xref: /linux/drivers/infiniband/hw/irdma/verbs.c (revision 9e7e6633458362db72427b48effad8d759131c35)
1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /* Copyright (c) 2015 - 2021 Intel Corporation */
3 #include "main.h"
4 
5 /**
6  * irdma_query_device - get device attributes
7  * @ibdev: device pointer from stack
8  * @props: returning device attributes
9  * @udata: user data
10  */
11 static int irdma_query_device(struct ib_device *ibdev,
12 			      struct ib_device_attr *props,
13 			      struct ib_udata *udata)
14 {
15 	struct irdma_device *iwdev = to_iwdev(ibdev);
16 	struct irdma_pci_f *rf = iwdev->rf;
17 	struct pci_dev *pcidev = iwdev->rf->pcidev;
18 	struct irdma_hw_attrs *hw_attrs = &rf->sc_dev.hw_attrs;
19 	int err;
20 
21 	err = ib_is_udata_in_empty(udata);
22 	if (err)
23 		return err;
24 
25 	memset(props, 0, sizeof(*props));
26 	addrconf_addr_eui48((u8 *)&props->sys_image_guid,
27 			    iwdev->netdev->dev_addr);
28 	props->fw_ver = (u64)irdma_fw_major_ver(&rf->sc_dev) << 32 |
29 			irdma_fw_minor_ver(&rf->sc_dev);
30 	props->device_cap_flags = IB_DEVICE_MEM_WINDOW |
31 				  IB_DEVICE_MEM_MGT_EXTENSIONS;
32 	if (hw_attrs->uk_attrs.hw_rev < IRDMA_GEN_3)
33 		props->kernel_cap_flags = IBK_LOCAL_DMA_LKEY;
34 	props->vendor_id = pcidev->vendor;
35 	props->vendor_part_id = pcidev->device;
36 
37 	props->hw_ver = rf->pcidev->revision;
38 	props->page_size_cap = hw_attrs->page_size_cap;
39 	props->max_mr_size = hw_attrs->max_mr_size;
40 	props->max_qp = rf->max_qp - rf->used_qps;
41 	props->max_qp_wr = hw_attrs->max_qp_wr;
42 	props->max_send_sge = hw_attrs->uk_attrs.max_hw_wq_frags;
43 	props->max_recv_sge = hw_attrs->uk_attrs.max_hw_wq_frags;
44 	props->max_cq = rf->max_cq - rf->used_cqs;
45 	props->max_cqe = rf->max_cqe - 1;
46 	props->max_mr = rf->max_mr - rf->used_mrs;
47 	if (hw_attrs->uk_attrs.hw_rev >= IRDMA_GEN_3)
48 		props->max_mw = props->max_mr;
49 	props->max_pd = rf->max_pd - rf->used_pds;
50 	props->max_sge_rd = hw_attrs->uk_attrs.max_hw_read_sges;
51 	props->max_qp_rd_atom = hw_attrs->max_hw_ird;
52 	props->max_qp_init_rd_atom = hw_attrs->max_hw_ord;
53 	if (rdma_protocol_roce(ibdev, 1)) {
54 		props->device_cap_flags |= IB_DEVICE_RC_RNR_NAK_GEN;
55 		props->max_pkeys = IRDMA_PKEY_TBL_SZ;
56 	}
57 
58 	props->max_ah = rf->max_ah;
59 	props->max_mcast_grp = rf->max_mcg;
60 	props->max_mcast_qp_attach = IRDMA_MAX_MGS_PER_CTX;
61 	props->max_total_mcast_qp_attach = rf->max_qp * IRDMA_MAX_MGS_PER_CTX;
62 	props->max_fast_reg_page_list_len = IRDMA_MAX_PAGES_PER_FMR;
63 	props->max_srq = rf->max_srq - rf->used_srqs;
64 	props->max_srq_wr = IRDMA_MAX_SRQ_WRS;
65 	props->max_srq_sge = hw_attrs->uk_attrs.max_hw_wq_frags;
66 	if (hw_attrs->uk_attrs.feature_flags & IRDMA_FEATURE_ATOMIC_OPS)
67 		props->atomic_cap = IB_ATOMIC_HCA;
68 	else
69 		props->atomic_cap = IB_ATOMIC_NONE;
70 	props->masked_atomic_cap = props->atomic_cap;
71 	if (hw_attrs->uk_attrs.hw_rev >= IRDMA_GEN_3) {
72 #define HCA_CORE_CLOCK_KHZ 1000000UL
73 		props->timestamp_mask = GENMASK(31, 0);
74 		props->hca_core_clock = HCA_CORE_CLOCK_KHZ;
75 	}
76 	if (hw_attrs->uk_attrs.hw_rev >= IRDMA_GEN_3)
77 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
78 
79 	return ib_respond_empty_udata(udata);
80 }
81 
82 /**
83  * irdma_query_port - get port attributes
84  * @ibdev: device pointer from stack
85  * @port: port number for query
86  * @props: returning device attributes
87  */
88 static int irdma_query_port(struct ib_device *ibdev, u32 port,
89 			    struct ib_port_attr *props)
90 {
91 	struct irdma_device *iwdev = to_iwdev(ibdev);
92 	struct net_device *netdev = iwdev->netdev;
93 
94 	/* no need to zero out pros here. done by caller */
95 
96 	props->max_mtu = IB_MTU_4096;
97 	props->active_mtu = ib_mtu_int_to_enum(netdev->mtu);
98 	props->lid = 1;
99 	props->lmc = 0;
100 	props->sm_lid = 0;
101 	props->sm_sl = 0;
102 	if (netif_carrier_ok(netdev) && netif_running(netdev)) {
103 		props->state = IB_PORT_ACTIVE;
104 		props->phys_state = IB_PORT_PHYS_STATE_LINK_UP;
105 	} else {
106 		props->state = IB_PORT_DOWN;
107 		props->phys_state = IB_PORT_PHYS_STATE_DISABLED;
108 	}
109 
110 	ib_get_eth_speed(ibdev, port, &props->active_speed,
111 			 &props->active_width);
112 
113 	if (rdma_protocol_roce(ibdev, 1)) {
114 		props->gid_tbl_len = 32;
115 		props->ip_gids = true;
116 		props->pkey_tbl_len = IRDMA_PKEY_TBL_SZ;
117 	} else {
118 		props->gid_tbl_len = 1;
119 	}
120 	props->qkey_viol_cntr = 0;
121 	props->port_cap_flags |= IB_PORT_CM_SUP | IB_PORT_REINIT_SUP;
122 	props->max_msg_sz = iwdev->rf->sc_dev.hw_attrs.max_hw_outbound_msg_size;
123 
124 	return 0;
125 }
126 
127 /**
128  * irdma_disassociate_ucontext - Disassociate user context
129  * @context: ib user context
130  */
131 static void irdma_disassociate_ucontext(struct ib_ucontext *context)
132 {
133 }
134 
135 static int irdma_mmap_legacy(struct irdma_ucontext *ucontext,
136 			     struct vm_area_struct *vma)
137 {
138 	u64 pfn;
139 
140 	if (vma->vm_pgoff || vma->vm_end - vma->vm_start != PAGE_SIZE)
141 		return -EINVAL;
142 
143 	vma->vm_private_data = ucontext;
144 	pfn = ((uintptr_t)ucontext->iwdev->rf->sc_dev.hw_regs[IRDMA_DB_ADDR_OFFSET] +
145 	       pci_resource_start(ucontext->iwdev->rf->pcidev, 0)) >> PAGE_SHIFT;
146 
147 	return rdma_user_mmap_io(&ucontext->ibucontext, vma, pfn, PAGE_SIZE,
148 				 pgprot_noncached(vma->vm_page_prot), NULL);
149 }
150 
151 static void irdma_mmap_free(struct rdma_user_mmap_entry *rdma_entry)
152 {
153 	struct irdma_user_mmap_entry *entry = to_irdma_mmap_entry(rdma_entry);
154 
155 	kfree(entry);
156 }
157 
158 static struct rdma_user_mmap_entry*
159 irdma_user_mmap_entry_insert(struct irdma_ucontext *ucontext, u64 bar_offset,
160 			     enum irdma_mmap_flag mmap_flag, u64 *mmap_offset)
161 {
162 	struct irdma_user_mmap_entry *entry = kzalloc_obj(*entry);
163 	int ret;
164 
165 	if (!entry)
166 		return NULL;
167 
168 	entry->bar_offset = bar_offset;
169 	entry->mmap_flag = mmap_flag;
170 
171 	ret = rdma_user_mmap_entry_insert(&ucontext->ibucontext,
172 					  &entry->rdma_entry, PAGE_SIZE);
173 	if (ret) {
174 		kfree(entry);
175 		return NULL;
176 	}
177 	*mmap_offset = rdma_user_mmap_get_offset(&entry->rdma_entry);
178 
179 	return &entry->rdma_entry;
180 }
181 
182 /**
183  * irdma_mmap - user memory map
184  * @context: context created during alloc
185  * @vma: kernel info for user memory map
186  */
187 static int irdma_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
188 {
189 	struct rdma_user_mmap_entry *rdma_entry;
190 	struct irdma_user_mmap_entry *entry;
191 	struct irdma_ucontext *ucontext;
192 	u64 pfn;
193 	int ret;
194 
195 	ucontext = to_ucontext(context);
196 
197 	/* Legacy support for libi40iw with hard-coded mmap key */
198 	if (ucontext->legacy_mode)
199 		return irdma_mmap_legacy(ucontext, vma);
200 
201 	rdma_entry = rdma_user_mmap_entry_get(&ucontext->ibucontext, vma);
202 	if (!rdma_entry) {
203 		ibdev_dbg(&ucontext->iwdev->ibdev,
204 			  "VERBS: pgoff[0x%lx] does not have valid entry\n",
205 			  vma->vm_pgoff);
206 		return -EINVAL;
207 	}
208 
209 	entry = to_irdma_mmap_entry(rdma_entry);
210 	ibdev_dbg(&ucontext->iwdev->ibdev,
211 		  "VERBS: bar_offset [0x%llx] mmap_flag [%d]\n",
212 		  entry->bar_offset, entry->mmap_flag);
213 
214 	pfn = (entry->bar_offset +
215 	      pci_resource_start(ucontext->iwdev->rf->pcidev, 0)) >> PAGE_SHIFT;
216 
217 	switch (entry->mmap_flag) {
218 	case IRDMA_MMAP_IO_NC:
219 		ret = rdma_user_mmap_io(context, vma, pfn, PAGE_SIZE,
220 					pgprot_noncached(vma->vm_page_prot),
221 					rdma_entry);
222 		break;
223 	case IRDMA_MMAP_IO_WC:
224 		ret = rdma_user_mmap_io(context, vma, pfn, PAGE_SIZE,
225 					pgprot_writecombine(vma->vm_page_prot),
226 					rdma_entry);
227 		break;
228 	default:
229 		ret = -EINVAL;
230 	}
231 
232 	if (ret)
233 		ibdev_dbg(&ucontext->iwdev->ibdev,
234 			  "VERBS: bar_offset [0x%llx] mmap_flag[%d] err[%d]\n",
235 			  entry->bar_offset, entry->mmap_flag, ret);
236 	rdma_user_mmap_entry_put(rdma_entry);
237 
238 	return ret;
239 }
240 
241 /**
242  * irdma_alloc_push_page - allocate a push page for qp
243  * @iwqp: qp pointer
244  */
245 static void irdma_alloc_push_page(struct irdma_qp *iwqp)
246 {
247 	struct irdma_cqp_request *cqp_request;
248 	struct cqp_cmds_info *cqp_info;
249 	struct irdma_device *iwdev = iwqp->iwdev;
250 	struct irdma_sc_qp *qp = &iwqp->sc_qp;
251 	int status;
252 
253 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
254 	if (!cqp_request)
255 		return;
256 
257 	cqp_info = &cqp_request->info;
258 	cqp_info->cqp_cmd = IRDMA_OP_MANAGE_PUSH_PAGE;
259 	cqp_info->post_sq = 1;
260 	cqp_info->in.u.manage_push_page.info.push_idx = 0;
261 	cqp_info->in.u.manage_push_page.info.qs_handle =
262 		qp->vsi->qos[qp->user_pri].qs_handle;
263 	cqp_info->in.u.manage_push_page.info.free_page = 0;
264 	cqp_info->in.u.manage_push_page.info.push_page_type = 0;
265 	cqp_info->in.u.manage_push_page.cqp = &iwdev->rf->cqp.sc_cqp;
266 	cqp_info->in.u.manage_push_page.scratch = (uintptr_t)cqp_request;
267 
268 	status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
269 	if (!status && cqp_request->compl_info.op_ret_val <
270 	    iwdev->rf->sc_dev.hw_attrs.max_hw_device_pages) {
271 		qp->push_idx = cqp_request->compl_info.op_ret_val;
272 		qp->push_offset = 0;
273 	}
274 
275 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
276 }
277 
278 /**
279  * irdma_alloc_ucontext - Allocate the user context data structure
280  * @uctx: uverbs context pointer
281  * @udata: user data
282  *
283  * This keeps track of all objects associated with a particular
284  * user-mode client.
285  */
286 static int irdma_alloc_ucontext(struct ib_ucontext *uctx,
287 				struct ib_udata *udata)
288 {
289 #define IRDMA_ALLOC_UCTX_MIN_RESP_LEN offsetofend(struct irdma_alloc_ucontext_resp, rsvd)
290 	struct ib_device *ibdev = uctx->device;
291 	struct irdma_device *iwdev = to_iwdev(ibdev);
292 	struct irdma_alloc_ucontext_req req = {};
293 	struct irdma_alloc_ucontext_resp uresp = {};
294 	struct irdma_ucontext *ucontext = to_ucontext(uctx);
295 	struct irdma_uk_attrs *uk_attrs = &iwdev->rf->sc_dev.hw_attrs.uk_attrs;
296 	int ret;
297 
298 	if (udata->outlen < IRDMA_ALLOC_UCTX_MIN_RESP_LEN)
299 		return -EINVAL;
300 
301 	ret = ib_copy_validate_udata_in_cm(udata, req, rsvd8,
302 					   IRDMA_ALLOC_UCTX_USE_RAW_ATTR |
303 						   IRDMA_SUPPORT_WQE_FORMAT_V2);
304 	if (ret)
305 		return ret;
306 
307 	if (req.userspace_ver < 4 || req.userspace_ver > IRDMA_ABI_VER)
308 		goto ver_error;
309 
310 	ucontext->iwdev = iwdev;
311 	ucontext->abi_ver = req.userspace_ver;
312 
313 	if (!(req.comp_mask & IRDMA_SUPPORT_WQE_FORMAT_V2) &&
314 	    uk_attrs->hw_rev >= IRDMA_GEN_3)
315 		return -EOPNOTSUPP;
316 
317 	if (req.comp_mask & IRDMA_ALLOC_UCTX_USE_RAW_ATTR)
318 		ucontext->use_raw_attrs = true;
319 
320 	/* GEN_1 legacy support with libi40iw */
321 	if (udata->outlen == IRDMA_ALLOC_UCTX_MIN_RESP_LEN) {
322 		if (uk_attrs->hw_rev != IRDMA_GEN_1)
323 			return -EOPNOTSUPP;
324 
325 		ucontext->legacy_mode = true;
326 		uresp.max_qps = iwdev->rf->max_qp;
327 		uresp.max_pds = iwdev->rf->sc_dev.hw_attrs.max_hw_pds;
328 		uresp.wq_size = iwdev->rf->sc_dev.hw_attrs.max_qp_wr * 2;
329 		uresp.kernel_ver = req.userspace_ver;
330 		ret = ib_respond_udata(udata, uresp);
331 		if (ret)
332 			return ret;
333 	} else {
334 		u64 bar_off = (uintptr_t)iwdev->rf->sc_dev.hw_regs[IRDMA_DB_ADDR_OFFSET];
335 
336 		ucontext->db_mmap_entry =
337 			irdma_user_mmap_entry_insert(ucontext, bar_off,
338 						     IRDMA_MMAP_IO_NC,
339 						     &uresp.db_mmap_key);
340 		if (!ucontext->db_mmap_entry)
341 			return -ENOMEM;
342 
343 		uresp.kernel_ver = IRDMA_ABI_VER;
344 		uresp.feature_flags = uk_attrs->feature_flags;
345 		uresp.max_hw_wq_frags = uk_attrs->max_hw_wq_frags;
346 		uresp.max_hw_read_sges = uk_attrs->max_hw_read_sges;
347 		uresp.max_hw_inline = uk_attrs->max_hw_inline;
348 		uresp.max_hw_rq_quanta = uk_attrs->max_hw_rq_quanta;
349 		uresp.max_hw_wq_quanta = uk_attrs->max_hw_wq_quanta;
350 		uresp.max_hw_sq_chunk = uk_attrs->max_hw_sq_chunk;
351 		uresp.max_hw_cq_size = uk_attrs->max_hw_cq_size;
352 		uresp.min_hw_cq_size = uk_attrs->min_hw_cq_size;
353 		uresp.hw_rev = uk_attrs->hw_rev;
354 		uresp.comp_mask |= IRDMA_ALLOC_UCTX_USE_RAW_ATTR;
355 		uresp.min_hw_wq_size = uk_attrs->min_hw_wq_size;
356 		uresp.comp_mask |= IRDMA_ALLOC_UCTX_MIN_HW_WQ_SIZE;
357 		uresp.max_hw_srq_quanta = uk_attrs->max_hw_srq_quanta;
358 		uresp.comp_mask |= IRDMA_ALLOC_UCTX_MAX_HW_SRQ_QUANTA;
359 		ret = ib_respond_udata(udata, uresp);
360 		if (ret) {
361 			rdma_user_mmap_entry_remove(ucontext->db_mmap_entry);
362 			return ret;
363 		}
364 	}
365 
366 	INIT_LIST_HEAD(&ucontext->cq_reg_mem_list);
367 	spin_lock_init(&ucontext->cq_reg_mem_list_lock);
368 	INIT_LIST_HEAD(&ucontext->qp_reg_mem_list);
369 	spin_lock_init(&ucontext->qp_reg_mem_list_lock);
370 	INIT_LIST_HEAD(&ucontext->srq_reg_mem_list);
371 	spin_lock_init(&ucontext->srq_reg_mem_list_lock);
372 
373 	return 0;
374 
375 ver_error:
376 	ibdev_err(&iwdev->ibdev,
377 		  "Invalid userspace driver version detected. Detected version %d, should be %d\n",
378 		  req.userspace_ver, IRDMA_ABI_VER);
379 	return -EINVAL;
380 }
381 
382 /**
383  * irdma_dealloc_ucontext - deallocate the user context data structure
384  * @context: user context created during alloc
385  */
386 static void irdma_dealloc_ucontext(struct ib_ucontext *context)
387 {
388 	struct irdma_ucontext *ucontext = to_ucontext(context);
389 
390 	rdma_user_mmap_entry_remove(ucontext->db_mmap_entry);
391 }
392 
393 /**
394  * irdma_alloc_pd - allocate protection domain
395  * @pd: PD pointer
396  * @udata: user data
397  */
398 static int irdma_alloc_pd(struct ib_pd *pd, struct ib_udata *udata)
399 {
400 #define IRDMA_ALLOC_PD_MIN_RESP_LEN offsetofend(struct irdma_alloc_pd_resp, rsvd)
401 	struct irdma_pd *iwpd = to_iwpd(pd);
402 	struct irdma_device *iwdev = to_iwdev(pd->device);
403 	struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
404 	struct irdma_pci_f *rf = iwdev->rf;
405 	struct irdma_alloc_pd_resp uresp = {};
406 	struct irdma_sc_pd *sc_pd;
407 	u32 pd_id = 0;
408 	int err;
409 
410 	if (udata && udata->outlen < IRDMA_ALLOC_PD_MIN_RESP_LEN)
411 		return -EINVAL;
412 
413 	err = irdma_alloc_rsrc(rf, rf->allocated_pds, rf->max_pd, &pd_id,
414 			       &rf->next_pd);
415 	if (err)
416 		return err;
417 
418 	sc_pd = &iwpd->sc_pd;
419 	if (udata) {
420 		struct irdma_ucontext *ucontext =
421 			rdma_udata_to_drv_context(udata, struct irdma_ucontext,
422 						  ibucontext);
423 		irdma_sc_pd_init(dev, sc_pd, pd_id, ucontext->abi_ver);
424 		uresp.pd_id = pd_id;
425 		err = ib_respond_udata(udata, uresp);
426 		if (err)
427 			goto error;
428 	} else {
429 		irdma_sc_pd_init(dev, sc_pd, pd_id, IRDMA_ABI_VER);
430 	}
431 
432 	return 0;
433 error:
434 	irdma_free_rsrc(rf, rf->allocated_pds, pd_id);
435 
436 	return err;
437 }
438 
439 /**
440  * irdma_dealloc_pd - deallocate pd
441  * @ibpd: ptr of pd to be deallocated
442  * @udata: user data
443  */
444 static int irdma_dealloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
445 {
446 	struct irdma_pd *iwpd = to_iwpd(ibpd);
447 	struct irdma_device *iwdev = to_iwdev(ibpd->device);
448 
449 	irdma_free_rsrc(iwdev->rf, iwdev->rf->allocated_pds, iwpd->sc_pd.pd_id);
450 
451 	return 0;
452 }
453 
454 /**
455  * irdma_get_pbl - Retrieve pbl from a list given a virtual
456  * address
457  * @va: user virtual address
458  * @pbl_list: pbl list to search in (QP's or CQ's)
459  */
460 static struct irdma_pbl *irdma_get_pbl(unsigned long va,
461 				       struct list_head *pbl_list)
462 {
463 	struct irdma_pbl *iwpbl;
464 
465 	list_for_each_entry (iwpbl, pbl_list, list) {
466 		if (iwpbl->user_base == va) {
467 			list_del(&iwpbl->list);
468 			iwpbl->on_list = false;
469 			return iwpbl;
470 		}
471 	}
472 
473 	return NULL;
474 }
475 
476 /**
477  * irdma_clean_cqes - clean cq entries for qp
478  * @iwqp: qp ptr (user or kernel)
479  * @iwcq: cq ptr
480  */
481 static void irdma_clean_cqes(struct irdma_qp *iwqp, struct irdma_cq *iwcq)
482 {
483 	struct irdma_cq_uk *ukcq = &iwcq->sc_cq.cq_uk;
484 	unsigned long flags;
485 
486 	spin_lock_irqsave(&iwcq->lock, flags);
487 	irdma_uk_clean_cq(&iwqp->sc_qp.qp_uk, ukcq);
488 	spin_unlock_irqrestore(&iwcq->lock, flags);
489 }
490 
491 static void irdma_remove_push_mmap_entries(struct irdma_qp *iwqp)
492 {
493 	if (iwqp->push_db_mmap_entry) {
494 		rdma_user_mmap_entry_remove(iwqp->push_db_mmap_entry);
495 		iwqp->push_db_mmap_entry = NULL;
496 	}
497 	if (iwqp->push_wqe_mmap_entry) {
498 		rdma_user_mmap_entry_remove(iwqp->push_wqe_mmap_entry);
499 		iwqp->push_wqe_mmap_entry = NULL;
500 	}
501 }
502 
503 static int irdma_setup_push_mmap_entries(struct irdma_ucontext *ucontext,
504 					 struct irdma_qp *iwqp,
505 					 u64 *push_wqe_mmap_key,
506 					 u64 *push_db_mmap_key)
507 {
508 	struct irdma_device *iwdev = ucontext->iwdev;
509 	u64 rsvd, bar_off;
510 
511 	rsvd = IRDMA_PF_BAR_RSVD;
512 	bar_off = (uintptr_t)iwdev->rf->sc_dev.hw_regs[IRDMA_DB_ADDR_OFFSET];
513 	/* skip over db page */
514 	bar_off += IRDMA_HW_PAGE_SIZE;
515 	/* push wqe page */
516 	bar_off += rsvd + iwqp->sc_qp.push_idx * IRDMA_HW_PAGE_SIZE;
517 	iwqp->push_wqe_mmap_entry = irdma_user_mmap_entry_insert(ucontext,
518 					bar_off, IRDMA_MMAP_IO_WC,
519 					push_wqe_mmap_key);
520 	if (!iwqp->push_wqe_mmap_entry)
521 		return -ENOMEM;
522 
523 	/* push doorbell page */
524 	bar_off += IRDMA_HW_PAGE_SIZE;
525 	iwqp->push_db_mmap_entry = irdma_user_mmap_entry_insert(ucontext,
526 					bar_off, IRDMA_MMAP_IO_NC,
527 					push_db_mmap_key);
528 	if (!iwqp->push_db_mmap_entry) {
529 		rdma_user_mmap_entry_remove(iwqp->push_wqe_mmap_entry);
530 		return -ENOMEM;
531 	}
532 
533 	return 0;
534 }
535 
536 /**
537  * irdma_destroy_qp - destroy qp
538  * @ibqp: qp's ib pointer also to get to device's qp address
539  * @udata: user data
540  */
541 static int irdma_destroy_qp(struct ib_qp *ibqp, struct ib_udata *udata)
542 {
543 	struct irdma_qp *iwqp = to_iwqp(ibqp);
544 	struct irdma_device *iwdev = iwqp->iwdev;
545 
546 	iwqp->sc_qp.qp_uk.destroy_pending = true;
547 
548 	if (iwqp->iwarp_state >= IRDMA_QP_STATE_IDLE)
549 		irdma_modify_qp_to_err(&iwqp->sc_qp);
550 
551 	if (!iwqp->user_mode)
552 		cancel_delayed_work_sync(&iwqp->dwork_flush);
553 
554 	if (!iwqp->user_mode) {
555 		if (iwqp->iwscq) {
556 			irdma_clean_cqes(iwqp, iwqp->iwscq);
557 			if (iwqp->iwrcq != iwqp->iwscq)
558 				irdma_clean_cqes(iwqp, iwqp->iwrcq);
559 		}
560 	}
561 
562 	irdma_qp_rem_ref(&iwqp->ibqp);
563 	if (!iwdev->rf->reset)
564 		wait_for_completion(&iwqp->free_qp);
565 	irdma_free_lsmm_rsrc(iwqp);
566 	irdma_cqp_qp_destroy_cmd(&iwdev->rf->sc_dev, &iwqp->sc_qp);
567 
568 	irdma_remove_push_mmap_entries(iwqp);
569 
570 	if (iwqp->sc_qp.qp_uk.qp_id == 1)
571 		iwdev->rf->hwqp1_rsvd = false;
572 	irdma_free_qp_rsrc(iwqp);
573 
574 	return 0;
575 }
576 
577 /**
578  * irdma_setup_virt_qp - setup for allocation of virtual qp
579  * @iwdev: irdma device
580  * @iwqp: qp ptr
581  * @init_info: initialize info to return
582  */
583 static void irdma_setup_virt_qp(struct irdma_device *iwdev,
584 			       struct irdma_qp *iwqp,
585 			       struct irdma_qp_init_info *init_info)
586 {
587 	struct irdma_pbl *iwpbl = iwqp->iwpbl;
588 	struct irdma_qp_mr *qpmr = &iwpbl->qp_mr;
589 
590 	iwqp->page = qpmr->sq_page;
591 	init_info->shadow_area_pa = qpmr->shadow;
592 	if (iwpbl->pbl_allocated) {
593 		init_info->virtual_map = true;
594 		init_info->sq_pa = qpmr->sq_pbl.idx;
595 		/* Need to use contiguous buffer for RQ of QP
596 		 * in case it is associated with SRQ.
597 		 */
598 		init_info->rq_pa = init_info->qp_uk_init_info.srq_uk ?
599 			qpmr->rq_pa : qpmr->rq_pbl.idx;
600 	} else {
601 		init_info->sq_pa = qpmr->sq_pbl.addr;
602 		init_info->rq_pa = qpmr->rq_pbl.addr;
603 	}
604 }
605 
606 /**
607  * irdma_setup_umode_qp - setup sq and rq size in user mode qp
608  * @udata: udata
609  * @iwdev: iwarp device
610  * @iwqp: qp ptr (user or kernel)
611  * @info: initialize info to return
612  * @init_attr: Initial QP create attributes
613  */
614 static int irdma_setup_umode_qp(struct ib_udata *udata,
615 				struct irdma_device *iwdev,
616 				struct irdma_qp *iwqp,
617 				struct irdma_qp_init_info *info,
618 				struct ib_qp_init_attr *init_attr)
619 {
620 	struct irdma_ucontext *ucontext = rdma_udata_to_drv_context(udata,
621 				struct irdma_ucontext, ibucontext);
622 	struct irdma_qp_uk_init_info *ukinfo = &info->qp_uk_init_info;
623 	struct irdma_create_qp_req req;
624 	unsigned long flags;
625 	int ret;
626 
627 	ret = ib_copy_from_udata(&req, udata,
628 				 min(sizeof(req), udata->inlen));
629 	if (ret) {
630 		ibdev_dbg(&iwdev->ibdev, "VERBS: ib_copy_from_data fail\n");
631 		return ret;
632 	}
633 
634 	iwqp->ctx_info.qp_compl_ctx = req.user_compl_ctx;
635 	iwqp->user_mode = 1;
636 	if (req.user_wqe_bufs) {
637 		spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags);
638 		iwqp->iwpbl = irdma_get_pbl((unsigned long)req.user_wqe_bufs,
639 					    &ucontext->qp_reg_mem_list);
640 		spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags);
641 
642 		if (!iwqp->iwpbl) {
643 			ret = -ENODATA;
644 			ibdev_dbg(&iwdev->ibdev, "VERBS: no pbl info\n");
645 			return ret;
646 		}
647 	}
648 
649 	if (!ucontext->use_raw_attrs) {
650 		/**
651 		 * Maintain backward compat with older ABI which passes sq and
652 		 * rq depth in quanta in cap.max_send_wr and cap.max_recv_wr.
653 		 * There is no way to compute the correct value of
654 		 * iwqp->max_send_wr/max_recv_wr in the kernel.
