xref: /freebsd/sys/dev/qlnx/qlnxr/qlnxr_verbs.c (revision b9f654b163bce26de79705e77b872427c9f2afa1)
1 /*
2  * Copyright (c) 2018-2019 Cavium, Inc.
3  * All rights reserved.
4  *
5  *  Redistribution and use in source and binary forms, with or without
6  *  modification, are permitted provided that the following conditions
7  *  are met:
8  *
9  *  1. Redistributions of source code must retain the above copyright
10  *     notice, this list of conditions and the following disclaimer.
11  *  2. Redistributions in binary form must reproduce the above copyright
12  *     notice, this list of conditions and the following disclaimer in the
13  *     documentation and/or other materials provided with the distribution.
14  *
15  *  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
16  *  AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  *  IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  *  ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
19  *  LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
20  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
21  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
22  *  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
23  *  CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
24  *  ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
25  *  POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 
29 /*
30  * File: qlnxr_verbs.c
31  */
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "qlnxr_def.h"
36 #include "rdma_common.h"
37 #include "qlnxr_roce.h"
38 #include "qlnxr_cm.h"
39 
40 #define upper_32_bits(x) (uint32_t)(x >> 32)
41 #define lower_32_bits(x) (uint32_t)(x)
42 #define HILO_U64(hi, lo)		((((u64)(hi)) << 32) + (lo))
43 
44 #define TYPEPTR_ADDR_SET(type_ptr, field, vaddr)			\
45 	do {								\
46 		(type_ptr)->field.hi = cpu_to_le32(upper_32_bits(vaddr));\
47 		(type_ptr)->field.lo = cpu_to_le32(lower_32_bits(vaddr));\
48 	} while (0)
49 
50 
51 #define RQ_SGE_SET(sge, vaddr, vlength, vflags)			\
52 	do {							\
53 		TYPEPTR_ADDR_SET(sge, addr, vaddr);		\
54 		(sge)->length = cpu_to_le32(vlength);		\
55 		(sge)->flags = cpu_to_le32(vflags);		\
56 	} while (0)
57 
58 #define SRQ_HDR_SET(hdr, vwr_id, num_sge)			\
59 	do {							\
60 		TYPEPTR_ADDR_SET(hdr, wr_id, vwr_id);		\
61 		(hdr)->num_sges = num_sge;			\
62 	} while (0)
63 
64 #define SRQ_SGE_SET(sge, vaddr, vlength, vlkey)			\
65 	do {							\
66 		TYPEPTR_ADDR_SET(sge, addr, vaddr);		\
67 		(sge)->length = cpu_to_le32(vlength);		\
68 		(sge)->l_key = cpu_to_le32(vlkey);		\
69 	} while (0)
70 
71 #define NIPQUAD(addr) \
72 	((unsigned char *)&addr)[0], \
73 	((unsigned char *)&addr)[1], \
74 	((unsigned char *)&addr)[2], \
75 	((unsigned char *)&addr)[3]
76 
77 static int
78 qlnxr_check_srq_params(struct ib_pd *ibpd,
79 	struct qlnxr_dev *dev,
80 	struct ib_srq_init_attr *attrs);
81 
82 static int
83 qlnxr_init_srq_user_params(struct ib_ucontext *ib_ctx,
84 	struct qlnxr_srq *srq,
85 	struct qlnxr_create_srq_ureq *ureq,
86 	int access, int dmasync);
87 
88 static int
89 qlnxr_alloc_srq_kernel_params(struct qlnxr_srq *srq,
90 	struct qlnxr_dev *dev,
91 	struct ib_srq_init_attr *init_attr);
92 
93 
94 static int
95 qlnxr_copy_srq_uresp(struct qlnxr_dev *dev,
96 	struct qlnxr_srq *srq,
97 	struct ib_udata *udata);
98 
99 static void
100 qlnxr_free_srq_user_params(struct qlnxr_srq *srq);
101 
102 static void
103 qlnxr_free_srq_kernel_params(struct qlnxr_srq *srq);
104 
105 
106 static u32
107 qlnxr_srq_elem_left(struct qlnxr_srq_hwq_info *hw_srq);
108 
109 int
110 qlnxr_iw_query_gid(struct ib_device *ibdev, u8 port, int index,
111 	union ib_gid *sgid)
112 {
113 	struct qlnxr_dev	*dev;
114 	qlnx_host_t		*ha;
115 
116 	dev = get_qlnxr_dev(ibdev);
117 	ha = dev->ha;
118 
119 	QL_DPRINT12(ha, "enter\n");
120 
121 	memset(sgid->raw, 0, sizeof(sgid->raw));
122 
123 	memcpy(sgid->raw, dev->ha->primary_mac, sizeof (dev->ha->primary_mac));
124 
125 	QL_DPRINT12(ha, "exit\n");
126 
127 	return 0;
128 }
129 
130 int
131 qlnxr_query_gid(struct ib_device *ibdev, u8 port, int index,
132 	union ib_gid *sgid)
133 {
134 	struct qlnxr_dev	*dev;
135 	qlnx_host_t		*ha;
136 
137 	dev = get_qlnxr_dev(ibdev);
138 	ha = dev->ha;
139 	QL_DPRINT12(ha, "enter index: %d\n", index);
140 #if 0
141 	int ret = 0;
142 	/* @@@: if DEFINE_ROCE_GID_TABLE to be used here */
143 	//if (!rdma_cap_roce_gid_table(ibdev, port)) {
144 	if (!(rdma_protocol_roce(ibdev, port) &&
145 		ibdev->add_gid && ibdev->del_gid)) {
146 		QL_DPRINT11(ha, "acquire gid failed\n");
147 		return -ENODEV;
148 	}
149 
150 	ret = ib_get_cached_gid(ibdev, port, index, sgid, NULL);
151 	if (ret == -EAGAIN) {
152 		memcpy(sgid, &zgid, sizeof(*sgid));
153 		return 0;
154 	}
155 #endif
156 	if ((index >= QLNXR_MAX_SGID) || (index < 0)) {
157 		QL_DPRINT12(ha, "invalid gid index %d\n", index);
158 		memset(sgid, 0, sizeof(*sgid));
159 		return -EINVAL;
160 	}
161 	memcpy(sgid, &dev->sgid_tbl[index], sizeof(*sgid));
162 
163 	QL_DPRINT12(ha, "exit : %p\n", sgid);
164 
165 	return 0;
166 }
167 
168 struct ib_srq *
169 qlnxr_create_srq(struct ib_pd *ibpd, struct ib_srq_init_attr *init_attr,
170 	struct ib_udata *udata)
171 {
172 	struct qlnxr_dev	*dev;
173 	qlnx_host_t		*ha;
174 	struct ecore_rdma_destroy_srq_in_params destroy_in_params;
175 	struct ecore_rdma_create_srq_out_params out_params;
176 	struct ecore_rdma_create_srq_in_params in_params;
177 	u64 pbl_base_addr, phy_prod_pair_addr;
178 	struct qlnxr_pd *pd = get_qlnxr_pd(ibpd);
179 	struct ib_ucontext *ib_ctx = NULL;
180 	struct qlnxr_srq_hwq_info *hw_srq;
181 	struct qlnxr_ucontext *ctx = NULL;
182 	struct qlnxr_create_srq_ureq ureq;
183 	u32 page_cnt, page_size;
184 	struct qlnxr_srq *srq;
185 	int ret = 0;
186 
187 	dev = get_qlnxr_dev((ibpd->device));
188 	ha = dev->ha;
189 
190 	QL_DPRINT12(ha, "enter\n");
191 
192 	ret = qlnxr_check_srq_params(ibpd, dev, init_attr);
193 
194 	srq = kzalloc(sizeof(*srq), GFP_KERNEL);
195 	if (!srq) {
196 		QL_DPRINT11(ha, "cannot allocate memory for srq\n");
197 		return NULL; //@@@ : TODO what to return here?
198 	}
199 
200 	srq->dev = dev;
201 	hw_srq = &srq->hw_srq;
202 	spin_lock_init(&srq->lock);
203 	memset(&in_params, 0, sizeof(in_params));
204 
205 	if (udata && ibpd->uobject && ibpd->uobject->context) {
206 		ib_ctx = ibpd->uobject->context;
207 		ctx = get_qlnxr_ucontext(ib_ctx);
208 
209 		memset(&ureq, 0, sizeof(ureq));
210 		if (ib_copy_from_udata(&ureq, udata, min(sizeof(ureq),
211 			udata->inlen))) {
212 			QL_DPRINT11(ha, "problem"
213 				" copying data from user space\n");
214 			goto err0;
215 		}
216 
217 		ret = qlnxr_init_srq_user_params(ib_ctx, srq, &ureq, 0, 0);
218 		if (ret)
219 			goto err0;
220 
221 		page_cnt = srq->usrq.pbl_info.num_pbes;
222 		pbl_base_addr = srq->usrq.pbl_tbl->pa;
223 		phy_prod_pair_addr = hw_srq->phy_prod_pair_addr;
224 		// @@@ : if DEFINE_IB_UMEM_PAGE_SHIFT
225 		// page_size = BIT(srq->usrq.umem->page_shift);
226 		// else
227 		page_size = srq->usrq.umem->page_size;
228 	} else {
229 		struct ecore_chain *pbl;
230 		ret = qlnxr_alloc_srq_kernel_params(srq, dev, init_attr);
231 		if (ret)
232 			goto err0;
233 		pbl = &hw_srq->pbl;
234 
235 		page_cnt = ecore_chain_get_page_cnt(pbl);
236 		pbl_base_addr = ecore_chain_get_pbl_phys(pbl);
237 		phy_prod_pair_addr = hw_srq->phy_prod_pair_addr;
238 		page_size = pbl->elem_per_page << 4;
239 	}
240 
241 	in_params.pd_id = pd->pd_id;
242 	in_params.pbl_base_addr = pbl_base_addr;
243 	in_params.prod_pair_addr = phy_prod_pair_addr;
244 	in_params.num_pages = page_cnt;
245 	in_params.page_size = page_size;
246 
247 	ret = ecore_rdma_create_srq(dev->rdma_ctx, &in_params, &out_params);
248 	if (ret)
249 		goto err1;
250 
251 	srq->srq_id = out_params.srq_id;
252 
253 	if (udata) {
254 		ret = qlnxr_copy_srq_uresp(dev, srq, udata);
255 		if (ret)
256 			goto err2;
257 	}
258 
259 	QL_DPRINT12(ha, "created srq with srq_id = 0x%0x\n", srq->srq_id);
260 	return &srq->ibsrq;
261 err2:
262 	memset(&in_params, 0, sizeof(in_params));
263 	destroy_in_params.srq_id = srq->srq_id;
264 	ecore_rdma_destroy_srq(dev->rdma_ctx, &destroy_in_params);
265 
266 err1:
267 	if (udata)
268 		qlnxr_free_srq_user_params(srq);
269 	else
270 		qlnxr_free_srq_kernel_params(srq);
271 
272 err0:
273 	kfree(srq);
274 	return ERR_PTR(-EFAULT);
275 }
276 
277 int
278 qlnxr_destroy_srq(struct ib_srq *ibsrq)
279 {
280 	struct qlnxr_dev	*dev;
281 	struct qlnxr_srq	*srq;
282 	qlnx_host_t		*ha;
283 	struct ecore_rdma_destroy_srq_in_params in_params;
284 
285 	srq = get_qlnxr_srq(ibsrq);
286 	dev = srq->dev;
287 	ha = dev->ha;
288 
289 	memset(&in_params, 0, sizeof(in_params));
290 	in_params.srq_id = srq->srq_id;
291 
292 	ecore_rdma_destroy_srq(dev->rdma_ctx, &in_params);
293 
294 	if (ibsrq->pd->uobject && ibsrq->pd->uobject->context)
295 		qlnxr_free_srq_user_params(srq);
296 	else
297 		qlnxr_free_srq_kernel_params(srq);
298 
299 	QL_DPRINT12(ha, "destroyed srq_id=0x%0x\n", srq->srq_id);
300 	kfree(srq);
301 	return 0;
302 }
303 
304 int
305 qlnxr_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
306 	enum ib_srq_attr_mask attr_mask, struct ib_udata *udata)
307 {
308 	struct qlnxr_dev	*dev;
309 	struct qlnxr_srq	*srq;
310 	qlnx_host_t		*ha;
311 	struct ecore_rdma_modify_srq_in_params in_params;
312 	int ret = 0;
313 
314 	srq = get_qlnxr_srq(ibsrq);
315 	dev = srq->dev;
316 	ha = dev->ha;
317 
318 	QL_DPRINT12(ha, "enter\n");
319 	if (attr_mask & IB_SRQ_MAX_WR) {
320 		QL_DPRINT12(ha, "invalid attribute mask=0x%x"
321 			" specified for %p\n", attr_mask, srq);
322 		return -EINVAL;
323 	}
324 
325 	if (attr_mask & IB_SRQ_LIMIT) {
326 		if (attr->srq_limit >= srq->hw_srq.max_wr) {
327 			QL_DPRINT12(ha, "invalid srq_limit=0x%x"
328 				" (max_srq_limit = 0x%x)\n",
329 			       attr->srq_limit, srq->hw_srq.max_wr);
330 			return -EINVAL;
331 		}
332 		memset(&in_params, 0, sizeof(in_params));
333 		in_params.srq_id = srq->srq_id;
334 		in_params.wqe_limit = attr->srq_limit;
335 		ret = ecore_rdma_modify_srq(dev->rdma_ctx, &in_params);
336 		if (ret)
337 			return ret;
338 	}
339 
340 	QL_DPRINT12(ha, "modified srq with srq_id = 0x%0x\n", srq->srq_id);
341 	return 0;
342 }
343 
344 int
345 qlnxr_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
346 {
347 	struct qlnxr_dev	*dev;
348 	struct qlnxr_srq	*srq;
349 	qlnx_host_t		*ha;
350 	struct ecore_rdma_device *qattr;
351 	srq = get_qlnxr_srq(ibsrq);
352 	dev = srq->dev;
353 	ha = dev->ha;
354 	//qattr = &dev->attr;
355 	qattr = ecore_rdma_query_device(dev->rdma_ctx);
356 	QL_DPRINT12(ha, "enter\n");
357 
358 	if (!dev->rdma_ctx) {
359 		QL_DPRINT12(ha, "called with invalid params"
360 			" rdma_ctx is NULL\n");
361 		return -EINVAL;
362 	}
363 
364 	srq_attr->srq_limit = qattr->max_srq;
365 	srq_attr->max_wr = qattr->max_srq_wr;
366 	srq_attr->max_sge = qattr->max_sge;
367 
368 	QL_DPRINT12(ha, "exit\n");
369 	return 0;
370 }
371 
372 /* Increment srq wr producer by one */
373 static
374 void qlnxr_inc_srq_wr_prod (struct qlnxr_srq_hwq_info *info)
375 {
376 	info->wr_prod_cnt++;
377 }
378 
379 /* Increment srq wr consumer by one */
380 static
381 void qlnxr_inc_srq_wr_cons(struct qlnxr_srq_hwq_info *info)
382 {
383         info->wr_cons_cnt++;
384 }
385 
386 /* get_port_immutable verb is not available in FreeBSD */
387 #if 0
388 int
389 qlnxr_roce_port_immutable(struct ib_device *ibdev, u8 port_num,
390 	struct ib_port_immutable *immutable)
391 {
392 	struct qlnxr_dev                *dev;
393 	qlnx_host_t                     *ha;
394 	dev = get_qlnxr_dev(ibdev);
395 	ha = dev->ha;
396 
397 	QL_DPRINT12(ha, "entered but not implemented!!!\n");
398 }
399 #endif
400 
401 int
402 qlnxr_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
403 	struct ib_recv_wr **bad_wr)
404 {
405 	struct qlnxr_dev	*dev;
406 	struct qlnxr_srq	*srq;
407 	qlnx_host_t		*ha;
408 	struct qlnxr_srq_hwq_info *hw_srq;
409 	struct ecore_chain *pbl;
410 	unsigned long flags;
411 	int status = 0;
412 	u32 num_sge, offset;
413 
414 	srq = get_qlnxr_srq(ibsrq);
415 	dev = srq->dev;
416 	ha = dev->ha;
417 	hw_srq = &srq->hw_srq;
418 
419 	QL_DPRINT12(ha, "enter\n");
420 	spin_lock_irqsave(&srq->lock, flags);
421 
422 	pbl = &srq->hw_srq.pbl;
423 	while (wr) {
424 		struct rdma_srq_wqe_header *hdr;
425 		int i;
426 
427 		if (!qlnxr_srq_elem_left(hw_srq) ||
428 		    wr->num_sge > srq->hw_srq.max_sges) {
429 			QL_DPRINT11(ha, "WR cannot be posted"
430 			    " (%d, %d) || (%d > %d)\n",
431 			    hw_srq->wr_prod_cnt, hw_srq->wr_cons_cnt,
432 			    wr->num_sge, srq->hw_srq.max_sges);
433 			status = -ENOMEM;
434 			*bad_wr = wr;
435 			break;
436 		}
437 
438 		hdr = ecore_chain_produce(pbl);
439 		num_sge = wr->num_sge;
440 		/* Set number of sge and WR id in header */
441 		SRQ_HDR_SET(hdr, wr->wr_id, num_sge);
442 
443                 /* PBL is maintained in case of WR granularity.
