = sqp_entry->qp1_qp;
2605 	orig_cqe = &sqp_entry->cqe;
2606 
2607 	wc->wr_id = sqp_entry->wrid;
2608 	wc->byte_len = orig_cqe->length;
2609 	wc->qp = &qp1_qp->ib_qp;
2610 
2611 	wc->ex.imm_data = orig_cqe->immdata;
2612 	wc->src_qp = orig_cqe->src_qp;
2613 	memcpy(wc->smac, orig_cqe->smac, ETH_ALEN);
2614 	wc->port_num = 1;
2615 	wc->vendor_err = orig_cqe->status;
2616 
2617 	wc->opcode = IB_WC_RECV;
2618 	wc->status = __rawqp1_to_ib_wc_status(orig_cqe->status);
2619 	wc->wc_flags |= IB_WC_GRH;
2620 
2621 	nw_type = bnxt_re_check_packet_type(orig_cqe->raweth_qp1_flags,
2622 					    orig_cqe->raweth_qp1_flags2);
2623 	if (nw_type >= 0) {
2624 		wc->network_hdr_type = bnxt_re_to_ib_nw_type(nw_type);
2625 		wc->wc_flags |= IB_WC_WITH_NETWORK_HDR_TYPE;
2626 	}
2627 }
2628 
2629 static void bnxt_re_process_res_ud_wc(struct ib_wc *wc,
2630 				      struct bnxt_qplib_cqe *cqe)
2631 {
2632 	wc->opcode = IB_WC_RECV;
2633 	wc->status = __rc_to_ib_wc_status(cqe->status);
2634 
2635 	if (cqe->flags & CQ_RES_RC_FLAGS_IMM)
2636 		wc->wc_flags |= IB_WC_WITH_IMM;
2637 	if (cqe->flags & CQ_RES_RC_FLAGS_INV)
2638 		wc->wc_flags |= IB_WC_WITH_INVALIDATE;
2639 	if ((cqe->flags & (CQ_RES_RC_FLAGS_RDMA | CQ_RES_RC_FLAGS_IMM)) ==
2640 	    (CQ_RES_RC_FLAGS_RDMA | CQ_RES_RC_FLAGS_IMM))
2641 		wc->opcode = IB_WC_RECV_RDMA_WITH_IMM;
2642 }
2643 
2644 int bnxt_re_poll_cq(struct ib_cq *ib_cq, int num_entries, struct ib_wc *wc)
2645 {
2646 	struct bnxt_re_cq *cq = container_of(ib_cq, struct bnxt_re_cq, ib_cq);
2647 	struct bnxt_re_qp *qp;
2648 	struct bnxt_qplib_cqe *cqe;
2649 	int i, ncqe, budget;
2650 	u32 tbl_idx;
2651 	struct bnxt_re_sqp_entries *sqp_entry = NULL;
2652 	unsigned long flags;
2653 
2654 	spin_lock_irqsave(&cq->cq_lock, flags);
2655 	budget = min_t(u32, num_entries, cq->max_cql);
2656 	if (!cq->cql) {
2657 		dev_err(rdev_to_dev(cq->rdev), "POLL CQ : no CQL to use");
2658 		goto exit;
2659 	}
2660 	cqe = &cq->cql[0];
2661 	while (budget) {
2662 		ncqe = bnxt_qplib_poll_cq(&cq->qplib_cq, cqe, budget);
2663 		if (!ncqe)
2664 			break;
2665 
2666 		for (i = 0; i < ncqe; i++, cqe++) {
2667 			/* Transcribe each qplib_wqe back to ib_wc */
2668 			memset(wc, 0, sizeof(*wc));
2669 
2670 			wc->wr_id = cqe->wr_id;
2671 			wc->byte_len = cqe->length;
2672 			qp = container_of
2673 				((struct bnxt_qplib_qp *)
2674 				 (unsigned long)(cqe->qp_handle),
2675 				 struct bnxt_re_qp, qplib_qp);
2676 			if (!qp) {
2677 				dev_err(rdev_to_dev(cq->rdev),
2678 					"POLL CQ : bad QP handle");
2679 				continue;
2680 			}
2681 			wc->qp = &qp->ib_qp;
2682 			wc->ex.imm_data = cqe->immdata;