655 		 */
656 		iwqp->max_send_wr = init_attr->cap.max_send_wr;
657 		iwqp->max_recv_wr = init_attr->cap.max_recv_wr;
658 		ukinfo->sq_size = init_attr->cap.max_send_wr;
659 		ukinfo->rq_size = init_attr->cap.max_recv_wr;
660 		irdma_uk_calc_shift_wq(ukinfo, &ukinfo->sq_shift,
661 				       &ukinfo->rq_shift);
662 	} else {
663 		ret = irdma_uk_calc_depth_shift_sq(ukinfo, &ukinfo->sq_depth,
664 						   &ukinfo->sq_shift);
665 		if (ret)
666 			return ret;
667 
668 		ret = irdma_uk_calc_depth_shift_rq(ukinfo, &ukinfo->rq_depth,
669 						   &ukinfo->rq_shift);
670 		if (ret)
671 			return ret;
672 
673 		iwqp->max_send_wr =
674 			(ukinfo->sq_depth - IRDMA_SQ_RSVD) >> ukinfo->sq_shift;
675 		iwqp->max_recv_wr =
676 			(ukinfo->rq_depth - IRDMA_RQ_RSVD) >> ukinfo->rq_shift;
677 		ukinfo->sq_size = ukinfo->sq_depth >> ukinfo->sq_shift;
678 		ukinfo->rq_size = ukinfo->rq_depth >> ukinfo->rq_shift;
679 	}
680 
681 	irdma_setup_virt_qp(iwdev, iwqp, info);
682 
683 	return 0;
684 }
685 
686 /**
687  * irdma_setup_kmode_qp - setup initialization for kernel mode qp
688  * @iwdev: iwarp device
689  * @iwqp: qp ptr (user or kernel)
690  * @info: initialize info to return
691  * @init_attr: Initial QP create attributes
692  */
693 static int irdma_setup_kmode_qp(struct irdma_device *iwdev,
694 				struct irdma_qp *iwqp,
695 				struct irdma_qp_init_info *info,
696 				struct ib_qp_init_attr *init_attr)
697 {
698 	struct irdma_dma_mem *mem = &iwqp->kqp.dma_mem;
699 	u32 size;
700 	int status;
701 	struct irdma_qp_uk_init_info *ukinfo = &info->qp_uk_init_info;
702 
703 	status = irdma_uk_calc_depth_shift_sq(ukinfo, &ukinfo->sq_depth,
704 					      &ukinfo->sq_shift);
705 	if (status)
706 		return status;
707 
708 	status = irdma_uk_calc_depth_shift_rq(ukinfo, &ukinfo->rq_depth,
709 					      &ukinfo->rq_shift);
710 	if (status)
711 		return status;
712 
713 	iwqp->kqp.sq_wrid_mem =
714 		kzalloc_objs(*iwqp->kqp.sq_wrid_mem, ukinfo->sq_depth);
715 	if (!iwqp->kqp.sq_wrid_mem)
716 		return -ENOMEM;
717 
718 	iwqp->kqp.rq_wrid_mem =
719 		kzalloc_objs(*iwqp->kqp.rq_wrid_mem, ukinfo->rq_depth);
720 
721 	if (!iwqp->kqp.rq_wrid_mem) {
722 		kfree(iwqp->kqp.sq_wrid_mem);
723 		iwqp->kqp.sq_wrid_mem = NULL;
724 		return -ENOMEM;
725 	}
726 
727 	ukinfo->sq_wrtrk_array = iwqp->kqp.sq_wrid_mem;
728 	ukinfo->rq_wrid_array = iwqp->kqp.rq_wrid_mem;
729 
730 	size = (ukinfo->sq_depth + ukinfo->rq_depth) * IRDMA_QP_WQE_MIN_SIZE;
731 	size += (IRDMA_SHADOW_AREA_SIZE << 3);
732 
733 	mem->size = ALIGN(size, 256);
734 	mem->va = dma_alloc_coherent(iwdev->rf->hw.device, mem->size,
735 				     &mem->pa, GFP_KERNEL);
736 	if (!mem->va) {
737 		kfree(iwqp->kqp.sq_wrid_mem);
738 		iwqp->kqp.sq_wrid_mem = NULL;
739 		kfree(iwqp->kqp.rq_wrid_mem);
740 		iwqp->kqp.rq_wrid_mem = NULL;
741 		return -ENOMEM;
742 	}
743 
744 	ukinfo->sq = mem->va;
745 	info->sq_pa = mem->pa;
746 	ukinfo->rq = &ukinfo->sq[ukinfo->sq_depth];
747 	info->rq_pa = info->sq_pa + (ukinfo->sq_depth * IRDMA_QP_WQE_MIN_SIZE);
748 	ukinfo->shadow_area = ukinfo->rq[ukinfo->rq_depth].elem;
749 	info->shadow_area_pa =
750 		info->rq_pa + (ukinfo->rq_depth * IRDMA_QP_WQE_MIN_SIZE);
751 	ukinfo->sq_size = ukinfo->sq_depth >> ukinfo->sq_shift;
752 	ukinfo->rq_size = ukinfo->rq_depth >> ukinfo->rq_shift;
753 	ukinfo->qp_id = info->qp_uk_init_info.qp_id;
754 
755 	iwqp->max_send_wr = (ukinfo->sq_depth - IRDMA_SQ_RSVD) >> ukinfo->sq_shift;
756 	iwqp->max_recv_wr = (ukinfo->rq_depth - IRDMA_RQ_RSVD) >> ukinfo->rq_shift;
757 	init_attr->cap.max_send_wr = iwqp->max_send_wr;
758 	init_attr->cap.max_recv_wr = iwqp->max_recv_wr;
759 
760 	return 0;
761 }
762 
763 static int irdma_cqp_create_qp_cmd(struct irdma_qp *iwqp)
764 {
765 	struct irdma_pci_f *rf = iwqp->iwdev->rf;
766 	struct irdma_cqp_request *cqp_request;
767 	struct cqp_cmds_info *cqp_info;
768 	struct irdma_create_qp_info *qp_info;
769 	int status;
770 
771 	cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
772 	if (!cqp_request)
773 		return -ENOMEM;
774 
775 	cqp_info = &cqp_request->info;
776 	qp_info = &cqp_request->info.in.u.qp_create.info;
777 	qp_info->mac_valid = true;
778 	qp_info->cq_num_valid = true;
779 	qp_info->next_iwarp_state = IRDMA_QP_STATE_IDLE;
780 
781 	cqp_info->cqp_cmd = IRDMA_OP_QP_CREATE;
782 	cqp_info->post_sq = 1;
783 	cqp_info->in.u.qp_create.qp = &iwqp->sc_qp;
784 	cqp_info->in.u.qp_create.scratch = (uintptr_t)cqp_request;
785 	status = irdma_handle_cqp_op(rf, cqp_request);
786 	irdma_put_cqp_request(&rf->cqp, cqp_request);
787 
788 	return status;
789 }
790 
791 static void irdma_roce_fill_and_set_qpctx_info(struct irdma_qp *iwqp,
792 					       struct irdma_qp_host_ctx_info *ctx_info)
793 {
794 	struct irdma_device *iwdev = iwqp->iwdev;
795 	struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
796 	struct irdma_roce_offload_info *roce_info;
797 	struct irdma_udp_offload_info *udp_info;
798 
799 	udp_info = &iwqp->udp_info;
800 	udp_info->snd_mss = ib_mtu_enum_to_int(ib_mtu_int_to_enum(iwdev->vsi.mtu));
801 	udp_info->cwnd = iwdev->roce_cwnd;
802 	udp_info->rexmit_thresh = 2;
803 	udp_info->rnr_nak_thresh = 2;
804 	udp_info->src_port = 0xc000;
805 	udp_info->dst_port = ROCE_V2_UDP_DPORT;
806 	roce_info = &iwqp->roce_info;
807 	ether_addr_copy(roce_info->mac_addr, iwdev->netdev->dev_addr);
808 
809 	if (iwqp->ibqp.qp_type == IB_QPT_GSI && iwqp->ibqp.qp_num != 1)
810 		roce_info->is_qp1 = true;
811 	roce_info->rd_en = true;
812 	roce_info->wr_rdresp_en = true;
813 	if (dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_3)
814 		roce_info->bind_en = true;
815 	roce_info->dcqcn_en = false;
816 	roce_info->rtomin = 5;
817 
818 	roce_info->ack_credits = iwdev->roce_ackcreds;
819 	roce_info->ird_size = dev->hw_attrs.max_hw_ird;
820 	roce_info->ord_size = dev->hw_attrs.max_hw_ord;
821 
822 	if (!iwqp->user_mode) {
823 		roce_info->priv_mode_en = true;
824 		roce_info->fast_reg_en = true;
825 		roce_info->udprivcq_en = true;
826 	}
827 	roce_info->roce_tver = 0;
828 
829 	ctx_info->roce_info = &iwqp->roce_info;
830 	ctx_info->udp_info = &iwqp->udp_info;
831 	irdma_sc_qp_setctx_roce(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
832 }
833 
834 static void irdma_iw_fill_and_set_qpctx_info(struct irdma_qp *iwqp,
835 					     struct irdma_qp_host_ctx_info *ctx_info)
836 {
837 	struct irdma_device *iwdev = iwqp->iwdev;
838 	struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
839 	struct irdma_iwarp_offload_info *iwarp_info;
840 
841 	iwarp_info = &iwqp->iwarp_info;
842 	ether_addr_copy(iwarp_info->mac_addr, iwdev->netdev->dev_addr);
843 	iwarp_info->rd_en = true;
844 	iwarp_info->wr_rdresp_en = true;
845 	iwarp_info->ecn_en = true;
846 	iwarp_info->rtomin = 5;
847 
848 	if (dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
849 		iwarp_info->ib_rd_en = true;
850 	if (!iwqp->user_mode) {
851 		iwarp_info->priv_mode_en = true;
852 		iwarp_info->fast_reg_en = true;
853 	}
854 	iwarp_info->ddp_ver = 1;
855 	iwarp_info->rdmap_ver = 1;
856 
857 	ctx_info->iwarp_info = &iwqp->iwarp_info;
858 	ctx_info->iwarp_info_valid = true;
859 	irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
860 	ctx_info->iwarp_info_valid = false;
861 }
862 
863 static int irdma_validate_qp_attrs(struct ib_qp_init_attr *init_attr,
864 				   struct irdma_device *iwdev)
865 {
866 	struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
867 	struct irdma_uk_attrs *uk_attrs = &dev->hw_attrs.uk_attrs;
868 
869 	if (init_attr->create_flags)
870 		return -EOPNOTSUPP;
871 
872 	if (init_attr->cap.max_inline_data > uk_attrs->max_hw_inline ||
873 	    init_attr->cap.max_send_sge > uk_attrs->max_hw_wq_frags ||
874 	    init_attr->cap.max_recv_sge > uk_attrs->max_hw_wq_frags ||
875 	    init_attr->cap.max_send_wr > uk_attrs->max_hw_wq_quanta ||
876 	    init_attr->cap.max_recv_wr > uk_attrs->max_hw_rq_quanta)
877 		return -EINVAL;
878 
879 	if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
880 		if (init_attr->qp_type != IB_QPT_RC &&
881 		    init_attr->qp_type != IB_QPT_UD &&
882 		    init_attr->qp_type != IB_QPT_GSI)
883 			return -EOPNOTSUPP;
884 	} else {
885 		if (init_attr->qp_type != IB_QPT_RC)
886 			return -EOPNOTSUPP;
887 	}
888 
889 	return 0;
890 }
891 
892 static void irdma_flush_worker(struct work_struct *work)
893 {
894 	struct delayed_work *dwork = to_delayed_work(work);
895 	struct irdma_qp *iwqp = container_of(dwork, struct irdma_qp, dwork_flush);
896 
897 	irdma_generate_flush_completions(iwqp);
898 }
899 
900 static int irdma_setup_gsi_qp_rsrc(struct irdma_qp *iwqp, u32 *qp_num)
901 {
902 	struct irdma_device *iwdev = iwqp->iwdev;
903 	struct irdma_pci_f *rf = iwdev->rf;
904 	unsigned long flags;
905 	int ret;
906 
907 	if (rf->rdma_ver <= IRDMA_GEN_2) {
908 		*qp_num = 1;
909 		return 0;
910 	}
911 
912 	spin_lock_irqsave(&rf->rsrc_lock, flags);
913 	if (!rf->hwqp1_rsvd) {
914 		*qp_num = 1;
915 		rf->hwqp1_rsvd = true;
916 		spin_unlock_irqrestore(&rf->rsrc_lock, flags);
917 	} else {
918 		spin_unlock_irqrestore(&rf->rsrc_lock, flags);
919 		ret = irdma_alloc_rsrc(rf, rf->allocated_qps, rf->max_qp,
920 				       qp_num, &rf->next_qp);
921 		if (ret)
922 			return ret;
923 	}
924 
925 	ret = irdma_vchnl_req_add_vport(&rf->sc_dev, iwdev->vport_id, *qp_num,
926 					(&iwdev->vsi)->qos);
927 	if (ret) {
928 		if (*qp_num != 1) {
929 			irdma_free_rsrc(rf, rf->allocated_qps, *qp_num);
930 		} else {
931 			spin_lock_irqsave(&rf->rsrc_lock, flags);
932 			rf->hwqp1_rsvd = false;
933 			spin_unlock_irqrestore(&rf->rsrc_lock, flags);
934 		}
935 		return ret;
936 	}
937 
938 	return 0;
939 }
940 
941 /**
942  * irdma_create_qp - create qp
943  * @ibqp: ptr of qp
944  * @init_attr: attributes for qp
945  * @udata: user data for create qp
946  */
947 static int irdma_create_qp(struct ib_qp *ibqp,
948 			   struct ib_qp_init_attr *init_attr,
949 			   struct ib_udata *udata)
950 {
951 #define IRDMA_CREATE_QP_MIN_REQ_LEN offsetofend(struct irdma_create_qp_req, user_compl_ctx)
952 #define IRDMA_CREATE_QP_MIN_RESP_LEN offsetofend(struct irdma_create_qp_resp, rsvd)
953 	struct ib_pd *ibpd = ibqp->pd;
954 	struct irdma_pd *iwpd = to_iwpd(ibpd);
955 	struct irdma_device *iwdev = to_iwdev(ibpd->device);
956 	struct irdma_pci_f *rf = iwdev->rf;
957 	struct irdma_qp *iwqp = to_iwqp(ibqp);
958 	struct irdma_create_qp_resp uresp = {};
959 	u32 qp_num = 0;
960 	int err_code;
961 	struct irdma_sc_qp *qp;
962 	struct irdma_sc_dev *dev = &rf->sc_dev;
963 	struct irdma_uk_attrs *uk_attrs = &dev->hw_attrs.uk_attrs;
964 	struct irdma_qp_init_info init_info = {};
965 	struct irdma_qp_host_ctx_info *ctx_info;
966 	struct irdma_srq *iwsrq;
967 	bool srq_valid = false;
968 	u32 srq_id = 0;
969 
970 	if (init_attr->srq) {
971 		iwsrq = to_iwsrq(init_attr->srq);
972 		srq_valid = true;
973 		srq_id = iwsrq->srq_num;
974 		init_attr->cap.max_recv_sge = uk_attrs->max_hw_wq_frags;
975 		init_attr->cap.max_recv_wr = 4;
976 		init_info.qp_uk_init_info.srq_uk = &iwsrq->sc_srq.srq_uk;
977 	}
978 
979 	err_code = irdma_validate_qp_attrs(init_attr, iwdev);
980 	if (err_code)
981 		return err_code;
982 
983 	if (udata && (udata->inlen < IRDMA_CREATE_QP_MIN_REQ_LEN ||
984 		      udata->outlen < IRDMA_CREATE_QP_MIN_RESP_LEN))
985 		return -EINVAL;
986 
987 	init_info.vsi = &iwdev->vsi;
988 	init_info.qp_uk_init_info.uk_attrs = uk_attrs;
989 	init_info.qp_uk_init_info.sq_size = init_attr->cap.max_send_wr;
990 	init_info.qp_uk_init_info.rq_size = init_attr->cap.max_recv_wr;
991 	init_info.qp_uk_init_info.max_sq_frag_cnt = init_attr->cap.max_send_sge;
992 	init_info.qp_uk_init_info.max_rq_frag_cnt = init_attr->cap.max_recv_sge;
993 	init_info.qp_uk_init_info.max_inline_data = init_attr->cap.max_inline_data;
994 
995 	qp = &iwqp->sc_qp;
996 	qp->qp_uk.back_qp = iwqp;
997 	qp->push_idx = IRDMA_INVALID_PUSH_PAGE_INDEX;
998 
999 	iwqp->iwdev = iwdev;
1000 	iwqp->q2_ctx_mem.size = ALIGN(IRDMA_Q2_BUF_SIZE + IRDMA_QP_CTX_SIZE,
1001 				      256);
1002 	iwqp->q2_ctx_mem.va = dma_alloc_coherent(dev->hw->device,
1003 						 iwqp->q2_ctx_mem.size,
1004 						 &iwqp->q2_ctx_mem.pa,
1005 						 GFP_KERNEL);
1006 	if (!iwqp->q2_ctx_mem.va)
1007 		return -ENOMEM;
1008 
1009 	init_info.q2 = iwqp->q2_ctx_mem.va;
1010 	init_info.q2_pa = iwqp->q2_ctx_mem.pa;
1011 	init_info.host_ctx = (__le64 *)(init_info.q2 + IRDMA_Q2_BUF_SIZE);
1012 	init_info.host_ctx_pa = init_info.q2_pa + IRDMA_Q2_BUF_SIZE;
1013 
1014 	if (init_attr->qp_type == IB_QPT_GSI) {
1015 		err_code = irdma_setup_gsi_qp_rsrc(iwqp, &qp_num);
1016 		if (err_code)
1017 			goto error;
1018 		iwqp->ibqp.qp_num = 1;
1019 	} else {
1020 		err_code = irdma_alloc_rsrc(rf, rf->allocated_qps, rf->max_qp,
1021 					    &qp_num, &rf->next_qp);
1022 		if (err_code)
1023 			goto error;
1024 		iwqp->ibqp.qp_num = qp_num;
1025 	}
1026 
1027 	iwqp->iwpd = iwpd;
1028 	qp = &iwqp->sc_qp;
1029 	iwqp->iwscq = to_iwcq(init_attr->send_cq);
1030 	iwqp->iwrcq = to_iwcq(init_attr->recv_cq);
1031 	iwqp->host_ctx.va = init_info.host_ctx;
1032 	iwqp->host_ctx.pa = init_info.host_ctx_pa;
1033 	iwqp->host_ctx.size = IRDMA_QP_CTX_SIZE;
1034 
1035 	init_info.pd = &iwpd->sc_pd;
1036 	init_info.qp_uk_init_info.qp_id = qp_num;
1037 	if (!rdma_protocol_roce(&iwdev->ibdev, 1))
1038 		init_info.qp_uk_init_info.first_sq_wq = 1;
1039 	iwqp->ctx_info.qp_compl_ctx = (uintptr_t)qp;
1040 	init_waitqueue_head(&iwqp->waitq);
1041 	init_waitqueue_head(&iwqp->mod_qp_waitq);
1042 
1043 	if (udata) {
1044 		init_info.qp_uk_init_info.abi_ver = iwpd->sc_pd.abi_ver;
1045 		err_code = irdma_setup_umode_qp(udata, iwdev, iwqp, &init_info,
1046 						init_attr);
1047 	} else {
1048 		INIT_DELAYED_WORK(&iwqp->dwork_flush, irdma_flush_worker);
1049 		init_info.qp_uk_init_info.abi_ver = IRDMA_ABI_VER;
1050 		err_code = irdma_setup_kmode_qp(iwdev, iwqp, &init_info, init_attr);
1051 	}
1052 
1053 	if (err_code) {
1054 		ibdev_dbg(&iwdev->ibdev, "VERBS: setup qp failed\n");
1055 		goto error;
1056 	}
1057 
1058 	if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
1059 		if (init_attr->qp_type == IB_QPT_RC) {
1060 			init_info.qp_uk_init_info.type = IRDMA_QP_TYPE_ROCE_RC;
1061 			init_info.qp_uk_init_info.qp_caps = IRDMA_SEND_WITH_IMM |
1062 							    IRDMA_WRITE_WITH_IMM |
1063 							    IRDMA_ROCE;
1064 		} else {
1065 			init_info.qp_uk_init_info.type = IRDMA_QP_TYPE_ROCE_UD;
1066 			init_info.qp_uk_init_info.qp_caps = IRDMA_SEND_WITH_IMM |
1067 							    IRDMA_ROCE;
1068 		}
1069 	} else {
1070 		init_info.qp_uk_init_info.type = IRDMA_QP_TYPE_IWARP;
1071 		init_info.qp_uk_init_info.qp_caps = IRDMA_WRITE_WITH_IMM;
1072 	}
1073 
1074 	if (dev->hw_attrs.uk_attrs.hw_rev > IRDMA_GEN_1)
1075 		init_info.qp_uk_init_info.qp_caps |= IRDMA_PUSH_MODE;
1076 
1077 	err_code = irdma_sc_qp_init(qp, &init_info);
1078 	if (err_code) {
1079 		ibdev_dbg(&iwdev->ibdev, "VERBS: qp_init fail\n");
1080 		goto error;
1081 	}
1082 
1083 	ctx_info = &iwqp->ctx_info;
1084 	ctx_info->srq_valid = srq_valid;
1085 	ctx_info->srq_id = srq_id;
1086 	ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
1087 	ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
1088 
1089 	if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
1090 		if (dev->ws_add(&iwdev->vsi, 0)) {
1091 			irdma_cqp_qp_destroy_cmd(&rf->sc_dev, &iwqp->sc_qp);
1092 			err_code = -EINVAL;
1093 			goto error;
1094 		}
1095 		irdma_qp_add_qos(&iwqp->sc_qp);
1096 		irdma_roce_fill_and_set_qpctx_info(iwqp, ctx_info);
1097 	} else {
1098 		irdma_iw_fill_and_set_qpctx_info(iwqp, ctx_info);
1099 	}
1100 
1101 	err_code = irdma_cqp_create_qp_cmd(iwqp);
1102 	if (err_code)
1103 		goto error;
1104 
1105 	refcount_set(&iwqp->refcnt, 1);
1106 	spin_lock_init(&iwqp->lock);
1107 	spin_lock_init(&iwqp->sc_qp.pfpdu.lock);
1108 	iwqp->sig_all = init_attr->sq_sig_type == IB_SIGNAL_ALL_WR;
1109 	rf->qp_table[qp_num] = iwqp;
1110 	init_completion(&iwqp->free_qp);
1111 
1112 	if (udata) {
1113 		/* GEN_1 legacy support with libi40iw does not have expanded uresp struct */
1114 		if (udata->outlen < sizeof(uresp)) {
1115 			uresp.lsmm = 1;
1116 			uresp.push_idx = IRDMA_INVALID_PUSH_PAGE_INDEX_GEN_1;
1117 		} else {
1118 			if (rdma_protocol_iwarp(&iwdev->ibdev, 1))
1119 				uresp.lsmm = 1;
1120 		}
1121 		uresp.actual_sq_size = init_info.qp_uk_init_info.sq_size;
1122 		uresp.actual_rq_size = init_info.qp_uk_init_info.rq_size;
1123 		uresp.qp_id = qp_num;
1124 		uresp.qp_caps = qp->qp_uk.qp_caps;
1125 
1126 		err_code = ib_respond_udata(udata, uresp);
1127 		if (err_code) {
1128 			irdma_destroy_qp(&iwqp->ibqp, udata);
1129 			return err_code;
1130 		}
1131 	}
1132 
1133 	return 0;
1134 
1135 error:
1136 	irdma_free_qp_rsrc(iwqp);
1137 	return err_code;
1138 }
1139 
1140 static int irdma_get_ib_acc_flags(struct irdma_qp *iwqp)
1141 {
1142 	int acc_flags = 0;
1143 
1144 	if (rdma_protocol_roce(iwqp->ibqp.device, 1)) {
1145 		if (iwqp->roce_info.wr_rdresp_en) {
1146 			acc_flags |= IB_ACCESS_LOCAL_WRITE;
1147 			acc_flags |= IB_ACCESS_REMOTE_WRITE;
1148 		}
1149 		if (iwqp->roce_info.rd_en)
1150 			acc_flags |= IB_ACCESS_REMOTE_READ;
1151 		if (iwqp->roce_info.bind_en)
1152 			acc_flags |= IB_ACCESS_MW_BIND;
1153 		if (iwqp->ctx_info.remote_atomics_en)
1154 			acc_flags |= IB_ACCESS_REMOTE_ATOMIC;
1155 	} else {
1156 		if (iwqp->iwarp_info.wr_rdresp_en) {
1157 			acc_flags |= IB_ACCESS_LOCAL_WRITE;
1158 			acc_flags |= IB_ACCESS_REMOTE_WRITE;
1159 		}
1160 		if (iwqp->iwarp_info.rd_en)
1161 			acc_flags |= IB_ACCESS_REMOTE_READ;
1162 		if (iwqp->ctx_info.remote_atomics_en)
1163 			acc_flags |= IB_ACCESS_REMOTE_ATOMIC;
1164 	}
1165 	return acc_flags;
1166 }
1167 
1168 /**
1169  * irdma_query_qp - query qp attributes
1170  * @ibqp: qp pointer
1171  * @attr: attributes pointer
1172  * @attr_mask: Not used
1173  * @init_attr: qp attributes to return
1174  */
1175 static int irdma_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1176 			  int attr_mask, struct ib_qp_init_attr *init_attr)
1177 {
1178 	struct irdma_qp *iwqp = to_iwqp(ibqp);
1179 	struct irdma_sc_qp *qp = &iwqp->sc_qp;
1180 
1181 	memset(attr, 0, sizeof(*attr));
1182 	memset(init_attr, 0, sizeof(*init_attr));
1183 
1184 	attr->qp_state = iwqp->ibqp_state;
1185 	attr->cur_qp_state = iwqp->ibqp_state;
1186 	attr->cap.max_send_wr = iwqp->max_send_wr;
1187 	attr->cap.max_recv_wr = iwqp->max_recv_wr;
1188 	attr->cap.max_inline_data = qp->qp_uk.max_inline_data;
1189 	attr->cap.max_send_sge = qp->qp_uk.max_sq_frag_cnt;
1190 	attr->cap.max_recv_sge = qp->qp_uk.max_rq_frag_cnt;
1191 	attr->qp_access_flags = irdma_get_ib_acc_flags(iwqp);
1192 	attr->port_num = 1;
1193 	if (rdma_protocol_roce(ibqp->device, 1)) {
1194 		attr->path_mtu = ib_mtu_int_to_enum(iwqp->udp_info.snd_mss);
1195 		attr->qkey = iwqp->roce_info.qkey;
1196 		attr->rq_psn = iwqp->udp_info.epsn;
1197 		attr->sq_psn = iwqp->udp_info.psn_nxt;
1198 		attr->dest_qp_num = iwqp->roce_info.dest_qp;
1199 		attr->pkey_index = iwqp->roce_info.p_key;
1200 		attr->retry_cnt = iwqp->udp_info.rexmit_thresh;
1201 		attr->rnr_retry = iwqp->udp_info.rnr_nak_thresh;
1202 		attr->min_rnr_timer = iwqp->udp_info.min_rnr_timer;
1203 		attr->max_rd_atomic = iwqp->roce_info.ord_size;
1204 		attr->max_dest_rd_atomic = iwqp->roce_info.ird_size;
1205 	}
1206 
1207 	init_attr->event_handler = iwqp->ibqp.event_handler;
1208 	init_attr->qp_context = iwqp->ibqp.qp_context;
1209 	init_attr->send_cq = iwqp->ibqp.send_cq;
1210 	init_attr->recv_cq = iwqp->ibqp.recv_cq;
1211 	init_attr->srq = iwqp->ibqp.srq;
1212 	init_attr->cap = attr->cap;
1213 
1214 	return 0;
1215 }
1216 
1217 /**
1218  * irdma_query_pkey - Query partition key
1219  * @ibdev: device pointer from stack
1220  * @port: port number
1221  * @index: index of pkey
1222  * @pkey: pointer to store the pkey
1223  */
1224 static int irdma_query_pkey(struct ib_device *ibdev, u32 port, u16 index,
1225 			    u16 *pkey)
1226 {
1227 	if (index >= IRDMA_PKEY_TBL_SZ)
1228 		return -EINVAL;
1229 
1230 	*pkey = IRDMA_DEFAULT_PKEY;
1231 	return 0;
1232 }
1233 
1234 static u8 irdma_roce_get_vlan_prio(const struct ib_gid_attr *attr, u8 prio)
1235 {
1236 	struct net_device *ndev;
1237 
1238 	rcu_read_lock();
1239 	ndev = rcu_dereference(attr->ndev);
1240 	if (!ndev)
1241 		goto exit;
1242 	if (is_vlan_dev(ndev)) {
1243 		u16 vlan_qos = vlan_dev_get_egress_qos_mask(ndev, prio);
1244 
1245 		prio = (vlan_qos & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
1246 	}
1247 exit:
1248 	rcu_read_unlock();
1249 	return prio;
1250 }
1251 
1252 static int irdma_wait_for_suspend(struct irdma_qp *iwqp)
1253 {
1254 	if (!wait_event_timeout(iwqp->iwdev->suspend_wq,
1255 				!iwqp->suspend_pending,
1256 				msecs_to_jiffies(IRDMA_EVENT_TIMEOUT_MS))) {
1257 		iwqp->suspend_pending = false;
1258 		ibdev_warn(&iwqp->iwdev->ibdev,
1259 			   "modify_qp timed out waiting for suspend. qp_id = %d, last_ae = 0x%x\n",
1260 			   iwqp->ibqp.qp_num, iwqp->last_aeq);
1261 		return -EBUSY;
1262 	}
1263 
1264 	return 0;
1265 }
1266 
1267 /**
1268  * irdma_modify_qp_roce - modify qp request
1269  * @ibqp: qp's pointer for modify
1270  * @attr: access attributes
1271  * @attr_mask: state mask
1272  * @udata: user data
1273  */
1274 int irdma_modify_qp_roce(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1275 			 int attr_mask, struct ib_udata *udata)
1276 {
1277 #define IRDMA_MODIFY_QP_MIN_REQ_LEN offsetofend(struct irdma_modify_qp_req, rq_flush)
1278 #define IRDMA_MODIFY_QP_MIN_RESP_LEN offsetofend(struct irdma_modify_qp_resp, push_valid)
1279 	struct irdma_pd *iwpd = to_iwpd(ibqp->pd);
1280 	struct irdma_qp *iwqp = to_iwqp(ibqp);
1281 	struct irdma_device *iwdev = iwqp->iwdev;
1282 	struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
1283 	struct irdma_qp_host_ctx_info *ctx_info;
1284 	struct irdma_roce_offload_info *roce_info;
1285 	struct irdma_udp_offload_info *udp_info;
1286 	struct irdma_modify_qp_info info = {};
1287 	struct irdma_modify_qp_resp uresp = {};
1288 	struct irdma_modify_qp_req ureq = {};
1289 	unsigned long flags;
1290 	u8 issue_modify_qp = 0;
1291 	int ret = 0;
1292 
1293 	ctx_info = &iwqp->ctx_info;
1294 	roce_info = &iwqp->roce_info;
1295 	udp_info = &iwqp->udp_info;
1296 
1297 	if (udata) {
1298 		/* udata inlen/outlen can be 0 when supporting legacy libi40iw */
1299 		if ((udata->inlen && udata->inlen < IRDMA_MODIFY_QP_MIN_REQ_LEN) ||
1300 		    (udata->outlen && udata->outlen < IRDMA_MODIFY_QP_MIN_RESP_LEN))
1301 			return -EINVAL;
1302 	}
1303 
1304 	if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
1305 		return -EOPNOTSUPP;
1306 
1307 	if (attr_mask & IB_QP_DEST_QPN)
1308 		roce_info->dest_qp = attr->dest_qp_num;
1309 
1310 	if (attr_mask & IB_QP_PKEY_INDEX) {
1311 		ret = irdma_query_pkey(ibqp->device, 0, attr->pkey_index,
1312 				       &roce_info->p_key);
1313 		if (ret)
1314 			return ret;
1315 	}
1316 
1317 	if (attr_mask & IB_QP_QKEY)
1318 		roce_info->qkey = attr->qkey;
1319 
1320 	if (attr_mask & IB_QP_PATH_MTU)
1321 		udp_info->snd_mss = ib_mtu_enum_to_int(attr->path_mtu);
1322 
1323 	if (attr_mask & IB_QP_SQ_PSN) {
1324 		udp_info->psn_nxt = attr->sq_psn;
1325 		udp_info->lsn =  0xffff;
1326 		udp_info->psn_una = attr->sq_psn;
1327 		udp_info->psn_max = attr->sq_psn;
1328 	}
1329 
1330 	if (attr_mask & IB_QP_RQ_PSN)
1331 		udp_info->epsn = attr->rq_psn;
1332 
1333 	if (attr_mask & IB_QP_RNR_RETRY)
1334 		udp_info->rnr_nak_thresh = attr->rnr_retry;
1335 
1336 	if (attr_mask & IB_QP_MIN_RNR_TIMER &&
1337 	    dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_3)
1338 		udp_info->min_rnr_timer = attr->min_rnr_timer;
1339 
1340 	if (attr_mask & IB_QP_RETRY_CNT)
1341 		udp_info->rexmit_thresh = attr->retry_cnt;
1342 
1343 	ctx_info->roce_info->pd_id = iwpd->sc_pd.pd_id;
1344 
1345 	if (attr_mask & IB_QP_AV) {
1346 		struct irdma_av *av = &iwqp->roce_ah.av;
1347 		const struct ib_gid_attr *sgid_attr =
1348 				attr->ah_attr.grh.sgid_attr;
1349 		u16 vlan_id = VLAN_N_VID;
1350 		u32 local_ip[4];
1351 
1352 		memset(&iwqp->roce_ah, 0, sizeof(iwqp->roce_ah));
1353 		if (attr->ah_attr.ah_flags & IB_AH_GRH) {
1354 			udp_info->ttl = attr->ah_attr.grh.hop_limit;
1355 			udp_info->flow_label = attr->ah_attr.grh.flow_label;
1356 			udp_info->tos = attr->ah_attr.grh.traffic_class;
1357 			udp_info->src_port =
1358 				rdma_get_udp_sport(udp_info->flow_label,
1359 						   ibqp->qp_num,
1360 						   roce_info->dest_qp);
1361 			irdma_qp_rem_qos(&iwqp->sc_qp);
1362 			dev->ws_remove(iwqp->sc_qp.vsi, ctx_info->user_pri);
1363 			if (iwqp->sc_qp.vsi->dscp_mode)
1364 				ctx_info->user_pri =
1365 				iwqp->sc_qp.vsi->dscp_map[irdma_tos2dscp(udp_info->tos)];
1366 			else
1367 				ctx_info->user_pri = rt_tos2priority(udp_info->tos);
1368 		}
1369 		ret = rdma_read_gid_l2_fields(sgid_attr, &vlan_id,
1370 					      ctx_info->roce_info->mac_addr);
1371 		if (ret)
1372 			return ret;
1373 		ctx_info->user_pri = irdma_roce_get_vlan_prio(sgid_attr,
1374 							      ctx_info->user_pri);
1375 		if (dev->ws_add(iwqp->sc_qp.vsi, ctx_info->user_pri))
1376 			return -ENOMEM;
1377 		iwqp->sc_qp.user_pri = ctx_info->user_pri;
1378 		irdma_qp_add_qos(&iwqp->sc_qp);
1379 
1380 		if (vlan_id >= VLAN_N_VID && iwdev->dcb_vlan_mode)
1381 			vlan_id = 0;
1382 		if (vlan_id < VLAN_N_VID) {
1383 			udp_info->insert_vlan_tag = true;
1384 			udp_info->vlan_tag = vlan_id |
1385 				ctx_info->user_pri << VLAN_PRIO_SHIFT;
1386 		} else {
1387 			udp_info->insert_vlan_tag = false;
1388 		}
1389 
1390 		av->attrs = attr->ah_attr;
1391 		rdma_gid2ip((struct sockaddr *)&av->sgid_addr, &sgid_attr->gid);
1392 		rdma_gid2ip((struct sockaddr *)&av->dgid_addr, &attr->ah_attr.grh.dgid);
1393 		av->net_type = rdma_gid_attr_network_type(sgid_attr);
1394 		if (av->net_type == RDMA_NETWORK_IPV6) {
1395 			__be32 *daddr =
1396 				av->dgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32;
1397 			__be32 *saddr =
1398 				av->sgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32;
1399 
1400 			irdma_copy_ip_ntohl(&udp_info->dest_ip_addr[0], daddr);
1401 			irdma_copy_ip_ntohl(&udp_info->local_ipaddr[0], saddr);
1402 
1403 			udp_info->ipv4 = false;
1404 			irdma_copy_ip_ntohl(local_ip, daddr);
1405 
1406 		} else if (av->net_type == RDMA_NETWORK_IPV4) {
1407 			__be32 saddr = av->sgid_addr.saddr_in.sin_addr.s_addr;
1408 			__be32 daddr = av->dgid_addr.saddr_in.sin_addr.s_addr;