444                  * So increment WR producer in case we post a WR.
445                  */
446 		qlnxr_inc_srq_wr_prod(hw_srq);
447 		hw_srq->wqe_prod++;
448 		hw_srq->sge_prod++;
449 
450 		QL_DPRINT12(ha, "SRQ WR : SGEs: %d with wr_id[%d] = %llx\n",
451 			wr->num_sge, hw_srq->wqe_prod, wr->wr_id);
452 
453 		for (i = 0; i < wr->num_sge; i++) {
454 			struct rdma_srq_sge *srq_sge =
455 			    ecore_chain_produce(pbl);
456 			/* Set SGE length, lkey and address */
457 			SRQ_SGE_SET(srq_sge, wr->sg_list[i].addr,
458 				wr->sg_list[i].length, wr->sg_list[i].lkey);
459 
460 			QL_DPRINT12(ha, "[%d]: len %d, key %x, addr %x:%x\n",
461 				i, srq_sge->length, srq_sge->l_key,
462 				srq_sge->addr.hi, srq_sge->addr.lo);
463 			hw_srq->sge_prod++;
464 		}
465 		wmb();
466 		/*
467 		 * SRQ prod is 8 bytes. Need to update SGE prod in index
468 		 * in first 4 bytes and need to update WQE prod in next
469 		 * 4 bytes.
470 		 */
471 		*(srq->hw_srq.virt_prod_pair_addr) = hw_srq->sge_prod;
472 		offset = offsetof(struct rdma_srq_producers, wqe_prod);
473 		*((u8 *)srq->hw_srq.virt_prod_pair_addr + offset) =
474 			hw_srq->wqe_prod;
475 		/* Flush prod after updating it */
476 		wmb();
477 		wr = wr->next;
478 	}
479 
480 	QL_DPRINT12(ha, "Elements in SRQ: %d\n",
481 		ecore_chain_get_elem_left(pbl));
482 
483 	spin_unlock_irqrestore(&srq->lock, flags);
484 	QL_DPRINT12(ha, "exit\n");
485 	return status;
486 }
487 
488 int
489 #if __FreeBSD_version < 1102000
490 qlnxr_query_device(struct ib_device *ibdev, struct ib_device_attr *attr)
491 #else
492 qlnxr_query_device(struct ib_device *ibdev, struct ib_device_attr *attr,
493 	struct ib_udata *udata)
494 #endif /* #if __FreeBSD_version < 1102000 */
495 
496 {
497 	struct qlnxr_dev		*dev;
498 	struct ecore_rdma_device	*qattr;
499 	qlnx_host_t			*ha;
500 
501 	dev = get_qlnxr_dev(ibdev);
502 	ha = dev->ha;
503 
504 	QL_DPRINT12(ha, "enter\n");
505 
506 #if __FreeBSD_version > 1102000
507 	if (udata->inlen || udata->outlen)
508 		return -EINVAL;
509 #endif /* #if __FreeBSD_version > 1102000 */
510 
511 	if (dev->rdma_ctx == NULL) {
512 		return -EINVAL;
513 	}
514 
515 	qattr = ecore_rdma_query_device(dev->rdma_ctx);
516 
517 	memset(attr, 0, sizeof *attr);
518 
519 	attr->fw_ver = qattr->fw_ver;
520 	attr->sys_image_guid = qattr->sys_image_guid;
521 	attr->max_mr_size = qattr->max_mr_size;
522 	attr->page_size_cap = qattr->page_size_caps;
523 	attr->vendor_id = qattr->vendor_id;
524 	attr->vendor_part_id = qattr->vendor_part_id;
525 	attr->hw_ver = qattr->hw_ver;
526 	attr->max_qp = qattr->max_qp;
527 	attr->device_cap_flags = IB_DEVICE_CURR_QP_STATE_MOD |
528 					IB_DEVICE_RC_RNR_NAK_GEN |
529 					IB_DEVICE_LOCAL_DMA_LKEY |
530 					IB_DEVICE_MEM_MGT_EXTENSIONS;
531 
532 	attr->max_sge = qattr->max_sge;
533 	attr->max_sge_rd = qattr->max_sge;
534 	attr->max_cq = qattr->max_cq;
535 	attr->max_cqe = qattr->max_cqe;
536 	attr->max_mr = qattr->max_mr;
537 	attr->max_mw = qattr->max_mw;
538 	attr->max_pd = qattr->max_pd;
539 	attr->atomic_cap = dev->atomic_cap;
540 	attr->max_fmr = qattr->max_fmr;
541 	attr->max_map_per_fmr = 16; /* TBD: FMR */
542 
543 	/* There is an implicit assumption in some of the ib_xxx apps that the
544 	 * qp_rd_atom is smaller than the qp_init_rd_atom. Specifically, in
545 	 * communication the qp_rd_atom is passed to the other side and used as
546 	 * init_rd_atom without check device capabilities for init_rd_atom.
547 	 * for this reason, we set the qp_rd_atom to be the minimum between the
548 	 * two...There is an additional assumption in mlx4 driver that the
549 	 * values are power of two, fls is performed on the value - 1, which
550 	 * in fact gives a larger power of two for values which are not a power
551 	 * of two. This should be fixed in mlx4 driver, but until then ->
552 	 * we provide a value that is a power of two in our code.
553 	 */
554 	attr->max_qp_init_rd_atom =
555 		1 << (fls(qattr->max_qp_req_rd_atomic_resc) - 1);
556 	attr->max_qp_rd_atom =
557 		min(1 << (fls(qattr->max_qp_resp_rd_atomic_resc) - 1),
558 		    attr->max_qp_init_rd_atom);
559 
560 	attr->max_srq = qattr->max_srq;
561 	attr->max_srq_sge = qattr->max_srq_sge;
562 	attr->max_srq_wr = qattr->max_srq_wr;
563 
564 	/* TODO: R&D to more properly configure the following */
565 	attr->local_ca_ack_delay = qattr->dev_ack_delay;
566 	attr->max_fast_reg_page_list_len = qattr->max_mr/8;
567 	attr->max_pkeys = QLNXR_ROCE_PKEY_MAX;
568 	attr->max_ah = qattr->max_ah;
569 
570 	QL_DPRINT12(ha, "exit\n");
571 	return 0;
572 }
573 
574 static inline void
575 get_link_speed_and_width(int speed, uint8_t *ib_speed, uint8_t *ib_width)
576 {
577 	switch (speed) {
578 	case 1000:
579 		*ib_speed = IB_SPEED_SDR;
580 		*ib_width = IB_WIDTH_1X;
581 		break;
582 	case 10000:
583 		*ib_speed = IB_SPEED_QDR;
584 		*ib_width = IB_WIDTH_1X;
585 		break;
586 
587 	case 20000:
588 		*ib_speed = IB_SPEED_DDR;
589 		*ib_width = IB_WIDTH_4X;
590 		break;
591 
592 	case 25000:
593 		*ib_speed = IB_SPEED_EDR;
594 		*ib_width = IB_WIDTH_1X;
595 		break;
596 
597 	case 40000:
598 		*ib_speed = IB_SPEED_QDR;
599 		*ib_width = IB_WIDTH_4X;
600 		break;
601 
602 	case 50000:
603 		*ib_speed = IB_SPEED_QDR;
604 		*ib_width = IB_WIDTH_4X; // TODO doesn't add up to 50...
605 		break;
606 
607 	case 100000:
608 		*ib_speed = IB_SPEED_EDR;
609 		*ib_width = IB_WIDTH_4X;
610 		break;
611 
612 	default:
613 		/* Unsupported */
614 		*ib_speed = IB_SPEED_SDR;
615 		*ib_width = IB_WIDTH_1X;
616 	}
617 	return;
618 }
619 
620 int
621 qlnxr_query_port(struct ib_device *ibdev, uint8_t port,
622 	struct ib_port_attr *attr)
623 {
624 	struct qlnxr_dev	*dev;
625 	struct ecore_rdma_port	*rdma_port;
626 	qlnx_host_t		*ha;
627 
628 	dev = get_qlnxr_dev(ibdev);
629 	ha = dev->ha;
630 
631 	QL_DPRINT12(ha, "enter\n");
632 
633 	if (port > 1) {
634 		QL_DPRINT12(ha, "port [%d] > 1 \n", port);
635 		return -EINVAL;
636 	}
637 
638 	if (dev->rdma_ctx == NULL) {
639 		QL_DPRINT12(ha, "rdma_ctx == NULL\n");
640 		return -EINVAL;
641 	}
642 
643 	rdma_port = ecore_rdma_query_port(dev->rdma_ctx);
644 	memset(attr, 0, sizeof *attr);
645 
646 	if (rdma_port->port_state == ECORE_RDMA_PORT_UP) {
647 		attr->state = IB_PORT_ACTIVE;
648 		attr->phys_state = 5;
649 	} else {
650 		attr->state = IB_PORT_DOWN;
651 		attr->phys_state = 3;
652 	}
653 
654 	attr->max_mtu = IB_MTU_4096;
655 	attr->active_mtu = iboe_get_mtu(dev->ha->ifp->if_mtu);
656 	attr->lid = 0;
657 	attr->lmc = 0;
658 	attr->sm_lid = 0;
659 	attr->sm_sl = 0;
660 	attr->port_cap_flags = 0;
661 
662 	if (QLNX_IS_IWARP(dev)) {
663 		attr->gid_tbl_len = 1;
664 		attr->pkey_tbl_len = 1;
665 	} else {
666 		attr->gid_tbl_len = QLNXR_MAX_SGID;
667 		attr->pkey_tbl_len = QLNXR_ROCE_PKEY_TABLE_LEN;
668 	}
669 
670 	attr->bad_pkey_cntr = rdma_port->pkey_bad_counter;
671 	attr->qkey_viol_cntr = 0;
672 
673 	get_link_speed_and_width(rdma_port->link_speed,
674 				 &attr->active_speed, &attr->active_width);
675 
676 	attr->max_msg_sz = rdma_port->max_msg_size;
677 	attr->max_vl_num = 4; /* TODO -> figure this one out... */
678 
679 	QL_DPRINT12(ha, "state = %d phys_state = %d "
680 		" link_speed = %d active_speed = %d active_width = %d"
681 		" attr->gid_tbl_len = %d attr->pkey_tbl_len = %d"
682 		" max_msg_sz = 0x%x max_vl_num = 0x%x \n",
683 		attr->state, attr->phys_state,
684 		rdma_port->link_speed, attr->active_speed,
685 		attr->active_width, attr->gid_tbl_len, attr->pkey_tbl_len,
686 		attr->max_msg_sz, attr->max_vl_num);
687 
688 	QL_DPRINT12(ha, "exit\n");
689 	return 0;
690 }
691 
692 int
693 qlnxr_modify_port(struct ib_device *ibdev, uint8_t port, int mask,
694 	struct ib_port_modify *props)
695 {
696 	struct qlnxr_dev	*dev;
697 	qlnx_host_t		*ha;
698 
699 	dev = get_qlnxr_dev(ibdev);
700 	ha = dev->ha;
701 
702 	QL_DPRINT12(ha, "enter\n");
703 
704 	if (port > 1) {
705 		QL_DPRINT12(ha, "port (%d) > 1\n", port);
706 		return -EINVAL;
707 	}
708 
709 	QL_DPRINT12(ha, "exit\n");
710 	return 0;
711 }
712 
713 enum rdma_link_layer
714 qlnxr_link_layer(struct ib_device *ibdev, uint8_t port_num)
715 {
716 	struct qlnxr_dev	*dev;
717 	qlnx_host_t		*ha;
718 
719 	dev = get_qlnxr_dev(ibdev);
720 	ha = dev->ha;
721 
722 	QL_DPRINT12(ha, "ibdev = %p port_num = 0x%x\n", ibdev, port_num);
723 
724         return IB_LINK_LAYER_ETHERNET;
725 }
726 
727 struct ib_pd *
728 qlnxr_alloc_pd(struct ib_device *ibdev, struct ib_ucontext *context,
729 	struct ib_udata *udata)
730 {
731 	struct qlnxr_pd		*pd = NULL;
732 	u16			pd_id;
733 	int			rc;
734 	struct qlnxr_dev	*dev;
735 	qlnx_host_t		*ha;
736 
737 	dev = get_qlnxr_dev(ibdev);
738 	ha = dev->ha;
739 
740 	QL_DPRINT12(ha, "ibdev = %p context = %p"
741 		" udata = %p enter\n", ibdev, context, udata);
742 
743 	if (dev->rdma_ctx == NULL) {
744 		QL_DPRINT11(ha, "dev->rdma_ctx = NULL\n");
745 		rc = -1;
746 		goto err;
747 	}
748 
749 	pd = kzalloc(sizeof(*pd), GFP_KERNEL);
750 	if (!pd) {
751 		rc = -ENOMEM;
752 		QL_DPRINT11(ha, "kzalloc(pd) = NULL\n");
753 		goto err;
754 	}
755 
756 	rc = ecore_rdma_alloc_pd(dev->rdma_ctx, &pd_id);
757 	if (rc)	{
758 		QL_DPRINT11(ha, "ecore_rdma_alloc_pd failed\n");
759 		goto err;
760 	}
761 
762 	pd->pd_id = pd_id;
763 
764 	if (udata && context) {
765 
766 		rc = ib_copy_to_udata(udata, &pd->pd_id, sizeof(pd->pd_id));
767 		if (rc) {
768 			QL_DPRINT11(ha, "ib_copy_to_udata failed\n");
769 			ecore_rdma_free_pd(dev->rdma_ctx, pd_id);
770 			goto err;
771 		}
772 
773 		pd->uctx = get_qlnxr_ucontext(context);
774 		pd->uctx->pd = pd;
775 	}
776 
777 	atomic_add_rel_32(&dev->pd_count, 1);
778 	QL_DPRINT12(ha, "exit [pd, pd_id, pd_count] = [%p, 0x%x, %d]\n",
779 		pd, pd_id, dev->pd_count);
780 
781 	return &pd->ibpd;
782 
783 err:
784 	kfree(pd);
785 	QL_DPRINT12(ha, "exit -1\n");
786 	return ERR_PTR(rc);
787 }
788 
789 int
790 qlnxr_dealloc_pd(struct ib_pd *ibpd)
791 {
792 	struct qlnxr_pd		*pd;
793 	struct qlnxr_dev	*dev;
794 	qlnx_host_t		*ha;
795 
796 	pd = get_qlnxr_pd(ibpd);
797 	dev = get_qlnxr_dev((ibpd->device));
798 	ha = dev->ha;