2683 			wc->src_qp = cqe->src_qp;
2684 			memcpy(wc->smac, cqe->smac, ETH_ALEN);
2685 			wc->port_num = 1;
2686 			wc->vendor_err = cqe->status;
2687 
2688 			switch (cqe->opcode) {
2689 			case CQ_BASE_CQE_TYPE_REQ:
2690 				if (qp->qplib_qp.id ==
2691 				    qp->rdev->qp1_sqp->qplib_qp.id) {
2692 					/* Handle this completion with
2693 					 * the stored completion
2694 					 */
2695 					memset(wc, 0, sizeof(*wc));
2696 					continue;
2697 				}
2698 				bnxt_re_process_req_wc(wc, cqe);
2699 				break;
2700 			case CQ_BASE_CQE_TYPE_RES_RAWETH_QP1:
2701 				if (!cqe->status) {
2702 					int rc = 0;
2703 
2704 					rc = bnxt_re_process_raw_qp_pkt_rx
2705 								(qp, cqe);
2706 					if (!rc) {
2707 						memset(wc, 0, sizeof(*wc));
2708 						continue;
2709 					}
2710 					cqe->status = -1;
2711 				}
2712 				/* Errors need not be looped back.
2713 				 * But change the wr_id to the one
2714 				 * stored in the table
2715 				 */
2716 				tbl_idx = cqe->wr_id;
2717 				sqp_entry = &cq->rdev->sqp_tbl[tbl_idx];
2718 				wc->wr_id = sqp_entry->wrid;
2719 				bnxt_re_process_res_rawqp1_wc(wc, cqe);
2720 				break;
2721 			case CQ_BASE_CQE_TYPE_RES_RC:
2722 				bnxt_re_process_res_rc_wc(wc, cqe);
2723 				break;
2724 			case CQ_BASE_CQE_TYPE_RES_UD:
2725 				if (qp->qplib_qp.id ==
2726 				    qp->rdev->qp1_sqp->qplib_qp.id) {
2727 					/* Handle this completion with
2728 					 * the stored completion
2729 					 */
2730 					if (cqe->status) {
2731 						continue;
2732 					} else {
2733 						bnxt_re_process_res_shadow_qp_wc
2734 								(qp, wc, cqe);
2735 						break;
2736 					}
2737 				}
2738 				bnxt_re_process_res_ud_wc(wc, cqe);
2739 				break;
2740 			default:
2741 				dev_err(rdev_to_dev(cq->rdev),
2742 					"POLL CQ : type 0x%x not handled",
2743 					cqe->opcode);
2744 				continue;
2745 			}
2746 			wc++;
2747 			budget--;
2748 		}
2749 	}
2750 exit:
2751 	spin_unlock_irqrestore(&cq->cq_lock, flags);
2752 	return num_entries - budget;
2753 }
2754 
2755 int bnxt_re_req_notify_cq(struct ib_cq *ib_cq,
2756 			  enum ib_cq_notify_flags ib_cqn_flags)
2757 {
2758 	struct bnxt_re_cq *cq = container_of(ib_cq, struct bnxt_re_cq, ib_cq);
2759 	int type = 0;
2760 
2761 	/* Trigger on the very next completion */
2762 	if (ib_cqn_flags & IB_CQ_NEXT_COMP)
2763 		type = DBR_DBR_TYPE_CQ_ARMALL;
2764 	/* Trigger on the next solicited completion */
2765 	else if (ib_cqn_flags & IB_CQ_SOLICITED)
2766 		type = DBR_DBR_TYPE_CQ_ARMSE;
2767 
2768 	bnxt_qplib_req_notify_cq(&cq->qplib_cq, type);
2769 
2770 	return 0;
2771 }
2772 
2773 /* Memory Regions */
