xref: /freebsd/sys/dev/irdma/irdma_verbs.c (revision 29fc4075e69fd27de0cded313ac6000165d99f8b)
1 /*-
2  * SPDX-License-Identifier: GPL-2.0 or Linux-OpenIB
3  *
4  * Copyright (c) 2015 - 2022 Intel Corporation
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenFabrics.org BSD license below:
11  *
12  *   Redistribution and use in source and binary forms, with or
13  *   without modification, are permitted provided that the following
14  *   conditions are met:
15  *
16  *    - Redistributions of source code must retain the above
17  *	copyright notice, this list of conditions and the following
18  *	disclaimer.
19  *
20  *    - Redistributions in binary form must reproduce the above
21  *	copyright notice, this list of conditions and the following
22  *	disclaimer in the documentation and/or other materials
23  *	provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 /*$FreeBSD$*/
35 
36 #include "irdma_main.h"
37 
38 /**
39  * irdma_query_device - get device attributes
40  * @ibdev: device pointer from stack
41  * @props: returning device attributes
42  * @udata: user data
43  */
44 static int
45 irdma_query_device(struct ib_device *ibdev,
46 		   struct ib_device_attr *props,
47 		   struct ib_udata *udata)
48 {
49 	struct irdma_device *iwdev = to_iwdev(ibdev);
50 	struct irdma_pci_f *rf = iwdev->rf;
51 	struct pci_dev *pcidev = iwdev->rf->pcidev;
52 	struct irdma_hw_attrs *hw_attrs = &rf->sc_dev.hw_attrs;
53 
54 	if (udata->inlen || udata->outlen)
55 		return -EINVAL;
56 
57 	memset(props, 0, sizeof(*props));
58 	ether_addr_copy((u8 *)&props->sys_image_guid, IF_LLADDR(iwdev->netdev));
59 	props->fw_ver = (u64)irdma_fw_major_ver(&rf->sc_dev) << 32 |
60 	    irdma_fw_minor_ver(&rf->sc_dev);
61 	props->device_cap_flags = iwdev->device_cap_flags;
62 	props->vendor_id = pcidev->vendor;
63 	props->vendor_part_id = pcidev->device;
64 	props->hw_ver = pcidev->revision;
65 	props->page_size_cap = SZ_4K | SZ_2M | SZ_1G;
66 	props->max_mr_size = hw_attrs->max_mr_size;
67 	props->max_qp = rf->max_qp - rf->used_qps;
68 	props->max_qp_wr = hw_attrs->max_qp_wr;
69 	set_max_sge(props, rf);
70 	props->max_cq = rf->max_cq - rf->used_cqs;
71 	props->max_cqe = rf->max_cqe;
72 	props->max_mr = rf->max_mr - rf->used_mrs;
73 	props->max_mw = props->max_mr;
74 	props->max_pd = rf->max_pd - rf->used_pds;
75 	props->max_sge_rd = hw_attrs->uk_attrs.max_hw_read_sges;
76 	props->max_qp_rd_atom = hw_attrs->max_hw_ird;
77 	props->max_qp_init_rd_atom = hw_attrs->max_hw_ord;
78 	if (rdma_protocol_roce(ibdev, 1))
79 		props->max_pkeys = IRDMA_PKEY_TBL_SZ;
80 	props->max_ah = rf->max_ah;
81 	props->max_mcast_grp = rf->max_mcg;
82 	props->max_mcast_qp_attach = IRDMA_MAX_MGS_PER_CTX;
83 	props->max_total_mcast_qp_attach = rf->max_qp * IRDMA_MAX_MGS_PER_CTX;
84 	props->max_fast_reg_page_list_len = IRDMA_MAX_PAGES_PER_FMR;
85 #define HCA_CLOCK_TIMESTAMP_MASK 0x1ffff
86 	if (hw_attrs->uk_attrs.hw_rev >= IRDMA_GEN_2)
87 		props->timestamp_mask = HCA_CLOCK_TIMESTAMP_MASK;
88 
89 	return 0;
90 }
91 
92 static int
93 irdma_mmap_legacy(struct irdma_ucontext *ucontext,
94 		  struct vm_area_struct *vma)
95 {
96 	u64 pfn;
97 
98 	if (vma->vm_pgoff || vma->vm_end - vma->vm_start != PAGE_SIZE)
99 		return -EINVAL;
100 
101 	vma->vm_private_data = ucontext;
102 	pfn = ((uintptr_t)ucontext->iwdev->rf->sc_dev.hw_regs[IRDMA_DB_ADDR_OFFSET] +
103 	       pci_resource_start(ucontext->iwdev->rf->pcidev, 0)) >> PAGE_SHIFT;
104 
105 	return rdma_user_mmap_io(&ucontext->ibucontext, vma, pfn, PAGE_SIZE,
106 				 pgprot_noncached(vma->vm_page_prot), NULL);
107 }
108 
109 static void
110 irdma_mmap_free(struct rdma_user_mmap_entry *rdma_entry)
111 {
112 	struct irdma_user_mmap_entry *entry = to_irdma_mmap_entry(rdma_entry);
113 
114 	kfree(entry);
115 }
116 
117 struct rdma_user_mmap_entry *
118 irdma_user_mmap_entry_insert(struct irdma_ucontext *ucontext, u64 bar_offset,
119 			     enum irdma_mmap_flag mmap_flag, u64 *mmap_offset)
120 {
121 	struct irdma_user_mmap_entry *entry = kzalloc(sizeof(*entry), GFP_KERNEL);
122 	int ret;
123 
124 	if (!entry)
125 		return NULL;
126 
127 	entry->bar_offset = bar_offset;
128 	entry->mmap_flag = mmap_flag;
129 
130 	ret = rdma_user_mmap_entry_insert(&ucontext->ibucontext,
131 					  &entry->rdma_entry, PAGE_SIZE);
132 	if (ret) {
133 		kfree(entry);
134 		return NULL;
135 	}
136 	*mmap_offset = rdma_user_mmap_get_offset(&entry->rdma_entry);
137 
138 	return &entry->rdma_entry;
139 }
140 
141 /**
142  * irdma_mmap - user memory map
143  * @context: context created during alloc
144  * @vma: kernel info for user memory map
145  */
146 static int
147 irdma_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
148 {
149 	struct rdma_user_mmap_entry *rdma_entry;
150 	struct irdma_user_mmap_entry *entry;
151 	struct irdma_ucontext *ucontext;
152 	u64 pfn;
153 	int ret;
154 
155 	ucontext = to_ucontext(context);
156 
157 	/* Legacy support for libi40iw with hard-coded mmap key */
158 	if (ucontext->legacy_mode)
159 		return irdma_mmap_legacy(ucontext, vma);
160 
161 	rdma_entry = rdma_user_mmap_entry_get(&ucontext->ibucontext, vma);
162 	if (!rdma_entry) {
163 		irdma_debug(iwdev_to_idev(ucontext->iwdev), IRDMA_DEBUG_VERBS,
164 			    "pgoff[0x%lx] does not have valid entry\n",
165 			    vma->vm_pgoff);
166 		return -EINVAL;
167 	}
168 
169 	entry = to_irdma_mmap_entry(rdma_entry);
170 	irdma_debug(iwdev_to_idev(ucontext->iwdev), IRDMA_DEBUG_VERBS,
171 		    "bar_offset [0x%lx] mmap_flag [%d]\n", entry->bar_offset,
172 		    entry->mmap_flag);
173 
174 	pfn = (entry->bar_offset +
175 	       pci_resource_start(ucontext->iwdev->rf->pcidev, 0)) >> PAGE_SHIFT;
176 
177 	switch (entry->mmap_flag) {
178 	case IRDMA_MMAP_IO_NC:
179 		ret = rdma_user_mmap_io(context, vma, pfn, PAGE_SIZE,
180 					pgprot_noncached(vma->vm_page_prot),
181 					rdma_entry);
182 		break;
183 	case IRDMA_MMAP_IO_WC:
184 		ret = rdma_user_mmap_io(context, vma, pfn, PAGE_SIZE,
185 					pgprot_writecombine(vma->vm_page_prot),
186 					rdma_entry);
187 		break;
188 	default:
189 		ret = -EINVAL;
190 	}
191 
192 	if (ret)
193 		irdma_debug(iwdev_to_idev(ucontext->iwdev), IRDMA_DEBUG_VERBS,
194 			    "bar_offset [0x%lx] mmap_flag[%d] err[%d]\n",
195 			    entry->bar_offset, entry->mmap_flag, ret);
196 	rdma_user_mmap_entry_put(rdma_entry);
197 
198 	return ret;
199 }
200 
201 /**
202  * irdma_alloc_push_page - allocate a push page for qp
203  * @iwqp: qp pointer
204  */
205 static void
206 irdma_alloc_push_page(struct irdma_qp *iwqp)
207 {
208 	struct irdma_cqp_request *cqp_request;
209 	struct cqp_cmds_info *cqp_info;
210 	struct irdma_device *iwdev = iwqp->iwdev;
211 	struct irdma_sc_qp *qp = &iwqp->sc_qp;
212 	int status;
213 
214 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
215 	if (!cqp_request)
216 		return;
217 
218 	cqp_info = &cqp_request->info;
219 	cqp_info->cqp_cmd = IRDMA_OP_MANAGE_PUSH_PAGE;
220 	cqp_info->post_sq = 1;
221 	cqp_info->in.u.manage_push_page.info.push_idx = 0;
222 	cqp_info->in.u.manage_push_page.info.qs_handle =
223 	    qp->vsi->qos[qp->user_pri].qs_handle;
224 	cqp_info->in.u.manage_push_page.info.free_page = 0;
225 	cqp_info->in.u.manage_push_page.info.push_page_type = 0;
226 	cqp_info->in.u.manage_push_page.cqp = &iwdev->rf->cqp.sc_cqp;
227 	cqp_info->in.u.manage_push_page.scratch = (uintptr_t)cqp_request;
228 
229 	status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
230 	if (!status && cqp_request->compl_info.op_ret_val <
231 	    iwdev->rf->sc_dev.hw_attrs.max_hw_device_pages) {
232 		qp->push_idx = cqp_request->compl_info.op_ret_val;
233 		qp->push_offset = 0;
234 	}
235 
236 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
237 }
238 
239 /**
240  * irdma_get_pbl - Retrieve pbl from a list given a virtual
241  * address
242  * @va: user virtual address
243  * @pbl_list: pbl list to search in (QP's or CQ's)
244  */
245 struct irdma_pbl *
246 irdma_get_pbl(unsigned long va,
247 	      struct list_head *pbl_list)
248 {
249 	struct irdma_pbl *iwpbl;
250 
251 	list_for_each_entry(iwpbl, pbl_list, list) {
252 		if (iwpbl->user_base == va) {
253 			list_del(&iwpbl->list);
254 			iwpbl->on_list = false;
255 			return iwpbl;
256 		}
257 	}
258 
259 	return NULL;
260 }
261 
262 /**
263  * irdma_clean_cqes - clean cq entries for qp
264  * @iwqp: qp ptr (user or kernel)
265  * @iwcq: cq ptr
266  */
267 void
268 irdma_clean_cqes(struct irdma_qp *iwqp, struct irdma_cq *iwcq)
269 {
270 	struct irdma_cq_uk *ukcq = &iwcq->sc_cq.cq_uk;
271 	unsigned long flags;
272 
273 	spin_lock_irqsave(&iwcq->lock, flags);
274 	irdma_uk_clean_cq(&iwqp->sc_qp.qp_uk, ukcq);
275 	spin_unlock_irqrestore(&iwcq->lock, flags);
276 }
277 
278 void
279 irdma_remove_push_mmap_entries(struct irdma_qp *iwqp)
280 {
281 	if (iwqp->push_db_mmap_entry) {
282 		rdma_user_mmap_entry_remove(iwqp->push_db_mmap_entry);
283 		iwqp->push_db_mmap_entry = NULL;
284 	}
285 	if (iwqp->push_wqe_mmap_entry) {
286 		rdma_user_mmap_entry_remove(iwqp->push_wqe_mmap_entry);
287 		iwqp->push_wqe_mmap_entry = NULL;
288 	}
289 }
290 
291 static int
292 irdma_setup_push_mmap_entries(struct irdma_ucontext *ucontext,
293 			      struct irdma_qp *iwqp,
294 			      u64 *push_wqe_mmap_key,
295 			      u64 *push_db_mmap_key)
296 {
297 	struct irdma_device *iwdev = ucontext->iwdev;
298 	u64 bar_off;
299 
300 	WARN_ON_ONCE(iwdev->rf->sc_dev.hw_attrs.uk_attrs.hw_rev < IRDMA_GEN_2);
301 
302 	bar_off = (uintptr_t)iwdev->rf->sc_dev.hw_regs[IRDMA_DB_ADDR_OFFSET];
303 
304 	if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.hw_rev == IRDMA_GEN_2) {
305 		/* skip over db page */
306 		bar_off += IRDMA_HW_PAGE_SIZE;
307 		/* skip over reserved space */
308 		bar_off += IRDMA_PF_BAR_RSVD;
309 	}
310 
311 	/* push wqe page */
312 	bar_off += iwqp->sc_qp.push_idx * IRDMA_HW_PAGE_SIZE;
313 	iwqp->push_wqe_mmap_entry = irdma_user_mmap_entry_insert(ucontext,
314 								 bar_off, IRDMA_MMAP_IO_WC,
315 								 push_wqe_mmap_key);
316 	if (!iwqp->push_wqe_mmap_entry)
317 		return -ENOMEM;
318 
319 	/* push doorbell page */
320 	bar_off += IRDMA_HW_PAGE_SIZE;
321 	iwqp->push_db_mmap_entry = irdma_user_mmap_entry_insert(ucontext,
322 								bar_off, IRDMA_MMAP_IO_NC,
323 								push_db_mmap_key);
324 	if (!iwqp->push_db_mmap_entry) {
325 		rdma_user_mmap_entry_remove(iwqp->push_wqe_mmap_entry);
326 		return -ENOMEM;
327 	}
328 
329 	return 0;
330 }
331 
332 /**
333  * irdma_setup_virt_qp - setup for allocation of virtual qp
334  * @iwdev: irdma device
335  * @iwqp: qp ptr
336  * @init_info: initialize info to return
337  */
338 void
339 irdma_setup_virt_qp(struct irdma_device *iwdev,
340 		    struct irdma_qp *iwqp,
341 		    struct irdma_qp_init_info *init_info)
342 {
343 	struct irdma_pbl *iwpbl = iwqp->iwpbl;
344 	struct irdma_qp_mr *qpmr = &iwpbl->qp_mr;
345 
346 	iwqp->page = qpmr->sq_page;
347 	init_info->shadow_area_pa = qpmr->shadow;
348 	if (iwpbl->pbl_allocated) {
349 		init_info->virtual_map = true;
350 		init_info->sq_pa = qpmr->sq_pbl.idx;
351 		init_info->rq_pa = qpmr->rq_pbl.idx;
352 	} else {
353 		init_info->sq_pa = qpmr->sq_pbl.addr;
354 		init_info->rq_pa = qpmr->rq_pbl.addr;
355 	}
356 }
357 
358 /**
359  * irdma_setup_kmode_qp - setup initialization for kernel mode qp
360  * @iwdev: iwarp device
361  * @iwqp: qp ptr (user or kernel)
362  * @info: initialize info to return
363  * @init_attr: Initial QP create attributes
364  */
365 int
366 irdma_setup_kmode_qp(struct irdma_device *iwdev,
367 		     struct irdma_qp *iwqp,
368 		     struct irdma_qp_init_info *info,
369 		     struct ib_qp_init_attr *init_attr)
370 {
371 	struct irdma_dma_mem *mem = &iwqp->kqp.dma_mem;
372 	u32 sqdepth, rqdepth;
373 	u8 sqshift, rqshift;
374 	u32 size;
375 	int status;
376 	struct irdma_qp_uk_init_info *ukinfo = &info->qp_uk_init_info;
377 	struct irdma_uk_attrs *uk_attrs = &iwdev->rf->sc_dev.hw_attrs.uk_attrs;
378 
379 	irdma_get_wqe_shift(uk_attrs,
380 			    uk_attrs->hw_rev >= IRDMA_GEN_2 ? ukinfo->max_sq_frag_cnt + 1 :
381 			    ukinfo->max_sq_frag_cnt,
382 			    ukinfo->max_inline_data, &sqshift);
383 	status = irdma_get_sqdepth(uk_attrs->max_hw_wq_quanta, ukinfo->sq_size,
384 				   sqshift, &sqdepth);
385 	if (status)
386 		return status;
387 
388 	if (uk_attrs->hw_rev == IRDMA_GEN_1)
389 		rqshift = IRDMA_MAX_RQ_WQE_SHIFT_GEN1;
390 	else
391 		irdma_get_wqe_shift(uk_attrs, ukinfo->max_rq_frag_cnt, 0,
392 				    &rqshift);
393 
394 	status = irdma_get_rqdepth(uk_attrs->max_hw_rq_quanta, ukinfo->rq_size,
395 				   rqshift, &rqdepth);
396 	if (status)
397 		return status;
398 
399 	iwqp->kqp.sq_wrid_mem =
400 	    kcalloc(sqdepth, sizeof(*iwqp->kqp.sq_wrid_mem), GFP_KERNEL);
401 	if (!iwqp->kqp.sq_wrid_mem)
402 		return -ENOMEM;
403 
404 	iwqp->kqp.rq_wrid_mem =
405 	    kcalloc(rqdepth, sizeof(*iwqp->kqp.rq_wrid_mem), GFP_KERNEL);
406 	if (!iwqp->kqp.rq_wrid_mem) {
407 		kfree(iwqp->kqp.sq_wrid_mem);
408 		iwqp->kqp.sq_wrid_mem = NULL;
409 		return -ENOMEM;
410 	}
411 
412 	iwqp->kqp.sig_trk_mem = kcalloc(sqdepth, sizeof(u32), GFP_KERNEL);
413 	memset(iwqp->kqp.sig_trk_mem, 0, sqdepth * sizeof(u32));
414 	if (!iwqp->kqp.sig_trk_mem) {
415 		kfree(iwqp->kqp.sq_wrid_mem);
416 		iwqp->kqp.sq_wrid_mem = NULL;
417 		kfree(iwqp->kqp.rq_wrid_mem);
418 		iwqp->kqp.rq_wrid_mem = NULL;
419 		return -ENOMEM;
420 	}
421 	ukinfo->sq_sigwrtrk_array = (void *)iwqp->kqp.sig_trk_mem;
422 	ukinfo->sq_wrtrk_array = iwqp->kqp.sq_wrid_mem;
423 	ukinfo->rq_wrid_array = iwqp->kqp.rq_wrid_mem;
424 
425 	size = (sqdepth + rqdepth) * IRDMA_QP_WQE_MIN_SIZE;
426 	size += (IRDMA_SHADOW_AREA_SIZE << 3);
427 
428 	mem->size = size;
429 	mem->va = irdma_allocate_dma_mem(&iwdev->rf->hw, mem, mem->size,
430 					 256);
431 	if (!mem->va) {
432 		kfree(iwqp->kqp.sq_wrid_mem);
433 		iwqp->kqp.sq_wrid_mem = NULL;
434 		kfree(iwqp->kqp.rq_wrid_mem);
435 		iwqp->kqp.rq_wrid_mem = NULL;
436 		return -ENOMEM;
437 	}
438 
439 	ukinfo->sq = mem->va;
440 	info->sq_pa = mem->pa;
441 	ukinfo->rq = &ukinfo->sq[sqdepth];
442 	info->rq_pa = info->sq_pa + (sqdepth * IRDMA_QP_WQE_MIN_SIZE);
443 	ukinfo->shadow_area = ukinfo->rq[rqdepth].elem;