799 
800 	QL_DPRINT12(ha, "enter\n");
801 
802 	if (pd == NULL) {
803 		QL_DPRINT11(ha, "pd = NULL\n");
804 	} else {
805 		ecore_rdma_free_pd(dev->rdma_ctx, pd->pd_id);
806 		kfree(pd);
807 		atomic_subtract_rel_32(&dev->pd_count, 1);
808 		QL_DPRINT12(ha, "exit [pd, pd_id, pd_count] = [%p, 0x%x, %d]\n",
809 			pd, pd->pd_id, dev->pd_count);
810 	}
811 
812 	QL_DPRINT12(ha, "exit\n");
813 	return 0;
814 }
815 
816 #define ROCE_WQE_ELEM_SIZE	sizeof(struct rdma_sq_sge)
817 #define	RDMA_MAX_SGE_PER_SRQ	(4) /* Should be part of HSI */
818 /* Should be part of HSI */
819 #define RDMA_MAX_SRQ_WQE_SIZE	(RDMA_MAX_SGE_PER_SRQ + 1) /* +1 for header */
820 #define DB_ADDR_SHIFT(addr)		((addr) << DB_PWM_ADDR_OFFSET_SHIFT)
821 
822 static void qlnxr_cleanup_user(struct qlnxr_dev *, struct qlnxr_qp *);
823 static void qlnxr_cleanup_kernel(struct qlnxr_dev *, struct qlnxr_qp *);
824 
825 int
826 qlnxr_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
827 {
828 	struct qlnxr_dev	*dev;
829 	qlnx_host_t		*ha;
830 
831 	dev = get_qlnxr_dev(ibdev);
832 	ha = dev->ha;
833 
834 	QL_DPRINT12(ha, "enter index = 0x%x\n", index);
835 
836 	if (index > QLNXR_ROCE_PKEY_TABLE_LEN)
837 		return -EINVAL;
838 
839 	*pkey = QLNXR_ROCE_PKEY_DEFAULT;
840 
841 	QL_DPRINT12(ha, "exit\n");
842 	return 0;
843 }
844 
845 
846 static inline bool
847 qlnxr_get_vlan_id_qp(qlnx_host_t *ha, struct ib_qp_attr *attr, int attr_mask,
848        u16 *vlan_id)
849 {
850 	bool ret = false;
851 
852 	QL_DPRINT12(ha, "enter \n");
853 
854 	*vlan_id = 0;
855 
856 #if __FreeBSD_version >= 1100000
857 	u16 tmp_vlan_id;
858 
859 #if __FreeBSD_version >= 1102000
860 	union ib_gid *dgid;
861 
862 	dgid = &attr->ah_attr.grh.dgid;
863 	tmp_vlan_id = (dgid->raw[11] << 8) | dgid->raw[12];
864 
865 	if (!(tmp_vlan_id & ~EVL_VLID_MASK)) {
866 		*vlan_id = tmp_vlan_id;
867 		ret = true;
868 	}
869 #else
870 	tmp_vlan_id = attr->vlan_id;
871 
872 	if ((attr_mask & IB_QP_VID) && (!(tmp_vlan_id & ~EVL_VLID_MASK))) {
873 		*vlan_id = tmp_vlan_id;
874 		ret = true;
875 	}
876 
877 #endif /* #if __FreeBSD_version > 1102000 */
878 
879 #else
880 	ret = true;
881 
882 #endif /* #if __FreeBSD_version >= 1100000 */
883 
884 	QL_DPRINT12(ha, "exit vlan_id = 0x%x ret = %d \n", *vlan_id, ret);
885 
886 	return (ret);
887 }
888 
889 static inline void
890 get_gid_info(struct ib_qp *ibqp, struct ib_qp_attr *attr,
891 	int attr_mask,
892 	struct qlnxr_dev *dev,
893 	struct qlnxr_qp *qp,
894 	struct ecore_rdma_modify_qp_in_params *qp_params)
895 {
896 	int		i;
897 	qlnx_host_t	*ha;
898 
899 	ha = dev->ha;
900 
901 	QL_DPRINT12(ha, "enter\n");
902 
903 	memcpy(&qp_params->sgid.bytes[0],
904 	       &dev->sgid_tbl[qp->sgid_idx].raw[0],
905 	       sizeof(qp_params->sgid.bytes));
906 	memcpy(&qp_params->dgid.bytes[0],
907 	       &attr->ah_attr.grh.dgid.raw[0],
908 	       sizeof(qp_params->dgid));
909 
910 	qlnxr_get_vlan_id_qp(ha, attr, attr_mask, &qp_params->vlan_id);
911 
912 	for (i = 0; i < (sizeof(qp_params->sgid.dwords)/sizeof(uint32_t)); i++) {
913 		qp_params->sgid.dwords[i] = ntohl(qp_params->sgid.dwords[i]);
914 		qp_params->dgid.dwords[i] = ntohl(qp_params->dgid.dwords[i]);
915 	}
916 
917 	QL_DPRINT12(ha, "exit\n");
918 	return;
919 }
920 
921 
922 
923 static int
924 qlnxr_add_mmap(struct qlnxr_ucontext *uctx, u64 phy_addr, unsigned long len)
925 {
926 	struct qlnxr_mm	*mm;
927 	qlnx_host_t	*ha;
928 
929 	ha = uctx->dev->ha;
930 
931 	QL_DPRINT12(ha, "enter\n");
932 
933 	mm = kzalloc(sizeof(*mm), GFP_KERNEL);
934 	if (mm == NULL) {
935 		QL_DPRINT11(ha, "mm = NULL\n");
936 		return -ENOMEM;
937 	}
938 
939 	mm->key.phy_addr = phy_addr;
940 
941 	/* This function might be called with a length which is not a multiple
942 	 * of PAGE_SIZE, while the mapping is PAGE_SIZE grained and the kernel
943 	 * forces this granularity by increasing the requested size if needed.
944 	 * When qedr_mmap is called, it will search the list with the updated
945 	 * length as a key. To prevent search failures, the length is rounded up
946 	 * in advance to PAGE_SIZE.
947 	 */
948 	mm->key.len = roundup(len, PAGE_SIZE);
949 	INIT_LIST_HEAD(&mm->entry);
950 
951 	mutex_lock(&uctx->mm_list_lock);
952 	list_add(&mm->entry, &uctx->mm_head);
953 	mutex_unlock(&uctx->mm_list_lock);
954 
955 	QL_DPRINT12(ha, "added (addr=0x%llx,len=0x%lx) for ctx=%p\n",
956 		(unsigned long long)mm->key.phy_addr,
957 		(unsigned long)mm->key.len, uctx);
958 
959 	return 0;
960 }
961 
962 static bool
963 qlnxr_search_mmap(struct qlnxr_ucontext *uctx, u64 phy_addr, unsigned long len)
964 {
965 	bool		found = false;
966 	struct qlnxr_mm	*mm;
967 	qlnx_host_t	*ha;
968 
969 	ha = uctx->dev->ha;
970 
971 	QL_DPRINT12(ha, "enter\n");
972 
973 	mutex_lock(&uctx->mm_list_lock);
974 	list_for_each_entry(mm, &uctx->mm_head, entry) {
975 		if (len != mm->key.len || phy_addr != mm->key.phy_addr)
976 			continue;
977 
978 		found = true;
979 		break;
980 	}
981 	mutex_unlock(&uctx->mm_list_lock);
982 
983 	QL_DPRINT12(ha,
984 		"searched for (addr=0x%llx,len=0x%lx) for ctx=%p, found=%d\n",
985 		mm->key.phy_addr, mm->key.len, uctx, found);
986 
987 	return found;
988 }
989 
990 struct
991 ib_ucontext *qlnxr_alloc_ucontext(struct ib_device *ibdev,
992                 struct ib_udata *udata)
993 {
994         int rc;
995         struct qlnxr_ucontext *ctx;
996         struct qlnxr_alloc_ucontext_resp uresp;
997         struct qlnxr_dev *dev = get_qlnxr_dev(ibdev);
998         qlnx_host_t *ha = dev->ha;
999         struct ecore_rdma_add_user_out_params oparams;
1000 
1001         if (!udata) {
1002                 return ERR_PTR(-EFAULT);
1003         }
1004 
1005 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1006 	if (!ctx)
1007 		return ERR_PTR(-ENOMEM);
1008 
1009 	rc = ecore_rdma_add_user(dev->rdma_ctx, &oparams);
1010 	if (rc) {
1011 		QL_DPRINT12(ha,
1012 			"Failed to allocate a DPI for a new RoCE application "
1013 			",rc = %d. To overcome this, consider to increase "
1014 			"the number of DPIs, increase the doorbell BAR size "
1015 			"or just close unnecessary RoCE applications. In "
1016 			"order to increase the number of DPIs consult the "
1017 			"README\n", rc);
1018 		goto err;
1019 	}
1020 
1021 	ctx->dpi = oparams.dpi;
1022 	ctx->dpi_addr = oparams.dpi_addr;
1023 	ctx->dpi_phys_addr = oparams.dpi_phys_addr;
1024 	ctx->dpi_size = oparams.dpi_size;
1025 	INIT_LIST_HEAD(&ctx->mm_head);
1026 	mutex_init(&ctx->mm_list_lock);
1027 
1028 	memset(&uresp, 0, sizeof(uresp));
1029 	uresp.dpm_enabled = offsetof(struct qlnxr_alloc_ucontext_resp, dpm_enabled)
1030 				< udata->outlen ? dev->user_dpm_enabled : 0; //TODO: figure this out
1031 	uresp.wids_enabled = offsetof(struct qlnxr_alloc_ucontext_resp, wids_enabled)
1032 				< udata->outlen ? 1 : 0; //TODO: figure this out
1033 	uresp.wid_count = offsetof(struct qlnxr_alloc_ucontext_resp, wid_count)
1034 				< udata->outlen ? oparams.wid_count : 0; //TODO: figure this out
1035         uresp.db_pa = ctx->dpi_phys_addr;
1036         uresp.db_size = ctx->dpi_size;
1037         uresp.max_send_wr = dev->attr.max_sqe;
1038         uresp.max_recv_wr = dev->attr.max_rqe;
1039         uresp.max_srq_wr = dev->attr.max_srq_wr;
1040         uresp.sges_per_send_wr = QLNXR_MAX_SQE_ELEMENTS_PER_SQE;
1041         uresp.sges_per_recv_wr = QLNXR_MAX_RQE_ELEMENTS_PER_RQE;
1042         uresp.sges_per_srq_wr = dev->attr.max_srq_sge;
1043         uresp.max_cqes = QLNXR_MAX_CQES;
1044 
1045 	rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
1046 	if (rc)
1047 		goto err;
1048 
1049 	ctx->dev = dev;
1050 
1051 	rc = qlnxr_add_mmap(ctx, ctx->dpi_phys_addr, ctx->dpi_size);
1052 	if (rc)
1053 		goto err;
1054 	QL_DPRINT12(ha, "Allocated user context %p\n",
1055 		&ctx->ibucontext);
1056 
1057 	return &ctx->ibucontext;
1058 err:
1059 	kfree(ctx);
1060 	return ERR_PTR(rc);
1061 }
1062 
1063 int
1064 qlnxr_dealloc_ucontext(struct ib_ucontext *ibctx)
1065 {
1066         struct qlnxr_ucontext *uctx = get_qlnxr_ucontext(ibctx);
1067         struct qlnxr_dev *dev = uctx->dev;
1068         qlnx_host_t *ha = dev->ha;
1069         struct qlnxr_mm *mm, *tmp;
1070         int status = 0;
1071 
1072         QL_DPRINT12(ha, "Deallocating user context %p\n",
1073                         uctx);
1074 
1075         if (dev) {
1076                 ecore_rdma_remove_user(uctx->dev->rdma_ctx, uctx->dpi);
1077         }
1078 
1079         list_for_each_entry_safe(mm, tmp, &uctx->mm_head, entry) {
1080                 QL_DPRINT12(ha, "deleted addr= 0x%llx, len = 0x%lx for"
1081                                 " ctx=%p\n",
1082                                 mm->key.phy_addr, mm->key.len, uctx);
1083                 list_del(&mm->entry);
1084                 kfree(mm);
1085         }
1086         kfree(uctx);
1087         return status;
1088 }
1089 
1090 int
1091 qlnxr_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1092 {
1093 	struct qlnxr_ucontext	*ucontext = get_qlnxr_ucontext(context);
1094 	struct qlnxr_dev	*dev = get_qlnxr_dev((context->device));
1095 	unsigned long		vm_page = vma->vm_pgoff << PAGE_SHIFT;
1096 	u64 			unmapped_db;
1097 	unsigned long 		len = (vma->vm_end - vma->vm_start);
1098 	int 			rc = 0;
1099 	bool 			found;
1100 	qlnx_host_t		*ha;
1101 
1102 	ha = dev->ha;
1103 
1104 #if __FreeBSD_version > 1102000
1105 	unmapped_db = dev->db_phys_addr + (ucontext->dpi * ucontext->dpi_size);
1106 #else
1107 	unmapped_db = dev->db_phys_addr;
1108 #endif /* #if __FreeBSD_version > 1102000 */
1109 
1110 	QL_DPRINT12(ha, "qedr_mmap enter vm_page=0x%lx"
1111 		" vm_pgoff=0x%lx unmapped_db=0x%llx db_size=%x, len=%lx\n",
1112 		vm_page, vma->vm_pgoff, unmapped_db,
1113 		dev->db_size, len);
1114 
1115 	if ((vma->vm_start & (PAGE_SIZE - 1)) || (len & (PAGE_SIZE - 1))) {
1116 		QL_DPRINT11(ha, "Vma_start not page aligned "
1117 			"vm_start = %ld vma_end = %ld\n", vma->vm_start,
1118 			vma->vm_end);
1119 		return -EINVAL;
1120 	}
1121 
1122 	found = qlnxr_search_mmap(ucontext, vm_page, len);
1123 	if (!found) {