2774 struct ib_mr *bnxt_re_get_dma_mr(struct ib_pd *ib_pd, int mr_access_flags)
2775 {
2776 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
2777 	struct bnxt_re_dev *rdev = pd->rdev;
2778 	struct bnxt_re_mr *mr;
2779 	u64 pbl = 0;
2780 	int rc;
2781 
2782 	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
2783 	if (!mr)
2784 		return ERR_PTR(-ENOMEM);
2785 
2786 	mr->rdev = rdev;
2787 	mr->qplib_mr.pd = &pd->qplib_pd;
2788 	mr->qplib_mr.flags = __from_ib_access_flags(mr_access_flags);
2789 	mr->qplib_mr.type = CMDQ_ALLOCATE_MRW_MRW_FLAGS_PMR;
2790 
2791 	/* Allocate and register 0 as the address */
2792 	rc = bnxt_qplib_alloc_mrw(&rdev->qplib_res, &mr->qplib_mr);
2793 	if (rc)
2794 		goto fail;
2795 
2796 	mr->qplib_mr.hwq.level = PBL_LVL_MAX;
2797 	mr->qplib_mr.total_size = -1; /* Infinte length */
2798 	rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, &pbl, 0, false);
2799 	if (rc)
2800 		goto fail_mr;
2801 
2802 	mr->ib_mr.lkey = mr->qplib_mr.lkey;
2803 	if (mr_access_flags & (IB_ACCESS_REMOTE_WRITE | IB_ACCESS_REMOTE_READ |
2804 			       IB_ACCESS_REMOTE_ATOMIC))
2805 		mr->ib_mr.rkey = mr->ib_mr.lkey;
2806 	atomic_inc(&rdev->mr_count);
2807 
2808 	return &mr->ib_mr;
2809 
2810 fail_mr:
2811 	bnxt_qplib_free_mrw(&rdev->qplib_res, &mr->qplib_mr);
2812 fail:
2813 	kfree(mr);
2814 	return ERR_PTR(rc);
2815 }
2816 
2817 int bnxt_re_dereg_mr(struct ib_mr *ib_mr)
2818 {
2819 	struct bnxt_re_mr *mr = container_of(ib_mr, struct bnxt_re_mr, ib_mr);
2820 	struct bnxt_re_dev *rdev = mr->rdev;
2821 	int rc = 0;
2822 
2823 	if (mr->npages && mr->pages) {
2824 		rc = bnxt_qplib_free_fast_reg_page_list(&rdev->qplib_res,
2825 							&mr->qplib_frpl);
2826 		kfree(mr->pages);
2827 		mr->npages = 0;
2828 		mr->pages = NULL;
2829 	}
2830 	rc = bnxt_qplib_free_mrw(&rdev->qplib_res, &mr->qplib_mr);
2831 
2832 	if (!IS_ERR(mr->ib_umem) && mr->ib_umem)
2833 		ib_umem_release(mr->ib_umem);
2834 
2835 	kfree(mr);
2836 	atomic_dec(&rdev->mr_count);
2837 	return rc;
2838 }
2839 
2840 static int bnxt_re_set_page(struct ib_mr *ib_mr, u64 addr)
2841 {
2842 	struct bnxt_re_mr *mr = container_of(ib_mr, struct bnxt_re_mr, ib_mr);
2843 
2844 	if (unlikely(mr->npages == mr->qplib_frpl.max_pg_ptrs))
2845 		return -ENOMEM;
2846 
2847 	mr->pages[mr->npages++] = addr;
2848 	return 0;
2849 }
2850 
2851 int bnxt_re_map_mr_sg(struct ib_mr *ib_mr, struct scatterlist *sg, int sg_nents,
2852 		      unsigned int *sg_offset)
2853 {
2854 	struct bnxt_re_mr *mr = container_of(ib_mr, struct bnxt_re_mr, ib_mr);
2855 
2856 	mr->npages = 0;