444 	info->shadow_area_pa = info->rq_pa + (rqdepth * IRDMA_QP_WQE_MIN_SIZE);
445 	ukinfo->sq_size = sqdepth >> sqshift;
446 	ukinfo->rq_size = rqdepth >> rqshift;
447 	ukinfo->qp_id = iwqp->ibqp.qp_num;
448 
449 	init_attr->cap.max_send_wr = (sqdepth - IRDMA_SQ_RSVD) >> sqshift;
450 	init_attr->cap.max_recv_wr = (rqdepth - IRDMA_RQ_RSVD) >> rqshift;
451 
452 	return 0;
453 }
454 
455 int
456 irdma_cqp_create_qp_cmd(struct irdma_qp *iwqp)
457 {
458 	struct irdma_pci_f *rf = iwqp->iwdev->rf;
459 	struct irdma_cqp_request *cqp_request;
460 	struct cqp_cmds_info *cqp_info;
461 	struct irdma_create_qp_info *qp_info;
462 	int status;
463 
464 	cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
465 	if (!cqp_request)
466 		return -ENOMEM;
467 
468 	cqp_info = &cqp_request->info;
469 	qp_info = &cqp_request->info.in.u.qp_create.info;
470 	memset(qp_info, 0, sizeof(*qp_info));
471 	qp_info->mac_valid = true;
472 	qp_info->cq_num_valid = true;
473 	qp_info->next_iwarp_state = IRDMA_QP_STATE_IDLE;
474 
475 	cqp_info->cqp_cmd = IRDMA_OP_QP_CREATE;
476 	cqp_info->post_sq = 1;
477 	cqp_info->in.u.qp_create.qp = &iwqp->sc_qp;
478 	cqp_info->in.u.qp_create.scratch = (uintptr_t)cqp_request;
479 	status = irdma_handle_cqp_op(rf, cqp_request);
480 	irdma_put_cqp_request(&rf->cqp, cqp_request);
481 
482 	return status;
483 }
484 
485 void
486 irdma_roce_fill_and_set_qpctx_info(struct irdma_qp *iwqp,
487 				   struct irdma_qp_host_ctx_info *ctx_info)
488 {
489 	struct irdma_device *iwdev = iwqp->iwdev;
490 	struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
491 	struct irdma_roce_offload_info *roce_info;
492 	struct irdma_udp_offload_info *udp_info;
493 
494 	udp_info = &iwqp->udp_info;
495 	udp_info->snd_mss = ib_mtu_enum_to_int(ib_mtu_int_to_enum(iwdev->vsi.mtu));
496 	udp_info->cwnd = iwdev->roce_cwnd;
497 	udp_info->rexmit_thresh = 2;
498 	udp_info->rnr_nak_thresh = 2;
499 	udp_info->src_port = 0xc000;
500 	udp_info->dst_port = ROCE_V2_UDP_DPORT;
501 	roce_info = &iwqp->roce_info;
502 	ether_addr_copy(roce_info->mac_addr, IF_LLADDR(iwdev->netdev));
503 
504 	roce_info->rd_en = true;
505 	roce_info->wr_rdresp_en = true;
506 	roce_info->bind_en = true;
507 	roce_info->dcqcn_en = false;
508 	roce_info->rtomin = 5;
509 
510 	roce_info->ack_credits = iwdev->roce_ackcreds;
511 	roce_info->ird_size = dev->hw_attrs.max_hw_ird;
512 	roce_info->ord_size = dev->hw_attrs.max_hw_ord;
513 
514 	if (!iwqp->user_mode) {
515 		roce_info->priv_mode_en = true;
516 		roce_info->fast_reg_en = true;
517 		roce_info->udprivcq_en = true;
518 	}
519 	roce_info->roce_tver = 0;
520 
521 	ctx_info->roce_info = &iwqp->roce_info;
522 	ctx_info->udp_info = &iwqp->udp_info;
523 	irdma_sc_qp_setctx_roce(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
524 }
525 
526 void
527 irdma_iw_fill_and_set_qpctx_info(struct irdma_qp *iwqp,
528 				 struct irdma_qp_host_ctx_info *ctx_info)
529 {
530 	struct irdma_device *iwdev = iwqp->iwdev;
531 	struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
532 	struct irdma_iwarp_offload_info *iwarp_info;
533 
534 	iwarp_info = &iwqp->iwarp_info;
535 	ether_addr_copy(iwarp_info->mac_addr, IF_LLADDR(iwdev->netdev));
536 	iwarp_info->rd_en = true;
537 	iwarp_info->wr_rdresp_en = true;
538 	iwarp_info->bind_en = true;
539 	iwarp_info->ecn_en = true;
540 	iwarp_info->rtomin = 5;
541 
542 	if (dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
543 		iwarp_info->ib_rd_en = true;
544 	if (!iwqp->user_mode) {
545 		iwarp_info->priv_mode_en = true;
546 		iwarp_info->fast_reg_en = true;
547 	}
548 	iwarp_info->ddp_ver = 1;
549 	iwarp_info->rdmap_ver = 1;
550 
551 	ctx_info->iwarp_info = &iwqp->iwarp_info;
552 	ctx_info->iwarp_info_valid = true;
553 	irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
554 	ctx_info->iwarp_info_valid = false;
555 }
556 
557 int
558 irdma_validate_qp_attrs(struct ib_qp_init_attr *init_attr,
559 			struct irdma_device *iwdev)
560 {
561 	struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
562 	struct irdma_uk_attrs *uk_attrs = &dev->hw_attrs.uk_attrs;
563 
564 	if (init_attr->create_flags)
565 		return -EOPNOTSUPP;
566 
567 	if (init_attr->cap.max_inline_data > uk_attrs->max_hw_inline ||
568 	    init_attr->cap.max_send_sge > uk_attrs->max_hw_wq_frags ||
569 	    init_attr->cap.max_recv_sge > uk_attrs->max_hw_wq_frags)
570 		return -EINVAL;
571 
572 	if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
573 		if (init_attr->qp_type != IB_QPT_RC &&
574 		    init_attr->qp_type != IB_QPT_UD &&
575 		    init_attr->qp_type != IB_QPT_GSI)
576 			return -EOPNOTSUPP;
577 	} else {
578 		if (init_attr->qp_type != IB_QPT_RC)
579 			return -EOPNOTSUPP;
580 	}
581 
582 	return 0;
583 }
584 
585 void
586 irdma_flush_worker(struct work_struct *work)
587 {
588 	struct delayed_work *dwork = to_delayed_work(work);
589 	struct irdma_qp *iwqp = container_of(dwork, struct irdma_qp, dwork_flush);
590 	unsigned long flags;
591 
592 	spin_lock_irqsave(&iwqp->lock, flags);	/* Don't allow more posting while generating completions */
593 	irdma_generate_flush_completions(iwqp);
594 	spin_unlock_irqrestore(&iwqp->lock, flags);
595 }
596 
597 static int
598 irdma_get_ib_acc_flags(struct irdma_qp *iwqp)
599 {
600 	int acc_flags = 0;
601 
602 	if (rdma_protocol_roce(iwqp->ibqp.device, 1)) {
603 		if (iwqp->roce_info.wr_rdresp_en) {
604 			acc_flags |= IB_ACCESS_LOCAL_WRITE;
605 			acc_flags |= IB_ACCESS_REMOTE_WRITE;
606 		}
607 		if (iwqp->roce_info.rd_en)
608 			acc_flags |= IB_ACCESS_REMOTE_READ;
609 		if (iwqp->roce_info.bind_en)
610 			acc_flags |= IB_ACCESS_MW_BIND;
611 	} else {
612 		if (iwqp->iwarp_info.wr_rdresp_en) {
613 			acc_flags |= IB_ACCESS_LOCAL_WRITE;
614 			acc_flags |= IB_ACCESS_REMOTE_WRITE;
615 		}
616 		if (iwqp->iwarp_info.rd_en)
617 			acc_flags |= IB_ACCESS_REMOTE_READ;
618 		if (iwqp->iwarp_info.bind_en)
619 			acc_flags |= IB_ACCESS_MW_BIND;
620 	}
621 	return acc_flags;
622 }
623 
624 /**
625  * irdma_query_qp - query qp attributes
626  * @ibqp: qp pointer
627  * @attr: attributes pointer
628  * @attr_mask: Not used
629  * @init_attr: qp attributes to return
630  */
631 static int
632 irdma_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
633 	       int attr_mask, struct ib_qp_init_attr *init_attr)
634 {
635 	struct irdma_qp *iwqp = to_iwqp(ibqp);
636 	struct irdma_sc_qp *qp = &iwqp->sc_qp;
637 
638 	memset(attr, 0, sizeof(*attr));
639 	memset(init_attr, 0, sizeof(*init_attr));
640 
641 	attr->qp_state = iwqp->ibqp_state;
642 	attr->cur_qp_state = iwqp->ibqp_state;
643 	attr->cap.max_send_wr = iwqp->max_send_wr;
644 	attr->cap.max_recv_wr = iwqp->max_recv_wr;
645 	attr->cap.max_inline_data = qp->qp_uk.max_inline_data;
646 	attr->cap.max_send_sge = qp->qp_uk.max_sq_frag_cnt;
647 	attr->cap.max_recv_sge = qp->qp_uk.max_rq_frag_cnt;
648 	attr->qp_access_flags = irdma_get_ib_acc_flags(iwqp);
649 	attr->port_num = 1;
650 	if (rdma_protocol_roce(ibqp->device, 1)) {
651 		attr->path_mtu = ib_mtu_int_to_enum(iwqp->udp_info.snd_mss);
652 		attr->qkey = iwqp->roce_info.qkey;
653 		attr->rq_psn = iwqp->udp_info.epsn;
654 		attr->sq_psn = iwqp->udp_info.psn_nxt;
655 		attr->dest_qp_num = iwqp->roce_info.dest_qp;
656 		attr->pkey_index = iwqp->roce_info.p_key;
657 		attr->retry_cnt = iwqp->udp_info.rexmit_thresh;
658 		attr->rnr_retry = iwqp->udp_info.rnr_nak_thresh;
659 		attr->max_rd_atomic = iwqp->roce_info.ord_size;
660 		attr->max_dest_rd_atomic = iwqp->roce_info.ird_size;
661 	}
662 
663 	init_attr->event_handler = iwqp->ibqp.event_handler;
664 	init_attr->qp_context = iwqp->ibqp.qp_context;
665 	init_attr->send_cq = iwqp->ibqp.send_cq;
666 	init_attr->recv_cq = iwqp->ibqp.recv_cq;
667 	init_attr->cap = attr->cap;
668 
669 	return 0;
670 }
671 
672 /**
673  * irdma_modify_qp_roce - modify qp request
674  * @ibqp: qp's pointer for modify
675  * @attr: access attributes
676  * @attr_mask: state mask
677  * @udata: user data
678  */
679 int
680 irdma_modify_qp_roce(struct ib_qp *ibqp, struct ib_qp_attr *attr,
681 		     int attr_mask, struct ib_udata *udata)
682 {
683 	struct irdma_pd *iwpd = to_iwpd(ibqp->pd);
684 	struct irdma_qp *iwqp = to_iwqp(ibqp);
685 	struct irdma_device *iwdev = iwqp->iwdev;
686 	struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
687 	struct irdma_qp_host_ctx_info *ctx_info;
688 	struct irdma_roce_offload_info *roce_info;
689 	struct irdma_udp_offload_info *udp_info;
690 	struct irdma_modify_qp_info info = {0};
691 	struct irdma_modify_qp_resp uresp = {};
692 	struct irdma_modify_qp_req ureq = {};
693 	unsigned long flags;
694 	u8 issue_modify_qp = 0;
695 	int ret = 0;
696 
697 	ctx_info = &iwqp->ctx_info;
698 	roce_info = &iwqp->roce_info;
699 	udp_info = &iwqp->udp_info;
700 
701 	if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
702 		return -EOPNOTSUPP;
703 
704 	if (attr_mask & IB_QP_DEST_QPN)
705 		roce_info->dest_qp = attr->dest_qp_num;
706 
707 	if (attr_mask & IB_QP_PKEY_INDEX) {
708 		ret = irdma_query_pkey(ibqp->device, 0, attr->pkey_index,
709 				       &roce_info->p_key);
710 		if (ret)
711 			return ret;
712 	}
713 
714 	if (attr_mask & IB_QP_QKEY)
715 		roce_info->qkey = attr->qkey;
716 
717 	if (attr_mask & IB_QP_PATH_MTU)
718 		udp_info->snd_mss = ib_mtu_enum_to_int(attr->path_mtu);
719 
720 	if (attr_mask & IB_QP_SQ_PSN) {
721 		udp_info->psn_nxt = attr->sq_psn;
722 		udp_info->lsn = 0xffff;
723 		udp_info->psn_una = attr->sq_psn;
724 		udp_info->psn_max = attr->sq_psn;
725 	}
726 
727 	if (attr_mask & IB_QP_RQ_PSN)
728 		udp_info->epsn = attr->rq_psn;
729 
730 	if (attr_mask & IB_QP_RNR_RETRY)
731 		udp_info->rnr_nak_thresh = attr->rnr_retry;
732 
733 	if (attr_mask & IB_QP_RETRY_CNT)
734 		udp_info->rexmit_thresh = attr->retry_cnt;
735 
736 	ctx_info->roce_info->pd_id = iwpd->sc_pd.pd_id;
737 
738 	if (attr_mask & IB_QP_AV) {
739 		struct irdma_av *av = &iwqp->roce_ah.av;
740 		u16 vlan_id = VLAN_N_VID;
741 		u32 local_ip[4] = {};
742 
743 		memset(&iwqp->roce_ah, 0, sizeof(iwqp->roce_ah));
744 		if (attr->ah_attr.ah_flags & IB_AH_GRH) {
745 			udp_info->ttl = attr->ah_attr.grh.hop_limit;
746 			udp_info->flow_label = attr->ah_attr.grh.flow_label;
747 			udp_info->tos = attr->ah_attr.grh.traffic_class;
748 			irdma_qp_rem_qos(&iwqp->sc_qp);
749 			dev->ws_remove(iwqp->sc_qp.vsi, ctx_info->user_pri);
750 			if (iwqp->sc_qp.vsi->dscp_mode)
751 				ctx_info->user_pri =
752 				    iwqp->sc_qp.vsi->dscp_map[irdma_tos2dscp(udp_info->tos)];
753 			else
754 				ctx_info->user_pri = rt_tos2priority(udp_info->tos);
755 			iwqp->sc_qp.user_pri = ctx_info->user_pri;
756 			if (dev->ws_add(iwqp->sc_qp.vsi, ctx_info->user_pri))
757 				return -ENOMEM;
758 			irdma_qp_add_qos(&iwqp->sc_qp);
759 		}
760 		ret = kc_irdma_set_roce_cm_info(iwqp, attr, &vlan_id);
761 		if (ret)
762 			return ret;
763 
764 		if (vlan_id >= VLAN_N_VID && iwdev->dcb_vlan_mode)
765 			vlan_id = 0;
766 		if (vlan_id < VLAN_N_VID) {
767 			udp_info->insert_vlan_tag = true;
768 			udp_info->vlan_tag = vlan_id |
769 			    ctx_info->user_pri << VLAN_PRIO_SHIFT;
770 		} else {
771 			udp_info->insert_vlan_tag = false;
772 		}
773 
774 		av->attrs = attr->ah_attr;
775 		rdma_gid2ip((struct sockaddr *)&av->dgid_addr, &attr->ah_attr.grh.dgid);
776 		roce_info->local_qp = ibqp->qp_num;
777 		if (av->sgid_addr.saddr.sa_family == AF_INET6) {
778 			__be32 *daddr =
779 			av->dgid_addr.saddr_in6.sin6_addr.__u6_addr.__u6_addr32;
780 			__be32 *saddr =
781 			av->sgid_addr.saddr_in6.sin6_addr.__u6_addr.__u6_addr32;
782 
783 			irdma_copy_ip_ntohl(&udp_info->dest_ip_addr[0], daddr);
784 			irdma_copy_ip_ntohl(&udp_info->local_ipaddr[0], saddr);
785 
786 			udp_info->ipv4 = false;
787 			irdma_copy_ip_ntohl(local_ip, daddr);
788 
789 			udp_info->arp_idx = irdma_arp_table(iwdev->rf, local_ip,
790 							    NULL, IRDMA_ARP_RESOLVE);
791 		} else {
792 			__be32 saddr = av->sgid_addr.saddr_in.sin_addr.s_addr;
793 			__be32 daddr = av->dgid_addr.saddr_in.sin_addr.s_addr;
794 
795 			local_ip[0] = ntohl(daddr);
796 
797 			udp_info->ipv4 = true;
798 			udp_info->dest_ip_addr[0] = 0;
799 			udp_info->dest_ip_addr[1] = 0;
800 			udp_info->dest_ip_addr[2] = 0;
801 			udp_info->dest_ip_addr[3] = local_ip[0];
802 
803 			udp_info->local_ipaddr[0] = 0;
804 			udp_info->local_ipaddr[1] = 0;
805 			udp_info->local_ipaddr[2] = 0;
806 			udp_info->local_ipaddr[3] = ntohl(saddr);
807 		}
808 		udp_info->arp_idx =
809 		    irdma_add_arp(iwdev->rf, local_ip,
810 				  ah_attr_to_dmac(attr->ah_attr));
811 	}
812 
813 	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
814 		if (attr->max_rd_atomic > dev->hw_attrs.max_hw_ord) {
815 			ibdev_err(&iwdev->ibdev,
816 				  "rd_atomic = %d, above max_hw_ord=%d\n",
817 				  attr->max_rd_atomic,
818 				  dev->hw_attrs.max_hw_ord);
819 			return -EINVAL;
820 		}
821 		if (attr->max_rd_atomic)
822 			roce_info->ord_size = attr->max_rd_atomic;
823 		info.ord_valid = true;
824 	}
825 
826 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
827 		if (attr->max_dest_rd_atomic > dev->hw_attrs.max_hw_ird) {
828 			ibdev_err(&iwdev->ibdev,
829 				  "rd_atomic = %d, above max_hw_ird=%d\n",
830 				  attr->max_rd_atomic,
831 				  dev->hw_attrs.max_hw_ird);
832 			return -EINVAL;
833 		}
834 		if (attr->max_dest_rd_atomic)
835 			roce_info->ird_size = attr->max_dest_rd_atomic;
836 	}
837 
838 	if (attr_mask & IB_QP_ACCESS_FLAGS) {
839 		if (attr->qp_access_flags & IB_ACCESS_LOCAL_WRITE)
840 			roce_info->wr_rdresp_en = true;
841 		if (attr->qp_access_flags & IB_ACCESS_REMOTE_WRITE)
842 			roce_info->wr_rdresp_en = true;
843 		if (attr->qp_access_flags & IB_ACCESS_REMOTE_READ)
844 			roce_info->rd_en = true;
845 	}
846 
847 	wait_event(iwqp->mod_qp_waitq, !atomic_read(&iwqp->hw_mod_qp_pend));
848 
849 	irdma_debug(dev, IRDMA_DEBUG_VERBS,
850 		    "caller: %pS qp_id=%d to_ibqpstate=%d ibqpstate=%d irdma_qpstate=%d attr_mask=0x%x\n",
851 		    __builtin_return_address(0), ibqp->qp_num, attr->qp_state, iwqp->ibqp_state,
852 		    iwqp->iwarp_state, attr_mask);
853 
854 	spin_lock_irqsave(&iwqp->lock, flags);