1409 
1410 			local_ip[0] = ntohl(daddr);
1411 
1412 			udp_info->ipv4 = true;
1413 			udp_info->dest_ip_addr[0] = 0;
1414 			udp_info->dest_ip_addr[1] = 0;
1415 			udp_info->dest_ip_addr[2] = 0;
1416 			udp_info->dest_ip_addr[3] = local_ip[0];
1417 
1418 			udp_info->local_ipaddr[0] = 0;
1419 			udp_info->local_ipaddr[1] = 0;
1420 			udp_info->local_ipaddr[2] = 0;
1421 			udp_info->local_ipaddr[3] = ntohl(saddr);
1422 		}
1423 		udp_info->arp_idx =
1424 			irdma_add_arp(iwdev->rf, local_ip, udp_info->ipv4,
1425 				      attr->ah_attr.roce.dmac);
1426 	}
1427 
1428 	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
1429 		if (attr->max_rd_atomic > dev->hw_attrs.max_hw_ord) {
1430 			ibdev_err(&iwdev->ibdev,
1431 				  "rd_atomic = %d, above max_hw_ord=%d\n",
1432 				  attr->max_rd_atomic,
1433 				  dev->hw_attrs.max_hw_ord);
1434 			return -EINVAL;
1435 		}
1436 		if (attr->max_rd_atomic)
1437 			roce_info->ord_size = attr->max_rd_atomic;
1438 		info.ord_valid = true;
1439 	}
1440 
1441 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
1442 		if (attr->max_dest_rd_atomic > dev->hw_attrs.max_hw_ird) {
1443 			ibdev_err(&iwdev->ibdev,
1444 				  "rd_atomic = %d, above max_hw_ird=%d\n",
1445 				   attr->max_dest_rd_atomic,
1446 				   dev->hw_attrs.max_hw_ird);
1447 			return -EINVAL;
1448 		}
1449 		if (attr->max_dest_rd_atomic)
1450 			roce_info->ird_size = attr->max_dest_rd_atomic;
1451 	}
1452 
1453 	if (attr_mask & IB_QP_ACCESS_FLAGS) {
1454 		if (attr->qp_access_flags & IB_ACCESS_LOCAL_WRITE)
1455 			roce_info->wr_rdresp_en = true;
1456 		if (attr->qp_access_flags & IB_ACCESS_REMOTE_WRITE)
1457 			roce_info->wr_rdresp_en = true;
1458 		if (attr->qp_access_flags & IB_ACCESS_REMOTE_READ)
1459 			roce_info->rd_en = true;
1460 		if (dev->hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_ATOMIC_OPS)
1461 			if (attr->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC)
1462 				ctx_info->remote_atomics_en = true;
1463 	}
1464 
1465 	ibdev_dbg(&iwdev->ibdev,
1466 		  "VERBS: caller: %pS qp_id=%d to_ibqpstate=%d ibqpstate=%d irdma_qpstate=%d attr_mask=0x%x\n",
1467 		  __builtin_return_address(0), ibqp->qp_num, attr->qp_state,
1468 		  iwqp->ibqp_state, iwqp->iwarp_state, attr_mask);
1469 
1470 	spin_lock_irqsave(&iwqp->lock, flags);
1471 	if (attr_mask & IB_QP_STATE) {
1472 		if (!ib_modify_qp_is_ok(iwqp->ibqp_state, attr->qp_state,
1473 					iwqp->ibqp.qp_type, attr_mask)) {
1474 			ibdev_warn(&iwdev->ibdev, "modify_qp invalid for qp_id=%d, old_state=0x%x, new_state=0x%x\n",
1475 				   iwqp->ibqp.qp_num, iwqp->ibqp_state,
1476 				   attr->qp_state);
1477 			ret = -EINVAL;
1478 			goto exit;
1479 		}
1480 		info.curr_iwarp_state = iwqp->iwarp_state;
1481 
1482 		switch (attr->qp_state) {
1483 		case IB_QPS_INIT:
1484 			if (iwqp->iwarp_state > IRDMA_QP_STATE_IDLE) {
1485 				ret = -EINVAL;
1486 				goto exit;
1487 			}
1488 
1489 			if (iwqp->iwarp_state == IRDMA_QP_STATE_INVALID) {
1490 				info.next_iwarp_state = IRDMA_QP_STATE_IDLE;
1491 				issue_modify_qp = 1;
1492 			}
1493 			break;
1494 		case IB_QPS_RTR:
1495 			if (iwqp->iwarp_state > IRDMA_QP_STATE_IDLE) {
1496 				ret = -EINVAL;
1497 				goto exit;
1498 			}
1499 			info.arp_cache_idx_valid = true;
1500 			info.cq_num_valid = true;
1501 			info.next_iwarp_state = IRDMA_QP_STATE_RTR;
1502 			issue_modify_qp = 1;
1503 			break;
1504 		case IB_QPS_RTS:
1505 			if (iwqp->ibqp_state < IB_QPS_RTR ||
1506 			    iwqp->ibqp_state == IB_QPS_ERR) {
1507 				ret = -EINVAL;
1508 				goto exit;
1509 			}
1510 
1511 			info.arp_cache_idx_valid = true;
1512 			info.cq_num_valid = true;
1513 			info.ord_valid = true;
1514 			info.next_iwarp_state = IRDMA_QP_STATE_RTS;
1515 			issue_modify_qp = 1;
1516 			if (iwdev->push_mode && udata &&
1517 			    iwqp->sc_qp.push_idx == IRDMA_INVALID_PUSH_PAGE_INDEX &&
1518 			    dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1519 				spin_unlock_irqrestore(&iwqp->lock, flags);
1520 				irdma_alloc_push_page(iwqp);
1521 				spin_lock_irqsave(&iwqp->lock, flags);
1522 			}
1523 			break;
1524 		case IB_QPS_SQD:
1525 			if (iwqp->iwarp_state == IRDMA_QP_STATE_SQD)
1526 				goto exit;
1527 
1528 			if (iwqp->iwarp_state != IRDMA_QP_STATE_RTS) {
1529 				ret = -EINVAL;
1530 				goto exit;
1531 			}
1532 
1533 			info.next_iwarp_state = IRDMA_QP_STATE_SQD;
1534 			issue_modify_qp = 1;
1535 			iwqp->suspend_pending = true;
1536 			break;
1537 		case IB_QPS_SQE:
1538 		case IB_QPS_ERR:
1539 		case IB_QPS_RESET:
1540 			if (iwqp->iwarp_state == IRDMA_QP_STATE_ERROR) {
1541 				iwqp->ibqp_state = attr->qp_state;
1542 				spin_unlock_irqrestore(&iwqp->lock, flags);
1543 				if (udata && udata->inlen) {
1544 					if (ib_copy_from_udata(&ureq, udata,
1545 					    min(sizeof(ureq), udata->inlen)))
1546 						return -EINVAL;
1547 
1548 					irdma_flush_wqes(iwqp,
1549 					    (ureq.sq_flush ? IRDMA_FLUSH_SQ : 0) |
1550 					    (ureq.rq_flush ? IRDMA_FLUSH_RQ : 0) |
1551 					    IRDMA_REFLUSH);
1552 				}
1553 				return 0;
1554 			}
1555 
1556 			info.next_iwarp_state = IRDMA_QP_STATE_ERROR;
1557 			issue_modify_qp = 1;
1558 			break;
1559 		default:
1560 			ret = -EINVAL;
1561 			goto exit;
1562 		}
1563 
1564 		iwqp->ibqp_state = attr->qp_state;
1565 	}
1566 
1567 	ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
1568 	ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
1569 	irdma_sc_qp_setctx_roce(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
1570 	spin_unlock_irqrestore(&iwqp->lock, flags);
1571 
1572 	if (attr_mask & IB_QP_STATE) {
1573 		if (issue_modify_qp) {
1574 			ctx_info->rem_endpoint_idx = udp_info->arp_idx;
1575 			if (irdma_hw_modify_qp(iwdev, iwqp, &info, true))
1576 				return -EINVAL;
1577 			if (info.next_iwarp_state == IRDMA_QP_STATE_SQD) {
1578 				ret = irdma_wait_for_suspend(iwqp);
1579 				if (ret)
1580 					return ret;
1581 			}
1582 			spin_lock_irqsave(&iwqp->lock, flags);
1583 			if (iwqp->iwarp_state == info.curr_iwarp_state) {
1584 				iwqp->iwarp_state = info.next_iwarp_state;
1585 				iwqp->ibqp_state = attr->qp_state;
1586 			}
1587 			if (iwqp->ibqp_state > IB_QPS_RTS &&
1588 			    !iwqp->flush_issued) {
1589 				spin_unlock_irqrestore(&iwqp->lock, flags);
1590 				irdma_flush_wqes(iwqp, IRDMA_FLUSH_SQ |
1591 						       IRDMA_FLUSH_RQ |
1592 						       IRDMA_FLUSH_WAIT);
1593 				iwqp->flush_issued = 1;
1594 			} else {
1595 				spin_unlock_irqrestore(&iwqp->lock, flags);
1596 			}
1597 		} else {
1598 			iwqp->ibqp_state = attr->qp_state;
1599 		}
1600 		if (udata && udata->outlen && dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1601 			struct irdma_ucontext *ucontext;
1602 
1603 			ucontext = rdma_udata_to_drv_context(udata,
1604 					struct irdma_ucontext, ibucontext);
1605 			if (iwqp->sc_qp.push_idx != IRDMA_INVALID_PUSH_PAGE_INDEX &&
1606 			    !iwqp->push_wqe_mmap_entry &&
1607 			    !irdma_setup_push_mmap_entries(ucontext, iwqp,
1608 				&uresp.push_wqe_mmap_key, &uresp.push_db_mmap_key)) {
1609 				uresp.push_valid = 1;
1610 				uresp.push_offset = iwqp->sc_qp.push_offset;
1611 			}
1612 			ret = ib_respond_udata(udata, uresp);
1613 			if (ret) {
1614 				irdma_remove_push_mmap_entries(iwqp);
1615 				return ret;
1616 			}
1617 		}
1618 	}
1619 
1620 	return 0;
1621 exit:
1622 	spin_unlock_irqrestore(&iwqp->lock, flags);
1623 
1624 	return ret;
1625 }
1626 
1627 /**
1628  * irdma_modify_qp - modify qp request
1629  * @ibqp: qp's pointer for modify
1630  * @attr: access attributes
1631  * @attr_mask: state mask
1632  * @udata: user data
1633  */
1634 int irdma_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask,
1635 		    struct ib_udata *udata)
1636 {
1637 #define IRDMA_MODIFY_QP_MIN_REQ_LEN offsetofend(struct irdma_modify_qp_req, rq_flush)
1638 #define IRDMA_MODIFY_QP_MIN_RESP_LEN offsetofend(struct irdma_modify_qp_resp, push_valid)
1639 	struct irdma_qp *iwqp = to_iwqp(ibqp);
1640 	struct irdma_device *iwdev = iwqp->iwdev;
1641 	struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
1642 	struct irdma_qp_host_ctx_info *ctx_info;
1643 	struct irdma_tcp_offload_info *tcp_info;
1644 	struct irdma_iwarp_offload_info *offload_info;
1645 	struct irdma_modify_qp_info info = {};
1646 	struct irdma_modify_qp_resp uresp = {};
1647 	struct irdma_modify_qp_req ureq = {};
1648 	u8 issue_modify_qp = 0;
1649 	u8 dont_wait = 0;
1650 	int err;
1651 	unsigned long flags;
1652 
1653 	if (udata) {
1654 		/* udata inlen/outlen can be 0 when supporting legacy libi40iw */
1655 		if ((udata->inlen && udata->inlen < IRDMA_MODIFY_QP_MIN_REQ_LEN) ||
1656 		    (udata->outlen && udata->outlen < IRDMA_MODIFY_QP_MIN_RESP_LEN))
1657 			return -EINVAL;
1658 	}
1659 
1660 	if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
1661 		return -EOPNOTSUPP;
1662 
1663 	ctx_info = &iwqp->ctx_info;
1664 	offload_info = &iwqp->iwarp_info;
1665 	tcp_info = &iwqp->tcp_info;
1666 	wait_event(iwqp->mod_qp_waitq, !atomic_read(&iwqp->hw_mod_qp_pend));
1667 	ibdev_dbg(&iwdev->ibdev,
1668 		  "VERBS: 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",
1669 		  __builtin_return_address(0), ibqp->qp_num, attr->qp_state,
1670 		  iwqp->ibqp_state, iwqp->iwarp_state, iwqp->last_aeq,
1671 		  iwqp->hw_tcp_state, iwqp->hw_iwarp_state, attr_mask);
1672 
1673 	spin_lock_irqsave(&iwqp->lock, flags);
1674 	if (attr_mask & IB_QP_STATE) {
1675 		info.curr_iwarp_state = iwqp->iwarp_state;
1676 		switch (attr->qp_state) {
1677 		case IB_QPS_INIT:
1678 		case IB_QPS_RTR:
1679 			if (iwqp->iwarp_state > IRDMA_QP_STATE_IDLE) {
1680 				err = -EINVAL;
1681 				goto exit;
1682 			}
1683 
1684 			if (iwqp->iwarp_state == IRDMA_QP_STATE_INVALID) {
1685 				info.next_iwarp_state = IRDMA_QP_STATE_IDLE;
1686 				issue_modify_qp = 1;
1687 			}
1688 			if (iwdev->push_mode && udata &&
1689 			    iwqp->sc_qp.push_idx == IRDMA_INVALID_PUSH_PAGE_INDEX &&
1690 			    dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1691 				spin_unlock_irqrestore(&iwqp->lock, flags);
1692 				irdma_alloc_push_page(iwqp);
1693 				spin_lock_irqsave(&iwqp->lock, flags);
1694 			}
1695 			break;
1696 		case IB_QPS_RTS:
1697 			if (iwqp->iwarp_state > IRDMA_QP_STATE_RTS ||
1698 			    !iwqp->cm_id) {
1699 				err = -EINVAL;
1700 				goto exit;
1701 			}
1702 
1703 			issue_modify_qp = 1;
1704 			iwqp->hw_tcp_state = IRDMA_TCP_STATE_ESTABLISHED;
1705 			iwqp->hte_added = 1;
1706 			info.next_iwarp_state = IRDMA_QP_STATE_RTS;
1707 			info.tcp_ctx_valid = true;
1708 			info.ord_valid = true;
1709 			info.arp_cache_idx_valid = true;
1710 			info.cq_num_valid = true;
1711 			break;
1712 		case IB_QPS_SQD:
1713 			if (iwqp->hw_iwarp_state > IRDMA_QP_STATE_RTS) {
1714 				err = 0;
1715 				goto exit;
1716 			}
1717 
1718 			if (iwqp->iwarp_state == IRDMA_QP_STATE_CLOSING ||
1719 			    iwqp->iwarp_state < IRDMA_QP_STATE_RTS) {
1720 				err = 0;
1721 				goto exit;
1722 			}
1723 
1724 			if (iwqp->iwarp_state > IRDMA_QP_STATE_CLOSING) {
1725 				err = -EINVAL;
1726 				goto exit;
1727 			}
1728 
1729 			info.next_iwarp_state = IRDMA_QP_STATE_CLOSING;
1730 			issue_modify_qp = 1;
1731 			break;
1732 		case IB_QPS_SQE:
1733 			if (iwqp->iwarp_state >= IRDMA_QP_STATE_TERMINATE) {
1734 				err = -EINVAL;
1735 				goto exit;
1736 			}
1737 
1738 			info.next_iwarp_state = IRDMA_QP_STATE_TERMINATE;
1739 			issue_modify_qp = 1;
1740 			break;
1741 		case IB_QPS_ERR:
1742 		case IB_QPS_RESET:
1743 			if (iwqp->iwarp_state == IRDMA_QP_STATE_ERROR) {
1744 				iwqp->ibqp_state = attr->qp_state;
1745 				spin_unlock_irqrestore(&iwqp->lock, flags);
1746 				if (udata && udata->inlen) {
1747 					if (ib_copy_from_udata(&ureq, udata,
1748 					    min(sizeof(ureq), udata->inlen)))
1749 						return -EINVAL;
1750 
1751 					irdma_flush_wqes(iwqp,
1752 					    (ureq.sq_flush ? IRDMA_FLUSH_SQ : 0) |
1753 					    (ureq.rq_flush ? IRDMA_FLUSH_RQ : 0) |
1754 					    IRDMA_REFLUSH);
1755 				}
1756 				return 0;
1757 			}
1758 
1759 			if (iwqp->sc_qp.term_flags) {
1760 				spin_unlock_irqrestore(&iwqp->lock, flags);
1761 				irdma_terminate_del_timer(&iwqp->sc_qp);
1762 				spin_lock_irqsave(&iwqp->lock, flags);
1763 			}
1764 			info.next_iwarp_state = IRDMA_QP_STATE_ERROR;
1765 			if (iwqp->hw_tcp_state > IRDMA_TCP_STATE_CLOSED &&
1766 			    iwdev->iw_status &&
1767 			    iwqp->hw_tcp_state != IRDMA_TCP_STATE_TIME_WAIT)
1768 				info.reset_tcp_conn = true;
1769 			else
1770 				dont_wait = 1;
1771 
1772 			issue_modify_qp = 1;
1773 			info.next_iwarp_state = IRDMA_QP_STATE_ERROR;
1774 			break;
1775 		default:
1776 			err = -EINVAL;
1777 			goto exit;
1778 		}
1779 
1780 		iwqp->ibqp_state = attr->qp_state;
1781 	}
1782 	if (attr_mask & IB_QP_ACCESS_FLAGS) {
1783 		ctx_info->iwarp_info_valid = true;
1784 		if (attr->qp_access_flags & IB_ACCESS_LOCAL_WRITE)
1785 			offload_info->wr_rdresp_en = true;
1786 		if (attr->qp_access_flags & IB_ACCESS_REMOTE_WRITE)
1787 			offload_info->wr_rdresp_en = true;
1788 		if (attr->qp_access_flags & IB_ACCESS_REMOTE_READ)
1789 			offload_info->rd_en = true;
1790 	}
1791 
1792 	if (ctx_info->iwarp_info_valid) {
1793 		ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
1794 		ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
1795 		irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
1796 	}
1797 	spin_unlock_irqrestore(&iwqp->lock, flags);
1798 
1799 	if (attr_mask & IB_QP_STATE) {
1800 		if (issue_modify_qp) {
1801 			ctx_info->rem_endpoint_idx = tcp_info->arp_idx;
1802 			if (irdma_hw_modify_qp(iwdev, iwqp, &info, true))
1803 				return -EINVAL;
1804 		}
1805 
1806 		spin_lock_irqsave(&iwqp->lock, flags);
1807 		if (iwqp->iwarp_state == info.curr_iwarp_state) {
1808 			iwqp->iwarp_state = info.next_iwarp_state;
1809 			iwqp->ibqp_state = attr->qp_state;
1810 		}
1811 		spin_unlock_irqrestore(&iwqp->lock, flags);
1812 	}
1813 
1814 	if (issue_modify_qp && iwqp->ibqp_state > IB_QPS_RTS) {
1815 		if (dont_wait) {
1816 			if (iwqp->hw_tcp_state) {
1817 				spin_lock_irqsave(&iwqp->lock, flags);
1818 				iwqp->hw_tcp_state = IRDMA_TCP_STATE_CLOSED;
1819 				iwqp->last_aeq = IRDMA_AE_RESET_SENT;
1820 				spin_unlock_irqrestore(&iwqp->lock, flags);
1821 			}
1822 			irdma_cm_disconn(iwqp);
1823 		} else {
1824 			int close_timer_started;
1825 
1826 			spin_lock_irqsave(&iwdev->cm_core.ht_lock, flags);
1827 
1828 			if (iwqp->cm_node) {
1829 				refcount_inc(&iwqp->cm_node->refcnt);
1830 				spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
1831 				close_timer_started = atomic_inc_return(&iwqp->close_timer_started);
1832 				if (iwqp->cm_id && close_timer_started == 1)
1833 					irdma_schedule_cm_timer(iwqp->cm_node,
1834 						(struct irdma_puda_buf *)iwqp,
1835 						IRDMA_TIMER_TYPE_CLOSE, 1, 0);
1836 
1837 				irdma_rem_ref_cm_node(iwqp->cm_node);
1838 			} else {
1839 				spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
1840 			}
1841 		}
1842 	}
1843 	if (attr_mask & IB_QP_STATE && udata && udata->outlen &&
1844 	    dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1845 		struct irdma_ucontext *ucontext;
1846 
1847 		ucontext = rdma_udata_to_drv_context(udata,
1848 					struct irdma_ucontext, ibucontext);
1849 		if (iwqp->sc_qp.push_idx != IRDMA_INVALID_PUSH_PAGE_INDEX &&
1850 		    !iwqp->push_wqe_mmap_entry &&
1851 		    !irdma_setup_push_mmap_entries(ucontext, iwqp,
1852 			&uresp.push_wqe_mmap_key, &uresp.push_db_mmap_key)) {
1853 			uresp.push_valid = 1;
1854 			uresp.push_offset = iwqp->sc_qp.push_offset;
1855 		}
1856 
1857 		err = ib_respond_udata(udata, uresp);
1858 		if (err) {
1859 			irdma_remove_push_mmap_entries(iwqp);
1860 			return err;
1861 		}
1862 	}
1863 
1864 	return 0;
1865 exit:
1866 	spin_unlock_irqrestore(&iwqp->lock, flags);
1867 
1868 	return err;
1869 }
1870 
1871 /**
1872  * irdma_srq_free_rsrc - free up resources for srq
1873  * @rf: RDMA PCI function
1874  * @iwsrq: srq ptr
1875  */
1876 static void irdma_srq_free_rsrc(struct irdma_pci_f *rf, struct irdma_srq *iwsrq)
1877 {
1878 	struct irdma_sc_srq *srq = &iwsrq->sc_srq;
1879 
1880 	if (!iwsrq->user_mode) {
1881 		dma_free_coherent(rf->sc_dev.hw->device, iwsrq->kmem.size,
1882 				  iwsrq->kmem.va, iwsrq->kmem.pa);
1883 		iwsrq->kmem.va = NULL;
1884 	}
1885 
1886 	irdma_free_rsrc(rf, rf->allocated_srqs, srq->srq_uk.srq_id);
1887 }
1888 
1889 /**
1890  * irdma_cq_free_rsrc - free up resources for cq
1891  * @rf: RDMA PCI function
1892  * @iwcq: cq ptr
1893  */
1894 static void irdma_cq_free_rsrc(struct irdma_pci_f *rf, struct irdma_cq *iwcq)
1895 {
1896 	struct irdma_sc_cq *cq = &iwcq->sc_cq;
1897 
1898 	if (!iwcq->user_mode) {
1899 		dma_free_coherent(rf->sc_dev.hw->device, iwcq->kmem.size,
1900 				  iwcq->kmem.va, iwcq->kmem.pa);
1901 		iwcq->kmem.va = NULL;
1902 		dma_free_coherent(rf->sc_dev.hw->device,
1903 				  iwcq->kmem_shadow.size,
1904 				  iwcq->kmem_shadow.va, iwcq->kmem_shadow.pa);
1905 		iwcq->kmem_shadow.va = NULL;
1906 	}
1907 
1908 	irdma_free_rsrc(rf, rf->allocated_cqs, cq->cq_uk.cq_id);
1909 }
1910 
1911 /**
1912  * irdma_free_cqbuf - worker to free a cq buffer
1913  * @work: provides access to the cq buffer to free
1914  */
1915 static void irdma_free_cqbuf(struct work_struct *work)
1916 {
1917 	struct irdma_cq_buf *cq_buf = container_of(work, struct irdma_cq_buf, work);
1918 
1919 	dma_free_coherent(cq_buf->hw->device, cq_buf->kmem_buf.size,
1920 			  cq_buf->kmem_buf.va, cq_buf->kmem_buf.pa);
1921 	cq_buf->kmem_buf.va = NULL;
1922 	kfree(cq_buf);
1923 }
1924 
1925 /**
1926  * irdma_process_resize_list - remove resized cq buffers from the resize_list
1927  * @iwcq: cq which owns the resize_list
1928  * @iwdev: irdma device
1929  * @lcqe_buf: the buffer where the last cqe is received
1930  */
1931 static int irdma_process_resize_list(struct irdma_cq *iwcq,
1932 				     struct irdma_device *iwdev,
1933 				     struct irdma_cq_buf *lcqe_buf)
1934 {
1935 	struct list_head *tmp_node, *list_node;
1936 	struct irdma_cq_buf *cq_buf;
1937 	int cnt = 0;
1938 
1939 	list_for_each_safe(list_node, tmp_node, &iwcq->resize_list) {
1940 		cq_buf = list_entry(list_node, struct irdma_cq_buf, list);
1941 		if (cq_buf == lcqe_buf)
1942 			return cnt;
1943 
1944 		list_del(&cq_buf->list);
1945 		queue_work(iwdev->cleanup_wq, &cq_buf->work);
1946 		cnt++;
1947 	}
1948 
1949 	return cnt;
1950 }
1951 
1952 /**
1953  * irdma_destroy_srq - destroy srq
1954  * @ibsrq: srq pointer
1955  * @udata: user data
1956  */
1957 static int irdma_destroy_srq(struct ib_srq *ibsrq, struct ib_udata *udata)
1958 {
1959 	struct irdma_device *iwdev = to_iwdev(ibsrq->device);
1960 	struct irdma_srq *iwsrq = to_iwsrq(ibsrq);
1961 	struct irdma_sc_srq *srq = &iwsrq->sc_srq;
1962 
1963 	irdma_srq_wq_destroy(iwdev->rf, srq);
1964 	irdma_srq_free_rsrc(iwdev->rf, iwsrq);
1965 	return 0;
1966 }
1967 
1968 /**
1969  * irdma_destroy_cq - destroy cq
1970  * @ib_cq: cq pointer
1971  * @udata: user data
1972  */
1973 static int irdma_destroy_cq(struct ib_cq *ib_cq, struct ib_udata *udata)
1974 {
1975 	struct irdma_device *iwdev = to_iwdev(ib_cq->device);
1976 	struct irdma_cq *iwcq = to_iwcq(ib_cq);
1977 	struct irdma_sc_cq *cq = &iwcq->sc_cq;
1978 	struct irdma_sc_dev *dev = cq->dev;
1979 	struct irdma_sc_ceq *ceq = dev->ceq[cq->ceq_id];
1980 	struct irdma_ceq *iwceq = container_of(ceq, struct irdma_ceq, sc_ceq);
1981 	unsigned long flags;
1982 
1983 	spin_lock_irqsave(&iwcq->lock, flags);
1984 	if (!list_empty(&iwcq->cmpl_generated))
1985 		irdma_remove_cmpls_list(iwcq);
1986 	if (!list_empty(&iwcq->resize_list))
1987 		irdma_process_resize_list(iwcq, iwdev, NULL);
1988 	spin_unlock_irqrestore(&iwcq->lock, flags);
1989 
1990 	irdma_cq_rem_ref(ib_cq);
1991 	wait_for_completion(&iwcq->free_cq);
1992 
1993 	irdma_cq_wq_destroy(iwdev->rf, cq);
1994 
1995 	spin_lock_irqsave(&iwceq->ce_lock, flags);
1996 	irdma_sc_cleanup_ceqes(cq, ceq);
1997 	spin_unlock_irqrestore(&iwceq->ce_lock, flags);
1998 	irdma_cq_free_rsrc(iwdev->rf, iwcq);
1999 
2000 	return 0;
2001 }
2002 
2003 /**
2004  * irdma_resize_cq - resize cq
2005  * @ibcq: cq to be resized
2006  * @entries: desired cq size
2007  * @udata: user data
2008  */
2009 static int irdma_resize_cq(struct ib_cq *ibcq, unsigned int entries,
2010 			   struct ib_udata *udata)
2011 {
2012 #define IRDMA_RESIZE_CQ_MIN_REQ_LEN offsetofend(struct irdma_resize_cq_req, user_cq_buffer)
2013 	struct irdma_cq *iwcq = to_iwcq(ibcq);
2014 	struct irdma_sc_dev *dev = iwcq->sc_cq.dev;
2015 	struct irdma_cqp_request *cqp_request;
2016 	struct cqp_cmds_info *cqp_info;
2017 	struct irdma_modify_cq_info *m_info;
2018 	struct irdma_modify_cq_info info = {};
2019 	struct irdma_dma_mem kmem_buf;
2020 	struct irdma_cq_mr *cqmr_buf;
2021 	struct irdma_pbl *iwpbl_buf;
2022 	struct irdma_device *iwdev;
2023 	struct irdma_pci_f *rf;
2024 	struct irdma_cq_buf *cq_buf = NULL;
2025 	unsigned long flags;
2026 	u8 cqe_size;
2027 	int ret;
2028 
2029 	iwdev = to_iwdev(ibcq->device);
2030 	rf = iwdev->rf;
2031 
2032 	if (!(rf->sc_dev.hw_attrs.uk_attrs.feature_flags &
2033 	    IRDMA_FEATURE_CQ_RESIZE))
2034 		return -EOPNOTSUPP;
2035 
2036 	if (udata && udata->inlen < IRDMA_RESIZE_CQ_MIN_REQ_LEN)
2037 		return -EINVAL;
2038 
2039 	if (entries > rf->max_cqe)
2040 		return -EINVAL;
2041 
2042 	if (!iwcq->user_mode) {
2043 		entries += 2;
2044 
2045 		if (!iwcq->sc_cq.cq_uk.avoid_mem_cflct &&
2046 		    dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
2047 			entries *= 2;
2048 
2049 		if (entries & 1)
2050 			entries += 1; /* cq size must be an even number */
2051 
2052 		cqe_size = iwcq->sc_cq.cq_uk.avoid_mem_cflct ? 64 : 32;
2053 		if (entries * cqe_size == IRDMA_HW_PAGE_SIZE)
2054 			entries += 2;
2055 	}
2056 
2057 	info.cq_size = max(entries, 4);
2058 
2059 	if (info.cq_size == iwcq->sc_cq.cq_uk.cq_size - 1)
2060 		return 0;
2061 
2062 	if (udata) {
2063 		struct irdma_resize_cq_req req = {};
2064 		struct irdma_ucontext *ucontext =
2065 			rdma_udata_to_drv_context(udata, struct irdma_ucontext,
2066 						  ibucontext);
2067 
2068 		if (ib_copy_from_udata(&req, udata,
2069 				       min(sizeof(req), udata->inlen)))
2070 			return -EINVAL;
2071 
2072 		spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
2073 		iwpbl_buf = irdma_get_pbl((unsigned long)req.user_cq_buffer,
2074 					  &ucontext->cq_reg_mem_list);
2075 		spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
2076 
2077 		if (!iwpbl_buf)
2078 			return -ENOMEM;
2079 
2080 		cqmr_buf = &iwpbl_buf->cq_mr;
2081 		if (iwpbl_buf->pbl_allocated) {
2082 			info.virtual_map = true;
2083 			info.pbl_chunk_size = 1;
2084 			info.first_pm_pbl_idx = cqmr_buf->cq_pbl.idx;
2085 		} else {
2086 			info.cq_pa = cqmr_buf->cq_pbl.addr;
2087 		}
2088 	} else {
2089 		/* Kmode CQ resize */
2090 		int rsize;
2091 
2092 		rsize = info.cq_size * sizeof(struct irdma_cqe);
2093 		kmem_buf.size = ALIGN(round_up(rsize, 256), 256);
2094 		kmem_buf.va = dma_alloc_coherent(dev->hw->device,
2095 						 kmem_buf.size, &kmem_buf.pa,
2096 						 GFP_KERNEL);
2097 		if (!kmem_buf.va)
2098 			return -ENOMEM;
2099 
2100 		info.cq_base = kmem_buf.va;
2101 		info.cq_pa = kmem_buf.pa;
2102 		cq_buf = kzalloc_obj(*cq_buf);
2103 		if (!cq_buf) {
2104 			ret = -ENOMEM;
2105 			goto error;
2106 		}
2107 	}
2108 
2109 	cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
2110 	if (!cqp_request) {
2111 		ret = -ENOMEM;
2112 		goto error;
2113 	}
2114 
2115 	info.shadow_read_threshold = iwcq->sc_cq.shadow_read_threshold;
2116 	info.cq_resize = true;
2117 
2118 	cqp_info = &cqp_request->info;
2119 	m_info = &cqp_info->in.u.cq_modify.info;
2120 	memcpy(m_info, &info, sizeof(*m_info));
2121 
2122 	cqp_info->cqp_cmd = IRDMA_OP_CQ_MODIFY;
2123 	cqp_info->in.u.cq_modify.cq = &iwcq->sc_cq;
2124 	cqp_info->in.u.cq_modify.scratch = (uintptr_t)cqp_request;
2125 	cqp_info->post_sq = 1;
2126 	ret = irdma_handle_cqp_op(rf, cqp_request);
2127 	irdma_put_cqp_request(&rf->cqp, cqp_request);
2128 	if (ret)
2129 		goto error;
2130 
2131 	spin_lock_irqsave(&iwcq->lock, flags);
2132 	if (cq_buf) {
2133 		cq_buf->kmem_buf = iwcq->kmem;
2134 		cq_buf->hw = dev->hw;
2135 		memcpy(&cq_buf->cq_uk, &iwcq->sc_cq.cq_uk, sizeof(cq_buf->cq_uk));
2136 		INIT_WORK(&cq_buf->work, irdma_free_cqbuf);
2137 		list_add_tail(&cq_buf->list, &iwcq->resize_list);
2138 		iwcq->kmem = kmem_buf;
2139 	}
2140 
2141 	irdma_sc_cq_resize(&iwcq->sc_cq, &info);
2142 	ibcq->cqe = info.cq_size - 1;
2143 	spin_unlock_irqrestore(&iwcq->lock, flags);
2144 
2145 	return 0;
2146 error:
2147 	if (!udata) {
2148 		dma_free_coherent(dev->hw->device, kmem_buf.size, kmem_buf.va,
2149 				  kmem_buf.pa);
2150 		kmem_buf.va = NULL;
2151 	}
2152 	kfree(cq_buf);
2153 
2154 	return ret;
2155 }
2156 
2157 /**
2158  * irdma_srq_event - event notification for srq limit
2159  * @srq: shared srq struct
2160  */
2161 void irdma_srq_event(struct irdma_sc_srq *srq)
2162 {
2163 	struct irdma_srq *iwsrq = container_of(srq, struct irdma_srq, sc_srq);
2164 	struct ib_srq *ibsrq = &iwsrq->ibsrq;
2165 	struct ib_event event;
2166 
2167 	srq->srq_limit = 0;
2168 
2169 	if (!ibsrq->event_handler)
2170 		return;
2171 
2172 	event.device = ibsrq->device;
2173 	event.element.port_num = 1;
2174 	event.element.srq = ibsrq;
2175 	event.event = IB_EVENT_SRQ_LIMIT_REACHED;
2176 	ibsrq->event_handler(&event, ibsrq->srq_context);
2177 }
2178 
2179 /**
2180  * irdma_modify_srq - modify srq request
2181  * @ibsrq: srq's pointer for modify
2182  * @attr: access attributes
2183  * @attr_mask: state mask
2184  * @udata: user data
2185  */
2186 static int irdma_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
2187 			    enum ib_srq_attr_mask attr_mask,
2188 			    struct ib_udata *udata)
2189 {
2190 	struct irdma_device *iwdev = to_iwdev(ibsrq->device);
2191 	struct irdma_srq *iwsrq = to_iwsrq(ibsrq);
2192 	struct irdma_cqp_request *cqp_request;
2193 	struct irdma_pci_f *rf = iwdev->rf;
2194 	struct irdma_modify_srq_info *info;
2195 	struct cqp_cmds_info *cqp_info;
2196 	int status;
2197 
2198 	if (attr_mask & IB_SRQ_MAX_WR)
2199 		return -EINVAL;
2200 
2201 	if (!(attr_mask & IB_SRQ_LIMIT))
2202 		return 0;
2203 
2204 	if (attr->srq_limit > iwsrq->sc_srq.srq_uk.srq_size)
2205 		return -EINVAL;
2206 
2207 	/* Execute this cqp op synchronously, so we can update srq_limit
2208 	 * upon successful completion.