2857 	return ib_sg_to_pages(ib_mr, sg, sg_nents, sg_offset, bnxt_re_set_page);
2858 }
2859 
2860 struct ib_mr *bnxt_re_alloc_mr(struct ib_pd *ib_pd, enum ib_mr_type type,
2861 			       u32 max_num_sg)
2862 {
2863 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
2864 	struct bnxt_re_dev *rdev = pd->rdev;
2865 	struct bnxt_re_mr *mr = NULL;
2866 	int rc;
2867 
2868 	if (type != IB_MR_TYPE_MEM_REG) {
2869 		dev_dbg(rdev_to_dev(rdev), "MR type 0x%x not supported", type);
2870 		return ERR_PTR(-EINVAL);
2871 	}
2872 	if (max_num_sg > MAX_PBL_LVL_1_PGS)
2873 		return ERR_PTR(-EINVAL);
2874 
2875 	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
2876 	if (!mr)
2877 		return ERR_PTR(-ENOMEM);
2878 
2879 	mr->rdev = rdev;
2880 	mr->qplib_mr.pd = &pd->qplib_pd;
2881 	mr->qplib_mr.flags = BNXT_QPLIB_FR_PMR;
2882 	mr->qplib_mr.type = CMDQ_ALLOCATE_MRW_MRW_FLAGS_PMR;
2883 
2884 	rc = bnxt_qplib_alloc_mrw(&rdev->qplib_res, &mr->qplib_mr);
2885 	if (rc)
2886 		goto fail;
2887 
2888 	mr->ib_mr.lkey = mr->qplib_mr.lkey;
2889 	mr->ib_mr.rkey = mr->ib_mr.lkey;
2890 
2891 	mr->pages = kcalloc(max_num_sg, sizeof(u64), GFP_KERNEL);
2892 	if (!mr->pages) {
2893 		rc = -ENOMEM;
2894 		goto fail;
2895 	}
2896 	rc = bnxt_qplib_alloc_fast_reg_page_list(&rdev->qplib_res,
2897 						 &mr->qplib_frpl, max_num_sg);
2898 	if (rc) {
2899 		dev_err(rdev_to_dev(rdev),
2900 			"Failed to allocate HW FR page list");
2901 		goto fail_mr;
2902 	}
2903 
2904 	atomic_inc(&rdev->mr_count);
2905 	return &mr->ib_mr;
2906 
2907 fail_mr:
2908 	bnxt_qplib_free_mrw(&rdev->qplib_res, &mr->qplib_mr);
2909 fail:
2910 	kfree(mr->pages);
2911 	kfree(mr);
2912 	return ERR_PTR(rc);
2913 }
2914 
2915 /* Fast Memory Regions */
2916 struct ib_fmr *bnxt_re_alloc_fmr(struct ib_pd *ib_pd, int mr_access_flags,
2917 				 struct ib_fmr_attr *fmr_attr)
2918 {
2919 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
2920 	struct bnxt_re_dev *rdev = pd->rdev;
2921 	struct bnxt_re_fmr *fmr;
2922 	int rc;
2923 
2924 	if (fmr_attr->max_pages > MAX_PBL_LVL_2_PGS ||
2925 	    fmr_attr->max_maps > rdev->dev_attr.max_map_per_fmr) {
2926 		dev_err(rdev_to_dev(rdev), "Allocate FMR exceeded Max limit");
2927 		return ERR_PTR(-ENOMEM);
2928 	}
2929 	fmr = kzalloc(sizeof(*fmr), GFP_KERNEL);
2930 	if (!fmr)
2931 		return ERR_PTR(-ENOMEM);
2932 
2933 	fmr->rdev = rdev;
2934 	fmr->qplib_fmr.pd = &pd->qplib_pd;
2935 	fmr->qplib_fmr.type = CMDQ_ALLOCATE_MRW_MRW_FLAGS_PMR;
2936 
2937 	rc = bnxt_qplib_alloc_mrw(&rdev->qplib_res, &fmr->qplib_fmr);
2938 	if (rc)
2939 		goto fail;
2940 
2941 	fmr->qplib_fmr.flags = __from_ib_access_flags(mr_access_flags);