1124 		QL_DPRINT11(ha, "Vma_pgoff not found in mapped array = %ld\n",
1125 			vma->vm_pgoff);
1126 		return -EINVAL;
1127 	}
1128 
1129 	QL_DPRINT12(ha, "Mapping doorbell bar\n");
1130 
1131 #if __FreeBSD_version > 1102000
1132 
1133 	if ((vm_page < unmapped_db) ||
1134 		((vm_page + len) > (unmapped_db + ucontext->dpi_size))) {
1135 		QL_DPRINT11(ha, "failed pages are outside of dpi;"
1136 			"page address=0x%lx, unmapped_db=0x%lx, dpi_size=0x%x\n",
1137 			vm_page, unmapped_db, ucontext->dpi_size);
1138 		return -EINVAL;
1139 	}
1140 
1141 	if (vma->vm_flags & VM_READ) {
1142 		QL_DPRINT11(ha, "failed mmap, cannot map doorbell bar for read\n");
1143 		return -EINVAL;
1144 	}
1145 
1146 	vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1147 	rc = io_remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff, len,
1148 			vma->vm_page_prot);
1149 
1150 #else
1151 
1152 	if ((vm_page >= unmapped_db) && (vm_page <= (unmapped_db +
1153 		dev->db_size))) {
1154 
1155 		QL_DPRINT12(ha, "Mapping doorbell bar\n");
1156 
1157 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1158 
1159 		rc = io_remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
1160 					    PAGE_SIZE, vma->vm_page_prot);
1161 	} else {
1162 		QL_DPRINT12(ha, "Mapping chains\n");
1163 		rc = io_remap_pfn_range(vma, vma->vm_start,
1164 					 vma->vm_pgoff, len, vma->vm_page_prot);
1165 	}
1166 
1167 #endif /* #if __FreeBSD_version > 1102000 */
1168 
1169 	QL_DPRINT12(ha, "exit [%d]\n", rc);
1170 	return rc;
1171 }
1172 
1173 struct ib_mr *
1174 qlnxr_get_dma_mr(struct ib_pd *ibpd, int acc)
1175 {
1176 	struct qlnxr_mr		*mr;
1177 	struct qlnxr_dev	*dev = get_qlnxr_dev((ibpd->device));
1178 	struct qlnxr_pd		*pd = get_qlnxr_pd(ibpd);
1179 	int			rc;
1180 	qlnx_host_t		*ha;
1181 
1182 	ha = dev->ha;
1183 
1184 	QL_DPRINT12(ha, "enter\n");
1185 
1186 	if (acc & IB_ACCESS_MW_BIND) {
1187 		QL_DPRINT12(ha, "Unsupported access flags received for dma mr\n");
1188 	}
1189 
1190 	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
1191 	if (!mr) {
1192 		rc = -ENOMEM;
1193 		QL_DPRINT12(ha, "kzalloc(mr) failed %d\n", rc);
1194 		goto err0;
1195 	}
1196 
1197 	mr->type = QLNXR_MR_DMA;
1198 
1199 	rc = ecore_rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
1200 	if (rc) {
1201 		QL_DPRINT12(ha, "ecore_rdma_alloc_tid failed %d\n", rc);
1202 		goto err1;
1203 	}
1204 
1205 	/* index only, 18 bit long, lkey = itid << 8 | key */
1206 	mr->hw_mr.tid_type = ECORE_RDMA_TID_REGISTERED_MR;
1207 	mr->hw_mr.pd = pd->pd_id;
1208 	mr->hw_mr.local_read = 1;
1209 	mr->hw_mr.local_write = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0;
1210 	mr->hw_mr.remote_read = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0;
1211 	mr->hw_mr.remote_write = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
1212 	mr->hw_mr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0;
1213 	mr->hw_mr.dma_mr = true;
1214 
1215 	rc = ecore_rdma_register_tid(dev->rdma_ctx, &mr->hw_mr);
1216 	if (rc) {
1217 		QL_DPRINT12(ha, "ecore_rdma_register_tid failed %d\n", rc);
1218 		goto err2;
1219 	}
1220 
1221 	mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
1222 
1223 	if (mr->hw_mr.remote_write || mr->hw_mr.remote_read ||
1224 		mr->hw_mr.remote_atomic) {
1225 		mr->ibmr.rkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
1226 	}
1227 
1228 	QL_DPRINT12(ha, "lkey = %x\n", mr->ibmr.lkey);
1229 
1230 	return &mr->ibmr;
1231 
1232 err2:
1233 	ecore_rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
1234 err1:
1235 	kfree(mr);
1236 err0:
1237 	QL_DPRINT12(ha, "exit [%d]\n", rc);
1238 
1239 	return ERR_PTR(rc);
1240 }
1241 
1242 static void
1243 qlnxr_free_pbl(struct qlnxr_dev *dev, struct qlnxr_pbl_info *pbl_info,
1244 	struct qlnxr_pbl *pbl)
1245 {
1246 	int		i;
1247 	qlnx_host_t	*ha;
1248 
1249 	ha = dev->ha;
1250 
1251 	QL_DPRINT12(ha, "enter\n");
1252 
1253 	for (i = 0; i < pbl_info->num_pbls; i++) {
1254 		if (!pbl[i].va)
1255 			continue;
1256 		qlnx_dma_free_coherent(&dev->ha->cdev, pbl[i].va, pbl[i].pa,
1257 			pbl_info->pbl_size);
1258 	}
1259 	kfree(pbl);
1260 
1261 	QL_DPRINT12(ha, "exit\n");
1262 	return;
1263 }
1264 
1265 #define MIN_FW_PBL_PAGE_SIZE (4*1024)
1266 #define MAX_FW_PBL_PAGE_SIZE (64*1024)
1267 
1268 #define NUM_PBES_ON_PAGE(_page_size) (_page_size / sizeof(u64))
1269 #define MAX_PBES_ON_PAGE NUM_PBES_ON_PAGE(MAX_FW_PBL_PAGE_SIZE)
1270 #define MAX_PBES_TWO_LAYER (MAX_PBES_ON_PAGE*MAX_PBES_ON_PAGE)
1271 
1272 static struct qlnxr_pbl *
1273 qlnxr_alloc_pbl_tbl(struct qlnxr_dev *dev,
1274 	struct qlnxr_pbl_info *pbl_info, gfp_t flags)
1275 {
1276 	void			*va;
1277 	dma_addr_t		pa;
1278 	dma_addr_t		*pbl_main_tbl;
1279 	struct qlnxr_pbl	*pbl_table;
1280 	int			i, rc = 0;
1281 	qlnx_host_t		*ha;
1282 
1283 	ha = dev->ha;
1284 
1285 	QL_DPRINT12(ha, "enter\n");
1286 
1287 	pbl_table = kzalloc(sizeof(*pbl_table) * pbl_info->num_pbls, flags);
1288 
1289 	if (!pbl_table) {
1290 		QL_DPRINT12(ha, "pbl_table = NULL\n");
1291 		return NULL;
1292 	}
1293 
1294 	for (i = 0; i < pbl_info->num_pbls; i++) {
1295 		va = qlnx_dma_alloc_coherent(&dev->ha->cdev, &pa, pbl_info->pbl_size);
1296 		if (!va) {
1297 			QL_DPRINT11(ha, "Failed to allocate pbl#%d\n", i);
1298 			rc = -ENOMEM;
1299 			goto err;
1300 		}
1301 		memset(va, 0, pbl_info->pbl_size);
1302 		pbl_table[i].va = va;
1303 		pbl_table[i].pa = pa;
1304 	}
1305 
1306 	/* Two-Layer PBLs, if we have more than one pbl we need to initialize
1307 	 * the first one with physical pointers to all of the rest
1308 	 */
1309 	pbl_main_tbl = (dma_addr_t *)pbl_table[0].va;
1310 	for (i = 0; i < pbl_info->num_pbls - 1; i++)
1311 		pbl_main_tbl[i] = pbl_table[i + 1].pa;
1312 
1313 	QL_DPRINT12(ha, "exit\n");
1314 	return pbl_table;
1315 
1316 err:
1317 	qlnxr_free_pbl(dev, pbl_info, pbl_table);
1318 
1319 	QL_DPRINT12(ha, "exit with error\n");
1320 	return NULL;
1321 }
1322 
1323 static int
1324 qlnxr_prepare_pbl_tbl(struct qlnxr_dev *dev,
1325 	struct qlnxr_pbl_info *pbl_info,
1326 	u32 num_pbes,
1327 	int two_layer_capable)
1328 {
1329 	u32		pbl_capacity;
1330 	u32		pbl_size;
1331 	u32		num_pbls;
1332 	qlnx_host_t	*ha;
1333 
1334 	ha = dev->ha;
1335 
1336 	QL_DPRINT12(ha, "enter\n");
1337 
1338 	if ((num_pbes > MAX_PBES_ON_PAGE) && two_layer_capable) {
1339 		if (num_pbes > MAX_PBES_TWO_LAYER) {
1340 			QL_DPRINT11(ha, "prepare pbl table: too many pages %d\n",
1341 				num_pbes);
1342 			return -EINVAL;
1343 		}
1344 
1345 		/* calculate required pbl page size */
1346 		pbl_size = MIN_FW_PBL_PAGE_SIZE;
1347 		pbl_capacity = NUM_PBES_ON_PAGE(pbl_size) *
1348 			NUM_PBES_ON_PAGE(pbl_size);
1349 
1350 		while (pbl_capacity < num_pbes) {
1351 			pbl_size *= 2;
1352 			pbl_capacity = pbl_size / sizeof(u64);
1353 			pbl_capacity = pbl_capacity * pbl_capacity;
1354 		}
1355 
1356 		num_pbls = DIV_ROUND_UP(num_pbes, NUM_PBES_ON_PAGE(pbl_size));
1357 		num_pbls++; /* One for the layer0 ( points to the pbls) */
1358 		pbl_info->two_layered = true;
1359 	} else {
1360 		/* One layered PBL */
1361 		num_pbls = 1;
1362 		pbl_size = max_t(u32, MIN_FW_PBL_PAGE_SIZE, \
1363 				roundup_pow_of_two((num_pbes * sizeof(u64))));
1364 		pbl_info->two_layered = false;
1365 	}
1366 
1367 	pbl_info->num_pbls = num_pbls;
1368 	pbl_info->pbl_size = pbl_size;
1369 	pbl_info->num_pbes = num_pbes;
1370 
1371 	QL_DPRINT12(ha, "prepare pbl table: num_pbes=%d, num_pbls=%d pbl_size=%d\n",
1372 		pbl_info->num_pbes, pbl_info->num_pbls, pbl_info->pbl_size);
1373 
1374 	return 0;
1375 }
1376 
1377 #define upper_32_bits(x) (uint32_t)(x >> 32)
1378 #define lower_32_bits(x) (uint32_t)(x)
1379 
1380 static void
1381 qlnxr_populate_pbls(struct qlnxr_dev *dev, struct ib_umem *umem,
1382 	struct qlnxr_pbl *pbl, struct qlnxr_pbl_info *pbl_info)
1383 {
1384 	struct regpair		*pbe;
1385 	struct qlnxr_pbl	*pbl_tbl;
1386 	struct scatterlist	*sg;
1387 	int			shift, pg_cnt, pages, pbe_cnt, total_num_pbes = 0;
1388 	qlnx_host_t		*ha;
1389 
1390 #ifdef DEFINE_IB_UMEM_WITH_CHUNK
1391         int                     i;
1392         struct                  ib_umem_chunk *chunk = NULL;
1393 #else
1394         int                     entry;
1395 #endif
1396 
1397 
1398 	ha = dev->ha;
1399 
1400 	QL_DPRINT12(ha, "enter\n");
1401 
1402 	if (!pbl_info) {
1403 		QL_DPRINT11(ha, "PBL_INFO not initialized\n");
1404 		return;
1405 	}
1406 
1407 	if (!pbl_info->num_pbes) {
1408 		QL_DPRINT11(ha, "pbl_info->num_pbes == 0\n");
1409 		return;
1410 	}
1411 
1412 	/* If we have a two layered pbl, the first pbl points to the rest
1413 	 * of the pbls and the first entry lays on the second pbl in the table
1414 	 */
1415 	if (pbl_info->two_layered)
1416 		pbl_tbl = &pbl[1];
1417 	else
1418 		pbl_tbl = pbl;
1419 
1420 	pbe = (struct regpair *)pbl_tbl->va;
1421 	if (!pbe) {
1422 		QL_DPRINT12(ha, "pbe is NULL\n");
1423 		return;
1424 	}
1425 
1426 	pbe_cnt = 0;
1427 
1428 	shift = ilog2(umem->page_size);
1429 
1430 #ifndef DEFINE_IB_UMEM_WITH_CHUNK
1431 
1432 	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
1433 
1434 #else
1435 	list_for_each_entry(chunk, &umem->chunk_list, list) {
1436 		/* get all the dma regions from the chunk. */
1437 		for (i = 0; i < chunk->nmap; i++) {
1438 			sg = &chunk->page_list[i];
1439 #endif
1440 			pages = sg_dma_len(sg) >> shift;
1441 			for (pg_cnt = 0; pg_cnt < pages; pg_cnt++) {
1442 				/* store the page address in pbe */
1443 				pbe->lo =
1444 				    cpu_to_le32(sg_dma_address(sg) +
1445 						(umem->page_size * pg_cnt));
1446 				pbe->hi =
1447 				    cpu_to_le32(upper_32_bits
1448 						((sg_dma_address(sg) +
1449 						  umem->page_size * pg_cnt)));
1450 
1451 				QL_DPRINT12(ha,
1452 					"Populate pbl table:"
1453 					" pbe->addr=0x%x:0x%x "
1454 					" pbe_cnt = %d total_num_pbes=%d"
1455 					" pbe=%p\n", pbe->lo, pbe->hi, pbe_cnt,
1456 					total_num_pbes, pbe);
1457 
1458 				pbe_cnt ++;
1459 				total_num_pbes ++;
1460 				pbe++;
1461 
1462 				if (total_num_pbes == pbl_info->num_pbes)
1463 					return;
1464 
1465 				/* if the given pbl is full storing the pbes,
1466 				 * move to next pbl.