855 	if (attr_mask & IB_QP_STATE) {
856 		if (!ib_modify_qp_is_ok(iwqp->ibqp_state, attr->qp_state, iwqp->ibqp.qp_type, attr_mask)) {
857 			irdma_print("modify_qp invalid for qp_id=%d, old_state=0x%x, new_state=0x%x\n",
858 				    iwqp->ibqp.qp_num, iwqp->ibqp_state,
859 				    attr->qp_state);
860 			ret = -EINVAL;
861 			goto exit;
862 		}
863 		info.curr_iwarp_state = iwqp->iwarp_state;
864 
865 		switch (attr->qp_state) {
866 		case IB_QPS_INIT:
867 			if (iwqp->iwarp_state > IRDMA_QP_STATE_IDLE) {
868 				ret = -EINVAL;
869 				goto exit;
870 			}
871 
872 			if (iwqp->iwarp_state == IRDMA_QP_STATE_INVALID) {
873 				info.next_iwarp_state = IRDMA_QP_STATE_IDLE;
874 				issue_modify_qp = 1;
875 			}
876 			break;
877 		case IB_QPS_RTR:
878 			if (iwqp->iwarp_state > IRDMA_QP_STATE_IDLE) {
879 				ret = -EINVAL;
880 				goto exit;
881 			}
882 			info.arp_cache_idx_valid = true;
883 			info.cq_num_valid = true;
884 			info.next_iwarp_state = IRDMA_QP_STATE_RTR;
885 			issue_modify_qp = 1;
886 			break;
887 		case IB_QPS_RTS:
888 			if (iwqp->ibqp_state < IB_QPS_RTR ||
889 			    iwqp->ibqp_state == IB_QPS_ERR) {
890 				ret = -EINVAL;
891 				goto exit;
892 			}
893 
894 			info.arp_cache_idx_valid = true;
895 			info.cq_num_valid = true;
896 			info.ord_valid = true;
897 			info.next_iwarp_state = IRDMA_QP_STATE_RTS;
898 			issue_modify_qp = 1;
899 			if (dev->hw_attrs.uk_attrs.hw_rev == IRDMA_GEN_2)
900 				iwdev->rf->check_fc(&iwdev->vsi, &iwqp->sc_qp);
901 			udp_info->cwnd = iwdev->roce_cwnd;
902 			roce_info->ack_credits = iwdev->roce_ackcreds;
903 			if (iwdev->push_mode && udata &&
904 			    iwqp->sc_qp.push_idx == IRDMA_INVALID_PUSH_PAGE_INDEX &&
905 			    dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
906 				spin_unlock_irqrestore(&iwqp->lock, flags);
907 				irdma_alloc_push_page(iwqp);
908 				spin_lock_irqsave(&iwqp->lock, flags);
909 			}
910 			break;
911 		case IB_QPS_SQD:
912 			if (iwqp->iwarp_state == IRDMA_QP_STATE_SQD)
913 				goto exit;
914 
915 			if (iwqp->iwarp_state != IRDMA_QP_STATE_RTS) {
916 				ret = -EINVAL;
917 				goto exit;
918 			}
919 
920 			info.next_iwarp_state = IRDMA_QP_STATE_SQD;
921 			issue_modify_qp = 1;
922 			break;
923 		case IB_QPS_SQE:
924 		case IB_QPS_ERR:
925 		case IB_QPS_RESET:
926 			if (iwqp->iwarp_state == IRDMA_QP_STATE_RTS) {
927 				spin_unlock_irqrestore(&iwqp->lock, flags);
928 				info.next_iwarp_state = IRDMA_QP_STATE_SQD;
929 				irdma_hw_modify_qp(iwdev, iwqp, &info, true);
930 				spin_lock_irqsave(&iwqp->lock, flags);
931 			}
932 
933 			if (iwqp->iwarp_state == IRDMA_QP_STATE_ERROR) {
934 				spin_unlock_irqrestore(&iwqp->lock, flags);
935 				if (udata) {
936 					if (ib_copy_from_udata(&ureq, udata,
937 							       min(sizeof(ureq), udata->inlen)))
938 						return -EINVAL;
939 
940 					irdma_flush_wqes(iwqp,
941 							 (ureq.sq_flush ? IRDMA_FLUSH_SQ : 0) |
942 							 (ureq.rq_flush ? IRDMA_FLUSH_RQ : 0) |
943 							 IRDMA_REFLUSH);
944 				}
945 				return 0;
946 			}
947 
948 			info.next_iwarp_state = IRDMA_QP_STATE_ERROR;
949 			issue_modify_qp = 1;
950 			break;
951 		default:
952 			ret = -EINVAL;
953 			goto exit;
954 		}
955 
956 		iwqp->ibqp_state = attr->qp_state;
957 	}
958 
959 	ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
960 	ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
961 	irdma_sc_qp_setctx_roce(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
962 	spin_unlock_irqrestore(&iwqp->lock, flags);
963 
964 	if (attr_mask & IB_QP_STATE) {
965 		if (issue_modify_qp) {
966 			ctx_info->rem_endpoint_idx = udp_info->arp_idx;
967 			if (irdma_hw_modify_qp(iwdev, iwqp, &info, true))
968 				return -EINVAL;
969 			spin_lock_irqsave(&iwqp->lock, flags);
970 			if (iwqp->iwarp_state == info.curr_iwarp_state) {
971 				iwqp->iwarp_state = info.next_iwarp_state;
972 				iwqp->ibqp_state = attr->qp_state;
973 			}
974 			if (iwqp->ibqp_state > IB_QPS_RTS &&
975 			    !iwqp->flush_issued) {
976 				iwqp->flush_issued = 1;
977 				if (!iwqp->user_mode)
978 					queue_delayed_work(iwqp->iwdev->cleanup_wq,
979 							   &iwqp->dwork_flush,
980 							   msecs_to_jiffies(IRDMA_FLUSH_DELAY_MS));
981 				spin_unlock_irqrestore(&iwqp->lock, flags);
982 				irdma_flush_wqes(iwqp, IRDMA_FLUSH_SQ |
983 						 IRDMA_FLUSH_RQ |
984 						 IRDMA_FLUSH_WAIT);
985 			} else {
986 				spin_unlock_irqrestore(&iwqp->lock, flags);
987 			}
988 		} else {
989 			iwqp->ibqp_state = attr->qp_state;
990 		}
991 		if (udata && dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
992 			struct irdma_ucontext *ucontext;
993 
994 			ucontext = rdma_udata_to_drv_context(udata,
995 							     struct irdma_ucontext, ibucontext);
996 			if (iwqp->sc_qp.push_idx != IRDMA_INVALID_PUSH_PAGE_INDEX &&
997 			    !iwqp->push_wqe_mmap_entry &&
998 			    !irdma_setup_push_mmap_entries(ucontext, iwqp,
999 							   &uresp.push_wqe_mmap_key, &uresp.push_db_mmap_key)) {
1000 				uresp.push_valid = 1;
1001 				uresp.push_offset = iwqp->sc_qp.push_offset;
1002 			}
1003 			uresp.rd_fence_rate = iwdev->rd_fence_rate;
1004 			ret = ib_copy_to_udata(udata, &uresp, min(sizeof(uresp),
1005 								  udata->outlen));
1006 			if (ret) {
1007 				irdma_remove_push_mmap_entries(iwqp);
1008 				irdma_debug(iwdev_to_idev(iwdev),
1009 					    IRDMA_DEBUG_VERBS,
1010 					    "copy_to_udata failed\n");
1011 				return ret;
1012 			}
1013 		}
1014 	}
1015 
1016 	return 0;
1017 exit:
1018 	spin_unlock_irqrestore(&iwqp->lock, flags);
1019 
1020 	return ret;
1021 }
1022 
1023 /**
1024  * irdma_modify_qp - modify qp request
1025  * @ibqp: qp's pointer for modify
1026  * @attr: access attributes
1027  * @attr_mask: state mask
1028  * @udata: user data
1029  */
1030 int
1031 irdma_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask,
1032 		struct ib_udata *udata)
1033 {
1034 	struct irdma_qp *iwqp = to_iwqp(ibqp);
1035 	struct irdma_device *iwdev = iwqp->iwdev;
1036 	struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
1037 	struct irdma_qp_host_ctx_info *ctx_info;
1038 	struct irdma_tcp_offload_info *tcp_info;
1039 	struct irdma_iwarp_offload_info *offload_info;
1040 	struct irdma_modify_qp_info info = {0};
1041 	struct irdma_modify_qp_resp uresp = {};
1042 	struct irdma_modify_qp_req ureq = {};
1043 	u8 issue_modify_qp = 0;
1044 	u8 dont_wait = 0;
1045 	int err;
1046 	unsigned long flags;
1047 
1048 	if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
1049 		return -EOPNOTSUPP;
1050 
1051 	ctx_info = &iwqp->ctx_info;
1052 	offload_info = &iwqp->iwarp_info;
1053 	tcp_info = &iwqp->tcp_info;
1054 	wait_event(iwqp->mod_qp_waitq, !atomic_read(&iwqp->hw_mod_qp_pend));
1055 	irdma_debug(dev, IRDMA_DEBUG_VERBS,
1056 		    "caller: %pS qp_id=%d to_ibqpstate=%d ibqpstate=%d irdma_qpstate=%d last_aeq=%d hw_tcp_state=%d hw_iwarp_state=%d attr_mask=0x%x\n",
1057 		    __builtin_return_address(0), ibqp->qp_num, attr->qp_state, iwqp->ibqp_state, iwqp->iwarp_state,
1058 		    iwqp->last_aeq, iwqp->hw_tcp_state, iwqp->hw_iwarp_state, attr_mask);
1059 
1060 	spin_lock_irqsave(&iwqp->lock, flags);
1061 	if (attr_mask & IB_QP_STATE) {
1062 		info.curr_iwarp_state = iwqp->iwarp_state;
1063 		switch (attr->qp_state) {
1064 		case IB_QPS_INIT:
1065 		case IB_QPS_RTR:
1066 			if (iwqp->iwarp_state > IRDMA_QP_STATE_IDLE) {
1067 				err = -EINVAL;
1068 				goto exit;
1069 			}
1070 
1071 			if (iwqp->iwarp_state == IRDMA_QP_STATE_INVALID) {
1072 				info.next_iwarp_state = IRDMA_QP_STATE_IDLE;
1073 				issue_modify_qp = 1;
1074 			}
1075 			if (iwdev->push_mode && udata &&
1076 			    iwqp->sc_qp.push_idx == IRDMA_INVALID_PUSH_PAGE_INDEX &&
1077 			    dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1078 				spin_unlock_irqrestore(&iwqp->lock, flags);
1079 				irdma_alloc_push_page(iwqp);
1080 				spin_lock_irqsave(&iwqp->lock, flags);
1081 			}
1082 			break;
1083 		case IB_QPS_RTS:
1084 			if (iwqp->iwarp_state > IRDMA_QP_STATE_RTS ||
1085 			    !iwqp->cm_id) {
1086 				err = -EINVAL;
1087 				goto exit;
1088 			}
1089 
1090 			issue_modify_qp = 1;
1091 			iwqp->hw_tcp_state = IRDMA_TCP_STATE_ESTABLISHED;
1092 			iwqp->hte_added = 1;
1093 			info.next_iwarp_state = IRDMA_QP_STATE_RTS;
1094 			info.tcp_ctx_valid = true;
1095 			info.ord_valid = true;
1096 			info.arp_cache_idx_valid = true;
1097 			info.cq_num_valid = true;
1098 			break;
1099 		case IB_QPS_SQD:
1100 			if (iwqp->hw_iwarp_state > IRDMA_QP_STATE_RTS) {
1101 				err = 0;
1102 				goto exit;
1103 			}
1104 
1105 			if (iwqp->iwarp_state == IRDMA_QP_STATE_CLOSING ||
1106 			    iwqp->iwarp_state < IRDMA_QP_STATE_RTS) {
1107 				err = 0;
1108 				goto exit;
1109 			}
1110 
1111 			if (iwqp->iwarp_state > IRDMA_QP_STATE_CLOSING) {
1112 				err = -EINVAL;
1113 				goto exit;
1114 			}
1115 
1116 			info.next_iwarp_state = IRDMA_QP_STATE_CLOSING;
1117 			issue_modify_qp = 1;
1118 			break;
1119 		case IB_QPS_SQE:
1120 			if (iwqp->iwarp_state >= IRDMA_QP_STATE_TERMINATE) {
1121 				err = -EINVAL;
1122 				goto exit;
1123 			}
1124 
1125 			/* fallthrough */
1126 		case IB_QPS_ERR:
1127 		case IB_QPS_RESET:
1128 			if (iwqp->iwarp_state == IRDMA_QP_STATE_ERROR) {
1129 				spin_unlock_irqrestore(&iwqp->lock, flags);
1130 				if (udata) {
1131 					if (ib_copy_from_udata(&ureq, udata,
1132 							       min(sizeof(ureq), udata->inlen)))
1133 						return -EINVAL;
1134 
1135 					irdma_flush_wqes(iwqp,
1136 							 (ureq.sq_flush ? IRDMA_FLUSH_SQ : 0) |
1137 							 (ureq.rq_flush ? IRDMA_FLUSH_RQ : 0) |
1138 							 IRDMA_REFLUSH);
1139 				}
1140 				return 0;
1141 			}
1142 
1143 			if (iwqp->sc_qp.term_flags) {
1144 				spin_unlock_irqrestore(&iwqp->lock, flags);
1145 				irdma_terminate_del_timer(&iwqp->sc_qp);
1146 				spin_lock_irqsave(&iwqp->lock, flags);
1147 			}
1148 			info.next_iwarp_state = IRDMA_QP_STATE_ERROR;
1149 			if (iwqp->hw_tcp_state > IRDMA_TCP_STATE_CLOSED &&
1150 			    iwdev->iw_status &&
1151 			    iwqp->hw_tcp_state != IRDMA_TCP_STATE_TIME_WAIT)
1152 				info.reset_tcp_conn = true;
1153 			else
1154 				dont_wait = 1;
1155 
1156 			issue_modify_qp = 1;
1157 			info.next_iwarp_state = IRDMA_QP_STATE_ERROR;
1158 			break;
1159 		default:
1160 			err = -EINVAL;
1161 			goto exit;
1162 		}
1163 
1164 		iwqp->ibqp_state = attr->qp_state;
1165 	}
1166 	if (attr_mask & IB_QP_ACCESS_FLAGS) {
1167 		ctx_info->iwarp_info_valid = true;
1168 		if (attr->qp_access_flags & IB_ACCESS_LOCAL_WRITE)
1169 			offload_info->wr_rdresp_en = true;
1170 		if (attr->qp_access_flags & IB_ACCESS_REMOTE_WRITE)
1171 			offload_info->wr_rdresp_en = true;
1172 		if (attr->qp_access_flags & IB_ACCESS_REMOTE_READ)
1173 			offload_info->rd_en = true;
1174 	}
1175 
1176 	if (ctx_info->iwarp_info_valid) {
1177 		ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
1178 		ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
1179 		irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
1180 	}
1181 	spin_unlock_irqrestore(&iwqp->lock, flags);
1182 
1183 	if (attr_mask & IB_QP_STATE) {
1184 		if (issue_modify_qp) {
1185 			ctx_info->rem_endpoint_idx = tcp_info->arp_idx;
1186 			if (irdma_hw_modify_qp(iwdev, iwqp, &info, true))
1187 				return -EINVAL;
1188 		}
1189 
1190 		spin_lock_irqsave(&iwqp->lock, flags);
1191 		if (iwqp->iwarp_state == info.curr_iwarp_state) {
1192 			iwqp->iwarp_state = info.next_iwarp_state;
1193 			iwqp->ibqp_state = attr->qp_state;
1194 		}
1195 		spin_unlock_irqrestore(&iwqp->lock, flags);
1196 	}
1197 
1198 	if (issue_modify_qp && iwqp->ibqp_state > IB_QPS_RTS) {
1199 		if (dont_wait) {
1200 			if (iwqp->cm_id && iwqp->hw_tcp_state) {
1201 				spin_lock_irqsave(&iwqp->lock, flags);
1202 				iwqp->hw_tcp_state = IRDMA_TCP_STATE_CLOSED;
1203 				iwqp->last_aeq = IRDMA_AE_RESET_SENT;
1204 				spin_unlock_irqrestore(&iwqp->lock, flags);
1205 				irdma_cm_disconn(iwqp);
1206 			}
1207 		} else {
1208 			int close_timer_started;
1209 
1210 			spin_lock_irqsave(&iwdev->cm_core.ht_lock, flags);
1211 
1212 			if (iwqp->cm_node) {
1213 				atomic_inc(&iwqp->cm_node->refcnt);
1214 				spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
1215 				close_timer_started = atomic_inc_return(&iwqp->close_timer_started);
1216 				if (iwqp->cm_id && close_timer_started == 1)
1217 					irdma_schedule_cm_timer(iwqp->cm_node,
1218 								(struct irdma_puda_buf *)iwqp,
1219 								IRDMA_TIMER_TYPE_CLOSE, 1, 0);
1220 
1221 				irdma_rem_ref_cm_node(iwqp->cm_node);
1222 			} else {
1223 				spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
1224 			}
1225 		}
1226 	}
1227 	if (attr_mask & IB_QP_STATE && udata &&
1228 	    dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1229 		struct irdma_ucontext *ucontext;
1230 
1231 		ucontext = rdma_udata_to_drv_context(udata,
1232 						     struct irdma_ucontext, ibucontext);
1233 		if (iwqp->sc_qp.push_idx != IRDMA_INVALID_PUSH_PAGE_INDEX &&
1234 		    !iwqp->push_wqe_mmap_entry &&
1235 		    !irdma_setup_push_mmap_entries(ucontext, iwqp,
1236 						   &uresp.push_wqe_mmap_key, &uresp.push_db_mmap_key)) {
1237 			uresp.push_valid = 1;
1238 			uresp.push_offset = iwqp->sc_qp.push_offset;
1239 		}
1240 		uresp.rd_fence_rate = iwdev->rd_fence_rate;
1241 
1242 		err = ib_copy_to_udata(udata, &uresp, min(sizeof(uresp),
1243 							  udata->outlen));
1244 		if (err) {
1245 			irdma_remove_push_mmap_entries(iwqp);
1246 			irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_VERBS,
1247 				    "copy_to_udata failed\n");
1248 			return err;
1249 		}
1250 	}
1251 
1252 	return 0;
1253 exit:
1254 	spin_unlock_irqrestore(&iwqp->lock, flags);
1255 
1256 	return err;
1257 }
1258 
1259 /**
1260  * irdma_cq_free_rsrc - free up resources for cq
1261  * @rf: RDMA PCI function
1262  * @iwcq: cq ptr
1263  */
1264 void
1265 irdma_cq_free_rsrc(struct irdma_pci_f *rf, struct irdma_cq *iwcq)
1266 {
1267 	struct irdma_sc_cq *cq = &iwcq->sc_cq;
1268 
1269 	if (!iwcq->user_mode) {
1270 		irdma_free_dma_mem(rf->sc_dev.hw, &iwcq->kmem);
1271 		irdma_free_dma_mem(rf->sc_dev.hw, &iwcq->kmem_shadow);
1272 	}
1273 
1274 	irdma_free_rsrc(rf, rf->allocated_cqs, cq->cq_uk.cq_id);
1275 }
