xref: /linux/drivers/infiniband/hw/bnxt_re/ib_verbs.c (revision b930d0bac9c671c053dd66229010ca9298e84aab)
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 #include <net/addrconf.h>
45 
46 #include <rdma/ib_verbs.h>
47 #include <rdma/ib_user_verbs.h>
48 #include <rdma/ib_umem.h>
49 #include <rdma/ib_addr.h>
50 #include <rdma/ib_mad.h>
51 #include <rdma/ib_cache.h>
52 #include <rdma/uverbs_ioctl.h>
53 #include <linux/hashtable.h>
54 
55 #include "bnxt_ulp.h"
56 
57 #include "roce_hsi.h"
58 #include "qplib_res.h"
59 #include "qplib_sp.h"
60 #include "qplib_fp.h"
61 #include "qplib_rcfw.h"
62 
63 #include "bnxt_re.h"
64 #include "ib_verbs.h"
65 
66 #include <rdma/uverbs_types.h>
67 #include <rdma/uverbs_std_types.h>
68 
69 #include <rdma/ib_user_ioctl_cmds.h>
70 
71 #define UVERBS_MODULE_NAME bnxt_re
72 #include <rdma/uverbs_named_ioctl.h>
73 
74 #include <rdma/bnxt_re-abi.h>
75 
76 static int __from_ib_access_flags(int iflags)
77 {
78 	int qflags = 0;
79 
80 	if (iflags & IB_ACCESS_LOCAL_WRITE)
81 		qflags |= BNXT_QPLIB_ACCESS_LOCAL_WRITE;
82 	if (iflags & IB_ACCESS_REMOTE_READ)
83 		qflags |= BNXT_QPLIB_ACCESS_REMOTE_READ;
84 	if (iflags & IB_ACCESS_REMOTE_WRITE)
85 		qflags |= BNXT_QPLIB_ACCESS_REMOTE_WRITE;
86 	if (iflags & IB_ACCESS_REMOTE_ATOMIC)
87 		qflags |= BNXT_QPLIB_ACCESS_REMOTE_ATOMIC;
88 	if (iflags & IB_ACCESS_MW_BIND)
89 		qflags |= BNXT_QPLIB_ACCESS_MW_BIND;
90 	if (iflags & IB_ZERO_BASED)
91 		qflags |= BNXT_QPLIB_ACCESS_ZERO_BASED;
92 	if (iflags & IB_ACCESS_ON_DEMAND)
93 		qflags |= BNXT_QPLIB_ACCESS_ON_DEMAND;
94 	return qflags;
95 };
96 
97 static enum ib_access_flags __to_ib_access_flags(int qflags)
98 {
99 	enum ib_access_flags iflags = 0;
100 
101 	if (qflags & BNXT_QPLIB_ACCESS_LOCAL_WRITE)
102 		iflags |= IB_ACCESS_LOCAL_WRITE;
103 	if (qflags & BNXT_QPLIB_ACCESS_REMOTE_WRITE)
104 		iflags |= IB_ACCESS_REMOTE_WRITE;
105 	if (qflags & BNXT_QPLIB_ACCESS_REMOTE_READ)
106 		iflags |= IB_ACCESS_REMOTE_READ;
107 	if (qflags & BNXT_QPLIB_ACCESS_REMOTE_ATOMIC)
108 		iflags |= IB_ACCESS_REMOTE_ATOMIC;
109 	if (qflags & BNXT_QPLIB_ACCESS_MW_BIND)
110 		iflags |= IB_ACCESS_MW_BIND;
111 	if (qflags & BNXT_QPLIB_ACCESS_ZERO_BASED)
112 		iflags |= IB_ZERO_BASED;
113 	if (qflags & BNXT_QPLIB_ACCESS_ON_DEMAND)
114 		iflags |= IB_ACCESS_ON_DEMAND;
115 	return iflags;
116 };
117 
118 static int bnxt_re_build_sgl(struct ib_sge *ib_sg_list,
119 			     struct bnxt_qplib_sge *sg_list, int num)
120 {
121 	int i, total = 0;
122 
123 	for (i = 0; i < num; i++) {
124 		sg_list[i].addr = ib_sg_list[i].addr;
125 		sg_list[i].lkey = ib_sg_list[i].lkey;
126 		sg_list[i].size = ib_sg_list[i].length;
127 		total += sg_list[i].size;
128 	}
129 	return total;
130 }
131 
132 /* Device */
133 int bnxt_re_query_device(struct ib_device *ibdev,
134 			 struct ib_device_attr *ib_attr,
135 			 struct ib_udata *udata)
136 {
137 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
138 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
139 
140 	memset(ib_attr, 0, sizeof(*ib_attr));
141 	memcpy(&ib_attr->fw_ver, dev_attr->fw_ver,
142 	       min(sizeof(dev_attr->fw_ver),
143 		   sizeof(ib_attr->fw_ver)));
144 	addrconf_addr_eui48((u8 *)&ib_attr->sys_image_guid,
145 			    rdev->netdev->dev_addr);
146 	ib_attr->max_mr_size = BNXT_RE_MAX_MR_SIZE;
147 	ib_attr->page_size_cap = BNXT_RE_PAGE_SIZE_SUPPORTED;
148 
149 	ib_attr->vendor_id = rdev->en_dev->pdev->vendor;
150 	ib_attr->vendor_part_id = rdev->en_dev->pdev->device;
151 	ib_attr->hw_ver = rdev->en_dev->pdev->subsystem_device;
152 	ib_attr->max_qp = dev_attr->max_qp;
153 	ib_attr->max_qp_wr = dev_attr->max_qp_wqes;
154 	ib_attr->device_cap_flags =
155 				    IB_DEVICE_CURR_QP_STATE_MOD
156 				    | IB_DEVICE_RC_RNR_NAK_GEN
157 				    | IB_DEVICE_SHUTDOWN_PORT
158 				    | IB_DEVICE_SYS_IMAGE_GUID
159 				    | IB_DEVICE_RESIZE_MAX_WR
160 				    | IB_DEVICE_PORT_ACTIVE_EVENT
161 				    | IB_DEVICE_N_NOTIFY_CQ
162 				    | IB_DEVICE_MEM_WINDOW
163 				    | IB_DEVICE_MEM_WINDOW_TYPE_2B
164 				    | IB_DEVICE_MEM_MGT_EXTENSIONS;
165 	ib_attr->kernel_cap_flags = IBK_LOCAL_DMA_LKEY;
166 	ib_attr->max_send_sge = dev_attr->max_qp_sges;
167 	ib_attr->max_recv_sge = dev_attr->max_qp_sges;
168 	ib_attr->max_sge_rd = dev_attr->max_qp_sges;
169 	ib_attr->max_cq = dev_attr->max_cq;
170 	ib_attr->max_cqe = dev_attr->max_cq_wqes;
171 	ib_attr->max_mr = dev_attr->max_mr;
172 	ib_attr->max_pd = dev_attr->max_pd;
173 	ib_attr->max_qp_rd_atom = dev_attr->max_qp_rd_atom;
174 	ib_attr->max_qp_init_rd_atom = dev_attr->max_qp_init_rd_atom;
175 	ib_attr->atomic_cap = IB_ATOMIC_NONE;
176 	ib_attr->masked_atomic_cap = IB_ATOMIC_NONE;
177 	if (dev_attr->is_atomic) {
178 		ib_attr->atomic_cap = IB_ATOMIC_GLOB;
179 		ib_attr->masked_atomic_cap = IB_ATOMIC_GLOB;
180 	}
181 
182 	ib_attr->max_ee_rd_atom = 0;
183 	ib_attr->max_res_rd_atom = 0;
184 	ib_attr->max_ee_init_rd_atom = 0;
185 	ib_attr->max_ee = 0;
186 	ib_attr->max_rdd = 0;
187 	ib_attr->max_mw = dev_attr->max_mw;
188 	ib_attr->max_raw_ipv6_qp = 0;
189 	ib_attr->max_raw_ethy_qp = dev_attr->max_raw_ethy_qp;
190 	ib_attr->max_mcast_grp = 0;
191 	ib_attr->max_mcast_qp_attach = 0;
192 	ib_attr->max_total_mcast_qp_attach = 0;
193 	ib_attr->max_ah = dev_attr->max_ah;
194 
195 	ib_attr->max_srq = dev_attr->max_srq;
196 	ib_attr->max_srq_wr = dev_attr->max_srq_wqes;
197 	ib_attr->max_srq_sge = dev_attr->max_srq_sges;
198 
199 	ib_attr->max_fast_reg_page_list_len = MAX_PBL_LVL_1_PGS;
200 
201 	ib_attr->max_pkeys = 1;
202 	ib_attr->local_ca_ack_delay = BNXT_RE_DEFAULT_ACK_DELAY;
203 	return 0;
204 }
205 
206 /* Port */
207 int bnxt_re_query_port(struct ib_device *ibdev, u32 port_num,
208 		       struct ib_port_attr *port_attr)
209 {
210 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
211 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
212 	int rc;
213 
214 	memset(port_attr, 0, sizeof(*port_attr));
215 
216 	if (netif_running(rdev->netdev) && netif_carrier_ok(rdev->netdev)) {
217 		port_attr->state = IB_PORT_ACTIVE;
218 		port_attr->phys_state = IB_PORT_PHYS_STATE_LINK_UP;
219 	} else {
220 		port_attr->state = IB_PORT_DOWN;
221 		port_attr->phys_state = IB_PORT_PHYS_STATE_DISABLED;
222 	}
223 	port_attr->max_mtu = IB_MTU_4096;
224 	port_attr->active_mtu = iboe_get_mtu(rdev->netdev->mtu);
225 	port_attr->gid_tbl_len = dev_attr->max_sgid;
226 	port_attr->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_REINIT_SUP |
227 				    IB_PORT_DEVICE_MGMT_SUP |
228 				    IB_PORT_VENDOR_CLASS_SUP;
229 	port_attr->ip_gids = true;
230 
231 	port_attr->max_msg_sz = (u32)BNXT_RE_MAX_MR_SIZE_LOW;
232 	port_attr->bad_pkey_cntr = 0;
233 	port_attr->qkey_viol_cntr = 0;
234 	port_attr->pkey_tbl_len = dev_attr->max_pkey;
235 	port_attr->lid = 0;
236 	port_attr->sm_lid = 0;
237 	port_attr->lmc = 0;
238 	port_attr->max_vl_num = 4;
239 	port_attr->sm_sl = 0;
240 	port_attr->subnet_timeout = 0;
241 	port_attr->init_type_reply = 0;
242 	rc = ib_get_eth_speed(&rdev->ibdev, port_num, &port_attr->active_speed,
243 			      &port_attr->active_width);
244 
245 	return rc;
246 }
247 
248 int bnxt_re_get_port_immutable(struct ib_device *ibdev, u32 port_num,
249 			       struct ib_port_immutable *immutable)
250 {
251 	struct ib_port_attr port_attr;
252 
253 	if (bnxt_re_query_port(ibdev, port_num, &port_attr))
254 		return -EINVAL;
255 
256 	immutable->pkey_tbl_len = port_attr.pkey_tbl_len;
257 	immutable->gid_tbl_len = port_attr.gid_tbl_len;
258 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
259 	immutable->core_cap_flags |= RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
260 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
261 	return 0;
262 }
263 
264 void bnxt_re_query_fw_str(struct ib_device *ibdev, char *str)
265 {
266 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
267 
268 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d.%d",
269 		 rdev->dev_attr.fw_ver[0], rdev->dev_attr.fw_ver[1],
270 		 rdev->dev_attr.fw_ver[2], rdev->dev_attr.fw_ver[3]);
271 }
272 
273 int bnxt_re_query_pkey(struct ib_device *ibdev, u32 port_num,
274 		       u16 index, u16 *pkey)
275 {
276 	if (index > 0)
277 		return -EINVAL;
278 
279 	*pkey = IB_DEFAULT_PKEY_FULL;
280 
281 	return 0;
282 }
283 
284 int bnxt_re_query_gid(struct ib_device *ibdev, u32 port_num,
285 		      int index, union ib_gid *gid)
286 {
287 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
288 	int rc;
289 
290 	/* Ignore port_num */
291 	memset(gid, 0, sizeof(*gid));
292 	rc = bnxt_qplib_get_sgid(&rdev->qplib_res,
293 				 &rdev->qplib_res.sgid_tbl, index,
294 				 (struct bnxt_qplib_gid *)gid);
295 	return rc;
296 }
297 
298 int bnxt_re_del_gid(const struct ib_gid_attr *attr, void **context)
299 {
300 	int rc = 0;
301 	struct bnxt_re_gid_ctx *ctx, **ctx_tbl;
302 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(attr->device, ibdev);
303 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
304 	struct bnxt_qplib_gid *gid_to_del;
305 	u16 vlan_id = 0xFFFF;
306 
307 	/* Delete the entry from the hardware */
308 	ctx = *context;
309 	if (!ctx)
310 		return -EINVAL;
311 
312 	if (sgid_tbl && sgid_tbl->active) {
313 		if (ctx->idx >= sgid_tbl->max)
314 			return -EINVAL;
315 		gid_to_del = &sgid_tbl->tbl[ctx->idx].gid;
316 		vlan_id = sgid_tbl->tbl[ctx->idx].vlan_id;
317 		/* DEL_GID is called in WQ context(netdevice_event_work_handler)
318 		 * or via the ib_unregister_device path. In the former case QP1
319 		 * may not be destroyed yet, in which case just return as FW
320 		 * needs that entry to be present and will fail it's deletion.
321 		 * We could get invoked again after QP1 is destroyed OR get an
322 		 * ADD_GID call with a different GID value for the same index
323 		 * where we issue MODIFY_GID cmd to update the GID entry -- TBD
324 		 */
325 		if (ctx->idx == 0 &&
326 		    rdma_link_local_addr((struct in6_addr *)gid_to_del) &&
327 		    ctx->refcnt == 1 && rdev->gsi_ctx.gsi_sqp) {
328 			ibdev_dbg(&rdev->ibdev,
329 				  "Trying to delete GID0 while QP1 is alive\n");
330 			return -EFAULT;
331 		}
332 		ctx->refcnt--;
333 		if (!ctx->refcnt) {
334 			rc = bnxt_qplib_del_sgid(sgid_tbl, gid_to_del,
335 						 vlan_id,  true);
336 			if (rc) {
337 				ibdev_err(&rdev->ibdev,
338 					  "Failed to remove GID: %#x", rc);
339 			} else {
340 				ctx_tbl = sgid_tbl->ctx;
341 				ctx_tbl[ctx->idx] = NULL;
342 				kfree(ctx);
343 			}
344 		}
345 	} else {
346 		return -EINVAL;
347 	}
348 	return rc;
349 }
350 
351 int bnxt_re_add_gid(const struct ib_gid_attr *attr, void **context)
352 {
353 	int rc;
354 	u32 tbl_idx = 0;
355 	u16 vlan_id = 0xFFFF;
356 	struct bnxt_re_gid_ctx *ctx, **ctx_tbl;
357 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(attr->device, ibdev);
358 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
359 
360 	rc = rdma_read_gid_l2_fields(attr, &vlan_id, NULL);
361 	if (rc)
362 		return rc;
363 
364 	rc = bnxt_qplib_add_sgid(sgid_tbl, (struct bnxt_qplib_gid *)&attr->gid,
365 				 rdev->qplib_res.netdev->dev_addr,
366 				 vlan_id, true, &tbl_idx);
367 	if (rc == -EALREADY) {
368 		ctx_tbl = sgid_tbl->ctx;
369 		ctx_tbl[tbl_idx]->refcnt++;
370 		*context = ctx_tbl[tbl_idx];
371 		return 0;
372 	}
373 
374 	if (rc < 0) {
375 		ibdev_err(&rdev->ibdev, "Failed to add GID: %#x", rc);
376 		return rc;
377 	}
378 
379 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
380 	if (!ctx)
381 		return -ENOMEM;
382 	ctx_tbl = sgid_tbl->ctx;
383 	ctx->idx = tbl_idx;
384 	ctx->refcnt = 1;
385 	ctx_tbl[tbl_idx] = ctx;
386 	*context = ctx;
387 
388 	return rc;
389 }
390 
391 enum rdma_link_layer bnxt_re_get_link_layer(struct ib_device *ibdev,
392 					    u32 port_num)
393 {
394 	return IB_LINK_LAYER_ETHERNET;
395 }
396 
397 #define	BNXT_RE_FENCE_PBL_SIZE	DIV_ROUND_UP(BNXT_RE_FENCE_BYTES, PAGE_SIZE)
398 
399 static void bnxt_re_create_fence_wqe(struct bnxt_re_pd *pd)
400 {
401 	struct bnxt_re_fence_data *fence = &pd->fence;
402 	struct ib_mr *ib_mr = &fence->mr->ib_mr;
403 	struct bnxt_qplib_swqe *wqe = &fence->bind_wqe;
404 	struct bnxt_re_dev *rdev = pd->rdev;
405 
406 	if (bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx))
407 		return;
408 
409 	memset(wqe, 0, sizeof(*wqe));
410 	wqe->type = BNXT_QPLIB_SWQE_TYPE_BIND_MW;
411 	wqe->wr_id = BNXT_QPLIB_FENCE_WRID;
412 	wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP;
413 	wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE;
414 	wqe->bind.zero_based = false;
415 	wqe->bind.parent_l_key = ib_mr->lkey;
416 	wqe->bind.va = (u64)(unsigned long)fence->va;
417 	wqe->bind.length = fence->size;
418 	wqe->bind.access_cntl = __from_ib_access_flags(IB_ACCESS_REMOTE_READ);
419 	wqe->bind.mw_type = SQ_BIND_MW_TYPE_TYPE1;
420 
421 	/* Save the initial rkey in fence structure for now;
422 	 * wqe->bind.r_key will be set at (re)bind time.
423 	 */
424 	fence->bind_rkey = ib_inc_rkey(fence->mw->rkey);
425 }
426 
427 static int bnxt_re_bind_fence_mw(struct bnxt_qplib_qp *qplib_qp)
428 {
429 	struct bnxt_re_qp *qp = container_of(qplib_qp, struct bnxt_re_qp,
430 					     qplib_qp);
431 	struct ib_pd *ib_pd = qp->ib_qp.pd;
432 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
433 	struct bnxt_re_fence_data *fence = &pd->fence;
434 	struct bnxt_qplib_swqe *fence_wqe = &fence->bind_wqe;
435 	struct bnxt_qplib_swqe wqe;
436 	int rc;
437 
438 	memcpy(&wqe, fence_wqe, sizeof(wqe));
439 	wqe.bind.r_key = fence->bind_rkey;
440 	fence->bind_rkey = ib_inc_rkey(fence->bind_rkey);
441 
442 	ibdev_dbg(&qp->rdev->ibdev,
443 		  "Posting bind fence-WQE: rkey: %#x QP: %d PD: %p\n",
444 		wqe.bind.r_key, qp->qplib_qp.id, pd);
445 	rc = bnxt_qplib_post_send(&qp->qplib_qp, &wqe);
446 	if (rc) {
447 		ibdev_err(&qp->rdev->ibdev, "Failed to bind fence-WQE\n");
448 		return rc;
449 	}
450 	bnxt_qplib_post_send_db(&qp->qplib_qp);
451 
452 	return rc;
453 }
454 
455 static void bnxt_re_destroy_fence_mr(struct bnxt_re_pd *pd)
456 {
457 	struct bnxt_re_fence_data *fence = &pd->fence;
458 	struct bnxt_re_dev *rdev = pd->rdev;
459 	struct device *dev = &rdev->en_dev->pdev->dev;
460 	struct bnxt_re_mr *mr = fence->mr;
461 
462 	if (bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx))
463 		return;
464 
465 	if (fence->mw) {
466 		bnxt_re_dealloc_mw(fence->mw);
467 		fence->mw = NULL;
468 	}
469 	if (mr) {
470 		if (mr->ib_mr.rkey)
471 			bnxt_qplib_dereg_mrw(&rdev->qplib_res, &mr->qplib_mr,
472 					     true);
473 		if (mr->ib_mr.lkey)
474 			bnxt_qplib_free_mrw(&rdev->qplib_res, &mr->qplib_mr);
475 		kfree(mr);
476 		fence->mr = NULL;
477 	}
478 	if (fence->dma_addr) {
479 		dma_unmap_single(dev, fence->dma_addr, BNXT_RE_FENCE_BYTES,
480 				 DMA_BIDIRECTIONAL);
481 		fence->dma_addr = 0;
482 	}
483 }
484 
485 static int bnxt_re_create_fence_mr(struct bnxt_re_pd *pd)
486 {
487 	int mr_access_flags = IB_ACCESS_LOCAL_WRITE | IB_ACCESS_MW_BIND;
488 	struct bnxt_re_fence_data *fence = &pd->fence;
489 	struct bnxt_re_dev *rdev = pd->rdev;
490 	struct device *dev = &rdev->en_dev->pdev->dev;
491 	struct bnxt_re_mr *mr = NULL;
492 	dma_addr_t dma_addr = 0;
493 	struct ib_mw *mw;
494 	int rc;
495 
496 	if (bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx))
497 		return 0;
498 
499 	dma_addr = dma_map_single(dev, fence->va, BNXT_RE_FENCE_BYTES,
500 				  DMA_BIDIRECTIONAL);
501 	rc = dma_mapping_error(dev, dma_addr);
502 	if (rc) {
503 		ibdev_err(&rdev->ibdev, "Failed to dma-map fence-MR-mem\n");
504 		rc = -EIO;
505 		fence->dma_addr = 0;
506 		goto fail;
507 	}
508 	fence->dma_addr = dma_addr;
509 
510 	/* Allocate a MR */
511 	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
512 	if (!mr) {
513 		rc = -ENOMEM;
514 		goto fail;
515 	}
516 	fence->mr = mr;
517 	mr->rdev = rdev;
518 	mr->qplib_mr.pd = &pd->qplib_pd;
519 	mr->qplib_mr.type = CMDQ_ALLOCATE_MRW_MRW_FLAGS_PMR;
520 	mr->qplib_mr.flags = __from_ib_access_flags(mr_access_flags);
521 	rc = bnxt_qplib_alloc_mrw(&rdev->qplib_res, &mr->qplib_mr);
522 	if (rc) {
523 		ibdev_err(&rdev->ibdev, "Failed to alloc fence-HW-MR\n");
524 		goto fail;
525 	}
526 
527 	/* Register MR */
528 	mr->ib_mr.lkey = mr->qplib_mr.lkey;
529 	mr->qplib_mr.va = (u64)(unsigned long)fence->va;
530 	mr->qplib_mr.total_size = BNXT_RE_FENCE_BYTES;
531 	rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, NULL,
532 			       BNXT_RE_FENCE_PBL_SIZE, PAGE_SIZE);
533 	if (rc) {
534 		ibdev_err(&rdev->ibdev, "Failed to register fence-MR\n");
535 		goto fail;
536 	}
537 	mr->ib_mr.rkey = mr->qplib_mr.rkey;
538 
539 	/* Create a fence MW only for kernel consumers */
540 	mw = bnxt_re_alloc_mw(&pd->ib_pd, IB_MW_TYPE_1, NULL);
541 	if (IS_ERR(mw)) {
542 		ibdev_err(&rdev->ibdev,
543 			  "Failed to create fence-MW for PD: %p\n", pd);
544 		rc = PTR_ERR(mw);
545 		goto fail;
546 	}
547 	fence->mw = mw;
548 
549 	bnxt_re_create_fence_wqe(pd);
550 	return 0;
551 
552 fail:
553 	bnxt_re_destroy_fence_mr(pd);
554 	return rc;
555 }
556 
557 static struct bnxt_re_user_mmap_entry*
558 bnxt_re_mmap_entry_insert(struct bnxt_re_ucontext *uctx, u64 mem_offset,
559 			  enum bnxt_re_mmap_flag mmap_flag, u64 *offset)
560 {
561 	struct bnxt_re_user_mmap_entry *entry;
562 	int ret;
563 
564 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
565 	if (!entry)
566 		return NULL;
567 
568 	entry->mem_offset = mem_offset;
569 	entry->mmap_flag = mmap_flag;
570 	entry->uctx = uctx;
571 
572 	switch (mmap_flag) {
573 	case BNXT_RE_MMAP_SH_PAGE:
574 		ret = rdma_user_mmap_entry_insert_exact(&uctx->ib_uctx,
575 							&entry->rdma_entry, PAGE_SIZE, 0);
576 		break;
577 	case BNXT_RE_MMAP_UC_DB:
578 	case BNXT_RE_MMAP_WC_DB:
579 	case BNXT_RE_MMAP_DBR_BAR:
580 	case BNXT_RE_MMAP_DBR_PAGE:
581 	case BNXT_RE_MMAP_TOGGLE_PAGE:
582 		ret = rdma_user_mmap_entry_insert(&uctx->ib_uctx,
583 						  &entry->rdma_entry, PAGE_SIZE);
584 		break;
585 	default:
586 		ret = -EINVAL;
587 		break;
588 	}
589 
590 	if (ret) {
591 		kfree(entry);
592 		return NULL;
593 	}
594 	if (offset)
595 		*offset = rdma_user_mmap_get_offset(&entry->rdma_entry);
596 
597 	return entry;
598 }
599 
600 /* Protection Domains */
601 int bnxt_re_dealloc_pd(struct ib_pd *ib_pd, struct ib_udata *udata)
602 {
603 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
604 	struct bnxt_re_dev *rdev = pd->rdev;
605 
606 	if (udata) {
607 		rdma_user_mmap_entry_remove(pd->pd_db_mmap);
608 		pd->pd_db_mmap = NULL;
609 	}
610 
611 	bnxt_re_destroy_fence_mr(pd);
612 
613 	if (pd->qplib_pd.id) {
614 		if (!bnxt_qplib_dealloc_pd(&rdev->qplib_res,
615 					   &rdev->qplib_res.pd_tbl,
616 					   &pd->qplib_pd))
617 			atomic_dec(&rdev->stats.res.pd_count);
618 	}
619 	return 0;
620 }
621 
622 int bnxt_re_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
623 {
624 	struct ib_device *ibdev = ibpd->device;
625 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
626 	struct bnxt_re_ucontext *ucntx = rdma_udata_to_drv_context(
627 		udata, struct bnxt_re_ucontext, ib_uctx);
628 	struct bnxt_re_pd *pd = container_of(ibpd, struct bnxt_re_pd, ib_pd);
629 	struct bnxt_re_user_mmap_entry *entry = NULL;
630 	u32 active_pds;
631 	int rc = 0;
632 
633 	pd->rdev = rdev;
634 	if (bnxt_qplib_alloc_pd(&rdev->qplib_res, &pd->qplib_pd)) {
635 		ibdev_err(&rdev->ibdev, "Failed to allocate HW PD");
636 		rc = -ENOMEM;
637 		goto fail;
638 	}
639 
640 	if (udata) {
641 		struct bnxt_re_pd_resp resp = {};
642 
643 		if (!ucntx->dpi.dbr) {
644 			/* Allocate DPI in alloc_pd to avoid failing of
645 			 * ibv_devinfo and family of application when DPIs
646 			 * are depleted.