1467 				 */
1468 				if (pbe_cnt ==
1469 					(pbl_info->pbl_size / sizeof(u64))) {
1470 					pbl_tbl++;
1471 					pbe = (struct regpair *)pbl_tbl->va;
1472 					pbe_cnt = 0;
1473 				}
1474 			}
1475 #ifdef DEFINE_IB_UMEM_WITH_CHUNK
1476 		}
1477 #endif
1478 	}
1479 	QL_DPRINT12(ha, "exit\n");
1480 	return;
1481 }
1482 
1483 static void
1484 free_mr_info(struct qlnxr_dev *dev, struct mr_info *info)
1485 {
1486 	struct qlnxr_pbl *pbl, *tmp;
1487 	qlnx_host_t		*ha;
1488 
1489 	ha = dev->ha;
1490 
1491 	QL_DPRINT12(ha, "enter\n");
1492 
1493 	if (info->pbl_table)
1494 		list_add_tail(&info->pbl_table->list_entry,
1495 			      &info->free_pbl_list);
1496 
1497 	if (!list_empty(&info->inuse_pbl_list))
1498 		list_splice(&info->inuse_pbl_list, &info->free_pbl_list);
1499 
1500 	list_for_each_entry_safe(pbl, tmp, &info->free_pbl_list, list_entry) {
1501 		list_del(&pbl->list_entry);
1502 		qlnxr_free_pbl(dev, &info->pbl_info, pbl);
1503 	}
1504 	QL_DPRINT12(ha, "exit\n");
1505 
1506 	return;
1507 }
1508 
1509 static int
1510 qlnxr_init_mr_info(struct qlnxr_dev *dev, struct mr_info *info,
1511 	size_t page_list_len, bool two_layered)
1512 {
1513 	int			rc;
1514 	struct qlnxr_pbl	*tmp;
1515 	qlnx_host_t		*ha;
1516 
1517 	ha = dev->ha;
1518 
1519 	QL_DPRINT12(ha, "enter\n");
1520 
1521 	INIT_LIST_HEAD(&info->free_pbl_list);
1522 	INIT_LIST_HEAD(&info->inuse_pbl_list);
1523 
1524 	rc = qlnxr_prepare_pbl_tbl(dev, &info->pbl_info,
1525 				  page_list_len, two_layered);
1526 	if (rc) {
1527 		QL_DPRINT11(ha, "qlnxr_prepare_pbl_tbl [%d]\n", rc);
1528 		goto done;
1529 	}
1530 
1531 	info->pbl_table = qlnxr_alloc_pbl_tbl(dev, &info->pbl_info, GFP_KERNEL);
1532 
1533 	if (!info->pbl_table) {
1534 		rc = -ENOMEM;
1535 		QL_DPRINT11(ha, "qlnxr_alloc_pbl_tbl returned NULL\n");
1536 		goto done;
1537 	}
1538 
1539 	QL_DPRINT12(ha, "pbl_table_pa = %pa\n", &info->pbl_table->pa);
1540 
1541 	/* in usual case we use 2 PBLs, so we add one to free
1542 	 * list and allocating another one
1543 	 */
1544 	tmp = qlnxr_alloc_pbl_tbl(dev, &info->pbl_info, GFP_KERNEL);
1545 
1546 	if (!tmp) {
1547 		QL_DPRINT11(ha, "Extra PBL is not allocated\n");
1548 		goto done; /* it's OK if second allocation fails, so rc = 0*/
1549 	}
1550 
1551 	list_add_tail(&tmp->list_entry, &info->free_pbl_list);
1552 
1553 	QL_DPRINT12(ha, "extra pbl_table_pa = %pa\n", &tmp->pa);
1554 
1555 done:
1556 	if (rc)
1557 		free_mr_info(dev, info);
1558 
1559 	QL_DPRINT12(ha, "exit [%d]\n", rc);
1560 
1561 	return rc;
1562 }
1563 
1564 
1565 struct ib_mr *
1566 #if __FreeBSD_version >= 1102000
1567 qlnxr_reg_user_mr(struct ib_pd *ibpd, u64 start, u64 len,
1568 	u64 usr_addr, int acc, struct ib_udata *udata)
1569 #else
1570 qlnxr_reg_user_mr(struct ib_pd *ibpd, u64 start, u64 len,
1571 	u64 usr_addr, int acc, struct ib_udata *udata, int mr_id)
1572 #endif /* #if __FreeBSD_version >= 1102000 */
1573 {
1574 	int		rc = -ENOMEM;
1575 	struct qlnxr_dev *dev = get_qlnxr_dev((ibpd->device));
1576 	struct qlnxr_mr *mr;
1577 	struct qlnxr_pd *pd;
1578 	qlnx_host_t	*ha;
1579 
1580 	ha = dev->ha;
1581 
1582 	QL_DPRINT12(ha, "enter\n");
1583 
1584 	pd = get_qlnxr_pd(ibpd);
1585 
1586 	QL_DPRINT12(ha, "qedr_register user mr pd = %d"
1587 		" start = %lld, len = %lld, usr_addr = %lld, acc = %d\n",
1588 		pd->pd_id, start, len, usr_addr, acc);
1589 
1590 	if (acc & IB_ACCESS_REMOTE_WRITE && !(acc & IB_ACCESS_LOCAL_WRITE)) {
1591 		QL_DPRINT11(ha,
1592 			"(acc & IB_ACCESS_REMOTE_WRITE &&"
1593 			" !(acc & IB_ACCESS_LOCAL_WRITE))\n");
1594 		return ERR_PTR(-EINVAL);
1595 	}
1596 
1597 	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
1598 	if (!mr) {
1599 		QL_DPRINT11(ha, "kzalloc(mr) failed\n");
1600 		return ERR_PTR(rc);
1601 	}
1602 
1603 	mr->type = QLNXR_MR_USER;
1604 
1605 	mr->umem = ib_umem_get(ibpd->uobject->context, start, len, acc, 0);
1606 	if (IS_ERR(mr->umem)) {
1607 		rc = -EFAULT;
1608 		QL_DPRINT11(ha, "ib_umem_get failed [%p]\n", mr->umem);
1609 		goto err0;
1610 	}
1611 
1612 	rc = qlnxr_init_mr_info(dev, &mr->info, ib_umem_page_count(mr->umem), 1);
1613 	if (rc) {
1614 		QL_DPRINT11(ha,
1615 			"qlnxr_init_mr_info failed [%d]\n", rc);
1616 		goto err1;
1617 	}
1618 
1619 	qlnxr_populate_pbls(dev, mr->umem, mr->info.pbl_table,
1620 			   &mr->info.pbl_info);
1621 
1622 	rc = ecore_rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
1623 
1624 	if (rc) {
1625 		QL_DPRINT11(ha, "roce alloc tid returned an error %d\n", rc);
1626 		goto err1;
1627 	}
1628 
1629 	/* index only, 18 bit long, lkey = itid << 8 | key */
1630 	mr->hw_mr.tid_type = ECORE_RDMA_TID_REGISTERED_MR;
1631 	mr->hw_mr.key = 0;
1632 	mr->hw_mr.pd = pd->pd_id;
1633 	mr->hw_mr.local_read = 1;
1634 	mr->hw_mr.local_write = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0;
1635 	mr->hw_mr.remote_read = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0;
1636 	mr->hw_mr.remote_write = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
1637 	mr->hw_mr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0;
1638 	mr->hw_mr.mw_bind = false; /* TBD MW BIND */
1639 	mr->hw_mr.pbl_ptr = mr->info.pbl_table[0].pa;
1640 	mr->hw_mr.pbl_two_level = mr->info.pbl_info.two_layered;
1641 	mr->hw_mr.pbl_page_size_log = ilog2(mr->info.pbl_info.pbl_size);
1642 	mr->hw_mr.page_size_log = ilog2(mr->umem->page_size); /* for the MR pages */
1643 
1644 #if __FreeBSD_version >= 1102000
1645 	mr->hw_mr.fbo = ib_umem_offset(mr->umem);
1646 #else
1647 	mr->hw_mr.fbo = mr->umem->offset;
1648 #endif
1649 	mr->hw_mr.length = len;
1650 	mr->hw_mr.vaddr = usr_addr;
1651 	mr->hw_mr.zbva = false; /* TBD figure when this should be true */
1652 	mr->hw_mr.phy_mr = false; /* Fast MR - True, Regular Register False */
1653 	mr->hw_mr.dma_mr = false;
1654 
1655 	rc = ecore_rdma_register_tid(dev->rdma_ctx, &mr->hw_mr);
1656 	if (rc) {
1657 		QL_DPRINT11(ha, "roce register tid returned an error %d\n", rc);
1658 		goto err2;
1659 	}
1660 
1661 	mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
1662 	if (mr->hw_mr.remote_write || mr->hw_mr.remote_read ||
1663 		mr->hw_mr.remote_atomic)
1664 		mr->ibmr.rkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
1665 
1666 	QL_DPRINT12(ha, "register user mr lkey: %x\n", mr->ibmr.lkey);
1667 
1668 	return (&mr->ibmr);
1669 
1670 err2:
1671 	ecore_rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
1672 err1:
1673 	qlnxr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table);
1674 err0:
1675 	kfree(mr);
1676 
1677 	QL_DPRINT12(ha, "exit [%d]\n", rc);
1678 	return (ERR_PTR(rc));
1679 }
1680 
1681 int
1682 qlnxr_dereg_mr(struct ib_mr *ib_mr)
1683 {
1684 	struct qlnxr_mr	*mr = get_qlnxr_mr(ib_mr);
1685 	struct qlnxr_dev *dev = get_qlnxr_dev((ib_mr->device));
1686 	int		rc = 0;
1687 	qlnx_host_t	*ha;
1688 
1689 	ha = dev->ha;
1690 
1691 	QL_DPRINT12(ha, "enter\n");
1692 
1693 	if ((mr->type != QLNXR_MR_DMA) && (mr->type != QLNXR_MR_FRMR))
1694 		qlnxr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table);
1695 
1696 	/* it could be user registered memory. */
1697 	if (mr->umem)
1698 		ib_umem_release(mr->umem);
1699 
1700 	kfree(mr->pages);
1701 
1702 	kfree(mr);
1703 
1704 	QL_DPRINT12(ha, "exit\n");
1705 	return rc;
1706 }
1707 
1708 static int
1709 qlnxr_copy_cq_uresp(struct qlnxr_dev *dev,
1710 	struct qlnxr_cq *cq, struct ib_udata *udata)
1711 {
1712 	struct qlnxr_create_cq_uresp	uresp;
1713 	int				rc;
1714 	qlnx_host_t			*ha;
1715 
1716 	ha = dev->ha;
1717 
1718 	QL_DPRINT12(ha, "enter\n");
1719 
1720 	memset(&uresp, 0, sizeof(uresp));
1721 
1722 	uresp.db_offset = DB_ADDR_SHIFT(DQ_PWM_OFFSET_UCM_RDMA_CQ_CONS_32BIT);
1723 	uresp.icid = cq->icid;
1724 
1725 	rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
1726 
1727 	if (rc) {
1728 		QL_DPRINT12(ha, "ib_copy_to_udata error cqid=0x%x[%d]\n",
1729 			cq->icid, rc);
1730 	}
1731 
1732 	QL_DPRINT12(ha, "exit [%d]\n", rc);
1733 	return rc;
1734 }
1735 
1736 static void
1737 consume_cqe(struct qlnxr_cq *cq)
1738 {
1739 
1740 	if (cq->latest_cqe == cq->toggle_cqe)
1741 		cq->pbl_toggle ^= RDMA_RESIZE_CQ_RAMROD_DATA_TOGGLE_BIT_MASK;
1742 
1743 	cq->latest_cqe = ecore_chain_consume(&cq->pbl);
1744 }
1745 
1746 static inline int
1747 qlnxr_align_cq_entries(int entries)
1748 {
1749 	u64 size, aligned_size;
1750 
1751 	/* We allocate an extra entry that we don't report to the FW.
1752 	 * Why?
1753 	 * The CQE size is 32 bytes but the FW writes in chunks of 64 bytes
1754 	 * (for performance purposes). Allocating an extra entry and telling
1755 	 * the FW we have less prevents overwriting the first entry in case of
1756 	 * a wrap i.e. when the FW writes the last entry and the application
1757 	 * hasn't read the first one.
1758 	 */
1759 	size = (entries + 1) * QLNXR_CQE_SIZE;
1760 
1761 	/* We align to PAGE_SIZE.
1762 	 * Why?
1763 	 * Since the CQ is going to be mapped and the mapping is anyhow in whole
1764 	 * kernel pages we benefit from the possibly extra CQEs.
1765 	 */
1766 	aligned_size = ALIGN(size, PAGE_SIZE);
1767 
1768 	/* note: for CQs created in user space the result of this function
1769 	 * should match the size mapped in user space
1770 	 */
1771 	return (aligned_size / QLNXR_CQE_SIZE);
1772 }
1773 
1774 static inline int
1775 qlnxr_init_user_queue(struct ib_ucontext *ib_ctx, struct qlnxr_dev *dev,
1776 	struct qlnxr_userq *q, u64 buf_addr, size_t buf_len,
1777 	int access, int dmasync, int alloc_and_init)
1778 {
1779 	int		page_cnt;
1780 	int		rc;
1781 	qlnx_host_t	*ha;
1782 
1783 	ha = dev->ha;
1784 
1785 	QL_DPRINT12(ha, "enter\n");
1786 
1787 	q->buf_addr = buf_addr;
1788 	q->buf_len = buf_len;
1789 
1790 	QL_DPRINT12(ha, "buf_addr : %llx, buf_len : %x, access : %x"
1791 	      " dmasync : %x\n", q->buf_addr, q->buf_len,
1792 		access, dmasync);
1793 
1794 	q->umem = ib_umem_get(ib_ctx, q->buf_addr, q->buf_len, access, dmasync);
1795 
1796 	if (IS_ERR(q->umem)) {
1797 		QL_DPRINT11(ha, "ib_umem_get failed [%lx]\n", PTR_ERR(q->umem));
1798 		return PTR_ERR(q->umem);
1799 	}
1800 
1801 	page_cnt = ib_umem_page_count(q->umem);
1802 	rc = qlnxr_prepare_pbl_tbl(dev, &q->pbl_info, page_cnt,
1803 				  0 /* SQ and RQ don't support dual layer pbl.
1804 				     * CQ may, but this is yet uncoded.