2209 	 */
2210 	cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
2211 	if (!cqp_request)
2212 		return -ENOMEM;
2213 
2214 	cqp_info = &cqp_request->info;
2215 	info = &cqp_info->in.u.srq_modify.info;
2216 	info->srq_limit = attr->srq_limit;
2217 	if (info->srq_limit > 0xFFF)
2218 		info->srq_limit = 0xFFF;
2219 	info->arm_limit_event = 1;
2220 
2221 	cqp_info->cqp_cmd = IRDMA_OP_SRQ_MODIFY;
2222 	cqp_info->post_sq = 1;
2223 	cqp_info->in.u.srq_modify.srq = &iwsrq->sc_srq;
2224 	cqp_info->in.u.srq_modify.scratch = (uintptr_t)cqp_request;
2225 	status = irdma_handle_cqp_op(rf, cqp_request);
2226 	irdma_put_cqp_request(&rf->cqp, cqp_request);
2227 	if (status)
2228 		return status;
2229 
2230 	iwsrq->sc_srq.srq_limit = info->srq_limit;
2231 
2232 	return 0;
2233 }
2234 
2235 static int irdma_setup_umode_srq(struct irdma_device *iwdev,
2236 				 struct irdma_srq *iwsrq,
2237 				 struct irdma_srq_init_info *info,
2238 				 struct ib_udata *udata)
2239 {
2240 #define IRDMA_CREATE_SRQ_MIN_REQ_LEN \
2241 	offsetofend(struct irdma_create_srq_req, user_shadow_area)
2242 	struct irdma_create_srq_req req = {};
2243 	struct irdma_ucontext *ucontext;
2244 	struct irdma_srq_mr *srqmr;
2245 	struct irdma_pbl *iwpbl;
2246 	unsigned long flags;
2247 
2248 	iwsrq->user_mode = true;
2249 	ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext,
2250 					     ibucontext);
2251 
2252 	if (udata->inlen < IRDMA_CREATE_SRQ_MIN_REQ_LEN)
2253 		return -EINVAL;
2254 
2255 	if (ib_copy_from_udata(&req, udata,
2256 			       min(sizeof(req), udata->inlen)))
2257 		return -EFAULT;
2258 
2259 	spin_lock_irqsave(&ucontext->srq_reg_mem_list_lock, flags);
2260 	iwpbl = irdma_get_pbl((unsigned long)req.user_srq_buf,
2261 			      &ucontext->srq_reg_mem_list);
2262 	spin_unlock_irqrestore(&ucontext->srq_reg_mem_list_lock, flags);
2263 	if (!iwpbl)
2264 		return -EPROTO;
2265 
2266 	iwsrq->iwpbl = iwpbl;
2267 	srqmr = &iwpbl->srq_mr;
2268 
2269 	if (iwpbl->pbl_allocated) {
2270 		info->virtual_map = true;
2271 		info->pbl_chunk_size = 1;
2272 		info->first_pm_pbl_idx = srqmr->srq_pbl.idx;
2273 		info->leaf_pbl_size = 1;
2274 	} else {
2275 		info->srq_pa = srqmr->srq_pbl.addr;
2276 	}
2277 	info->shadow_area_pa = srqmr->shadow;
2278 
2279 	return 0;
2280 }
2281 
2282 static int irdma_setup_kmode_srq(struct irdma_device *iwdev,
2283 				 struct irdma_srq *iwsrq,
2284 				 struct irdma_srq_init_info *info, u32 depth,
2285 				 u8 shift)
2286 {
2287 	struct irdma_srq_uk_init_info *ukinfo = &info->srq_uk_init_info;
2288 	struct irdma_dma_mem *mem = &iwsrq->kmem;
2289 	u32 size, ring_size;
2290 
2291 	ring_size = depth * IRDMA_QP_WQE_MIN_SIZE;
2292 	size = ring_size + (IRDMA_SHADOW_AREA_SIZE << 3);
2293 
2294 	mem->size = ALIGN(size, 256);
2295 	mem->va = dma_alloc_coherent(iwdev->rf->hw.device, mem->size,
2296 				     &mem->pa, GFP_KERNEL);
2297 	if (!mem->va)
2298 		return -ENOMEM;
2299 
2300 	ukinfo->srq = mem->va;
2301 	ukinfo->srq_size = depth >> shift;
2302 	ukinfo->shadow_area = mem->va + ring_size;
2303 
2304 	info->srq_pa = mem->pa;
2305 	info->shadow_area_pa = info->srq_pa + ring_size;
2306 
2307 	return 0;
2308 }
2309 
2310 /**
2311  * irdma_create_srq - create srq
2312  * @ibsrq: ib's srq pointer
2313  * @initattrs: attributes for srq
2314  * @udata: user data for create srq
2315  */
2316 static int irdma_create_srq(struct ib_srq *ibsrq,
2317 			    struct ib_srq_init_attr *initattrs,
2318 			    struct ib_udata *udata)
2319 {
2320 	struct irdma_device *iwdev = to_iwdev(ibsrq->device);
2321 	struct ib_srq_attr *attr = &initattrs->attr;
2322 	struct irdma_pd *iwpd = to_iwpd(ibsrq->pd);
2323 	struct irdma_srq *iwsrq = to_iwsrq(ibsrq);
2324 	struct irdma_srq_uk_init_info *ukinfo;
2325 	struct irdma_cqp_request *cqp_request;
2326 	struct irdma_srq_init_info info = {};
2327 	struct irdma_pci_f *rf = iwdev->rf;
2328 	struct irdma_uk_attrs *uk_attrs;
2329 	struct cqp_cmds_info *cqp_info;
2330 	int err_code = 0;
2331 	u32 depth;
2332 	u8 shift;
2333 
2334 	uk_attrs = &rf->sc_dev.hw_attrs.uk_attrs;
2335 	ukinfo = &info.srq_uk_init_info;
2336 
2337 	if (initattrs->srq_type != IB_SRQT_BASIC)
2338 		return -EOPNOTSUPP;
2339 
2340 	if (!(uk_attrs->feature_flags & IRDMA_FEATURE_SRQ) ||
2341 	    attr->max_sge > uk_attrs->max_hw_wq_frags)
2342 		return -EINVAL;
2343 
2344 	refcount_set(&iwsrq->refcnt, 1);
2345 	spin_lock_init(&iwsrq->lock);
2346 	err_code = irdma_alloc_rsrc(rf, rf->allocated_srqs, rf->max_srq,
2347 				    &iwsrq->srq_num, &rf->next_srq);
2348 	if (err_code)
2349 		return err_code;
2350 
2351 	ukinfo->max_srq_frag_cnt = attr->max_sge;
2352 	ukinfo->uk_attrs = uk_attrs;
2353 	ukinfo->srq_id = iwsrq->srq_num;
2354 
2355 	irdma_get_wqe_shift(ukinfo->uk_attrs, ukinfo->max_srq_frag_cnt, 0,
2356 			    &shift);
2357 
2358 	err_code = irdma_get_srqdepth(ukinfo->uk_attrs, attr->max_wr,
2359 				      shift, &depth);
2360 	if (err_code)
2361 		return err_code;
2362 
2363 	/* Actual SRQ size in WRs for ring and HW */
2364 	ukinfo->srq_size = depth >> shift;
2365 
2366 	/* Max postable WRs to SRQ */
2367 	iwsrq->max_wr = (depth - IRDMA_RQ_RSVD) >> shift;
2368 	attr->max_wr = iwsrq->max_wr;
2369 
2370 	if (udata)
2371 		err_code = irdma_setup_umode_srq(iwdev, iwsrq, &info, udata);
2372 	else
2373 		err_code = irdma_setup_kmode_srq(iwdev, iwsrq, &info, depth,
2374 						 shift);
2375 
2376 	if (err_code)
2377 		goto free_rsrc;
2378 
2379 	info.vsi = &iwdev->vsi;
2380 	info.pd = &iwpd->sc_pd;
2381 
2382 	iwsrq->sc_srq.srq_uk.lock = &iwsrq->lock;
2383 	err_code = irdma_sc_srq_init(&iwsrq->sc_srq, &info);
2384 	if (err_code)
2385 		goto free_dmem;
2386 
2387 	cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
2388 	if (!cqp_request) {
2389 		err_code = -ENOMEM;
2390 		goto free_dmem;
2391 	}
2392 
2393 	cqp_info = &cqp_request->info;
2394 	cqp_info->cqp_cmd = IRDMA_OP_SRQ_CREATE;
2395 	cqp_info->post_sq = 1;
2396 	cqp_info->in.u.srq_create.srq = &iwsrq->sc_srq;
2397 	cqp_info->in.u.srq_create.scratch = (uintptr_t)cqp_request;
2398 	err_code = irdma_handle_cqp_op(rf, cqp_request);
2399 	irdma_put_cqp_request(&rf->cqp, cqp_request);
2400 	if (err_code)
2401 		goto free_dmem;
2402 
2403 	if (udata) {
2404 		struct irdma_create_srq_resp resp = {};
2405 
2406 		resp.srq_id = iwsrq->srq_num;
2407 		resp.srq_size = ukinfo->srq_size;
2408 		err_code = ib_respond_udata(udata, resp);
2409 		if (err_code)
2410 			goto srq_destroy;
2411 	}
2412 
2413 	return 0;
2414 
2415 srq_destroy:
2416 	irdma_srq_wq_destroy(rf, &iwsrq->sc_srq);
2417 
2418 free_dmem:
2419 	if (!iwsrq->user_mode)
2420 		dma_free_coherent(rf->hw.device, iwsrq->kmem.size,
2421 				  iwsrq->kmem.va, iwsrq->kmem.pa);
2422 free_rsrc:
2423 	irdma_free_rsrc(rf, rf->allocated_srqs, iwsrq->srq_num);
2424 	return err_code;
2425 }
2426 
2427 /**
2428  * irdma_query_srq - get SRQ attributes
2429  * @ibsrq: the SRQ to query
2430  * @attr: the attributes of the SRQ
2431  */
2432 static int irdma_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr)
2433 {
2434 	struct irdma_srq *iwsrq = to_iwsrq(ibsrq);
2435 
2436 	attr->max_wr = iwsrq->max_wr;
2437 	attr->max_sge = iwsrq->sc_srq.srq_uk.max_srq_frag_cnt;
2438 	attr->srq_limit = iwsrq->sc_srq.srq_limit;
2439 
2440 	return 0;
2441 }
2442 
2443 static inline int cq_validate_flags(u32 flags, u8 hw_rev)
2444 {
2445 	/* GEN1/2 does not support CQ create flags */
2446 	if (hw_rev <= IRDMA_GEN_2)
2447 		return flags ? -EOPNOTSUPP : 0;
2448 
2449 	return flags & ~IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION ? -EOPNOTSUPP : 0;
2450 }
2451 
2452 /**
2453  * irdma_create_cq - create cq
2454  * @ibcq: CQ allocated
2455  * @attr: attributes for cq
2456  * @attrs: uverbs attribute bundle
2457  */
2458 static int irdma_create_cq(struct ib_cq *ibcq,
2459 			   const struct ib_cq_init_attr *attr,
2460 			   struct uverbs_attr_bundle *attrs)
2461 {
2462 #define IRDMA_CREATE_CQ_MIN_REQ_LEN offsetofend(struct irdma_create_cq_req, user_cq_buf)
2463 #define IRDMA_CREATE_CQ_MIN_RESP_LEN offsetofend(struct irdma_create_cq_resp, cq_size)
2464 	struct ib_udata *udata = &attrs->driver_udata;
2465 	struct ib_device *ibdev = ibcq->device;
2466 	struct irdma_device *iwdev = to_iwdev(ibdev);
2467 	struct irdma_pci_f *rf = iwdev->rf;
2468 	struct irdma_cq *iwcq = to_iwcq(ibcq);
2469 	u32 cq_num = 0;
2470 	struct irdma_sc_cq *cq;
2471 	struct irdma_sc_dev *dev = &rf->sc_dev;
2472 	struct irdma_cq_init_info info = {};
2473 	struct irdma_cqp_request *cqp_request;
2474 	struct cqp_cmds_info *cqp_info;
2475 	struct irdma_cq_uk_init_info *ukinfo = &info.cq_uk_init_info;
2476 	unsigned long flags;
2477 	int err_code;
2478 	int entries = attr->cqe;
2479 	bool cqe_64byte_ena;
2480 	u8 cqe_size;
2481 
2482 	err_code = cq_validate_flags(attr->flags, dev->hw_attrs.uk_attrs.hw_rev);
2483 	if (err_code)
2484 		return err_code;
2485 
2486 	if (udata && (udata->inlen < IRDMA_CREATE_CQ_MIN_REQ_LEN ||
2487 		      udata->outlen < IRDMA_CREATE_CQ_MIN_RESP_LEN))
2488 		return -EINVAL;
2489 
2490 	err_code = irdma_alloc_rsrc(rf, rf->allocated_cqs, rf->max_cq, &cq_num,
2491 				    &rf->next_cq);
2492 	if (err_code)
2493 		return err_code;
2494 
2495 	cq = &iwcq->sc_cq;
2496 	cq->back_cq = iwcq;
2497 	refcount_set(&iwcq->refcnt, 1);
2498 	spin_lock_init(&iwcq->lock);
2499 	INIT_LIST_HEAD(&iwcq->resize_list);
2500 	INIT_LIST_HEAD(&iwcq->cmpl_generated);
2501 	iwcq->cq_num = cq_num;
2502 	info.dev = dev;
2503 	ukinfo->cq_size = max(entries, 4);
2504 	ukinfo->cq_id = cq_num;
2505 	cqe_64byte_ena = dev->hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_64_BYTE_CQE ?
2506 			 true : false;
2507 	cqe_size = cqe_64byte_ena ? 64 : 32;
2508 	ukinfo->avoid_mem_cflct = cqe_64byte_ena;
2509 	iwcq->ibcq.cqe = info.cq_uk_init_info.cq_size;
2510 	if (attr->comp_vector < rf->ceqs_count)
2511 		info.ceq_id = attr->comp_vector;
2512 	info.ceq_id_valid = true;
2513 	info.ceqe_mask = 1;
2514 	info.type = IRDMA_CQ_TYPE_IWARP;
2515 	info.vsi = &iwdev->vsi;
2516 
2517 	if (udata) {
2518 		struct irdma_ucontext *ucontext;
2519 		struct irdma_create_cq_req req = {};
2520 		struct irdma_cq_mr *cqmr;
2521 		struct irdma_pbl *iwpbl;
2522 		struct irdma_pbl *iwpbl_shadow;
2523 		struct irdma_cq_mr *cqmr_shadow;
2524 
2525 		iwcq->user_mode = true;
2526 		ucontext =
2527 			rdma_udata_to_drv_context(udata, struct irdma_ucontext,
2528 						  ibucontext);
2529 		if (ib_copy_from_udata(&req, udata,
2530 				       min(sizeof(req), udata->inlen))) {
2531 			err_code = -EFAULT;
2532 			goto cq_free_rsrc;
2533 		}
2534 
2535 		spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
2536 		iwpbl = irdma_get_pbl((unsigned long)req.user_cq_buf,
2537 				      &ucontext->cq_reg_mem_list);
2538 		spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
2539 		if (!iwpbl) {
2540 			err_code = -EPROTO;
2541 			goto cq_free_rsrc;
2542 		}
2543 
2544 		cqmr = &iwpbl->cq_mr;
2545 
2546 		if (rf->sc_dev.hw_attrs.uk_attrs.feature_flags &
2547 		    IRDMA_FEATURE_CQ_RESIZE) {
2548 			spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
2549 			iwpbl_shadow = irdma_get_pbl(
2550 					(unsigned long)req.user_shadow_area,
2551 					&ucontext->cq_reg_mem_list);
2552 			spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
2553 
2554 			if (!iwpbl_shadow) {
2555 				err_code = -EPROTO;
2556 				goto cq_free_rsrc;
2557 			}
2558 			cqmr_shadow = &iwpbl_shadow->cq_mr;
2559 			info.shadow_area_pa = cqmr_shadow->cq_pbl.addr;
2560 		} else {
2561 			info.shadow_area_pa = cqmr->shadow;
2562 		}
2563 		if (iwpbl->pbl_allocated) {
2564 			info.virtual_map = true;
2565 			info.pbl_chunk_size = 1;
2566 			info.first_pm_pbl_idx = cqmr->cq_pbl.idx;
2567 		} else {
2568 			info.cq_base_pa = cqmr->cq_pbl.addr;
2569 		}
2570 	} else {
2571 		/* Kmode allocations */
2572 		int rsize;
2573 
2574 		if (entries < 1 || entries > rf->max_cqe) {
2575 			err_code = -EINVAL;
2576 			goto cq_free_rsrc;
2577 		}
2578 
2579 		entries += 2;
2580 		if (!cqe_64byte_ena && dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
2581 			entries *= 2;
2582 
2583 		if (entries & 1)
2584 			entries += 1; /* cq size must be an even number */
2585 
2586 		if (entries * cqe_size == IRDMA_HW_PAGE_SIZE)
2587 			entries += 2;
2588 
2589 		ukinfo->cq_size = entries;
2590 
2591 		if (cqe_64byte_ena)
2592 			rsize = info.cq_uk_init_info.cq_size * sizeof(struct irdma_extended_cqe);
2593 		else
2594 			rsize = info.cq_uk_init_info.cq_size * sizeof(struct irdma_cqe);
2595 		iwcq->kmem.size = ALIGN(round_up(rsize, 256), 256);
2596 		iwcq->kmem.va = dma_alloc_coherent(dev->hw->device,
2597 						   iwcq->kmem.size,
2598 						   &iwcq->kmem.pa, GFP_KERNEL);
2599 		if (!iwcq->kmem.va) {
2600 			err_code = -ENOMEM;
2601 			goto cq_free_rsrc;
2602 		}
2603 
2604 		iwcq->kmem_shadow.size = ALIGN(IRDMA_SHADOW_AREA_SIZE << 3,
2605 					       64);
2606 		iwcq->kmem_shadow.va = dma_alloc_coherent(dev->hw->device,
2607 							  iwcq->kmem_shadow.size,
2608 							  &iwcq->kmem_shadow.pa,
2609 							  GFP_KERNEL);
2610 		if (!iwcq->kmem_shadow.va) {
2611 			err_code = -ENOMEM;
2612 			goto cq_free_rsrc;
2613 		}
2614 		info.shadow_area_pa = iwcq->kmem_shadow.pa;
2615 		ukinfo->shadow_area = iwcq->kmem_shadow.va;
2616 		ukinfo->cq_base = iwcq->kmem.va;
2617 		info.cq_base_pa = iwcq->kmem.pa;
2618 	}
2619 
2620 	info.shadow_read_threshold = min(info.cq_uk_init_info.cq_size / 2,
2621 					 (u32)IRDMA_MAX_CQ_READ_THRESH);
2622 
2623 	if (irdma_sc_cq_init(cq, &info)) {
2624 		ibdev_dbg(&iwdev->ibdev, "VERBS: init cq fail\n");
2625 		err_code = -EPROTO;
2626 		goto cq_free_rsrc;
2627 	}
2628 
2629 	cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
2630 	if (!cqp_request) {
2631 		err_code = -ENOMEM;
2632 		goto cq_free_rsrc;
2633 	}
2634 
2635 	cqp_info = &cqp_request->info;
2636 	cqp_info->cqp_cmd = IRDMA_OP_CQ_CREATE;
2637 	cqp_info->post_sq = 1;
2638 	cqp_info->in.u.cq_create.cq = cq;
2639 	cqp_info->in.u.cq_create.check_overflow = true;
2640 	cqp_info->in.u.cq_create.scratch = (uintptr_t)cqp_request;
2641 	err_code = irdma_handle_cqp_op(rf, cqp_request);
2642 	irdma_put_cqp_request(&rf->cqp, cqp_request);
2643 	if (err_code)
2644 		goto cq_free_rsrc;
2645 
2646 	if (udata) {
2647 		struct irdma_create_cq_resp resp = {};
2648 
2649 		resp.cq_id = info.cq_uk_init_info.cq_id;
2650 		resp.cq_size = info.cq_uk_init_info.cq_size;
2651 		err_code = ib_respond_udata(udata, resp);
2652 		if (err_code)
2653 			goto cq_destroy;
2654 	}
2655 
2656 	init_completion(&iwcq->free_cq);
2657 
2658 	/* Populate table entry after CQ is fully created. */
2659 	smp_store_release(&rf->cq_table[cq_num], iwcq);
2660 
2661 	return 0;
2662 cq_destroy:
2663 	irdma_cq_wq_destroy(rf, cq);
2664 cq_free_rsrc:
2665 	irdma_cq_free_rsrc(rf, iwcq);
2666 
2667 	return err_code;
2668 }
2669 
2670 /**
2671  * irdma_get_mr_access - get hw MR access permissions from IB access flags
2672  * @access: IB access flags
2673  * @hw_rev: Hardware version
2674  */
2675 static inline u16 irdma_get_mr_access(int access, u8 hw_rev)
2676 {
2677 	u16 hw_access = 0;
2678 
2679 	hw_access |= (access & IB_ACCESS_LOCAL_WRITE) ?
2680 		     IRDMA_ACCESS_FLAGS_LOCALWRITE : 0;
2681 	hw_access |= (access & IB_ACCESS_REMOTE_WRITE) ?
2682 		     IRDMA_ACCESS_FLAGS_REMOTEWRITE : 0;
2683 	hw_access |= (access & IB_ACCESS_REMOTE_READ) ?
2684 		     IRDMA_ACCESS_FLAGS_REMOTEREAD : 0;
2685 	if (hw_rev >= IRDMA_GEN_3) {
2686 		hw_access |= (access & IB_ACCESS_MW_BIND) ?
2687 			     IRDMA_ACCESS_FLAGS_BIND_WINDOW : 0;
2688 	}
2689 	hw_access |= (access & IB_ZERO_BASED) ?