1276 
1277 /**
1278  * irdma_free_cqbuf - worker to free a cq buffer
1279  * @work: provides access to the cq buffer to free
1280  */
1281 static void
1282 irdma_free_cqbuf(struct work_struct *work)
1283 {
1284 	struct irdma_cq_buf *cq_buf = container_of(work, struct irdma_cq_buf, work);
1285 
1286 	irdma_free_dma_mem(cq_buf->hw, &cq_buf->kmem_buf);
1287 	kfree(cq_buf);
1288 }
1289 
1290 /**
1291  * irdma_process_resize_list - remove resized cq buffers from the resize_list
1292  * @iwcq: cq which owns the resize_list
1293  * @iwdev: irdma device
1294  * @lcqe_buf: the buffer where the last cqe is received
1295  */
1296 int
1297 irdma_process_resize_list(struct irdma_cq *iwcq,
1298 			  struct irdma_device *iwdev,
1299 			  struct irdma_cq_buf *lcqe_buf)
1300 {
1301 	struct list_head *tmp_node, *list_node;
1302 	struct irdma_cq_buf *cq_buf;
1303 	int cnt = 0;
1304 
1305 	list_for_each_safe(list_node, tmp_node, &iwcq->resize_list) {
1306 		cq_buf = list_entry(list_node, struct irdma_cq_buf, list);
1307 		if (cq_buf == lcqe_buf)
1308 			return cnt;
1309 
1310 		list_del(&cq_buf->list);
1311 		queue_work(iwdev->cleanup_wq, &cq_buf->work);
1312 		cnt++;
1313 	}
1314 
1315 	return cnt;
1316 }
1317 
1318 /**
1319  * irdma_resize_cq - resize cq
1320  * @ibcq: cq to be resized
1321  * @entries: desired cq size
1322  * @udata: user data
1323  */
1324 static int
1325 irdma_resize_cq(struct ib_cq *ibcq, int entries,
1326 		struct ib_udata *udata)
1327 {
1328 	struct irdma_cq *iwcq = to_iwcq(ibcq);
1329 	struct irdma_sc_dev *dev = iwcq->sc_cq.dev;
1330 	struct irdma_cqp_request *cqp_request;
1331 	struct cqp_cmds_info *cqp_info;
1332 	struct irdma_modify_cq_info *m_info;
1333 	struct irdma_modify_cq_info info = {0};
1334 	struct irdma_dma_mem kmem_buf;
1335 	struct irdma_cq_mr *cqmr_buf;
1336 	struct irdma_pbl *iwpbl_buf;
1337 	struct irdma_device *iwdev;
1338 	struct irdma_pci_f *rf;
1339 	struct irdma_cq_buf *cq_buf = NULL;
1340 	unsigned long flags;
1341 	int ret;
1342 
1343 	iwdev = to_iwdev(ibcq->device);
1344 	rf = iwdev->rf;
1345 
1346 	if (!(rf->sc_dev.hw_attrs.uk_attrs.feature_flags &
1347 	      IRDMA_FEATURE_CQ_RESIZE))
1348 		return -EOPNOTSUPP;
1349 
1350 	if (entries > rf->max_cqe)
1351 		return -EINVAL;
1352 
1353 	if (!iwcq->user_mode) {
1354 		entries++;
1355 		if (rf->sc_dev.hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
1356 			entries *= 2;
1357 	}
1358 
1359 	info.cq_size = max(entries, 4);
1360 
1361 	if (info.cq_size == iwcq->sc_cq.cq_uk.cq_size - 1)
1362 		return 0;
1363 
1364 	if (udata) {
1365 		struct irdma_resize_cq_req req = {0};
1366 		struct irdma_ucontext *ucontext =
1367 		rdma_udata_to_drv_context(udata, struct irdma_ucontext,
1368 					  ibucontext);
1369 
1370 		/* CQ resize not supported with legacy GEN_1 libi40iw */
1371 		if (ucontext->legacy_mode)
1372 			return -EOPNOTSUPP;
1373 
1374 		if (ib_copy_from_udata(&req, udata,
1375 				       min(sizeof(req), udata->inlen)))
1376 			return -EINVAL;
1377 
1378 		spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
1379 		iwpbl_buf = irdma_get_pbl((unsigned long)req.user_cq_buffer,
1380 					  &ucontext->cq_reg_mem_list);
1381 		spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
1382 
1383 		if (!iwpbl_buf)
1384 			return -ENOMEM;
1385 
1386 		cqmr_buf = &iwpbl_buf->cq_mr;
1387 		if (iwpbl_buf->pbl_allocated) {
1388 			info.virtual_map = true;
1389 			info.pbl_chunk_size = 1;
1390 			info.first_pm_pbl_idx = cqmr_buf->cq_pbl.idx;
1391 		} else {
1392 			info.cq_pa = cqmr_buf->cq_pbl.addr;
1393 		}
1394 	} else {
1395 		/* Kmode CQ resize */
1396 		int rsize;
1397 
1398 		rsize = info.cq_size * sizeof(struct irdma_cqe);
1399 		kmem_buf.size = round_up(rsize, 256);
1400 		kmem_buf.va = irdma_allocate_dma_mem(dev->hw, &kmem_buf,
1401 						     kmem_buf.size, 256);
1402 		if (!kmem_buf.va)
1403 			return -ENOMEM;
1404 
1405 		info.cq_base = kmem_buf.va;
1406 		info.cq_pa = kmem_buf.pa;
1407 		cq_buf = kzalloc(sizeof(*cq_buf), GFP_KERNEL);
1408 		if (!cq_buf) {
1409 			ret = -ENOMEM;
1410 			goto error;
1411 		}
1412 	}
1413 
1414 	cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
1415 	if (!cqp_request) {
1416 		ret = -ENOMEM;
1417 		goto error;
1418 	}
1419 
1420 	info.shadow_read_threshold = iwcq->sc_cq.shadow_read_threshold;
1421 	info.cq_resize = true;
1422 
1423 	cqp_info = &cqp_request->info;
1424 	m_info = &cqp_info->in.u.cq_modify.info;
1425 	memcpy(m_info, &info, sizeof(*m_info));
1426 
1427 	cqp_info->cqp_cmd = IRDMA_OP_CQ_MODIFY;
1428 	cqp_info->in.u.cq_modify.cq = &iwcq->sc_cq;
1429 	cqp_info->in.u.cq_modify.scratch = (uintptr_t)cqp_request;
1430 	cqp_info->post_sq = 1;
1431 	ret = irdma_handle_cqp_op(rf, cqp_request);
1432 	irdma_put_cqp_request(&rf->cqp, cqp_request);
1433 	if (ret)
1434 		goto error;
1435 
1436 	spin_lock_irqsave(&iwcq->lock, flags);
1437 	if (cq_buf) {
1438 		cq_buf->kmem_buf = iwcq->kmem;
1439 		cq_buf->hw = dev->hw;
1440 		memcpy(&cq_buf->cq_uk, &iwcq->sc_cq.cq_uk, sizeof(cq_buf->cq_uk));
1441 		INIT_WORK(&cq_buf->work, irdma_free_cqbuf);
1442 		list_add_tail(&cq_buf->list, &iwcq->resize_list);
1443 		iwcq->kmem = kmem_buf;
1444 	}
1445 
1446 	irdma_sc_cq_resize(&iwcq->sc_cq, &info);
1447 	ibcq->cqe = info.cq_size - 1;
1448 	spin_unlock_irqrestore(&iwcq->lock, flags);
1449 
1450 	return 0;
1451 error:
1452 	if (!udata)
1453 		irdma_free_dma_mem(dev->hw, &kmem_buf);
1454 	kfree(cq_buf);
1455 
1456 	return ret;
1457 }
1458 
1459 /**
1460  * irdma_get_mr_access - get hw MR access permissions from IB access flags
1461  * @access: IB access flags
1462  */
1463 static inline u16 irdma_get_mr_access(int access){
1464 	u16 hw_access = 0;
1465 
1466 	hw_access |= (access & IB_ACCESS_LOCAL_WRITE) ?
1467 	    IRDMA_ACCESS_FLAGS_LOCALWRITE : 0;
1468 	hw_access |= (access & IB_ACCESS_REMOTE_WRITE) ?
1469 	    IRDMA_ACCESS_FLAGS_REMOTEWRITE : 0;
1470 	hw_access |= (access & IB_ACCESS_REMOTE_READ) ?
1471 	    IRDMA_ACCESS_FLAGS_REMOTEREAD : 0;
1472 	hw_access |= (access & IB_ACCESS_MW_BIND) ?
1473 	    IRDMA_ACCESS_FLAGS_BIND_WINDOW : 0;
1474 	hw_access |= (access & IB_ZERO_BASED) ?
1475 	    IRDMA_ACCESS_FLAGS_ZERO_BASED : 0;
1476 	hw_access |= IRDMA_ACCESS_FLAGS_LOCALREAD;
1477 
1478 	return hw_access;
1479 }
1480 
1481 /**
1482  * irdma_free_stag - free stag resource
1483  * @iwdev: irdma device
1484  * @stag: stag to free
1485  */
1486 void
1487 irdma_free_stag(struct irdma_device *iwdev, u32 stag)
1488 {
1489 	u32 stag_idx;
1490 
1491 	stag_idx = (stag & iwdev->rf->mr_stagmask) >> IRDMA_CQPSQ_STAG_IDX_S;
1492 	irdma_free_rsrc(iwdev->rf, iwdev->rf->allocated_mrs, stag_idx);
1493 }
1494 
1495 /**
1496  * irdma_create_stag - create random stag
1497  * @iwdev: irdma device
1498  */
1499 u32
1500 irdma_create_stag(struct irdma_device *iwdev)
1501 {
1502 	u32 stag = 0;
1503 	u32 stag_index = 0;
1504 	u32 next_stag_index;
1505 	u32 driver_key;
1506 	u32 random;
1507 	u8 consumer_key;
1508 	int ret;
1509 
1510 	get_random_bytes(&random, sizeof(random));
1511 	consumer_key = (u8)random;
1512 
1513 	driver_key = random & ~iwdev->rf->mr_stagmask;
1514 	next_stag_index = (random & iwdev->rf->mr_stagmask) >> 8;
1515 	next_stag_index %= iwdev->rf->max_mr;
1516 
1517 	ret = irdma_alloc_rsrc(iwdev->rf, iwdev->rf->allocated_mrs,
1518 			       iwdev->rf->max_mr, &stag_index,
1519 			       &next_stag_index);
1520 	if (ret)
1521 		return stag;
1522 	stag = stag_index << IRDMA_CQPSQ_STAG_IDX_S;
1523 	stag |= driver_key;
1524 	stag += (u32)consumer_key;
1525 
1526 	return stag;
1527 }
1528 
1529 /**
1530  * irdma_check_mem_contiguous - check if pbls stored in arr are contiguous
1531  * @arr: lvl1 pbl array
1532  * @npages: page count
1533  * @pg_size: page size
1534  *
1535  */
1536 static bool
1537 irdma_check_mem_contiguous(u64 *arr, u32 npages, u32 pg_size)
1538 {
1539 	u32 pg_idx;
1540 
1541 	for (pg_idx = 0; pg_idx < npages; pg_idx++) {
1542 		if ((*arr + (pg_size * pg_idx)) != arr[pg_idx])
1543 			return false;
1544 	}
1545 
1546 	return true;
1547 }
1548 
1549 /**
1550  * irdma_check_mr_contiguous - check if MR is physically contiguous
1551  * @palloc: pbl allocation struct
1552  * @pg_size: page size
1553  */
1554 static bool
1555 irdma_check_mr_contiguous(struct irdma_pble_alloc *palloc,
1556 			  u32 pg_size)
1557 {
1558 	struct irdma_pble_level2 *lvl2 = &palloc->level2;
1559 	struct irdma_pble_info *leaf = lvl2->leaf;
1560 	u64 *arr = NULL;
1561 	u64 *start_addr = NULL;
1562 	int i;
1563 	bool ret;
1564 
1565 	if (palloc->level == PBLE_LEVEL_1) {
1566 		arr = palloc->level1.addr;
1567 		ret = irdma_check_mem_contiguous(arr, palloc->total_cnt,
1568 						 pg_size);
1569 		return ret;
1570 	}
1571 
1572 	start_addr = leaf->addr;
1573 
1574 	for (i = 0; i < lvl2->leaf_cnt; i++, leaf++) {
1575 		arr = leaf->addr;
1576 		if ((*start_addr + (i * pg_size * PBLE_PER_PAGE)) != *arr)
1577 			return false;
1578 		ret = irdma_check_mem_contiguous(arr, leaf->cnt, pg_size);
1579 		if (!ret)
1580 			return false;
1581 	}
1582 
1583 	return true;
1584 }
1585 
1586 /**
1587  * irdma_setup_pbles - copy user pg address to pble's
1588  * @rf: RDMA PCI function
1589  * @iwmr: mr pointer for this memory registration
1590  * @use_pbles: flag if to use pble's
1591  */
1592 static int
1593 irdma_setup_pbles(struct irdma_pci_f *rf, struct irdma_mr *iwmr,
1594 		  bool use_pbles)
1595 {
1596 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
1597 	struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
1598 	struct irdma_pble_info *pinfo;
1599 	u64 *pbl;
1600 	int status;
1601 	enum irdma_pble_level level = PBLE_LEVEL_1;
1602 
1603 	if (use_pbles) {
1604 		status = irdma_get_pble(rf->pble_rsrc, palloc, iwmr->page_cnt,
1605 					false);
1606 		if (status)
1607 			return status;
1608 
1609 		iwpbl->pbl_allocated = true;
1610 		level = palloc->level;
1611 		pinfo = (level == PBLE_LEVEL_1) ? &palloc->level1 :
1612 		    palloc->level2.leaf;
1613 		pbl = pinfo->addr;
1614 	} else {
1615 		pbl = iwmr->pgaddrmem;
1616 	}
1617 
1618 	irdma_copy_user_pgaddrs(iwmr, pbl, level);
1619 
1620 	if (use_pbles)
1621 		iwmr->pgaddrmem[0] = *pbl;
1622 
1623 	return 0;
1624 }
1625 
1626 /**
1627  * irdma_handle_q_mem - handle memory for qp and cq
1628  * @iwdev: irdma device
1629  * @req: information for q memory management
1630  * @iwpbl: pble struct
1631  * @use_pbles: flag to use pble
1632  */
1633 static int
1634 irdma_handle_q_mem(struct irdma_device *iwdev,
1635 		   struct irdma_mem_reg_req *req,
1636 		   struct irdma_pbl *iwpbl, bool use_pbles)
1637 {
1638 	struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
1639 	struct irdma_mr *iwmr = iwpbl->iwmr;
1640 	struct irdma_qp_mr *qpmr = &iwpbl->qp_mr;
1641 	struct irdma_cq_mr *cqmr = &iwpbl->cq_mr;
1642 	struct irdma_hmc_pble *hmc_p;
1643 	u64 *arr = iwmr->pgaddrmem;
1644 	u32 pg_size, total;
1645 	int err = 0;
1646 	bool ret = true;
1647 
1648 	pg_size = iwmr->page_size;
1649 	err = irdma_setup_pbles(iwdev->rf, iwmr, use_pbles);
1650 	if (err)
1651 		return err;
1652 
1653 	if (use_pbles && palloc->level != PBLE_LEVEL_1) {
1654 		irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
1655 		iwpbl->pbl_allocated = false;
1656 		return -ENOMEM;
1657 	}
1658 
1659 	if (use_pbles)
1660 		arr = palloc->level1.addr;
1661 
1662 	switch (iwmr->type) {
1663 	case IRDMA_MEMREG_TYPE_QP:
1664 		total = req->sq_pages + req->rq_pages;
1665 		hmc_p = &qpmr->sq_pbl;
1666 		qpmr->shadow = (dma_addr_t) arr[total];
1667 		if (use_pbles) {
1668 			ret = irdma_check_mem_contiguous(arr, req->sq_pages,
1669 							 pg_size);
1670 			if (ret)
1671 				ret = irdma_check_mem_contiguous(&arr[req->sq_pages],
1672 								 req->rq_pages,
1673 								 pg_size);
1674 		}
1675 
1676 		if (!ret) {
1677 			hmc_p->idx = palloc->level1.idx;
1678 			hmc_p = &qpmr->rq_pbl;
1679 			hmc_p->idx = palloc->level1.idx + req->sq_pages;
1680 		} else {
1681 			hmc_p->addr = arr[0];
1682 			hmc_p = &qpmr->rq_pbl;
1683 			hmc_p->addr = arr[req->sq_pages];
1684 		}
1685 		break;
1686 	case IRDMA_MEMREG_TYPE_CQ:
1687 		hmc_p = &cqmr->cq_pbl;
1688 
1689 		if (!cqmr->split)
1690 			cqmr->shadow = (dma_addr_t) arr[req->cq_pages];
1691 
1692 		if (use_pbles)
1693 			ret = irdma_check_mem_contiguous(arr, req->cq_pages,
1694 							 pg_size);
1695 
1696 		if (!ret)
1697 			hmc_p->idx = palloc->level1.idx;
1698 		else
1699 			hmc_p->addr = arr[0];
1700 		break;
1701 	default:
1702 		irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_VERBS,
1703 			    "MR type error\n");
1704 		err = -EINVAL;
1705 	}
1706 
1707 	if (use_pbles && ret) {
1708 		irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
1709 		iwpbl->pbl_allocated = false;
1710 	}
1711 
1712 	return err;
1713 }
1714 
1715 /**
1716  * irdma_hw_alloc_mw - create the hw memory window
1717  * @iwdev: irdma device
1718  * @iwmr: pointer to memory window info
1719  */
1720 int
1721 irdma_hw_alloc_mw(struct irdma_device *iwdev, struct irdma_mr *iwmr)
1722 {
1723 	struct irdma_mw_alloc_info *info;
1724 	struct irdma_pd *iwpd = to_iwpd(iwmr->ibmr.pd);
1725 	struct irdma_cqp_request *cqp_request;
1726 	struct cqp_cmds_info *cqp_info;
1727 	int status;
1728 
1729 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
1730 	if (!cqp_request)
1731 		return -ENOMEM;
1732 
1733 	cqp_info = &cqp_request->info;
1734 	info = &cqp_info->in.u.mw_alloc.info;
1735 	memset(info, 0, sizeof(*info));
1736 	if (iwmr->ibmw.type == IB_MW_TYPE_1)
1737 		info->mw_wide = true;
1738 
1739 	info->page_size = PAGE_SIZE;
1740 	info->mw_stag_index = iwmr->stag >> IRDMA_CQPSQ_STAG_IDX_S;
1741 	info->pd_id = iwpd->sc_pd.pd_id;
1742 	info->remote_access = true;
1743 	cqp_info->cqp_cmd = IRDMA_OP_MW_ALLOC;
1744 	cqp_info->post_sq = 1;
1745 	cqp_info->in.u.mw_alloc.dev = &iwdev->rf->sc_dev;
1746 	cqp_info->in.u.mw_alloc.scratch = (uintptr_t)cqp_request;
1747 	status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
1748 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
1749 
1750 	return status;
1751 }
1752 
1753 /**
1754  * irdma_dealloc_mw - Dealloc memory window
1755  * @ibmw: memory window structure.