2942 	fmr->ib_fmr.lkey = fmr->qplib_fmr.lkey;
2943 	fmr->ib_fmr.rkey = fmr->ib_fmr.lkey;
2944 
2945 	atomic_inc(&rdev->mr_count);
2946 	return &fmr->ib_fmr;
2947 fail:
2948 	kfree(fmr);
2949 	return ERR_PTR(rc);
2950 }
2951 
2952 int bnxt_re_map_phys_fmr(struct ib_fmr *ib_fmr, u64 *page_list, int list_len,
2953 			 u64 iova)
2954 {
2955 	struct bnxt_re_fmr *fmr = container_of(ib_fmr, struct bnxt_re_fmr,
2956 					     ib_fmr);
2957 	struct bnxt_re_dev *rdev = fmr->rdev;
2958 	int rc;
2959 
2960 	fmr->qplib_fmr.va = iova;
2961 	fmr->qplib_fmr.total_size = list_len * PAGE_SIZE;
2962 
2963 	rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &fmr->qplib_fmr, page_list,
2964 			       list_len, true);
2965 	if (rc)
2966 		dev_err(rdev_to_dev(rdev), "Failed to map FMR for lkey = 0x%x!",
2967 			fmr->ib_fmr.lkey);
2968 	return rc;
2969 }
2970 
2971 int bnxt_re_unmap_fmr(struct list_head *fmr_list)
2972 {
2973 	struct bnxt_re_dev *rdev;
2974 	struct bnxt_re_fmr *fmr;
2975 	struct ib_fmr *ib_fmr;
2976 	int rc = 0;
2977 
2978 	/* Validate each FMRs inside the fmr_list */
2979 	list_for_each_entry(ib_fmr, fmr_list, list) {
2980 		fmr = container_of(ib_fmr, struct bnxt_re_fmr, ib_fmr);
2981 		rdev = fmr->rdev;
2982 
2983 		if (rdev) {
2984 			rc = bnxt_qplib_dereg_mrw(&rdev->qplib_res,
2985 						  &fmr->qplib_fmr, true);
2986 			if (rc)
2987 				break;
2988 		}
2989 	}
2990 	return rc;
2991 }
2992 
2993 int bnxt_re_dealloc_fmr(struct ib_fmr *ib_fmr)
2994 {
2995 	struct bnxt_re_fmr *fmr = container_of(ib_fmr, struct bnxt_re_fmr,
2996 					       ib_fmr);
2997 	struct bnxt_re_dev *rdev = fmr->rdev;
2998 	int rc;
2999 
3000 	rc = bnxt_qplib_free_mrw(&rdev->qplib_res, &fmr->qplib_fmr);
3001 	if (rc)
3002 		dev_err(rdev_to_dev(rdev), "Failed to free FMR");
3003 
3004 	kfree(fmr);
3005 	atomic_dec(&rdev->mr_count);
3006 	return rc;
3007 }
3008 
3009 /* uverbs */
3010 struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length,
3011 				  u64 virt_addr, int mr_access_flags,
3012 				  struct ib_udata *udata)
3013 {
3014 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
3015 	struct bnxt_re_dev *rdev = pd->rdev;
3016 	struct bnxt_re_mr *mr;
3017 	struct ib_umem *umem;
3018 	u64 *pbl_tbl, *pbl_tbl_orig;
3019 	int i, umem_pgs, pages, page_shift, rc;
3020 	struct scatterlist *sg;
3021 	int entry;
3022 
3023 	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
3024 	if (!mr)
3025 		return ERR_PTR(-ENOMEM);
3026 
3027 	mr->rdev = rdev;
3028 	mr->qplib_mr.pd = &pd->qplib_pd;
3029 	mr->qplib_mr.flags = __from_ib_access_flags(mr_access_flags);