1805 				     */);
1806 	if (rc) {
1807 		QL_DPRINT11(ha, "qlnxr_prepare_pbl_tbl failed [%d]\n", rc);
1808 		goto err;
1809 	}
1810 
1811 	if (alloc_and_init) {
1812 		q->pbl_tbl = qlnxr_alloc_pbl_tbl(dev, &q->pbl_info, GFP_KERNEL);
1813 
1814 		if (!q->pbl_tbl) {
1815 			QL_DPRINT11(ha, "qlnxr_alloc_pbl_tbl failed\n");
1816 			rc = -ENOMEM;
1817 			goto err;
1818 		}
1819 
1820 		qlnxr_populate_pbls(dev, q->umem, q->pbl_tbl, &q->pbl_info);
1821 	} else {
1822 		q->pbl_tbl = kzalloc(sizeof(*q->pbl_tbl), GFP_KERNEL);
1823 
1824 		if (!q->pbl_tbl) {
1825 			QL_DPRINT11(ha, "qlnxr_alloc_pbl_tbl failed\n");
1826 			rc = -ENOMEM;
1827 			goto err;
1828 		}
1829 	}
1830 
1831 	QL_DPRINT12(ha, "exit\n");
1832 	return 0;
1833 
1834 err:
1835 	ib_umem_release(q->umem);
1836 	q->umem = NULL;
1837 
1838 	QL_DPRINT12(ha, "exit [%d]\n", rc);
1839 	return rc;
1840 }
1841 
1842 #if __FreeBSD_version >= 1102000
1843 
1844 struct ib_cq *
1845 qlnxr_create_cq(struct ib_device *ibdev,
1846 	const struct ib_cq_init_attr *attr,
1847 	struct ib_ucontext *ib_ctx,
1848 	struct ib_udata *udata)
1849 
1850 #else
1851 
1852 #if __FreeBSD_version >= 1100000
1853 
1854 struct ib_cq *
1855 qlnxr_create_cq(struct ib_device *ibdev,
1856 	struct ib_cq_init_attr *attr,
1857 	struct ib_ucontext *ib_ctx,
1858 	struct ib_udata *udata)
1859 
1860 #else
1861 
1862 struct ib_cq *
1863 qlnxr_create_cq(struct ib_device *ibdev,
1864 	int entries,
1865 	int vector,
1866 	struct ib_ucontext *ib_ctx,
1867 	struct ib_udata *udata)
1868 #endif /* #if __FreeBSD_version >= 1100000 */
1869 
1870 #endif /* #if __FreeBSD_version >= 1102000 */
1871 {
1872 	struct qlnxr_ucontext			*ctx;
1873 	struct ecore_rdma_destroy_cq_out_params destroy_oparams;
1874 	struct ecore_rdma_destroy_cq_in_params	destroy_iparams;
1875 	struct qlnxr_dev			*dev;
1876 	struct ecore_rdma_create_cq_in_params	params;
1877 	struct qlnxr_create_cq_ureq		ureq;
1878 
1879 #if __FreeBSD_version >= 1100000
1880 	int					vector = attr->comp_vector;
1881 	int					entries = attr->cqe;
1882 #endif
1883 	struct qlnxr_cq				*cq;
1884 	int					chain_entries, rc, page_cnt;
1885 	u64					pbl_ptr;
1886 	u16					icid;
1887 	qlnx_host_t				*ha;
1888 
1889 	dev = get_qlnxr_dev(ibdev);
1890 	ha = dev->ha;
1891 
1892 	QL_DPRINT12(ha, "called from %s. entries = %d, "
1893 		"vector = %d\n",
1894 		(udata ? "User Lib" : "Kernel"), entries, vector);
1895 
1896         memset(&params, 0, sizeof(struct ecore_rdma_create_cq_in_params));
1897         memset(&destroy_iparams, 0, sizeof(struct ecore_rdma_destroy_cq_in_params));
1898         memset(&destroy_oparams, 0, sizeof(struct ecore_rdma_destroy_cq_out_params));
1899 
1900 	if (entries > QLNXR_MAX_CQES) {
1901 		QL_DPRINT11(ha,
1902 			"the number of entries %d is too high. "
1903 			"Must be equal or below %d.\n",
1904 			entries, QLNXR_MAX_CQES);
1905 		return ERR_PTR(-EINVAL);
1906 	}
1907 	chain_entries = qlnxr_align_cq_entries(entries);
1908 	chain_entries = min_t(int, chain_entries, QLNXR_MAX_CQES);
1909 
1910 	cq = qlnx_zalloc((sizeof(struct qlnxr_cq)));
1911 
1912 	if (!cq)
1913 		return ERR_PTR(-ENOMEM);
1914 
1915 	if (udata) {
1916 		memset(&ureq, 0, sizeof(ureq));
1917 
1918 		if (ib_copy_from_udata(&ureq, udata,
1919 			min(sizeof(ureq), udata->inlen))) {
1920 			QL_DPRINT11(ha, "ib_copy_from_udata failed\n");
1921 			goto err0;
1922 		}
1923 
1924 		if (!ureq.len) {
1925 			QL_DPRINT11(ha, "ureq.len == 0\n");
1926 			goto err0;
1927 		}
1928 
1929 		cq->cq_type = QLNXR_CQ_TYPE_USER;
1930 
1931 		qlnxr_init_user_queue(ib_ctx, dev, &cq->q, ureq.addr, ureq.len,
1932 				     IB_ACCESS_LOCAL_WRITE, 1, 1);
1933 
1934 		pbl_ptr = cq->q.pbl_tbl->pa;
1935 		page_cnt = cq->q.pbl_info.num_pbes;
1936 		cq->ibcq.cqe = chain_entries;
1937 	} else {
1938 		cq->cq_type = QLNXR_CQ_TYPE_KERNEL;
1939 
1940                 rc = ecore_chain_alloc(&dev->ha->cdev,
1941                            ECORE_CHAIN_USE_TO_CONSUME,
1942                            ECORE_CHAIN_MODE_PBL,
1943                            ECORE_CHAIN_CNT_TYPE_U32,
1944                            chain_entries,
1945                            sizeof(union roce_cqe),
1946                            &cq->pbl, NULL);
1947 
1948 		if (rc)
1949 			goto err1;
1950 
1951 		page_cnt = ecore_chain_get_page_cnt(&cq->pbl);
1952 		pbl_ptr = ecore_chain_get_pbl_phys(&cq->pbl);
1953 		cq->ibcq.cqe = cq->pbl.capacity;
1954 	}
1955 
1956         params.cq_handle_hi = upper_32_bits((uintptr_t)cq);
1957         params.cq_handle_lo = lower_32_bits((uintptr_t)cq);
1958         params.cnq_id = vector;
1959         params.cq_size = chain_entries - 1;
1960         params.pbl_num_pages = page_cnt;
1961         params.pbl_ptr = pbl_ptr;
1962         params.pbl_two_level = 0;
1963 
1964 	if (ib_ctx != NULL) {
1965 		ctx = get_qlnxr_ucontext(ib_ctx);
1966         	params.dpi = ctx->dpi;
1967 	} else {
1968         	params.dpi = dev->dpi;
1969 	}
1970 
1971 	rc = ecore_rdma_create_cq(dev->rdma_ctx, &params, &icid);
1972 	if (rc)
1973 		goto err2;
1974 
1975 	cq->icid = icid;
1976 	cq->sig = QLNXR_CQ_MAGIC_NUMBER;
1977 	spin_lock_init(&cq->cq_lock);
1978 
1979 	if (ib_ctx) {
1980 		rc = qlnxr_copy_cq_uresp(dev, cq, udata);
1981 		if (rc)
1982 			goto err3;
1983 	} else {
1984 		/* Generate doorbell address.
1985 		 * Configure bits 3-9 with DQ_PWM_OFFSET_UCM_RDMA_CQ_CONS_32BIT.
1986 		 * TODO: consider moving to device scope as it is a function of
1987 		 *       the device.
1988 		 * TODO: add ifdef if plan to support 16 bit.
1989 		 */
1990 		cq->db_addr = dev->db_addr +
1991 			DB_ADDR_SHIFT(DQ_PWM_OFFSET_UCM_RDMA_CQ_CONS_32BIT);
1992 		cq->db.data.icid = cq->icid;
1993 		cq->db.data.params = DB_AGG_CMD_SET <<
1994 				     RDMA_PWM_VAL32_DATA_AGG_CMD_SHIFT;
1995 
1996 		/* point to the very last element, passing it we will toggle */
1997 		cq->toggle_cqe = ecore_chain_get_last_elem(&cq->pbl);
1998 		cq->pbl_toggle = RDMA_RESIZE_CQ_RAMROD_DATA_TOGGLE_BIT_MASK;
1999 
2000 		/* must be different from pbl_toggle */
2001 		cq->latest_cqe = NULL;
2002 		consume_cqe(cq);
2003 		cq->cq_cons = ecore_chain_get_cons_idx_u32(&cq->pbl);
2004 	}
2005 
2006 	QL_DPRINT12(ha, "exit icid = 0x%0x, addr = %p,"
2007 		" number of entries = 0x%x\n",
2008 		cq->icid, cq, params.cq_size);
2009 	QL_DPRINT12(ha,"cq_addr = %p\n", cq);
2010 	return &cq->ibcq;
2011 
2012 err3:
2013 	destroy_iparams.icid = cq->icid;
2014 	ecore_rdma_destroy_cq(dev->rdma_ctx, &destroy_iparams, &destroy_oparams);
2015 err2:
2016 	if (udata)
2017 		qlnxr_free_pbl(dev, &cq->q.pbl_info, cq->q.pbl_tbl);
2018 	else
2019 		ecore_chain_free(&dev->ha->cdev, &cq->pbl);
2020 err1:
2021 	if (udata)
2022 		ib_umem_release(cq->q.umem);
2023 err0:
2024 	kfree(cq);
2025 
2026 	QL_DPRINT12(ha, "exit error\n");
2027 
2028 	return ERR_PTR(-EINVAL);
2029 }
2030 
2031 int qlnxr_resize_cq(struct ib_cq *ibcq, int new_cnt, struct ib_udata *udata)
2032 {
2033 	int			status = 0;
2034 	struct qlnxr_dev	*dev = get_qlnxr_dev((ibcq->device));
2035 	qlnx_host_t		*ha;
2036 
2037 	ha = dev->ha;
2038 
2039 	QL_DPRINT12(ha, "enter/exit\n");
2040 
2041 	return status;
2042 }
2043 
2044 int
2045 qlnxr_destroy_cq(struct ib_cq *ibcq)
2046 {
2047 	struct qlnxr_dev			*dev = get_qlnxr_dev((ibcq->device));
2048 	struct ecore_rdma_destroy_cq_out_params oparams;
2049 	struct ecore_rdma_destroy_cq_in_params	iparams;
2050 	struct qlnxr_cq				*cq = get_qlnxr_cq(ibcq);
2051 	int					rc = 0;
2052 	qlnx_host_t				*ha;
2053 
2054 	ha = dev->ha;
2055 
2056 	QL_DPRINT12(ha, "enter cq_id = %d\n", cq->icid);
2057 
2058 	cq->destroyed = 1;
2059 
2060 	/* TODO: Syncronize irq of the CNQ the CQ belongs to for validation
2061 	 * that all completions with notification are dealt with. The rest
2062 	 * of the completions are not interesting
2063 	 */
2064 
2065 	/* GSIs CQs are handled by driver, so they don't exist in the FW */
2066 
2067 	if (cq->cq_type != QLNXR_CQ_TYPE_GSI) {
2068 
2069 		iparams.icid = cq->icid;
2070 
2071 		rc = ecore_rdma_destroy_cq(dev->rdma_ctx, &iparams, &oparams);
2072 
2073 		if (rc) {
2074 			QL_DPRINT12(ha, "ecore_rdma_destroy_cq failed cq_id = %d\n",
2075 				cq->icid);
2076 			return rc;
2077 		}
2078 
2079 		QL_DPRINT12(ha, "free cq->pbl cq_id = %d\n", cq->icid);
2080 		ecore_chain_free(&dev->ha->cdev, &cq->pbl);
2081 	}
2082 
2083 	if (ibcq->uobject && ibcq->uobject->context) {
2084 		qlnxr_free_pbl(dev, &cq->q.pbl_info, cq->q.pbl_tbl);
2085 		ib_umem_release(cq->q.umem);
2086 	}
2087 
2088 	cq->sig = ~cq->sig;
2089 
2090 	kfree(cq);
2091 
2092 	QL_DPRINT12(ha, "exit cq_id = %d\n", cq->icid);
2093 
2094 	return rc;
2095 }
2096 
2097 static int
2098 qlnxr_check_qp_attrs(struct ib_pd *ibpd,
2099 	struct qlnxr_dev *dev,
2100 	struct ib_qp_init_attr *attrs,
2101 	struct ib_udata *udata)
2102 {
2103 	struct ecore_rdma_device	*qattr;
2104 	qlnx_host_t			*ha;
2105 
2106 	qattr = ecore_rdma_query_device(dev->rdma_ctx);
2107 	ha = dev->ha;
2108 
2109 	QL_DPRINT12(ha, "enter\n");
2110 
2111 	QL_DPRINT12(ha, "attrs->sq_sig_type = %d\n", attrs->sq_sig_type);
2112 	QL_DPRINT12(ha, "attrs->qp_type = %d\n", attrs->qp_type);
2113 	QL_DPRINT12(ha, "attrs->create_flags = %d\n", attrs->create_flags);
2114 
2115 #if __FreeBSD_version < 1102000
2116 	QL_DPRINT12(ha, "attrs->qpg_type = %d\n", attrs->qpg_type);
2117 #endif
2118 
2119 	QL_DPRINT12(ha, "attrs->port_num = %d\n", attrs->port_num);
2120 	QL_DPRINT12(ha, "attrs->cap.max_send_wr = 0x%x\n", attrs->cap.max_send_wr);
2121 	QL_DPRINT12(ha, "attrs->cap.max_recv_wr = 0x%x\n", attrs->cap.max_recv_wr);
2122 	QL_DPRINT12(ha, "attrs->cap.max_send_sge = 0x%x\n", attrs->cap.max_send_sge);
2123 	QL_DPRINT12(ha, "attrs->cap.max_recv_sge = 0x%x\n", attrs->cap.max_recv_sge);
2124 	QL_DPRINT12(ha, "attrs->cap.max_inline_data = 0x%x\n",
2125 		attrs->cap.max_inline_data);
2126 
2127 #if __FreeBSD_version < 1102000
2128 	QL_DPRINT12(ha, "attrs->cap.qpg_tss_mask_sz = 0x%x\n",
2129 		attrs->cap.qpg_tss_mask_sz);
2130 #endif
2131 
2132 	QL_DPRINT12(ha, "\n\nqattr->vendor_id = 0x%x\n", qattr->vendor_id);
2133 	QL_DPRINT12(ha, "qattr->vendor_part_id = 0x%x\n", qattr->vendor_part_id);
2134 	QL_DPRINT12(ha, "qattr->hw_ver = 0x%x\n", qattr->hw_ver);
2135 	QL_DPRINT12(ha, "qattr->fw_ver = %p\n", (void *)qattr->fw_ver);
2136 	QL_DPRINT12(ha, "qattr->node_guid = %p\n", (void *)qattr->node_guid);
2137 	QL_DPRINT12(ha, "qattr->sys_image_guid = %p\n",
2138 		(void *)qattr->sys_image_guid);
2139 	QL_DPRINT12(ha, "qattr->max_cnq = 0x%x\n", qattr->max_cnq);
2140 	QL_DPRINT12(ha, "qattr->max_sge = 0x%x\n", qattr->max_sge);