647 			 */
648 			if (bnxt_qplib_alloc_dpi(&rdev->qplib_res,
649 						 &ucntx->dpi, ucntx, BNXT_QPLIB_DPI_TYPE_UC)) {
650 				rc = -ENOMEM;
651 				goto dbfail;
652 			}
653 		}
654 
655 		resp.pdid = pd->qplib_pd.id;
656 		/* Still allow mapping this DBR to the new user PD. */
657 		resp.dpi = ucntx->dpi.dpi;
658 
659 		entry = bnxt_re_mmap_entry_insert(ucntx, (u64)ucntx->dpi.umdbr,
660 						  BNXT_RE_MMAP_UC_DB, &resp.dbr);
661 
662 		if (!entry) {
663 			rc = -ENOMEM;
664 			goto dbfail;
665 		}
666 
667 		pd->pd_db_mmap = &entry->rdma_entry;
668 
669 		rc = ib_copy_to_udata(udata, &resp, min(sizeof(resp), udata->outlen));
670 		if (rc) {
671 			rdma_user_mmap_entry_remove(pd->pd_db_mmap);
672 			rc = -EFAULT;
673 			goto dbfail;
674 		}
675 	}
676 
677 	if (!udata)
678 		if (bnxt_re_create_fence_mr(pd))
679 			ibdev_warn(&rdev->ibdev,
680 				   "Failed to create Fence-MR\n");
681 	active_pds = atomic_inc_return(&rdev->stats.res.pd_count);
682 	if (active_pds > rdev->stats.res.pd_watermark)
683 		rdev->stats.res.pd_watermark = active_pds;
684 
685 	return 0;
686 dbfail:
687 	bnxt_qplib_dealloc_pd(&rdev->qplib_res, &rdev->qplib_res.pd_tbl,
688 			      &pd->qplib_pd);
689 fail:
690 	return rc;
691 }
692 
693 /* Address Handles */
694 int bnxt_re_destroy_ah(struct ib_ah *ib_ah, u32 flags)
695 {
696 	struct bnxt_re_ah *ah = container_of(ib_ah, struct bnxt_re_ah, ib_ah);
697 	struct bnxt_re_dev *rdev = ah->rdev;
698 	bool block = true;
699 	int rc;
700 
701 	block = !(flags & RDMA_DESTROY_AH_SLEEPABLE);
702 	rc = bnxt_qplib_destroy_ah(&rdev->qplib_res, &ah->qplib_ah, block);
703 	if (BNXT_RE_CHECK_RC(rc)) {
704 		if (rc == -ETIMEDOUT)
705 			rc = 0;
706 		else
707 			goto fail;
708 	}
709 	atomic_dec(&rdev->stats.res.ah_count);
710 fail:
711 	return rc;
712 }
713 
714 static u8 bnxt_re_stack_to_dev_nw_type(enum rdma_network_type ntype)
715 {
716 	u8 nw_type;
717 
718 	switch (ntype) {
719 	case RDMA_NETWORK_IPV4:
720 		nw_type = CMDQ_CREATE_AH_TYPE_V2IPV4;
721 		break;
722 	case RDMA_NETWORK_IPV6:
723 		nw_type = CMDQ_CREATE_AH_TYPE_V2IPV6;
724 		break;
725 	default:
726 		nw_type = CMDQ_CREATE_AH_TYPE_V1;
727 		break;
728 	}
729 	return nw_type;
730 }
731 
732 int bnxt_re_create_ah(struct ib_ah *ib_ah, struct rdma_ah_init_attr *init_attr,
733 		      struct ib_udata *udata)
734 {
735 	struct ib_pd *ib_pd = ib_ah->pd;
736 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
737 	struct rdma_ah_attr *ah_attr = init_attr->ah_attr;
738 	const struct ib_global_route *grh = rdma_ah_read_grh(ah_attr);
739 	struct bnxt_re_dev *rdev = pd->rdev;
740 	const struct ib_gid_attr *sgid_attr;
741 	struct bnxt_re_gid_ctx *ctx;
742 	struct bnxt_re_ah *ah = container_of(ib_ah, struct bnxt_re_ah, ib_ah);
743 	u32 active_ahs;
744 	u8 nw_type;
745 	int rc;
746 
747 	if (!(rdma_ah_get_ah_flags(ah_attr) & IB_AH_GRH)) {
748 		ibdev_err(&rdev->ibdev, "Failed to alloc AH: GRH not set");
749 		return -EINVAL;
750 	}
751 
752 	ah->rdev = rdev;
753 	ah->qplib_ah.pd = &pd->qplib_pd;
754 
755 	/* Supply the configuration for the HW */
756 	memcpy(ah->qplib_ah.dgid.data, grh->dgid.raw,
757 	       sizeof(union ib_gid));
758 	sgid_attr = grh->sgid_attr;
759 	/* Get the HW context of the GID. The reference
760 	 * of GID table entry is already taken by the caller.
761 	 */
762 	ctx = rdma_read_gid_hw_context(sgid_attr);
763 	ah->qplib_ah.sgid_index = ctx->idx;
764 	ah->qplib_ah.host_sgid_index = grh->sgid_index;
765 	ah->qplib_ah.traffic_class = grh->traffic_class;
766 	ah->qplib_ah.flow_label = grh->flow_label;
767 	ah->qplib_ah.hop_limit = grh->hop_limit;
768 	ah->qplib_ah.sl = rdma_ah_get_sl(ah_attr);
769 
770 	/* Get network header type for this GID */
771 	nw_type = rdma_gid_attr_network_type(sgid_attr);
772 	ah->qplib_ah.nw_type = bnxt_re_stack_to_dev_nw_type(nw_type);
773 
774 	memcpy(ah->qplib_ah.dmac, ah_attr->roce.dmac, ETH_ALEN);
775 	rc = bnxt_qplib_create_ah(&rdev->qplib_res, &ah->qplib_ah,
776 				  !(init_attr->flags &
777 				    RDMA_CREATE_AH_SLEEPABLE));
778 	if (rc) {
779 		ibdev_err(&rdev->ibdev, "Failed to allocate HW AH");
780 		return rc;
781 	}
782 
783 	/* Write AVID to shared page. */
784 	if (udata) {
785 		struct bnxt_re_ucontext *uctx = rdma_udata_to_drv_context(
786 			udata, struct bnxt_re_ucontext, ib_uctx);
787 		unsigned long flag;
788 		u32 *wrptr;
789 
790 		spin_lock_irqsave(&uctx->sh_lock, flag);
791 		wrptr = (u32 *)(uctx->shpg + BNXT_RE_AVID_OFFT);
792 		*wrptr = ah->qplib_ah.id;
793 		wmb(); /* make sure cache is updated. */
794 		spin_unlock_irqrestore(&uctx->sh_lock, flag);
795 	}
796 	active_ahs = atomic_inc_return(&rdev->stats.res.ah_count);
797 	if (active_ahs > rdev->stats.res.ah_watermark)
798 		rdev->stats.res.ah_watermark = active_ahs;
799 
800 	return 0;
801 }
802 
803 int bnxt_re_query_ah(struct ib_ah *ib_ah, struct rdma_ah_attr *ah_attr)
804 {
805 	struct bnxt_re_ah *ah = container_of(ib_ah, struct bnxt_re_ah, ib_ah);
806 
807 	ah_attr->type = ib_ah->type;
808 	rdma_ah_set_sl(ah_attr, ah->qplib_ah.sl);
809 	memcpy(ah_attr->roce.dmac, ah->qplib_ah.dmac, ETH_ALEN);
810 	rdma_ah_set_grh(ah_attr, NULL, 0,
811 			ah->qplib_ah.host_sgid_index,
812 			0, ah->qplib_ah.traffic_class);
813 	rdma_ah_set_dgid_raw(ah_attr, ah->qplib_ah.dgid.data);
814 	rdma_ah_set_port_num(ah_attr, 1);
815 	rdma_ah_set_static_rate(ah_attr, 0);
816 	return 0;
817 }
818 
819 unsigned long bnxt_re_lock_cqs(struct bnxt_re_qp *qp)
820 	__acquires(&qp->scq->cq_lock) __acquires(&qp->rcq->cq_lock)
821 {
822 	unsigned long flags;
823 
824 	spin_lock_irqsave(&qp->scq->cq_lock, flags);
825 	if (qp->rcq != qp->scq)
826 		spin_lock(&qp->rcq->cq_lock);
827 	else
828 		__acquire(&qp->rcq->cq_lock);
829 
830 	return flags;
831 }
832 
833 void bnxt_re_unlock_cqs(struct bnxt_re_qp *qp,
834 			unsigned long flags)
835 	__releases(&qp->scq->cq_lock) __releases(&qp->rcq->cq_lock)
836 {
837 	if (qp->rcq != qp->scq)
838 		spin_unlock(&qp->rcq->cq_lock);
839 	else
840 		__release(&qp->rcq->cq_lock);
841 	spin_unlock_irqrestore(&qp->scq->cq_lock, flags);
842 }
843 
844 static int bnxt_re_destroy_gsi_sqp(struct bnxt_re_qp *qp)
845 {
846 	struct bnxt_re_qp *gsi_sqp;
847 	struct bnxt_re_ah *gsi_sah;
848 	struct bnxt_re_dev *rdev;
849 	int rc;
850 
851 	rdev = qp->rdev;
852 	gsi_sqp = rdev->gsi_ctx.gsi_sqp;
853 	gsi_sah = rdev->gsi_ctx.gsi_sah;
854 
855 	ibdev_dbg(&rdev->ibdev, "Destroy the shadow AH\n");
856 	bnxt_qplib_destroy_ah(&rdev->qplib_res,
857 			      &gsi_sah->qplib_ah,
858 			      true);
859 	atomic_dec(&rdev->stats.res.ah_count);
860 	bnxt_qplib_clean_qp(&qp->qplib_qp);
861 
862 	ibdev_dbg(&rdev->ibdev, "Destroy the shadow QP\n");
863 	rc = bnxt_qplib_destroy_qp(&rdev->qplib_res, &gsi_sqp->qplib_qp);
864 	if (rc) {
865 		ibdev_err(&rdev->ibdev, "Destroy Shadow QP failed");
866 		goto fail;
867 	}
868 	bnxt_qplib_free_qp_res(&rdev->qplib_res, &gsi_sqp->qplib_qp);
869 
870 	/* remove from active qp list */
871 	mutex_lock(&rdev->qp_lock);
872 	list_del(&gsi_sqp->list);
873 	mutex_unlock(&rdev->qp_lock);
874 	atomic_dec(&rdev->stats.res.qp_count);
875 
876 	kfree(rdev->gsi_ctx.sqp_tbl);
877 	kfree(gsi_sah);
878 	kfree(gsi_sqp);
879 	rdev->gsi_ctx.gsi_sqp = NULL;
880 	rdev->gsi_ctx.gsi_sah = NULL;
881 	rdev->gsi_ctx.sqp_tbl = NULL;
882 
883 	return 0;
884 fail:
885 	return rc;
886 }
887 
888 /* Queue Pairs */
889 int bnxt_re_destroy_qp(struct ib_qp *ib_qp, struct ib_udata *udata)
890 {
891 	struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
892 	struct bnxt_qplib_qp *qplib_qp = &qp->qplib_qp;
893 	struct bnxt_re_dev *rdev = qp->rdev;
894 	struct bnxt_qplib_nq *scq_nq = NULL;
895 	struct bnxt_qplib_nq *rcq_nq = NULL;
896 	unsigned int flags;
897 	int rc;
898 
899 	bnxt_qplib_flush_cqn_wq(&qp->qplib_qp);
900 
901 	rc = bnxt_qplib_destroy_qp(&rdev->qplib_res, &qp->qplib_qp);
902 	if (rc) {
903 		ibdev_err(&rdev->ibdev, "Failed to destroy HW QP");
904 		return rc;
905 	}
906 
907 	if (rdma_is_kernel_res(&qp->ib_qp.res)) {
908 		flags = bnxt_re_lock_cqs(qp);
909 		bnxt_qplib_clean_qp(&qp->qplib_qp);
910 		bnxt_re_unlock_cqs(qp, flags);
911 	}
912 
913 	bnxt_qplib_free_qp_res(&rdev->qplib_res, &qp->qplib_qp);
914 
915 	if (ib_qp->qp_type == IB_QPT_GSI && rdev->gsi_ctx.gsi_sqp) {
916 		rc = bnxt_re_destroy_gsi_sqp(qp);
917 		if (rc)
918 			return rc;
919 	}
920 
921 	mutex_lock(&rdev->qp_lock);
922 	list_del(&qp->list);
923 	mutex_unlock(&rdev->qp_lock);
924 	atomic_dec(&rdev->stats.res.qp_count);
925 	if (qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_RC)
926 		atomic_dec(&rdev->stats.res.rc_qp_count);
927 	else if (qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_UD)
928 		atomic_dec(&rdev->stats.res.ud_qp_count);
929 
930 	ib_umem_release(qp->rumem);
931 	ib_umem_release(qp->sumem);
932 
933 	/* Flush all the entries of notification queue associated with
934 	 * given qp.
935 	 */
936 	scq_nq = qplib_qp->scq->nq;
937 	rcq_nq = qplib_qp->rcq->nq;
938 	bnxt_re_synchronize_nq(scq_nq);
939 	if (scq_nq != rcq_nq)
940 		bnxt_re_synchronize_nq(rcq_nq);
941 
942 	return 0;
943 }
944 
945 static u8 __from_ib_qp_type(enum ib_qp_type type)
946 {
947 	switch (type) {
948 	case IB_QPT_GSI:
949 		return CMDQ_CREATE_QP1_TYPE_GSI;
950 	case IB_QPT_RC:
951 		return CMDQ_CREATE_QP_TYPE_RC;
952 	case IB_QPT_UD:
953 		return CMDQ_CREATE_QP_TYPE_UD;
954 	default:
955 		return IB_QPT_MAX;
956 	}
957 }
958 
959 static u16 bnxt_re_setup_rwqe_size(struct bnxt_qplib_qp *qplqp,
960 				   int rsge, int max)
961 {
962 	if (qplqp->wqe_mode == BNXT_QPLIB_WQE_MODE_STATIC)
963 		rsge = max;
964 	return bnxt_re_get_rwqe_size(rsge);
965 }
966 
967 static u16 bnxt_re_get_wqe_size(int ilsize, int nsge)
968 {
969 	u16 wqe_size, calc_ils;
970 
971 	wqe_size = bnxt_re_get_swqe_size(nsge);
972 	if (ilsize) {
973 		calc_ils = sizeof(struct sq_send_hdr) + ilsize;
974 		wqe_size = max_t(u16, calc_ils, wqe_size);
975 		wqe_size = ALIGN(wqe_size, sizeof(struct sq_send_hdr));
976 	}
977 	return wqe_size;
978 }
979 
980 static int bnxt_re_setup_swqe_size(struct bnxt_re_qp *qp,
981 				   struct ib_qp_init_attr *init_attr)
982 {
983 	struct bnxt_qplib_dev_attr *dev_attr;
984 	struct bnxt_qplib_qp *qplqp;
985 	struct bnxt_re_dev *rdev;
986 	struct bnxt_qplib_q *sq;
987 	int align, ilsize;
988 
989 	rdev = qp->rdev;
990 	qplqp = &qp->qplib_qp;
991 	sq = &qplqp->sq;
992 	dev_attr = &rdev->dev_attr;
993 
994 	align = sizeof(struct sq_send_hdr);
995 	ilsize = ALIGN(init_attr->cap.max_inline_data, align);
996 
997 	sq->wqe_size = bnxt_re_get_wqe_size(ilsize, sq->max_sge);
998 	if (sq->wqe_size > bnxt_re_get_swqe_size(dev_attr->max_qp_sges))
999 		return -EINVAL;
1000 	/* For gen p4 and gen p5 backward compatibility mode
1001 	 * wqe size is fixed to 128 bytes
1002 	 */
1003 	if (sq->wqe_size < bnxt_re_get_swqe_size(dev_attr->max_qp_sges) &&
1004 			qplqp->wqe_mode == BNXT_QPLIB_WQE_MODE_STATIC)
1005 		sq->wqe_size = bnxt_re_get_swqe_size(dev_attr->max_qp_sges);
1006 
1007 	if (init_attr->cap.max_inline_data) {
1008 		qplqp->max_inline_data = sq->wqe_size -
1009 			sizeof(struct sq_send_hdr);
1010 		init_attr->cap.max_inline_data = qplqp->max_inline_data;
1011 		if (qplqp->wqe_mode == BNXT_QPLIB_WQE_MODE_STATIC)
1012 			sq->max_sge = qplqp->max_inline_data /
1013 				sizeof(struct sq_sge);
1014 	}
1015 
1016 	return 0;
1017 }
1018 
1019 static int bnxt_re_init_user_qp(struct bnxt_re_dev *rdev, struct bnxt_re_pd *pd,
1020 				struct bnxt_re_qp *qp, struct ib_udata *udata)
1021 {
1022 	struct bnxt_qplib_qp *qplib_qp;
1023 	struct bnxt_re_ucontext *cntx;
1024 	struct bnxt_re_qp_req ureq;
1025 	int bytes = 0, psn_sz;
1026 	struct ib_umem *umem;
1027 	int psn_nume;
1028 
1029 	qplib_qp = &qp->qplib_qp;
1030 	cntx = rdma_udata_to_drv_context(udata, struct bnxt_re_ucontext,
1031 					 ib_uctx);
1032 	if (ib_copy_from_udata(&ureq, udata, sizeof(ureq)))
1033 		return -EFAULT;
1034 
1035 	bytes = (qplib_qp->sq.max_wqe * qplib_qp->sq.wqe_size);
1036 	/* Consider mapping PSN search memory only for RC QPs. */
1037 	if (qplib_qp->type == CMDQ_CREATE_QP_TYPE_RC) {
1038 		psn_sz = bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx) ?
1039 						   sizeof(struct sq_psn_search_ext) :
1040 						   sizeof(struct sq_psn_search);
1041 		psn_nume = (qplib_qp->wqe_mode == BNXT_QPLIB_WQE_MODE_STATIC) ?
1042 			    qplib_qp->sq.max_wqe :
1043 			    ((qplib_qp->sq.max_wqe * qplib_qp->sq.wqe_size) /
1044 			      sizeof(struct bnxt_qplib_sge));
1045 		if (_is_host_msn_table(rdev->qplib_res.dattr->dev_cap_flags2))
1046 			psn_nume = roundup_pow_of_two(psn_nume);
1047 		bytes += (psn_nume * psn_sz);
1048 	}
1049 
1050 	bytes = PAGE_ALIGN(bytes);
1051 	umem = ib_umem_get(&rdev->ibdev, ureq.qpsva, bytes,
1052 			   IB_ACCESS_LOCAL_WRITE);
1053 	if (IS_ERR(umem))
1054 		return PTR_ERR(umem);
1055 
1056 	qp->sumem = umem;
1057 	qplib_qp->sq.sg_info.umem = umem;
1058 	qplib_qp->sq.sg_info.pgsize = PAGE_SIZE;
1059 	qplib_qp->sq.sg_info.pgshft = PAGE_SHIFT;
1060 	qplib_qp->qp_handle = ureq.qp_handle;
1061 
1062 	if (!qp->qplib_qp.srq) {
1063 		bytes = (qplib_qp->rq.max_wqe * qplib_qp->rq.wqe_size);
1064 		bytes = PAGE_ALIGN(bytes);
1065 		umem = ib_umem_get(&rdev->ibdev, ureq.qprva, bytes,
1066 				   IB_ACCESS_LOCAL_WRITE);
1067 		if (IS_ERR(umem))
1068 			goto rqfail;
1069 		qp->rumem = umem;
1070 		qplib_qp->rq.sg_info.umem = umem;
1071 		qplib_qp->rq.sg_info.pgsize = PAGE_SIZE;
1072 		qplib_qp->rq.sg_info.pgshft = PAGE_SHIFT;
1073 	}
1074 
1075 	qplib_qp->dpi = &cntx->dpi;
1076 	return 0;
1077 rqfail:
1078 	ib_umem_release(qp->sumem);
1079 	qp->sumem = NULL;
1080 	memset(&qplib_qp->sq.sg_info, 0, sizeof(qplib_qp->sq.sg_info));
1081 
1082 	return PTR_ERR(umem);
1083 }
1084 
1085 static struct bnxt_re_ah *bnxt_re_create_shadow_qp_ah
1086 				(struct bnxt_re_pd *pd,
1087 				 struct bnxt_qplib_res *qp1_res,
1088 				 struct bnxt_qplib_qp *qp1_qp)
1089 {
1090 	struct bnxt_re_dev *rdev = pd->rdev;
1091 	struct bnxt_re_ah *ah;
1092 	union ib_gid sgid;
1093 	int rc;
1094 
1095 	ah = kzalloc(sizeof(*ah), GFP_KERNEL);
1096 	if (!ah)
1097 		return NULL;
1098 
1099 	ah->rdev = rdev;
1100 	ah->qplib_ah.pd = &pd->qplib_pd;
1101 
1102 	rc = bnxt_re_query_gid(&rdev->ibdev, 1, 0, &sgid);
1103 	if (rc)
1104 		goto fail;
1105 
1106 	/* supply the dgid data same as sgid */
1107 	memcpy(ah->qplib_ah.dgid.data, &sgid.raw,
1108 	       sizeof(union ib_gid));
1109 	ah->qplib_ah.sgid_index = 0;
1110 
1111 	ah->qplib_ah.traffic_class = 0;
1112 	ah->qplib_ah.flow_label = 0;
1113 	ah->qplib_ah.hop_limit = 1;
1114 	ah->qplib_ah.sl = 0;
1115 	/* Have DMAC same as SMAC */
1116 	ether_addr_copy(ah->qplib_ah.dmac, rdev->netdev->dev_addr);
1117 
1118 	rc = bnxt_qplib_create_ah(&rdev->qplib_res, &ah->qplib_ah, false);
1119 	if (rc) {
1120 		ibdev_err(&rdev->ibdev,
1121 			  "Failed to allocate HW AH for Shadow QP");
1122 		goto fail;
1123 	}
1124 	atomic_inc(&rdev->stats.res.ah_count);
1125 
1126 	return ah;
1127 
1128 fail:
1129 	kfree(ah);
1130 	return NULL;
1131 }
1132 
1133 static struct bnxt_re_qp *bnxt_re_create_shadow_qp
1134 				(struct bnxt_re_pd *pd,
1135 				 struct bnxt_qplib_res *qp1_res,
1136 				 struct bnxt_qplib_qp *qp1_qp)
1137 {
1138 	struct bnxt_re_dev *rdev = pd->rdev;
1139 	struct bnxt_re_qp *qp;
1140 	int rc;
1141 
1142 	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
1143 	if (!qp)
1144 		return NULL;
1145 
1146 	qp->rdev = rdev;
1147 
1148 	/* Initialize the shadow QP structure from the QP1 values */
1149 	ether_addr_copy(qp->qplib_qp.smac, rdev->netdev->dev_addr);
1150 
1151 	qp->qplib_qp.pd = &pd->qplib_pd;
1152 	qp->qplib_qp.qp_handle = (u64)(unsigned long)(&qp->qplib_qp);
1153 	qp->qplib_qp.type = IB_QPT_UD;
1154 
1155 	qp->qplib_qp.max_inline_data = 0;
1156 	qp->qplib_qp.sig_type = true;
1157 
1158 	/* Shadow QP SQ depth should be same as QP1 RQ depth */
1159 	qp->qplib_qp.sq.wqe_size = bnxt_re_get_wqe_size(0, 6);
1160 	qp->qplib_qp.sq.max_wqe = qp1_qp->rq.max_wqe;
1161 	qp->qplib_qp.sq.max_sw_wqe = qp1_qp->rq.max_wqe;
1162 	qp->qplib_qp.sq.max_sge = 2;
1163 	/* Q full delta can be 1 since it is internal QP */
1164 	qp->qplib_qp.sq.q_full_delta = 1;
1165 	qp->qplib_qp.sq.sg_info.pgsize = PAGE_SIZE;
1166 	qp->qplib_qp.sq.sg_info.pgshft = PAGE_SHIFT;
1167 
1168 	qp->qplib_qp.scq = qp1_qp->scq;
1169 	qp->qplib_qp.rcq = qp1_qp->rcq;
1170 
1171 	qp->qplib_qp.rq.wqe_size = bnxt_re_get_rwqe_size(6);
1172 	qp->qplib_qp.rq.max_wqe = qp1_qp->rq.max_wqe;
1173 	qp->qplib_qp.rq.max_sw_wqe = qp1_qp->rq.max_wqe;
1174 	qp->qplib_qp.rq.max_sge = qp1_qp->rq.max_sge;
1175 	/* Q full delta can be 1 since it is internal QP */
1176 	qp->qplib_qp.rq.q_full_delta = 1;
1177 	qp->qplib_qp.rq.sg_info.pgsize = PAGE_SIZE;
1178 	qp->qplib_qp.rq.sg_info.pgshft = PAGE_SHIFT;
1179 
1180 	qp->qplib_qp.mtu = qp1_qp->mtu;
1181 
1182 	qp->qplib_qp.sq_hdr_buf_size = 0;
1183 	qp->qplib_qp.rq_hdr_buf_size = BNXT_QPLIB_MAX_GRH_HDR_SIZE_IPV6;
1184 	qp->qplib_qp.dpi = &rdev->dpi_privileged;
1185 
1186 	rc = bnxt_qplib_create_qp(qp1_res, &qp->qplib_qp);
1187 	if (rc)
1188 		goto fail;
1189 
1190 	spin_lock_init(&qp->sq_lock);
1191 	INIT_LIST_HEAD(&qp->list);
1192 	mutex_lock(&rdev->qp_lock);
1193 	list_add_tail(&qp->list, &rdev->qp_list);
1194 	atomic_inc(&rdev->stats.res.qp_count);
1195 	mutex_unlock(&rdev->qp_lock);
1196 	return qp;
1197 fail:
1198 	kfree(qp);
1199 	return NULL;
1200 }
1201 
1202 static int bnxt_re_init_rq_attr(struct bnxt_re_qp *qp,
1203 				struct ib_qp_init_attr *init_attr,
1204 				struct bnxt_re_ucontext *uctx)
1205 {
1206 	struct bnxt_qplib_dev_attr *dev_attr;
1207 	struct bnxt_qplib_qp *qplqp;
1208 	struct bnxt_re_dev *rdev;
1209 	struct bnxt_qplib_q *rq;
1210 	int entries;
1211 
1212 	rdev = qp->rdev;
1213 	qplqp = &qp->qplib_qp;
1214 	rq = &qplqp->rq;
1215 	dev_attr = &rdev->dev_attr;
1216 
1217 	if (init_attr->srq) {
1218 		struct bnxt_re_srq *srq;
1219 
1220 		srq = container_of(init_attr->srq, struct bnxt_re_srq, ib_srq);
1221 		qplqp->srq = &srq->qplib_srq;
1222 		rq->max_wqe = 0;
1223 	} else {
1224 		rq->max_sge = init_attr->cap.max_recv_sge;
1225 		if (rq->max_sge > dev_attr->max_qp_sges)
1226 			rq->max_sge = dev_attr->max_qp_sges;
1227 		init_attr->cap.max_recv_sge = rq->max_sge;
1228 		rq->wqe_size = bnxt_re_setup_rwqe_size(qplqp, rq->max_sge,
1229 						       dev_attr->max_qp_sges);
1230 		/* Allocate 1 more than what's provided so posting max doesn't
1231 		 * mean empty.