2690 		     IRDMA_ACCESS_FLAGS_ZERO_BASED : 0;
2691 	hw_access |= IRDMA_ACCESS_FLAGS_LOCALREAD;
2692 
2693 	return hw_access;
2694 }
2695 
2696 /**
2697  * irdma_free_stag - free stag resource
2698  * @iwdev: irdma device
2699  * @stag: stag to free
2700  */
2701 static void irdma_free_stag(struct irdma_device *iwdev, u32 stag)
2702 {
2703 	u32 stag_idx;
2704 
2705 	stag_idx = (stag & iwdev->rf->mr_stagmask) >> IRDMA_CQPSQ_STAG_IDX_S;
2706 	irdma_free_rsrc(iwdev->rf, iwdev->rf->allocated_mrs, stag_idx);
2707 }
2708 
2709 /**
2710  * irdma_create_stag - create random stag
2711  * @iwdev: irdma device
2712  */
2713 static u32 irdma_create_stag(struct irdma_device *iwdev)
2714 {
2715 	u32 stag = 0;
2716 	u32 stag_index = 0;
2717 	u32 next_stag_index;
2718 	u32 driver_key;
2719 	u32 random;
2720 	u8 consumer_key;
2721 	int ret;
2722 
2723 	get_random_bytes(&random, sizeof(random));
2724 	consumer_key = (u8)random;
2725 
2726 	driver_key = random & ~iwdev->rf->mr_stagmask;
2727 	next_stag_index = (random & iwdev->rf->mr_stagmask) >> 8;
2728 	next_stag_index %= iwdev->rf->max_mr;
2729 
2730 	ret = irdma_alloc_rsrc(iwdev->rf, iwdev->rf->allocated_mrs,
2731 			       iwdev->rf->max_mr, &stag_index,
2732 			       &next_stag_index);
2733 	if (ret)
2734 		return stag;
2735 	stag = stag_index << IRDMA_CQPSQ_STAG_IDX_S;
2736 	stag |= driver_key;
2737 	stag += (u32)consumer_key;
2738 
2739 	return stag;
2740 }
2741 
2742 /**
2743  * irdma_next_pbl_addr - Get next pbl address
2744  * @pbl: pointer to a pble
2745  * @pinfo: info pointer
2746  * @idx: index
2747  */
2748 static inline u64 *irdma_next_pbl_addr(u64 *pbl, struct irdma_pble_info **pinfo,
2749 				       u32 *idx)
2750 {
2751 	*idx += 1;
2752 	if (!(*pinfo) || *idx != (*pinfo)->cnt)
2753 		return ++pbl;
2754 	*idx = 0;
2755 	(*pinfo)++;
2756 
2757 	return (*pinfo)->addr;
2758 }
2759 
2760 /**
2761  * irdma_copy_user_pgaddrs - copy user page address to pble's os locally
2762  * @iwmr: iwmr for IB's user page addresses
2763  * @pbl: ple pointer to save 1 level or 0 level pble
2764  * @pbl_len: Max number of PBL entries to populate
2765  * @level: indicated level 0, 1 or 2
2766  */
2767 static void irdma_copy_user_pgaddrs(struct irdma_mr *iwmr, u64 *pbl,
2768 				    u32 pbl_len, enum irdma_pble_level level)
2769 {
2770 	struct ib_umem *region = iwmr->region;
2771 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
2772 	struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
2773 	struct irdma_pble_info *pinfo;
2774 	struct ib_block_iter biter;
2775 	u32 idx = 0;
2776 
2777 	if (!pbl_len)
2778 		return;
2779 
2780 	pinfo = (level == PBLE_LEVEL_1) ? NULL : palloc->level2.leaf;
2781 
2782 	if (iwmr->type == IRDMA_MEMREG_TYPE_QP)
2783 		iwpbl->qp_mr.sq_page = sg_page(region->sgt_append.sgt.sgl);
2784 
2785 	rdma_umem_for_each_dma_block(region, &biter, iwmr->page_size) {
2786 		*pbl = rdma_block_iter_dma_address(&biter);
2787 		if (!--pbl_len)
2788 			break;
2789 		pbl = irdma_next_pbl_addr(pbl, &pinfo, &idx);
2790 	}
2791 }
2792 
2793 /**
2794  * irdma_check_mem_contiguous - check if pbls stored in arr are contiguous
2795  * @arr: lvl1 pbl array
2796  * @npages: page count
2797  * @pg_size: page size
2798  *
2799  */
2800 static bool irdma_check_mem_contiguous(u64 *arr, u32 npages, u32 pg_size)
2801 {
2802 	u32 pg_idx;
2803 
2804 	for (pg_idx = 0; pg_idx < npages; pg_idx++) {
2805 		if ((*arr + (pg_size * pg_idx)) != arr[pg_idx])
2806 			return false;
2807 	}
2808 
2809 	return true;
2810 }
2811 
2812 /**
2813  * irdma_check_mr_contiguous - check if MR is physically contiguous
2814  * @palloc: pbl allocation struct
2815  * @pg_size: page size
2816  */
2817 static bool irdma_check_mr_contiguous(struct irdma_pble_alloc *palloc,
2818 				      u32 pg_size)
2819 {
2820 	struct irdma_pble_level2 *lvl2 = &palloc->level2;
2821 	struct irdma_pble_info *leaf = lvl2->leaf;
2822 	u64 *arr = NULL;
2823 	u64 *start_addr = NULL;
2824 	int i;
2825 	bool ret;
2826 
2827 	if (palloc->level == PBLE_LEVEL_1) {
2828 		arr = palloc->level1.addr;
2829 		ret = irdma_check_mem_contiguous(arr, palloc->total_cnt,
2830 						 pg_size);
2831 		return ret;
2832 	}
2833 
2834 	start_addr = leaf->addr;
2835 
2836 	for (i = 0; i < lvl2->leaf_cnt; i++, leaf++) {
2837 		arr = leaf->addr;
2838 		if ((*start_addr + (i * pg_size * PBLE_PER_PAGE)) != *arr)
2839 			return false;
2840 		ret = irdma_check_mem_contiguous(arr, leaf->cnt, pg_size);
2841 		if (!ret)
2842 			return false;
2843 	}
2844 
2845 	return true;
2846 }
2847 
2848 /**
2849  * irdma_setup_pbles - copy user pg address to pble's
2850  * @rf: RDMA PCI function
2851  * @iwmr: mr pointer for this memory registration
2852  * @lvl: requested pble levels
2853  */
2854 static int irdma_setup_pbles(struct irdma_pci_f *rf, struct irdma_mr *iwmr,
2855 			     u8 lvl)
2856 {
2857 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
2858 	struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
2859 	struct irdma_pble_info *pinfo;
2860 	u64 *pbl;
2861 	int status;
2862 	enum irdma_pble_level level = PBLE_LEVEL_1;
2863 	u32 pbl_len;
2864 
2865 	if (lvl) {
2866 		status = irdma_get_pble(rf->pble_rsrc, palloc, iwmr->page_cnt,
2867 					lvl);
2868 		if (status)
2869 			return status;
2870 
2871 		pbl_len = palloc->total_cnt;
2872 		iwpbl->pbl_allocated = true;
2873 		level = palloc->level;
2874 		pinfo = (level == PBLE_LEVEL_1) ? &palloc->level1 :
2875 						  palloc->level2.leaf;
2876 		pbl = pinfo->addr;
2877 	} else {
2878 		pbl_len = IRDMA_MAX_SAVED_PHY_PGADDR;
2879 		pbl = iwmr->pgaddrmem;
2880 	}
2881 
2882 	irdma_copy_user_pgaddrs(iwmr, pbl, pbl_len, level);
2883 
2884 	if (lvl)
2885 		iwmr->pgaddrmem[0] = *pbl;
2886 
2887 	return 0;
2888 }
2889 
2890 /**
2891  * irdma_handle_q_mem - handle memory for qp and cq
2892  * @iwdev: irdma device
2893  * @req: information for q memory management
2894  * @iwpbl: pble struct
2895  * @lvl: pble level mask
2896  */
2897 static int irdma_handle_q_mem(struct irdma_device *iwdev,
2898 			      struct irdma_mem_reg_req *req,
2899 			      struct irdma_pbl *iwpbl, u8 lvl)
2900 {
2901 	struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
2902 	struct irdma_mr *iwmr = iwpbl->iwmr;
2903 	struct irdma_qp_mr *qpmr = &iwpbl->qp_mr;
2904 	struct irdma_cq_mr *cqmr = &iwpbl->cq_mr;
2905 	struct irdma_srq_mr *srqmr = &iwpbl->srq_mr;
2906 	struct irdma_hmc_pble *hmc_p;
2907 	u64 *arr = iwmr->pgaddrmem;
2908 	u32 pg_size, total;
2909 	int err = 0;
2910 	bool ret = true;
2911 
2912 	pg_size = iwmr->page_size;
2913 	err = irdma_setup_pbles(iwdev->rf, iwmr, lvl);
2914 	if (err)
2915 		return err;
2916 
2917 	if (lvl)
2918 		arr = palloc->level1.addr;
2919 
2920 	switch (iwmr->type) {
2921 	case IRDMA_MEMREG_TYPE_QP:
2922 		total = req->sq_pages + req->rq_pages;
2923 		hmc_p = &qpmr->sq_pbl;
2924 		qpmr->shadow = (dma_addr_t)arr[total];
2925 		/* Need to use physical address for RQ of QP
2926 		 * in case it is associated with SRQ.
2927 		 */
2928 		qpmr->rq_pa = (dma_addr_t)arr[req->sq_pages];
2929 		if (lvl) {
2930 			ret = irdma_check_mem_contiguous(arr, req->sq_pages,
2931 							 pg_size);
2932 			if (ret)
2933 				ret = irdma_check_mem_contiguous(&arr[req->sq_pages],
2934 								 req->rq_pages,
2935 								 pg_size);
2936 		}
2937 
2938 		if (!ret) {
2939 			hmc_p->idx = palloc->level1.idx;
2940 			hmc_p = &qpmr->rq_pbl;
2941 			hmc_p->idx = palloc->level1.idx + req->sq_pages;
2942 		} else {
2943 			hmc_p->addr = arr[0];
2944 			hmc_p = &qpmr->rq_pbl;
2945 			hmc_p->addr = arr[req->sq_pages];
2946 		}
2947 		break;
2948 	case IRDMA_MEMREG_TYPE_SRQ:
2949 		hmc_p = &srqmr->srq_pbl;
2950 		srqmr->shadow = (dma_addr_t)arr[req->rq_pages];
2951 		if (lvl)
2952 			ret = irdma_check_mem_contiguous(arr, req->rq_pages,
2953 							 pg_size);
2954 
2955 		if (!ret)
2956 			hmc_p->idx = palloc->level1.idx;
2957 		else
2958 			hmc_p->addr = arr[0];
2959 	break;
2960 	case IRDMA_MEMREG_TYPE_CQ:
2961 		hmc_p = &cqmr->cq_pbl;
2962 
2963 		if (!(iwdev->rf->sc_dev.hw_attrs.uk_attrs.feature_flags &
2964 		      IRDMA_FEATURE_CQ_RESIZE))
2965 			cqmr->shadow = (dma_addr_t)arr[req->cq_pages];
2966 
2967 		if (lvl)
2968 			ret = irdma_check_mem_contiguous(arr, req->cq_pages,
2969 							 pg_size);
2970 
2971 		if (!ret)
2972 			hmc_p->idx = palloc->level1.idx;
2973 		else
2974 			hmc_p->addr = arr[0];
2975 	break;
2976 	default:
2977 		ibdev_dbg(&iwdev->ibdev, "VERBS: MR type error\n");
2978 		err = -EINVAL;
2979 	}
2980 
2981 	if (lvl && ret) {
2982 		irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
2983 		iwpbl->pbl_allocated = false;
2984 	}
2985 
2986 	return err;
2987 }
2988 
2989 /**
2990  * irdma_hw_alloc_mw - create the hw memory window
2991  * @iwdev: irdma device
2992  * @iwmr: pointer to memory window info
2993  */
2994 static int irdma_hw_alloc_mw(struct irdma_device *iwdev, struct irdma_mr *iwmr)
2995 {
2996 	struct irdma_mw_alloc_info *info;
2997 	struct irdma_pd *iwpd = to_iwpd(iwmr->ibmr.pd);
2998 	struct irdma_cqp_request *cqp_request;
2999 	struct cqp_cmds_info *cqp_info;
3000 	int status;
3001 
3002 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
3003 	if (!cqp_request)
3004 		return -ENOMEM;
3005 
3006 	cqp_info = &cqp_request->info;
3007 	info = &cqp_info->in.u.mw_alloc.info;
3008 	memset(info, 0, sizeof(*info));
3009 	if (iwmr->ibmw.type == IB_MW_TYPE_1)
3010 		info->mw_wide = true;
3011 
3012 	info->page_size = PAGE_SIZE;
3013 	info->mw_stag_index = iwmr->stag >> IRDMA_CQPSQ_STAG_IDX_S;
3014 	info->pd_id = iwpd->sc_pd.pd_id;
3015 	info->remote_access = true;
3016 	cqp_info->cqp_cmd = IRDMA_OP_MW_ALLOC;
3017 	cqp_info->post_sq = 1;
3018 	cqp_info->in.u.mw_alloc.dev = &iwdev->rf->sc_dev;
3019 	cqp_info->in.u.mw_alloc.scratch = (uintptr_t)cqp_request;
3020 	status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
3021 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
3022 
3023 	return status;
3024 }
3025 
3026 /**
3027  * irdma_alloc_mw - Allocate memory window
3028  * @ibmw: Memory Window
3029  * @udata: user data pointer
3030  */
3031 static int irdma_alloc_mw(struct ib_mw *ibmw, struct ib_udata *udata)
3032 {
3033 	struct irdma_device *iwdev = to_iwdev(ibmw->device);
3034 	struct irdma_mr *iwmr = to_iwmw(ibmw);
3035 	int err_code;
3036 	u32 stag;
3037 
3038 	stag = irdma_create_stag(iwdev);
3039 	if (!stag)
3040 		return -ENOMEM;
3041 
3042 	iwmr->stag = stag;
3043 	ibmw->rkey = stag;
3044 
3045 	err_code = irdma_hw_alloc_mw(iwdev, iwmr);
3046 	if (err_code) {
3047 		irdma_free_stag(iwdev, stag);
3048 		return err_code;
3049 	}
3050 
3051 	return 0;
3052 }
3053 
3054 /**
3055  * irdma_dealloc_mw - Dealloc memory window
3056  * @ibmw: memory window structure.
3057  */
3058 static int irdma_dealloc_mw(struct ib_mw *ibmw)
3059 {
3060 	struct ib_pd *ibpd = ibmw->pd;
3061 	struct irdma_pd *iwpd = to_iwpd(ibpd);
3062 	struct irdma_mr *iwmr = to_iwmr((struct ib_mr *)ibmw);
3063 	struct irdma_device *iwdev = to_iwdev(ibmw->device);
3064 	struct irdma_cqp_request *cqp_request;
3065 	struct cqp_cmds_info *cqp_info;
3066 	struct irdma_dealloc_stag_info *info;
3067 
3068 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
3069 	if (!cqp_request)
3070 		return -ENOMEM;
3071 
3072 	cqp_info = &cqp_request->info;
3073 	info = &cqp_info->in.u.dealloc_stag.info;
3074 	memset(info, 0, sizeof(*info));
3075 	info->pd_id = iwpd->sc_pd.pd_id;
3076 	info->stag_idx = ibmw->rkey >> IRDMA_CQPSQ_STAG_IDX_S;
3077 	info->mr = false;
3078 	cqp_info->cqp_cmd = IRDMA_OP_DEALLOC_STAG;
3079 	cqp_info->post_sq = 1;
3080 	cqp_info->in.u.dealloc_stag.dev = &iwdev->rf->sc_dev;
3081 	cqp_info->in.u.dealloc_stag.scratch = (uintptr_t)cqp_request;
3082 	irdma_handle_cqp_op(iwdev->rf, cqp_request);
3083 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
3084 	irdma_free_stag(iwdev, iwmr->stag);
3085 
3086 	return 0;
3087 }
3088 
3089 /**
3090  * irdma_hw_alloc_stag - cqp command to allocate stag
3091  * @iwdev: irdma device
3092  * @iwmr: irdma mr pointer
3093  */
3094 static int irdma_hw_alloc_stag(struct irdma_device *iwdev,
3095 			       struct irdma_mr *iwmr)
3096 {
3097 	struct irdma_allocate_stag_info *info;
3098 	struct ib_pd *pd = iwmr->ibmr.pd;
3099 	struct irdma_pd *iwpd = to_iwpd(pd);
3100 	int status;
3101 	struct irdma_cqp_request *cqp_request;
3102 	struct cqp_cmds_info *cqp_info;
3103 
3104 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
3105 	if (!cqp_request)
3106 		return -ENOMEM;
3107 
3108 	cqp_info = &cqp_request->info;
3109 	info = &cqp_info->in.u.alloc_stag.info;
3110 	info->page_size = PAGE_SIZE;
3111 	info->stag_idx = iwmr->stag >> IRDMA_CQPSQ_STAG_IDX_S;
3112 	info->pd_id = iwpd->sc_pd.pd_id;
3113 	info->total_len = iwmr->len;
3114 	info->remote_access = true;
3115 	cqp_info->cqp_cmd = IRDMA_OP_ALLOC_STAG;
3116 	cqp_info->post_sq = 1;
3117 	cqp_info->in.u.alloc_stag.dev = &iwdev->rf->sc_dev;
3118 	cqp_info->in.u.alloc_stag.scratch = (uintptr_t)cqp_request;
3119 	status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
3120 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
3121 	if (status)
3122 		return status;
3123 
3124 	iwmr->is_hwreg = true;
3125 	return 0;
3126 }
3127 
3128 /**
3129  * irdma_alloc_mr - register stag for fast memory registration
3130  * @pd: ibpd pointer
3131  * @mr_type: memory for stag registrion
3132  * @max_num_sg: man number of pages
3133  */
3134 static struct ib_mr *irdma_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
3135 				    u32 max_num_sg)
3136 {
3137 	struct irdma_device *iwdev = to_iwdev(pd->device);
3138 	struct irdma_pble_alloc *palloc;
3139 	struct irdma_pbl *iwpbl;
3140 	struct irdma_mr *iwmr;
3141 	u32 stag;
3142 	int err_code;
3143 
3144 	iwmr = kzalloc_obj(*iwmr);
3145 	if (!iwmr)
3146 		return ERR_PTR(-ENOMEM);
3147 
3148 	stag = irdma_create_stag(iwdev);
3149 	if (!stag) {
3150 		err_code = -ENOMEM;
3151 		goto err;
3152 	}
3153 
3154 	iwmr->stag = stag;
3155 	iwmr->ibmr.rkey = stag;
3156 	iwmr->ibmr.lkey = stag;
3157 	iwmr->ibmr.pd = pd;
3158 	iwmr->ibmr.device = pd->device;
3159 	iwpbl = &iwmr->iwpbl;
3160 	iwpbl->iwmr = iwmr;
3161 	iwmr->type = IRDMA_MEMREG_TYPE_MEM;
3162 	palloc = &iwpbl->pble_alloc;
3163 	iwmr->page_cnt = max_num_sg;
3164 	/* Use system PAGE_SIZE as the sg page sizes are unknown at this point */
3165 	iwmr->len = max_num_sg * PAGE_SIZE;
3166 	err_code = irdma_get_pble(iwdev->rf->pble_rsrc, palloc, iwmr->page_cnt,
3167 				  false);
3168 	if (err_code)
3169 		goto err_get_pble;
3170 
3171 	err_code = irdma_hw_alloc_stag(iwdev, iwmr);
3172 	if (err_code)
3173 		goto err_alloc_stag;
3174 
3175 	iwpbl->pbl_allocated = true;
3176 
3177 	return &iwmr->ibmr;
3178 err_alloc_stag:
3179 	irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
3180 err_get_pble:
3181 	irdma_free_stag(iwdev, stag);
3182 err:
3183 	kfree(iwmr);
3184 
3185 	return ERR_PTR(err_code);
3186 }
3187 
3188 /**
3189  * irdma_set_page - populate pbl list for fmr
3190  * @ibmr: ib mem to access iwarp mr pointer
3191  * @addr: page dma address fro pbl list
3192  */
3193 static int irdma_set_page(struct ib_mr *ibmr, u64 addr)
3194 {
3195 	struct irdma_mr *iwmr = to_iwmr(ibmr);
3196 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
3197 	struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
3198 	u64 *pbl;
3199 
3200 	if (unlikely(iwmr->npages == iwmr->page_cnt))
3201 		return -ENOMEM;
3202 
3203 	if (palloc->level == PBLE_LEVEL_2) {
3204 		struct irdma_pble_info *palloc_info =
3205 			palloc->level2.leaf + (iwmr->npages >> PBLE_512_SHIFT);
3206 
3207 		palloc_info->addr[iwmr->npages & (PBLE_PER_PAGE - 1)] = addr;
3208 	} else {
3209 		pbl = palloc->level1.addr;
3210 		pbl[iwmr->npages] = addr;
3211 	}
3212 	iwmr->npages++;
3213 
3214 	return 0;
3215 }
3216 
3217 /**
3218  * irdma_map_mr_sg - map of sg list for fmr
3219  * @ibmr: ib mem to access iwarp mr pointer
3220  * @sg: scatter gather list
3221  * @sg_nents: number of sg pages
3222  * @sg_offset: scatter gather list for fmr
3223  */
3224 static int irdma_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
3225 			   int sg_nents, unsigned int *sg_offset)
3226 {
3227 	struct irdma_mr *iwmr = to_iwmr(ibmr);
3228 
3229 	iwmr->npages = 0;
3230 
3231 	return ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, irdma_set_page);
3232 }
3233 
3234 /**
3235  * irdma_hwreg_mr - send cqp command for memory registration
3236  * @iwdev: irdma device
3237  * @iwmr: irdma mr pointer
3238  * @access: access for MR
3239  */
3240 static int irdma_hwreg_mr(struct irdma_device *iwdev, struct irdma_mr *iwmr,
3241 			  u16 access)
3242 {
3243 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
3244 	struct irdma_reg_ns_stag_info *stag_info;
3245 	struct ib_pd *pd = iwmr->ibmr.pd;
3246 	struct irdma_pd *iwpd = to_iwpd(pd);
3247 	struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
3248 	struct irdma_cqp_request *cqp_request;
3249 	struct cqp_cmds_info *cqp_info;
3250 	int ret;
3251 
3252 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
3253 	if (!cqp_request)
3254 		return -ENOMEM;
3255 
3256 	cqp_info = &cqp_request->info;
3257 	stag_info = &cqp_info->in.u.mr_reg_non_shared.info;
3258 	stag_info->va = iwpbl->user_base;
3259 	stag_info->stag_idx = iwmr->stag >> IRDMA_CQPSQ_STAG_IDX_S;
3260 	stag_info->stag_key = (u8)iwmr->stag;
3261 	stag_info->total_len = iwmr->len;
3262 	stag_info->access_rights = irdma_get_mr_access(access,
3263 						       iwdev->rf->sc_dev.hw_attrs.uk_attrs.hw_rev);
3264 	if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_ATOMIC_OPS)
3265 		stag_info->remote_atomics_en = (access & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0;
3266 	stag_info->pd_id = iwpd->sc_pd.pd_id;
3267 	stag_info->all_memory = iwmr->dma_mr;
3268 	if (stag_info->access_rights & IRDMA_ACCESS_FLAGS_ZERO_BASED)
3269 		stag_info->addr_type = IRDMA_ADDR_TYPE_ZERO_BASED;
3270 	else
3271 		stag_info->addr_type = IRDMA_ADDR_TYPE_VA_BASED;
3272 	stag_info->page_size = iwmr->page_size;
3273 
3274 	if (iwpbl->pbl_allocated) {
3275 		if (palloc->level == PBLE_LEVEL_1) {
3276 			stag_info->first_pm_pbl_index = palloc->level1.idx;
3277 			stag_info->chunk_size = 1;
3278 		} else {
3279 			stag_info->first_pm_pbl_index = palloc->level2.root.idx;
3280 			stag_info->chunk_size = 3;
3281 		}
3282 	} else {
3283 		stag_info->reg_addr_pa = iwmr->pgaddrmem[0];
3284 	}
3285 
3286 	cqp_info->cqp_cmd = IRDMA_OP_MR_REG_NON_SHARED;
3287 	cqp_info->post_sq = 1;
3288 	cqp_info->in.u.mr_reg_non_shared.dev = &iwdev->rf->sc_dev;
3289 	cqp_info->in.u.mr_reg_non_shared.scratch = (uintptr_t)cqp_request;
3290 	ret = irdma_handle_cqp_op(iwdev->rf, cqp_request);
3291 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
3292 
3293 	if (!ret)
3294 		iwmr->is_hwreg = true;
3295 
3296 	return ret;
3297 }
3298 
3299 static int irdma_reg_user_mr_type_mem(struct irdma_mr *iwmr, int access,
3300 				      bool create_stag)
3301 {
3302 	struct irdma_device *iwdev = to_iwdev(iwmr->ibmr.device);
3303 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
3304 	u32 stag = 0;
3305 	u8 lvl;
3306 	int err;
3307 
3308 	lvl = iwmr->page_cnt != 1 ? PBLE_LEVEL_1 | PBLE_LEVEL_2 : PBLE_LEVEL_0;
3309 	iwmr->access = access;
3310 
3311 	err = irdma_setup_pbles(iwdev->rf, iwmr, lvl);
3312 	if (err)
3313 		return err;
3314 
3315 	if (lvl) {
3316 		err = irdma_check_mr_contiguous(&iwpbl->pble_alloc,
3317 						iwmr->page_size);
3318 		if (err) {
3319 			irdma_free_pble(iwdev->rf->pble_rsrc, &iwpbl->pble_alloc);
3320 			iwpbl->pbl_allocated = false;
3321 		}
3322 	}
3323 
3324 	if (create_stag) {
3325 		stag = irdma_create_stag(iwdev);
3326 		if (!stag) {
3327 			err = -ENOMEM;
3328 			goto free_pble;
3329 		}
3330 
3331 		iwmr->stag = stag;
3332 		iwmr->ibmr.rkey = stag;
3333 		iwmr->ibmr.lkey = stag;
3334 	}
3335 
3336 	err = irdma_hwreg_mr(iwdev, iwmr, access);
3337 	if (err)
3338 		goto err_hwreg;
3339 
3340 	return 0;
3341 
3342 err_hwreg:
3343 	if (stag)
3344 		irdma_free_stag(iwdev, stag);
3345 
3346 free_pble:
3347 	if (iwpbl->pble_alloc.level != PBLE_LEVEL_0 && iwpbl->pbl_allocated)
3348 		irdma_free_pble(iwdev->rf->pble_rsrc, &iwpbl->pble_alloc);
3349 
3350 	return err;
3351 }
3352 
3353 static struct irdma_mr *irdma_alloc_iwmr(struct ib_umem *region,
3354 					 struct ib_pd *pd, u64 virt,
3355 					 enum irdma_memreg_type reg_type)
3356 {
3357 	struct irdma_device *iwdev = to_iwdev(pd->device);
3358 	struct irdma_pbl *iwpbl;
3359 	struct irdma_mr *iwmr;
3360 	unsigned long pgsz_bitmap;
3361 
3362 	iwmr = kzalloc_obj(*iwmr);
3363 	if (!iwmr)
3364 		return ERR_PTR(-ENOMEM);
3365 
3366 	iwpbl = &iwmr->iwpbl;
3367 	iwpbl->iwmr = iwmr;
3368 	iwmr->region = region;
3369 	iwmr->ibmr.pd = pd;
3370 	iwmr->ibmr.device = pd->device;
3371 	iwmr->ibmr.iova = virt;
3372 	iwmr->type = reg_type;
3373 
3374 	pgsz_bitmap = (reg_type == IRDMA_MEMREG_TYPE_MEM) ?
3375 		iwdev->rf->sc_dev.hw_attrs.page_size_cap : SZ_4K;
3376 
3377 	iwmr->page_size = ib_umem_find_best_pgsz(region, pgsz_bitmap, virt);
3378 	if (unlikely(!iwmr->page_size)) {
3379 		kfree(iwmr);
3380 		return ERR_PTR(-EOPNOTSUPP);
3381 	}
3382 
3383 	iwmr->len = region->length;
3384 	iwpbl->user_base = virt;
3385 	iwmr->page_cnt = ib_umem_num_dma_blocks(region, iwmr->page_size);
3386 
3387 	return iwmr;
3388 }
3389 
3390 static void irdma_free_iwmr(struct irdma_mr *iwmr)
3391 {
3392 	kfree(iwmr);
3393 }
3394 
3395 static int irdma_reg_user_mr_type_qp(struct irdma_mem_reg_req req,
3396 				     struct ib_udata *udata,
3397 				     struct irdma_mr *iwmr)
3398 {
3399 	struct irdma_device *iwdev = to_iwdev(iwmr->ibmr.device);
3400 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
3401 	struct irdma_ucontext *ucontext = NULL;
3402 	unsigned long flags;
3403 	u32 total;
3404 	int err;
3405 	u8 lvl;
3406 
3407 	/* iWarp: Catch page not starting on OS page boundary */
3408 	if (!rdma_protocol_roce(&iwdev->ibdev, 1) &&
3409 	    ib_umem_offset(iwmr->region))
3410 		return -EINVAL;
3411 
3412 	total = req.sq_pages + req.rq_pages + 1;
3413 	if (total > iwmr->page_cnt)
3414 		return -EINVAL;
3415 
3416 	total = req.sq_pages + req.rq_pages;
3417 	lvl = total > 2 ? PBLE_LEVEL_1 : PBLE_LEVEL_0;
3418 	err = irdma_handle_q_mem(iwdev, &req, iwpbl, lvl);
3419 	if (err)
3420 		return err;
3421 
3422 	ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext,
3423 					     ibucontext);
3424 	spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags);
3425 	list_add_tail(&iwpbl->list, &ucontext->qp_reg_mem_list);
3426 	iwpbl->on_list = true;
3427 	spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags);
3428 
3429 	return 0;
3430 }
3431 
3432 static int irdma_reg_user_mr_type_srq(struct irdma_mem_reg_req req,
3433 				      struct ib_udata *udata,
3434 				      struct irdma_mr *iwmr)
3435 {
3436 	struct irdma_device *iwdev = to_iwdev(iwmr->ibmr.device);
3437 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
3438 	struct irdma_ucontext *ucontext;
3439 	unsigned long flags;
3440 	u32 total;
3441 	int err;
3442 	u8 lvl;
3443 
3444 	total = req.rq_pages + IRDMA_SHADOW_PGCNT;
3445 	if (total > iwmr->page_cnt)
3446 		return -EINVAL;
3447 
3448 	lvl = req.rq_pages > 1 ? PBLE_LEVEL_1 : PBLE_LEVEL_0;
3449 	err = irdma_handle_q_mem(iwdev, &req, iwpbl, lvl);
3450 	if (err)
3451 		return err;
3452 
3453 	ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext,
3454 					     ibucontext);
3455 	spin_lock_irqsave(&ucontext->srq_reg_mem_list_lock, flags);
3456 	list_add_tail(&iwpbl->list, &ucontext->srq_reg_mem_list);
3457 	iwpbl->on_list = true;
3458 	spin_unlock_irqrestore(&ucontext->srq_reg_mem_list_lock, flags);
3459 
3460 	return 0;
3461 }
3462 
3463 static int irdma_reg_user_mr_type_cq(struct irdma_mem_reg_req req,
3464 				     struct ib_udata *udata,
3465 				     struct irdma_mr *iwmr)
3466 {
3467 	struct irdma_device *iwdev = to_iwdev(iwmr->ibmr.device);
3468 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
3469 	struct irdma_ucontext *ucontext = NULL;
3470 	u8 shadow_pgcnt = 1;
3471 	unsigned long flags;
3472 	u32 total;
3473 	int err;
3474 	u8 lvl;
3475 
3476 	if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_CQ_RESIZE)
3477 		shadow_pgcnt = 0;
3478 	total = req.cq_pages + shadow_pgcnt;
3479 	if (total > iwmr->page_cnt)
3480 		return -EINVAL;
3481 
3482 	lvl = req.cq_pages > 1 ? PBLE_LEVEL_1 : PBLE_LEVEL_0;
3483 	err = irdma_handle_q_mem(iwdev, &req, iwpbl, lvl);
3484 	if (err)
3485 		return err;
3486 
3487 	ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext,
3488 					     ibucontext);
3489 	spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
3490 	list_add_tail(&iwpbl->list, &ucontext->cq_reg_mem_list);
3491 	iwpbl->on_list = true;
3492 	spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
3493 
3494 	return 0;
3495 }
3496 
3497 /**
3498  * irdma_reg_user_mr - Register a user memory region
3499  * @pd: ptr of pd
3500  * @start: virtual start address
3501  * @len: length of mr
3502  * @virt: virtual address
3503  * @access: access of mr
3504  * @dmah: dma handle
3505  * @udata: user data
3506  */
3507 static struct ib_mr *irdma_reg_user_mr(struct ib_pd *pd, u64 start, u64 len,
3508 				       u64 virt, int access,
3509 				       struct ib_dmah *dmah,
3510 				       struct ib_udata *udata)
3511 {
3512 #define IRDMA_MEM_REG_MIN_REQ_LEN offsetofend(struct irdma_mem_reg_req, sq_pages)
3513 	struct irdma_device *iwdev = to_iwdev(pd->device);
3514 	struct irdma_mem_reg_req req = {};
3515 	struct ib_umem *region = NULL;
3516 	struct irdma_mr *iwmr = NULL;
3517 	int err;
3518 
3519 	if (dmah)
3520 		return ERR_PTR(-EOPNOTSUPP);
3521 
3522 	if (len > iwdev->rf->sc_dev.hw_attrs.max_mr_size)
3523 		return ERR_PTR(-EINVAL);
3524 
3525 	if (udata->inlen < IRDMA_MEM_REG_MIN_REQ_LEN)
3526 		return ERR_PTR(-EINVAL);
3527 
3528 	region = ib_umem_get_va(pd->device, start, len, access);
3529 
3530 	if (IS_ERR(region)) {
3531 		ibdev_dbg(&iwdev->ibdev,
3532 			  "VERBS: Failed to create ib_umem region\n");
3533 		return (struct ib_mr *)region;
3534 	}
3535 
3536 	if (ib_copy_from_udata(&req, udata, min(sizeof(req), udata->inlen))) {
3537 		ib_umem_release(region);
3538 		return ERR_PTR(-EFAULT);
3539 	}
3540 
3541 	iwmr = irdma_alloc_iwmr(region, pd, virt, req.reg_type);
3542 	if (IS_ERR(iwmr)) {
3543 		ib_umem_release(region);
3544 		return (struct ib_mr *)iwmr;
3545 	}
3546 
3547 	switch (req.reg_type) {
3548 	case IRDMA_MEMREG_TYPE_QP:
3549 		err = irdma_reg_user_mr_type_qp(req, udata, iwmr);
3550 		if (err)
3551 			goto error;
3552 
3553 		break;
3554 	case IRDMA_MEMREG_TYPE_SRQ:
3555 		err = irdma_reg_user_mr_type_srq(req, udata, iwmr);
3556 		if (err)
3557 			goto error;
3558 
3559 		break;
3560 	case IRDMA_MEMREG_TYPE_CQ:
3561 		err = irdma_reg_user_mr_type_cq(req, udata, iwmr);
3562 		if (err)
3563 			goto error;
3564 		break;
3565 	case IRDMA_MEMREG_TYPE_MEM:
3566 		err = irdma_reg_user_mr_type_mem(iwmr, access, true);
3567 		if (err)
3568 			goto error;
3569 
3570 		break;
3571 	default:
3572 		err = -EINVAL;
3573 		goto error;
3574 	}
3575 
3576 	return &iwmr->ibmr;
3577 error:
3578 	ib_umem_release(region);
3579 	irdma_free_iwmr(iwmr);
3580 
3581 	return ERR_PTR(err);
3582 }
3583 
3584 static int irdma_hwdereg_mr(struct ib_mr *ib_mr);
3585 
3586 static void irdma_umem_dmabuf_revoke(void *priv)
3587 {
3588 	/* priv is guaranteed to be valid any time this callback is invoked
3589 	 * because we do not set the callback until after successful iwmr
3590 	 * allocation and initialization.