2141 	QL_DPRINT12(ha, "qattr->max_srq_sge = 0x%x\n", qattr->max_srq_sge);
2142 	QL_DPRINT12(ha, "qattr->max_inline = 0x%x\n", qattr->max_inline);
2143 	QL_DPRINT12(ha, "qattr->max_wqe = 0x%x\n", qattr->max_wqe);
2144 	QL_DPRINT12(ha, "qattr->max_srq_wqe = 0x%x\n", qattr->max_srq_wqe);
2145 	QL_DPRINT12(ha, "qattr->max_qp_resp_rd_atomic_resc = 0x%x\n",
2146 		qattr->max_qp_resp_rd_atomic_resc);
2147 	QL_DPRINT12(ha, "qattr->max_qp_req_rd_atomic_resc = 0x%x\n",
2148 		qattr->max_qp_req_rd_atomic_resc);
2149 	QL_DPRINT12(ha, "qattr->max_dev_resp_rd_atomic_resc = 0x%x\n",
2150 		qattr->max_dev_resp_rd_atomic_resc);
2151 	QL_DPRINT12(ha, "qattr->max_cq = 0x%x\n", qattr->max_cq);
2152 	QL_DPRINT12(ha, "qattr->max_qp = 0x%x\n", qattr->max_qp);
2153 	QL_DPRINT12(ha, "qattr->max_srq = 0x%x\n", qattr->max_srq);
2154 	QL_DPRINT12(ha, "qattr->max_mr = 0x%x\n", qattr->max_mr);
2155 	QL_DPRINT12(ha, "qattr->max_mr_size = %p\n", (void *)qattr->max_mr_size);
2156 	QL_DPRINT12(ha, "qattr->max_cqe = 0x%x\n", qattr->max_cqe);
2157 	QL_DPRINT12(ha, "qattr->max_mw = 0x%x\n", qattr->max_mw);
2158 	QL_DPRINT12(ha, "qattr->max_fmr = 0x%x\n", qattr->max_fmr);
2159 	QL_DPRINT12(ha, "qattr->max_mr_mw_fmr_pbl = 0x%x\n",
2160 		qattr->max_mr_mw_fmr_pbl);
2161 	QL_DPRINT12(ha, "qattr->max_mr_mw_fmr_size = %p\n",
2162 		(void *)qattr->max_mr_mw_fmr_size);
2163 	QL_DPRINT12(ha, "qattr->max_pd = 0x%x\n", qattr->max_pd);
2164 	QL_DPRINT12(ha, "qattr->max_ah = 0x%x\n", qattr->max_ah);
2165 	QL_DPRINT12(ha, "qattr->max_pkey = 0x%x\n", qattr->max_pkey);
2166 	QL_DPRINT12(ha, "qattr->max_srq_wr = 0x%x\n", qattr->max_srq_wr);
2167 	QL_DPRINT12(ha, "qattr->max_stats_queues = 0x%x\n",
2168 		qattr->max_stats_queues);
2169 	//QL_DPRINT12(ha, "qattr->dev_caps = 0x%x\n", qattr->dev_caps);
2170 	QL_DPRINT12(ha, "qattr->page_size_caps = %p\n",
2171 		(void *)qattr->page_size_caps);
2172 	QL_DPRINT12(ha, "qattr->dev_ack_delay = 0x%x\n", qattr->dev_ack_delay);
2173 	QL_DPRINT12(ha, "qattr->reserved_lkey = 0x%x\n", qattr->reserved_lkey);
2174 	QL_DPRINT12(ha, "qattr->bad_pkey_counter = 0x%x\n",
2175 		qattr->bad_pkey_counter);
2176 
2177 	if ((attrs->qp_type == IB_QPT_GSI) && udata) {
2178 		QL_DPRINT12(ha, "unexpected udata when creating GSI QP\n");
2179 		return -EINVAL;
2180 	}
2181 
2182 	if (udata && !(ibpd->uobject && ibpd->uobject->context)) {
2183 		QL_DPRINT12(ha, "called from user without context\n");
2184 		return -EINVAL;
2185 	}
2186 
2187 	/* QP0... attrs->qp_type == IB_QPT_GSI */
2188 	if (attrs->qp_type != IB_QPT_RC && attrs->qp_type != IB_QPT_GSI) {
2189 		QL_DPRINT12(ha, "unsupported qp type=0x%x requested\n",
2190 			   attrs->qp_type);
2191 		return -EINVAL;
2192 	}
2193 	if (attrs->qp_type == IB_QPT_GSI && attrs->srq) {
2194 		QL_DPRINT12(ha, "cannot create GSI qp with SRQ\n");
2195 		return -EINVAL;
2196 	}
2197 	/* Skip the check for QP1 to support CM size of 128 */
2198 	if (attrs->cap.max_send_wr > qattr->max_wqe) {
2199 		QL_DPRINT12(ha, "cannot create a SQ with %d elements "
2200 			" (max_send_wr=0x%x)\n",
2201 			attrs->cap.max_send_wr, qattr->max_wqe);
2202 		return -EINVAL;
2203 	}
2204 	if (!attrs->srq && (attrs->cap.max_recv_wr > qattr->max_wqe)) {
2205 		QL_DPRINT12(ha, "cannot create a RQ with %d elements"
2206 			" (max_recv_wr=0x%x)\n",
2207 			attrs->cap.max_recv_wr, qattr->max_wqe);
2208 		return -EINVAL;
2209 	}
2210 	if (attrs->cap.max_inline_data > qattr->max_inline) {
2211 		QL_DPRINT12(ha,
2212 			"unsupported inline data size=0x%x "
2213 			"requested (max_inline=0x%x)\n",
2214 			attrs->cap.max_inline_data, qattr->max_inline);
2215 		return -EINVAL;
2216 	}
2217 	if (attrs->cap.max_send_sge > qattr->max_sge) {
2218 		QL_DPRINT12(ha,
2219 			"unsupported send_sge=0x%x "
2220 			"requested (max_send_sge=0x%x)\n",
2221 			attrs->cap.max_send_sge, qattr->max_sge);
2222 		return -EINVAL;
2223 	}
2224 	if (attrs->cap.max_recv_sge > qattr->max_sge) {
2225 		QL_DPRINT12(ha,
2226 			"unsupported recv_sge=0x%x requested "
2227 			" (max_recv_sge=0x%x)\n",
2228 			attrs->cap.max_recv_sge, qattr->max_sge);
2229 		return -EINVAL;
2230 	}
2231 	/* unprivileged user space cannot create special QP */
2232 	if (ibpd->uobject && attrs->qp_type == IB_QPT_GSI) {
2233 		QL_DPRINT12(ha,
2234 			"userspace can't create special QPs of type=0x%x\n",
2235 			attrs->qp_type);
2236 		return -EINVAL;
2237 	}
2238 	/* allow creating only one GSI type of QP */
2239 	if (attrs->qp_type == IB_QPT_GSI && dev->gsi_qp_created) {
2240 		QL_DPRINT12(ha,
2241 			"create qp: GSI special QPs already created.\n");
2242 		return -EINVAL;
2243 	}
2244 
2245 	/* verify consumer QPs are not trying to use GSI QP's CQ */
2246 	if ((attrs->qp_type != IB_QPT_GSI) && (dev->gsi_qp_created)) {
2247 		struct qlnxr_cq *send_cq = get_qlnxr_cq(attrs->send_cq);
2248 		struct qlnxr_cq *recv_cq = get_qlnxr_cq(attrs->recv_cq);
2249 
2250 		if ((send_cq->cq_type == QLNXR_CQ_TYPE_GSI) ||
2251 		    (recv_cq->cq_type == QLNXR_CQ_TYPE_GSI)) {
2252 			QL_DPRINT11(ha, "consumer QP cannot use GSI CQs.\n");
2253 			return -EINVAL;
2254 		}
2255 	}
2256 	QL_DPRINT12(ha, "exit\n");
2257 	return 0;
2258 }
2259 
2260 static int
2261 qlnxr_copy_srq_uresp(struct qlnxr_dev *dev,
2262 	struct qlnxr_srq *srq,
2263 	struct ib_udata *udata)
2264 {
2265 	struct qlnxr_create_srq_uresp	uresp;
2266 	qlnx_host_t			*ha;
2267 	int				rc;
2268 
2269 	ha = dev->ha;
2270 
2271 	QL_DPRINT12(ha, "enter\n");
2272 
2273 	memset(&uresp, 0, sizeof(uresp));
2274 
2275 	uresp.srq_id = srq->srq_id;
2276 
2277 	rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
2278 
2279 	QL_DPRINT12(ha, "exit [%d]\n", rc);
2280 	return rc;
2281 }
2282 
2283 static void
2284 qlnxr_copy_rq_uresp(struct qlnxr_dev *dev,
2285 	struct qlnxr_create_qp_uresp *uresp,
2286 	struct qlnxr_qp *qp)
2287 {
2288 	qlnx_host_t	*ha;
2289 
2290 	ha = dev->ha;
2291 
2292 	/* Return if QP is associated with SRQ instead of RQ */
2293 	QL_DPRINT12(ha, "enter qp->srq = %p\n", qp->srq);
2294 
2295 	if (qp->srq)
2296 		return;
2297 
2298 	/* iWARP requires two doorbells per RQ. */
2299 	if (QLNX_IS_IWARP(dev)) {
2300 
2301 		uresp->rq_db_offset =
2302 			DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_IWARP_RQ_PROD);
2303 		uresp->rq_db2_offset =
2304 			DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_FLAGS);
2305 
2306 		QL_DPRINT12(ha, "uresp->rq_db_offset = 0x%x "
2307 			"uresp->rq_db2_offset = 0x%x\n",
2308 			uresp->rq_db_offset, uresp->rq_db2_offset);
2309 	} else {
2310 		uresp->rq_db_offset =
2311 			DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_ROCE_RQ_PROD);
2312 	}
2313 	uresp->rq_icid = qp->icid;
2314 
2315 	QL_DPRINT12(ha, "exit\n");
2316 	return;
2317 }
2318 
2319 static void
2320 qlnxr_copy_sq_uresp(struct qlnxr_dev *dev,
2321 	struct qlnxr_create_qp_uresp *uresp,
2322 	struct qlnxr_qp *qp)
2323 {
2324 	qlnx_host_t	*ha;
2325 
2326 	ha = dev->ha;
2327 
2328 	QL_DPRINT12(ha, "enter\n");
2329 
2330 	uresp->sq_db_offset = DB_ADDR_SHIFT(DQ_PWM_OFFSET_XCM_RDMA_SQ_PROD);
2331 
2332 	/* iWARP uses the same cid for rq and sq*/
2333 	if (QLNX_IS_IWARP(dev)) {
2334 		uresp->sq_icid = qp->icid;
2335 		QL_DPRINT12(ha, "uresp->sq_icid = 0x%x\n", uresp->sq_icid);
2336 	} else
2337 		uresp->sq_icid = qp->icid + 1;
2338 
2339 	QL_DPRINT12(ha, "exit\n");
2340 	return;
2341 }
2342 
2343 static int
2344 qlnxr_copy_qp_uresp(struct qlnxr_dev *dev,
2345 	struct qlnxr_qp *qp,
2346 	struct ib_udata *udata)
2347 {
2348 	int				rc;
2349 	struct qlnxr_create_qp_uresp	uresp;
2350 	qlnx_host_t			*ha;
2351 
2352 	ha = dev->ha;
2353 
2354 	QL_DPRINT12(ha, "enter qp->icid =0x%x\n", qp->icid);
2355 
2356 	memset(&uresp, 0, sizeof(uresp));
2357 	qlnxr_copy_sq_uresp(dev, &uresp, qp);
2358 	qlnxr_copy_rq_uresp(dev, &uresp, qp);
2359 
2360 	uresp.atomic_supported = dev->atomic_cap != IB_ATOMIC_NONE;
2361 	uresp.qp_id = qp->qp_id;
2362 
2363 	rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
2364 
2365 	QL_DPRINT12(ha, "exit [%d]\n", rc);
2366 	return rc;
2367 }
2368 
2369 
2370 static void
2371 qlnxr_set_common_qp_params(struct qlnxr_dev *dev,
2372 	struct qlnxr_qp *qp,
2373 	struct qlnxr_pd *pd,
2374 	struct ib_qp_init_attr *attrs)
2375 {
2376 	qlnx_host_t			*ha;
2377 
2378 	ha = dev->ha;
2379 
2380 	QL_DPRINT12(ha, "enter\n");
2381 
2382 	spin_lock_init(&qp->q_lock);
2383 
2384 	atomic_set(&qp->refcnt, 1);
2385 	qp->pd = pd;
2386 	qp->sig = QLNXR_QP_MAGIC_NUMBER;
2387 	qp->qp_type = attrs->qp_type;
2388 	qp->max_inline_data = ROCE_REQ_MAX_INLINE_DATA_SIZE;
2389 	qp->sq.max_sges = attrs->cap.max_send_sge;
2390 	qp->state = ECORE_ROCE_QP_STATE_RESET;
2391 	qp->signaled = (attrs->sq_sig_type == IB_SIGNAL_ALL_WR) ? true : false;
2392 	qp->sq_cq = get_qlnxr_cq(attrs->send_cq);
2393 	qp->rq_cq = get_qlnxr_cq(attrs->recv_cq);
2394 	qp->dev = dev;
2395 
2396 	if (!attrs->srq) {
2397 		/* QP is associated with RQ instead of SRQ */
2398 		qp->rq.max_sges = attrs->cap.max_recv_sge;
2399 		QL_DPRINT12(ha, "RQ params:\trq_max_sges = %d, rq_cq_id = %d\n",
2400 			qp->rq.max_sges, qp->rq_cq->icid);
2401 	} else {
2402 		qp->srq = get_qlnxr_srq(attrs->srq);
2403 	}
2404 
2405 	QL_DPRINT12(ha,
2406 		"QP params:\tpd = %d, qp_type = %d, max_inline_data = %d,"
2407 		" state = %d, signaled = %d, use_srq=%d\n",
2408 		pd->pd_id, qp->qp_type, qp->max_inline_data,
2409 		qp->state, qp->signaled, ((attrs->srq) ? 1 : 0));
2410 	QL_DPRINT12(ha, "SQ params:\tsq_max_sges = %d, sq_cq_id = %d\n",
2411 		qp->sq.max_sges, qp->sq_cq->icid);
2412 	return;
2413 }
2414 
2415 static int
2416 qlnxr_check_srq_params(struct ib_pd *ibpd,
2417 	struct qlnxr_dev *dev,
2418 	struct ib_srq_init_attr *attrs)
2419 {
2420 	struct ecore_rdma_device *qattr;
2421 	qlnx_host_t		*ha;
2422 
2423 	ha = dev->ha;
2424 	qattr = ecore_rdma_query_device(dev->rdma_ctx);
2425 
2426 	QL_DPRINT12(ha, "enter\n");
2427 
2428 	if (attrs->attr.max_wr > qattr->max_srq_wqe) {
2429 		QL_DPRINT12(ha, "unsupported srq_wr=0x%x"
2430 			" requested (max_srq_wr=0x%x)\n",
2431 			attrs->attr.max_wr, qattr->max_srq_wr);
2432 		return -EINVAL;
2433 	}
2434 
2435 	if (attrs->attr.max_sge > qattr->max_sge) {
2436 		QL_DPRINT12(ha,
2437 			"unsupported sge=0x%x requested (max_srq_sge=0x%x)\n",
2438 			attrs->attr.max_sge, qattr->max_sge);
2439 		return -EINVAL;
2440 	}
2441 
2442 	if (attrs->attr.srq_limit > attrs->attr.max_wr) {
2443 		QL_DPRINT12(ha,
2444 		       "unsupported srq_limit=0x%x requested"
2445 			" (max_srq_limit=0x%x)\n",
2446 			attrs->attr.srq_limit, attrs->attr.srq_limit);
2447 		return -EINVAL;
2448 	}
2449 
2450 	QL_DPRINT12(ha, "exit\n");
2451 	return 0;