1232 		 */
1233 		entries = bnxt_re_init_depth(init_attr->cap.max_recv_wr + 1, uctx);
1234 		rq->max_wqe = min_t(u32, entries, dev_attr->max_qp_wqes + 1);
1235 		rq->max_sw_wqe = rq->max_wqe;
1236 		rq->q_full_delta = 0;
1237 		rq->sg_info.pgsize = PAGE_SIZE;
1238 		rq->sg_info.pgshft = PAGE_SHIFT;
1239 	}
1240 
1241 	return 0;
1242 }
1243 
1244 static void bnxt_re_adjust_gsi_rq_attr(struct bnxt_re_qp *qp)
1245 {
1246 	struct bnxt_qplib_dev_attr *dev_attr;
1247 	struct bnxt_qplib_qp *qplqp;
1248 	struct bnxt_re_dev *rdev;
1249 
1250 	rdev = qp->rdev;
1251 	qplqp = &qp->qplib_qp;
1252 	dev_attr = &rdev->dev_attr;
1253 
1254 	if (!bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx)) {
1255 		qplqp->rq.max_sge = dev_attr->max_qp_sges;
1256 		if (qplqp->rq.max_sge > dev_attr->max_qp_sges)
1257 			qplqp->rq.max_sge = dev_attr->max_qp_sges;
1258 		qplqp->rq.max_sge = 6;
1259 	}
1260 }
1261 
1262 static int bnxt_re_init_sq_attr(struct bnxt_re_qp *qp,
1263 				struct ib_qp_init_attr *init_attr,
1264 				struct bnxt_re_ucontext *uctx)
1265 {
1266 	struct bnxt_qplib_dev_attr *dev_attr;
1267 	struct bnxt_qplib_qp *qplqp;
1268 	struct bnxt_re_dev *rdev;
1269 	struct bnxt_qplib_q *sq;
1270 	int entries;
1271 	int diff;
1272 	int rc;
1273 
1274 	rdev = qp->rdev;
1275 	qplqp = &qp->qplib_qp;
1276 	sq = &qplqp->sq;
1277 	dev_attr = &rdev->dev_attr;
1278 
1279 	sq->max_sge = init_attr->cap.max_send_sge;
1280 	if (sq->max_sge > dev_attr->max_qp_sges) {
1281 		sq->max_sge = dev_attr->max_qp_sges;
1282 		init_attr->cap.max_send_sge = sq->max_sge;
1283 	}
1284 
1285 	rc = bnxt_re_setup_swqe_size(qp, init_attr);
1286 	if (rc)
1287 		return rc;
1288 
1289 	entries = init_attr->cap.max_send_wr;
1290 	/* Allocate 128 + 1 more than what's provided */
1291 	diff = (qplqp->wqe_mode == BNXT_QPLIB_WQE_MODE_VARIABLE) ?
1292 		0 : BNXT_QPLIB_RESERVED_QP_WRS;
1293 	entries = bnxt_re_init_depth(entries + diff + 1, uctx);
1294 	sq->max_wqe = min_t(u32, entries, dev_attr->max_qp_wqes + diff + 1);
1295 	sq->max_sw_wqe = bnxt_qplib_get_depth(sq, qplqp->wqe_mode, true);
1296 	sq->q_full_delta = diff + 1;
1297 	/*
1298 	 * Reserving one slot for Phantom WQE. Application can
1299 	 * post one extra entry in this case. But allowing this to avoid
1300 	 * unexpected Queue full condition
1301 	 */
1302 	qplqp->sq.q_full_delta -= 1;
1303 	qplqp->sq.sg_info.pgsize = PAGE_SIZE;
1304 	qplqp->sq.sg_info.pgshft = PAGE_SHIFT;
1305 
1306 	return 0;
1307 }
1308 
1309 static void bnxt_re_adjust_gsi_sq_attr(struct bnxt_re_qp *qp,
1310 				       struct ib_qp_init_attr *init_attr,
1311 				       struct bnxt_re_ucontext *uctx)
1312 {
1313 	struct bnxt_qplib_dev_attr *dev_attr;
1314 	struct bnxt_qplib_qp *qplqp;
1315 	struct bnxt_re_dev *rdev;
1316 	int entries;
1317 
1318 	rdev = qp->rdev;
1319 	qplqp = &qp->qplib_qp;
1320 	dev_attr = &rdev->dev_attr;
1321 
1322 	if (!bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx)) {
1323 		entries = bnxt_re_init_depth(init_attr->cap.max_send_wr + 1, uctx);
1324 		qplqp->sq.max_wqe = min_t(u32, entries,
1325 					  dev_attr->max_qp_wqes + 1);
1326 		qplqp->sq.q_full_delta = qplqp->sq.max_wqe -
1327 			init_attr->cap.max_send_wr;
1328 		qplqp->sq.max_sge++; /* Need one extra sge to put UD header */
1329 		if (qplqp->sq.max_sge > dev_attr->max_qp_sges)
1330 			qplqp->sq.max_sge = dev_attr->max_qp_sges;
1331 	}
1332 }
1333 
1334 static int bnxt_re_init_qp_type(struct bnxt_re_dev *rdev,
1335 				struct ib_qp_init_attr *init_attr)
1336 {
1337 	struct bnxt_qplib_chip_ctx *chip_ctx;
1338 	int qptype;
1339 
1340 	chip_ctx = rdev->chip_ctx;
1341 
1342 	qptype = __from_ib_qp_type(init_attr->qp_type);
1343 	if (qptype == IB_QPT_MAX) {
1344 		ibdev_err(&rdev->ibdev, "QP type 0x%x not supported", qptype);
1345 		qptype = -EOPNOTSUPP;
1346 		goto out;
1347 	}
1348 
1349 	if (bnxt_qplib_is_chip_gen_p5_p7(chip_ctx) &&
1350 	    init_attr->qp_type == IB_QPT_GSI)
1351 		qptype = CMDQ_CREATE_QP_TYPE_GSI;
1352 out:
1353 	return qptype;
1354 }
1355 
1356 static int bnxt_re_init_qp_attr(struct bnxt_re_qp *qp, struct bnxt_re_pd *pd,
1357 				struct ib_qp_init_attr *init_attr,
1358 				struct ib_udata *udata)
1359 {
1360 	struct bnxt_qplib_dev_attr *dev_attr;
1361 	struct bnxt_re_ucontext *uctx;
1362 	struct bnxt_qplib_qp *qplqp;
1363 	struct bnxt_re_dev *rdev;
1364 	struct bnxt_re_cq *cq;
1365 	int rc = 0, qptype;
1366 
1367 	rdev = qp->rdev;
1368 	qplqp = &qp->qplib_qp;
1369 	dev_attr = &rdev->dev_attr;
1370 
1371 	uctx = rdma_udata_to_drv_context(udata, struct bnxt_re_ucontext, ib_uctx);
1372 	/* Setup misc params */
1373 	ether_addr_copy(qplqp->smac, rdev->netdev->dev_addr);
1374 	qplqp->pd = &pd->qplib_pd;
1375 	qplqp->qp_handle = (u64)qplqp;
1376 	qplqp->max_inline_data = init_attr->cap.max_inline_data;
1377 	qplqp->sig_type = init_attr->sq_sig_type == IB_SIGNAL_ALL_WR;
1378 	qptype = bnxt_re_init_qp_type(rdev, init_attr);
1379 	if (qptype < 0) {
1380 		rc = qptype;
1381 		goto out;
1382 	}
1383 	qplqp->type = (u8)qptype;
1384 	qplqp->wqe_mode = rdev->chip_ctx->modes.wqe_mode;
1385 
1386 	if (init_attr->qp_type == IB_QPT_RC) {
1387 		qplqp->max_rd_atomic = dev_attr->max_qp_rd_atom;
1388 		qplqp->max_dest_rd_atomic = dev_attr->max_qp_init_rd_atom;
1389 	}
1390 	qplqp->mtu = ib_mtu_enum_to_int(iboe_get_mtu(rdev->netdev->mtu));
1391 	qplqp->dpi = &rdev->dpi_privileged; /* Doorbell page */
1392 	if (init_attr->create_flags) {
1393 		ibdev_dbg(&rdev->ibdev,
1394 			  "QP create flags 0x%x not supported",
1395 			  init_attr->create_flags);
1396 		return -EOPNOTSUPP;
1397 	}
1398 
1399 	/* Setup CQs */
1400 	if (init_attr->send_cq) {
1401 		cq = container_of(init_attr->send_cq, struct bnxt_re_cq, ib_cq);
1402 		qplqp->scq = &cq->qplib_cq;
1403 		qp->scq = cq;
1404 	}
1405 
1406 	if (init_attr->recv_cq) {
1407 		cq = container_of(init_attr->recv_cq, struct bnxt_re_cq, ib_cq);
1408 		qplqp->rcq = &cq->qplib_cq;
1409 		qp->rcq = cq;
1410 	}
1411 
1412 	/* Setup RQ/SRQ */
1413 	rc = bnxt_re_init_rq_attr(qp, init_attr, uctx);
1414 	if (rc)
1415 		goto out;
1416 	if (init_attr->qp_type == IB_QPT_GSI)
1417 		bnxt_re_adjust_gsi_rq_attr(qp);
1418 
1419 	/* Setup SQ */
1420 	rc = bnxt_re_init_sq_attr(qp, init_attr, uctx);
1421 	if (rc)
1422 		goto out;
1423 	if (init_attr->qp_type == IB_QPT_GSI)
1424 		bnxt_re_adjust_gsi_sq_attr(qp, init_attr, uctx);
1425 
1426 	if (udata) /* This will update DPI and qp_handle */
1427 		rc = bnxt_re_init_user_qp(rdev, pd, qp, udata);
1428 out:
1429 	return rc;
1430 }
1431 
1432 static int bnxt_re_create_shadow_gsi(struct bnxt_re_qp *qp,
1433 				     struct bnxt_re_pd *pd)
1434 {
1435 	struct bnxt_re_sqp_entries *sqp_tbl;
1436 	struct bnxt_re_dev *rdev;
1437 	struct bnxt_re_qp *sqp;
1438 	struct bnxt_re_ah *sah;
1439 	int rc = 0;
1440 
1441 	rdev = qp->rdev;
1442 	/* Create a shadow QP to handle the QP1 traffic */
1443 	sqp_tbl = kcalloc(BNXT_RE_MAX_GSI_SQP_ENTRIES, sizeof(*sqp_tbl),
1444 			  GFP_KERNEL);
1445 	if (!sqp_tbl)
1446 		return -ENOMEM;
1447 	rdev->gsi_ctx.sqp_tbl = sqp_tbl;
1448 
1449 	sqp = bnxt_re_create_shadow_qp(pd, &rdev->qplib_res, &qp->qplib_qp);
1450 	if (!sqp) {
1451 		rc = -ENODEV;
1452 		ibdev_err(&rdev->ibdev, "Failed to create Shadow QP for QP1");
1453 		goto out;
1454 	}
1455 	rdev->gsi_ctx.gsi_sqp = sqp;
1456 
1457 	sqp->rcq = qp->rcq;
1458 	sqp->scq = qp->scq;
1459 	sah = bnxt_re_create_shadow_qp_ah(pd, &rdev->qplib_res,
1460 					  &qp->qplib_qp);
1461 	if (!sah) {
1462 		bnxt_qplib_destroy_qp(&rdev->qplib_res,
1463 				      &sqp->qplib_qp);
1464 		rc = -ENODEV;
1465 		ibdev_err(&rdev->ibdev,
1466 			  "Failed to create AH entry for ShadowQP");
1467 		goto out;
1468 	}
1469 	rdev->gsi_ctx.gsi_sah = sah;
1470 
1471 	return 0;
1472 out:
1473 	kfree(sqp_tbl);
1474 	return rc;
1475 }
1476 
1477 static int bnxt_re_create_gsi_qp(struct bnxt_re_qp *qp, struct bnxt_re_pd *pd,
1478 				 struct ib_qp_init_attr *init_attr)
1479 {
1480 	struct bnxt_re_dev *rdev;
1481 	struct bnxt_qplib_qp *qplqp;
1482 	int rc;
1483 
1484 	rdev = qp->rdev;
1485 	qplqp = &qp->qplib_qp;
1486 
1487 	qplqp->rq_hdr_buf_size = BNXT_QPLIB_MAX_QP1_RQ_HDR_SIZE_V2;
1488 	qplqp->sq_hdr_buf_size = BNXT_QPLIB_MAX_QP1_SQ_HDR_SIZE_V2;
1489 
1490 	rc = bnxt_qplib_create_qp1(&rdev->qplib_res, qplqp);
1491 	if (rc) {
1492 		ibdev_err(&rdev->ibdev, "create HW QP1 failed!");
1493 		goto out;
1494 	}
1495 
1496 	rc = bnxt_re_create_shadow_gsi(qp, pd);
1497 out:
1498 	return rc;
1499 }
1500 
1501 static bool bnxt_re_test_qp_limits(struct bnxt_re_dev *rdev,
1502 				   struct ib_qp_init_attr *init_attr,
1503 				   struct bnxt_qplib_dev_attr *dev_attr)
1504 {
1505 	bool rc = true;
1506 
1507 	if (init_attr->cap.max_send_wr > dev_attr->max_qp_wqes ||
1508 	    init_attr->cap.max_recv_wr > dev_attr->max_qp_wqes ||
1509 	    init_attr->cap.max_send_sge > dev_attr->max_qp_sges ||
1510 	    init_attr->cap.max_recv_sge > dev_attr->max_qp_sges ||
1511 	    init_attr->cap.max_inline_data > dev_attr->max_inline_data) {
1512 		ibdev_err(&rdev->ibdev,
1513 			  "Create QP failed - max exceeded! 0x%x/0x%x 0x%x/0x%x 0x%x/0x%x 0x%x/0x%x 0x%x/0x%x",
1514 			  init_attr->cap.max_send_wr, dev_attr->max_qp_wqes,
1515 			  init_attr->cap.max_recv_wr, dev_attr->max_qp_wqes,
1516 			  init_attr->cap.max_send_sge, dev_attr->max_qp_sges,
1517 			  init_attr->cap.max_recv_sge, dev_attr->max_qp_sges,
1518 			  init_attr->cap.max_inline_data,
1519 			  dev_attr->max_inline_data);
1520 		rc = false;
1521 	}
1522 	return rc;
1523 }
1524 
1525 int bnxt_re_create_qp(struct ib_qp *ib_qp, struct ib_qp_init_attr *qp_init_attr,
1526 		      struct ib_udata *udata)
1527 {
1528 	struct ib_pd *ib_pd = ib_qp->pd;
1529 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
1530 	struct bnxt_re_dev *rdev = pd->rdev;
1531 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
1532 	struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
1533 	u32 active_qps;
1534 	int rc;
1535 
1536 	rc = bnxt_re_test_qp_limits(rdev, qp_init_attr, dev_attr);
1537 	if (!rc) {
1538 		rc = -EINVAL;
1539 		goto fail;
1540 	}
1541 
1542 	qp->rdev = rdev;
1543 	rc = bnxt_re_init_qp_attr(qp, pd, qp_init_attr, udata);
1544 	if (rc)
1545 		goto fail;
1546 
1547 	if (qp_init_attr->qp_type == IB_QPT_GSI &&
1548 	    !(bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx))) {
1549 		rc = bnxt_re_create_gsi_qp(qp, pd, qp_init_attr);
1550 		if (rc == -ENODEV)
1551 			goto qp_destroy;
1552 		if (rc)
1553 			goto fail;
1554 	} else {
1555 		rc = bnxt_qplib_create_qp(&rdev->qplib_res, &qp->qplib_qp);
1556 		if (rc) {
1557 			ibdev_err(&rdev->ibdev, "Failed to create HW QP");
1558 			goto free_umem;
1559 		}
1560 		if (udata) {
1561 			struct bnxt_re_qp_resp resp;
1562 
1563 			resp.qpid = qp->qplib_qp.id;
1564 			resp.rsvd = 0;
1565 			rc = ib_copy_to_udata(udata, &resp, sizeof(resp));
1566 			if (rc) {
1567 				ibdev_err(&rdev->ibdev, "Failed to copy QP udata");
1568 				goto qp_destroy;
1569 			}
1570 		}
1571 	}
1572 
1573 	qp->ib_qp.qp_num = qp->qplib_qp.id;
1574 	if (qp_init_attr->qp_type == IB_QPT_GSI)
1575 		rdev->gsi_ctx.gsi_qp = qp;
1576 	spin_lock_init(&qp->sq_lock);
1577 	spin_lock_init(&qp->rq_lock);
1578 	INIT_LIST_HEAD(&qp->list);
1579 	mutex_lock(&rdev->qp_lock);
1580 	list_add_tail(&qp->list, &rdev->qp_list);
1581 	mutex_unlock(&rdev->qp_lock);
1582 	active_qps = atomic_inc_return(&rdev->stats.res.qp_count);
1583 	if (active_qps > rdev->stats.res.qp_watermark)
1584 		rdev->stats.res.qp_watermark = active_qps;
1585 	if (qp_init_attr->qp_type == IB_QPT_RC) {
1586 		active_qps = atomic_inc_return(&rdev->stats.res.rc_qp_count);
1587 		if (active_qps > rdev->stats.res.rc_qp_watermark)
1588 			rdev->stats.res.rc_qp_watermark = active_qps;
1589 	} else if (qp_init_attr->qp_type == IB_QPT_UD) {
1590 		active_qps = atomic_inc_return(&rdev->stats.res.ud_qp_count);
1591 		if (active_qps > rdev->stats.res.ud_qp_watermark)
1592 			rdev->stats.res.ud_qp_watermark = active_qps;
1593 	}
1594 
1595 	return 0;
1596 qp_destroy:
1597 	bnxt_qplib_destroy_qp(&rdev->qplib_res, &qp->qplib_qp);
1598 free_umem:
1599 	ib_umem_release(qp->rumem);
1600 	ib_umem_release(qp->sumem);
1601 fail:
1602 	return rc;
1603 }
1604 
1605 static u8 __from_ib_qp_state(enum ib_qp_state state)
1606 {
1607 	switch (state) {
1608 	case IB_QPS_RESET:
1609 		return CMDQ_MODIFY_QP_NEW_STATE_RESET;
1610 	case IB_QPS_INIT:
1611 		return CMDQ_MODIFY_QP_NEW_STATE_INIT;
1612 	case IB_QPS_RTR:
1613 		return CMDQ_MODIFY_QP_NEW_STATE_RTR;
1614 	case IB_QPS_RTS:
1615 		return CMDQ_MODIFY_QP_NEW_STATE_RTS;
1616 	case IB_QPS_SQD:
1617 		return CMDQ_MODIFY_QP_NEW_STATE_SQD;
1618 	case IB_QPS_SQE:
1619 		return CMDQ_MODIFY_QP_NEW_STATE_SQE;
1620 	case IB_QPS_ERR:
1621 	default:
1622 		return CMDQ_MODIFY_QP_NEW_STATE_ERR;
1623 	}
1624 }
1625 
1626 static enum ib_qp_state __to_ib_qp_state(u8 state)
1627 {
1628 	switch (state) {
1629 	case CMDQ_MODIFY_QP_NEW_STATE_RESET:
1630 		return IB_QPS_RESET;
1631 	case CMDQ_MODIFY_QP_NEW_STATE_INIT:
1632 		return IB_QPS_INIT;
1633 	case CMDQ_MODIFY_QP_NEW_STATE_RTR:
1634 		return IB_QPS_RTR;
1635 	case CMDQ_MODIFY_QP_NEW_STATE_RTS:
1636 		return IB_QPS_RTS;
1637 	case CMDQ_MODIFY_QP_NEW_STATE_SQD:
1638 		return IB_QPS_SQD;
1639 	case CMDQ_MODIFY_QP_NEW_STATE_SQE:
1640 		return IB_QPS_SQE;
1641 	case CMDQ_MODIFY_QP_NEW_STATE_ERR:
1642 	default:
1643 		return IB_QPS_ERR;
1644 	}
1645 }
1646 
1647 static u32 __from_ib_mtu(enum ib_mtu mtu)
1648 {
1649 	switch (mtu) {
1650 	case IB_MTU_256:
1651 		return CMDQ_MODIFY_QP_PATH_MTU_MTU_256;
1652 	case IB_MTU_512:
1653 		return CMDQ_MODIFY_QP_PATH_MTU_MTU_512;
1654 	case IB_MTU_1024:
1655 		return CMDQ_MODIFY_QP_PATH_MTU_MTU_1024;
1656 	case IB_MTU_2048:
1657 		return CMDQ_MODIFY_QP_PATH_MTU_MTU_2048;
1658 	case IB_MTU_4096:
1659 		return CMDQ_MODIFY_QP_PATH_MTU_MTU_4096;
1660 	default:
1661 		return CMDQ_MODIFY_QP_PATH_MTU_MTU_2048;
1662 	}
1663 }
1664 
1665 static enum ib_mtu __to_ib_mtu(u32 mtu)
1666 {
1667 	switch (mtu & CREQ_QUERY_QP_RESP_SB_PATH_MTU_MASK) {
1668 	case CMDQ_MODIFY_QP_PATH_MTU_MTU_256:
1669 		return IB_MTU_256;
1670 	case CMDQ_MODIFY_QP_PATH_MTU_MTU_512:
1671 		return IB_MTU_512;
1672 	case CMDQ_MODIFY_QP_PATH_MTU_MTU_1024:
1673 		return IB_MTU_1024;
1674 	case CMDQ_MODIFY_QP_PATH_MTU_MTU_2048:
1675 		return IB_MTU_2048;
1676 	case CMDQ_MODIFY_QP_PATH_MTU_MTU_4096:
1677 		return IB_MTU_4096;
1678 	default:
1679 		return IB_MTU_2048;
1680 	}
1681 }
1682 
1683 /* Shared Receive Queues */
1684 int bnxt_re_destroy_srq(struct ib_srq *ib_srq, struct ib_udata *udata)
1685 {
1686 	struct bnxt_re_srq *srq = container_of(ib_srq, struct bnxt_re_srq,
1687 					       ib_srq);
1688 	struct bnxt_re_dev *rdev = srq->rdev;
1689 	struct bnxt_qplib_srq *qplib_srq = &srq->qplib_srq;
1690 	struct bnxt_qplib_nq *nq = NULL;
1691 
1692 	if (qplib_srq->cq)
1693 		nq = qplib_srq->cq->nq;
1694 	bnxt_qplib_destroy_srq(&rdev->qplib_res, qplib_srq);
1695 	ib_umem_release(srq->umem);
1696 	atomic_dec(&rdev->stats.res.srq_count);
1697 	if (nq)
1698 		nq->budget--;
1699 	return 0;
1700 }
1701 
1702 static int bnxt_re_init_user_srq(struct bnxt_re_dev *rdev,
1703 				 struct bnxt_re_pd *pd,
1704 				 struct bnxt_re_srq *srq,
1705 				 struct ib_udata *udata)
1706 {
1707 	struct bnxt_re_srq_req ureq;
1708 	struct bnxt_qplib_srq *qplib_srq = &srq->qplib_srq;
1709 	struct ib_umem *umem;
1710 	int bytes = 0;
1711 	struct bnxt_re_ucontext *cntx = rdma_udata_to_drv_context(
1712 		udata, struct bnxt_re_ucontext, ib_uctx);
1713 
1714 	if (ib_copy_from_udata(&ureq, udata, sizeof(ureq)))
1715 		return -EFAULT;
1716 
1717 	bytes = (qplib_srq->max_wqe * qplib_srq->wqe_size);
1718 	bytes = PAGE_ALIGN(bytes);
1719 	umem = ib_umem_get(&rdev->ibdev, ureq.srqva, bytes,
1720 			   IB_ACCESS_LOCAL_WRITE);
1721 	if (IS_ERR(umem))
1722 		return PTR_ERR(umem);
1723 
1724 	srq->umem = umem;
1725 	qplib_srq->sg_info.umem = umem;
1726 	qplib_srq->sg_info.pgsize = PAGE_SIZE;
1727 	qplib_srq->sg_info.pgshft = PAGE_SHIFT;
1728 	qplib_srq->srq_handle = ureq.srq_handle;
1729 	qplib_srq->dpi = &cntx->dpi;
1730 
1731 	return 0;
1732 }
1733 
1734 int bnxt_re_create_srq(struct ib_srq *ib_srq,
1735 		       struct ib_srq_init_attr *srq_init_attr,
1736 		       struct ib_udata *udata)
1737 {
1738 	struct bnxt_qplib_dev_attr *dev_attr;
1739 	struct bnxt_qplib_nq *nq = NULL;
1740 	struct bnxt_re_ucontext *uctx;
1741 	struct bnxt_re_dev *rdev;
1742 	struct bnxt_re_srq *srq;
1743 	struct bnxt_re_pd *pd;
1744 	struct ib_pd *ib_pd;
1745 	u32 active_srqs;
1746 	int rc, entries;
1747 
1748 	ib_pd = ib_srq->pd;
1749 	pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
1750 	rdev = pd->rdev;
1751 	dev_attr = &rdev->dev_attr;
1752 	srq = container_of(ib_srq, struct bnxt_re_srq, ib_srq);
1753 
1754 	if (srq_init_attr->attr.max_wr >= dev_attr->max_srq_wqes) {
1755 		ibdev_err(&rdev->ibdev, "Create CQ failed - max exceeded");
1756 		rc = -EINVAL;
1757 		goto exit;
1758 	}
1759 
1760 	if (srq_init_attr->srq_type != IB_SRQT_BASIC) {
1761 		rc = -EOPNOTSUPP;
1762 		goto exit;
1763 	}
1764 
1765 	uctx = rdma_udata_to_drv_context(udata, struct bnxt_re_ucontext, ib_uctx);
1766 	srq->rdev = rdev;
1767 	srq->qplib_srq.pd = &pd->qplib_pd;
1768 	srq->qplib_srq.dpi = &rdev->dpi_privileged;
1769 	/* Allocate 1 more than what's provided so posting max doesn't
1770 	 * mean empty
1771 	 */
1772 	entries = bnxt_re_init_depth(srq_init_attr->attr.max_wr + 1, uctx);
1773 	if (entries > dev_attr->max_srq_wqes + 1)
1774 		entries = dev_attr->max_srq_wqes + 1;
1775 	srq->qplib_srq.max_wqe = entries;
1776 
1777 	srq->qplib_srq.max_sge = srq_init_attr->attr.max_sge;
1778 	 /* 128 byte wqe size for SRQ . So use max sges */
1779 	srq->qplib_srq.wqe_size = bnxt_re_get_rwqe_size(dev_attr->max_srq_sges);
1780 	srq->qplib_srq.threshold = srq_init_attr->attr.srq_limit;
1781 	srq->srq_limit = srq_init_attr->attr.srq_limit;
1782 	srq->qplib_srq.eventq_hw_ring_id = rdev->nq[0].ring_id;
1783 	nq = &rdev->nq[0];
1784 
1785 	if (udata) {
1786 		rc = bnxt_re_init_user_srq(rdev, pd, srq, udata);
1787 		if (rc)
1788 			goto fail;
1789 	}
1790 
1791 	rc = bnxt_qplib_create_srq(&rdev->qplib_res, &srq->qplib_srq);
1792 	if (rc) {
1793 		ibdev_err(&rdev->ibdev, "Create HW SRQ failed!");
1794 		goto fail;
1795 	}
1796 
1797 	if (udata) {
1798 		struct bnxt_re_srq_resp resp;
1799 
1800 		resp.srqid = srq->qplib_srq.id;
1801 		rc = ib_copy_to_udata(udata, &resp, sizeof(resp));
1802 		if (rc) {
1803 			ibdev_err(&rdev->ibdev, "SRQ copy to udata failed!");
1804 			bnxt_qplib_destroy_srq(&rdev->qplib_res,
1805 					       &srq->qplib_srq);
1806 			goto fail;
1807 		}
1808 	}
1809 	if (nq)
1810 		nq->budget++;
1811 	active_srqs = atomic_inc_return(&rdev->stats.res.srq_count);
1812 	if (active_srqs > rdev->stats.res.srq_watermark)
1813 		rdev->stats.res.srq_watermark = active_srqs;
1814 	spin_lock_init(&srq->lock);
1815 
1816 	return 0;
1817 
1818 fail:
1819 	ib_umem_release(srq->umem);
1820 exit:
1821 	return rc;
1822 }
1823 
1824 int bnxt_re_modify_srq(struct ib_srq *ib_srq, struct ib_srq_attr *srq_attr,
1825 		       enum ib_srq_attr_mask srq_attr_mask,
1826 		       struct ib_udata *udata)
1827 {
1828 	struct bnxt_re_srq *srq = container_of(ib_srq, struct bnxt_re_srq,
1829 					       ib_srq);
1830 	struct bnxt_re_dev *rdev = srq->rdev;
1831 	int rc;
1832 
1833 	switch (srq_attr_mask) {
1834 	case IB_SRQ_MAX_WR:
1835 		/* SRQ resize is not supported */
1836 		return -EINVAL;
1837 	case IB_SRQ_LIMIT:
1838 		/* Change the SRQ threshold */
1839 		if (srq_attr->srq_limit > srq->qplib_srq.max_wqe)
1840 			return -EINVAL;
1841 
1842 		srq->qplib_srq.threshold = srq_attr->srq_limit;
1843 		rc = bnxt_qplib_modify_srq(&rdev->qplib_res, &srq->qplib_srq);
1844 		if (rc) {
1845 			ibdev_err(&rdev->ibdev, "Modify HW SRQ failed!");
1846 			return rc;
1847 		}
1848 		/* On success, update the shadow */
1849 		srq->srq_limit = srq_attr->srq_limit;
1850 		/* No need to Build and send response back to udata */
1851 		return 0;
1852 	default:
1853 		ibdev_err(&rdev->ibdev,
1854 			  "Unsupported srq_attr_mask 0x%x", srq_attr_mask);