3591 	 */
3592 	struct irdma_mr *iwmr = priv;
3593 	int err;
3594 
3595 	/* Invalidate the key in hardware. This does not actually release the
3596 	 * key for potential reuse - that only occurs when the region is fully
3597 	 * deregistered.
3598 	 *
3599 	 * The irdma_hwdereg_mr call is a no-op if the region is not currently
3600 	 * registered with hardware.
3601 	 */
3602 	err = irdma_hwdereg_mr(&iwmr->ibmr);
3603 	if (err) {
3604 		struct irdma_device *iwdev = to_iwdev(iwmr->ibmr.device);
3605 
3606 		ibdev_err(&iwdev->ibdev, "dmabuf mr revoke failed %d", err);
3607 		if (!iwdev->rf->reset) {
3608 			iwdev->rf->reset = true;
3609 			iwdev->rf->gen_ops.request_reset(iwdev->rf);
3610 		}
3611 	}
3612 }
3613 
3614 static struct ib_mr *irdma_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start,
3615 					      u64 len, u64 virt,
3616 					      int fd, int access,
3617 					      struct ib_dmah *dmah,
3618 					      struct uverbs_attr_bundle *attrs)
3619 {
3620 	struct irdma_device *iwdev = to_iwdev(pd->device);
3621 	struct ib_umem_dmabuf *umem_dmabuf;
3622 	struct irdma_mr *iwmr;
3623 	int err;
3624 
3625 	if (dmah)
3626 		return ERR_PTR(-EOPNOTSUPP);
3627 
3628 	if (len > iwdev->rf->sc_dev.hw_attrs.max_mr_size)
3629 		return ERR_PTR(-EINVAL);
3630 
3631 	umem_dmabuf =
3632 		ib_umem_dmabuf_get_pinned_revocable_and_lock(pd->device, start,
3633 							     len, fd, access);
3634 	if (IS_ERR(umem_dmabuf)) {
3635 		ibdev_dbg(&iwdev->ibdev, "Failed to get dmabuf umem[%pe]\n",
3636 			  umem_dmabuf);
3637 		return ERR_CAST(umem_dmabuf);
3638 	}
3639 
3640 	iwmr = irdma_alloc_iwmr(&umem_dmabuf->umem, pd, virt, IRDMA_MEMREG_TYPE_MEM);
3641 	if (IS_ERR(iwmr)) {
3642 		err = PTR_ERR(iwmr);
3643 		goto err_release;
3644 	}
3645 
3646 	err = irdma_reg_user_mr_type_mem(iwmr, access, true);
3647 	if (err)
3648 		goto err_iwmr;
3649 
3650 	ib_umem_dmabuf_set_revoke_locked(umem_dmabuf, irdma_umem_dmabuf_revoke,
3651 					 iwmr);
3652 	ib_umem_dmabuf_revoke_unlock(umem_dmabuf);
3653 	return &iwmr->ibmr;
3654 
3655 err_iwmr:
3656 	irdma_free_iwmr(iwmr);
3657 
3658 err_release:
3659 	ib_umem_dmabuf_revoke_unlock(umem_dmabuf);
3660 
3661 	/* Will result in a call to revoke, but driver callback is not set and
3662 	 * is therefore skipped.
3663 	 */
3664 	ib_umem_release(&umem_dmabuf->umem);
3665 
3666 	return ERR_PTR(err);
3667 }
3668 
3669 static int irdma_hwdereg_mr(struct ib_mr *ib_mr)
3670 {
3671 	struct irdma_device *iwdev = to_iwdev(ib_mr->device);
3672 	struct irdma_mr *iwmr = to_iwmr(ib_mr);
3673 	struct irdma_pd *iwpd = to_iwpd(ib_mr->pd);
3674 	struct irdma_dealloc_stag_info *info;
3675 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
3676 	struct irdma_cqp_request *cqp_request;
3677 	struct cqp_cmds_info *cqp_info;
3678 	int status;
3679 
3680 	/* Skip HW MR de-register when it is already de-registered
3681 	 * during an MR re-reregister and the re-registration fails
3682 	 */
3683 	if (!iwmr->is_hwreg)
3684 		return 0;
3685 
3686 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
3687 	if (!cqp_request)
3688 		return -ENOMEM;
3689 
3690 	cqp_info = &cqp_request->info;
3691 	info = &cqp_info->in.u.dealloc_stag.info;
3692 	info->pd_id = iwpd->sc_pd.pd_id;
3693 	info->stag_idx = ib_mr->rkey >> IRDMA_CQPSQ_STAG_IDX_S;
3694 	info->mr = true;
3695 	if (iwpbl->pbl_allocated)
3696 		info->dealloc_pbl = true;
3697 
3698 	cqp_info->cqp_cmd = IRDMA_OP_DEALLOC_STAG;
3699 	cqp_info->post_sq = 1;
3700 	cqp_info->in.u.dealloc_stag.dev = &iwdev->rf->sc_dev;
3701 	cqp_info->in.u.dealloc_stag.scratch = (uintptr_t)cqp_request;
3702 	status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
3703 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
3704 	if (status)
3705 		return status;
3706 
3707 	iwmr->is_hwreg = false;
3708 	return 0;
3709 }
3710 
3711 /*
3712  * irdma_rereg_mr_trans - Re-register a user MR for a change translation.
3713  * @iwmr: ptr of iwmr
3714  * @start: virtual start address
3715  * @len: length of mr
3716  * @virt: virtual address
3717  *
3718  * Re-register a user memory region when a change translation is requested.
3719  * Re-register a new region while reusing the stag from the original registration.
3720  */
3721 static int irdma_rereg_mr_trans(struct irdma_mr *iwmr, u64 start, u64 len,
3722 				u64 virt)
3723 {
3724 	struct irdma_device *iwdev = to_iwdev(iwmr->ibmr.device);
3725 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
3726 	struct ib_pd *pd = iwmr->ibmr.pd;
3727 	struct ib_umem *region;
3728 	int err;
3729 
3730 	region = ib_umem_get_va(pd->device, start, len, iwmr->access);
3731 	if (IS_ERR(region))
3732 		return PTR_ERR(region);
3733 
3734 	iwmr->region = region;
3735 	iwmr->ibmr.iova = virt;
3736 	iwmr->ibmr.pd = pd;
3737 	iwmr->page_size = ib_umem_find_best_pgsz(region,
3738 				iwdev->rf->sc_dev.hw_attrs.page_size_cap,
3739 				virt);
3740 	if (unlikely(!iwmr->page_size)) {
3741 		err = -EOPNOTSUPP;
3742 		goto err;
3743 	}
3744 
3745 	iwmr->len = region->length;
3746 	iwpbl->user_base = virt;
3747 	iwmr->page_cnt = ib_umem_num_dma_blocks(region, iwmr->page_size);
3748 
3749 	err = irdma_reg_user_mr_type_mem(iwmr, iwmr->access, false);
3750 	if (err)
3751 		goto err;
3752 
3753 	return 0;
3754 
3755 err:
3756 	ib_umem_release(region);
3757 	iwmr->region = NULL;
3758 	return err;
3759 }
3760 
3761 /*
3762  *  irdma_rereg_user_mr - Re-Register a user memory region(MR)
3763  *  @ibmr: ib mem to access iwarp mr pointer
3764  *  @flags: bit mask to indicate which of the attr's of MR modified
3765  *  @start: virtual start address
3766  *  @len: length of mr
3767  *  @virt: virtual address
3768  *  @new_access: bit mask of access flags
3769  *  @new_pd: ptr of pd
3770  *  @udata: user data
3771  *
3772  *  Return:
3773  *  NULL - Success, existing MR updated
3774  *  ERR_PTR - error occurred
3775  */
3776 static struct ib_mr *irdma_rereg_user_mr(struct ib_mr *ib_mr, int flags,
3777 					 u64 start, u64 len, u64 virt,
3778 					 int new_access, struct ib_pd *new_pd,
3779 					 struct ib_udata *udata)
3780 {
3781 	struct irdma_device *iwdev = to_iwdev(ib_mr->device);
3782 	struct irdma_mr *iwmr = to_iwmr(ib_mr);
3783 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
3784 	bool dmabuf_revocable = iwmr->region && iwmr->region->is_dmabuf;
3785 	struct ib_umem_dmabuf *umem_dmabuf;
3786 	int ret;
3787 
3788 	if (len > iwdev->rf->sc_dev.hw_attrs.max_mr_size)
3789 		return ERR_PTR(-EINVAL);
3790 
3791 	if (flags & ~(IB_MR_REREG_TRANS | IB_MR_REREG_PD | IB_MR_REREG_ACCESS))
3792 		return ERR_PTR(-EOPNOTSUPP);
3793 
3794 	ret = ib_umem_check_rereg(iwmr->region, flags, new_access);
3795 	if (ret)
3796 		return ERR_PTR(ret);
3797 
3798 	if (dmabuf_revocable) {
3799 		umem_dmabuf = to_ib_umem_dmabuf(iwmr->region);
3800 
3801 		ib_umem_dmabuf_revoke_lock(umem_dmabuf);
3802 
3803 		/* If the dmabuf has been revoked, it means that the region has
3804 		 * been invalidated in HW. We must not allow it to become valid
3805 		 * again unless the user is requesting a change in translation
3806 		 * which will end up dropping the umem dmabuf and allocating an
3807 		 * entirely new umem anyway.
3808 		 */
3809 		if (umem_dmabuf->revoked && !(flags & IB_MR_REREG_TRANS)) {
3810 			ret = -EINVAL;
3811 			goto err_unlock;
3812 		}
3813 	}
3814 
3815 	ret = irdma_hwdereg_mr(ib_mr);
3816 	if (ret)
3817 		goto err_unlock;
3818 
3819 	if (flags & IB_MR_REREG_ACCESS)
3820 		iwmr->access = new_access;
3821 
3822 	if (flags & IB_MR_REREG_PD) {
3823 		iwmr->ibmr.pd = new_pd;
3824 		iwmr->ibmr.device = new_pd->device;
3825 	}
3826 
3827 	if (flags & IB_MR_REREG_TRANS) {
3828 		if (iwpbl->pbl_allocated) {
3829 			irdma_free_pble(iwdev->rf->pble_rsrc,
3830 					&iwpbl->pble_alloc);
3831 			iwpbl->pbl_allocated = false;
3832 		}
3833 
3834 		if (dmabuf_revocable) {
3835 			/* Must unlock before release to prevent deadlock */
3836 			ib_umem_dmabuf_revoke_unlock(umem_dmabuf);
3837 			dmabuf_revocable = false;
3838 		}
3839 
3840 		if (iwmr->region) {
3841 			ib_umem_release(iwmr->region);
3842 			iwmr->region = NULL;
3843 		}
3844 
3845 		ret = irdma_rereg_mr_trans(iwmr, start, len, virt);
3846 	} else {
3847 		ret = irdma_hwreg_mr(iwdev, iwmr, iwmr->access);
3848 	}
3849 
3850 err_unlock:
3851 	if (dmabuf_revocable)
3852 		ib_umem_dmabuf_revoke_unlock(umem_dmabuf);
3853 
3854 	return ret ? ERR_PTR(ret) : NULL;
3855 }
3856 
3857 /**
3858  * irdma_reg_phys_mr - register kernel physical memory
3859  * @pd: ibpd pointer
3860  * @addr: physical address of memory to register
3861  * @size: size of memory to register
3862  * @access: Access rights
3863  * @iova_start: start of virtual address for physical buffers
3864  * @dma_mr: Flag indicating whether this region is a PD DMA MR
3865  */
3866 struct ib_mr *irdma_reg_phys_mr(struct ib_pd *pd, u64 addr, u64 size, int access,
3867 				u64 *iova_start, bool dma_mr)
3868 {
3869 	struct irdma_device *iwdev = to_iwdev(pd->device);
3870 	struct irdma_pbl *iwpbl;
3871 	struct irdma_mr *iwmr;
3872 	u32 stag;
3873 	int ret;
3874 
3875 	iwmr = kzalloc_obj(*iwmr);
3876 	if (!iwmr)
3877 		return ERR_PTR(-ENOMEM);
3878 
3879 	iwmr->ibmr.pd = pd;
3880 	iwmr->ibmr.device = pd->device;
3881 	iwpbl = &iwmr->iwpbl;
3882 	iwpbl->iwmr = iwmr;
3883 	iwmr->type = IRDMA_MEMREG_TYPE_MEM;
3884 	iwmr->dma_mr = dma_mr;
3885 	iwpbl->user_base = *iova_start;
3886 	stag = irdma_create_stag(iwdev);
3887 	if (!stag) {
3888 		ret = -ENOMEM;
3889 		goto err;
3890 	}
3891 
3892 	iwmr->stag = stag;
3893 	iwmr->ibmr.iova = *iova_start;
3894 	iwmr->ibmr.rkey = stag;
3895 	iwmr->ibmr.lkey = stag;
3896 	iwmr->page_cnt = 1;
3897 	iwmr->pgaddrmem[0] = addr;
3898 	iwmr->len = size;
3899 	iwmr->page_size = SZ_4K;
3900 	ret = irdma_hwreg_mr(iwdev, iwmr, access);
3901 	if (ret) {
3902 		irdma_free_stag(iwdev, stag);
3903 		goto err;
3904 	}
3905 
3906 	return &iwmr->ibmr;
3907 
3908 err:
3909 	kfree(iwmr);
3910 
3911 	return ERR_PTR(ret);
3912 }
3913 
3914 /**
3915  * irdma_get_dma_mr - register physical mem
3916  * @pd: ptr of pd
3917  * @acc: access for memory
3918  */
3919 static struct ib_mr *irdma_get_dma_mr(struct ib_pd *pd, int acc)
3920 {
3921 	u64 kva = 0;
3922 
3923 	return irdma_reg_phys_mr(pd, 0, 0, acc, &kva, true);
3924 }
3925 
3926 /**
3927  * irdma_del_memlist - Deleting pbl list entries for CQ/QP
3928  * @iwmr: iwmr for IB's user page addresses
3929  * @ucontext: ptr to user context
3930  */
3931 static void irdma_del_memlist(struct irdma_mr *iwmr,
3932 			      struct irdma_ucontext *ucontext)
3933 {
3934 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
3935 	unsigned long flags;
3936 
3937 	switch (iwmr->type) {
3938 	case IRDMA_MEMREG_TYPE_CQ:
3939 		spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
3940 		if (iwpbl->on_list) {
3941 			iwpbl->on_list = false;
3942 			list_del(&iwpbl->list);
3943 		}
3944 		spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
3945 		break;
3946 	case IRDMA_MEMREG_TYPE_QP:
3947 		spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags);
3948 		if (iwpbl->on_list) {
3949 			iwpbl->on_list = false;
3950 			list_del(&iwpbl->list);
3951 		}
3952 		spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags);
3953 		break;
3954 	case IRDMA_MEMREG_TYPE_SRQ:
3955 		spin_lock_irqsave(&ucontext->srq_reg_mem_list_lock, flags);
3956 		if (iwpbl->on_list) {
3957 			iwpbl->on_list = false;
3958 			list_del(&iwpbl->list);
3959 		}
3960 		spin_unlock_irqrestore(&ucontext->srq_reg_mem_list_lock, flags);
3961 		break;
3962 	default:
3963 		break;
3964 	}
3965 }
3966 
3967 /**
3968  * irdma_dereg_mr - deregister mr
3969  * @ib_mr: mr ptr for dereg
3970  * @udata: user data
3971  */
3972 static int irdma_dereg_mr(struct ib_mr *ib_mr, struct ib_udata *udata)
3973 {
3974 	struct irdma_mr *iwmr = to_iwmr(ib_mr);
3975 	struct irdma_device *iwdev = to_iwdev(ib_mr->device);
3976 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
3977 	bool dmabuf_revocable = iwmr->region && iwmr->region->is_dmabuf;
3978 	int ret;
3979 
3980 	if (iwmr->type != IRDMA_MEMREG_TYPE_MEM) {
3981 		if (iwmr->region) {
3982 			struct irdma_ucontext *ucontext;
3983 
3984 			ucontext = rdma_udata_to_drv_context(udata,
3985 						struct irdma_ucontext,
3986 						ibucontext);
3987 			irdma_del_memlist(iwmr, ucontext);
3988 		}
3989 		goto done;
3990 	}
3991 
3992 	if (!dmabuf_revocable) {
3993 		ret = irdma_hwdereg_mr(ib_mr);
3994 		if (ret)
3995 			return ret;
3996 
3997 		irdma_free_stag(iwdev, iwmr->stag);
3998 	}
3999 done:
4000 	if (iwmr->region)
4001 		/* For dmabuf MRs, ib_umem_release will trigger a synchronous
4002 		 * call to the revoke callback which will perform the actual HW
4003 		 * invalidation via irdma_hwdereg_mr. We rely on this for its
4004 		 * implicit serialization w.r.t. concurrent revocations. This
4005 		 * must be done before freeing the PBLEs.
4006 		 */
4007 		ib_umem_release(iwmr->region);
4008 
4009 	if (iwpbl->pbl_allocated)
4010 		irdma_free_pble(iwdev->rf->pble_rsrc, &iwpbl->pble_alloc);
4011 
4012 	if (dmabuf_revocable)
4013 		irdma_free_stag(iwdev, iwmr->stag);
4014 
4015 	kfree(iwmr);
4016 
4017 	return 0;
4018 }
4019 
4020 /**
4021  * irdma_post_send -  kernel application wr
4022  * @ibqp: qp ptr for wr
4023  * @ib_wr: work request ptr
4024  * @bad_wr: return of bad wr if err
4025  */
4026 static int irdma_post_send(struct ib_qp *ibqp,
4027 			   const struct ib_send_wr *ib_wr,
4028 			   const struct ib_send_wr **bad_wr)
4029 {
4030 	struct irdma_qp *iwqp;
4031 	struct irdma_qp_uk *ukqp;
4032 	struct irdma_sc_dev *dev;
4033 	struct irdma_post_sq_info info;
4034 	int err = 0;
4035 	unsigned long flags;
4036 	bool inv_stag;
4037 	struct irdma_ah *ah;
4038 
4039 	iwqp = to_iwqp(ibqp);
4040 	ukqp = &iwqp->sc_qp.qp_uk;
4041 	dev = &iwqp->iwdev->rf->sc_dev;
4042 
4043 	spin_lock_irqsave(&iwqp->lock, flags);
4044 	while (ib_wr) {
4045 		memset(&info, 0, sizeof(info));
4046 		inv_stag = false;
4047 		info.wr_id = (ib_wr->wr_id);
4048 		if ((ib_wr->send_flags & IB_SEND_SIGNALED) || iwqp->sig_all)
4049 			info.signaled = true;
4050 		if (ib_wr->send_flags & IB_SEND_FENCE)
4051 			info.read_fence = true;
4052 		switch (ib_wr->opcode) {
4053 		case IB_WR_ATOMIC_CMP_AND_SWP:
4054 			if (unlikely(!(dev->hw_attrs.uk_attrs.feature_flags &
4055 				       IRDMA_FEATURE_ATOMIC_OPS))) {
4056 				err = -EINVAL;
4057 				break;
4058 			}
4059 			info.op_type = IRDMA_OP_TYPE_ATOMIC_COMPARE_AND_SWAP;
4060 			info.op.atomic_compare_swap.tagged_offset = ib_wr->sg_list[0].addr;
4061 			info.op.atomic_compare_swap.remote_tagged_offset =
4062 				atomic_wr(ib_wr)->remote_addr;
4063 			info.op.atomic_compare_swap.swap_data_bytes = atomic_wr(ib_wr)->swap;
4064 			info.op.atomic_compare_swap.compare_data_bytes =
4065 				atomic_wr(ib_wr)->compare_add;
4066 			info.op.atomic_compare_swap.stag = ib_wr->sg_list[0].lkey;
4067 			info.op.atomic_compare_swap.remote_stag = atomic_wr(ib_wr)->rkey;
4068 			err = irdma_uk_atomic_compare_swap(ukqp, &info, false);
4069 			break;
4070 		case IB_WR_ATOMIC_FETCH_AND_ADD:
4071 			if (unlikely(!(dev->hw_attrs.uk_attrs.feature_flags &
4072 				       IRDMA_FEATURE_ATOMIC_OPS))) {
4073 				err = -EINVAL;
4074 				break;
4075 			}
4076 			info.op_type = IRDMA_OP_TYPE_ATOMIC_FETCH_AND_ADD;
4077 			info.op.atomic_fetch_add.tagged_offset = ib_wr->sg_list[0].addr;
4078 			info.op.atomic_fetch_add.remote_tagged_offset =
4079 				atomic_wr(ib_wr)->remote_addr;
4080 			info.op.atomic_fetch_add.fetch_add_data_bytes =
4081 				atomic_wr(ib_wr)->compare_add;
4082 			info.op.atomic_fetch_add.stag = ib_wr->sg_list[0].lkey;
4083 			info.op.atomic_fetch_add.remote_stag =
4084 				atomic_wr(ib_wr)->rkey;
4085 			err = irdma_uk_atomic_fetch_add(ukqp, &info, false);
4086 			break;
4087 		case IB_WR_SEND_WITH_IMM:
4088 			if (ukqp->qp_caps & IRDMA_SEND_WITH_IMM) {
4089 				info.imm_data_valid = true;
4090 				info.imm_data = ntohl(ib_wr->ex.imm_data);
4091 			} else {
4092 				err = -EINVAL;
4093 				break;
4094 			}
4095 			fallthrough;
4096 		case IB_WR_SEND:
4097 		case IB_WR_SEND_WITH_INV:
4098 			if (ib_wr->opcode == IB_WR_SEND ||
4099 			    ib_wr->opcode == IB_WR_SEND_WITH_IMM) {
4100 				if (ib_wr->send_flags & IB_SEND_SOLICITED)
4101 					info.op_type = IRDMA_OP_TYPE_SEND_SOL;
4102 				else
4103 					info.op_type = IRDMA_OP_TYPE_SEND;
4104 			} else {
4105 				if (ib_wr->send_flags & IB_SEND_SOLICITED)
4106 					info.op_type = IRDMA_OP_TYPE_SEND_SOL_INV;
4107 				else
4108 					info.op_type = IRDMA_OP_TYPE_SEND_INV;
4109 				info.stag_to_inv = ib_wr->ex.invalidate_rkey;
4110 			}
4111 
4112 			info.op.send.num_sges = ib_wr->num_sge;
4113 			info.op.send.sg_list = ib_wr->sg_list;
4114 			if (iwqp->ibqp.qp_type == IB_QPT_UD ||
4115 			    iwqp->ibqp.qp_type == IB_QPT_GSI) {
4116 				ah = to_iwah(ud_wr(ib_wr)->ah);
4117 				info.op.send.ah_id = ah->sc_ah.ah_info.ah_idx;
4118 				info.op.send.qkey = ud_wr(ib_wr)->remote_qkey;
4119 				info.op.send.dest_qp = ud_wr(ib_wr)->remote_qpn;
4120 			}
4121 
4122 			if (ib_wr->send_flags & IB_SEND_INLINE)
4123 				err = irdma_uk_inline_send(ukqp, &info, false);
4124 			else
4125 				err = irdma_uk_send(ukqp, &info, false);
4126 			break;
4127 		case IB_WR_RDMA_WRITE_WITH_IMM:
4128 			if (ukqp->qp_caps & IRDMA_WRITE_WITH_IMM) {
4129 				info.imm_data_valid = true;
4130 				info.imm_data = ntohl(ib_wr->ex.imm_data);
4131 			} else {
4132 				err = -EINVAL;
4133 				break;
4134 			}
4135 			fallthrough;
4136 		case IB_WR_RDMA_WRITE:
4137 			if (ib_wr->send_flags & IB_SEND_SOLICITED)
4138 				info.op_type = IRDMA_OP_TYPE_RDMA_WRITE_SOL;
4139 			else
4140 				info.op_type = IRDMA_OP_TYPE_RDMA_WRITE;
4141 
4142 			info.op.rdma_write.num_lo_sges = ib_wr->num_sge;
4143 			info.op.rdma_write.lo_sg_list = ib_wr->sg_list;
4144 			info.op.rdma_write.rem_addr.addr =
4145 				rdma_wr(ib_wr)->remote_addr;
4146 			info.op.rdma_write.rem_addr.lkey = rdma_wr(ib_wr)->rkey;
4147 			if (ib_wr->send_flags & IB_SEND_INLINE)
4148 				err = irdma_uk_inline_rdma_write(ukqp, &info, false);
4149 			else
4150 				err = irdma_uk_rdma_write(ukqp, &info, false);
4151 			break;
4152 		case IB_WR_RDMA_READ_WITH_INV:
4153 			inv_stag = true;
4154 			fallthrough;
4155 		case IB_WR_RDMA_READ:
4156 			if (ib_wr->num_sge >
4157 			    dev->hw_attrs.uk_attrs.max_hw_read_sges) {
4158 				err = -EINVAL;
4159 				break;
4160 			}
4161 			info.op_type = IRDMA_OP_TYPE_RDMA_READ;
4162 			info.op.rdma_read.rem_addr.addr = rdma_wr(ib_wr)->remote_addr;
4163 			info.op.rdma_read.rem_addr.lkey = rdma_wr(ib_wr)->rkey;
4164 			info.op.rdma_read.lo_sg_list = (void *)ib_wr->sg_list;
4165 			info.op.rdma_read.num_lo_sges = ib_wr->num_sge;
4166 			err = irdma_uk_rdma_read(ukqp, &info, inv_stag, false);
4167 			break;
4168 		case IB_WR_LOCAL_INV:
4169 			info.op_type = IRDMA_OP_TYPE_INV_STAG;
4170 			info.local_fence = true;
4171 			info.op.inv_local_stag.target_stag = ib_wr->ex.invalidate_rkey;
4172 			err = irdma_uk_stag_local_invalidate(ukqp, &info, true);
4173 			break;
4174 		case IB_WR_REG_MR: {
4175 			struct irdma_mr *iwmr = to_iwmr(reg_wr(ib_wr)->mr);
4176 			struct irdma_pble_alloc *palloc = &iwmr->iwpbl.pble_alloc;
4177 			struct irdma_fast_reg_stag_info stag_info = {};
4178 
4179 			stag_info.signaled = info.signaled;
4180 			stag_info.read_fence = info.read_fence;
4181 			stag_info.access_rights =
4182 				irdma_get_mr_access(reg_wr(ib_wr)->access,
4183 						    dev->hw_attrs.uk_attrs.hw_rev);
4184 			stag_info.stag_key = reg_wr(ib_wr)->key & 0xff;
4185 			stag_info.stag_idx = reg_wr(ib_wr)->key >> 8;
4186 			stag_info.page_size = reg_wr(ib_wr)->mr->page_size;
4187 			stag_info.wr_id = ib_wr->wr_id;
4188 			stag_info.addr_type = IRDMA_ADDR_TYPE_VA_BASED;
4189 			stag_info.va = (void *)(uintptr_t)iwmr->ibmr.iova;
4190 			stag_info.total_len = iwmr->ibmr.length;
4191 			stag_info.reg_addr_pa = *palloc->level1.addr;
4192 			stag_info.first_pm_pbl_index = palloc->level1.idx;
4193 			stag_info.local_fence = ib_wr->send_flags & IB_SEND_FENCE;
4194 			if (iwmr->npages > IRDMA_MIN_PAGES_PER_FMR)
4195 				stag_info.chunk_size = 1;
4196 			err = irdma_sc_mr_fast_register(&iwqp->sc_qp, &stag_info,
4197 							true);
4198 			break;
4199 		}
4200 		default:
4201 			err = -EINVAL;
4202 			ibdev_dbg(&iwqp->iwdev->ibdev,
4203 				  "VERBS: upost_send bad opcode = 0x%x\n",
4204 				  ib_wr->opcode);
4205 			break;
4206 		}
4207 
4208 		if (err)
4209 			break;
4210 		ib_wr = ib_wr->next;
4211 	}
4212 
4213 	if (!iwqp->flush_issued) {
4214 		if (iwqp->hw_iwarp_state <= IRDMA_QP_STATE_RTS)
4215 			irdma_uk_qp_post_wr(ukqp);
4216 		spin_unlock_irqrestore(&iwqp->lock, flags);
4217 	} else {
4218 		spin_unlock_irqrestore(&iwqp->lock, flags);
4219 		mod_delayed_work(iwqp->iwdev->cleanup_wq, &iwqp->dwork_flush,
4220 				 msecs_to_jiffies(IRDMA_FLUSH_DELAY_MS));
4221 	}
4222 
4223 	if (err)
4224 		*bad_wr = ib_wr;
4225 
4226 	return err;
4227 }
4228 
4229 /**
4230  * irdma_post_srq_recv - post receive wr for kernel application
4231  * @ibsrq: ib srq pointer
4232  * @ib_wr: work request for receive
4233  * @bad_wr: bad wr caused an error
4234  */
4235 static int irdma_post_srq_recv(struct ib_srq *ibsrq,
4236 			       const struct ib_recv_wr *ib_wr,
4237 			       const struct ib_recv_wr **bad_wr)
4238 {
4239 	struct irdma_srq *iwsrq = to_iwsrq(ibsrq);
4240 	struct irdma_srq_uk *uksrq = &iwsrq->sc_srq.srq_uk;
4241 	struct irdma_post_rq_info post_recv = {};
4242 	unsigned long flags;
4243 	int err = 0;
4244 
4245 	spin_lock_irqsave(&iwsrq->lock, flags);
4246 	while (ib_wr) {
4247 		if (ib_wr->num_sge > uksrq->max_srq_frag_cnt) {
4248 			err = -EINVAL;
4249 			goto out;
4250 		}
4251 		post_recv.num_sges = ib_wr->num_sge;
4252 		post_recv.wr_id = ib_wr->wr_id;
4253 		post_recv.sg_list = ib_wr->sg_list;
4254 		err = irdma_uk_srq_post_receive(uksrq, &post_recv);
4255 		if (err)
4256 			goto out;
4257 
4258 		ib_wr = ib_wr->next;
4259 	}
4260 
4261 out:
4262 	spin_unlock_irqrestore(&iwsrq->lock, flags);
4263 
4264 	if (err)
4265 		*bad_wr = ib_wr;
4266 
4267 	return err;
4268 }
4269 
4270 /**
4271  * irdma_post_recv - post receive wr for kernel application
4272  * @ibqp: ib qp pointer
4273  * @ib_wr: work request for receive
4274  * @bad_wr: bad wr caused an error
4275  */
4276 static int irdma_post_recv(struct ib_qp *ibqp,
4277 			   const struct ib_recv_wr *ib_wr,
4278 			   const struct ib_recv_wr **bad_wr)
4279 {
4280 	struct irdma_qp *iwqp;
4281 	struct irdma_qp_uk *ukqp;
4282 	struct irdma_post_rq_info post_recv = {};
4283 	unsigned long flags;
4284 	int err = 0;
4285 
4286 	iwqp = to_iwqp(ibqp);
4287 	ukqp = &iwqp->sc_qp.qp_uk;
4288 
4289 	if (ukqp->srq_uk) {
4290 		*bad_wr = ib_wr;
4291 		return -EINVAL;
4292 	}
4293 
4294 	spin_lock_irqsave(&iwqp->lock, flags);
4295 	while (ib_wr) {
4296 		post_recv.num_sges = ib_wr->num_sge;
4297 		post_recv.wr_id = ib_wr->wr_id;
4298 		post_recv.sg_list = ib_wr->sg_list;
4299 		err = irdma_uk_post_receive(ukqp, &post_recv);
4300 		if (err) {
4301 			ibdev_dbg(&iwqp->iwdev->ibdev,
4302 				  "VERBS: post_recv err %d\n", err);
4303 			goto out;
4304 		}
4305 
4306 		ib_wr = ib_wr->next;
4307 	}
4308 
4309 out:
4310 	spin_unlock_irqrestore(&iwqp->lock, flags);
4311 	if (iwqp->flush_issued)
4312 		mod_delayed_work(iwqp->iwdev->cleanup_wq, &iwqp->dwork_flush,
4313 				 msecs_to_jiffies(IRDMA_FLUSH_DELAY_MS));
4314 
4315 	if (err)
4316 		*bad_wr = ib_wr;
4317 
4318 	return err;
4319 }
4320 
4321 /**
4322  * irdma_flush_err_to_ib_wc_status - return change flush error code to IB status
4323  * @opcode: iwarp flush code
4324  */
4325 static enum ib_wc_status irdma_flush_err_to_ib_wc_status(enum irdma_flush_opcode opcode)
4326 {
4327 	switch (opcode) {
4328 	case FLUSH_PROT_ERR:
4329 		return IB_WC_LOC_PROT_ERR;
4330 	case FLUSH_REM_ACCESS_ERR:
4331 		return IB_WC_REM_ACCESS_ERR;
4332 	case FLUSH_LOC_QP_OP_ERR:
4333 		return IB_WC_LOC_QP_OP_ERR;
4334 	case FLUSH_REM_OP_ERR:
4335 		return IB_WC_REM_OP_ERR;
4336 	case FLUSH_LOC_LEN_ERR:
4337 		return IB_WC_LOC_LEN_ERR;
4338 	case FLUSH_GENERAL_ERR:
4339 		return IB_WC_WR_FLUSH_ERR;
4340 	case FLUSH_RETRY_EXC_ERR:
4341 		return IB_WC_RETRY_EXC_ERR;
4342 	case FLUSH_MW_BIND_ERR:
4343 		return IB_WC_MW_BIND_ERR;
4344 	case FLUSH_REM_INV_REQ_ERR:
4345 		return IB_WC_REM_INV_REQ_ERR;
4346 	case FLUSH_RNR_RETRY_EXC_ERR:
4347 		return IB_WC_RNR_RETRY_EXC_ERR;
4348 	case FLUSH_FATAL_ERR:
4349 	default:
4350 		return IB_WC_FATAL_ERR;
4351 	}
4352 }
4353 
4354 /**
4355  * irdma_process_cqe - process cqe info
4356  * @entry: processed cqe
4357  * @cq_poll_info: cqe info
4358  */
4359 static void irdma_process_cqe(struct ib_wc *entry,
4360 			      struct irdma_cq_poll_info *cq_poll_info)
4361 {
4362 	struct irdma_sc_qp *qp;
4363 
4364 	entry->wc_flags = 0;
4365 	entry->pkey_index = 0;
4366 	entry->wr_id = cq_poll_info->wr_id;
4367 
4368 	qp = cq_poll_info->qp_handle;
4369 	entry->qp = qp->qp_uk.back_qp;
4370 
4371 	if (cq_poll_info->error) {
4372 		entry->status = (cq_poll_info->comp_status == IRDMA_COMPL_STATUS_FLUSHED) ?