1756  */
1757 static int
1758 irdma_dealloc_mw(struct ib_mw *ibmw)
1759 {
1760 	struct ib_pd *ibpd = ibmw->pd;
1761 	struct irdma_pd *iwpd = to_iwpd(ibpd);
1762 	struct irdma_mr *iwmr = to_iwmr((struct ib_mr *)ibmw);
1763 	struct irdma_device *iwdev = to_iwdev(ibmw->device);
1764 	struct irdma_cqp_request *cqp_request;
1765 	struct cqp_cmds_info *cqp_info;
1766 	struct irdma_dealloc_stag_info *info;
1767 
1768 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
1769 	if (!cqp_request)
1770 		return -ENOMEM;
1771 
1772 	cqp_info = &cqp_request->info;
1773 	info = &cqp_info->in.u.dealloc_stag.info;
1774 	memset(info, 0, sizeof(*info));
1775 	info->pd_id = iwpd->sc_pd.pd_id & 0x00007fff;
1776 	info->stag_idx = RS_64_1(ibmw->rkey, IRDMA_CQPSQ_STAG_IDX_S);
1777 	info->mr = false;
1778 	cqp_info->cqp_cmd = IRDMA_OP_DEALLOC_STAG;
1779 	cqp_info->post_sq = 1;
1780 	cqp_info->in.u.dealloc_stag.dev = &iwdev->rf->sc_dev;
1781 	cqp_info->in.u.dealloc_stag.scratch = (uintptr_t)cqp_request;
1782 	irdma_handle_cqp_op(iwdev->rf, cqp_request);
1783 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
1784 	irdma_free_stag(iwdev, iwmr->stag);
1785 	kfree(iwmr);
1786 
1787 	return 0;
1788 }
1789 
1790 /**
1791  * irdma_hw_alloc_stag - cqp command to allocate stag
1792  * @iwdev: irdma device
1793  * @iwmr: irdma mr pointer
1794  */
1795 int
1796 irdma_hw_alloc_stag(struct irdma_device *iwdev,
1797 		    struct irdma_mr *iwmr)
1798 {
1799 	struct irdma_allocate_stag_info *info;
1800 	struct irdma_pd *iwpd = to_iwpd(iwmr->ibmr.pd);
1801 	int status;
1802 	struct irdma_cqp_request *cqp_request;
1803 	struct cqp_cmds_info *cqp_info;
1804 
1805 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
1806 	if (!cqp_request)
1807 		return -ENOMEM;
1808 
1809 	cqp_info = &cqp_request->info;
1810 	info = &cqp_info->in.u.alloc_stag.info;
1811 	memset(info, 0, sizeof(*info));
1812 	info->page_size = PAGE_SIZE;
1813 	info->stag_idx = iwmr->stag >> IRDMA_CQPSQ_STAG_IDX_S;
1814 	info->pd_id = iwpd->sc_pd.pd_id;
1815 	info->total_len = iwmr->len;
1816 	info->remote_access = true;
1817 	cqp_info->cqp_cmd = IRDMA_OP_ALLOC_STAG;
1818 	cqp_info->post_sq = 1;
1819 	cqp_info->in.u.alloc_stag.dev = &iwdev->rf->sc_dev;
1820 	cqp_info->in.u.alloc_stag.scratch = (uintptr_t)cqp_request;
1821 	status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
1822 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
1823 
1824 	return status;
1825 }
1826 
1827 /**
1828  * irdma_set_page - populate pbl list for fmr
1829  * @ibmr: ib mem to access iwarp mr pointer
1830  * @addr: page dma address fro pbl list
1831  */
1832 static int
1833 irdma_set_page(struct ib_mr *ibmr, u64 addr)
1834 {
1835 	struct irdma_mr *iwmr = to_iwmr(ibmr);
1836 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
1837 	struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
1838 	u64 *pbl;
1839 
1840 	if (unlikely(iwmr->npages == iwmr->page_cnt))
1841 		return -ENOMEM;
1842 
1843 	pbl = palloc->level1.addr;
1844 	pbl[iwmr->npages++] = addr;
1845 
1846 	return 0;
1847 }
1848 
1849 /**
1850  * irdma_map_mr_sg - map of sg list for fmr
1851  * @ibmr: ib mem to access iwarp mr pointer
1852  * @sg: scatter gather list
1853  * @sg_nents: number of sg pages
1854  * @sg_offset: scatter gather list for fmr
1855  */
1856 static int
1857 irdma_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
1858 		int sg_nents, unsigned int *sg_offset)
1859 {
1860 	struct irdma_mr *iwmr = to_iwmr(ibmr);
1861 
1862 	iwmr->npages = 0;
1863 
1864 	return ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, irdma_set_page);
1865 }
1866 
1867 /**
1868  * irdma_hwreg_mr - send cqp command for memory registration
1869  * @iwdev: irdma device
1870  * @iwmr: irdma mr pointer
1871  * @access: access for MR
1872  */
1873 static int
1874 irdma_hwreg_mr(struct irdma_device *iwdev, struct irdma_mr *iwmr,
1875 	       u16 access)
1876 {
1877 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
1878 	struct irdma_reg_ns_stag_info *stag_info;
1879 	struct irdma_pd *iwpd = to_iwpd(iwmr->ibmr.pd);
1880 	struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
1881 	struct irdma_cqp_request *cqp_request;
1882 	struct cqp_cmds_info *cqp_info;
1883 	int ret;
1884 
1885 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
1886 	if (!cqp_request)
1887 		return -ENOMEM;
1888 
1889 	cqp_info = &cqp_request->info;
1890 	stag_info = &cqp_info->in.u.mr_reg_non_shared.info;
1891 	memset(stag_info, 0, sizeof(*stag_info));
1892 	stag_info->va = iwpbl->user_base;
1893 	stag_info->stag_idx = iwmr->stag >> IRDMA_CQPSQ_STAG_IDX_S;
1894 	stag_info->stag_key = (u8)iwmr->stag;
1895 	stag_info->total_len = iwmr->len;
1896 	stag_info->access_rights = irdma_get_mr_access(access);
1897 	stag_info->pd_id = iwpd->sc_pd.pd_id;
1898 	if (stag_info->access_rights & IRDMA_ACCESS_FLAGS_ZERO_BASED)
1899 		stag_info->addr_type = IRDMA_ADDR_TYPE_ZERO_BASED;
1900 	else
1901 		stag_info->addr_type = IRDMA_ADDR_TYPE_VA_BASED;
1902 	stag_info->page_size = iwmr->page_size;
1903 
1904 	if (iwpbl->pbl_allocated) {
1905 		if (palloc->level == PBLE_LEVEL_1) {
1906 			stag_info->first_pm_pbl_index = palloc->level1.idx;
1907 			stag_info->chunk_size = 1;
1908 		} else {
1909 			stag_info->first_pm_pbl_index = palloc->level2.root.idx;
1910 			stag_info->chunk_size = 3;
1911 		}
1912 	} else {
1913 		stag_info->reg_addr_pa = iwmr->pgaddrmem[0];
1914 	}
1915 
1916 	cqp_info->cqp_cmd = IRDMA_OP_MR_REG_NON_SHARED;
1917 	cqp_info->post_sq = 1;
1918 	cqp_info->in.u.mr_reg_non_shared.dev = &iwdev->rf->sc_dev;
1919 	cqp_info->in.u.mr_reg_non_shared.scratch = (uintptr_t)cqp_request;
1920 	ret = irdma_handle_cqp_op(iwdev->rf, cqp_request);
1921 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
1922 
1923 	return ret;
1924 }
1925 
1926 /**
1927  * irdma_reg_user_mr - Register a user memory region
1928  * @pd: ptr of pd
1929  * @start: virtual start address
1930  * @len: length of mr
1931  * @virt: virtual address
1932  * @access: access of mr
1933  * @udata: user data
1934  */
1935 static struct ib_mr *
1936 irdma_reg_user_mr(struct ib_pd *pd, u64 start, u64 len,
1937 		  u64 virt, int access,
1938 		  struct ib_udata *udata)
1939 {
1940 	struct irdma_device *iwdev = to_iwdev(pd->device);
1941 	struct irdma_ucontext *ucontext;
1942 	struct irdma_pble_alloc *palloc;
1943 	struct irdma_pbl *iwpbl;
1944 	struct irdma_mr *iwmr;
1945 	struct ib_umem *region;
1946 	struct irdma_mem_reg_req req;
1947 	u32 total, stag = 0;
1948 	u8 shadow_pgcnt = 1;
1949 	bool use_pbles = false;
1950 	unsigned long flags;
1951 	int err = -EINVAL;
1952 	int ret;
1953 
1954 	if (!len || len > iwdev->rf->sc_dev.hw_attrs.max_mr_size)
1955 		return ERR_PTR(-EINVAL);
1956 
1957 	region = ib_umem_get(pd->uobject->context, start, len, access, 0);
1958 
1959 	if (IS_ERR(region)) {
1960 		irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_VERBS,
1961 			    "Failed to create ib_umem region\n");
1962 		return (struct ib_mr *)region;
1963 	}
1964 
1965 	if (ib_copy_from_udata(&req, udata, min(sizeof(req), udata->inlen))) {
1966 		ib_umem_release(region);
1967 		return ERR_PTR(-EFAULT);
1968 	}
1969 
1970 	iwmr = kzalloc(sizeof(*iwmr), GFP_KERNEL);
1971 	if (!iwmr) {
1972 		ib_umem_release(region);
1973 		return ERR_PTR(-ENOMEM);
1974 	}
1975 
1976 	iwpbl = &iwmr->iwpbl;
1977 	iwpbl->iwmr = iwmr;
1978 	iwmr->region = region;
1979 	iwmr->ibmr.pd = pd;
1980 	iwmr->ibmr.device = pd->device;
1981 	iwmr->ibmr.iova = virt;
1982 	iwmr->page_size = PAGE_SIZE;
1983 
1984 	iwmr->page_msk = PAGE_MASK;
1985 	iwmr->len = region->length;
1986 	iwpbl->user_base = virt;
1987 	palloc = &iwpbl->pble_alloc;
1988 	iwmr->type = req.reg_type;
1989 	iwmr->page_cnt = irdma_ib_umem_num_dma_blocks(region, iwmr->page_size, virt);
1990 
1991 	switch (req.reg_type) {
1992 	case IRDMA_MEMREG_TYPE_QP:
1993 		total = req.sq_pages + req.rq_pages + shadow_pgcnt;
1994 		if (total > iwmr->page_cnt) {
1995 			err = -EINVAL;
1996 			goto error;
1997 		}
1998 		total = req.sq_pages + req.rq_pages;
1999 		use_pbles = (total > 2);
2000 		err = irdma_handle_q_mem(iwdev, &req, iwpbl, use_pbles);
2001 		if (err)
2002 			goto error;
2003 
2004 		ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext,
2005 						     ibucontext);
2006 		spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags);
2007 		list_add_tail(&iwpbl->list, &ucontext->qp_reg_mem_list);
2008 		iwpbl->on_list = true;
2009 		spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags);
2010 		break;
2011 	case IRDMA_MEMREG_TYPE_CQ:
2012 		if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_CQ_RESIZE)
2013 			shadow_pgcnt = 0;
2014 		total = req.cq_pages + shadow_pgcnt;
2015 		if (total > iwmr->page_cnt) {
2016 			err = -EINVAL;
2017 			goto error;
2018 		}
2019 
2020 		use_pbles = (req.cq_pages > 1);
2021 		err = irdma_handle_q_mem(iwdev, &req, iwpbl, use_pbles);
2022 		if (err)
2023 			goto error;
2024 
2025 		ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext,
2026 						     ibucontext);
2027 		spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
2028 		list_add_tail(&iwpbl->list, &ucontext->cq_reg_mem_list);
2029 		iwpbl->on_list = true;
2030 		spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
2031 		break;
2032 	case IRDMA_MEMREG_TYPE_MEM:
2033 		use_pbles = (iwmr->page_cnt != 1);
2034 
2035 		err = irdma_setup_pbles(iwdev->rf, iwmr, use_pbles);
2036 		if (err)
2037 			goto error;
2038 
2039 		if (use_pbles) {
2040 			ret = irdma_check_mr_contiguous(palloc,
2041 							iwmr->page_size);
2042 			if (ret) {
2043 				irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
2044 				iwpbl->pbl_allocated = false;
2045 			}
2046 		}
2047 
2048 		stag = irdma_create_stag(iwdev);
2049 		if (!stag) {
2050 			err = -ENOMEM;
2051 			goto error;
2052 		}
2053 
2054 		iwmr->stag = stag;
2055 		iwmr->ibmr.rkey = stag;
2056 		iwmr->ibmr.lkey = stag;
2057 		err = irdma_hwreg_mr(iwdev, iwmr, access);
2058 		if (err) {
2059 			irdma_free_stag(iwdev, stag);
2060 			goto error;
2061 		}
2062 
2063 		break;
2064 	default:
2065 		goto error;
2066 	}
2067 
2068 	iwmr->type = req.reg_type;
2069 
2070 	return &iwmr->ibmr;
2071 
2072 error:
2073 	if (palloc->level != PBLE_LEVEL_0 && iwpbl->pbl_allocated)
2074 		irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
2075 	ib_umem_release(region);
2076 	kfree(iwmr);
2077 
2078 	return ERR_PTR(err);
2079 }
2080 
2081 /**
2082  * irdma_reg_phys_mr - register kernel physical memory
2083  * @pd: ibpd pointer
2084  * @addr: physical address of memory to register
2085  * @size: size of memory to register
2086  * @access: Access rights
2087  * @iova_start: start of virtual address for physical buffers
2088  */
2089 struct ib_mr *
2090 irdma_reg_phys_mr(struct ib_pd *pd, u64 addr, u64 size, int access,
2091 		  u64 *iova_start)
2092 {
2093 	struct irdma_device *iwdev = to_iwdev(pd->device);
2094 	struct irdma_pbl *iwpbl;
2095 	struct irdma_mr *iwmr;
2096 	u32 stag;
2097 	int ret;
2098 
2099 	iwmr = kzalloc(sizeof(*iwmr), GFP_KERNEL);
2100 	if (!iwmr)
2101 		return ERR_PTR(-ENOMEM);
2102 
2103 	iwmr->ibmr.pd = pd;
2104 	iwmr->ibmr.device = pd->device;
2105 	iwpbl = &iwmr->iwpbl;
2106 	iwpbl->iwmr = iwmr;
2107 	iwmr->type = IRDMA_MEMREG_TYPE_MEM;
2108 	iwpbl->user_base = *iova_start;
2109 	stag = irdma_create_stag(iwdev);
2110 	if (!stag) {
2111 		ret = -ENOMEM;
2112 		goto err;
2113 	}
2114 
2115 	iwmr->stag = stag;
2116 	iwmr->ibmr.iova = *iova_start;
2117 	iwmr->ibmr.rkey = stag;
2118 	iwmr->ibmr.lkey = stag;
2119 	iwmr->page_cnt = 1;
2120 	iwmr->pgaddrmem[0] = addr;
2121 	iwmr->len = size;
2122 	iwmr->page_size = SZ_4K;
2123 	ret = irdma_hwreg_mr(iwdev, iwmr, access);
2124 	if (ret) {
2125 		irdma_free_stag(iwdev, stag);
2126 		goto err;
2127 	}
2128 
2129 	return &iwmr->ibmr;
2130 
2131 err:
2132 	kfree(iwmr);
2133 
2134 	return ERR_PTR(ret);
2135 }
2136 
2137 /**
2138  * irdma_get_dma_mr - register physical mem
2139  * @pd: ptr of pd
2140  * @acc: access for memory
2141  */
2142 static struct ib_mr *
2143 irdma_get_dma_mr(struct ib_pd *pd, int acc)
2144 {
2145 	u64 kva = 0;
2146 
2147 	return irdma_reg_phys_mr(pd, 0, 0, acc, &kva);
2148 }
2149 
2150 /**
2151  * irdma_del_memlist - Deleting pbl list entries for CQ/QP
2152  * @iwmr: iwmr for IB's user page addresses
2153  * @ucontext: ptr to user context
2154  */
2155 void
2156 irdma_del_memlist(struct irdma_mr *iwmr,
2157 		  struct irdma_ucontext *ucontext)
2158 {
2159 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
2160 	unsigned long flags;
2161 
2162 	switch (iwmr->type) {
2163 	case IRDMA_MEMREG_TYPE_CQ:
2164 		spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
2165 		if (iwpbl->on_list) {
2166 			iwpbl->on_list = false;
2167 			list_del(&iwpbl->list);
2168 		}
2169 		spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
2170 		break;
2171 	case IRDMA_MEMREG_TYPE_QP:
2172 		spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags);
2173 		if (iwpbl->on_list) {
2174 			iwpbl->on_list = false;