1855 		return -EINVAL;
1856 	}
1857 }
1858 
1859 int bnxt_re_query_srq(struct ib_srq *ib_srq, struct ib_srq_attr *srq_attr)
1860 {
1861 	struct bnxt_re_srq *srq = container_of(ib_srq, struct bnxt_re_srq,
1862 					       ib_srq);
1863 	struct bnxt_re_srq tsrq;
1864 	struct bnxt_re_dev *rdev = srq->rdev;
1865 	int rc;
1866 
1867 	/* Get live SRQ attr */
1868 	tsrq.qplib_srq.id = srq->qplib_srq.id;
1869 	rc = bnxt_qplib_query_srq(&rdev->qplib_res, &tsrq.qplib_srq);
1870 	if (rc) {
1871 		ibdev_err(&rdev->ibdev, "Query HW SRQ failed!");
1872 		return rc;
1873 	}
1874 	srq_attr->max_wr = srq->qplib_srq.max_wqe;
1875 	srq_attr->max_sge = srq->qplib_srq.max_sge;
1876 	srq_attr->srq_limit = tsrq.qplib_srq.threshold;
1877 
1878 	return 0;
1879 }
1880 
1881 int bnxt_re_post_srq_recv(struct ib_srq *ib_srq, const struct ib_recv_wr *wr,
1882 			  const struct ib_recv_wr **bad_wr)
1883 {
1884 	struct bnxt_re_srq *srq = container_of(ib_srq, struct bnxt_re_srq,
1885 					       ib_srq);
1886 	struct bnxt_qplib_swqe wqe;
1887 	unsigned long flags;
1888 	int rc = 0;
1889 
1890 	spin_lock_irqsave(&srq->lock, flags);
1891 	while (wr) {
1892 		/* Transcribe each ib_recv_wr to qplib_swqe */
1893 		wqe.num_sge = wr->num_sge;
1894 		bnxt_re_build_sgl(wr->sg_list, wqe.sg_list, wr->num_sge);
1895 		wqe.wr_id = wr->wr_id;
1896 		wqe.type = BNXT_QPLIB_SWQE_TYPE_RECV;
1897 
1898 		rc = bnxt_qplib_post_srq_recv(&srq->qplib_srq, &wqe);
1899 		if (rc) {
1900 			*bad_wr = wr;
1901 			break;
1902 		}
1903 		wr = wr->next;
1904 	}
1905 	spin_unlock_irqrestore(&srq->lock, flags);
1906 
1907 	return rc;
1908 }
1909 static int bnxt_re_modify_shadow_qp(struct bnxt_re_dev *rdev,
1910 				    struct bnxt_re_qp *qp1_qp,
1911 				    int qp_attr_mask)
1912 {
1913 	struct bnxt_re_qp *qp = rdev->gsi_ctx.gsi_sqp;
1914 	int rc;
1915 
1916 	if (qp_attr_mask & IB_QP_STATE) {
1917 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_STATE;
1918 		qp->qplib_qp.state = qp1_qp->qplib_qp.state;
1919 	}
1920 	if (qp_attr_mask & IB_QP_PKEY_INDEX) {
1921 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_PKEY;
1922 		qp->qplib_qp.pkey_index = qp1_qp->qplib_qp.pkey_index;
1923 	}
1924 
1925 	if (qp_attr_mask & IB_QP_QKEY) {
1926 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_QKEY;
1927 		/* Using a Random  QKEY */
1928 		qp->qplib_qp.qkey = 0x81818181;
1929 	}
1930 	if (qp_attr_mask & IB_QP_SQ_PSN) {
1931 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_SQ_PSN;
1932 		qp->qplib_qp.sq.psn = qp1_qp->qplib_qp.sq.psn;
1933 	}
1934 
1935 	rc = bnxt_qplib_modify_qp(&rdev->qplib_res, &qp->qplib_qp);
1936 	if (rc)
1937 		ibdev_err(&rdev->ibdev, "Failed to modify Shadow QP for QP1");
1938 	return rc;
1939 }
1940 
1941 int bnxt_re_modify_qp(struct ib_qp *ib_qp, struct ib_qp_attr *qp_attr,
1942 		      int qp_attr_mask, struct ib_udata *udata)
1943 {
1944 	struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
1945 	struct bnxt_re_dev *rdev = qp->rdev;
1946 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
1947 	enum ib_qp_state curr_qp_state, new_qp_state;
1948 	int rc, entries;
1949 	unsigned int flags;
1950 	u8 nw_type;
1951 
1952 	if (qp_attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
1953 		return -EOPNOTSUPP;
1954 
1955 	qp->qplib_qp.modify_flags = 0;
1956 	if (qp_attr_mask & IB_QP_STATE) {
1957 		curr_qp_state = __to_ib_qp_state(qp->qplib_qp.cur_qp_state);
1958 		new_qp_state = qp_attr->qp_state;
1959 		if (!ib_modify_qp_is_ok(curr_qp_state, new_qp_state,
1960 					ib_qp->qp_type, qp_attr_mask)) {
1961 			ibdev_err(&rdev->ibdev,
1962 				  "Invalid attribute mask: %#x specified ",
1963 				  qp_attr_mask);
1964 			ibdev_err(&rdev->ibdev,
1965 				  "for qpn: %#x type: %#x",
1966 				  ib_qp->qp_num, ib_qp->qp_type);
1967 			ibdev_err(&rdev->ibdev,
1968 				  "curr_qp_state=0x%x, new_qp_state=0x%x\n",
1969 				  curr_qp_state, new_qp_state);
1970 			return -EINVAL;
1971 		}
1972 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_STATE;
1973 		qp->qplib_qp.state = __from_ib_qp_state(qp_attr->qp_state);
1974 
1975 		if (!qp->sumem &&
1976 		    qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1977 			ibdev_dbg(&rdev->ibdev,
1978 				  "Move QP = %p to flush list\n", qp);
1979 			flags = bnxt_re_lock_cqs(qp);
1980 			bnxt_qplib_add_flush_qp(&qp->qplib_qp);
1981 			bnxt_re_unlock_cqs(qp, flags);
1982 		}
1983 		if (!qp->sumem &&
1984 		    qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_RESET) {
1985 			ibdev_dbg(&rdev->ibdev,
1986 				  "Move QP = %p out of flush list\n", qp);
1987 			flags = bnxt_re_lock_cqs(qp);
1988 			bnxt_qplib_clean_qp(&qp->qplib_qp);
1989 			bnxt_re_unlock_cqs(qp, flags);
1990 		}
1991 	}
1992 	if (qp_attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY) {
1993 		qp->qplib_qp.modify_flags |=
1994 				CMDQ_MODIFY_QP_MODIFY_MASK_EN_SQD_ASYNC_NOTIFY;
1995 		qp->qplib_qp.en_sqd_async_notify = true;
1996 	}
1997 	if (qp_attr_mask & IB_QP_ACCESS_FLAGS) {
1998 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_ACCESS;
1999 		qp->qplib_qp.access =
2000 			__from_ib_access_flags(qp_attr->qp_access_flags);
2001 		/* LOCAL_WRITE access must be set to allow RC receive */
2002 		qp->qplib_qp.access |= BNXT_QPLIB_ACCESS_LOCAL_WRITE;
2003 		/* Temp: Set all params on QP as of now */
2004 		qp->qplib_qp.access |= CMDQ_MODIFY_QP_ACCESS_REMOTE_WRITE;
2005 		qp->qplib_qp.access |= CMDQ_MODIFY_QP_ACCESS_REMOTE_READ;
2006 	}
2007 	if (qp_attr_mask & IB_QP_PKEY_INDEX) {
2008 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_PKEY;
2009 		qp->qplib_qp.pkey_index = qp_attr->pkey_index;
2010 	}
2011 	if (qp_attr_mask & IB_QP_QKEY) {
2012 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_QKEY;
2013 		qp->qplib_qp.qkey = qp_attr->qkey;
2014 	}
2015 	if (qp_attr_mask & IB_QP_AV) {
2016 		const struct ib_global_route *grh =
2017 			rdma_ah_read_grh(&qp_attr->ah_attr);
2018 		const struct ib_gid_attr *sgid_attr;
2019 		struct bnxt_re_gid_ctx *ctx;
2020 
2021 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_DGID |
2022 				     CMDQ_MODIFY_QP_MODIFY_MASK_FLOW_LABEL |
2023 				     CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX |
2024 				     CMDQ_MODIFY_QP_MODIFY_MASK_HOP_LIMIT |
2025 				     CMDQ_MODIFY_QP_MODIFY_MASK_TRAFFIC_CLASS |
2026 				     CMDQ_MODIFY_QP_MODIFY_MASK_DEST_MAC |
2027 				     CMDQ_MODIFY_QP_MODIFY_MASK_VLAN_ID;
2028 		memcpy(qp->qplib_qp.ah.dgid.data, grh->dgid.raw,
2029 		       sizeof(qp->qplib_qp.ah.dgid.data));
2030 		qp->qplib_qp.ah.flow_label = grh->flow_label;
2031 		sgid_attr = grh->sgid_attr;
2032 		/* Get the HW context of the GID. The reference
2033 		 * of GID table entry is already taken by the caller.
2034 		 */
2035 		ctx = rdma_read_gid_hw_context(sgid_attr);
2036 		qp->qplib_qp.ah.sgid_index = ctx->idx;
2037 		qp->qplib_qp.ah.host_sgid_index = grh->sgid_index;
2038 		qp->qplib_qp.ah.hop_limit = grh->hop_limit;
2039 		qp->qplib_qp.ah.traffic_class = grh->traffic_class;
2040 		qp->qplib_qp.ah.sl = rdma_ah_get_sl(&qp_attr->ah_attr);
2041 		ether_addr_copy(qp->qplib_qp.ah.dmac,
2042 				qp_attr->ah_attr.roce.dmac);
2043 
2044 		rc = rdma_read_gid_l2_fields(sgid_attr, NULL,
2045 					     &qp->qplib_qp.smac[0]);
2046 		if (rc)
2047 			return rc;
2048 
2049 		nw_type = rdma_gid_attr_network_type(sgid_attr);
2050 		switch (nw_type) {
2051 		case RDMA_NETWORK_IPV4:
2052 			qp->qplib_qp.nw_type =
2053 				CMDQ_MODIFY_QP_NETWORK_TYPE_ROCEV2_IPV4;
2054 			break;
2055 		case RDMA_NETWORK_IPV6:
2056 			qp->qplib_qp.nw_type =
2057 				CMDQ_MODIFY_QP_NETWORK_TYPE_ROCEV2_IPV6;
2058 			break;
2059 		default:
2060 			qp->qplib_qp.nw_type =
2061 				CMDQ_MODIFY_QP_NETWORK_TYPE_ROCEV1;
2062 			break;
2063 		}
2064 	}
2065 
2066 	if (qp_attr_mask & IB_QP_PATH_MTU) {
2067 		qp->qplib_qp.modify_flags |=
2068 				CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU;
2069 		qp->qplib_qp.path_mtu = __from_ib_mtu(qp_attr->path_mtu);
2070 		qp->qplib_qp.mtu = ib_mtu_enum_to_int(qp_attr->path_mtu);
2071 	} else if (qp_attr->qp_state == IB_QPS_RTR) {
2072 		qp->qplib_qp.modify_flags |=
2073 			CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU;
2074 		qp->qplib_qp.path_mtu =
2075 			__from_ib_mtu(iboe_get_mtu(rdev->netdev->mtu));
2076 		qp->qplib_qp.mtu =
2077 			ib_mtu_enum_to_int(iboe_get_mtu(rdev->netdev->mtu));
2078 	}
2079 
2080 	if (qp_attr_mask & IB_QP_TIMEOUT) {
2081 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_TIMEOUT;
2082 		qp->qplib_qp.timeout = qp_attr->timeout;
2083 	}
2084 	if (qp_attr_mask & IB_QP_RETRY_CNT) {
2085 		qp->qplib_qp.modify_flags |=
2086 				CMDQ_MODIFY_QP_MODIFY_MASK_RETRY_CNT;
2087 		qp->qplib_qp.retry_cnt = qp_attr->retry_cnt;
2088 	}
2089 	if (qp_attr_mask & IB_QP_RNR_RETRY) {
2090 		qp->qplib_qp.modify_flags |=
2091 				CMDQ_MODIFY_QP_MODIFY_MASK_RNR_RETRY;
2092 		qp->qplib_qp.rnr_retry = qp_attr->rnr_retry;
2093 	}
2094 	if (qp_attr_mask & IB_QP_MIN_RNR_TIMER) {
2095 		qp->qplib_qp.modify_flags |=
2096 				CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER;
2097 		qp->qplib_qp.min_rnr_timer = qp_attr->min_rnr_timer;
2098 	}
2099 	if (qp_attr_mask & IB_QP_RQ_PSN) {
2100 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_RQ_PSN;
2101 		qp->qplib_qp.rq.psn = qp_attr->rq_psn;
2102 	}
2103 	if (qp_attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
2104 		qp->qplib_qp.modify_flags |=
2105 				CMDQ_MODIFY_QP_MODIFY_MASK_MAX_RD_ATOMIC;
2106 		/* Cap the max_rd_atomic to device max */
2107 		qp->qplib_qp.max_rd_atomic = min_t(u32, qp_attr->max_rd_atomic,
2108 						   dev_attr->max_qp_rd_atom);
2109 	}
2110 	if (qp_attr_mask & IB_QP_SQ_PSN) {
2111 		qp->qplib_qp.modify_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_SQ_PSN;
2112 		qp->qplib_qp.sq.psn = qp_attr->sq_psn;
2113 	}
2114 	if (qp_attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
2115 		if (qp_attr->max_dest_rd_atomic >
2116 		    dev_attr->max_qp_init_rd_atom) {
2117 			ibdev_err(&rdev->ibdev,
2118 				  "max_dest_rd_atomic requested%d is > dev_max%d",
2119 				  qp_attr->max_dest_rd_atomic,
2120 				  dev_attr->max_qp_init_rd_atom);
2121 			return -EINVAL;
2122 		}
2123 
2124 		qp->qplib_qp.modify_flags |=
2125 				CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC;
2126 		qp->qplib_qp.max_dest_rd_atomic = qp_attr->max_dest_rd_atomic;
2127 	}
2128 	if (qp_attr_mask & IB_QP_CAP) {
2129 		struct bnxt_re_ucontext *uctx =
2130 			rdma_udata_to_drv_context(udata, struct bnxt_re_ucontext, ib_uctx);
2131 
2132 		qp->qplib_qp.modify_flags |=
2133 				CMDQ_MODIFY_QP_MODIFY_MASK_SQ_SIZE |
2134 				CMDQ_MODIFY_QP_MODIFY_MASK_RQ_SIZE |
2135 				CMDQ_MODIFY_QP_MODIFY_MASK_SQ_SGE |
2136 				CMDQ_MODIFY_QP_MODIFY_MASK_RQ_SGE |
2137 				CMDQ_MODIFY_QP_MODIFY_MASK_MAX_INLINE_DATA;
2138 		if ((qp_attr->cap.max_send_wr >= dev_attr->max_qp_wqes) ||
2139 		    (qp_attr->cap.max_recv_wr >= dev_attr->max_qp_wqes) ||
2140 		    (qp_attr->cap.max_send_sge >= dev_attr->max_qp_sges) ||
2141 		    (qp_attr->cap.max_recv_sge >= dev_attr->max_qp_sges) ||
2142 		    (qp_attr->cap.max_inline_data >=
2143 						dev_attr->max_inline_data)) {
2144 			ibdev_err(&rdev->ibdev,
2145 				  "Create QP failed - max exceeded");
2146 			return -EINVAL;
2147 		}
2148 		entries = bnxt_re_init_depth(qp_attr->cap.max_send_wr, uctx);
2149 		qp->qplib_qp.sq.max_wqe = min_t(u32, entries,
2150 						dev_attr->max_qp_wqes + 1);
2151 		qp->qplib_qp.sq.q_full_delta = qp->qplib_qp.sq.max_wqe -
2152 						qp_attr->cap.max_send_wr;
2153 		/*
2154 		 * Reserving one slot for Phantom WQE. Some application can
2155 		 * post one extra entry in this case. Allowing this to avoid
2156 		 * unexpected Queue full condition
2157 		 */
2158 		qp->qplib_qp.sq.q_full_delta -= 1;
2159 		qp->qplib_qp.sq.max_sge = qp_attr->cap.max_send_sge;
2160 		if (qp->qplib_qp.rq.max_wqe) {
2161 			entries = bnxt_re_init_depth(qp_attr->cap.max_recv_wr, uctx);
2162 			qp->qplib_qp.rq.max_wqe =
2163 				min_t(u32, entries, dev_attr->max_qp_wqes + 1);
2164 			qp->qplib_qp.rq.max_sw_wqe = qp->qplib_qp.rq.max_wqe;
2165 			qp->qplib_qp.rq.q_full_delta = qp->qplib_qp.rq.max_wqe -
2166 						       qp_attr->cap.max_recv_wr;
2167 			qp->qplib_qp.rq.max_sge = qp_attr->cap.max_recv_sge;
2168 		} else {
2169 			/* SRQ was used prior, just ignore the RQ caps */
2170 		}
2171 	}
2172 	if (qp_attr_mask & IB_QP_DEST_QPN) {
2173 		qp->qplib_qp.modify_flags |=
2174 				CMDQ_MODIFY_QP_MODIFY_MASK_DEST_QP_ID;
2175 		qp->qplib_qp.dest_qpn = qp_attr->dest_qp_num;
2176 	}
2177 	rc = bnxt_qplib_modify_qp(&rdev->qplib_res, &qp->qplib_qp);
2178 	if (rc) {
2179 		ibdev_err(&rdev->ibdev, "Failed to modify HW QP");
2180 		return rc;
2181 	}
2182 	if (ib_qp->qp_type == IB_QPT_GSI && rdev->gsi_ctx.gsi_sqp)
2183 		rc = bnxt_re_modify_shadow_qp(rdev, qp, qp_attr_mask);
2184 	return rc;
2185 }
2186 
2187 int bnxt_re_query_qp(struct ib_qp *ib_qp, struct ib_qp_attr *qp_attr,
2188 		     int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr)
2189 {
2190 	struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
2191 	struct bnxt_re_dev *rdev = qp->rdev;
2192 	struct bnxt_qplib_qp *qplib_qp;
2193 	int rc;
2194 
2195 	qplib_qp = kzalloc(sizeof(*qplib_qp), GFP_KERNEL);
2196 	if (!qplib_qp)
2197 		return -ENOMEM;
2198 
2199 	qplib_qp->id = qp->qplib_qp.id;
2200 	qplib_qp->ah.host_sgid_index = qp->qplib_qp.ah.host_sgid_index;
2201 
2202 	rc = bnxt_qplib_query_qp(&rdev->qplib_res, qplib_qp);
2203 	if (rc) {
2204 		ibdev_err(&rdev->ibdev, "Failed to query HW QP");
2205 		goto out;
2206 	}
2207 	qp_attr->qp_state = __to_ib_qp_state(qplib_qp->state);
2208 	qp_attr->cur_qp_state = __to_ib_qp_state(qplib_qp->cur_qp_state);
2209 	qp_attr->en_sqd_async_notify = qplib_qp->en_sqd_async_notify ? 1 : 0;
2210 	qp_attr->qp_access_flags = __to_ib_access_flags(qplib_qp->access);
2211 	qp_attr->pkey_index = qplib_qp->pkey_index;
2212 	qp_attr->qkey = qplib_qp->qkey;
2213 	qp_attr->ah_attr.type = RDMA_AH_ATTR_TYPE_ROCE;
2214 	rdma_ah_set_grh(&qp_attr->ah_attr, NULL, qplib_qp->ah.flow_label,
2215 			qplib_qp->ah.host_sgid_index,
2216 			qplib_qp->ah.hop_limit,
2217 			qplib_qp->ah.traffic_class);
2218 	rdma_ah_set_dgid_raw(&qp_attr->ah_attr, qplib_qp->ah.dgid.data);
2219 	rdma_ah_set_sl(&qp_attr->ah_attr, qplib_qp->ah.sl);
2220 	ether_addr_copy(qp_attr->ah_attr.roce.dmac, qplib_qp->ah.dmac);
2221 	qp_attr->path_mtu = __to_ib_mtu(qplib_qp->path_mtu);
2222 	qp_attr->timeout = qplib_qp->timeout;
2223 	qp_attr->retry_cnt = qplib_qp->retry_cnt;
2224 	qp_attr->rnr_retry = qplib_qp->rnr_retry;
2225 	qp_attr->min_rnr_timer = qplib_qp->min_rnr_timer;
2226 	qp_attr->rq_psn = qplib_qp->rq.psn;
2227 	qp_attr->max_rd_atomic = qplib_qp->max_rd_atomic;
2228 	qp_attr->sq_psn = qplib_qp->sq.psn;
2229 	qp_attr->max_dest_rd_atomic = qplib_qp->max_dest_rd_atomic;
2230 	qp_init_attr->sq_sig_type = qplib_qp->sig_type ? IB_SIGNAL_ALL_WR :
2231 							 IB_SIGNAL_REQ_WR;
2232 	qp_attr->dest_qp_num = qplib_qp->dest_qpn;
2233 
2234 	qp_attr->cap.max_send_wr = qp->qplib_qp.sq.max_wqe;
2235 	qp_attr->cap.max_send_sge = qp->qplib_qp.sq.max_sge;
2236 	qp_attr->cap.max_recv_wr = qp->qplib_qp.rq.max_wqe;
2237 	qp_attr->cap.max_recv_sge = qp->qplib_qp.rq.max_sge;
2238 	qp_attr->cap.max_inline_data = qp->qplib_qp.max_inline_data;
2239 	qp_init_attr->cap = qp_attr->cap;
2240 
2241 out:
2242 	kfree(qplib_qp);
2243 	return rc;
2244 }
2245 
2246 /* Routine for sending QP1 packets for RoCE V1 an V2
2247  */
2248 static int bnxt_re_build_qp1_send_v2(struct bnxt_re_qp *qp,
2249 				     const struct ib_send_wr *wr,
2250 				     struct bnxt_qplib_swqe *wqe,
2251 				     int payload_size)
2252 {
2253 	struct bnxt_re_ah *ah = container_of(ud_wr(wr)->ah, struct bnxt_re_ah,
2254 					     ib_ah);
2255 	struct bnxt_qplib_ah *qplib_ah = &ah->qplib_ah;
2256 	const struct ib_gid_attr *sgid_attr = ah->ib_ah.sgid_attr;
2257 	struct bnxt_qplib_sge sge;
2258 	u8 nw_type;
2259 	u16 ether_type;
2260 	union ib_gid dgid;
2261 	bool is_eth = false;
2262 	bool is_vlan = false;
2263 	bool is_grh = false;
2264 	bool is_udp = false;
2265 	u8 ip_version = 0;
2266 	u16 vlan_id = 0xFFFF;
2267 	void *buf;
2268 	int i, rc;
2269 
2270 	memset(&qp->qp1_hdr, 0, sizeof(qp->qp1_hdr));
2271 
2272 	rc = rdma_read_gid_l2_fields(sgid_attr, &vlan_id, NULL);
2273 	if (rc)
2274 		return rc;
2275 
2276 	/* Get network header type for this GID */
2277 	nw_type = rdma_gid_attr_network_type(sgid_attr);
2278 	switch (nw_type) {
2279 	case RDMA_NETWORK_IPV4:
2280 		nw_type = BNXT_RE_ROCEV2_IPV4_PACKET;
2281 		break;
2282 	case RDMA_NETWORK_IPV6:
2283 		nw_type = BNXT_RE_ROCEV2_IPV6_PACKET;
2284 		break;
2285 	default:
2286 		nw_type = BNXT_RE_ROCE_V1_PACKET;
2287 		break;
2288 	}
2289 	memcpy(&dgid.raw, &qplib_ah->dgid, 16);
2290 	is_udp = sgid_attr->gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP;
2291 	if (is_udp) {
2292 		if (ipv6_addr_v4mapped((struct in6_addr *)&sgid_attr->gid)) {
2293 			ip_version = 4;
2294 			ether_type = ETH_P_IP;
2295 		} else {
2296 			ip_version = 6;
2297 			ether_type = ETH_P_IPV6;
2298 		}
2299 		is_grh = false;
2300 	} else {
2301 		ether_type = ETH_P_IBOE;
2302 		is_grh = true;
2303 	}
2304 
2305 	is_eth = true;
2306 	is_vlan = vlan_id && (vlan_id < 0x1000);
2307 
2308 	ib_ud_header_init(payload_size, !is_eth, is_eth, is_vlan, is_grh,
2309 			  ip_version, is_udp, 0, &qp->qp1_hdr);
2310 
2311 	/* ETH */
2312 	ether_addr_copy(qp->qp1_hdr.eth.dmac_h, ah->qplib_ah.dmac);
2313 	ether_addr_copy(qp->qp1_hdr.eth.smac_h, qp->qplib_qp.smac);
2314 
2315 	/* For vlan, check the sgid for vlan existence */
2316 
2317 	if (!is_vlan) {
2318 		qp->qp1_hdr.eth.type = cpu_to_be16(ether_type);
2319 	} else {
2320 		qp->qp1_hdr.vlan.type = cpu_to_be16(ether_type);
2321 		qp->qp1_hdr.vlan.tag = cpu_to_be16(vlan_id);
2322 	}
2323 
2324 	if (is_grh || (ip_version == 6)) {
2325 		memcpy(qp->qp1_hdr.grh.source_gid.raw, sgid_attr->gid.raw,
2326 		       sizeof(sgid_attr->gid));
2327 		memcpy(qp->qp1_hdr.grh.destination_gid.raw, qplib_ah->dgid.data,
2328 		       sizeof(sgid_attr->gid));
2329 		qp->qp1_hdr.grh.hop_limit     = qplib_ah->hop_limit;
2330 	}
2331 
2332 	if (ip_version == 4) {
2333 		qp->qp1_hdr.ip4.tos = 0;
2334 		qp->qp1_hdr.ip4.id = 0;
2335 		qp->qp1_hdr.ip4.frag_off = htons(IP_DF);
2336 		qp->qp1_hdr.ip4.ttl = qplib_ah->hop_limit;
2337 
2338 		memcpy(&qp->qp1_hdr.ip4.saddr, sgid_attr->gid.raw + 12, 4);
2339 		memcpy(&qp->qp1_hdr.ip4.daddr, qplib_ah->dgid.data + 12, 4);
2340 		qp->qp1_hdr.ip4.check = ib_ud_ip4_csum(&qp->qp1_hdr);
2341 	}
2342 
2343 	if (is_udp) {
2344 		qp->qp1_hdr.udp.dport = htons(ROCE_V2_UDP_DPORT);
2345 		qp->qp1_hdr.udp.sport = htons(0x8CD1);
2346 		qp->qp1_hdr.udp.csum = 0;
2347 	}
2348 
2349 	/* BTH */
2350 	if (wr->opcode == IB_WR_SEND_WITH_IMM) {
2351 		qp->qp1_hdr.bth.opcode = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE;
2352 		qp->qp1_hdr.immediate_present = 1;
2353 	} else {
2354 		qp->qp1_hdr.bth.opcode = IB_OPCODE_UD_SEND_ONLY;
2355 	}
2356 	if (wr->send_flags & IB_SEND_SOLICITED)
2357 		qp->qp1_hdr.bth.solicited_event = 1;
2358 	/* pad_count */
2359 	qp->qp1_hdr.bth.pad_count = (4 - payload_size) & 3;
2360 
2361 	/* P_key for QP1 is for all members */
2362 	qp->qp1_hdr.bth.pkey = cpu_to_be16(0xFFFF);
2363 	qp->qp1_hdr.bth.destination_qpn = IB_QP1;
2364 	qp->qp1_hdr.bth.ack_req = 0;
2365 	qp->send_psn++;
2366 	qp->send_psn &= BTH_PSN_MASK;
2367 	qp->qp1_hdr.bth.psn = cpu_to_be32(qp->send_psn);
2368 	/* DETH */
2369 	/* Use the priviledged Q_Key for QP1 */
2370 	qp->qp1_hdr.deth.qkey = cpu_to_be32(IB_QP1_QKEY);
2371 	qp->qp1_hdr.deth.source_qpn = IB_QP1;
2372 
2373 	/* Pack the QP1 to the transmit buffer */
2374 	buf = bnxt_qplib_get_qp1_sq_buf(&qp->qplib_qp, &sge);
2375 	if (buf) {
2376 		ib_ud_header_pack(&qp->qp1_hdr, buf);
2377 		for (i = wqe->num_sge; i; i--) {
2378 			wqe->sg_list[i].addr = wqe->sg_list[i - 1].addr;
2379 			wqe->sg_list[i].lkey = wqe->sg_list[i - 1].lkey;
2380 			wqe->sg_list[i].size = wqe->sg_list[i - 1].size;
2381 		}
2382 
2383 		/*
2384 		 * Max Header buf size for IPV6 RoCE V2 is 86,
2385 		 * which is same as the QP1 SQ header buffer.