4373 				irdma_flush_err_to_ib_wc_status(cq_poll_info->minor_err) : IB_WC_GENERAL_ERR;
4374 
4375 		entry->vendor_err = cq_poll_info->major_err << 16 |
4376 				    cq_poll_info->minor_err;
4377 	} else {
4378 		entry->status = IB_WC_SUCCESS;
4379 		if (cq_poll_info->imm_valid) {
4380 			entry->ex.imm_data = htonl(cq_poll_info->imm_data);
4381 			entry->wc_flags |= IB_WC_WITH_IMM;
4382 		}
4383 		if (cq_poll_info->ud_smac_valid) {
4384 			ether_addr_copy(entry->smac, cq_poll_info->ud_smac);
4385 			entry->wc_flags |= IB_WC_WITH_SMAC;
4386 		}
4387 
4388 		if (cq_poll_info->ud_vlan_valid) {
4389 			u16 vlan = cq_poll_info->ud_vlan & VLAN_VID_MASK;
4390 
4391 			entry->sl = cq_poll_info->ud_vlan >> VLAN_PRIO_SHIFT;
4392 			if (vlan) {
4393 				entry->vlan_id = vlan;
4394 				entry->wc_flags |= IB_WC_WITH_VLAN;
4395 			}
4396 		} else {
4397 			entry->sl = 0;
4398 		}
4399 	}
4400 
4401 	if (cq_poll_info->q_type == IRDMA_CQE_QTYPE_SQ) {
4402 		set_ib_wc_op_sq(cq_poll_info, entry);
4403 	} else {
4404 		if (qp->dev->hw_attrs.uk_attrs.hw_rev <= IRDMA_GEN_2)
4405 			set_ib_wc_op_rq(cq_poll_info, entry,
4406 					qp->qp_uk.qp_caps & IRDMA_SEND_WITH_IMM ?
4407 					true : false);
4408 		else
4409 			set_ib_wc_op_rq_gen_3(cq_poll_info, entry);
4410 		if (qp->qp_uk.qp_type != IRDMA_QP_TYPE_ROCE_UD &&
4411 		    cq_poll_info->stag_invalid_set) {
4412 			entry->ex.invalidate_rkey = cq_poll_info->inv_stag;
4413 			entry->wc_flags |= IB_WC_WITH_INVALIDATE;
4414 		}
4415 	}
4416 
4417 	if (qp->qp_uk.qp_type == IRDMA_QP_TYPE_ROCE_UD) {
4418 		entry->src_qp = cq_poll_info->ud_src_qpn;
4419 		entry->slid = 0;
4420 		entry->wc_flags |=
4421 			(IB_WC_GRH | IB_WC_WITH_NETWORK_HDR_TYPE);
4422 		entry->network_hdr_type = cq_poll_info->ipv4 ?
4423 						  RDMA_NETWORK_IPV4 :
4424 						  RDMA_NETWORK_IPV6;
4425 	} else {
4426 		entry->src_qp = cq_poll_info->qp_id;
4427 	}
4428 
4429 	entry->byte_len = cq_poll_info->bytes_xfered;
4430 }
4431 
4432 /**
4433  * irdma_poll_one - poll one entry of the CQ
4434  * @ukcq: ukcq to poll
4435  * @cur_cqe: current CQE info to be filled in
4436  * @entry: ibv_wc object to be filled for non-extended CQ or NULL for extended CQ
4437  *
4438  * Returns the internal irdma device error code or 0 on success
4439  */
4440 static inline int irdma_poll_one(struct irdma_cq_uk *ukcq,
4441 				 struct irdma_cq_poll_info *cur_cqe,
4442 				 struct ib_wc *entry)
4443 {
4444 	int ret = irdma_uk_cq_poll_cmpl(ukcq, cur_cqe);
4445 
4446 	if (ret)
4447 		return ret;
4448 
4449 	irdma_process_cqe(entry, cur_cqe);
4450 
4451 	return 0;
4452 }
4453 
4454 /**
4455  * __irdma_poll_cq - poll cq for completion (kernel apps)
4456  * @iwcq: cq to poll
4457  * @num_entries: number of entries to poll
4458  * @entry: wr of a completed entry
4459  */
4460 static int __irdma_poll_cq(struct irdma_cq *iwcq, int num_entries, struct ib_wc *entry)
4461 {
4462 	struct list_head *tmp_node, *list_node;
4463 	struct irdma_cq_buf *last_buf = NULL;
4464 	struct irdma_cq_poll_info *cur_cqe = &iwcq->cur_cqe;
4465 	struct irdma_cq_buf *cq_buf;
4466 	int ret;
4467 	struct irdma_device *iwdev;
4468 	struct irdma_cq_uk *ukcq;
4469 	bool cq_new_cqe = false;
4470 	int resized_bufs = 0;
4471 	int npolled = 0;
4472 
4473 	iwdev = to_iwdev(iwcq->ibcq.device);
4474 	ukcq = &iwcq->sc_cq.cq_uk;
4475 
4476 	/* go through the list of previously resized CQ buffers */
4477 	list_for_each_safe(list_node, tmp_node, &iwcq->resize_list) {
4478 		cq_buf = container_of(list_node, struct irdma_cq_buf, list);
4479 		while (npolled < num_entries) {
4480 			ret = irdma_poll_one(&cq_buf->cq_uk, cur_cqe, entry + npolled);
4481 			if (!ret) {
4482 				++npolled;
4483 				cq_new_cqe = true;
4484 				continue;
4485 			}
4486 			if (ret == -ENOENT)
4487 				break;
4488 			 /* QP using the CQ is destroyed. Skip reporting this CQE */
4489 			if (ret == -EFAULT) {
4490 				cq_new_cqe = true;
4491 				continue;
4492 			}
4493 			goto error;
4494 		}
4495 
4496 		/* save the resized CQ buffer which received the last cqe */
4497 		if (cq_new_cqe)
4498 			last_buf = cq_buf;
4499 		cq_new_cqe = false;
4500 	}
4501 
4502 	/* check the current CQ for new cqes */
4503 	while (npolled < num_entries) {
4504 		ret = irdma_poll_one(ukcq, cur_cqe, entry + npolled);
4505 		if (ret == -ENOENT) {
4506 			ret = irdma_generated_cmpls(iwcq, cur_cqe);
4507 			if (!ret)
4508 				irdma_process_cqe(entry + npolled, cur_cqe);
4509 		}
4510 		if (!ret) {
4511 			++npolled;
4512 			cq_new_cqe = true;
4513 			continue;
4514 		}
4515 
4516 		if (ret == -ENOENT)
4517 			break;
4518 		/* QP using the CQ is destroyed. Skip reporting this CQE */
4519 		if (ret == -EFAULT) {
4520 			cq_new_cqe = true;
4521 			continue;
4522 		}
4523 		goto error;
4524 	}
4525 
4526 	if (cq_new_cqe)
4527 		/* all previous CQ resizes are complete */
4528 		resized_bufs = irdma_process_resize_list(iwcq, iwdev, NULL);
4529 	else if (last_buf)
4530 		/* only CQ resizes up to the last_buf are complete */
4531 		resized_bufs = irdma_process_resize_list(iwcq, iwdev, last_buf);
4532 	if (resized_bufs)
4533 		/* report to the HW the number of complete CQ resizes */
4534 		irdma_uk_cq_set_resized_cnt(ukcq, resized_bufs);
4535 
4536 	return npolled;
4537 error:
4538 	ibdev_dbg(&iwdev->ibdev, "%s: Error polling CQ, irdma_err: %d\n",
4539 		  __func__, ret);
4540 
4541 	return ret;
4542 }
4543 
4544 /**
4545  * irdma_poll_cq - poll cq for completion (kernel apps)
4546  * @ibcq: cq to poll
4547  * @num_entries: number of entries to poll
4548  * @entry: wr of a completed entry
4549  */
4550 static int irdma_poll_cq(struct ib_cq *ibcq, int num_entries,
4551 			 struct ib_wc *entry)
4552 {
4553 	struct irdma_cq *iwcq;
4554 	unsigned long flags;
4555 	int ret;
4556 
4557 	iwcq = to_iwcq(ibcq);
4558 
4559 	spin_lock_irqsave(&iwcq->lock, flags);
4560 	ret = __irdma_poll_cq(iwcq, num_entries, entry);
4561 	spin_unlock_irqrestore(&iwcq->lock, flags);
4562 
4563 	return ret;
4564 }
4565 
4566 /**
4567  * irdma_req_notify_cq - arm cq kernel application
4568  * @ibcq: cq to arm
4569  * @notify_flags: notofication flags
4570  */
4571 static int irdma_req_notify_cq(struct ib_cq *ibcq,
4572 			       enum ib_cq_notify_flags notify_flags)
4573 {
4574 	struct irdma_cq *iwcq;
4575 	struct irdma_cq_uk *ukcq;
4576 	unsigned long flags;
4577 	enum irdma_cmpl_notify cq_notify;
4578 	bool promo_event = false;
4579 	int ret = 0;
4580 
4581 	cq_notify = notify_flags == IB_CQ_SOLICITED ?
4582 		    IRDMA_CQ_COMPL_SOLICITED : IRDMA_CQ_COMPL_EVENT;
4583 	iwcq = to_iwcq(ibcq);
4584 	ukcq = &iwcq->sc_cq.cq_uk;
4585 
4586 	spin_lock_irqsave(&iwcq->lock, flags);
4587 	/* Only promote to arm the CQ for any event if the last arm event was solicited. */
4588 	if (iwcq->last_notify == IRDMA_CQ_COMPL_SOLICITED && notify_flags != IB_CQ_SOLICITED)
4589 		promo_event = true;
4590 
4591 	if (!atomic_cmpxchg(&iwcq->armed, 0, 1) || promo_event) {
4592 		iwcq->last_notify = cq_notify;
4593 		irdma_uk_cq_request_notification(ukcq, cq_notify);
4594 	}
4595 
4596 	if ((notify_flags & IB_CQ_REPORT_MISSED_EVENTS) &&
4597 	    (!irdma_uk_cq_empty(ukcq) || !list_empty(&iwcq->cmpl_generated)))
4598 		ret = 1;
4599 	spin_unlock_irqrestore(&iwcq->lock, flags);
4600 
4601 	return ret;
4602 }
4603 
4604 static const struct rdma_stat_desc irdma_hw_stat_descs[] = {
4605 	/* gen1 - 32-bit */
4606 	[IRDMA_HW_STAT_INDEX_IP4RXDISCARD].name		= "ip4InDiscards",
4607 	[IRDMA_HW_STAT_INDEX_IP4RXTRUNC].name		= "ip4InTruncatedPkts",
4608 	[IRDMA_HW_STAT_INDEX_IP4TXNOROUTE].name		= "ip4OutNoRoutes",
4609 	[IRDMA_HW_STAT_INDEX_IP6RXDISCARD].name		= "ip6InDiscards",
4610 	[IRDMA_HW_STAT_INDEX_IP6RXTRUNC].name		= "ip6InTruncatedPkts",
4611 	[IRDMA_HW_STAT_INDEX_IP6TXNOROUTE].name		= "ip6OutNoRoutes",
4612 	[IRDMA_HW_STAT_INDEX_RXVLANERR].name		= "rxVlanErrors",
4613 	/* gen1 - 64-bit */
4614 	[IRDMA_HW_STAT_INDEX_IP4RXOCTS].name		= "ip4InOctets",
4615 	[IRDMA_HW_STAT_INDEX_IP4RXPKTS].name		= "ip4InPkts",
4616 	[IRDMA_HW_STAT_INDEX_IP4RXFRAGS].name		= "ip4InReasmRqd",
4617 	[IRDMA_HW_STAT_INDEX_IP4RXMCPKTS].name		= "ip4InMcastPkts",
4618 	[IRDMA_HW_STAT_INDEX_IP4TXOCTS].name		= "ip4OutOctets",
4619 	[IRDMA_HW_STAT_INDEX_IP4TXPKTS].name		= "ip4OutPkts",
4620 	[IRDMA_HW_STAT_INDEX_IP4TXFRAGS].name		= "ip4OutSegRqd",
4621 	[IRDMA_HW_STAT_INDEX_IP4TXMCPKTS].name		= "ip4OutMcastPkts",
4622 	[IRDMA_HW_STAT_INDEX_IP6RXOCTS].name		= "ip6InOctets",
4623 	[IRDMA_HW_STAT_INDEX_IP6RXPKTS].name		= "ip6InPkts",
4624 	[IRDMA_HW_STAT_INDEX_IP6RXFRAGS].name		= "ip6InReasmRqd",
4625 	[IRDMA_HW_STAT_INDEX_IP6RXMCPKTS].name		= "ip6InMcastPkts",
4626 	[IRDMA_HW_STAT_INDEX_IP6TXOCTS].name		= "ip6OutOctets",
4627 	[IRDMA_HW_STAT_INDEX_IP6TXPKTS].name		= "ip6OutPkts",
4628 	[IRDMA_HW_STAT_INDEX_IP6TXFRAGS].name		= "ip6OutSegRqd",
4629 	[IRDMA_HW_STAT_INDEX_IP6TXMCPKTS].name		= "ip6OutMcastPkts",
4630 	[IRDMA_HW_STAT_INDEX_RDMARXRDS].name		= "InRdmaReads",
4631 	[IRDMA_HW_STAT_INDEX_RDMARXSNDS].name		= "InRdmaSends",
4632 	[IRDMA_HW_STAT_INDEX_RDMARXWRS].name		= "InRdmaWrites",
4633 	[IRDMA_HW_STAT_INDEX_RDMATXRDS].name		= "OutRdmaReads",
4634 	[IRDMA_HW_STAT_INDEX_RDMATXSNDS].name		= "OutRdmaSends",
4635 	[IRDMA_HW_STAT_INDEX_RDMATXWRS].name		= "OutRdmaWrites",
4636 	[IRDMA_HW_STAT_INDEX_RDMAVBND].name		= "RdmaBnd",
4637 	[IRDMA_HW_STAT_INDEX_RDMAVINV].name		= "RdmaInv",
4638 
4639 	/* gen2 - 32-bit */
4640 	[IRDMA_HW_STAT_INDEX_RXRPCNPHANDLED].name	= "cnpHandled",
4641 	[IRDMA_HW_STAT_INDEX_RXRPCNPIGNORED].name	= "cnpIgnored",
4642 	[IRDMA_HW_STAT_INDEX_TXNPCNPSENT].name		= "cnpSent",
4643 	/* gen2 - 64-bit */
4644 	[IRDMA_HW_STAT_INDEX_IP4RXMCOCTS].name		= "ip4InMcastOctets",
4645 	[IRDMA_HW_STAT_INDEX_IP4TXMCOCTS].name		= "ip4OutMcastOctets",
4646 	[IRDMA_HW_STAT_INDEX_IP6RXMCOCTS].name		= "ip6InMcastOctets",
4647 	[IRDMA_HW_STAT_INDEX_IP6TXMCOCTS].name		= "ip6OutMcastOctets",
4648 	[IRDMA_HW_STAT_INDEX_UDPRXPKTS].name		= "RxUDP",
4649 	[IRDMA_HW_STAT_INDEX_UDPTXPKTS].name		= "TxUDP",
4650 	[IRDMA_HW_STAT_INDEX_RXNPECNMARKEDPKTS].name	= "RxECNMrkd",
4651 	[IRDMA_HW_STAT_INDEX_TCPRTXSEG].name		= "RetransSegs",
4652 	[IRDMA_HW_STAT_INDEX_TCPRXOPTERR].name		= "InOptErrors",
4653 	[IRDMA_HW_STAT_INDEX_TCPRXPROTOERR].name	= "InProtoErrors",
4654 	[IRDMA_HW_STAT_INDEX_TCPRXSEGS].name		= "InSegs",
4655 	[IRDMA_HW_STAT_INDEX_TCPTXSEG].name		= "OutSegs",
4656 
4657 	/* gen3 */
4658 	[IRDMA_HW_STAT_INDEX_RNR_SENT].name		= "RNR sent",
4659 	[IRDMA_HW_STAT_INDEX_RNR_RCVD].name		= "RNR received",
4660 	[IRDMA_HW_STAT_INDEX_RDMAORDLMTCNT].name	= "ord limit count",
4661 	[IRDMA_HW_STAT_INDEX_RDMAIRDLMTCNT].name	= "ird limit count",
4662 	[IRDMA_HW_STAT_INDEX_RDMARXATS].name		= "Rx atomics",
4663 	[IRDMA_HW_STAT_INDEX_RDMATXATS].name		= "Tx atomics",
4664 	[IRDMA_HW_STAT_INDEX_NAKSEQERR].name		= "Nak Sequence Error",
4665 	[IRDMA_HW_STAT_INDEX_NAKSEQERR_IMPLIED].name	= "Nak Sequence Error Implied",
4666 	[IRDMA_HW_STAT_INDEX_RTO].name			= "RTO",
4667 	[IRDMA_HW_STAT_INDEX_RXOOOPKTS].name		= "Rcvd Out of order packets",
4668 	[IRDMA_HW_STAT_INDEX_ICRCERR].name		= "CRC errors",
4669 };
4670 
4671 static int irdma_roce_port_immutable(struct ib_device *ibdev, u32 port_num,
4672 				     struct ib_port_immutable *immutable)
4673 {
4674 	struct ib_port_attr attr;
4675 	int err;
4676 
4677 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
4678 	err = ib_query_port(ibdev, port_num, &attr);
4679 	if (err)
4680 		return err;
4681 
4682 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
4683 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
4684 	immutable->gid_tbl_len = attr.gid_tbl_len;
4685 
4686 	return 0;
4687 }
4688 
4689 static int irdma_iw_port_immutable(struct ib_device *ibdev, u32 port_num,
4690 				   struct ib_port_immutable *immutable)
4691 {
4692 	struct ib_port_attr attr;
4693 	int err;
4694 
4695 	immutable->core_cap_flags = RDMA_CORE_PORT_IWARP;
4696 	err = ib_query_port(ibdev, port_num, &attr);
4697 	if (err)
4698 		return err;
4699 	immutable->gid_tbl_len = attr.gid_tbl_len;
4700 
4701 	return 0;
4702 }
4703 
4704 static void irdma_get_dev_fw_str(struct ib_device *dev, char *str)
4705 {
4706 	struct irdma_device *iwdev = to_iwdev(dev);
4707 
4708 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%u.%u",
4709 		 irdma_fw_major_ver(&iwdev->rf->sc_dev),
4710 		 irdma_fw_minor_ver(&iwdev->rf->sc_dev));
4711 }
4712 
4713 /**
4714  * irdma_alloc_hw_port_stats - Allocate a hw stats structure
4715  * @ibdev: device pointer from stack
4716  * @port_num: port number
4717  */
4718 static struct rdma_hw_stats *irdma_alloc_hw_port_stats(struct ib_device *ibdev,
4719 						       u32 port_num)
4720 {
4721 	struct irdma_device *iwdev = to_iwdev(ibdev);
4722 	struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
4723 
4724 	int num_counters = dev->hw_attrs.max_stat_idx;
4725 	unsigned long lifespan = RDMA_HW_STATS_DEFAULT_LIFESPAN;
4726 
4727 	return rdma_alloc_hw_stats_struct(irdma_hw_stat_descs, num_counters,
4728 					  lifespan);
4729 }
4730 
4731 /**
4732  * irdma_get_hw_stats - Populates the rdma_hw_stats structure
4733  * @ibdev: device pointer from stack
4734  * @stats: stats pointer from stack
4735  * @port_num: port number
4736  * @index: which hw counter the stack is requesting we update
4737  */
4738 static int irdma_get_hw_stats(struct ib_device *ibdev,
4739 			      struct rdma_hw_stats *stats, u32 port_num,
4740 			      int index)
4741 {
4742 	struct irdma_device *iwdev = to_iwdev(ibdev);
4743 	struct irdma_dev_hw_stats *hw_stats = &iwdev->vsi.pestat->hw_stats;
4744 
4745 	if (iwdev->rf->rdma_ver >= IRDMA_GEN_2)
4746 		irdma_cqp_gather_stats_cmd(&iwdev->rf->sc_dev, iwdev->vsi.pestat, true);
4747 	else
4748 		irdma_cqp_gather_stats_gen1(&iwdev->rf->sc_dev, iwdev->vsi.pestat);
4749 
4750 	memcpy(&stats->value[0], hw_stats, sizeof(u64) * stats->num_counters);
4751 
4752 	return stats->num_counters;
4753 }
4754 
4755 /**
4756  * irdma_query_gid - Query port GID
4757  * @ibdev: device pointer from stack
4758  * @port: port number
4759  * @index: Entry index
4760  * @gid: Global ID
4761  */
4762 static int irdma_query_gid(struct ib_device *ibdev, u32 port, int index,
4763 			   union ib_gid *gid)
4764 {
4765 	struct irdma_device *iwdev = to_iwdev(ibdev);
4766 
4767 	memset(gid->raw, 0, sizeof(gid->raw));
4768 	ether_addr_copy(gid->raw, iwdev->netdev->dev_addr);
4769 
4770 	return 0;
4771 }
4772 
4773 /**
4774  * mcast_list_add -  Add a new mcast item to list
4775  * @rf: RDMA PCI function
4776  * @new_elem: pointer to element to add
4777  */
4778 static void mcast_list_add(struct irdma_pci_f *rf,
4779 			   struct mc_table_list *new_elem)
4780 {
4781 	list_add(&new_elem->list, &rf->mc_qht_list.list);
4782 }
4783 
4784 /**
4785  * mcast_list_del - Remove an mcast item from list
4786  * @mc_qht_elem: pointer to mcast table list element
4787  */
4788 static void mcast_list_del(struct mc_table_list *mc_qht_elem)
4789 {
4790 	if (mc_qht_elem)
4791 		list_del(&mc_qht_elem->list);
4792 }
4793 
4794 /**
4795  * mcast_list_lookup_ip - Search mcast list for address
4796  * @rf: RDMA PCI function
4797  * @ip_mcast: pointer to mcast IP address
4798  */
4799 static struct mc_table_list *mcast_list_lookup_ip(struct irdma_pci_f *rf,
4800 						  u32 *ip_mcast)
4801 {
4802 	struct mc_table_list *mc_qht_el;
4803 	struct list_head *pos, *q;
4804 
4805 	list_for_each_safe (pos, q, &rf->mc_qht_list.list) {
4806 		mc_qht_el = list_entry(pos, struct mc_table_list, list);
4807 		if (!memcmp(mc_qht_el->mc_info.dest_ip, ip_mcast,
4808 			    sizeof(mc_qht_el->mc_info.dest_ip)))
4809 			return mc_qht_el;
4810 	}
4811 
4812 	return NULL;
4813 }
4814 
4815 /**
4816  * irdma_mcast_cqp_op - perform a mcast cqp operation
4817  * @iwdev: irdma device
4818  * @mc_grp_ctx: mcast group info
4819  * @op: operation
4820  *
4821  * returns error status
4822  */
4823 static int irdma_mcast_cqp_op(struct irdma_device *iwdev,
4824 			      struct irdma_mcast_grp_info *mc_grp_ctx, u8 op)
4825 {
4826 	struct cqp_cmds_info *cqp_info;
4827 	struct irdma_cqp_request *cqp_request;
4828 	int status;
4829 
4830 	cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
4831 	if (!cqp_request)
4832 		return -ENOMEM;
4833 
4834 	cqp_request->info.in.u.mc_create.info = *mc_grp_ctx;
4835 	cqp_info = &cqp_request->info;
4836 	cqp_info->cqp_cmd = op;
4837 	cqp_info->post_sq = 1;
4838 	cqp_info->in.u.mc_create.scratch = (uintptr_t)cqp_request;
4839 	cqp_info->in.u.mc_create.cqp = &iwdev->rf->cqp.sc_cqp;
4840 	status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
4841 	irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
4842 
4843 	return status;
4844 }
4845 
4846 /**
4847  * irdma_mcast_mac - Get the multicast MAC for an IP address
4848  * @ip_addr: IPv4 or IPv6 address
4849  * @mac: pointer to result MAC address
4850  * @ipv4: flag indicating IPv4 or IPv6
4851  *
4852  */
4853 void irdma_mcast_mac(u32 *ip_addr, u8 *mac, bool ipv4)
4854 {
4855 	u8 *ip = (u8 *)ip_addr;
4856 
4857 	if (ipv4) {
4858 		unsigned char mac4[ETH_ALEN] = {0x01, 0x00, 0x5E, 0x00,
4859 						0x00, 0x00};
4860 
4861 		mac4[3] = ip[2] & 0x7F;
4862 		mac4[4] = ip[1];
4863 		mac4[5] = ip[0];
4864 		ether_addr_copy(mac, mac4);
4865 	} else {
4866 		unsigned char mac6[ETH_ALEN] = {0x33, 0x33, 0x00, 0x00,
4867 						0x00, 0x00};
4868 
4869 		mac6[2] = ip[3];
4870 		mac6[3] = ip[2];
4871 		mac6[4] = ip[1];
4872 		mac6[5] = ip[0];
4873 		ether_addr_copy(mac, mac6);
4874 	}
4875 }
4876 
4877 /**
4878  * irdma_attach_mcast - attach a qp to a multicast group
4879  * @ibqp: ptr to qp
4880  * @ibgid: pointer to global ID
4881  * @lid: local ID
4882  *
4883  * returns error status
4884  */
4885 static int irdma_attach_mcast(struct ib_qp *ibqp, union ib_gid *ibgid, u16 lid)
4886 {
4887 	struct irdma_qp *iwqp = to_iwqp(ibqp);
4888 	struct irdma_device *iwdev = iwqp->iwdev;
4889 	struct irdma_pci_f *rf = iwdev->rf;
4890 	struct mc_table_list *mc_qht_elem;
4891 	struct irdma_mcast_grp_ctx_entry_info mcg_info = {};
4892 	unsigned long flags;
4893 	u32 ip_addr[4] = {};
4894 	u32 mgn;
4895 	u32 no_mgs;
4896 	int ret = 0;
4897 	bool ipv4;
4898 	u16 vlan_id;
4899 	union irdma_sockaddr sgid_addr;
4900 	unsigned char dmac[ETH_ALEN];
4901 
4902 	rdma_gid2ip((struct sockaddr *)&sgid_addr, ibgid);
4903 
4904 	if (!ipv6_addr_v4mapped((struct in6_addr *)ibgid)) {
4905 		irdma_copy_ip_ntohl(ip_addr,
4906 				    sgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32);
4907 		irdma_get_vlan_mac_ipv6(ip_addr, &vlan_id, NULL);
4908 		ipv4 = false;
4909 		ibdev_dbg(&iwdev->ibdev,
4910 			  "VERBS: qp_id=%d, IP6address=%pI6\n", ibqp->qp_num,
4911 			  ip_addr);
4912 		irdma_mcast_mac(ip_addr, dmac, false);
4913 	} else {
4914 		ip_addr[0] = ntohl(sgid_addr.saddr_in.sin_addr.s_addr);
4915 		ipv4 = true;
4916 		vlan_id = irdma_get_vlan_ipv4(ip_addr);
4917 		irdma_mcast_mac(ip_addr, dmac, true);
4918 		ibdev_dbg(&iwdev->ibdev,
4919 			  "VERBS: qp_id=%d, IP4address=%pI4, MAC=%pM\n",
4920 			  ibqp->qp_num, ip_addr, dmac);
4921 	}
4922 
4923 	spin_lock_irqsave(&rf->qh_list_lock, flags);
4924 	mc_qht_elem = mcast_list_lookup_ip(rf, ip_addr);
4925 	if (!mc_qht_elem) {
4926 		struct irdma_dma_mem *dma_mem_mc;
4927 
4928 		spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4929 		mc_qht_elem = kzalloc_obj(*mc_qht_elem);
4930 		if (!mc_qht_elem)
4931 			return -ENOMEM;
4932 
4933 		mc_qht_elem->mc_info.ipv4_valid = ipv4;
4934 		memcpy(mc_qht_elem->mc_info.dest_ip, ip_addr,
4935 		       sizeof(mc_qht_elem->mc_info.dest_ip));
4936 		ret = irdma_alloc_rsrc(rf, rf->allocated_mcgs, rf->max_mcg,
4937 				       &mgn, &rf->next_mcg);
4938 		if (ret) {
4939 			kfree(mc_qht_elem);
4940 			return -ENOMEM;
4941 		}
4942 
4943 		mc_qht_elem->mc_info.mgn = mgn;
4944 		dma_mem_mc = &mc_qht_elem->mc_grp_ctx.dma_mem_mc;
4945 		dma_mem_mc->size = ALIGN(sizeof(u64) * IRDMA_MAX_MGS_PER_CTX,
4946 					 IRDMA_HW_PAGE_SIZE);
4947 		dma_mem_mc->va = dma_alloc_coherent(rf->hw.device,
4948 						    dma_mem_mc->size,
4949 						    &dma_mem_mc->pa,
4950 						    GFP_KERNEL);
4951 		if (!dma_mem_mc->va) {
4952 			irdma_free_rsrc(rf, rf->allocated_mcgs, mgn);
4953 			kfree(mc_qht_elem);
4954 			return -ENOMEM;
4955 		}
4956 
4957 		mc_qht_elem->mc_grp_ctx.mg_id = (u16)mgn;
4958 		memcpy(mc_qht_elem->mc_grp_ctx.dest_ip_addr, ip_addr,
4959 		       sizeof(mc_qht_elem->mc_grp_ctx.dest_ip_addr));
4960 		mc_qht_elem->mc_grp_ctx.ipv4_valid = ipv4;
4961 		mc_qht_elem->mc_grp_ctx.vlan_id = vlan_id;
4962 		if (vlan_id < VLAN_N_VID)
4963 			mc_qht_elem->mc_grp_ctx.vlan_valid = true;
4964 		mc_qht_elem->mc_grp_ctx.hmc_fcn_id = iwdev->rf->sc_dev.hmc_fn_id;
4965 		mc_qht_elem->mc_grp_ctx.qs_handle =
4966 			iwqp->sc_qp.vsi->qos[iwqp->sc_qp.user_pri].qs_handle;
4967 		ether_addr_copy(mc_qht_elem->mc_grp_ctx.dest_mac_addr, dmac);
4968 
4969 		spin_lock_irqsave(&rf->qh_list_lock, flags);
4970 		mcast_list_add(rf, mc_qht_elem);
4971 	} else {
4972 		if (mc_qht_elem->mc_grp_ctx.no_of_mgs ==