2175 			list_del(&iwpbl->list);
2176 		}
2177 		spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags);
2178 		break;
2179 	default:
2180 		break;
2181 	}
2182 }
2183 
2184 /**
2185  * irdma_copy_sg_list - copy sg list for qp
2186  * @sg_list: copied into sg_list
2187  * @sgl: copy from sgl
2188  * @num_sges: count of sg entries
2189  */
2190 static void
2191 irdma_copy_sg_list(struct irdma_sge *sg_list, struct ib_sge *sgl,
2192 		   int num_sges)
2193 {
2194 	unsigned int i;
2195 
2196 	for (i = 0; i < num_sges; i++) {
2197 		sg_list[i].tag_off = sgl[i].addr;
2198 		sg_list[i].len = sgl[i].length;
2199 		sg_list[i].stag = sgl[i].lkey;
2200 	}
2201 }
2202 
2203 /**
2204  * irdma_post_send -  kernel application wr
2205  * @ibqp: qp ptr for wr
2206  * @ib_wr: work request ptr
2207  * @bad_wr: return of bad wr if err
2208  */
2209 static int
2210 irdma_post_send(struct ib_qp *ibqp,
2211 		const struct ib_send_wr *ib_wr,
2212 		const struct ib_send_wr **bad_wr)
2213 {
2214 	struct irdma_qp *iwqp;
2215 	struct irdma_qp_uk *ukqp;
2216 	struct irdma_sc_dev *dev;
2217 	struct irdma_post_sq_info info;
2218 	int err = 0;
2219 	unsigned long flags;
2220 	bool inv_stag;
2221 	struct irdma_ah *ah;
2222 
2223 	iwqp = to_iwqp(ibqp);
2224 	ukqp = &iwqp->sc_qp.qp_uk;
2225 	dev = &iwqp->iwdev->rf->sc_dev;
2226 
2227 	spin_lock_irqsave(&iwqp->lock, flags);
2228 	while (ib_wr) {
2229 		memset(&info, 0, sizeof(info));
2230 		inv_stag = false;
2231 		info.wr_id = (ib_wr->wr_id);
2232 		if ((ib_wr->send_flags & IB_SEND_SIGNALED) || iwqp->sig_all)
2233 			info.signaled = true;
2234 		if (ib_wr->send_flags & IB_SEND_FENCE)
2235 			info.read_fence = true;
2236 		switch (ib_wr->opcode) {
2237 		case IB_WR_SEND_WITH_IMM:
2238 			if (ukqp->qp_caps & IRDMA_SEND_WITH_IMM) {
2239 				info.imm_data_valid = true;
2240 				info.imm_data = ntohl(ib_wr->ex.imm_data);
2241 			} else {
2242 				err = -EINVAL;
2243 				break;
2244 			}
2245 			/* fallthrough */
2246 		case IB_WR_SEND:
2247 		case IB_WR_SEND_WITH_INV:
2248 			if (ib_wr->opcode == IB_WR_SEND ||
2249 			    ib_wr->opcode == IB_WR_SEND_WITH_IMM) {
2250 				if (ib_wr->send_flags & IB_SEND_SOLICITED)
2251 					info.op_type = IRDMA_OP_TYPE_SEND_SOL;
2252 				else
2253 					info.op_type = IRDMA_OP_TYPE_SEND;
2254 			} else {
2255 				if (ib_wr->send_flags & IB_SEND_SOLICITED)
2256 					info.op_type = IRDMA_OP_TYPE_SEND_SOL_INV;
2257 				else
2258 					info.op_type = IRDMA_OP_TYPE_SEND_INV;
2259 				info.stag_to_inv = ib_wr->ex.invalidate_rkey;
2260 			}
2261 
2262 			if (ib_wr->send_flags & IB_SEND_INLINE) {
2263 				info.op.inline_send.data = (void *)(unsigned long)
2264 				    ib_wr->sg_list[0].addr;
2265 				info.op.inline_send.len = ib_wr->sg_list[0].length;
2266 				if (iwqp->ibqp.qp_type == IB_QPT_UD ||
2267 				    iwqp->ibqp.qp_type == IB_QPT_GSI) {
2268 					ah = to_iwah(ud_wr(ib_wr)->ah);
2269 					info.op.inline_send.ah_id = ah->sc_ah.ah_info.ah_idx;
2270 					info.op.inline_send.qkey = ud_wr(ib_wr)->remote_qkey;
2271 					info.op.inline_send.dest_qp = ud_wr(ib_wr)->remote_qpn;
2272 				}
2273 				err = irdma_uk_inline_send(ukqp, &info, false);
2274 			} else {
2275 				info.op.send.num_sges = ib_wr->num_sge;
2276 				info.op.send.sg_list = (struct irdma_sge *)
2277 				    ib_wr->sg_list;
2278 				if (iwqp->ibqp.qp_type == IB_QPT_UD ||
2279 				    iwqp->ibqp.qp_type == IB_QPT_GSI) {
2280 					ah = to_iwah(ud_wr(ib_wr)->ah);
2281 					info.op.send.ah_id = ah->sc_ah.ah_info.ah_idx;
2282 					info.op.send.qkey = ud_wr(ib_wr)->remote_qkey;
2283 					info.op.send.dest_qp = ud_wr(ib_wr)->remote_qpn;
2284 				}
2285 				err = irdma_uk_send(ukqp, &info, false);
2286 			}
2287 			break;
2288 		case IB_WR_RDMA_WRITE_WITH_IMM:
2289 			if (ukqp->qp_caps & IRDMA_WRITE_WITH_IMM) {
2290 				info.imm_data_valid = true;
2291 				info.imm_data = ntohl(ib_wr->ex.imm_data);
2292 			} else {
2293 				err = -EINVAL;
2294 				break;
2295 			}
2296 			/* fallthrough */
2297 		case IB_WR_RDMA_WRITE:
2298 			if (ib_wr->send_flags & IB_SEND_SOLICITED)
2299 				info.op_type = IRDMA_OP_TYPE_RDMA_WRITE_SOL;
2300 			else
2301 				info.op_type = IRDMA_OP_TYPE_RDMA_WRITE;
2302 
2303 			if (ib_wr->send_flags & IB_SEND_INLINE) {
2304 				info.op.inline_rdma_write.data = (void *)(uintptr_t)ib_wr->sg_list[0].addr;
2305 				info.op.inline_rdma_write.len = ib_wr->sg_list[0].length;
2306 				info.op.inline_rdma_write.rem_addr.tag_off = rdma_wr(ib_wr)->remote_addr;
2307 				info.op.inline_rdma_write.rem_addr.stag = rdma_wr(ib_wr)->rkey;
2308 				err = irdma_uk_inline_rdma_write(ukqp, &info, false);
2309 			} else {
2310 				info.op.rdma_write.lo_sg_list = (void *)ib_wr->sg_list;
2311 				info.op.rdma_write.num_lo_sges = ib_wr->num_sge;
2312 				info.op.rdma_write.rem_addr.tag_off = rdma_wr(ib_wr)->remote_addr;
2313 				info.op.rdma_write.rem_addr.stag = rdma_wr(ib_wr)->rkey;
2314 				err = irdma_uk_rdma_write(ukqp, &info, false);
2315 			}
2316 			break;
2317 		case IB_WR_RDMA_READ_WITH_INV:
2318 			inv_stag = true;
2319 			/* fallthrough */
2320 		case IB_WR_RDMA_READ:
2321 			if (ib_wr->num_sge >
2322 			    dev->hw_attrs.uk_attrs.max_hw_read_sges) {
2323 				err = -EINVAL;
2324 				break;
2325 			}
2326 			info.op_type = IRDMA_OP_TYPE_RDMA_READ;
2327 			info.op.rdma_read.rem_addr.tag_off = rdma_wr(ib_wr)->remote_addr;
2328 			info.op.rdma_read.rem_addr.stag = rdma_wr(ib_wr)->rkey;
2329 			info.op.rdma_read.lo_sg_list = (void *)ib_wr->sg_list;
2330 			info.op.rdma_read.num_lo_sges = ib_wr->num_sge;
2331 			err = irdma_uk_rdma_read(ukqp, &info, inv_stag, false);
2332 			break;
2333 		case IB_WR_LOCAL_INV:
2334 			info.op_type = IRDMA_OP_TYPE_INV_STAG;
2335 			info.op.inv_local_stag.target_stag = ib_wr->ex.invalidate_rkey;
2336 			err = irdma_uk_stag_local_invalidate(ukqp, &info, true);
2337 			break;
2338 		case IB_WR_REG_MR:{
2339 				struct irdma_mr *iwmr = to_iwmr(reg_wr(ib_wr)->mr);
2340 				struct irdma_pble_alloc *palloc = &iwmr->iwpbl.pble_alloc;
2341 				struct irdma_fast_reg_stag_info stag_info = {0};
2342 
2343 				stag_info.signaled = info.signaled;
2344 				stag_info.read_fence = info.read_fence;
2345 				stag_info.access_rights = irdma_get_mr_access(reg_wr(ib_wr)->access);
2346 				stag_info.stag_key = reg_wr(ib_wr)->key & 0xff;
2347 				stag_info.stag_idx = reg_wr(ib_wr)->key >> 8;
2348 				stag_info.page_size = reg_wr(ib_wr)->mr->page_size;
2349 				stag_info.wr_id = ib_wr->wr_id;
2350 				stag_info.addr_type = IRDMA_ADDR_TYPE_VA_BASED;
2351 				stag_info.va = (void *)(uintptr_t)iwmr->ibmr.iova;
2352 				stag_info.total_len = iwmr->ibmr.length;
2353 				stag_info.reg_addr_pa = *palloc->level1.addr;
2354 				stag_info.first_pm_pbl_index = palloc->level1.idx;
2355 				stag_info.local_fence = ib_wr->send_flags & IB_SEND_FENCE;
2356 				if (iwmr->npages > IRDMA_MIN_PAGES_PER_FMR)
2357 					stag_info.chunk_size = 1;
2358 				err = irdma_sc_mr_fast_register(&iwqp->sc_qp, &stag_info,
2359 								true);
2360 				break;
2361 			}
2362 		default:
2363 			err = -EINVAL;
2364 			irdma_debug(iwdev_to_idev(iwqp->iwdev),
2365 				    IRDMA_DEBUG_VERBS,
2366 				    "upost_send bad opcode = 0x%x\n",
2367 				    ib_wr->opcode);
2368 			break;
2369 		}
2370 
2371 		if (err)
2372 			break;
2373 		ib_wr = ib_wr->next;
2374 	}
2375 
2376 	if (!iwqp->flush_issued && iwqp->hw_iwarp_state <= IRDMA_QP_STATE_RTS)
2377 		irdma_uk_qp_post_wr(ukqp);
2378 	else if (iwqp->flush_issued)
2379 		mod_delayed_work(iwqp->iwdev->cleanup_wq, &iwqp->dwork_flush, IRDMA_FLUSH_DELAY_MS);
2380 	spin_unlock_irqrestore(&iwqp->lock, flags);
2381 	if (err)
2382 		*bad_wr = ib_wr;
2383 
2384 	return err;
2385 }
2386 
2387 /**
2388  * irdma_post_recv - post receive wr for kernel application
2389  * @ibqp: ib qp pointer
2390  * @ib_wr: work request for receive
2391  * @bad_wr: bad wr caused an error
2392  */
2393 static int
2394 irdma_post_recv(struct ib_qp *ibqp,
2395 		const struct ib_recv_wr *ib_wr,
2396 		const struct ib_recv_wr **bad_wr)
2397 {
2398 	struct irdma_qp *iwqp = to_iwqp(ibqp);
2399 	struct irdma_qp_uk *ukqp = &iwqp->sc_qp.qp_uk;
2400 	struct irdma_post_rq_info post_recv = {0};
2401 	struct irdma_sge *sg_list = iwqp->sg_list;
2402 	unsigned long flags;
2403 	int err = 0;
2404 
2405 	spin_lock_irqsave(&iwqp->lock, flags);
2406 
2407 	while (ib_wr) {
2408 		if (ib_wr->num_sge > ukqp->max_rq_frag_cnt) {
2409 			err = -EINVAL;
2410 			goto out;
2411 		}
2412 		post_recv.num_sges = ib_wr->num_sge;
2413 		post_recv.wr_id = ib_wr->wr_id;
2414 		irdma_copy_sg_list(sg_list, ib_wr->sg_list, ib_wr->num_sge);
2415 		post_recv.sg_list = sg_list;
2416 		err = irdma_uk_post_receive(ukqp, &post_recv);
2417 		if (err) {
2418 			irdma_debug(iwdev_to_idev(iwqp->iwdev),
2419 				    IRDMA_DEBUG_VERBS, "post_recv err %d\n",
2420 				    err);
2421 			goto out;
2422 		}
2423 
2424 		ib_wr = ib_wr->next;
2425 	}
2426 
2427 out:
2428 	if (iwqp->flush_issued)
2429 		mod_delayed_work(iwqp->iwdev->cleanup_wq, &iwqp->dwork_flush, IRDMA_FLUSH_DELAY_MS);
2430 	spin_unlock_irqrestore(&iwqp->lock, flags);
2431 	if (err)
2432 		*bad_wr = ib_wr;
2433 
2434 	return err;
2435 }
2436 
2437 /**
2438  * irdma_flush_err_to_ib_wc_status - return change flush error code to IB status
2439  * @opcode: iwarp flush code
2440  */
2441 static enum ib_wc_status
2442 irdma_flush_err_to_ib_wc_status(enum irdma_flush_opcode opcode)
2443 {
2444 	switch (opcode) {
2445 	case FLUSH_PROT_ERR:
2446 		return IB_WC_LOC_PROT_ERR;
2447 	case FLUSH_REM_ACCESS_ERR:
2448 		return IB_WC_REM_ACCESS_ERR;
2449 	case FLUSH_LOC_QP_OP_ERR:
2450 		return IB_WC_LOC_QP_OP_ERR;
2451 	case FLUSH_REM_OP_ERR:
2452 		return IB_WC_REM_OP_ERR;
2453 	case FLUSH_LOC_LEN_ERR:
2454 		return IB_WC_LOC_LEN_ERR;
2455 	case FLUSH_GENERAL_ERR:
2456 		return IB_WC_WR_FLUSH_ERR;
2457 	case FLUSH_MW_BIND_ERR:
2458 		return IB_WC_MW_BIND_ERR;
2459 	case FLUSH_RETRY_EXC_ERR:
2460 		return IB_WC_RETRY_EXC_ERR;
2461 	case FLUSH_FATAL_ERR:
2462 	default:
2463 		return IB_WC_FATAL_ERR;
2464 	}
2465 }
2466 
2467 /**
2468  * irdma_process_cqe - process cqe info
2469  * @entry: processed cqe
2470  * @cq_poll_info: cqe info
2471  */
2472 static void
2473 irdma_process_cqe(struct ib_wc *entry,
2474 		  struct irdma_cq_poll_info *cq_poll_info)
2475 {
2476 	struct irdma_sc_qp *qp;
2477 
2478 	entry->wc_flags = 0;
2479 	entry->pkey_index = 0;
2480 	entry->wr_id = cq_poll_info->wr_id;
2481 
2482 	qp = cq_poll_info->qp_handle;
2483 	entry->qp = qp->qp_uk.back_qp;
2484 
2485 	if (cq_poll_info->error) {
2486 		entry->status = (cq_poll_info->comp_status == IRDMA_COMPL_STATUS_FLUSHED) ?
2487 		    irdma_flush_err_to_ib_wc_status(cq_poll_info->minor_err) : IB_WC_GENERAL_ERR;
2488 
2489 		entry->vendor_err = cq_poll_info->major_err << 16 |
2490 		    cq_poll_info->minor_err;
2491 	} else {
2492 		entry->status = IB_WC_SUCCESS;
2493 		if (cq_poll_info->imm_valid) {
2494 			entry->ex.imm_data = htonl(cq_poll_info->imm_data);
2495 			entry->wc_flags |= IB_WC_WITH_IMM;
2496 		}
2497 		if (cq_poll_info->ud_smac_valid) {
2498 			ether_addr_copy(entry->smac, cq_poll_info->ud_smac);
2499 			entry->wc_flags |= IB_WC_WITH_SMAC;
2500 		}
2501 
2502 		if (cq_poll_info->ud_vlan_valid) {
2503 			u16 vlan = cq_poll_info->ud_vlan & EVL_VLID_MASK;
2504 
2505 			entry->sl = cq_poll_info->ud_vlan >> VLAN_PRIO_SHIFT;
2506 			if (vlan) {
2507 				entry->vlan_id = vlan;
2508 				entry->wc_flags |= IB_WC_WITH_VLAN;
2509 			}
2510 		} else {
2511 			entry->sl = 0;
2512 		}
2513 	}
2514 
2515 	switch (cq_poll_info->op_type) {
2516 	case IRDMA_OP_TYPE_RDMA_WRITE:
2517 	case IRDMA_OP_TYPE_RDMA_WRITE_SOL:
2518 		entry->opcode = IB_WC_RDMA_WRITE;
2519 		break;
2520 	case IRDMA_OP_TYPE_RDMA_READ_INV_STAG:
2521 	case IRDMA_OP_TYPE_RDMA_READ:
2522 		entry->opcode = IB_WC_RDMA_READ;
2523 		break;
2524 	case IRDMA_OP_TYPE_SEND_INV:
2525 	case IRDMA_OP_TYPE_SEND_SOL:
2526 	case IRDMA_OP_TYPE_SEND_SOL_INV:
2527 	case IRDMA_OP_TYPE_SEND:
2528 		entry->opcode = IB_WC_SEND;
2529 		break;
2530 	case IRDMA_OP_TYPE_FAST_REG_NSMR:
2531 		entry->opcode = IB_WC_REG_MR;
2532 		break;
2533 	case IRDMA_OP_TYPE_INV_STAG:
2534 		entry->opcode = IB_WC_LOCAL_INV;
2535 		break;
2536 	case IRDMA_OP_TYPE_REC_IMM:
2537 	case IRDMA_OP_TYPE_REC:
2538 		entry->opcode = cq_poll_info->op_type == IRDMA_OP_TYPE_REC_IMM ?
2539 		    IB_WC_RECV_RDMA_WITH_IMM : IB_WC_RECV;
2540 		if (qp->qp_uk.qp_type != IRDMA_QP_TYPE_ROCE_UD &&
2541 		    cq_poll_info->stag_invalid_set) {
2542 			entry->ex.invalidate_rkey = cq_poll_info->inv_stag;
2543 			entry->wc_flags |= IB_WC_WITH_INVALIDATE;
2544 		}
2545 		break;
2546 	default:
2547 		ibdev_err(&iwqp->iwdev->ibdev,
2548 			  "Invalid opcode = %d in CQE\n", cq_poll_info->op_type);
2549 		entry->status = IB_WC_GENERAL_ERR;
2550 		return;
2551 	}
2552 
2553 	if (qp->qp_uk.qp_type == IRDMA_QP_TYPE_ROCE_UD) {
2554 		entry->src_qp = cq_poll_info->ud_src_qpn;
2555 		entry->slid = 0;
2556 		entry->wc_flags |=
2557 		    (IB_WC_GRH | IB_WC_WITH_NETWORK_HDR_TYPE);
2558 		entry->network_hdr_type = cq_poll_info->ipv4 ?