2386 		 * Header buf size for IPV4 RoCE V2 can be 66.
2387 		 * ETH(14) + VLAN(4)+ IP(20) + UDP (8) + BTH(20).
2388 		 * Subtract 20 bytes from QP1 SQ header buf size
2389 		 */
2390 		if (is_udp && ip_version == 4)
2391 			sge.size -= 20;
2392 		/*
2393 		 * Max Header buf size for RoCE V1 is 78.
2394 		 * ETH(14) + VLAN(4) + GRH(40) + BTH(20).
2395 		 * Subtract 8 bytes from QP1 SQ header buf size
2396 		 */
2397 		if (!is_udp)
2398 			sge.size -= 8;
2399 
2400 		/* Subtract 4 bytes for non vlan packets */
2401 		if (!is_vlan)
2402 			sge.size -= 4;
2403 
2404 		wqe->sg_list[0].addr = sge.addr;
2405 		wqe->sg_list[0].lkey = sge.lkey;
2406 		wqe->sg_list[0].size = sge.size;
2407 		wqe->num_sge++;
2408 
2409 	} else {
2410 		ibdev_err(&qp->rdev->ibdev, "QP1 buffer is empty!");
2411 		rc = -ENOMEM;
2412 	}
2413 	return rc;
2414 }
2415 
2416 /* For the MAD layer, it only provides the recv SGE the size of
2417  * ib_grh + MAD datagram.  No Ethernet headers, Ethertype, BTH, DETH,
2418  * nor RoCE iCRC.  The Cu+ solution must provide buffer for the entire
2419  * receive packet (334 bytes) with no VLAN and then copy the GRH
2420  * and the MAD datagram out to the provided SGE.
2421  */
2422 static int bnxt_re_build_qp1_shadow_qp_recv(struct bnxt_re_qp *qp,
2423 					    const struct ib_recv_wr *wr,
2424 					    struct bnxt_qplib_swqe *wqe,
2425 					    int payload_size)
2426 {
2427 	struct bnxt_re_sqp_entries *sqp_entry;
2428 	struct bnxt_qplib_sge ref, sge;
2429 	struct bnxt_re_dev *rdev;
2430 	u32 rq_prod_index;
2431 
2432 	rdev = qp->rdev;
2433 
2434 	rq_prod_index = bnxt_qplib_get_rq_prod_index(&qp->qplib_qp);
2435 
2436 	if (!bnxt_qplib_get_qp1_rq_buf(&qp->qplib_qp, &sge))
2437 		return -ENOMEM;
2438 
2439 	/* Create 1 SGE to receive the entire
2440 	 * ethernet packet
2441 	 */
2442 	/* Save the reference from ULP */
2443 	ref.addr = wqe->sg_list[0].addr;
2444 	ref.lkey = wqe->sg_list[0].lkey;
2445 	ref.size = wqe->sg_list[0].size;
2446 
2447 	sqp_entry = &rdev->gsi_ctx.sqp_tbl[rq_prod_index];
2448 
2449 	/* SGE 1 */
2450 	wqe->sg_list[0].addr = sge.addr;
2451 	wqe->sg_list[0].lkey = sge.lkey;
2452 	wqe->sg_list[0].size = BNXT_QPLIB_MAX_QP1_RQ_HDR_SIZE_V2;
2453 	sge.size -= wqe->sg_list[0].size;
2454 
2455 	sqp_entry->sge.addr = ref.addr;
2456 	sqp_entry->sge.lkey = ref.lkey;
2457 	sqp_entry->sge.size = ref.size;
2458 	/* Store the wrid for reporting completion */
2459 	sqp_entry->wrid = wqe->wr_id;
2460 	/* change the wqe->wrid to table index */
2461 	wqe->wr_id = rq_prod_index;
2462 	return 0;
2463 }
2464 
2465 static int is_ud_qp(struct bnxt_re_qp *qp)
2466 {
2467 	return (qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_UD ||
2468 		qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_GSI);
2469 }
2470 
2471 static int bnxt_re_build_send_wqe(struct bnxt_re_qp *qp,
2472 				  const struct ib_send_wr *wr,
2473 				  struct bnxt_qplib_swqe *wqe)
2474 {
2475 	struct bnxt_re_ah *ah = NULL;
2476 
2477 	if (is_ud_qp(qp)) {
2478 		ah = container_of(ud_wr(wr)->ah, struct bnxt_re_ah, ib_ah);
2479 		wqe->send.q_key = ud_wr(wr)->remote_qkey;
2480 		wqe->send.dst_qp = ud_wr(wr)->remote_qpn;
2481 		wqe->send.avid = ah->qplib_ah.id;
2482 	}
2483 	switch (wr->opcode) {
2484 	case IB_WR_SEND:
2485 		wqe->type = BNXT_QPLIB_SWQE_TYPE_SEND;
2486 		break;
2487 	case IB_WR_SEND_WITH_IMM:
2488 		wqe->type = BNXT_QPLIB_SWQE_TYPE_SEND_WITH_IMM;
2489 		wqe->send.imm_data = be32_to_cpu(wr->ex.imm_data);
2490 		break;
2491 	case IB_WR_SEND_WITH_INV:
2492 		wqe->type = BNXT_QPLIB_SWQE_TYPE_SEND_WITH_INV;
2493 		wqe->send.inv_key = wr->ex.invalidate_rkey;
2494 		break;
2495 	default:
2496 		return -EINVAL;
2497 	}
2498 	if (wr->send_flags & IB_SEND_SIGNALED)
2499 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP;
2500 	if (wr->send_flags & IB_SEND_FENCE)
2501 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE;
2502 	if (wr->send_flags & IB_SEND_SOLICITED)
2503 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT;
2504 	if (wr->send_flags & IB_SEND_INLINE)
2505 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_INLINE;
2506 
2507 	return 0;
2508 }
2509 
2510 static int bnxt_re_build_rdma_wqe(const struct ib_send_wr *wr,
2511 				  struct bnxt_qplib_swqe *wqe)
2512 {
2513 	switch (wr->opcode) {
2514 	case IB_WR_RDMA_WRITE:
2515 		wqe->type = BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE;
2516 		break;
2517 	case IB_WR_RDMA_WRITE_WITH_IMM:
2518 		wqe->type = BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE_WITH_IMM;
2519 		wqe->rdma.imm_data = be32_to_cpu(wr->ex.imm_data);
2520 		break;
2521 	case IB_WR_RDMA_READ:
2522 		wqe->type = BNXT_QPLIB_SWQE_TYPE_RDMA_READ;
2523 		wqe->rdma.inv_key = wr->ex.invalidate_rkey;
2524 		break;
2525 	default:
2526 		return -EINVAL;
2527 	}
2528 	wqe->rdma.remote_va = rdma_wr(wr)->remote_addr;
2529 	wqe->rdma.r_key = rdma_wr(wr)->rkey;
2530 	if (wr->send_flags & IB_SEND_SIGNALED)
2531 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP;
2532 	if (wr->send_flags & IB_SEND_FENCE)
2533 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE;
2534 	if (wr->send_flags & IB_SEND_SOLICITED)
2535 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT;
2536 	if (wr->send_flags & IB_SEND_INLINE)
2537 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_INLINE;
2538 
2539 	return 0;
2540 }
2541 
2542 static int bnxt_re_build_atomic_wqe(const struct ib_send_wr *wr,
2543 				    struct bnxt_qplib_swqe *wqe)
2544 {
2545 	switch (wr->opcode) {
2546 	case IB_WR_ATOMIC_CMP_AND_SWP:
2547 		wqe->type = BNXT_QPLIB_SWQE_TYPE_ATOMIC_CMP_AND_SWP;
2548 		wqe->atomic.cmp_data = atomic_wr(wr)->compare_add;
2549 		wqe->atomic.swap_data = atomic_wr(wr)->swap;
2550 		break;
2551 	case IB_WR_ATOMIC_FETCH_AND_ADD:
2552 		wqe->type = BNXT_QPLIB_SWQE_TYPE_ATOMIC_FETCH_AND_ADD;
2553 		wqe->atomic.cmp_data = atomic_wr(wr)->compare_add;
2554 		break;
2555 	default:
2556 		return -EINVAL;
2557 	}
2558 	wqe->atomic.remote_va = atomic_wr(wr)->remote_addr;
2559 	wqe->atomic.r_key = atomic_wr(wr)->rkey;
2560 	if (wr->send_flags & IB_SEND_SIGNALED)
2561 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP;
2562 	if (wr->send_flags & IB_SEND_FENCE)
2563 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE;
2564 	if (wr->send_flags & IB_SEND_SOLICITED)
2565 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT;
2566 	return 0;
2567 }
2568 
2569 static int bnxt_re_build_inv_wqe(const struct ib_send_wr *wr,
2570 				 struct bnxt_qplib_swqe *wqe)
2571 {
2572 	wqe->type = BNXT_QPLIB_SWQE_TYPE_LOCAL_INV;
2573 	wqe->local_inv.inv_l_key = wr->ex.invalidate_rkey;
2574 
2575 	if (wr->send_flags & IB_SEND_SIGNALED)
2576 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP;
2577 	if (wr->send_flags & IB_SEND_SOLICITED)
2578 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SOLICIT_EVENT;
2579 
2580 	return 0;
2581 }
2582 
2583 static int bnxt_re_build_reg_wqe(const struct ib_reg_wr *wr,
2584 				 struct bnxt_qplib_swqe *wqe)
2585 {
2586 	struct bnxt_re_mr *mr = container_of(wr->mr, struct bnxt_re_mr, ib_mr);
2587 	struct bnxt_qplib_frpl *qplib_frpl = &mr->qplib_frpl;
2588 	int access = wr->access;
2589 
2590 	wqe->frmr.pbl_ptr = (__le64 *)qplib_frpl->hwq.pbl_ptr[0];
2591 	wqe->frmr.pbl_dma_ptr = qplib_frpl->hwq.pbl_dma_ptr[0];
2592 	wqe->frmr.page_list = mr->pages;
2593 	wqe->frmr.page_list_len = mr->npages;
2594 	wqe->frmr.levels = qplib_frpl->hwq.level;
2595 	wqe->type = BNXT_QPLIB_SWQE_TYPE_REG_MR;
2596 
2597 	if (wr->wr.send_flags & IB_SEND_SIGNALED)
2598 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_SIGNAL_COMP;
2599 
2600 	if (access & IB_ACCESS_LOCAL_WRITE)
2601 		wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_LOCAL_WRITE;
2602 	if (access & IB_ACCESS_REMOTE_READ)
2603 		wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_REMOTE_READ;
2604 	if (access & IB_ACCESS_REMOTE_WRITE)
2605 		wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_REMOTE_WRITE;
2606 	if (access & IB_ACCESS_REMOTE_ATOMIC)
2607 		wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_REMOTE_ATOMIC;
2608 	if (access & IB_ACCESS_MW_BIND)
2609 		wqe->frmr.access_cntl |= SQ_FR_PMR_ACCESS_CNTL_WINDOW_BIND;
2610 
2611 	wqe->frmr.l_key = wr->key;
2612 	wqe->frmr.length = wr->mr->length;
2613 	wqe->frmr.pbl_pg_sz_log = ilog2(PAGE_SIZE >> PAGE_SHIFT_4K);
2614 	wqe->frmr.pg_sz_log = ilog2(wr->mr->page_size >> PAGE_SHIFT_4K);
2615 	wqe->frmr.va = wr->mr->iova;
2616 	return 0;
2617 }
2618 
2619 static int bnxt_re_copy_inline_data(struct bnxt_re_dev *rdev,
2620 				    const struct ib_send_wr *wr,
2621 				    struct bnxt_qplib_swqe *wqe)
2622 {
2623 	/*  Copy the inline data to the data  field */
2624 	u8 *in_data;
2625 	u32 i, sge_len;
2626 	void *sge_addr;
2627 
2628 	in_data = wqe->inline_data;
2629 	for (i = 0; i < wr->num_sge; i++) {
2630 		sge_addr = (void *)(unsigned long)
2631 				wr->sg_list[i].addr;
2632 		sge_len = wr->sg_list[i].length;
2633 
2634 		if ((sge_len + wqe->inline_len) >
2635 		    BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH) {
2636 			ibdev_err(&rdev->ibdev,
2637 				  "Inline data size requested > supported value");
2638 			return -EINVAL;
2639 		}
2640 		sge_len = wr->sg_list[i].length;
2641 
2642 		memcpy(in_data, sge_addr, sge_len);
2643 		in_data += wr->sg_list[i].length;
2644 		wqe->inline_len += wr->sg_list[i].length;
2645 	}
2646 	return wqe->inline_len;
2647 }
2648 
2649 static int bnxt_re_copy_wr_payload(struct bnxt_re_dev *rdev,
2650 				   const struct ib_send_wr *wr,
2651 				   struct bnxt_qplib_swqe *wqe)
2652 {
2653 	int payload_sz = 0;
2654 
2655 	if (wr->send_flags & IB_SEND_INLINE)
2656 		payload_sz = bnxt_re_copy_inline_data(rdev, wr, wqe);
2657 	else
2658 		payload_sz = bnxt_re_build_sgl(wr->sg_list, wqe->sg_list,
2659 					       wqe->num_sge);
2660 
2661 	return payload_sz;
2662 }
2663 
2664 static void bnxt_ud_qp_hw_stall_workaround(struct bnxt_re_qp *qp)
2665 {
2666 	if ((qp->ib_qp.qp_type == IB_QPT_UD ||
2667 	     qp->ib_qp.qp_type == IB_QPT_GSI ||
2668 	     qp->ib_qp.qp_type == IB_QPT_RAW_ETHERTYPE) &&
2669 	     qp->qplib_qp.wqe_cnt == BNXT_RE_UD_QP_HW_STALL) {
2670 		int qp_attr_mask;
2671 		struct ib_qp_attr qp_attr;
2672 
2673 		qp_attr_mask = IB_QP_STATE;
2674 		qp_attr.qp_state = IB_QPS_RTS;
2675 		bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, qp_attr_mask, NULL);
2676 		qp->qplib_qp.wqe_cnt = 0;
2677 	}
2678 }
2679 
2680 static int bnxt_re_post_send_shadow_qp(struct bnxt_re_dev *rdev,
2681 				       struct bnxt_re_qp *qp,
2682 				       const struct ib_send_wr *wr)
2683 {
2684 	int rc = 0, payload_sz = 0;
2685 	unsigned long flags;
2686 
2687 	spin_lock_irqsave(&qp->sq_lock, flags);
2688 	while (wr) {
2689 		struct bnxt_qplib_swqe wqe = {};
2690 
2691 		/* Common */
2692 		wqe.num_sge = wr->num_sge;
2693 		if (wr->num_sge > qp->qplib_qp.sq.max_sge) {
2694 			ibdev_err(&rdev->ibdev,
2695 				  "Limit exceeded for Send SGEs");
2696 			rc = -EINVAL;
2697 			goto bad;
2698 		}
2699 
2700 		payload_sz = bnxt_re_copy_wr_payload(qp->rdev, wr, &wqe);
2701 		if (payload_sz < 0) {
2702 			rc = -EINVAL;
2703 			goto bad;
2704 		}
2705 		wqe.wr_id = wr->wr_id;
2706 
2707 		wqe.type = BNXT_QPLIB_SWQE_TYPE_SEND;
2708 
2709 		rc = bnxt_re_build_send_wqe(qp, wr, &wqe);
2710 		if (!rc)
2711 			rc = bnxt_qplib_post_send(&qp->qplib_qp, &wqe);
2712 bad:
2713 		if (rc) {
2714 			ibdev_err(&rdev->ibdev,
2715 				  "Post send failed opcode = %#x rc = %d",
2716 				  wr->opcode, rc);
2717 			break;
2718 		}
2719 		wr = wr->next;
2720 	}
2721 	bnxt_qplib_post_send_db(&qp->qplib_qp);
2722 	bnxt_ud_qp_hw_stall_workaround(qp);
2723 	spin_unlock_irqrestore(&qp->sq_lock, flags);
2724 	return rc;
2725 }
2726 
2727 static void bnxt_re_legacy_set_uc_fence(struct bnxt_qplib_swqe *wqe)
2728 {
2729 	/* Need unconditional fence for non-wire memory opcode
2730 	 * to work as expected.
2731 	 */
2732 	if (wqe->type == BNXT_QPLIB_SWQE_TYPE_LOCAL_INV ||
2733 	    wqe->type == BNXT_QPLIB_SWQE_TYPE_FAST_REG_MR ||
2734 	    wqe->type == BNXT_QPLIB_SWQE_TYPE_REG_MR ||
2735 	    wqe->type == BNXT_QPLIB_SWQE_TYPE_BIND_MW)
2736 		wqe->flags |= BNXT_QPLIB_SWQE_FLAGS_UC_FENCE;
2737 }
2738 
2739 int bnxt_re_post_send(struct ib_qp *ib_qp, const struct ib_send_wr *wr,
2740 		      const struct ib_send_wr **bad_wr)
2741 {
2742 	struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
2743 	struct bnxt_qplib_swqe wqe;
2744 	int rc = 0, payload_sz = 0;
2745 	unsigned long flags;
2746 
2747 	spin_lock_irqsave(&qp->sq_lock, flags);
2748 	while (wr) {
2749 		/* House keeping */
2750 		memset(&wqe, 0, sizeof(wqe));
2751 
2752 		/* Common */
2753 		wqe.num_sge = wr->num_sge;
2754 		if (wr->num_sge > qp->qplib_qp.sq.max_sge) {
2755 			ibdev_err(&qp->rdev->ibdev,
2756 				  "Limit exceeded for Send SGEs");
2757 			rc = -EINVAL;
2758 			goto bad;
2759 		}
2760 
2761 		payload_sz = bnxt_re_copy_wr_payload(qp->rdev, wr, &wqe);
2762 		if (payload_sz < 0) {
2763 			rc = -EINVAL;
2764 			goto bad;
2765 		}
2766 		wqe.wr_id = wr->wr_id;
2767 
2768 		switch (wr->opcode) {
2769 		case IB_WR_SEND:
2770 		case IB_WR_SEND_WITH_IMM:
2771 			if (qp->qplib_qp.type == CMDQ_CREATE_QP1_TYPE_GSI) {
2772 				rc = bnxt_re_build_qp1_send_v2(qp, wr, &wqe,
2773 							       payload_sz);
2774 				if (rc)
2775 					goto bad;
2776 				wqe.rawqp1.lflags |=
2777 					SQ_SEND_RAWETH_QP1_LFLAGS_ROCE_CRC;
2778 			}
2779 			switch (wr->send_flags) {
2780 			case IB_SEND_IP_CSUM:
2781 				wqe.rawqp1.lflags |=
2782 					SQ_SEND_RAWETH_QP1_LFLAGS_IP_CHKSUM;
2783 				break;
2784 			default:
2785 				break;
2786 			}
2787 			fallthrough;
2788 		case IB_WR_SEND_WITH_INV:
2789 			rc = bnxt_re_build_send_wqe(qp, wr, &wqe);
2790 			break;
2791 		case IB_WR_RDMA_WRITE:
2792 		case IB_WR_RDMA_WRITE_WITH_IMM:
2793 		case IB_WR_RDMA_READ:
2794 			rc = bnxt_re_build_rdma_wqe(wr, &wqe);
2795 			break;
2796 		case IB_WR_ATOMIC_CMP_AND_SWP:
2797 		case IB_WR_ATOMIC_FETCH_AND_ADD:
2798 			rc = bnxt_re_build_atomic_wqe(wr, &wqe);
2799 			break;
2800 		case IB_WR_RDMA_READ_WITH_INV:
2801 			ibdev_err(&qp->rdev->ibdev,
2802 				  "RDMA Read with Invalidate is not supported");
2803 			rc = -EINVAL;
2804 			goto bad;
2805 		case IB_WR_LOCAL_INV:
2806 			rc = bnxt_re_build_inv_wqe(wr, &wqe);
2807 			break;
2808 		case IB_WR_REG_MR:
2809 			rc = bnxt_re_build_reg_wqe(reg_wr(wr), &wqe);
2810 			break;
2811 		default:
2812 			/* Unsupported WRs */
2813 			ibdev_err(&qp->rdev->ibdev,
2814 				  "WR (%#x) is not supported", wr->opcode);
2815 			rc = -EINVAL;
2816 			goto bad;
2817 		}
2818 		if (!rc) {
2819 			if (!bnxt_qplib_is_chip_gen_p5_p7(qp->rdev->chip_ctx))
2820 				bnxt_re_legacy_set_uc_fence(&wqe);
2821 			rc = bnxt_qplib_post_send(&qp->qplib_qp, &wqe);
2822 		}
2823 bad:
2824 		if (rc) {
2825 			ibdev_err(&qp->rdev->ibdev,
2826 				  "post_send failed op:%#x qps = %#x rc = %d\n",
2827 				  wr->opcode, qp->qplib_qp.state, rc);
2828 			*bad_wr = wr;
2829 			break;
2830 		}
2831 		wr = wr->next;
2832 	}
2833 	bnxt_qplib_post_send_db(&qp->qplib_qp);
2834 	bnxt_ud_qp_hw_stall_workaround(qp);
2835 	spin_unlock_irqrestore(&qp->sq_lock, flags);
2836 
2837 	return rc;
2838 }
2839 
2840 static int bnxt_re_post_recv_shadow_qp(struct bnxt_re_dev *rdev,
2841 				       struct bnxt_re_qp *qp,
2842 				       const struct ib_recv_wr *wr)
2843 {
2844 	struct bnxt_qplib_swqe wqe;
2845 	int rc = 0;
2846 
2847 	while (wr) {
2848 		/* House keeping */
2849 		memset(&wqe, 0, sizeof(wqe));
2850 
2851 		/* Common */
2852 		wqe.num_sge = wr->num_sge;
2853 		if (wr->num_sge > qp->qplib_qp.rq.max_sge) {
2854 			ibdev_err(&rdev->ibdev,
2855 				  "Limit exceeded for Receive SGEs");
2856 			rc = -EINVAL;
2857 			break;
2858 		}
2859 		bnxt_re_build_sgl(wr->sg_list, wqe.sg_list, wr->num_sge);
2860 		wqe.wr_id = wr->wr_id;
2861 		wqe.type = BNXT_QPLIB_SWQE_TYPE_RECV;
2862 
2863 		rc = bnxt_qplib_post_recv(&qp->qplib_qp, &wqe);
2864 		if (rc)
2865 			break;
2866 
2867 		wr = wr->next;
2868 	}
2869 	if (!rc)
2870 		bnxt_qplib_post_recv_db(&qp->qplib_qp);
2871 	return rc;
2872 }
2873 
2874 int bnxt_re_post_recv(struct ib_qp *ib_qp, const struct ib_recv_wr *wr,
2875 		      const struct ib_recv_wr **bad_wr)
2876 {
2877 	struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
2878 	struct bnxt_qplib_swqe wqe;
2879 	int rc = 0, payload_sz = 0;
2880 	unsigned long flags;
2881 	u32 count = 0;
2882 
2883 	spin_lock_irqsave(&qp->rq_lock, flags);
2884 	while (wr) {
2885 		/* House keeping */
2886 		memset(&wqe, 0, sizeof(wqe));
2887 
2888 		/* Common */
2889 		wqe.num_sge = wr->num_sge;
2890 		if (wr->num_sge > qp->qplib_qp.rq.max_sge) {
2891 			ibdev_err(&qp->rdev->ibdev,
2892 				  "Limit exceeded for Receive SGEs");
2893 			rc = -EINVAL;
2894 			*bad_wr = wr;
2895 			break;
2896 		}
2897 
2898 		payload_sz = bnxt_re_build_sgl(wr->sg_list, wqe.sg_list,
2899 					       wr->num_sge);
2900 		wqe.wr_id = wr->wr_id;
2901 		wqe.type = BNXT_QPLIB_SWQE_TYPE_RECV;
2902 
2903 		if (ib_qp->qp_type == IB_QPT_GSI &&
2904 		    qp->qplib_qp.type != CMDQ_CREATE_QP_TYPE_GSI)
2905 			rc = bnxt_re_build_qp1_shadow_qp_recv(qp, wr, &wqe,
2906 							      payload_sz);
2907 		if (!rc)
2908 			rc = bnxt_qplib_post_recv(&qp->qplib_qp, &wqe);
2909 		if (rc) {
2910 			*bad_wr = wr;
2911 			break;
2912 		}
2913 
2914 		/* Ring DB if the RQEs posted reaches a threshold value */
2915 		if (++count >= BNXT_RE_RQ_WQE_THRESHOLD) {
2916 			bnxt_qplib_post_recv_db(&qp->qplib_qp);
2917 			count = 0;
2918 		}
2919 
2920 		wr = wr->next;
2921 	}
2922 
2923 	if (count)
2924 		bnxt_qplib_post_recv_db(&qp->qplib_qp);
2925 
2926 	spin_unlock_irqrestore(&qp->rq_lock, flags);
2927 
2928 	return rc;
2929 }
2930 
2931 /* Completion Queues */
2932 int bnxt_re_destroy_cq(struct ib_cq *ib_cq, struct ib_udata *udata)
2933 {
2934 	struct bnxt_qplib_chip_ctx *cctx;
2935 	struct bnxt_qplib_nq *nq;
2936 	struct bnxt_re_dev *rdev;
2937 	struct bnxt_re_cq *cq;
2938 
2939 	cq = container_of(ib_cq, struct bnxt_re_cq, ib_cq);
2940 	rdev = cq->rdev;
2941 	nq = cq->qplib_cq.nq;
2942 	cctx = rdev->chip_ctx;
2943 
2944 	if (cctx->modes.toggle_bits & BNXT_QPLIB_CQ_TOGGLE_BIT) {
2945 		free_page((unsigned long)cq->uctx_cq_page);
2946 		hash_del(&cq->hash_entry);
2947 	}
2948 	bnxt_qplib_destroy_cq(&rdev->qplib_res, &cq->qplib_cq);
2949 	ib_umem_release(cq->umem);
2950 
2951 	atomic_dec(&rdev->stats.res.cq_count);
2952 	nq->budget--;
2953 	kfree(cq->cql);
2954 	return 0;
2955 }
2956 
2957 int bnxt_re_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
2958 		      struct uverbs_attr_bundle *attrs)
2959 {
2960 	struct bnxt_re_cq *cq = container_of(ibcq, struct bnxt_re_cq, ib_cq);
2961 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibcq->device, ibdev);
2962 	struct ib_udata *udata = &attrs->driver_udata;
2963 	struct bnxt_re_ucontext *uctx =
2964 		rdma_udata_to_drv_context(udata, struct bnxt_re_ucontext, ib_uctx);
2965 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
2966 	struct bnxt_qplib_chip_ctx *cctx;
2967 	struct bnxt_qplib_nq *nq = NULL;
2968 	unsigned int nq_alloc_cnt;
2969 	int cqe = attr->cqe;
2970 	int rc, entries;
2971 	u32 active_cqs;
2972 
2973 	if (attr->flags)
2974 		return -EOPNOTSUPP;
2975 
2976 	/* Validate CQ fields */
2977 	if (cqe < 1 || cqe > dev_attr->max_cq_wqes) {
2978 		ibdev_err(&rdev->ibdev, "Failed to create CQ -max exceeded");
2979 		return -EINVAL;
2980 	}
2981 
2982 	cq->rdev = rdev;
2983 	cctx = rdev->chip_ctx;
2984 	cq->qplib_cq.cq_handle = (u64)(unsigned long)(&cq->qplib_cq);
2985 
2986 	entries = bnxt_re_init_depth(cqe + 1, uctx);
2987 	if (entries > dev_attr->max_cq_wqes + 1)
2988 		entries = dev_attr->max_cq_wqes + 1;
2989 
2990 	cq->qplib_cq.sg_info.pgsize = PAGE_SIZE;
2991 	cq->qplib_cq.sg_info.pgshft = PAGE_SHIFT;
2992 	if (udata) {
2993 		struct bnxt_re_cq_req req;
2994 		if (ib_copy_from_udata(&req, udata, sizeof(req))) {
2995 			rc = -EFAULT;
2996 			goto fail;
2997 		}
2998 
2999 		cq->umem = ib_umem_get(&rdev->ibdev, req.cq_va,
3000 				       entries * sizeof(struct cq_base),
3001 				       IB_ACCESS_LOCAL_WRITE);
3002 		if (IS_ERR(cq->umem)) {
3003 			rc = PTR_ERR(cq->umem);
3004 			goto fail;
3005 		}
3006 		cq->qplib_cq.sg_info.umem = cq->umem;
3007 		cq->qplib_cq.dpi = &uctx->dpi;
3008 	} else {
3009 		cq->max_cql = min_t(u32, entries, MAX_CQL_PER_POLL);
3010 		cq->cql = kcalloc(cq->max_cql, sizeof(struct bnxt_qplib_cqe),
3011 				  GFP_KERNEL);
3012 		if (!cq->cql) {
3013 			rc = -ENOMEM;
3014 			goto fail;
3015 		}
3016 
3017 		cq->qplib_cq.dpi = &rdev->dpi_privileged;
3018 	}
3019 	/*
3020 	 * Allocating the NQ in a round robin fashion. nq_alloc_cnt is a
3021 	 * used for getting the NQ index.