3030 	mr->qplib_mr.type = CMDQ_ALLOCATE_MRW_MRW_FLAGS_MR;
3031 
3032 	umem = ib_umem_get(ib_pd->uobject->context, start, length,
3033 			   mr_access_flags, 0);
3034 	if (IS_ERR(umem)) {
3035 		dev_err(rdev_to_dev(rdev), "Failed to get umem");
3036 		rc = -EFAULT;
3037 		goto free_mr;
3038 	}
3039 	mr->ib_umem = umem;
3040 
3041 	rc = bnxt_qplib_alloc_mrw(&rdev->qplib_res, &mr->qplib_mr);
3042 	if (rc) {
3043 		dev_err(rdev_to_dev(rdev), "Failed to allocate MR");
3044 		goto release_umem;
3045 	}
3046 	/* The fixed portion of the rkey is the same as the lkey */
3047 	mr->ib_mr.rkey = mr->qplib_mr.rkey;
3048 
3049 	mr->qplib_mr.va = virt_addr;
3050 	umem_pgs = ib_umem_page_count(umem);
3051 	if (!umem_pgs) {
3052 		dev_err(rdev_to_dev(rdev), "umem is invalid!");
3053 		rc = -EINVAL;
3054 		goto free_mrw;
3055 	}
3056 	mr->qplib_mr.total_size = length;
3057 
3058 	pbl_tbl = kcalloc(umem_pgs, sizeof(u64 *), GFP_KERNEL);
3059 	if (!pbl_tbl) {
3060 		rc = -EINVAL;
3061 		goto free_mrw;
3062 	}
3063 	pbl_tbl_orig = pbl_tbl;
3064 
3065 	page_shift = ilog2(umem->page_size);
3066 	if (umem->hugetlb) {
3067 		dev_err(rdev_to_dev(rdev), "umem hugetlb not supported!");
3068 		rc = -EFAULT;
3069 		goto fail;
3070 	}
3071 	if (umem->page_size != PAGE_SIZE) {
3072 		dev_err(rdev_to_dev(rdev), "umem page size unsupported!");
3073 		rc = -EFAULT;
3074 		goto fail;
3075 	}
3076 	/* Map umem buf ptrs to the PBL */
3077 	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
3078 		pages = sg_dma_len(sg) >> page_shift;
3079 		for (i = 0; i < pages; i++, pbl_tbl++)
3080 			*pbl_tbl = sg_dma_address(sg) + (i << page_shift);
3081 	}
3082 	rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, pbl_tbl_orig,
3083 			       umem_pgs, false);
3084 	if (rc) {
3085 		dev_err(rdev_to_dev(rdev), "Failed to register user MR");
3086 		goto fail;
3087 	}
3088 
3089 	kfree(pbl_tbl_orig);
3090 
3091 	mr->ib_mr.lkey = mr->qplib_mr.lkey;
3092 	mr->ib_mr.rkey = mr->qplib_mr.lkey;
3093 	atomic_inc(&rdev->mr_count);
3094 
3095 	return &mr->ib_mr;
3096 fail:
3097 	kfree(pbl_tbl_orig);
3098 free_mrw:
3099 	bnxt_qplib_free_mrw(&rdev->qplib_res, &mr->qplib_mr);
3100 release_umem:
3101 	ib_umem_release(umem);
3102 free_mr:
3103 	kfree(mr);
3104 	return ERR_PTR(rc);
3105 }
3106 
3107 struct ib_ucontext *bnxt_re_alloc_ucontext(struct ib_device *ibdev,
3108 					   struct ib_udata *udata)
3109 {
3110 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
3111 	struct bnxt_re_uctx_resp resp;
3112 	struct bnxt_re_ucontext *uctx;
3113 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
3114 	int rc;
3115 