2559 		    RDMA_NETWORK_IPV4 :
2560 		    RDMA_NETWORK_IPV6;
2561 	} else {
2562 		entry->src_qp = cq_poll_info->qp_id;
2563 	}
2564 
2565 	entry->byte_len = cq_poll_info->bytes_xfered;
2566 }
2567 
2568 /**
2569  * irdma_poll_one - poll one entry of the CQ
2570  * @ukcq: ukcq to poll
2571  * @cur_cqe: current CQE info to be filled in
2572  * @entry: ibv_wc object to be filled for non-extended CQ or NULL for extended CQ
2573  *
2574  * Returns the internal irdma device error code or 0 on success
2575  */
2576 static inline int
2577 irdma_poll_one(struct irdma_cq_uk *ukcq,
2578 	       struct irdma_cq_poll_info *cur_cqe,
2579 	       struct ib_wc *entry)
2580 {
2581 	int ret = irdma_uk_cq_poll_cmpl(ukcq, cur_cqe);
2582 
2583 	if (ret)
2584 		return ret;
2585 
2586 	irdma_process_cqe(entry, cur_cqe);
2587 
2588 	return 0;
2589 }
2590 
2591 /**
2592  * __irdma_poll_cq - poll cq for completion (kernel apps)
2593  * @iwcq: cq to poll
2594  * @num_entries: number of entries to poll
2595  * @entry: wr of a completed entry
2596  */
2597 static int
2598 __irdma_poll_cq(struct irdma_cq *iwcq, int num_entries, struct ib_wc *entry)
2599 {
2600 	struct list_head *tmp_node, *list_node;
2601 	struct irdma_cq_buf *last_buf = NULL;
2602 	struct irdma_cq_poll_info *cur_cqe = &iwcq->cur_cqe;
2603 	struct irdma_cq_buf *cq_buf;
2604 	int ret;
2605 	struct irdma_device *iwdev;
2606 	struct irdma_cq_uk *ukcq;
2607 	bool cq_new_cqe = false;
2608 	int resized_bufs = 0;
2609 	int npolled = 0;
2610 
2611 	iwdev = to_iwdev(iwcq->ibcq.device);
2612 	ukcq = &iwcq->sc_cq.cq_uk;
2613 
2614 	/* go through the list of previously resized CQ buffers */
2615 	list_for_each_safe(list_node, tmp_node, &iwcq->resize_list) {
2616 		cq_buf = container_of(list_node, struct irdma_cq_buf, list);
2617 		while (npolled < num_entries) {
2618 			ret = irdma_poll_one(&cq_buf->cq_uk, cur_cqe, entry + npolled);
2619 			if (!ret) {
2620 				++npolled;
2621 				cq_new_cqe = true;
2622 				continue;
2623 			}
2624 			if (ret == -ENOENT)
2625 				break;
2626 			/* QP using the CQ is destroyed. Skip reporting this CQE */
2627 			if (ret == -EFAULT) {
2628 				cq_new_cqe = true;
2629 				continue;
2630 			}
2631 			goto error;
2632 		}
2633 
2634 		/* save the resized CQ buffer which received the last cqe */
2635 		if (cq_new_cqe)
2636 			last_buf = cq_buf;
2637 		cq_new_cqe = false;
2638 	}
2639 
2640 	/* check the current CQ for new cqes */
2641 	while (npolled < num_entries) {
2642 		ret = irdma_poll_one(ukcq, cur_cqe, entry + npolled);
2643 		if (ret == -ENOENT) {
2644 			ret = irdma_generated_cmpls(iwcq, cur_cqe);
2645 			if (!ret)
2646 				irdma_process_cqe(entry + npolled, cur_cqe);
2647 		}
2648 		if (!ret) {
2649 			++npolled;
2650 			cq_new_cqe = true;
2651 			continue;
2652 		}
2653 
2654 		if (ret == -ENOENT)
2655 			break;
2656 		/* QP using the CQ is destroyed. Skip reporting this CQE */
2657 		if (ret == -EFAULT) {
2658 			cq_new_cqe = true;
2659 			continue;
2660 		}
2661 		goto error;
2662 	}
2663 
2664 	if (cq_new_cqe)
2665 		/* all previous CQ resizes are complete */
2666 		resized_bufs = irdma_process_resize_list(iwcq, iwdev, NULL);
2667 	else if (last_buf)
2668 		/* only CQ resizes up to the last_buf are complete */
2669 		resized_bufs = irdma_process_resize_list(iwcq, iwdev, last_buf);
2670 	if (resized_bufs)
2671 		/* report to the HW the number of complete CQ resizes */
2672 		irdma_uk_cq_set_resized_cnt(ukcq, resized_bufs);
2673 
2674 	return npolled;
2675 error:
2676 	irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_VERBS,
2677 		    "%s: Error polling CQ, irdma_err: %d\n", __func__, ret);
2678 
2679 	return ret;
2680 }
2681 
2682 /**
2683  * irdma_poll_cq - poll cq for completion (kernel apps)
2684  * @ibcq: cq to poll
2685  * @num_entries: number of entries to poll
2686  * @entry: wr of a completed entry
2687  */
2688 static int
2689 irdma_poll_cq(struct ib_cq *ibcq, int num_entries,
2690 	      struct ib_wc *entry)
2691 {
2692 	struct irdma_cq *iwcq;
2693 	unsigned long flags;
2694 	int ret;
2695 
2696 	iwcq = to_iwcq(ibcq);
2697 
2698 	spin_lock_irqsave(&iwcq->lock, flags);
2699 	ret = __irdma_poll_cq(iwcq, num_entries, entry);
2700 	spin_unlock_irqrestore(&iwcq->lock, flags);
2701 
2702 	return ret;
2703 }
2704 
2705 /**
2706  * irdma_req_notify_cq - arm cq kernel application
2707  * @ibcq: cq to arm
2708  * @notify_flags: notofication flags
2709  */
2710 static int
2711 irdma_req_notify_cq(struct ib_cq *ibcq,
2712 		    enum ib_cq_notify_flags notify_flags)
2713 {
2714 	struct irdma_cq *iwcq;
2715 	struct irdma_cq_uk *ukcq;
2716 	unsigned long flags;
2717 	enum irdma_cmpl_notify cq_notify = IRDMA_CQ_COMPL_EVENT;
2718 	bool promo_event = false;
2719 	int ret = 0;
2720 
2721 	iwcq = to_iwcq(ibcq);
2722 	ukcq = &iwcq->sc_cq.cq_uk;
2723 
2724 	spin_lock_irqsave(&iwcq->lock, flags);
2725 	if (notify_flags == IB_CQ_SOLICITED) {
2726 		cq_notify = IRDMA_CQ_COMPL_SOLICITED;
2727 	} else {
2728 		if (iwcq->last_notify == IRDMA_CQ_COMPL_SOLICITED)
2729 			promo_event = true;
2730 	}
2731 
2732 	if (!iwcq->armed || promo_event) {
2733 		iwcq->armed = true;
2734 		iwcq->last_notify = cq_notify;
2735 		irdma_uk_cq_request_notification(ukcq, cq_notify);
2736 	}
2737 
2738 	if ((notify_flags & IB_CQ_REPORT_MISSED_EVENTS) && !irdma_cq_empty(iwcq))
2739 		ret = 1;
2740 	spin_unlock_irqrestore(&iwcq->lock, flags);
2741 
2742 	return ret;
2743 }
2744 
2745 const char *const irdma_hw_stat_names[] = {
2746 	/* gen1 - 32-bit */
2747 	[IRDMA_HW_STAT_INDEX_IP4RXDISCARD] = "ip4InDiscards",
2748 	[IRDMA_HW_STAT_INDEX_IP4RXTRUNC] = "ip4InTruncatedPkts",
2749 	[IRDMA_HW_STAT_INDEX_IP4TXNOROUTE] = "ip4OutNoRoutes",
2750 	[IRDMA_HW_STAT_INDEX_IP6RXDISCARD] = "ip6InDiscards",
2751 	[IRDMA_HW_STAT_INDEX_IP6RXTRUNC] = "ip6InTruncatedPkts",
2752 	[IRDMA_HW_STAT_INDEX_IP6TXNOROUTE] = "ip6OutNoRoutes",
2753 	[IRDMA_HW_STAT_INDEX_TCPRTXSEG] = "tcpRetransSegs",
2754 	[IRDMA_HW_STAT_INDEX_TCPRXOPTERR] = "tcpInOptErrors",
2755 	[IRDMA_HW_STAT_INDEX_TCPRXPROTOERR] = "tcpInProtoErrors",
2756 	[IRDMA_HW_STAT_INDEX_RXVLANERR] = "rxVlanErrors",
2757 	/* gen1 - 64-bit */
2758 	[IRDMA_HW_STAT_INDEX_IP4RXOCTS] = "ip4InOctets",
2759 	[IRDMA_HW_STAT_INDEX_IP4RXPKTS] = "ip4InPkts",
2760 	[IRDMA_HW_STAT_INDEX_IP4RXFRAGS] = "ip4InReasmRqd",
2761 	[IRDMA_HW_STAT_INDEX_IP4RXMCPKTS] = "ip4InMcastPkts",
2762 	[IRDMA_HW_STAT_INDEX_IP4TXOCTS] = "ip4OutOctets",
2763 	[IRDMA_HW_STAT_INDEX_IP4TXPKTS] = "ip4OutPkts",
2764 	[IRDMA_HW_STAT_INDEX_IP4TXFRAGS] = "ip4OutSegRqd",
2765 	[IRDMA_HW_STAT_INDEX_IP4TXMCPKTS] = "ip4OutMcastPkts",
2766 	[IRDMA_HW_STAT_INDEX_IP6RXOCTS] = "ip6InOctets",
2767 	[IRDMA_HW_STAT_INDEX_IP6RXPKTS] = "ip6InPkts",
2768 	[IRDMA_HW_STAT_INDEX_IP6RXFRAGS] = "ip6InReasmRqd",
2769 	[IRDMA_HW_STAT_INDEX_IP6RXMCPKTS] = "ip6InMcastPkts",
2770 	[IRDMA_HW_STAT_INDEX_IP6TXOCTS] = "ip6OutOctets",
2771 	[IRDMA_HW_STAT_INDEX_IP6TXPKTS] = "ip6OutPkts",
2772 	[IRDMA_HW_STAT_INDEX_IP6TXFRAGS] = "ip6OutSegRqd",
2773 	[IRDMA_HW_STAT_INDEX_IP6TXMCPKTS] = "ip6OutMcastPkts",
2774 	[IRDMA_HW_STAT_INDEX_TCPRXSEGS] = "tcpInSegs",
2775 	[IRDMA_HW_STAT_INDEX_TCPTXSEG] = "tcpOutSegs",
2776 	[IRDMA_HW_STAT_INDEX_RDMARXRDS] = "iwInRdmaReads",
2777 	[IRDMA_HW_STAT_INDEX_RDMARXSNDS] = "iwInRdmaSends",
2778 	[IRDMA_HW_STAT_INDEX_RDMARXWRS] = "iwInRdmaWrites",
2779 	[IRDMA_HW_STAT_INDEX_RDMATXRDS] = "iwOutRdmaReads",
2780 	[IRDMA_HW_STAT_INDEX_RDMATXSNDS] = "iwOutRdmaSends",
2781 	[IRDMA_HW_STAT_INDEX_RDMATXWRS] = "iwOutRdmaWrites",
2782 	[IRDMA_HW_STAT_INDEX_RDMAVBND] = "iwRdmaBnd",
2783 	[IRDMA_HW_STAT_INDEX_RDMAVINV] = "iwRdmaInv",
2784 
2785 	/* gen2 - 32-bit */
2786 	[IRDMA_HW_STAT_INDEX_RXRPCNPHANDLED] = "cnpHandled",
2787 	[IRDMA_HW_STAT_INDEX_RXRPCNPIGNORED] = "cnpIgnored",
2788 	[IRDMA_HW_STAT_INDEX_TXNPCNPSENT] = "cnpSent",
2789 	/* gen2 - 64-bit */
2790 	[IRDMA_HW_STAT_INDEX_IP4RXMCOCTS] = "ip4InMcastOctets",
2791 	[IRDMA_HW_STAT_INDEX_IP4TXMCOCTS] = "ip4OutMcastOctets",
2792 	[IRDMA_HW_STAT_INDEX_IP6RXMCOCTS] = "ip6InMcastOctets",
2793 	[IRDMA_HW_STAT_INDEX_IP6TXMCOCTS] = "ip6OutMcastOctets",
2794 	[IRDMA_HW_STAT_INDEX_UDPRXPKTS] = "RxUDP",
2795 	[IRDMA_HW_STAT_INDEX_UDPTXPKTS] = "TxUDP",
2796 	[IRDMA_HW_STAT_INDEX_RXNPECNMARKEDPKTS] = "RxECNMrkd",
2797 };
2798 
2799 /**
2800  * mcast_list_add -  Add a new mcast item to list
2801  * @rf: RDMA PCI function
2802  * @new_elem: pointer to element to add
2803  */
2804 static void
2805 mcast_list_add(struct irdma_pci_f *rf,
2806 	       struct mc_table_list *new_elem)
2807 {
2808 	list_add(&new_elem->list, &rf->mc_qht_list.list);
2809 }
2810 
2811 /**
2812  * mcast_list_del - Remove an mcast item from list
2813  * @mc_qht_elem: pointer to mcast table list element
2814  */
2815 static void
2816 mcast_list_del(struct mc_table_list *mc_qht_elem)
2817 {
2818 	if (mc_qht_elem)
2819 		list_del(&mc_qht_elem->list);
2820 }
2821 
2822 /**
2823  * mcast_list_lookup_ip - Search mcast list for address
2824  * @rf: RDMA PCI function
2825  * @ip_mcast: pointer to mcast IP address
2826  */
2827 static struct mc_table_list *
2828 mcast_list_lookup_ip(struct irdma_pci_f *rf,
2829 		     u32 *ip_mcast)
2830 {
2831 	struct mc_table_list *mc_qht_el;
2832 	struct list_head *pos, *q;
2833 
2834 	list_for_each_safe(pos, q, &rf->mc_qht_list.list) {
2835 		mc_qht_el = list_entry(pos, struct mc_table_list, list);
2836 		if (!memcmp(mc_qht_el->mc_info.dest_ip, ip_mcast,
2837 			    sizeof(mc_qht_el->mc_info.dest_ip)))
2838 			return mc_qht_el;
2839 	}
2840 
2841 	return NULL;
2842 }
2843 
2844 /**
2845  * irdma_mcast_cqp_op - perform a mcast cqp operation
2846  * @iwdev: irdma device
2847  * @mc_grp_ctx: mcast group info
2848  * @op: operation
2849  *
2850  * returns error status
2851  */
2852 static int
2853 irdma_mcast_cqp_op(struct irdma_device *iwdev,
2854 		   struct irdma_mcast_grp_info *mc_grp_ctx, u8 op)
2855 {
2856 	struct cqp_cmds_info *cqp_info;
2857 	struct irdma_cqp_request *cqp_request;
2858 	int status;
2859 
2860 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
2861 	if (!cqp_request)
2862 		return -ENOMEM;
2863 
2864 	cqp_request->info.in.u.mc_create.info = *mc_grp_ctx;
2865 	cqp_info = &cqp_request->info;
2866 	cqp_info->cqp_cmd = op;
2867 	cqp_info->post_sq = 1;
2868 	cqp_info->in.u.mc_create.scratch = (uintptr_t)cqp_request;
2869 	cqp_info->in.u.mc_create.cqp = &iwdev->rf->cqp.sc_cqp;
2870 	status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
2871 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
2872 
2873 	return status;
2874 }
2875 
2876 /**
2877  * irdma_attach_mcast - attach a qp to a multicast group
2878  * @ibqp: ptr to qp
2879  * @ibgid: pointer to global ID
2880  * @lid: local ID
2881  *
2882  * returns error status
2883  */
2884 static int
2885 irdma_attach_mcast(struct ib_qp *ibqp, union ib_gid *ibgid, u16 lid)
2886 {
2887 	struct irdma_qp *iwqp = to_iwqp(ibqp);
2888 	struct irdma_device *iwdev = iwqp->iwdev;
2889 	struct irdma_pci_f *rf = iwdev->rf;
2890 	struct mc_table_list *mc_qht_elem;
2891 	struct irdma_mcast_grp_ctx_entry_info mcg_info = {0};
2892 	unsigned long flags;
2893 	u32 ip_addr[4] = {0};
2894 	u32 mgn;
2895 	u32 no_mgs;
2896 	int ret = 0;
2897 	bool ipv4;
2898 	u16 vlan_id;
2899 	union {
2900 		struct sockaddr saddr;
2901 		struct sockaddr_in saddr_in;
2902 		struct sockaddr_in6 saddr_in6;
2903 	} sgid_addr;
2904 	unsigned char dmac[ETH_ALEN];
2905 
2906 	rdma_gid2ip((struct sockaddr *)&sgid_addr, ibgid);
2907 
2908 	if (!ipv6_addr_v4mapped((struct in6_addr *)ibgid)) {
2909 		irdma_copy_ip_ntohl(ip_addr,
2910 				    sgid_addr.saddr_in6.sin6_addr.__u6_addr.__u6_addr32);
2911 		irdma_netdev_vlan_ipv6(ip_addr, &vlan_id, NULL);
2912 		ipv4 = false;
2913 		irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_VERBS,
2914 			    "qp_id=%d, IP6address=%pI6\n", ibqp->qp_num,
2915 			    ip_addr);
2916 		irdma_mcast_mac_v6(ip_addr, dmac);
2917 	} else {
2918 		ip_addr[0] = ntohl(sgid_addr.saddr_in.sin_addr.s_addr);
2919 		ipv4 = true;
2920 		vlan_id = irdma_get_vlan_ipv4(ip_addr);
2921 		irdma_mcast_mac_v4(ip_addr, dmac);
2922 		irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_VERBS,
2923 			    "qp_id=%d, IP4address=%pI4, MAC=%pM\n",
2924 			    ibqp->qp_num, ip_addr, dmac);
2925 	}
2926 
2927 	spin_lock_irqsave(&rf->qh_list_lock, flags);
2928 	mc_qht_elem = mcast_list_lookup_ip(rf, ip_addr);
2929 	if (!mc_qht_elem) {
2930 		struct irdma_dma_mem *dma_mem_mc;
2931 
2932 		spin_unlock_irqrestore(&rf->qh_list_lock, flags);
2933 		mc_qht_elem = kzalloc(sizeof(*mc_qht_elem), GFP_KERNEL);
2934 		if (!mc_qht_elem)
2935 			return -ENOMEM;
2936 
2937 		mc_qht_elem->mc_info.ipv4_valid = ipv4;
2938 		memcpy(mc_qht_elem->mc_info.dest_ip, ip_addr,
2939 		       sizeof(mc_qht_elem->mc_info.dest_ip));
2940 		ret = irdma_alloc_rsrc(rf, rf->allocated_mcgs, rf->max_mcg,
2941 				       &mgn, &rf->next_mcg);
2942 		if (ret) {
2943 			kfree(mc_qht_elem);
2944 			return -ENOMEM;
2945 		}
2946 
2947 		mc_qht_elem->mc_info.mgn = mgn;
2948 		dma_mem_mc = &mc_qht_elem->mc_grp_ctx.dma_mem_mc;
2949 		dma_mem_mc->size = sizeof(u64)* IRDMA_MAX_MGS_PER_CTX;
2950 		dma_mem_mc->va = irdma_allocate_dma_mem(&rf->hw, dma_mem_mc,
2951 							dma_mem_mc->size,
2952 							IRDMA_HW_PAGE_SIZE);
2953 		if (!dma_mem_mc->va) {
2954 			irdma_free_rsrc(rf, rf->allocated_mcgs, mgn);
2955 			kfree(mc_qht_elem);
2956 			return -ENOMEM;