3022 	 */
3023 	nq_alloc_cnt = atomic_inc_return(&rdev->nq_alloc_cnt);
3024 	nq = &rdev->nq[nq_alloc_cnt % (rdev->num_msix - 1)];
3025 	cq->qplib_cq.max_wqe = entries;
3026 	cq->qplib_cq.cnq_hw_ring_id = nq->ring_id;
3027 	cq->qplib_cq.nq	= nq;
3028 
3029 	rc = bnxt_qplib_create_cq(&rdev->qplib_res, &cq->qplib_cq);
3030 	if (rc) {
3031 		ibdev_err(&rdev->ibdev, "Failed to create HW CQ");
3032 		goto fail;
3033 	}
3034 
3035 	cq->ib_cq.cqe = entries;
3036 	cq->cq_period = cq->qplib_cq.period;
3037 	nq->budget++;
3038 
3039 	active_cqs = atomic_inc_return(&rdev->stats.res.cq_count);
3040 	if (active_cqs > rdev->stats.res.cq_watermark)
3041 		rdev->stats.res.cq_watermark = active_cqs;
3042 	spin_lock_init(&cq->cq_lock);
3043 
3044 	if (udata) {
3045 		struct bnxt_re_cq_resp resp = {};
3046 
3047 		if (cctx->modes.toggle_bits & BNXT_QPLIB_CQ_TOGGLE_BIT) {
3048 			hash_add(rdev->cq_hash, &cq->hash_entry, cq->qplib_cq.id);
3049 			/* Allocate a page */
3050 			cq->uctx_cq_page = (void *)get_zeroed_page(GFP_KERNEL);
3051 			if (!cq->uctx_cq_page) {
3052 				rc = -ENOMEM;
3053 				goto c2fail;
3054 			}
3055 			resp.comp_mask |= BNXT_RE_CQ_TOGGLE_PAGE_SUPPORT;
3056 		}
3057 		resp.cqid = cq->qplib_cq.id;
3058 		resp.tail = cq->qplib_cq.hwq.cons;
3059 		resp.phase = cq->qplib_cq.period;
3060 		resp.rsvd = 0;
3061 		rc = ib_copy_to_udata(udata, &resp, min(sizeof(resp), udata->outlen));
3062 		if (rc) {
3063 			ibdev_err(&rdev->ibdev, "Failed to copy CQ udata");
3064 			bnxt_qplib_destroy_cq(&rdev->qplib_res, &cq->qplib_cq);
3065 			goto free_mem;
3066 		}
3067 	}
3068 
3069 	return 0;
3070 
3071 free_mem:
3072 	free_page((unsigned long)cq->uctx_cq_page);
3073 c2fail:
3074 	ib_umem_release(cq->umem);
3075 fail:
3076 	kfree(cq->cql);
3077 	return rc;
3078 }
3079 
3080 static void bnxt_re_resize_cq_complete(struct bnxt_re_cq *cq)
3081 {
3082 	struct bnxt_re_dev *rdev = cq->rdev;
3083 
3084 	bnxt_qplib_resize_cq_complete(&rdev->qplib_res, &cq->qplib_cq);
3085 
3086 	cq->qplib_cq.max_wqe = cq->resize_cqe;
3087 	if (cq->resize_umem) {
3088 		ib_umem_release(cq->umem);
3089 		cq->umem = cq->resize_umem;
3090 		cq->resize_umem = NULL;
3091 		cq->resize_cqe = 0;
3092 	}
3093 }
3094 
3095 int bnxt_re_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata)
3096 {
3097 	struct bnxt_qplib_sg_info sg_info = {};
3098 	struct bnxt_qplib_dpi *orig_dpi = NULL;
3099 	struct bnxt_qplib_dev_attr *dev_attr;
3100 	struct bnxt_re_ucontext *uctx = NULL;
3101 	struct bnxt_re_resize_cq_req req;
3102 	struct bnxt_re_dev *rdev;
3103 	struct bnxt_re_cq *cq;
3104 	int rc, entries;
3105 
3106 	cq =  container_of(ibcq, struct bnxt_re_cq, ib_cq);
3107 	rdev = cq->rdev;
3108 	dev_attr = &rdev->dev_attr;
3109 	if (!ibcq->uobject) {
3110 		ibdev_err(&rdev->ibdev, "Kernel CQ Resize not supported");
3111 		return -EOPNOTSUPP;
3112 	}
3113 
3114 	if (cq->resize_umem) {
3115 		ibdev_err(&rdev->ibdev, "Resize CQ %#x failed - Busy",
3116 			  cq->qplib_cq.id);
3117 		return -EBUSY;
3118 	}
3119 
3120 	/* Check the requested cq depth out of supported depth */
3121 	if (cqe < 1 || cqe > dev_attr->max_cq_wqes) {
3122 		ibdev_err(&rdev->ibdev, "Resize CQ %#x failed - out of range cqe %d",
3123 			  cq->qplib_cq.id, cqe);
3124 		return -EINVAL;
3125 	}
3126 
3127 	uctx = rdma_udata_to_drv_context(udata, struct bnxt_re_ucontext, ib_uctx);
3128 	entries = bnxt_re_init_depth(cqe + 1, uctx);
3129 	if (entries > dev_attr->max_cq_wqes + 1)
3130 		entries = dev_attr->max_cq_wqes + 1;
3131 
3132 	/* uverbs consumer */
3133 	if (ib_copy_from_udata(&req, udata, sizeof(req))) {
3134 		rc = -EFAULT;
3135 		goto fail;
3136 	}
3137 
3138 	cq->resize_umem = ib_umem_get(&rdev->ibdev, req.cq_va,
3139 				      entries * sizeof(struct cq_base),
3140 				      IB_ACCESS_LOCAL_WRITE);
3141 	if (IS_ERR(cq->resize_umem)) {
3142 		rc = PTR_ERR(cq->resize_umem);
3143 		cq->resize_umem = NULL;
3144 		ibdev_err(&rdev->ibdev, "%s: ib_umem_get failed! rc = %d\n",
3145 			  __func__, rc);
3146 		goto fail;
3147 	}
3148 	cq->resize_cqe = entries;
3149 	memcpy(&sg_info, &cq->qplib_cq.sg_info, sizeof(sg_info));
3150 	orig_dpi = cq->qplib_cq.dpi;
3151 
3152 	cq->qplib_cq.sg_info.umem = cq->resize_umem;
3153 	cq->qplib_cq.sg_info.pgsize = PAGE_SIZE;
3154 	cq->qplib_cq.sg_info.pgshft = PAGE_SHIFT;
3155 	cq->qplib_cq.dpi = &uctx->dpi;
3156 
3157 	rc = bnxt_qplib_resize_cq(&rdev->qplib_res, &cq->qplib_cq, entries);
3158 	if (rc) {
3159 		ibdev_err(&rdev->ibdev, "Resize HW CQ %#x failed!",
3160 			  cq->qplib_cq.id);
3161 		goto fail;
3162 	}
3163 
3164 	cq->ib_cq.cqe = cq->resize_cqe;
3165 	atomic_inc(&rdev->stats.res.resize_count);
3166 
3167 	return 0;
3168 
3169 fail:
3170 	if (cq->resize_umem) {
3171 		ib_umem_release(cq->resize_umem);
3172 		cq->resize_umem = NULL;
3173 		cq->resize_cqe = 0;
3174 		memcpy(&cq->qplib_cq.sg_info, &sg_info, sizeof(sg_info));
3175 		cq->qplib_cq.dpi = orig_dpi;
3176 	}
3177 	return rc;
3178 }
3179 
3180 static u8 __req_to_ib_wc_status(u8 qstatus)
3181 {
3182 	switch (qstatus) {
3183 	case CQ_REQ_STATUS_OK:
3184 		return IB_WC_SUCCESS;
3185 	case CQ_REQ_STATUS_BAD_RESPONSE_ERR:
3186 		return IB_WC_BAD_RESP_ERR;
3187 	case CQ_REQ_STATUS_LOCAL_LENGTH_ERR:
3188 		return IB_WC_LOC_LEN_ERR;
3189 	case CQ_REQ_STATUS_LOCAL_QP_OPERATION_ERR:
3190 		return IB_WC_LOC_QP_OP_ERR;
3191 	case CQ_REQ_STATUS_LOCAL_PROTECTION_ERR:
3192 		return IB_WC_LOC_PROT_ERR;
3193 	case CQ_REQ_STATUS_MEMORY_MGT_OPERATION_ERR:
3194 		return IB_WC_GENERAL_ERR;
3195 	case CQ_REQ_STATUS_REMOTE_INVALID_REQUEST_ERR:
3196 		return IB_WC_REM_INV_REQ_ERR;
3197 	case CQ_REQ_STATUS_REMOTE_ACCESS_ERR:
3198 		return IB_WC_REM_ACCESS_ERR;
3199 	case CQ_REQ_STATUS_REMOTE_OPERATION_ERR:
3200 		return IB_WC_REM_OP_ERR;
3201 	case CQ_REQ_STATUS_RNR_NAK_RETRY_CNT_ERR:
3202 		return IB_WC_RNR_RETRY_EXC_ERR;
3203 	case CQ_REQ_STATUS_TRANSPORT_RETRY_CNT_ERR:
3204 		return IB_WC_RETRY_EXC_ERR;
3205 	case CQ_REQ_STATUS_WORK_REQUEST_FLUSHED_ERR:
3206 		return IB_WC_WR_FLUSH_ERR;
3207 	default:
3208 		return IB_WC_GENERAL_ERR;
3209 	}
3210 	return 0;
3211 }
3212 
3213 static u8 __rawqp1_to_ib_wc_status(u8 qstatus)
3214 {
3215 	switch (qstatus) {
3216 	case CQ_RES_RAWETH_QP1_STATUS_OK:
3217 		return IB_WC_SUCCESS;
3218 	case CQ_RES_RAWETH_QP1_STATUS_LOCAL_ACCESS_ERROR:
3219 		return IB_WC_LOC_ACCESS_ERR;
3220 	case CQ_RES_RAWETH_QP1_STATUS_HW_LOCAL_LENGTH_ERR:
3221 		return IB_WC_LOC_LEN_ERR;
3222 	case CQ_RES_RAWETH_QP1_STATUS_LOCAL_PROTECTION_ERR:
3223 		return IB_WC_LOC_PROT_ERR;
3224 	case CQ_RES_RAWETH_QP1_STATUS_LOCAL_QP_OPERATION_ERR:
3225 		return IB_WC_LOC_QP_OP_ERR;
3226 	case CQ_RES_RAWETH_QP1_STATUS_MEMORY_MGT_OPERATION_ERR:
3227 		return IB_WC_GENERAL_ERR;
3228 	case CQ_RES_RAWETH_QP1_STATUS_WORK_REQUEST_FLUSHED_ERR:
3229 		return IB_WC_WR_FLUSH_ERR;
3230 	case CQ_RES_RAWETH_QP1_STATUS_HW_FLUSH_ERR:
3231 		return IB_WC_WR_FLUSH_ERR;
3232 	default:
3233 		return IB_WC_GENERAL_ERR;
3234 	}
3235 }
3236 
3237 static u8 __rc_to_ib_wc_status(u8 qstatus)
3238 {
3239 	switch (qstatus) {
3240 	case CQ_RES_RC_STATUS_OK:
3241 		return IB_WC_SUCCESS;
3242 	case CQ_RES_RC_STATUS_LOCAL_ACCESS_ERROR:
3243 		return IB_WC_LOC_ACCESS_ERR;
3244 	case CQ_RES_RC_STATUS_LOCAL_LENGTH_ERR:
3245 		return IB_WC_LOC_LEN_ERR;
3246 	case CQ_RES_RC_STATUS_LOCAL_PROTECTION_ERR:
3247 		return IB_WC_LOC_PROT_ERR;
3248 	case CQ_RES_RC_STATUS_LOCAL_QP_OPERATION_ERR:
3249 		return IB_WC_LOC_QP_OP_ERR;
3250 	case CQ_RES_RC_STATUS_MEMORY_MGT_OPERATION_ERR:
3251 		return IB_WC_GENERAL_ERR;
3252 	case CQ_RES_RC_STATUS_REMOTE_INVALID_REQUEST_ERR:
3253 		return IB_WC_REM_INV_REQ_ERR;
3254 	case CQ_RES_RC_STATUS_WORK_REQUEST_FLUSHED_ERR:
3255 		return IB_WC_WR_FLUSH_ERR;
3256 	case CQ_RES_RC_STATUS_HW_FLUSH_ERR:
3257 		return IB_WC_WR_FLUSH_ERR;
3258 	default:
3259 		return IB_WC_GENERAL_ERR;
3260 	}
3261 }
3262 
3263 static void bnxt_re_process_req_wc(struct ib_wc *wc, struct bnxt_qplib_cqe *cqe)
3264 {
3265 	switch (cqe->type) {
3266 	case BNXT_QPLIB_SWQE_TYPE_SEND:
3267 		wc->opcode = IB_WC_SEND;
3268 		break;
3269 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_IMM:
3270 		wc->opcode = IB_WC_SEND;
3271 		wc->wc_flags |= IB_WC_WITH_IMM;
3272 		break;
3273 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_INV:
3274 		wc->opcode = IB_WC_SEND;
3275 		wc->wc_flags |= IB_WC_WITH_INVALIDATE;
3276 		break;
3277 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE:
3278 		wc->opcode = IB_WC_RDMA_WRITE;
3279 		break;
3280 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE_WITH_IMM:
3281 		wc->opcode = IB_WC_RDMA_WRITE;
3282 		wc->wc_flags |= IB_WC_WITH_IMM;
3283 		break;
3284 	case BNXT_QPLIB_SWQE_TYPE_RDMA_READ:
3285 		wc->opcode = IB_WC_RDMA_READ;
3286 		break;
3287 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_CMP_AND_SWP:
3288 		wc->opcode = IB_WC_COMP_SWAP;
3289 		break;
3290 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_FETCH_AND_ADD:
3291 		wc->opcode = IB_WC_FETCH_ADD;
3292 		break;
3293 	case BNXT_QPLIB_SWQE_TYPE_LOCAL_INV:
3294 		wc->opcode = IB_WC_LOCAL_INV;
3295 		break;
3296 	case BNXT_QPLIB_SWQE_TYPE_REG_MR:
3297 		wc->opcode = IB_WC_REG_MR;
3298 		break;
3299 	default:
3300 		wc->opcode = IB_WC_SEND;
3301 		break;
3302 	}
3303 
3304 	wc->status = __req_to_ib_wc_status(cqe->status);
3305 }
3306 
3307 static int bnxt_re_check_packet_type(u16 raweth_qp1_flags,
3308 				     u16 raweth_qp1_flags2)
3309 {
3310 	bool is_ipv6 = false, is_ipv4 = false;
3311 
3312 	/* raweth_qp1_flags Bit 9-6 indicates itype */
3313 	if ((raweth_qp1_flags & CQ_RES_RAWETH_QP1_RAWETH_QP1_FLAGS_ITYPE_ROCE)
3314 	    != CQ_RES_RAWETH_QP1_RAWETH_QP1_FLAGS_ITYPE_ROCE)
3315 		return -1;
3316 
3317 	if (raweth_qp1_flags2 &
3318 	    CQ_RES_RAWETH_QP1_RAWETH_QP1_FLAGS2_IP_CS_CALC &&
3319 	    raweth_qp1_flags2 &
3320 	    CQ_RES_RAWETH_QP1_RAWETH_QP1_FLAGS2_L4_CS_CALC) {
3321 		/* raweth_qp1_flags2 Bit 8 indicates ip_type. 0-v4 1 - v6 */
3322 		(raweth_qp1_flags2 &
3323 		 CQ_RES_RAWETH_QP1_RAWETH_QP1_FLAGS2_IP_TYPE) ?
3324 			(is_ipv6 = true) : (is_ipv4 = true);
3325 		return ((is_ipv6) ?
3326 			 BNXT_RE_ROCEV2_IPV6_PACKET :
3327 			 BNXT_RE_ROCEV2_IPV4_PACKET);
3328 	} else {
3329 		return BNXT_RE_ROCE_V1_PACKET;
3330 	}
3331 }
3332 
3333 static int bnxt_re_to_ib_nw_type(int nw_type)
3334 {
3335 	u8 nw_hdr_type = 0xFF;
3336 
3337 	switch (nw_type) {
3338 	case BNXT_RE_ROCE_V1_PACKET:
3339 		nw_hdr_type = RDMA_NETWORK_ROCE_V1;
3340 		break;
3341 	case BNXT_RE_ROCEV2_IPV4_PACKET:
3342 		nw_hdr_type = RDMA_NETWORK_IPV4;
3343 		break;
3344 	case BNXT_RE_ROCEV2_IPV6_PACKET:
3345 		nw_hdr_type = RDMA_NETWORK_IPV6;
3346 		break;
3347 	}
3348 	return nw_hdr_type;
3349 }
3350 
3351 static bool bnxt_re_is_loopback_packet(struct bnxt_re_dev *rdev,
3352 				       void *rq_hdr_buf)
3353 {
3354 	u8 *tmp_buf = NULL;
3355 	struct ethhdr *eth_hdr;
3356 	u16 eth_type;
3357 	bool rc = false;
3358 
3359 	tmp_buf = (u8 *)rq_hdr_buf;
3360 	/*
3361 	 * If dest mac is not same as I/F mac, this could be a
3362 	 * loopback address or multicast address, check whether
3363 	 * it is a loopback packet
3364 	 */
3365 	if (!ether_addr_equal(tmp_buf, rdev->netdev->dev_addr)) {
3366 		tmp_buf += 4;
3367 		/* Check the  ether type */
3368 		eth_hdr = (struct ethhdr *)tmp_buf;
3369 		eth_type = ntohs(eth_hdr->h_proto);
3370 		switch (eth_type) {
3371 		case ETH_P_IBOE:
3372 			rc = true;
3373 			break;
3374 		case ETH_P_IP:
3375 		case ETH_P_IPV6: {
3376 			u32 len;
3377 			struct udphdr *udp_hdr;
3378 
3379 			len = (eth_type == ETH_P_IP ? sizeof(struct iphdr) :
3380 						      sizeof(struct ipv6hdr));
3381 			tmp_buf += sizeof(struct ethhdr) + len;
3382 			udp_hdr = (struct udphdr *)tmp_buf;
3383 			if (ntohs(udp_hdr->dest) ==
3384 				    ROCE_V2_UDP_DPORT)
3385 				rc = true;
3386 			break;
3387 			}
3388 		default:
3389 			break;
3390 		}
3391 	}
3392 
3393 	return rc;
3394 }
3395 
3396 static int bnxt_re_process_raw_qp_pkt_rx(struct bnxt_re_qp *gsi_qp,
3397 					 struct bnxt_qplib_cqe *cqe)
3398 {
3399 	struct bnxt_re_dev *rdev = gsi_qp->rdev;
3400 	struct bnxt_re_sqp_entries *sqp_entry = NULL;
3401 	struct bnxt_re_qp *gsi_sqp = rdev->gsi_ctx.gsi_sqp;
3402 	dma_addr_t shrq_hdr_buf_map;
3403 	struct ib_sge s_sge[2] = {};
3404 	struct ib_sge r_sge[2] = {};
3405 	struct bnxt_re_ah *gsi_sah;
3406 	struct ib_recv_wr rwr = {};
3407 	dma_addr_t rq_hdr_buf_map;
3408 	struct ib_ud_wr udwr = {};
3409 	struct ib_send_wr *swr;
3410 	u32 skip_bytes = 0;
3411 	int pkt_type = 0;
3412 	void *rq_hdr_buf;
3413 	u32 offset = 0;
3414 	u32 tbl_idx;
3415 	int rc;
3416 
3417 	swr = &udwr.wr;
3418 	tbl_idx = cqe->wr_id;
3419 
3420 	rq_hdr_buf = gsi_qp->qplib_qp.rq_hdr_buf +
3421 			(tbl_idx * gsi_qp->qplib_qp.rq_hdr_buf_size);
3422 	rq_hdr_buf_map = bnxt_qplib_get_qp_buf_from_index(&gsi_qp->qplib_qp,
3423 							  tbl_idx);
3424 
3425 	/* Shadow QP header buffer */
3426 	shrq_hdr_buf_map = bnxt_qplib_get_qp_buf_from_index(&gsi_qp->qplib_qp,
3427 							    tbl_idx);
3428 	sqp_entry = &rdev->gsi_ctx.sqp_tbl[tbl_idx];
3429 
3430 	/* Store this cqe */
3431 	memcpy(&sqp_entry->cqe, cqe, sizeof(struct bnxt_qplib_cqe));
3432 	sqp_entry->qp1_qp = gsi_qp;
3433 
3434 	/* Find packet type from the cqe */
3435 
3436 	pkt_type = bnxt_re_check_packet_type(cqe->raweth_qp1_flags,
3437 					     cqe->raweth_qp1_flags2);
3438 	if (pkt_type < 0) {
3439 		ibdev_err(&rdev->ibdev, "Invalid packet\n");
3440 		return -EINVAL;
3441 	}
3442 
3443 	/* Adjust the offset for the user buffer and post in the rq */
3444 
3445 	if (pkt_type == BNXT_RE_ROCEV2_IPV4_PACKET)
3446 		offset = 20;
3447 
3448 	/*
3449 	 * QP1 loopback packet has 4 bytes of internal header before
3450 	 * ether header. Skip these four bytes.