4973 		    IRDMA_MAX_MGS_PER_CTX) {
4974 			spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4975 			return -ENOMEM;
4976 		}
4977 	}
4978 
4979 	mcg_info.qp_id = iwqp->ibqp.qp_num;
4980 	no_mgs = mc_qht_elem->mc_grp_ctx.no_of_mgs;
4981 	irdma_sc_add_mcast_grp(&mc_qht_elem->mc_grp_ctx, &mcg_info);
4982 	spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4983 
4984 	/* Only if there is a change do we need to modify or create */
4985 	if (!no_mgs) {
4986 		ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
4987 					 IRDMA_OP_MC_CREATE);
4988 	} else if (no_mgs != mc_qht_elem->mc_grp_ctx.no_of_mgs) {
4989 		ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
4990 					 IRDMA_OP_MC_MODIFY);
4991 	} else {
4992 		return 0;
4993 	}
4994 
4995 	if (ret)
4996 		goto error;
4997 
4998 	return 0;
4999 
5000 error:
5001 	irdma_sc_del_mcast_grp(&mc_qht_elem->mc_grp_ctx, &mcg_info);
5002 	if (!mc_qht_elem->mc_grp_ctx.no_of_mgs) {
5003 		mcast_list_del(mc_qht_elem);
5004 		dma_free_coherent(rf->hw.device,
5005 				  mc_qht_elem->mc_grp_ctx.dma_mem_mc.size,
5006 				  mc_qht_elem->mc_grp_ctx.dma_mem_mc.va,
5007 				  mc_qht_elem->mc_grp_ctx.dma_mem_mc.pa);
5008 		mc_qht_elem->mc_grp_ctx.dma_mem_mc.va = NULL;
5009 		irdma_free_rsrc(rf, rf->allocated_mcgs,
5010 				mc_qht_elem->mc_grp_ctx.mg_id);
5011 		kfree(mc_qht_elem);
5012 	}
5013 
5014 	return ret;
5015 }
5016 
5017 /**
5018  * irdma_detach_mcast - detach a qp from a multicast group
5019  * @ibqp: ptr to qp
5020  * @ibgid: pointer to global ID
5021  * @lid: local ID
5022  *
5023  * returns error status
5024  */
5025 static int irdma_detach_mcast(struct ib_qp *ibqp, union ib_gid *ibgid, u16 lid)
5026 {
5027 	struct irdma_qp *iwqp = to_iwqp(ibqp);
5028 	struct irdma_device *iwdev = iwqp->iwdev;
5029 	struct irdma_pci_f *rf = iwdev->rf;
5030 	u32 ip_addr[4] = {};
5031 	struct mc_table_list *mc_qht_elem;
5032 	struct irdma_mcast_grp_ctx_entry_info mcg_info = {};
5033 	int ret;
5034 	unsigned long flags;
5035 	union irdma_sockaddr sgid_addr;
5036 
5037 	rdma_gid2ip((struct sockaddr *)&sgid_addr, ibgid);
5038 	if (!ipv6_addr_v4mapped((struct in6_addr *)ibgid))
5039 		irdma_copy_ip_ntohl(ip_addr,
5040 				    sgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32);
5041 	else
5042 		ip_addr[0] = ntohl(sgid_addr.saddr_in.sin_addr.s_addr);
5043 
5044 	spin_lock_irqsave(&rf->qh_list_lock, flags);
5045 	mc_qht_elem = mcast_list_lookup_ip(rf, ip_addr);
5046 	if (!mc_qht_elem) {
5047 		spin_unlock_irqrestore(&rf->qh_list_lock, flags);
5048 		ibdev_dbg(&iwdev->ibdev,
5049 			  "VERBS: address not found MCG\n");
5050 		return 0;
5051 	}
5052 
5053 	mcg_info.qp_id = iwqp->ibqp.qp_num;
5054 	irdma_sc_del_mcast_grp(&mc_qht_elem->mc_grp_ctx, &mcg_info);
5055 	if (!mc_qht_elem->mc_grp_ctx.no_of_mgs) {
5056 		mcast_list_del(mc_qht_elem);
5057 		spin_unlock_irqrestore(&rf->qh_list_lock, flags);
5058 		ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
5059 					 IRDMA_OP_MC_DESTROY);
5060 		if (ret) {
5061 			ibdev_dbg(&iwdev->ibdev,
5062 				  "VERBS: failed MC_DESTROY MCG\n");
5063 			spin_lock_irqsave(&rf->qh_list_lock, flags);
5064 			mcast_list_add(rf, mc_qht_elem);
5065 			spin_unlock_irqrestore(&rf->qh_list_lock, flags);
5066 			return -EAGAIN;
5067 		}
5068 
5069 		dma_free_coherent(rf->hw.device,
5070 				  mc_qht_elem->mc_grp_ctx.dma_mem_mc.size,
5071 				  mc_qht_elem->mc_grp_ctx.dma_mem_mc.va,
5072 				  mc_qht_elem->mc_grp_ctx.dma_mem_mc.pa);
5073 		mc_qht_elem->mc_grp_ctx.dma_mem_mc.va = NULL;
5074 		irdma_free_rsrc(rf, rf->allocated_mcgs,
5075 				mc_qht_elem->mc_grp_ctx.mg_id);
5076 		kfree(mc_qht_elem);
5077 	} else {
5078 		spin_unlock_irqrestore(&rf->qh_list_lock, flags);
5079 		ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
5080 					 IRDMA_OP_MC_MODIFY);
5081 		if (ret) {
5082 			ibdev_dbg(&iwdev->ibdev,
5083 				  "VERBS: failed Modify MCG\n");
5084 			return ret;
5085 		}
5086 	}
5087 
5088 	return 0;
5089 }
5090 
5091 static int irdma_create_hw_ah(struct irdma_device *iwdev, struct irdma_ah *ah, bool sleep)
5092 {
5093 	struct irdma_pci_f *rf = iwdev->rf;
5094 	int err;
5095 
5096 	err = irdma_alloc_rsrc(rf, rf->allocated_ahs, rf->max_ah, &ah->sc_ah.ah_info.ah_idx,
5097 			       &rf->next_ah);
5098 	if (err)
5099 		return err;
5100 
5101 	err = irdma_ah_cqp_op(rf, &ah->sc_ah, IRDMA_OP_AH_CREATE, sleep,
5102 			      irdma_gsi_ud_qp_ah_cb, &ah->sc_ah);
5103 
5104 	if (err) {
5105 		ibdev_dbg(&iwdev->ibdev, "VERBS: CQP-OP Create AH fail");
5106 		goto err_ah_create;
5107 	}
5108 
5109 	if (!sleep) {
5110 		const u64 tmout_ms = irdma_get_timeout_threshold(&rf->sc_dev) *
5111 			CQP_COMPL_WAIT_TIME_MS;
5112 
5113 		if (poll_timeout_us_atomic(irdma_cqp_ce_handler(rf,
5114 								&rf->ccq.sc_cq),
5115 					   ah->sc_ah.ah_info.ah_valid, 1,
5116 					   tmout_ms * USEC_PER_MSEC, false)) {
5117 			ibdev_dbg(&iwdev->ibdev,
5118 				  "VERBS: CQP create AH timed out");
5119 			err = -ETIMEDOUT;
5120 			goto err_ah_create;
5121 		}
5122 	}
5123 	return 0;
5124 
5125 err_ah_create:
5126 	irdma_free_rsrc(iwdev->rf, iwdev->rf->allocated_ahs, ah->sc_ah.ah_info.ah_idx);
5127 
5128 	return err;
5129 }
5130 
5131 static int irdma_setup_ah(struct ib_ah *ibah, struct rdma_ah_init_attr *attr)
5132 {
5133 	struct irdma_pd *pd = to_iwpd(ibah->pd);
5134 	struct irdma_ah *ah = container_of(ibah, struct irdma_ah, ibah);
5135 	struct rdma_ah_attr *ah_attr = attr->ah_attr;
5136 	const struct ib_gid_attr *sgid_attr;
5137 	struct irdma_device *iwdev = to_iwdev(ibah->pd->device);
5138 	struct irdma_pci_f *rf = iwdev->rf;
5139 	struct irdma_sc_ah *sc_ah;
5140 	struct irdma_ah_info *ah_info;
5141 	union irdma_sockaddr sgid_addr, dgid_addr;
5142 	int err;
5143 	u8 dmac[ETH_ALEN];
5144 
5145 	ah->pd = pd;
5146 	sc_ah = &ah->sc_ah;
5147 	sc_ah->ah_info.vsi = &iwdev->vsi;
5148 	irdma_sc_init_ah(&rf->sc_dev, sc_ah);
5149 	ah->sgid_index = ah_attr->grh.sgid_index;
5150 	sgid_attr = ah_attr->grh.sgid_attr;
5151 	memcpy(&ah->dgid, &ah_attr->grh.dgid, sizeof(ah->dgid));
5152 	rdma_gid2ip((struct sockaddr *)&sgid_addr, &sgid_attr->gid);
5153 	rdma_gid2ip((struct sockaddr *)&dgid_addr, &ah_attr->grh.dgid);
5154 	ah->av.attrs = *ah_attr;
5155 	ah->av.net_type = rdma_gid_attr_network_type(sgid_attr);
5156 	ah_info = &sc_ah->ah_info;
5157 	ah_info->pd_idx = pd->sc_pd.pd_id;
5158 	if (ah_attr->ah_flags & IB_AH_GRH) {
5159 		ah_info->flow_label = ah_attr->grh.flow_label;
5160 		ah_info->hop_ttl = ah_attr->grh.hop_limit;
5161 		ah_info->tc_tos = ah_attr->grh.traffic_class;
5162 	}
5163 
5164 	ether_addr_copy(dmac, ah_attr->roce.dmac);
5165 	if (ah->av.net_type == RDMA_NETWORK_IPV4) {
5166 		ah_info->ipv4_valid = true;
5167 		ah_info->dest_ip_addr[0] =
5168 			ntohl(dgid_addr.saddr_in.sin_addr.s_addr);
5169 		ah_info->src_ip_addr[0] =
5170 			ntohl(sgid_addr.saddr_in.sin_addr.s_addr);
5171 		ah_info->do_lpbk = irdma_ipv4_is_lpb(ah_info->src_ip_addr[0],
5172 						     ah_info->dest_ip_addr[0]);
5173 		if (ipv4_is_multicast(dgid_addr.saddr_in.sin_addr.s_addr)) {
5174 			ah_info->do_lpbk = true;
5175 			irdma_mcast_mac(ah_info->dest_ip_addr, dmac, true);
5176 		}
5177 	} else {
5178 		irdma_copy_ip_ntohl(ah_info->dest_ip_addr,
5179 				    dgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32);
5180 		irdma_copy_ip_ntohl(ah_info->src_ip_addr,
5181 				    sgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32);
5182 		ah_info->do_lpbk = irdma_ipv6_is_lpb(ah_info->src_ip_addr,
5183 						     ah_info->dest_ip_addr);
5184 		if (rdma_is_multicast_addr(&dgid_addr.saddr_in6.sin6_addr)) {
5185 			ah_info->do_lpbk = true;
5186 			irdma_mcast_mac(ah_info->dest_ip_addr, dmac, false);
5187 		}
5188 	}
5189 
5190 	err = rdma_read_gid_l2_fields(sgid_attr, &ah_info->vlan_tag,
5191 				      ah_info->mac_addr);
5192 	if (err)
5193 		return err;
5194 
5195 	ah_info->dst_arpindex = irdma_add_arp(iwdev->rf, ah_info->dest_ip_addr,
5196 					      ah_info->ipv4_valid, dmac);
5197 
5198 	if (ah_info->dst_arpindex == -1)
5199 		return -EINVAL;
5200 
5201 	if (ah_info->vlan_tag >= VLAN_N_VID && iwdev->dcb_vlan_mode)
5202 		ah_info->vlan_tag = 0;
5203 
5204 	if (ah_info->vlan_tag < VLAN_N_VID) {
5205 		u8 prio = rt_tos2priority(ah_info->tc_tos);
5206 
5207 		prio = irdma_roce_get_vlan_prio(sgid_attr, prio);
5208 
5209 		ah_info->vlan_tag |= (u16)prio << VLAN_PRIO_SHIFT;
5210 		ah_info->insert_vlan_tag = true;
5211 	}
5212 
5213 	return 0;
5214 }
5215 
5216 /**
5217  * irdma_ah_exists - Check for existing identical AH
5218  * @iwdev: irdma device
5219  * @new_ah: AH to check for
5220  *
5221  * returns true if AH is found, false if not found.
5222  */
5223 static bool irdma_ah_exists(struct irdma_device *iwdev,
5224 			    struct irdma_ah *new_ah)
5225 {
5226 	struct irdma_ah *ah;
5227 	u32 key = new_ah->sc_ah.ah_info.dest_ip_addr[0] ^
5228 		  new_ah->sc_ah.ah_info.dest_ip_addr[1] ^
5229 		  new_ah->sc_ah.ah_info.dest_ip_addr[2] ^
5230 		  new_ah->sc_ah.ah_info.dest_ip_addr[3];
5231 
5232 	hash_for_each_possible(iwdev->rf->ah_hash_tbl, ah, list, key) {
5233 		/* Set ah_valid and ah_id the same so memcmp can work */
5234 		new_ah->sc_ah.ah_info.ah_idx = ah->sc_ah.ah_info.ah_idx;
5235 		new_ah->sc_ah.ah_info.ah_valid = ah->sc_ah.ah_info.ah_valid;
5236 		if (!memcmp(&ah->sc_ah.ah_info, &new_ah->sc_ah.ah_info,
5237 			    sizeof(ah->sc_ah.ah_info))) {
5238 			refcount_inc(&ah->refcnt);
5239 			new_ah->parent_ah = ah;
5240 			return true;
5241 		}
5242 	}
5243 
5244 	return false;
5245 }
5246 
5247 /**
5248  * irdma_destroy_ah - Destroy address handle
5249  * @ibah: pointer to address handle
5250  * @ah_flags: flags for sleepable
5251  */
5252 static int irdma_destroy_ah(struct ib_ah *ibah, u32 ah_flags)
5253 {
5254 	struct irdma_device *iwdev = to_iwdev(ibah->device);
5255 	struct irdma_ah *ah = to_iwah(ibah);
5256 
5257 	if ((ah_flags & RDMA_DESTROY_AH_SLEEPABLE) && ah->parent_ah) {
5258 		mutex_lock(&iwdev->rf->ah_tbl_lock);
5259 		if (!refcount_dec_and_test(&ah->parent_ah->refcnt)) {
5260 			mutex_unlock(&iwdev->rf->ah_tbl_lock);
5261 			return 0;
5262 		}
5263 		hash_del(&ah->parent_ah->list);
5264 		kfree(ah->parent_ah);
5265 		mutex_unlock(&iwdev->rf->ah_tbl_lock);
5266 	}
5267 
5268 	irdma_ah_cqp_op(iwdev->rf, &ah->sc_ah, IRDMA_OP_AH_DESTROY,
5269 			false, NULL, ah);
5270 
5271 	irdma_free_rsrc(iwdev->rf, iwdev->rf->allocated_ahs,
5272 			ah->sc_ah.ah_info.ah_idx);
5273 
5274 	return 0;
5275 }
5276 
5277 /**
5278  * irdma_create_user_ah - create user address handle
5279  * @ibah: address handle
5280  * @attr: address handle attributes
5281  * @udata: User data
5282  *
5283  * returns 0 on success, error otherwise
5284  */
5285 static int irdma_create_user_ah(struct ib_ah *ibah,
5286 				struct rdma_ah_init_attr *attr,
5287 				struct ib_udata *udata)
5288 {
5289 #define IRDMA_CREATE_AH_MIN_RESP_LEN offsetofend(struct irdma_create_ah_resp, rsvd)
5290 	struct irdma_ah *ah = container_of(ibah, struct irdma_ah, ibah);
5291 	struct irdma_device *iwdev = to_iwdev(ibah->pd->device);
5292 	struct irdma_create_ah_resp uresp = {};
5293 	struct irdma_ah *parent_ah;
5294 	int err;
5295 
5296 	if (udata->outlen < IRDMA_CREATE_AH_MIN_RESP_LEN)
5297 		return -EINVAL;
5298 
5299 	err = irdma_setup_ah(ibah, attr);
5300 	if (err)
5301 		return err;
5302 	mutex_lock(&iwdev->rf->ah_tbl_lock);
5303 	if (!irdma_ah_exists(iwdev, ah)) {
5304 		err = irdma_create_hw_ah(iwdev, ah, true);
5305 		if (err) {
5306 			mutex_unlock(&iwdev->rf->ah_tbl_lock);
5307 			return err;
5308 		}
5309 		/* Add new AH to list */
5310 		parent_ah = kmemdup(ah, sizeof(*ah), GFP_KERNEL);
5311 		if (parent_ah) {
5312 			u32 key = parent_ah->sc_ah.ah_info.dest_ip_addr[0] ^
5313 				  parent_ah->sc_ah.ah_info.dest_ip_addr[1] ^
5314 				  parent_ah->sc_ah.ah_info.dest_ip_addr[2] ^
5315 				  parent_ah->sc_ah.ah_info.dest_ip_addr[3];
5316 
5317 			ah->parent_ah = parent_ah;
5318 			hash_add(iwdev->rf->ah_hash_tbl, &parent_ah->list, key);
5319 			refcount_set(&parent_ah->refcnt, 1);
5320 		}
5321 	}
5322 	mutex_unlock(&iwdev->rf->ah_tbl_lock);
5323 
5324 	uresp.ah_id = ah->sc_ah.ah_info.ah_idx;
5325 	err = ib_respond_udata(udata, uresp);
5326 	if (err)
5327 		irdma_destroy_ah(ibah, attr->flags);
5328 
5329 	return err;
5330 }
5331 
5332 /**
5333  * irdma_create_ah - create address handle
5334  * @ibah: address handle
5335  * @attr: address handle attributes
5336  * @udata: NULL
5337  *
5338  * returns 0 on success, error otherwise
5339  */
5340 static int irdma_create_ah(struct ib_ah *ibah, struct rdma_ah_init_attr *attr,
5341 			   struct ib_udata *udata)
5342 {
5343 	struct irdma_ah *ah = container_of(ibah, struct irdma_ah, ibah);
5344 	struct irdma_device *iwdev = to_iwdev(ibah->pd->device);
5345 	int err;
5346 
5347 	err = irdma_setup_ah(ibah, attr);
5348 	if (err)
5349 		return err;
5350 	err = irdma_create_hw_ah(iwdev, ah, attr->flags & RDMA_CREATE_AH_SLEEPABLE);
5351 
5352 	return err;
5353 }
5354 
5355 /**
5356  * irdma_query_ah - Query address handle
5357  * @ibah: pointer to address handle
5358  * @ah_attr: address handle attributes
5359  */
5360 static int irdma_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr)
5361 {
5362 	struct irdma_ah *ah = to_iwah(ibah);
5363 
5364 	memset(ah_attr, 0, sizeof(*ah_attr));
5365 	if (ah->av.attrs.ah_flags & IB_AH_GRH) {
5366 		ah_attr->ah_flags = IB_AH_GRH;
5367 		ah_attr->grh.flow_label = ah->sc_ah.ah_info.flow_label;
5368 		ah_attr->grh.traffic_class = ah->sc_ah.ah_info.tc_tos;
5369 		ah_attr->grh.hop_limit = ah->sc_ah.ah_info.hop_ttl;
5370 		ah_attr->grh.sgid_index = ah->sgid_index;
5371 		memcpy(&ah_attr->grh.dgid, &ah->dgid,
5372 		       sizeof(ah_attr->grh.dgid));
5373 	}
5374 
5375 	return 0;
5376 }
5377 
5378 static enum rdma_link_layer irdma_get_link_layer(struct ib_device *ibdev,
5379 						 u32 port_num)
5380 {
5381 	return IB_LINK_LAYER_ETHERNET;
5382 }
5383 
5384 static const struct ib_device_ops irdma_gen1_dev_ops = {
5385 	.dealloc_driver = irdma_ib_dealloc_device,
5386 };
5387 
5388 static const struct ib_device_ops irdma_gen3_dev_ops = {
5389 	.alloc_mw = irdma_alloc_mw,
5390 	.create_srq = irdma_create_srq,
5391 	.dealloc_mw = irdma_dealloc_mw,
5392 	.destroy_srq = irdma_destroy_srq,
5393 	.modify_srq = irdma_modify_srq,
5394 	.post_srq_recv = irdma_post_srq_recv,
5395 	.query_srq = irdma_query_srq,
5396 };
5397 
5398 static const struct ib_device_ops irdma_roce_dev_ops = {
5399 	.attach_mcast = irdma_attach_mcast,
5400 	.create_ah = irdma_create_ah,
5401 	.create_user_ah = irdma_create_user_ah,
5402 	.destroy_ah = irdma_destroy_ah,
5403 	.detach_mcast = irdma_detach_mcast,
5404 	.get_link_layer = irdma_get_link_layer,
5405 	.get_port_immutable = irdma_roce_port_immutable,
5406 	.modify_qp = irdma_modify_qp_roce,
5407 	.query_ah = irdma_query_ah,
5408 	.query_pkey = irdma_query_pkey,
5409 };
5410 
5411 static const struct ib_device_ops irdma_iw_dev_ops = {
5412 	.get_port_immutable = irdma_iw_port_immutable,
5413 	.iw_accept = irdma_accept,
5414 	.iw_add_ref = irdma_qp_add_ref,
5415 	.iw_connect = irdma_connect,
5416 	.iw_create_listen = irdma_create_listen,
5417 	.iw_destroy_listen = irdma_destroy_listen,
5418 	.iw_get_qp = irdma_get_qp,
5419 	.iw_reject = irdma_reject,
5420 	.iw_rem_ref = irdma_qp_rem_ref,
5421 	.modify_qp = irdma_modify_qp,
5422 	.query_gid = irdma_query_gid,
5423 };
5424 
5425 static const struct ib_device_ops irdma_dev_ops = {
5426 	.owner = THIS_MODULE,
5427 	.driver_id = RDMA_DRIVER_IRDMA,
5428 	.uverbs_abi_ver = IRDMA_ABI_VER,
5429 
5430 	.alloc_hw_port_stats = irdma_alloc_hw_port_stats,
5431 	.alloc_mr = irdma_alloc_mr,
5432 	.alloc_pd = irdma_alloc_pd,
5433 	.alloc_ucontext = irdma_alloc_ucontext,
5434 	.create_cq = irdma_create_cq,
5435 	.create_qp = irdma_create_qp,
5436 	.dealloc_driver = irdma_ib_dealloc_device,
5437 	.dealloc_mw = irdma_dealloc_mw,
5438 	.dealloc_pd = irdma_dealloc_pd,
5439 	.dealloc_ucontext = irdma_dealloc_ucontext,
5440 	.dereg_mr = irdma_dereg_mr,
5441 	.destroy_cq = irdma_destroy_cq,
5442 	.destroy_qp = irdma_destroy_qp,
5443 	.disassociate_ucontext = irdma_disassociate_ucontext,
5444 	.get_dev_fw_str = irdma_get_dev_fw_str,
5445 	.get_dma_mr = irdma_get_dma_mr,
5446 	.get_hw_stats = irdma_get_hw_stats,
5447 	.map_mr_sg = irdma_map_mr_sg,
5448 	.mmap = irdma_mmap,
5449 	.mmap_free = irdma_mmap_free,
5450 	.poll_cq = irdma_poll_cq,
5451 	.post_recv = irdma_post_recv,
5452 	.post_send = irdma_post_send,
5453 	.query_device = irdma_query_device,
5454 	.query_port = irdma_query_port,
5455 	.query_qp = irdma_query_qp,
5456 	.reg_user_mr = irdma_reg_user_mr,
5457 	.reg_user_mr_dmabuf = irdma_reg_user_mr_dmabuf,
5458 	.rereg_user_mr = irdma_rereg_user_mr,
5459 	.req_notify_cq = irdma_req_notify_cq,
5460 	.resize_user_cq = irdma_resize_cq,
5461 	INIT_RDMA_OBJ_SIZE(ib_pd, irdma_pd, ibpd),
5462 	INIT_RDMA_OBJ_SIZE(ib_ucontext, irdma_ucontext, ibucontext),
5463 	INIT_RDMA_OBJ_SIZE(ib_ah, irdma_ah, ibah),
5464 	INIT_RDMA_OBJ_SIZE(ib_cq, irdma_cq, ibcq),
5465 	INIT_RDMA_OBJ_SIZE(ib_mw, irdma_mr, ibmw),
5466 	INIT_RDMA_OBJ_SIZE(ib_qp, irdma_qp, ibqp),
5467 	INIT_RDMA_OBJ_SIZE(ib_srq, irdma_srq, ibsrq),
5468 };
5469 
5470 /**
5471  * irdma_init_roce_device - initialization of roce rdma device
5472  * @iwdev: irdma device
5473  */
5474 static void irdma_init_roce_device(struct irdma_device *iwdev)
5475 {
5476 	iwdev->ibdev.node_type = RDMA_NODE_IB_CA;
5477 	addrconf_addr_eui48((u8 *)&iwdev->ibdev.node_guid,
5478 			    iwdev->netdev->dev_addr);
5479 	ib_set_device_ops(&iwdev->ibdev, &irdma_roce_dev_ops);
5480 }
5481 
5482 /**
5483  * irdma_init_iw_device - initialization of iwarp rdma device
5484  * @iwdev: irdma device
5485  */
5486 static void irdma_init_iw_device(struct irdma_device *iwdev)
5487 {
5488 	struct net_device *netdev = iwdev->netdev;
5489 
5490 	iwdev->ibdev.node_type = RDMA_NODE_RNIC;
5491 	addrconf_addr_eui48((u8 *)&iwdev->ibdev.node_guid,
5492 			    netdev->dev_addr);
5493 	memcpy(iwdev->ibdev.iw_ifname, netdev->name,
5494 	       sizeof(iwdev->ibdev.iw_ifname));
5495 	ib_set_device_ops(&iwdev->ibdev, &irdma_iw_dev_ops);
5496 }
5497 
5498 /**
5499  * irdma_init_rdma_device - initialization of rdma device
5500  * @iwdev: irdma device
5501  */
5502 static void irdma_init_rdma_device(struct irdma_device *iwdev)
5503 {
5504 	struct pci_dev *pcidev = iwdev->rf->pcidev;
5505 
5506 	if (iwdev->roce_mode)
5507 		irdma_init_roce_device(iwdev);
5508 	else
5509 		irdma_init_iw_device(iwdev);
5510 
5511 	iwdev->ibdev.phys_port_cnt = 1;
5512 	iwdev->ibdev.num_comp_vectors = iwdev->rf->ceqs_count;
5513 	iwdev->ibdev.dev.parent = &pcidev->dev;
5514 	ib_set_device_ops(&iwdev->ibdev, &irdma_dev_ops);
5515 	if (iwdev->rf->rdma_ver == IRDMA_GEN_1)
5516 		ib_set_device_ops(&iwdev->ibdev, &irdma_gen1_dev_ops);
5517 	if (iwdev->rf->rdma_ver >= IRDMA_GEN_3)
5518 		ib_set_device_ops(&iwdev->ibdev, &irdma_gen3_dev_ops);
5519 }
5520 
5521 /**
5522  * irdma_port_ibevent - indicate port event
5523  * @iwdev: irdma device
5524  */
5525 void irdma_port_ibevent(struct irdma_device *iwdev)
5526 {
5527 	struct ib_event event;
5528 
5529 	event.device = &iwdev->ibdev;
5530 	event.element.port_num = 1;
5531 	event.event =
5532 		iwdev->iw_status ? IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
5533 	ib_dispatch_event(&event);
5534 }
5535 
5536 /**
5537  * irdma_ib_unregister_device - unregister rdma device from IB
5538  * core
5539  * @iwdev: irdma device
5540  */
5541 void irdma_ib_unregister_device(struct irdma_device *iwdev)
5542 {
5543 	iwdev->iw_status = 0;
5544 	irdma_port_ibevent(iwdev);
5545 	ib_unregister_device(&iwdev->ibdev);
5546 }
5547 
5548 /**
5549  * irdma_ib_register_device - register irdma device to IB core
5550  * @iwdev: irdma device
5551  */
5552 int irdma_ib_register_device(struct irdma_device *iwdev)
5553 {
5554 	int ret;
5555 
5556 	irdma_init_rdma_device(iwdev);
5557 
5558 	ret = ib_device_set_netdev(&iwdev->ibdev, iwdev->netdev, 1);
5559 	if (ret)
5560 		goto error;
5561 	dma_set_max_seg_size(iwdev->rf->hw.device, UINT_MAX);
5562 	ret = ib_register_device(&iwdev->ibdev, "irdma%d", iwdev->rf->hw.device);
5563 	if (ret)
5564 		goto error;
5565 
5566 	iwdev->iw_status = 1;
5567 	irdma_port_ibevent(iwdev);
5568 
5569 	return 0;
5570 
5571 error:
5572 	if (ret)
5573 		ibdev_dbg(&iwdev->ibdev, "VERBS: Register RDMA device fail\n");
5574 
5575 	return ret;
5576 }
5577 
5578 /**
5579  * irdma_ib_dealloc_device
5580  * @ibdev: ib device
5581  *
5582  * callback from ibdev dealloc_driver to deallocate resources
5583  * unber irdma device
5584  */
5585 void irdma_ib_dealloc_device(struct ib_device *ibdev)
5586 {
5587 	struct irdma_device *iwdev = to_iwdev(ibdev);
5588 
5589 	irdma_rt_deinit_hw(iwdev);
5590 	if (!iwdev->is_vport) {
5591 		irdma_ctrl_deinit_hw(iwdev->rf);
5592 		if (iwdev->rf->vchnl_wq) {
5593 			destroy_workqueue(iwdev->rf->vchnl_wq);
5594 			mutex_destroy(&iwdev->rf->sc_dev.vchnl_mutex);
5595 		}
5596 	}
5597 }
5598