2957 		}
2958 
2959 		mc_qht_elem->mc_grp_ctx.mg_id = (u16)mgn;
2960 		memcpy(mc_qht_elem->mc_grp_ctx.dest_ip_addr, ip_addr,
2961 		       sizeof(mc_qht_elem->mc_grp_ctx.dest_ip_addr));
2962 		mc_qht_elem->mc_grp_ctx.ipv4_valid = ipv4;
2963 		mc_qht_elem->mc_grp_ctx.vlan_id = vlan_id;
2964 		if (vlan_id < VLAN_N_VID)
2965 			mc_qht_elem->mc_grp_ctx.vlan_valid = true;
2966 		mc_qht_elem->mc_grp_ctx.hmc_fcn_id = iwdev->vsi.fcn_id;
2967 		mc_qht_elem->mc_grp_ctx.qs_handle =
2968 		    iwqp->sc_qp.vsi->qos[iwqp->sc_qp.user_pri].qs_handle;
2969 		ether_addr_copy(mc_qht_elem->mc_grp_ctx.dest_mac_addr, dmac);
2970 
2971 		spin_lock_irqsave(&rf->qh_list_lock, flags);
2972 		mcast_list_add(rf, mc_qht_elem);
2973 	} else {
2974 		if (mc_qht_elem->mc_grp_ctx.no_of_mgs ==
2975 		    IRDMA_MAX_MGS_PER_CTX) {
2976 			spin_unlock_irqrestore(&rf->qh_list_lock, flags);
2977 			return -ENOMEM;
2978 		}
2979 	}
2980 
2981 	mcg_info.qp_id = iwqp->ibqp.qp_num;
2982 	no_mgs = mc_qht_elem->mc_grp_ctx.no_of_mgs;
2983 	irdma_sc_add_mcast_grp(&mc_qht_elem->mc_grp_ctx, &mcg_info);
2984 	spin_unlock_irqrestore(&rf->qh_list_lock, flags);
2985 
2986 	/* Only if there is a change do we need to modify or create */
2987 	if (!no_mgs) {
2988 		ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
2989 					 IRDMA_OP_MC_CREATE);
2990 	} else if (no_mgs != mc_qht_elem->mc_grp_ctx.no_of_mgs) {
2991 		ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
2992 					 IRDMA_OP_MC_MODIFY);
2993 	} else {
2994 		return 0;
2995 	}
2996 
2997 	if (ret)
2998 		goto error;
2999 
3000 	return 0;
3001 
3002 error:
3003 	irdma_sc_del_mcast_grp(&mc_qht_elem->mc_grp_ctx, &mcg_info);
3004 	if (!mc_qht_elem->mc_grp_ctx.no_of_mgs) {
3005 		mcast_list_del(mc_qht_elem);
3006 		irdma_free_dma_mem(&rf->hw,
3007 				   &mc_qht_elem->mc_grp_ctx.dma_mem_mc);
3008 		irdma_free_rsrc(rf, rf->allocated_mcgs,
3009 				mc_qht_elem->mc_grp_ctx.mg_id);
3010 		kfree(mc_qht_elem);
3011 	}
3012 
3013 	return ret;
3014 }
3015 
3016 /**
3017  * irdma_detach_mcast - detach a qp from a multicast group
3018  * @ibqp: ptr to qp
3019  * @ibgid: pointer to global ID
3020  * @lid: local ID
3021  *
3022  * returns error status
3023  */
3024 static int
3025 irdma_detach_mcast(struct ib_qp *ibqp, union ib_gid *ibgid, u16 lid)
3026 {
3027 	struct irdma_qp *iwqp = to_iwqp(ibqp);
3028 	struct irdma_device *iwdev = iwqp->iwdev;
3029 	struct irdma_pci_f *rf = iwdev->rf;
3030 	u32 ip_addr[4] = {0};
3031 	struct mc_table_list *mc_qht_elem;
3032 	struct irdma_mcast_grp_ctx_entry_info mcg_info = {0};
3033 	int ret;
3034 	unsigned long flags;
3035 	union {
3036 		struct sockaddr saddr;
3037 		struct sockaddr_in saddr_in;
3038 		struct sockaddr_in6 saddr_in6;
3039 	} sgid_addr;
3040 
3041 	rdma_gid2ip((struct sockaddr *)&sgid_addr, ibgid);
3042 	if (!ipv6_addr_v4mapped((struct in6_addr *)ibgid))
3043 		irdma_copy_ip_ntohl(ip_addr,
3044 				    sgid_addr.saddr_in6.sin6_addr.__u6_addr.__u6_addr32);
3045 	else
3046 		ip_addr[0] = ntohl(sgid_addr.saddr_in.sin_addr.s_addr);
3047 
3048 	spin_lock_irqsave(&rf->qh_list_lock, flags);
3049 	mc_qht_elem = mcast_list_lookup_ip(rf, ip_addr);
3050 	if (!mc_qht_elem) {
3051 		spin_unlock_irqrestore(&rf->qh_list_lock, flags);
3052 		irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_VERBS,
3053 			    "address not found MCG\n");
3054 		return 0;
3055 	}
3056 
3057 	mcg_info.qp_id = iwqp->ibqp.qp_num;
3058 	irdma_sc_del_mcast_grp(&mc_qht_elem->mc_grp_ctx, &mcg_info);
3059 	if (!mc_qht_elem->mc_grp_ctx.no_of_mgs) {
3060 		mcast_list_del(mc_qht_elem);
3061 		spin_unlock_irqrestore(&rf->qh_list_lock, flags);
3062 		ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
3063 					 IRDMA_OP_MC_DESTROY);
3064 		if (ret) {
3065 			irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_VERBS,
3066 				    "failed MC_DESTROY MCG\n");
3067 			spin_lock_irqsave(&rf->qh_list_lock, flags);
3068 			mcast_list_add(rf, mc_qht_elem);
3069 			spin_unlock_irqrestore(&rf->qh_list_lock, flags);
3070 			return -EAGAIN;
3071 		}
3072 
3073 		irdma_free_dma_mem(&rf->hw,
3074 				   &mc_qht_elem->mc_grp_ctx.dma_mem_mc);
3075 		irdma_free_rsrc(rf, rf->allocated_mcgs,
3076 				mc_qht_elem->mc_grp_ctx.mg_id);
3077 		kfree(mc_qht_elem);
3078 	} else {
3079 		spin_unlock_irqrestore(&rf->qh_list_lock, flags);
3080 		ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
3081 					 IRDMA_OP_MC_MODIFY);
3082 		if (ret) {
3083 			irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_VERBS,
3084 				    "failed Modify MCG\n");
3085 			return ret;
3086 		}
3087 	}
3088 
3089 	return 0;
3090 }
3091 
3092 /**
3093  * irdma_query_ah - Query address handle
3094  * @ibah: pointer to address handle
3095  * @ah_attr: address handle attributes
3096  */
3097 static int
3098 irdma_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr)
3099 {
3100 	struct irdma_ah *ah = to_iwah(ibah);
3101 
3102 	memset(ah_attr, 0, sizeof(*ah_attr));
3103 	if (ah->av.attrs.ah_flags & IB_AH_GRH) {
3104 		ah_attr->ah_flags = IB_AH_GRH;
3105 		ah_attr->grh.flow_label = ah->sc_ah.ah_info.flow_label;
3106 		ah_attr->grh.traffic_class = ah->sc_ah.ah_info.tc_tos;
3107 		ah_attr->grh.hop_limit = ah->sc_ah.ah_info.hop_ttl;
3108 		ah_attr->grh.sgid_index = ah->sgid_index;
3109 		ah_attr->grh.sgid_index = ah->sgid_index;
3110 		memcpy(&ah_attr->grh.dgid, &ah->dgid,
3111 		       sizeof(ah_attr->grh.dgid));
3112 	}
3113 
3114 	return 0;
3115 }
3116 
3117 static __be64 irdma_mac_to_guid(struct ifnet *ndev){
3118 	unsigned char *mac = IF_LLADDR(ndev);
3119 	__be64 guid;
3120 	unsigned char *dst = (unsigned char *)&guid;
3121 
3122 	dst[0] = mac[0] ^ 2;
3123 	dst[1] = mac[1];
3124 	dst[2] = mac[2];
3125 	dst[3] = 0xff;
3126 	dst[4] = 0xfe;
3127 	dst[5] = mac[3];
3128 	dst[6] = mac[4];
3129 	dst[7] = mac[5];
3130 
3131 	return guid;
3132 }
3133 
3134 static struct ifnet *
3135 irdma_get_netdev(struct ib_device *ibdev, u8 port_num)
3136 {
3137 	struct irdma_device *iwdev = to_iwdev(ibdev);
3138 
3139 	if (iwdev->netdev) {
3140 		dev_hold(iwdev->netdev);
3141 		return iwdev->netdev;
3142 	}
3143 
3144 	return NULL;
3145 }
3146 
3147 static void
3148 irdma_set_device_ops(struct ib_device *ibdev)
3149 {
3150 	struct ib_device *dev_ops = ibdev;
3151 
3152 	dev_ops->ops.driver_id = RDMA_DRIVER_I40IW;
3153 	dev_ops->ops.size_ib_ah = IRDMA_SET_RDMA_OBJ_SIZE(ib_ah, irdma_ah, ibah);
3154 	dev_ops->ops.size_ib_cq = IRDMA_SET_RDMA_OBJ_SIZE(ib_cq, irdma_cq, ibcq);
3155 	dev_ops->ops.size_ib_pd = IRDMA_SET_RDMA_OBJ_SIZE(ib_pd, irdma_pd, ibpd);
3156 	dev_ops->ops.size_ib_ucontext = IRDMA_SET_RDMA_OBJ_SIZE(ib_ucontext,
3157 								irdma_ucontext,
3158 								ibucontext);
3159 
3160 	dev_ops->alloc_hw_stats = irdma_alloc_hw_stats;
3161 	dev_ops->alloc_mr = irdma_alloc_mr;
3162 	dev_ops->alloc_mw = irdma_alloc_mw;
3163 	dev_ops->alloc_pd = irdma_alloc_pd;
3164 	dev_ops->alloc_ucontext = irdma_alloc_ucontext;
3165 	dev_ops->create_cq = irdma_create_cq;
3166 	dev_ops->create_qp = irdma_create_qp;
3167 	dev_ops->dealloc_mw = irdma_dealloc_mw;
3168 	dev_ops->dealloc_pd = irdma_dealloc_pd;
3169 	dev_ops->dealloc_ucontext = irdma_dealloc_ucontext;
3170 	dev_ops->dereg_mr = irdma_dereg_mr;
3171 	dev_ops->destroy_cq = irdma_destroy_cq;
3172 	dev_ops->destroy_qp = irdma_destroy_qp;
3173 	dev_ops->disassociate_ucontext = irdma_disassociate_ucontext;
3174 	dev_ops->get_dev_fw_str = irdma_get_dev_fw_str;
3175 	dev_ops->get_dma_mr = irdma_get_dma_mr;
3176 	dev_ops->get_hw_stats = irdma_get_hw_stats;
3177 	dev_ops->get_netdev = irdma_get_netdev;
3178 	dev_ops->map_mr_sg = irdma_map_mr_sg;
3179 	dev_ops->mmap = irdma_mmap;
3180 	dev_ops->mmap_free = irdma_mmap_free;
3181 	dev_ops->poll_cq = irdma_poll_cq;
3182 	dev_ops->post_recv = irdma_post_recv;
3183 	dev_ops->post_send = irdma_post_send;
3184 	dev_ops->query_device = irdma_query_device;
3185 	dev_ops->query_port = irdma_query_port;
3186 	dev_ops->modify_port = irdma_modify_port;
3187 	dev_ops->query_qp = irdma_query_qp;
3188 	dev_ops->reg_user_mr = irdma_reg_user_mr;
3189 	dev_ops->req_notify_cq = irdma_req_notify_cq;
3190 	dev_ops->resize_cq = irdma_resize_cq;
3191 }
3192 
3193 static void
3194 irdma_set_device_roce_ops(struct ib_device *ibdev)
3195 {
3196 	struct ib_device *dev_ops = ibdev;
3197 
3198 	dev_ops->attach_mcast = irdma_attach_mcast;
3199 	dev_ops->create_ah = irdma_create_ah;
3200 	dev_ops->destroy_ah = irdma_destroy_ah;
3201 	dev_ops->detach_mcast = irdma_detach_mcast;
3202 	dev_ops->get_link_layer = irdma_get_link_layer;
3203 	dev_ops->get_port_immutable = irdma_roce_port_immutable;
3204 	dev_ops->modify_qp = irdma_modify_qp_roce;
3205 	dev_ops->query_ah = irdma_query_ah;
3206 	dev_ops->query_gid = irdma_query_gid_roce;
3207 	dev_ops->query_pkey = irdma_query_pkey;
3208 	ibdev->add_gid = irdma_add_gid;
3209 	ibdev->del_gid = irdma_del_gid;
3210 }
3211 
3212 static void
3213 irdma_set_device_iw_ops(struct ib_device *ibdev)
3214 {
3215 	struct ib_device *dev_ops = ibdev;
3216 
3217 	ibdev->uverbs_cmd_mask |=
3218 	    (1ull << IB_USER_VERBS_CMD_CREATE_AH) |
3219 	    (1ull << IB_USER_VERBS_CMD_DESTROY_AH);
3220 
3221 	dev_ops->create_ah = irdma_create_ah_stub;
3222 	dev_ops->destroy_ah = irdma_destroy_ah_stub;
3223 	dev_ops->get_port_immutable = irdma_iw_port_immutable;
3224 	dev_ops->modify_qp = irdma_modify_qp;
3225 	dev_ops->query_gid = irdma_query_gid;
3226 	dev_ops->query_pkey = irdma_iw_query_pkey;
3227 }
3228 
3229 /**
3230  * irdma_init_roce_device - initialization of roce rdma device
3231  * @iwdev: irdma device
3232  */
3233 static void
3234 irdma_init_roce_device(struct irdma_device *iwdev)
3235 {
3236 	kc_set_roce_uverbs_cmd_mask(iwdev);
3237 	iwdev->ibdev.node_type = RDMA_NODE_IB_CA;
3238 	iwdev->ibdev.node_guid = irdma_mac_to_guid(iwdev->netdev);
3239 	irdma_set_device_roce_ops(&iwdev->ibdev);
3240 }
3241 
3242 /**
3243  * irdma_init_iw_device - initialization of iwarp rdma device
3244  * @iwdev: irdma device
3245  */
3246 static int
3247 irdma_init_iw_device(struct irdma_device *iwdev)
3248 {
3249 	struct ifnet *netdev = iwdev->netdev;
3250 
3251 	iwdev->ibdev.node_type = RDMA_NODE_RNIC;
3252 	ether_addr_copy((u8 *)&iwdev->ibdev.node_guid, IF_LLADDR(netdev));
3253 	iwdev->ibdev.iwcm = kzalloc(sizeof(*iwdev->ibdev.iwcm), GFP_KERNEL);
3254 	if (!iwdev->ibdev.iwcm)
3255 		return -ENOMEM;
3256 
3257 	iwdev->ibdev.iwcm->add_ref = irdma_qp_add_ref;
3258 	iwdev->ibdev.iwcm->rem_ref = irdma_qp_rem_ref;
3259 	iwdev->ibdev.iwcm->get_qp = irdma_get_qp;
3260 	iwdev->ibdev.iwcm->connect = irdma_connect;
3261 	iwdev->ibdev.iwcm->accept = irdma_accept;
3262 	iwdev->ibdev.iwcm->reject = irdma_reject;
3263 	iwdev->ibdev.iwcm->create_listen = irdma_create_listen;
3264 	iwdev->ibdev.iwcm->destroy_listen = irdma_destroy_listen;
3265 	memcpy(iwdev->ibdev.iwcm->ifname, if_name(netdev),
3266 	       sizeof(iwdev->ibdev.iwcm->ifname));
3267 	irdma_set_device_iw_ops(&iwdev->ibdev);
3268 
3269 	return 0;
3270 }
3271 
3272 /**
3273  * irdma_init_rdma_device - initialization of rdma device
3274  * @iwdev: irdma device
3275  */
3276 static int
3277 irdma_init_rdma_device(struct irdma_device *iwdev)
3278 {
3279 	struct pci_dev *pcidev = iwdev->rf->pcidev;
3280 	int ret;
3281 
3282 	iwdev->ibdev.owner = THIS_MODULE;
3283 	iwdev->ibdev.uverbs_abi_ver = IRDMA_ABI_VER;
3284 	kc_set_rdma_uverbs_cmd_mask(iwdev);
3285 
3286 	if (iwdev->roce_mode) {
3287 		irdma_init_roce_device(iwdev);
3288 	} else {
3289 		ret = irdma_init_iw_device(iwdev);
3290 		if (ret)
3291 			return ret;
3292 	}
3293 	iwdev->ibdev.phys_port_cnt = 1;
3294 	iwdev->ibdev.num_comp_vectors = iwdev->rf->ceqs_count;
3295 	iwdev->ibdev.dev.parent = iwdev->rf->dev_ctx.dev;
3296 	set_ibdev_dma_device(iwdev->ibdev, &pcidev->dev);
3297 	irdma_set_device_ops(&iwdev->ibdev);
3298 
3299 	return 0;
3300 }
3301 
3302 /**
3303  * irdma_port_ibevent - indicate port event
3304  * @iwdev: irdma device
3305  */
3306 void
3307 irdma_port_ibevent(struct irdma_device *iwdev)
3308 {
3309 	struct ib_event event;
3310 
3311 	event.device = &iwdev->ibdev;
3312 	event.element.port_num = 1;
3313 	event.event =
3314 	    iwdev->iw_status ? IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3315 	ib_dispatch_event(&event);
3316 }
3317 
3318 /**
3319  * irdma_ib_unregister_device - unregister rdma device from IB
3320  * core
3321  * @iwdev: irdma device
3322  */
3323 void
3324 irdma_ib_unregister_device(struct irdma_device *iwdev)
3325 {
3326 	iwdev->iw_status = 0;
3327 	irdma_port_ibevent(iwdev);
3328 	ib_unregister_device(&iwdev->ibdev);
3329 	kfree(iwdev->ibdev.iwcm);
3330 	iwdev->ibdev.iwcm = NULL;
3331 }
3332 
3333 /**
3334  * irdma_ib_register_device - register irdma device to IB core
3335  * @iwdev: irdma device
3336  */
3337 int
3338 irdma_ib_register_device(struct irdma_device *iwdev)
3339 {
3340 	int ret;
3341 
3342 	ret = irdma_init_rdma_device(iwdev);
3343 	if (ret)
3344 		return ret;
3345 
3346 	sprintf(iwdev->ibdev.name, "irdma-%s", if_name(iwdev->netdev));
3347 	ret = ib_register_device(&iwdev->ibdev, NULL);
3348 	if (ret)
3349 		goto error;
3350 
3351 	iwdev->iw_status = 1;
3352 	irdma_port_ibevent(iwdev);
3353 
3354 	return 0;
3355 
3356 error:
3357 	kfree(iwdev->ibdev.iwcm);
3358 	iwdev->ibdev.iwcm = NULL;
3359 	if (ret)
3360 		irdma_debug(&iwdev->rf->sc_dev, IRDMA_DEBUG_VERBS,
3361 			    "Register RDMA device fail\n");
3362 
3363 	return ret;
3364 }
3365