3451 	 */
3452 	if (bnxt_re_is_loopback_packet(rdev, rq_hdr_buf))
3453 		skip_bytes = 4;
3454 
3455 	/* First send SGE . Skip the ether header*/
3456 	s_sge[0].addr = rq_hdr_buf_map + BNXT_QPLIB_MAX_QP1_RQ_ETH_HDR_SIZE
3457 			+ skip_bytes;
3458 	s_sge[0].lkey = 0xFFFFFFFF;
3459 	s_sge[0].length = offset ? BNXT_QPLIB_MAX_GRH_HDR_SIZE_IPV4 :
3460 				BNXT_QPLIB_MAX_GRH_HDR_SIZE_IPV6;
3461 
3462 	/* Second Send SGE */
3463 	s_sge[1].addr = s_sge[0].addr + s_sge[0].length +
3464 			BNXT_QPLIB_MAX_QP1_RQ_BDETH_HDR_SIZE;
3465 	if (pkt_type != BNXT_RE_ROCE_V1_PACKET)
3466 		s_sge[1].addr += 8;
3467 	s_sge[1].lkey = 0xFFFFFFFF;
3468 	s_sge[1].length = 256;
3469 
3470 	/* First recv SGE */
3471 
3472 	r_sge[0].addr = shrq_hdr_buf_map;
3473 	r_sge[0].lkey = 0xFFFFFFFF;
3474 	r_sge[0].length = 40;
3475 
3476 	r_sge[1].addr = sqp_entry->sge.addr + offset;
3477 	r_sge[1].lkey = sqp_entry->sge.lkey;
3478 	r_sge[1].length = BNXT_QPLIB_MAX_GRH_HDR_SIZE_IPV6 + 256 - offset;
3479 
3480 	/* Create receive work request */
3481 	rwr.num_sge = 2;
3482 	rwr.sg_list = r_sge;
3483 	rwr.wr_id = tbl_idx;
3484 	rwr.next = NULL;
3485 
3486 	rc = bnxt_re_post_recv_shadow_qp(rdev, gsi_sqp, &rwr);
3487 	if (rc) {
3488 		ibdev_err(&rdev->ibdev,
3489 			  "Failed to post Rx buffers to shadow QP");
3490 		return -ENOMEM;
3491 	}
3492 
3493 	swr->num_sge = 2;
3494 	swr->sg_list = s_sge;
3495 	swr->wr_id = tbl_idx;
3496 	swr->opcode = IB_WR_SEND;
3497 	swr->next = NULL;
3498 	gsi_sah = rdev->gsi_ctx.gsi_sah;
3499 	udwr.ah = &gsi_sah->ib_ah;
3500 	udwr.remote_qpn = gsi_sqp->qplib_qp.id;
3501 	udwr.remote_qkey = gsi_sqp->qplib_qp.qkey;
3502 
3503 	/* post data received  in the send queue */
3504 	return bnxt_re_post_send_shadow_qp(rdev, gsi_sqp, swr);
3505 }
3506 
3507 static void bnxt_re_process_res_rawqp1_wc(struct ib_wc *wc,
3508 					  struct bnxt_qplib_cqe *cqe)
3509 {
3510 	wc->opcode = IB_WC_RECV;
3511 	wc->status = __rawqp1_to_ib_wc_status(cqe->status);
3512 	wc->wc_flags |= IB_WC_GRH;
3513 }
3514 
3515 static bool bnxt_re_check_if_vlan_valid(struct bnxt_re_dev *rdev,
3516 					u16 vlan_id)
3517 {
3518 	/*
3519 	 * Check if the vlan is configured in the host.  If not configured, it
3520 	 * can be a transparent VLAN. So dont report the vlan id.
3521 	 */
3522 	if (!__vlan_find_dev_deep_rcu(rdev->netdev,
3523 				      htons(ETH_P_8021Q), vlan_id))
3524 		return false;
3525 	return true;
3526 }
3527 
3528 static bool bnxt_re_is_vlan_pkt(struct bnxt_qplib_cqe *orig_cqe,
3529 				u16 *vid, u8 *sl)
3530 {
3531 	bool ret = false;
3532 	u32 metadata;
3533 	u16 tpid;
3534 
3535 	metadata = orig_cqe->raweth_qp1_metadata;
3536 	if (orig_cqe->raweth_qp1_flags2 &
3537 		CQ_RES_RAWETH_QP1_RAWETH_QP1_FLAGS2_META_FORMAT_VLAN) {
3538 		tpid = ((metadata &
3539 			 CQ_RES_RAWETH_QP1_RAWETH_QP1_METADATA_TPID_MASK) >>
3540 			 CQ_RES_RAWETH_QP1_RAWETH_QP1_METADATA_TPID_SFT);
3541 		if (tpid == ETH_P_8021Q) {
3542 			*vid = metadata &
3543 			       CQ_RES_RAWETH_QP1_RAWETH_QP1_METADATA_VID_MASK;
3544 			*sl = (metadata &
3545 			       CQ_RES_RAWETH_QP1_RAWETH_QP1_METADATA_PRI_MASK) >>
3546 			       CQ_RES_RAWETH_QP1_RAWETH_QP1_METADATA_PRI_SFT;
3547 			ret = true;
3548 		}
3549 	}
3550 
3551 	return ret;
3552 }
3553 
3554 static void bnxt_re_process_res_rc_wc(struct ib_wc *wc,
3555 				      struct bnxt_qplib_cqe *cqe)
3556 {
3557 	wc->opcode = IB_WC_RECV;
3558 	wc->status = __rc_to_ib_wc_status(cqe->status);
3559 
3560 	if (cqe->flags & CQ_RES_RC_FLAGS_IMM)
3561 		wc->wc_flags |= IB_WC_WITH_IMM;
3562 	if (cqe->flags & CQ_RES_RC_FLAGS_INV)
3563 		wc->wc_flags |= IB_WC_WITH_INVALIDATE;
3564 	if ((cqe->flags & (CQ_RES_RC_FLAGS_RDMA | CQ_RES_RC_FLAGS_IMM)) ==
3565 	    (CQ_RES_RC_FLAGS_RDMA | CQ_RES_RC_FLAGS_IMM))
3566 		wc->opcode = IB_WC_RECV_RDMA_WITH_IMM;
3567 }
3568 
3569 static void bnxt_re_process_res_shadow_qp_wc(struct bnxt_re_qp *gsi_sqp,
3570 					     struct ib_wc *wc,
3571 					     struct bnxt_qplib_cqe *cqe)
3572 {
3573 	struct bnxt_re_dev *rdev = gsi_sqp->rdev;
3574 	struct bnxt_re_qp *gsi_qp = NULL;
3575 	struct bnxt_qplib_cqe *orig_cqe = NULL;
3576 	struct bnxt_re_sqp_entries *sqp_entry = NULL;
3577 	int nw_type;
3578 	u32 tbl_idx;
3579 	u16 vlan_id;
3580 	u8 sl;
3581 
3582 	tbl_idx = cqe->wr_id;
3583 
3584 	sqp_entry = &rdev->gsi_ctx.sqp_tbl[tbl_idx];
3585 	gsi_qp = sqp_entry->qp1_qp;
3586 	orig_cqe = &sqp_entry->cqe;
3587 
3588 	wc->wr_id = sqp_entry->wrid;
3589 	wc->byte_len = orig_cqe->length;
3590 	wc->qp = &gsi_qp->ib_qp;
3591 
3592 	wc->ex.imm_data = cpu_to_be32(le32_to_cpu(orig_cqe->immdata));
3593 	wc->src_qp = orig_cqe->src_qp;
3594 	memcpy(wc->smac, orig_cqe->smac, ETH_ALEN);
3595 	if (bnxt_re_is_vlan_pkt(orig_cqe, &vlan_id, &sl)) {
3596 		if (bnxt_re_check_if_vlan_valid(rdev, vlan_id)) {
3597 			wc->vlan_id = vlan_id;
3598 			wc->sl = sl;
3599 			wc->wc_flags |= IB_WC_WITH_VLAN;
3600 		}
3601 	}
3602 	wc->port_num = 1;
3603 	wc->vendor_err = orig_cqe->status;
3604 
3605 	wc->opcode = IB_WC_RECV;
3606 	wc->status = __rawqp1_to_ib_wc_status(orig_cqe->status);
3607 	wc->wc_flags |= IB_WC_GRH;
3608 
3609 	nw_type = bnxt_re_check_packet_type(orig_cqe->raweth_qp1_flags,
3610 					    orig_cqe->raweth_qp1_flags2);
3611 	if (nw_type >= 0) {
3612 		wc->network_hdr_type = bnxt_re_to_ib_nw_type(nw_type);
3613 		wc->wc_flags |= IB_WC_WITH_NETWORK_HDR_TYPE;
3614 	}
3615 }
3616 
3617 static void bnxt_re_process_res_ud_wc(struct bnxt_re_qp *qp,
3618 				      struct ib_wc *wc,
3619 				      struct bnxt_qplib_cqe *cqe)
3620 {
3621 	struct bnxt_re_dev *rdev;
3622 	u16 vlan_id = 0;
3623 	u8 nw_type;
3624 
3625 	rdev = qp->rdev;
3626 	wc->opcode = IB_WC_RECV;
3627 	wc->status = __rc_to_ib_wc_status(cqe->status);
3628 
3629 	if (cqe->flags & CQ_RES_UD_FLAGS_IMM)
3630 		wc->wc_flags |= IB_WC_WITH_IMM;
3631 	/* report only on GSI QP for Thor */
3632 	if (qp->qplib_qp.type == CMDQ_CREATE_QP_TYPE_GSI) {
3633 		wc->wc_flags |= IB_WC_GRH;
3634 		memcpy(wc->smac, cqe->smac, ETH_ALEN);
3635 		wc->wc_flags |= IB_WC_WITH_SMAC;
3636 		if (cqe->flags & CQ_RES_UD_FLAGS_META_FORMAT_VLAN) {
3637 			vlan_id = (cqe->cfa_meta & 0xFFF);
3638 		}
3639 		/* Mark only if vlan_id is non zero */
3640 		if (vlan_id && bnxt_re_check_if_vlan_valid(rdev, vlan_id)) {
3641 			wc->vlan_id = vlan_id;
3642 			wc->wc_flags |= IB_WC_WITH_VLAN;
3643 		}
3644 		nw_type = (cqe->flags & CQ_RES_UD_FLAGS_ROCE_IP_VER_MASK) >>
3645 			   CQ_RES_UD_FLAGS_ROCE_IP_VER_SFT;
3646 		wc->network_hdr_type = bnxt_re_to_ib_nw_type(nw_type);
3647 		wc->wc_flags |= IB_WC_WITH_NETWORK_HDR_TYPE;
3648 	}
3649 
3650 }
3651 
3652 static int send_phantom_wqe(struct bnxt_re_qp *qp)
3653 {
3654 	struct bnxt_qplib_qp *lib_qp = &qp->qplib_qp;
3655 	unsigned long flags;
3656 	int rc;
3657 
3658 	spin_lock_irqsave(&qp->sq_lock, flags);
3659 
3660 	rc = bnxt_re_bind_fence_mw(lib_qp);
3661 	if (!rc) {
3662 		lib_qp->sq.phantom_wqe_cnt++;
3663 		ibdev_dbg(&qp->rdev->ibdev,
3664 			  "qp %#x sq->prod %#x sw_prod %#x phantom_wqe_cnt %d\n",
3665 			  lib_qp->id, lib_qp->sq.hwq.prod,
3666 			  HWQ_CMP(lib_qp->sq.hwq.prod, &lib_qp->sq.hwq),
3667 			  lib_qp->sq.phantom_wqe_cnt);
3668 	}
3669 
3670 	spin_unlock_irqrestore(&qp->sq_lock, flags);
3671 	return rc;
3672 }
3673 
3674 int bnxt_re_poll_cq(struct ib_cq *ib_cq, int num_entries, struct ib_wc *wc)
3675 {
3676 	struct bnxt_re_cq *cq = container_of(ib_cq, struct bnxt_re_cq, ib_cq);
3677 	struct bnxt_re_qp *qp, *sh_qp;
3678 	struct bnxt_qplib_cqe *cqe;
3679 	int i, ncqe, budget;
3680 	struct bnxt_qplib_q *sq;
3681 	struct bnxt_qplib_qp *lib_qp;
3682 	u32 tbl_idx;
3683 	struct bnxt_re_sqp_entries *sqp_entry = NULL;
3684 	unsigned long flags;
3685 
3686 	/* User CQ; the only processing we do is to
3687 	 * complete any pending CQ resize operation.
3688 	 */
3689 	if (cq->umem) {
3690 		if (cq->resize_umem)
3691 			bnxt_re_resize_cq_complete(cq);
3692 		return 0;
3693 	}
3694 
3695 	spin_lock_irqsave(&cq->cq_lock, flags);
3696 	budget = min_t(u32, num_entries, cq->max_cql);
3697 	num_entries = budget;
3698 	if (!cq->cql) {
3699 		ibdev_err(&cq->rdev->ibdev, "POLL CQ : no CQL to use");
3700 		goto exit;
3701 	}
3702 	cqe = &cq->cql[0];
3703 	while (budget) {
3704 		lib_qp = NULL;
3705 		ncqe = bnxt_qplib_poll_cq(&cq->qplib_cq, cqe, budget, &lib_qp);
3706 		if (lib_qp) {
3707 			sq = &lib_qp->sq;
3708 			if (sq->send_phantom) {
3709 				qp = container_of(lib_qp,
3710 						  struct bnxt_re_qp, qplib_qp);
3711 				if (send_phantom_wqe(qp) == -ENOMEM)
3712 					ibdev_err(&cq->rdev->ibdev,
3713 						  "Phantom failed! Scheduled to send again\n");
3714 				else
3715 					sq->send_phantom = false;
3716 			}
3717 		}
3718 		if (ncqe < budget)
3719 			ncqe += bnxt_qplib_process_flush_list(&cq->qplib_cq,
3720 							      cqe + ncqe,
3721 							      budget - ncqe);
3722 
3723 		if (!ncqe)
3724 			break;
3725 
3726 		for (i = 0; i < ncqe; i++, cqe++) {
3727 			/* Transcribe each qplib_wqe back to ib_wc */
3728 			memset(wc, 0, sizeof(*wc));
3729 
3730 			wc->wr_id = cqe->wr_id;
3731 			wc->byte_len = cqe->length;
3732 			qp = container_of
3733 				((struct bnxt_qplib_qp *)
3734 				 (unsigned long)(cqe->qp_handle),
3735 				 struct bnxt_re_qp, qplib_qp);
3736 			wc->qp = &qp->ib_qp;
3737 			wc->ex.imm_data = cpu_to_be32(le32_to_cpu(cqe->immdata));
3738 			wc->src_qp = cqe->src_qp;
3739 			memcpy(wc->smac, cqe->smac, ETH_ALEN);
3740 			wc->port_num = 1;
3741 			wc->vendor_err = cqe->status;
3742 
3743 			switch (cqe->opcode) {
3744 			case CQ_BASE_CQE_TYPE_REQ:
3745 				sh_qp = qp->rdev->gsi_ctx.gsi_sqp;
3746 				if (sh_qp &&
3747 				    qp->qplib_qp.id == sh_qp->qplib_qp.id) {
3748 					/* Handle this completion with
3749 					 * the stored completion
3750 					 */
3751 					memset(wc, 0, sizeof(*wc));
3752 					continue;
3753 				}
3754 				bnxt_re_process_req_wc(wc, cqe);
3755 				break;
3756 			case CQ_BASE_CQE_TYPE_RES_RAWETH_QP1:
3757 				if (!cqe->status) {
3758 					int rc = 0;
3759 
3760 					rc = bnxt_re_process_raw_qp_pkt_rx
3761 								(qp, cqe);
3762 					if (!rc) {
3763 						memset(wc, 0, sizeof(*wc));
3764 						continue;
3765 					}
3766 					cqe->status = -1;
3767 				}
3768 				/* Errors need not be looped back.
3769 				 * But change the wr_id to the one
3770 				 * stored in the table
3771 				 */
3772 				tbl_idx = cqe->wr_id;
3773 				sqp_entry = &cq->rdev->gsi_ctx.sqp_tbl[tbl_idx];
3774 				wc->wr_id = sqp_entry->wrid;
3775 				bnxt_re_process_res_rawqp1_wc(wc, cqe);
3776 				break;
3777 			case CQ_BASE_CQE_TYPE_RES_RC:
3778 				bnxt_re_process_res_rc_wc(wc, cqe);
3779 				break;
3780 			case CQ_BASE_CQE_TYPE_RES_UD:
3781 				sh_qp = qp->rdev->gsi_ctx.gsi_sqp;
3782 				if (sh_qp &&
3783 				    qp->qplib_qp.id == sh_qp->qplib_qp.id) {
3784 					/* Handle this completion with
3785 					 * the stored completion
3786 					 */
3787 					if (cqe->status) {
3788 						continue;
3789 					} else {
3790 						bnxt_re_process_res_shadow_qp_wc
3791 								(qp, wc, cqe);
3792 						break;
3793 					}
3794 				}
3795 				bnxt_re_process_res_ud_wc(qp, wc, cqe);
3796 				break;
3797 			default:
3798 				ibdev_err(&cq->rdev->ibdev,
3799 					  "POLL CQ : type 0x%x not handled",
3800 					  cqe->opcode);
3801 				continue;
3802 			}
3803 			wc++;
3804 			budget--;
3805 		}
3806 	}
3807 exit:
3808 	spin_unlock_irqrestore(&cq->cq_lock, flags);
3809 	return num_entries - budget;
3810 }
3811 
3812 int bnxt_re_req_notify_cq(struct ib_cq *ib_cq,
3813 			  enum ib_cq_notify_flags ib_cqn_flags)
3814 {
3815 	struct bnxt_re_cq *cq = container_of(ib_cq, struct bnxt_re_cq, ib_cq);
3816 	int type = 0, rc = 0;
3817 	unsigned long flags;
3818 
3819 	spin_lock_irqsave(&cq->cq_lock, flags);
3820 	/* Trigger on the very next completion */
3821 	if (ib_cqn_flags & IB_CQ_NEXT_COMP)
3822 		type = DBC_DBC_TYPE_CQ_ARMALL;
3823 	/* Trigger on the next solicited completion */
3824 	else if (ib_cqn_flags & IB_CQ_SOLICITED)
3825 		type = DBC_DBC_TYPE_CQ_ARMSE;
3826 
3827 	/* Poll to see if there are missed events */
3828 	if ((ib_cqn_flags & IB_CQ_REPORT_MISSED_EVENTS) &&
3829 	    !(bnxt_qplib_is_cq_empty(&cq->qplib_cq))) {
3830 		rc = 1;
3831 		goto exit;
3832 	}
3833 	bnxt_qplib_req_notify_cq(&cq->qplib_cq, type);
3834 
3835 exit:
3836 	spin_unlock_irqrestore(&cq->cq_lock, flags);
3837 	return rc;
3838 }
3839 
3840 /* Memory Regions */
3841 struct ib_mr *bnxt_re_get_dma_mr(struct ib_pd *ib_pd, int mr_access_flags)
3842 {
3843 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
3844 	struct bnxt_re_dev *rdev = pd->rdev;
3845 	struct bnxt_re_mr *mr;
3846 	u32 active_mrs;
3847 	int rc;
3848 
3849 	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
3850 	if (!mr)
3851 		return ERR_PTR(-ENOMEM);
3852 
3853 	mr->rdev = rdev;
3854 	mr->qplib_mr.pd = &pd->qplib_pd;
3855 	mr->qplib_mr.flags = __from_ib_access_flags(mr_access_flags);
3856 	mr->qplib_mr.type = CMDQ_ALLOCATE_MRW_MRW_FLAGS_PMR;
3857 
3858 	/* Allocate and register 0 as the address */
3859 	rc = bnxt_qplib_alloc_mrw(&rdev->qplib_res, &mr->qplib_mr);
3860 	if (rc)
3861 		goto fail;
3862 
3863 	mr->qplib_mr.hwq.level = PBL_LVL_MAX;
3864 	mr->qplib_mr.total_size = -1; /* Infinte length */
3865 	rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, NULL, 0,
3866 			       PAGE_SIZE);
3867 	if (rc)
3868 		goto fail_mr;
3869 
3870 	mr->ib_mr.lkey = mr->qplib_mr.lkey;
3871 	if (mr_access_flags & (IB_ACCESS_REMOTE_WRITE | IB_ACCESS_REMOTE_READ |
3872 			       IB_ACCESS_REMOTE_ATOMIC))
3873 		mr->ib_mr.rkey = mr->ib_mr.lkey;
3874 	active_mrs = atomic_inc_return(&rdev->stats.res.mr_count);
3875 	if (active_mrs > rdev->stats.res.mr_watermark)
3876 		rdev->stats.res.mr_watermark = active_mrs;
3877 
3878 	return &mr->ib_mr;
3879 
3880 fail_mr:
3881 	bnxt_qplib_free_mrw(&rdev->qplib_res, &mr->qplib_mr);
3882 fail:
3883 	kfree(mr);
3884 	return ERR_PTR(rc);
3885 }
3886 
3887 int bnxt_re_dereg_mr(struct ib_mr *ib_mr, struct ib_udata *udata)
3888 {
3889 	struct bnxt_re_mr *mr = container_of(ib_mr, struct bnxt_re_mr, ib_mr);
3890 	struct bnxt_re_dev *rdev = mr->rdev;
3891 	int rc;
3892 
3893 	rc = bnxt_qplib_free_mrw(&rdev->qplib_res, &mr->qplib_mr);
3894 	if (rc) {
3895 		ibdev_err(&rdev->ibdev, "Dereg MR failed: %#x\n", rc);
3896 		return rc;
3897 	}
3898 
3899 	if (mr->pages) {
3900 		rc = bnxt_qplib_free_fast_reg_page_list(&rdev->qplib_res,
3901 							&mr->qplib_frpl);
3902 		kfree(mr->pages);
3903 		mr->npages = 0;
3904 		mr->pages = NULL;
3905 	}
3906 	ib_umem_release(mr->ib_umem);
3907 
3908 	kfree(mr);
3909 	atomic_dec(&rdev->stats.res.mr_count);
3910 	return rc;
3911 }
3912 
3913 static int bnxt_re_set_page(struct ib_mr *ib_mr, u64 addr)
3914 {
3915 	struct bnxt_re_mr *mr = container_of(ib_mr, struct bnxt_re_mr, ib_mr);
3916 
3917 	if (unlikely(mr->npages == mr->qplib_frpl.max_pg_ptrs))
3918 		return -ENOMEM;
3919 
3920 	mr->pages[mr->npages++] = addr;
3921 	return 0;
3922 }
3923 
3924 int bnxt_re_map_mr_sg(struct ib_mr *ib_mr, struct scatterlist *sg, int sg_nents,
3925 		      unsigned int *sg_offset)
3926 {
3927 	struct bnxt_re_mr *mr = container_of(ib_mr, struct bnxt_re_mr, ib_mr);
3928 
3929 	mr->npages = 0;
3930 	return ib_sg_to_pages(ib_mr, sg, sg_nents, sg_offset, bnxt_re_set_page);
3931 }
3932 
3933 struct ib_mr *bnxt_re_alloc_mr(struct ib_pd *ib_pd, enum ib_mr_type type,
3934 			       u32 max_num_sg)
3935 {
3936 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
3937 	struct bnxt_re_dev *rdev = pd->rdev;
3938 	struct bnxt_re_mr *mr = NULL;
3939 	u32 active_mrs;
3940 	int rc;
3941 
3942 	if (type != IB_MR_TYPE_MEM_REG) {
3943 		ibdev_dbg(&rdev->ibdev, "MR type 0x%x not supported", type);
3944 		return ERR_PTR(-EINVAL);
3945 	}
3946 	if (max_num_sg > MAX_PBL_LVL_1_PGS)
3947 		return ERR_PTR(-EINVAL);
3948 
3949 	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
3950 	if (!mr)
3951 		return ERR_PTR(-ENOMEM);
3952 
3953 	mr->rdev = rdev;
3954 	mr->qplib_mr.pd = &pd->qplib_pd;
3955 	mr->qplib_mr.flags = BNXT_QPLIB_FR_PMR;
3956 	mr->qplib_mr.type = CMDQ_ALLOCATE_MRW_MRW_FLAGS_PMR;
3957 
3958 	rc = bnxt_qplib_alloc_mrw(&rdev->qplib_res, &mr->qplib_mr);
3959 	if (rc)
3960 		goto bail;
3961 
3962 	mr->ib_mr.lkey = mr->qplib_mr.lkey;
3963 	mr->ib_mr.rkey = mr->ib_mr.lkey;
3964 
3965 	mr->pages = kcalloc(max_num_sg, sizeof(u64), GFP_KERNEL);
3966 	if (!mr->pages) {
3967 		rc = -ENOMEM;
3968 		goto fail;
3969 	}
3970 	rc = bnxt_qplib_alloc_fast_reg_page_list(&rdev->qplib_res,
3971 						 &mr->qplib_frpl, max_num_sg);
3972 	if (rc) {
3973 		ibdev_err(&rdev->ibdev,
3974 			  "Failed to allocate HW FR page list");
3975 		goto fail_mr;
3976 	}
3977 
3978 	active_mrs = atomic_inc_return(&rdev->stats.res.mr_count);
3979 	if (active_mrs > rdev->stats.res.mr_watermark)
3980 		rdev->stats.res.mr_watermark = active_mrs;
3981 	return &mr->ib_mr;
3982 
3983 fail_mr:
3984 	kfree(mr->pages);
3985 fail:
3986 	bnxt_qplib_free_mrw(&rdev->qplib_res, &mr->qplib_mr);
3987 bail:
3988 	kfree(mr);
3989 	return ERR_PTR(rc);
3990 }
3991 
3992 struct ib_mw *bnxt_re_alloc_mw(struct ib_pd *ib_pd, enum ib_mw_type type,
3993 			       struct ib_udata *udata)
3994 {
3995 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
3996 	struct bnxt_re_dev *rdev = pd->rdev;
3997 	struct bnxt_re_mw *mw;
3998 	u32 active_mws;
3999 	int rc;
4000 
4001 	mw = kzalloc(sizeof(*mw), GFP_KERNEL);
4002 	if (!mw)
4003 		return ERR_PTR(-ENOMEM);
4004 	mw->rdev = rdev;
4005 	mw->qplib_mw.pd = &pd->qplib_pd;
4006 
4007 	mw->qplib_mw.type = (type == IB_MW_TYPE_1 ?