3116 	dev_dbg(rdev_to_dev(rdev), "ABI version requested %d",
3117 		ibdev->uverbs_abi_ver);
3118 
3119 	if (ibdev->uverbs_abi_ver != BNXT_RE_ABI_VERSION) {
3120 		dev_dbg(rdev_to_dev(rdev), " is different from the device %d ",
3121 			BNXT_RE_ABI_VERSION);
3122 		return ERR_PTR(-EPERM);
3123 	}
3124 
3125 	uctx = kzalloc(sizeof(*uctx), GFP_KERNEL);
3126 	if (!uctx)
3127 		return ERR_PTR(-ENOMEM);
3128 
3129 	uctx->rdev = rdev;
3130 
3131 	uctx->shpg = (void *)__get_free_page(GFP_KERNEL);
3132 	if (!uctx->shpg) {
3133 		rc = -ENOMEM;
3134 		goto fail;
3135 	}
3136 	spin_lock_init(&uctx->sh_lock);
3137 
3138 	resp.dev_id = rdev->en_dev->pdev->devfn; /*Temp, Use idr_alloc instead*/
3139 	resp.max_qp = rdev->qplib_ctx.qpc_count;
3140 	resp.pg_size = PAGE_SIZE;
3141 	resp.cqe_sz = sizeof(struct cq_base);
3142 	resp.max_cqd = dev_attr->max_cq_wqes;
3143 	resp.rsvd    = 0;
3144 
3145 	rc = ib_copy_to_udata(udata, &resp, sizeof(resp));
3146 	if (rc) {
3147 		dev_err(rdev_to_dev(rdev), "Failed to copy user context");
3148 		rc = -EFAULT;
3149 		goto cfail;
3150 	}
3151 
3152 	return &uctx->ib_uctx;
3153 cfail:
3154 	free_page((unsigned long)uctx->shpg);
3155 	uctx->shpg = NULL;
3156 fail:
3157 	kfree(uctx);
3158 	return ERR_PTR(rc);
3159 }
3160 
3161 int bnxt_re_dealloc_ucontext(struct ib_ucontext *ib_uctx)
3162 {
3163 	struct bnxt_re_ucontext *uctx = container_of(ib_uctx,
3164 						   struct bnxt_re_ucontext,
3165 						   ib_uctx);
3166 	if (uctx->shpg)
3167 		free_page((unsigned long)uctx->shpg);
3168 	kfree(uctx);
3169 	return 0;
3170 }
3171 
3172 /* Helper function to mmap the virtual memory from user app */
3173 int bnxt_re_mmap(struct ib_ucontext *ib_uctx, struct vm_area_struct *vma)
3174 {
3175 	struct bnxt_re_ucontext *uctx = container_of(ib_uctx,
3176 						   struct bnxt_re_ucontext,
3177 						   ib_uctx);
3178 	struct bnxt_re_dev *rdev = uctx->rdev;
3179 	u64 pfn;
3180 
3181 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
3182 		return -EINVAL;
3183 
3184 	if (vma->vm_pgoff) {
3185 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
3186 		if (io_remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
3187 				       PAGE_SIZE, vma->vm_page_prot)) {
3188 			dev_err(rdev_to_dev(rdev), "Failed to map DPI");
3189 			return -EAGAIN;
3190 		}
3191 	} else {
3192 		pfn = virt_to_phys(uctx->shpg) >> PAGE_SHIFT;
3193 		if (remap_pfn_range(vma, vma->vm_start,
3194 				    pfn, PAGE_SIZE, vma->vm_page_prot)) {
3195 			dev_err(rdev_to_dev(rdev),
3196 				"Failed to map shared page");
3197 			return -EAGAIN;
3198 		}
3199 	}
3200 
3201 	return 0;
3202 }
3203