4008 			       CMDQ_ALLOCATE_MRW_MRW_FLAGS_MW_TYPE1 :
4009 			       CMDQ_ALLOCATE_MRW_MRW_FLAGS_MW_TYPE2B);
4010 	rc = bnxt_qplib_alloc_mrw(&rdev->qplib_res, &mw->qplib_mw);
4011 	if (rc) {
4012 		ibdev_err(&rdev->ibdev, "Allocate MW failed!");
4013 		goto fail;
4014 	}
4015 	mw->ib_mw.rkey = mw->qplib_mw.rkey;
4016 
4017 	active_mws = atomic_inc_return(&rdev->stats.res.mw_count);
4018 	if (active_mws > rdev->stats.res.mw_watermark)
4019 		rdev->stats.res.mw_watermark = active_mws;
4020 	return &mw->ib_mw;
4021 
4022 fail:
4023 	kfree(mw);
4024 	return ERR_PTR(rc);
4025 }
4026 
4027 int bnxt_re_dealloc_mw(struct ib_mw *ib_mw)
4028 {
4029 	struct bnxt_re_mw *mw = container_of(ib_mw, struct bnxt_re_mw, ib_mw);
4030 	struct bnxt_re_dev *rdev = mw->rdev;
4031 	int rc;
4032 
4033 	rc = bnxt_qplib_free_mrw(&rdev->qplib_res, &mw->qplib_mw);
4034 	if (rc) {
4035 		ibdev_err(&rdev->ibdev, "Free MW failed: %#x\n", rc);
4036 		return rc;
4037 	}
4038 
4039 	kfree(mw);
4040 	atomic_dec(&rdev->stats.res.mw_count);
4041 	return rc;
4042 }
4043 
4044 static struct ib_mr *__bnxt_re_user_reg_mr(struct ib_pd *ib_pd, u64 length, u64 virt_addr,
4045 					   int mr_access_flags, struct ib_umem *umem)
4046 {
4047 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
4048 	struct bnxt_re_dev *rdev = pd->rdev;
4049 	unsigned long page_size;
4050 	struct bnxt_re_mr *mr;
4051 	int umem_pgs, rc;
4052 	u32 active_mrs;
4053 
4054 	if (length > BNXT_RE_MAX_MR_SIZE) {
4055 		ibdev_err(&rdev->ibdev, "MR Size: %lld > Max supported:%lld\n",
4056 			  length, BNXT_RE_MAX_MR_SIZE);
4057 		return ERR_PTR(-ENOMEM);
4058 	}
4059 
4060 	page_size = ib_umem_find_best_pgsz(umem, BNXT_RE_PAGE_SIZE_SUPPORTED, virt_addr);
4061 	if (!page_size) {
4062 		ibdev_err(&rdev->ibdev, "umem page size unsupported!");
4063 		return ERR_PTR(-EINVAL);
4064 	}
4065 
4066 	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
4067 	if (!mr)
4068 		return ERR_PTR(-ENOMEM);
4069 
4070 	mr->rdev = rdev;
4071 	mr->qplib_mr.pd = &pd->qplib_pd;
4072 	mr->qplib_mr.flags = __from_ib_access_flags(mr_access_flags);
4073 	mr->qplib_mr.type = CMDQ_ALLOCATE_MRW_MRW_FLAGS_MR;
4074 
4075 	rc = bnxt_qplib_alloc_mrw(&rdev->qplib_res, &mr->qplib_mr);
4076 	if (rc) {
4077 		ibdev_err(&rdev->ibdev, "Failed to allocate MR rc = %d", rc);
4078 		rc = -EIO;
4079 		goto free_mr;
4080 	}
4081 	/* The fixed portion of the rkey is the same as the lkey */
4082 	mr->ib_mr.rkey = mr->qplib_mr.rkey;
4083 	mr->ib_umem = umem;
4084 	mr->qplib_mr.va = virt_addr;
4085 	mr->qplib_mr.total_size = length;
4086 
4087 	umem_pgs = ib_umem_num_dma_blocks(umem, page_size);
4088 	rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, umem,
4089 			       umem_pgs, page_size);
4090 	if (rc) {
4091 		ibdev_err(&rdev->ibdev, "Failed to register user MR - rc = %d\n", rc);
4092 		rc = -EIO;
4093 		goto free_mrw;
4094 	}
4095 
4096 	mr->ib_mr.lkey = mr->qplib_mr.lkey;
4097 	mr->ib_mr.rkey = mr->qplib_mr.lkey;
4098 	active_mrs = atomic_inc_return(&rdev->stats.res.mr_count);
4099 	if (active_mrs > rdev->stats.res.mr_watermark)
4100 		rdev->stats.res.mr_watermark = active_mrs;
4101 
4102 	return &mr->ib_mr;
4103 
4104 free_mrw:
4105 	bnxt_qplib_free_mrw(&rdev->qplib_res, &mr->qplib_mr);
4106 free_mr:
4107 	kfree(mr);
4108 	return ERR_PTR(rc);
4109 }
4110 
4111 struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length,
4112 				  u64 virt_addr, int mr_access_flags,
4113 				  struct ib_udata *udata)
4114 {
4115 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
4116 	struct bnxt_re_dev *rdev = pd->rdev;
4117 	struct ib_umem *umem;
4118 	struct ib_mr *ib_mr;
4119 
4120 	umem = ib_umem_get(&rdev->ibdev, start, length, mr_access_flags);
4121 	if (IS_ERR(umem))
4122 		return ERR_CAST(umem);
4123 
4124 	ib_mr = __bnxt_re_user_reg_mr(ib_pd, length, virt_addr, mr_access_flags, umem);
4125 	if (IS_ERR(ib_mr))
4126 		ib_umem_release(umem);
4127 	return ib_mr;
4128 }
4129 
4130 struct ib_mr *bnxt_re_reg_user_mr_dmabuf(struct ib_pd *ib_pd, u64 start,
4131 					 u64 length, u64 virt_addr, int fd,
4132 					 int mr_access_flags, struct ib_udata *udata)
4133 {
4134 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
4135 	struct bnxt_re_dev *rdev = pd->rdev;
4136 	struct ib_umem_dmabuf *umem_dmabuf;
4137 	struct ib_umem *umem;
4138 	struct ib_mr *ib_mr;
4139 
4140 	umem_dmabuf = ib_umem_dmabuf_get_pinned(&rdev->ibdev, start, length,
4141 						fd, mr_access_flags);
4142 	if (IS_ERR(umem_dmabuf))
4143 		return ERR_CAST(umem_dmabuf);
4144 
4145 	umem = &umem_dmabuf->umem;
4146 
4147 	ib_mr = __bnxt_re_user_reg_mr(ib_pd, length, virt_addr, mr_access_flags, umem);
4148 	if (IS_ERR(ib_mr))
4149 		ib_umem_release(umem);
4150 	return ib_mr;
4151 }
4152 
4153 int bnxt_re_alloc_ucontext(struct ib_ucontext *ctx, struct ib_udata *udata)
4154 {
4155 	struct ib_device *ibdev = ctx->device;
4156 	struct bnxt_re_ucontext *uctx =
4157 		container_of(ctx, struct bnxt_re_ucontext, ib_uctx);
4158 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibdev, ibdev);
4159 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
4160 	struct bnxt_re_user_mmap_entry *entry;
4161 	struct bnxt_re_uctx_resp resp = {};
4162 	struct bnxt_re_uctx_req ureq = {};
4163 	u32 chip_met_rev_num = 0;
4164 	int rc;
4165 
4166 	ibdev_dbg(ibdev, "ABI version requested %u", ibdev->ops.uverbs_abi_ver);
4167 
4168 	if (ibdev->ops.uverbs_abi_ver != BNXT_RE_ABI_VERSION) {
4169 		ibdev_dbg(ibdev, " is different from the device %d ",
4170 			  BNXT_RE_ABI_VERSION);
4171 		return -EPERM;
4172 	}
4173 
4174 	uctx->rdev = rdev;
4175 
4176 	uctx->shpg = (void *)__get_free_page(GFP_KERNEL);
4177 	if (!uctx->shpg) {
4178 		rc = -ENOMEM;
4179 		goto fail;
4180 	}
4181 	spin_lock_init(&uctx->sh_lock);
4182 
4183 	resp.comp_mask = BNXT_RE_UCNTX_CMASK_HAVE_CCTX;
4184 	chip_met_rev_num = rdev->chip_ctx->chip_num;
4185 	chip_met_rev_num |= ((u32)rdev->chip_ctx->chip_rev & 0xFF) <<
4186 			     BNXT_RE_CHIP_ID0_CHIP_REV_SFT;
4187 	chip_met_rev_num |= ((u32)rdev->chip_ctx->chip_metal & 0xFF) <<
4188 			     BNXT_RE_CHIP_ID0_CHIP_MET_SFT;
4189 	resp.chip_id0 = chip_met_rev_num;
4190 	/*Temp, Use xa_alloc instead */
4191 	resp.dev_id = rdev->en_dev->pdev->devfn;
4192 	resp.max_qp = rdev->qplib_ctx.qpc_count;
4193 	resp.pg_size = PAGE_SIZE;
4194 	resp.cqe_sz = sizeof(struct cq_base);
4195 	resp.max_cqd = dev_attr->max_cq_wqes;
4196 
4197 	if (rdev->chip_ctx->modes.db_push)
4198 		resp.comp_mask |= BNXT_RE_UCNTX_CMASK_WC_DPI_ENABLED;
4199 
4200 	entry = bnxt_re_mmap_entry_insert(uctx, 0, BNXT_RE_MMAP_SH_PAGE, NULL);
4201 	if (!entry) {
4202 		rc = -ENOMEM;
4203 		goto cfail;
4204 	}
4205 	uctx->shpage_mmap = &entry->rdma_entry;
4206 	if (rdev->pacing.dbr_pacing)
4207 		resp.comp_mask |= BNXT_RE_UCNTX_CMASK_DBR_PACING_ENABLED;
4208 
4209 	if (_is_host_msn_table(rdev->qplib_res.dattr->dev_cap_flags2))
4210 		resp.comp_mask |= BNXT_RE_UCNTX_CMASK_MSN_TABLE_ENABLED;
4211 
4212 	if (udata->inlen >= sizeof(ureq)) {
4213 		rc = ib_copy_from_udata(&ureq, udata, min(udata->inlen, sizeof(ureq)));
4214 		if (rc)
4215 			goto cfail;
4216 		if (ureq.comp_mask & BNXT_RE_COMP_MASK_REQ_UCNTX_POW2_SUPPORT) {
4217 			resp.comp_mask |= BNXT_RE_UCNTX_CMASK_POW2_DISABLED;
4218 			uctx->cmask |= BNXT_RE_UCNTX_CMASK_POW2_DISABLED;
4219 		}
4220 	}
4221 
4222 	rc = ib_copy_to_udata(udata, &resp, min(udata->outlen, sizeof(resp)));
4223 	if (rc) {
4224 		ibdev_err(ibdev, "Failed to copy user context");
4225 		rc = -EFAULT;
4226 		goto cfail;
4227 	}
4228 
4229 	return 0;
4230 cfail:
4231 	free_page((unsigned long)uctx->shpg);
4232 	uctx->shpg = NULL;
4233 fail:
4234 	return rc;
4235 }
4236 
4237 void bnxt_re_dealloc_ucontext(struct ib_ucontext *ib_uctx)
4238 {
4239 	struct bnxt_re_ucontext *uctx = container_of(ib_uctx,
4240 						   struct bnxt_re_ucontext,
4241 						   ib_uctx);
4242 
4243 	struct bnxt_re_dev *rdev = uctx->rdev;
4244 
4245 	rdma_user_mmap_entry_remove(uctx->shpage_mmap);
4246 	uctx->shpage_mmap = NULL;
4247 	if (uctx->shpg)
4248 		free_page((unsigned long)uctx->shpg);
4249 
4250 	if (uctx->dpi.dbr) {
4251 		/* Free DPI only if this is the first PD allocated by the
4252 		 * application and mark the context dpi as NULL
4253 		 */
4254 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res, &uctx->dpi);
4255 		uctx->dpi.dbr = NULL;
4256 	}
4257 }
4258 
4259 static struct bnxt_re_cq *bnxt_re_search_for_cq(struct bnxt_re_dev *rdev, u32 cq_id)
4260 {
4261 	struct bnxt_re_cq *cq = NULL, *tmp_cq;
4262 
4263 	hash_for_each_possible(rdev->cq_hash, tmp_cq, hash_entry, cq_id) {
4264 		if (tmp_cq->qplib_cq.id == cq_id) {
4265 			cq = tmp_cq;
4266 			break;
4267 		}
4268 	}
4269 	return cq;
4270 }
4271 
4272 /* Helper function to mmap the virtual memory from user app */
4273 int bnxt_re_mmap(struct ib_ucontext *ib_uctx, struct vm_area_struct *vma)
4274 {
4275 	struct bnxt_re_ucontext *uctx = container_of(ib_uctx,
4276 						   struct bnxt_re_ucontext,
4277 						   ib_uctx);
4278 	struct bnxt_re_user_mmap_entry *bnxt_entry;
4279 	struct rdma_user_mmap_entry *rdma_entry;
4280 	int ret = 0;
4281 	u64 pfn;
4282 
4283 	rdma_entry = rdma_user_mmap_entry_get(&uctx->ib_uctx, vma);
4284 	if (!rdma_entry)
4285 		return -EINVAL;
4286 
4287 	bnxt_entry = container_of(rdma_entry, struct bnxt_re_user_mmap_entry,
4288 				  rdma_entry);
4289 
4290 	switch (bnxt_entry->mmap_flag) {
4291 	case BNXT_RE_MMAP_WC_DB:
4292 		pfn = bnxt_entry->mem_offset >> PAGE_SHIFT;
4293 		ret = rdma_user_mmap_io(ib_uctx, vma, pfn, PAGE_SIZE,
4294 					pgprot_writecombine(vma->vm_page_prot),
4295 					rdma_entry);
4296 		break;
4297 	case BNXT_RE_MMAP_UC_DB:
4298 		pfn = bnxt_entry->mem_offset >> PAGE_SHIFT;
4299 		ret = rdma_user_mmap_io(ib_uctx, vma, pfn, PAGE_SIZE,
4300 					pgprot_noncached(vma->vm_page_prot),
4301 				rdma_entry);
4302 		break;
4303 	case BNXT_RE_MMAP_SH_PAGE:
4304 		ret = vm_insert_page(vma, vma->vm_start, virt_to_page(uctx->shpg));
4305 		break;
4306 	case BNXT_RE_MMAP_DBR_BAR:
4307 		pfn = bnxt_entry->mem_offset >> PAGE_SHIFT;
4308 		ret = rdma_user_mmap_io(ib_uctx, vma, pfn, PAGE_SIZE,
4309 					pgprot_noncached(vma->vm_page_prot),
4310 					rdma_entry);
4311 		break;
4312 	case BNXT_RE_MMAP_DBR_PAGE:
4313 	case BNXT_RE_MMAP_TOGGLE_PAGE:
4314 		/* Driver doesn't expect write access for user space */
4315 		if (vma->vm_flags & VM_WRITE)
4316 			return -EFAULT;
4317 		ret = vm_insert_page(vma, vma->vm_start,
4318 				     virt_to_page((void *)bnxt_entry->mem_offset));
4319 		break;
4320 	default:
4321 		ret = -EINVAL;
4322 		break;
4323 	}
4324 
4325 	rdma_user_mmap_entry_put(rdma_entry);
4326 	return ret;
4327 }
4328 
4329 void bnxt_re_mmap_free(struct rdma_user_mmap_entry *rdma_entry)
4330 {
4331 	struct bnxt_re_user_mmap_entry *bnxt_entry;
4332 
4333 	bnxt_entry = container_of(rdma_entry, struct bnxt_re_user_mmap_entry,
4334 				  rdma_entry);
4335 
4336 	kfree(bnxt_entry);
4337 }
4338 
4339 static int UVERBS_HANDLER(BNXT_RE_METHOD_NOTIFY_DRV)(struct uverbs_attr_bundle *attrs)
4340 {
4341 	struct bnxt_re_ucontext *uctx;
4342 
4343 	uctx = container_of(ib_uverbs_get_ucontext(attrs), struct bnxt_re_ucontext, ib_uctx);
4344 	bnxt_re_pacing_alert(uctx->rdev);
4345 	return 0;
4346 }
4347 
4348 static int UVERBS_HANDLER(BNXT_RE_METHOD_ALLOC_PAGE)(struct uverbs_attr_bundle *attrs)
4349 {
4350 	struct ib_uobject *uobj = uverbs_attr_get_uobject(attrs, BNXT_RE_ALLOC_PAGE_HANDLE);
4351 	enum bnxt_re_alloc_page_type alloc_type;
4352 	struct bnxt_re_user_mmap_entry *entry;
4353 	enum bnxt_re_mmap_flag mmap_flag;
4354 	struct bnxt_qplib_chip_ctx *cctx;
4355 	struct bnxt_re_ucontext *uctx;
4356 	struct bnxt_re_dev *rdev;
4357 	u64 mmap_offset;
4358 	u32 length;
4359 	u32 dpi;
4360 	u64 addr;
4361 	int err;
4362 
4363 	uctx = container_of(ib_uverbs_get_ucontext(attrs), struct bnxt_re_ucontext, ib_uctx);
4364 	if (IS_ERR(uctx))
4365 		return PTR_ERR(uctx);
4366 
4367 	err = uverbs_get_const(&alloc_type, attrs, BNXT_RE_ALLOC_PAGE_TYPE);
4368 	if (err)
4369 		return err;
4370 
4371 	rdev = uctx->rdev;
4372 	cctx = rdev->chip_ctx;
4373 
4374 	switch (alloc_type) {
4375 	case BNXT_RE_ALLOC_WC_PAGE:
4376 		if (cctx->modes.db_push)  {
4377 			if (bnxt_qplib_alloc_dpi(&rdev->qplib_res, &uctx->wcdpi,
4378 						 uctx, BNXT_QPLIB_DPI_TYPE_WC))
4379 				return -ENOMEM;
4380 			length = PAGE_SIZE;
4381 			dpi = uctx->wcdpi.dpi;
4382 			addr = (u64)uctx->wcdpi.umdbr;
4383 			mmap_flag = BNXT_RE_MMAP_WC_DB;
4384 		} else {
4385 			return -EINVAL;
4386 		}
4387 
4388 		break;
4389 	case BNXT_RE_ALLOC_DBR_BAR_PAGE:
4390 		length = PAGE_SIZE;
4391 		addr = (u64)rdev->pacing.dbr_bar_addr;
4392 		mmap_flag = BNXT_RE_MMAP_DBR_BAR;
4393 		break;
4394 
4395 	case BNXT_RE_ALLOC_DBR_PAGE:
4396 		length = PAGE_SIZE;
4397 		addr = (u64)rdev->pacing.dbr_page;
4398 		mmap_flag = BNXT_RE_MMAP_DBR_PAGE;
4399 		break;
4400 
4401 	default:
4402 		return -EOPNOTSUPP;
4403 	}
4404 
4405 	entry = bnxt_re_mmap_entry_insert(uctx, addr, mmap_flag, &mmap_offset);
4406 	if (!entry)
4407 		return -ENOMEM;
4408 
4409 	uobj->object = entry;
4410 	uverbs_finalize_uobj_create(attrs, BNXT_RE_ALLOC_PAGE_HANDLE);
4411 	err = uverbs_copy_to(attrs, BNXT_RE_ALLOC_PAGE_MMAP_OFFSET,
4412 			     &mmap_offset, sizeof(mmap_offset));
4413 	if (err)
4414 		return err;
4415 
4416 	err = uverbs_copy_to(attrs, BNXT_RE_ALLOC_PAGE_MMAP_LENGTH,
4417 			     &length, sizeof(length));
4418 	if (err)
4419 		return err;
4420 
4421 	err = uverbs_copy_to(attrs, BNXT_RE_ALLOC_PAGE_DPI,
4422 			     &dpi, sizeof(length));
4423 	if (err)
4424 		return err;
4425 
4426 	return 0;
4427 }
4428 
4429 static int alloc_page_obj_cleanup(struct ib_uobject *uobject,
4430 				  enum rdma_remove_reason why,
4431 			    struct uverbs_attr_bundle *attrs)
4432 {
4433 	struct  bnxt_re_user_mmap_entry *entry = uobject->object;
4434 	struct bnxt_re_ucontext *uctx = entry->uctx;
4435 
4436 	switch (entry->mmap_flag) {
4437 	case BNXT_RE_MMAP_WC_DB:
4438 		if (uctx && uctx->wcdpi.dbr) {
4439 			struct bnxt_re_dev *rdev = uctx->rdev;
4440 
4441 			bnxt_qplib_dealloc_dpi(&rdev->qplib_res, &uctx->wcdpi);
4442 			uctx->wcdpi.dbr = NULL;
4443 		}
4444 		break;
4445 	case BNXT_RE_MMAP_DBR_BAR:
4446 	case BNXT_RE_MMAP_DBR_PAGE:
4447 		break;
4448 	default:
4449 		goto exit;
4450 	}
4451 	rdma_user_mmap_entry_remove(&entry->rdma_entry);
4452 exit:
4453 	return 0;
4454 }
4455 
4456 DECLARE_UVERBS_NAMED_METHOD(BNXT_RE_METHOD_ALLOC_PAGE,
4457 			    UVERBS_ATTR_IDR(BNXT_RE_ALLOC_PAGE_HANDLE,
4458 					    BNXT_RE_OBJECT_ALLOC_PAGE,
4459 					    UVERBS_ACCESS_NEW,
4460 					    UA_MANDATORY),
4461 			    UVERBS_ATTR_CONST_IN(BNXT_RE_ALLOC_PAGE_TYPE,
4462 						 enum bnxt_re_alloc_page_type,
4463 						 UA_MANDATORY),
4464 			    UVERBS_ATTR_PTR_OUT(BNXT_RE_ALLOC_PAGE_MMAP_OFFSET,
4465 						UVERBS_ATTR_TYPE(u64),
4466 						UA_MANDATORY),
4467 			    UVERBS_ATTR_PTR_OUT(BNXT_RE_ALLOC_PAGE_MMAP_LENGTH,
4468 						UVERBS_ATTR_TYPE(u32),
4469 						UA_MANDATORY),
4470 			    UVERBS_ATTR_PTR_OUT(BNXT_RE_ALLOC_PAGE_DPI,
4471 						UVERBS_ATTR_TYPE(u32),
4472 						UA_MANDATORY));
4473 
4474 DECLARE_UVERBS_NAMED_METHOD_DESTROY(BNXT_RE_METHOD_DESTROY_PAGE,
4475 				    UVERBS_ATTR_IDR(BNXT_RE_DESTROY_PAGE_HANDLE,
4476 						    BNXT_RE_OBJECT_ALLOC_PAGE,
4477 						    UVERBS_ACCESS_DESTROY,
4478 						    UA_MANDATORY));
4479 
4480 DECLARE_UVERBS_NAMED_OBJECT(BNXT_RE_OBJECT_ALLOC_PAGE,
4481 			    UVERBS_TYPE_ALLOC_IDR(alloc_page_obj_cleanup),
4482 			    &UVERBS_METHOD(BNXT_RE_METHOD_ALLOC_PAGE),
4483 			    &UVERBS_METHOD(BNXT_RE_METHOD_DESTROY_PAGE));
4484 
4485 DECLARE_UVERBS_NAMED_METHOD(BNXT_RE_METHOD_NOTIFY_DRV);
4486 
4487 DECLARE_UVERBS_GLOBAL_METHODS(BNXT_RE_OBJECT_NOTIFY_DRV,
4488 			      &UVERBS_METHOD(BNXT_RE_METHOD_NOTIFY_DRV));
4489 
4490 /* Toggle MEM */
4491 static int UVERBS_HANDLER(BNXT_RE_METHOD_GET_TOGGLE_MEM)(struct uverbs_attr_bundle *attrs)
4492 {
4493 	struct ib_uobject *uobj = uverbs_attr_get_uobject(attrs, BNXT_RE_TOGGLE_MEM_HANDLE);
4494 	enum bnxt_re_mmap_flag mmap_flag = BNXT_RE_MMAP_TOGGLE_PAGE;
4495 	enum bnxt_re_get_toggle_mem_type res_type;
4496 	struct bnxt_re_user_mmap_entry *entry;
4497 	struct bnxt_re_ucontext *uctx;
4498 	struct ib_ucontext *ib_uctx;
4499 	struct bnxt_re_dev *rdev;
4500 	struct bnxt_re_cq *cq;
4501 	u64 mem_offset;
4502 	u64 addr = 0;
4503 	u32 length;
4504 	u32 offset;
4505 	u32 cq_id;
4506 	int err;
4507 
4508 	ib_uctx = ib_uverbs_get_ucontext(attrs);
4509 	if (IS_ERR(ib_uctx))
4510 		return PTR_ERR(ib_uctx);
4511 
4512 	err = uverbs_get_const(&res_type, attrs, BNXT_RE_TOGGLE_MEM_TYPE);
4513 	if (err)
4514 		return err;
4515 
4516 	uctx = container_of(ib_uctx, struct bnxt_re_ucontext, ib_uctx);
4517 	rdev = uctx->rdev;
4518 
4519 	switch (res_type) {
4520 	case BNXT_RE_CQ_TOGGLE_MEM:
4521 		err = uverbs_copy_from(&cq_id, attrs, BNXT_RE_TOGGLE_MEM_RES_ID);
4522 		if (err)
4523 			return err;
4524 
4525 		cq = bnxt_re_search_for_cq(rdev, cq_id);
4526 		if (!cq)
4527 			return -EINVAL;
4528 
4529 		length = PAGE_SIZE;
4530 		addr = (u64)cq->uctx_cq_page;
4531 		mmap_flag = BNXT_RE_MMAP_TOGGLE_PAGE;
4532 		offset = 0;
4533 		break;
4534 	case BNXT_RE_SRQ_TOGGLE_MEM:
4535 		break;
4536 
4537 	default:
4538 		return -EOPNOTSUPP;
4539 	}
4540 
4541 	entry = bnxt_re_mmap_entry_insert(uctx, addr, mmap_flag, &mem_offset);
4542 	if (!entry)
4543 		return -ENOMEM;
4544 
4545 	uobj->object = entry;
4546 	uverbs_finalize_uobj_create(attrs, BNXT_RE_TOGGLE_MEM_HANDLE);
4547 	err = uverbs_copy_to(attrs, BNXT_RE_TOGGLE_MEM_MMAP_PAGE,
4548 			     &mem_offset, sizeof(mem_offset));
4549 	if (err)
4550 		return err;
4551 
4552 	err = uverbs_copy_to(attrs, BNXT_RE_TOGGLE_MEM_MMAP_LENGTH,
4553 			     &length, sizeof(length));
4554 	if (err)
4555 		return err;
4556 
4557 	err = uverbs_copy_to(attrs, BNXT_RE_TOGGLE_MEM_MMAP_OFFSET,
4558 			     &offset, sizeof(length));
4559 	if (err)
4560 		return err;
4561 
4562 	return 0;
4563 }
4564 
4565 static int get_toggle_mem_obj_cleanup(struct ib_uobject *uobject,
4566 				      enum rdma_remove_reason why,
4567 				      struct uverbs_attr_bundle *attrs)
4568 {
4569 	struct  bnxt_re_user_mmap_entry *entry = uobject->object;
4570 
4571 	rdma_user_mmap_entry_remove(&entry->rdma_entry);
4572 	return 0;
4573 }
4574 
4575 DECLARE_UVERBS_NAMED_METHOD(BNXT_RE_METHOD_GET_TOGGLE_MEM,
4576 			    UVERBS_ATTR_IDR(BNXT_RE_TOGGLE_MEM_HANDLE,
4577 					    BNXT_RE_OBJECT_GET_TOGGLE_MEM,
4578 					    UVERBS_ACCESS_NEW,
4579 					    UA_MANDATORY),
4580 			    UVERBS_ATTR_CONST_IN(BNXT_RE_TOGGLE_MEM_TYPE,
4581 						 enum bnxt_re_get_toggle_mem_type,
4582 						 UA_MANDATORY),
4583 			    UVERBS_ATTR_PTR_IN(BNXT_RE_TOGGLE_MEM_RES_ID,
4584 					       UVERBS_ATTR_TYPE(u32),
4585 					       UA_MANDATORY),
4586 			    UVERBS_ATTR_PTR_OUT(BNXT_RE_TOGGLE_MEM_MMAP_PAGE,
4587 						UVERBS_ATTR_TYPE(u64),
4588 						UA_MANDATORY),
4589 			    UVERBS_ATTR_PTR_OUT(BNXT_RE_TOGGLE_MEM_MMAP_OFFSET,
4590 						UVERBS_ATTR_TYPE(u32),
4591 						UA_MANDATORY),
4592 			    UVERBS_ATTR_PTR_OUT(BNXT_RE_TOGGLE_MEM_MMAP_LENGTH,
4593 						UVERBS_ATTR_TYPE(u32),
4594 						UA_MANDATORY));
4595 
4596 DECLARE_UVERBS_NAMED_METHOD_DESTROY(BNXT_RE_METHOD_RELEASE_TOGGLE_MEM,
4597 				    UVERBS_ATTR_IDR(BNXT_RE_RELEASE_TOGGLE_MEM_HANDLE,
4598 						    BNXT_RE_OBJECT_GET_TOGGLE_MEM,
4599 						    UVERBS_ACCESS_DESTROY,
4600 						    UA_MANDATORY));
4601 
4602 DECLARE_UVERBS_NAMED_OBJECT(BNXT_RE_OBJECT_GET_TOGGLE_MEM,
4603 			    UVERBS_TYPE_ALLOC_IDR(get_toggle_mem_obj_cleanup),
4604 			    &UVERBS_METHOD(BNXT_RE_METHOD_GET_TOGGLE_MEM),
4605 			    &UVERBS_METHOD(BNXT_RE_METHOD_RELEASE_TOGGLE_MEM));
4606 
4607 const struct uapi_definition bnxt_re_uapi_defs[] = {
4608 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(BNXT_RE_OBJECT_ALLOC_PAGE),
4609 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(BNXT_RE_OBJECT_NOTIFY_DRV),
4610 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(BNXT_RE_OBJECT_GET_TOGGLE_MEM),
4611 	{}
4612 };
4613