2452 }
2453 
2454 
2455 static void
2456 qlnxr_free_srq_user_params(struct qlnxr_srq *srq)
2457 {
2458 	struct qlnxr_dev	*dev = srq->dev;
2459 	qlnx_host_t		*ha;
2460 
2461 	ha = dev->ha;
2462 
2463 	QL_DPRINT12(ha, "enter\n");
2464 
2465 	qlnxr_free_pbl(srq->dev, &srq->usrq.pbl_info, srq->usrq.pbl_tbl);
2466 	ib_umem_release(srq->usrq.umem);
2467 	ib_umem_release(srq->prod_umem);
2468 
2469 	QL_DPRINT12(ha, "exit\n");
2470 	return;
2471 }
2472 
2473 static void
2474 qlnxr_free_srq_kernel_params(struct qlnxr_srq *srq)
2475 {
2476 	struct qlnxr_srq_hwq_info *hw_srq  = &srq->hw_srq;
2477 	struct qlnxr_dev	*dev = srq->dev;
2478 	qlnx_host_t		*ha;
2479 
2480 	ha = dev->ha;
2481 
2482 	QL_DPRINT12(ha, "enter\n");
2483 
2484 	ecore_chain_free(dev->cdev, &hw_srq->pbl);
2485 
2486 	qlnx_dma_free_coherent(&dev->cdev,
2487 		hw_srq->virt_prod_pair_addr,
2488 		hw_srq->phy_prod_pair_addr,
2489 		sizeof(struct rdma_srq_producers));
2490 
2491 	QL_DPRINT12(ha, "exit\n");
2492 
2493 	return;
2494 }
2495 
2496 static int
2497 qlnxr_init_srq_user_params(struct ib_ucontext *ib_ctx,
2498 	struct qlnxr_srq *srq,
2499 	struct qlnxr_create_srq_ureq *ureq,
2500 	int access, int dmasync)
2501 {
2502 #ifdef DEFINE_IB_UMEM_WITH_CHUNK
2503 	struct ib_umem_chunk	*chunk;
2504 #endif
2505 	struct scatterlist	*sg;
2506 	int			rc;
2507 	struct qlnxr_dev	*dev = srq->dev;
2508 	qlnx_host_t		*ha;
2509 
2510 	ha = dev->ha;
2511 
2512 	QL_DPRINT12(ha, "enter\n");
2513 
2514 	rc = qlnxr_init_user_queue(ib_ctx, srq->dev, &srq->usrq, ureq->srq_addr,
2515 				  ureq->srq_len, access, dmasync, 1);
2516 	if (rc)
2517 		return rc;
2518 
2519 	srq->prod_umem = ib_umem_get(ib_ctx, ureq->prod_pair_addr,
2520 				     sizeof(struct rdma_srq_producers),
2521 				     access, dmasync);
2522 	if (IS_ERR(srq->prod_umem)) {
2523 
2524 		qlnxr_free_pbl(srq->dev, &srq->usrq.pbl_info, srq->usrq.pbl_tbl);
2525 		ib_umem_release(srq->usrq.umem);
2526 
2527 		QL_DPRINT12(ha, "ib_umem_get failed for producer [%p]\n",
2528 			PTR_ERR(srq->prod_umem));
2529 
2530 		return PTR_ERR(srq->prod_umem);
2531 	}
2532 
2533 #ifdef DEFINE_IB_UMEM_WITH_CHUNK
2534 	chunk = container_of((&srq->prod_umem->chunk_list)->next,
2535 			     typeof(*chunk), list);
2536 	sg = &chunk->page_list[0];
2537 #else
2538 	sg = srq->prod_umem->sg_head.sgl;
2539 #endif
2540 	srq->hw_srq.phy_prod_pair_addr = sg_dma_address(sg);
2541 
2542 	QL_DPRINT12(ha, "exit\n");
2543 	return 0;
2544 }
2545 
2546 
2547 static int
2548 qlnxr_alloc_srq_kernel_params(struct qlnxr_srq *srq,
2549 	struct qlnxr_dev *dev,
2550 	struct ib_srq_init_attr *init_attr)
2551 {
2552 	struct qlnxr_srq_hwq_info	*hw_srq  = &srq->hw_srq;
2553 	dma_addr_t			phy_prod_pair_addr;
2554 	u32				num_elems, max_wr;
2555 	void				*va;
2556 	int				rc;
2557 	qlnx_host_t			*ha;
2558 
2559 	ha = dev->ha;
2560 
2561 	QL_DPRINT12(ha, "enter\n");
2562 
2563 	va = qlnx_dma_alloc_coherent(&dev->cdev,
2564 			&phy_prod_pair_addr,
2565 			sizeof(struct rdma_srq_producers));
2566 	if (!va) {
2567 		QL_DPRINT11(ha, "qlnx_dma_alloc_coherent failed for produceer\n");
2568 		return -ENOMEM;
2569 	}
2570 
2571 	hw_srq->phy_prod_pair_addr = phy_prod_pair_addr;
2572 	hw_srq->virt_prod_pair_addr = va;
2573 
2574 	max_wr = init_attr->attr.max_wr;
2575 
2576 	num_elems = max_wr * RDMA_MAX_SRQ_WQE_SIZE;
2577 
2578         rc = ecore_chain_alloc(dev->cdev,
2579                    ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
2580                    ECORE_CHAIN_MODE_PBL,
2581                    ECORE_CHAIN_CNT_TYPE_U32,
2582                    num_elems,
2583                    ECORE_RDMA_SRQ_WQE_ELEM_SIZE,
2584                    &hw_srq->pbl, NULL);
2585 
2586 	if (rc) {
2587 		QL_DPRINT11(ha, "ecore_chain_alloc failed [%d]\n", rc);
2588 		goto err0;
2589 	}
2590 
2591 	hw_srq->max_wr = max_wr;
2592 	hw_srq->num_elems = num_elems;
2593 	hw_srq->max_sges = RDMA_MAX_SGE_PER_SRQ;
2594 
2595 	QL_DPRINT12(ha, "exit\n");
2596 	return 0;
2597 
2598 err0:
2599 	qlnx_dma_free_coherent(&dev->cdev, va, phy_prod_pair_addr,
2600 		sizeof(struct rdma_srq_producers));
2601 
2602 	QL_DPRINT12(ha, "exit [%d]\n", rc);
2603 	return rc;
2604 }
2605 
2606 static inline void
2607 qlnxr_init_common_qp_in_params(struct qlnxr_dev *dev,
2608 	struct qlnxr_pd *pd,
2609 	struct qlnxr_qp *qp,
2610 	struct ib_qp_init_attr *attrs,
2611 	bool fmr_and_reserved_lkey,
2612 	struct ecore_rdma_create_qp_in_params *params)
2613 {
2614 	qlnx_host_t	*ha;
2615 
2616 	ha = dev->ha;
2617 
2618 	QL_DPRINT12(ha, "enter\n");
2619 
2620 	/* QP handle to be written in an async event */
2621 	params->qp_handle_async_lo = lower_32_bits((uintptr_t)qp);
2622 	params->qp_handle_async_hi = upper_32_bits((uintptr_t)qp);
2623 
2624 	params->signal_all = (attrs->sq_sig_type == IB_SIGNAL_ALL_WR);
2625 	params->fmr_and_reserved_lkey = fmr_and_reserved_lkey;
2626 	params->pd = pd->pd_id;
2627 	params->dpi = pd->uctx ? pd->uctx->dpi : dev->dpi;
2628 	params->sq_cq_id = get_qlnxr_cq(attrs->send_cq)->icid;
2629 	params->stats_queue = 0;
2630 
2631 	params->rq_cq_id = get_qlnxr_cq(attrs->recv_cq)->icid;
2632 
2633 	if (qp->srq) {
2634 		/* QP is associated with SRQ instead of RQ */
2635 		params->srq_id = qp->srq->srq_id;
2636 		params->use_srq = true;
2637 		QL_DPRINT11(ha, "exit srq_id = 0x%x use_srq = 0x%x\n",
2638 			params->srq_id, params->use_srq);
2639 		return;
2640 	}
2641 
2642 	params->srq_id = 0;
2643 	params->use_srq = false;
2644 
2645 	QL_DPRINT12(ha, "exit\n");
2646 	return;
2647 }
2648 
2649 
2650 static inline void
2651 qlnxr_qp_user_print( struct qlnxr_dev *dev,
2652 	struct qlnxr_qp *qp)
2653 {
2654 	QL_DPRINT12((dev->ha), "qp=%p. sq_addr=0x%llx, sq_len=%zd, "
2655 		"rq_addr=0x%llx, rq_len=%zd\n",
2656 		qp, qp->usq.buf_addr, qp->usq.buf_len, qp->urq.buf_addr,
2657 		qp->urq.buf_len);
2658 	return;
2659 }
2660 
2661 static int
2662 qlnxr_idr_add(struct qlnxr_dev *dev, void *ptr, u32 id)
2663 {
2664 	u32		newid;
2665 	int		rc;
2666 	qlnx_host_t	*ha;
2667 
2668 	ha = dev->ha;
2669 
2670 	QL_DPRINT12(ha, "enter\n");
2671 
2672 	if (!QLNX_IS_IWARP(dev))
2673 		return 0;
2674 
2675 	do {
2676 		if (!idr_pre_get(&dev->qpidr, GFP_KERNEL)) {
2677 			QL_DPRINT11(ha, "idr_pre_get failed\n");
2678 			return -ENOMEM;
2679 		}
2680 
2681 		mtx_lock(&dev->idr_lock);
2682 
2683 		rc = idr_get_new_above(&dev->qpidr, ptr, id, &newid);
2684 
2685 		mtx_unlock(&dev->idr_lock);
2686 
2687 	} while (rc == -EAGAIN);
2688 
2689 	QL_DPRINT12(ha, "exit [%d]\n", rc);
2690 
2691 	return rc;
2692 }
2693 
2694 static void
2695 qlnxr_idr_remove(struct qlnxr_dev *dev, u32 id)
2696 {
2697 	qlnx_host_t	*ha;
2698 
2699 	ha = dev->ha;
2700 
2701 	QL_DPRINT12(ha, "enter\n");
2702 
2703 	if (!QLNX_IS_IWARP(dev))
2704 		return;
2705 
2706 	mtx_lock(&dev->idr_lock);
2707 	idr_remove(&dev->qpidr, id);
2708 	mtx_unlock(&dev->idr_lock);
2709 
2710 	QL_DPRINT12(ha, "exit \n");
2711 
2712 	return;
2713 }
2714 
2715 static inline void
2716 qlnxr_iwarp_populate_user_qp(struct qlnxr_dev *dev,
2717 	struct qlnxr_qp *qp,
2718 	struct ecore_rdma_create_qp_out_params *out_params)
2719 {
2720 	qlnx_host_t	*ha;
2721 
2722 	ha = dev->ha;
2723 
2724 	QL_DPRINT12(ha, "enter\n");
2725 
2726 	qp->usq.pbl_tbl->va = out_params->sq_pbl_virt;
2727 	qp->usq.pbl_tbl->pa = out_params->sq_pbl_phys;
2728 
2729 	qlnxr_populate_pbls(dev, qp->usq.umem, qp->usq.pbl_tbl,
2730 			   &qp->usq.pbl_info);
2731 
2732 	if (qp->srq) {
2733 		QL_DPRINT11(ha, "qp->srq = %p\n", qp->srq);
2734 		return;
2735 	}
2736 
2737 	qp->urq.pbl_tbl->va = out_params->rq_pbl_virt;
2738 	qp->urq.pbl_tbl->pa = out_params->rq_pbl_phys;
2739 
2740 	qlnxr_populate_pbls(dev, qp->urq.umem, qp->urq.pbl_tbl,
2741 			   &qp->urq.pbl_info);
2742 
2743 	QL_DPRINT12(ha, "exit\n");
2744 	return;
2745 }
2746 
2747 static int
2748 qlnxr_create_user_qp(struct qlnxr_dev *dev,
2749 	struct qlnxr_qp *qp,
2750 	struct ib_pd *ibpd,
2751 	struct ib_udata *udata,
2752 	struct ib_qp_init_attr *attrs)
2753 {
2754 	struct ecore_rdma_destroy_qp_out_params d_out_params;
2755 	struct ecore_rdma_create_qp_in_params in_params;
2756 	struct ecore_rdma_create_qp_out_params out_params;
2757 	struct qlnxr_pd *pd = get_qlnxr_pd(ibpd);
2758 	struct ib_ucontext *ib_ctx = NULL;
2759 	struct qlnxr_ucontext *ctx = NULL;
2760 	struct qlnxr_create_qp_ureq ureq;
2761 	int alloc_and_init = QLNX_IS_ROCE(dev);
2762 	int rc = -EINVAL;
2763 	qlnx_host_t	*ha;
2764 
2765 	ha = dev->ha;
2766 
2767 	QL_DPRINT12(ha, "enter\n");
2768 
2769 	ib_ctx = ibpd->uobject->context;
2770 	ctx = get_qlnxr_ucontext(ib_ctx);
2771 
2772 	memset(&ureq, 0, sizeof(ureq));
2773 	rc = ib_copy_from_udata(&ureq, udata, sizeof(ureq));
2774 
2775 	if (rc) {
2776 		QL_DPRINT11(ha, "ib_copy_from_udata failed [%d]\n", rc);
2777 		return rc;
2778 	}
2779 
2780 	/* SQ - read access only (0), dma sync not required (0) */
2781 	rc = qlnxr_init_user_queue(ib_ctx, dev, &qp->usq, ureq.sq_addr,
2782 				  ureq.sq_len, 0, 0,
2783 				  alloc_and_init);
2784 	if (rc) {
2785 		QL_DPRINT11(ha, "qlnxr_init_user_queue failed [%d]\n", rc);
2786 		return rc;
2787 	}
2788 
2789 	if (!qp->srq) {
2790 		/* RQ - read access only (0), dma sync not required (0) */
2791 		rc = qlnxr_init_user_queue(ib_ctx, dev, &qp->urq, ureq.rq_addr,
2792 					  ureq.rq_len, 0, 0,
2793 					  alloc_and_init);
2794 
2795 		if (rc) {
2796 			QL_DPRINT11(ha, "qlnxr_init_user_queue failed [%d]\n", rc);
2797 			return rc;
2798 		}
2799 	}
2800 
2801 	memset(&in_params, 0, sizeof(in_params));
2802 	qlnxr_init_common_qp_in_params(dev, pd, qp, attrs, false, &in_params);
2803 	in_params.qp_handle_lo = ureq.qp_handle_lo;
2804 	in_params.qp_handle_hi = ureq.qp_handle_hi;
2805 	in_params.sq_num_pages = qp->usq.pbl_info.num_pbes;
2806 	in_params.sq_pbl_ptr = qp->usq.pbl_tbl->pa;
2807 
2808 	if (!qp->srq) {
2809 		in_params.rq_num_pages = qp->urq.pbl_info.num_pbes;
2810 		in_params.rq_pbl_ptr = qp->urq.pbl_tbl->pa;
2811 	}
2812 
2813 	qp->ecore_qp = ecore_rdma_create_qp(dev->rdma_ctx, &in_params, &out_params);
2814 
2815 	if (!qp->ecore_qp) {
2816 		rc = -ENOMEM;
2817 		QL_DPRINT11(ha, "ecore_rdma_create_qp failed\n");
2818 		goto err1;
2819 	}
2820 
2821 	if (QLNX_IS_IWARP(dev))
2822 		qlnxr_iwarp_populate_user_qp(dev, qp, &out_params);
2823 
2824 	qp->qp_id = out_params.qp_id;
2825 	qp->icid = out_params.icid;
2826 
2827 	rc = qlnxr_copy_qp_uresp(dev, qp, udata);
2828 
2829 	if (rc) {
2830 		QL_DPRINT11(ha, "qlnxr_copy_qp_uresp failed\n");
2831 		goto err;
2832 	}
2833 
2834 	qlnxr_qp_user_print(dev, qp);
2835 
2836 	QL_DPRINT12(ha, "exit\n");
2837 	return 0;
2838 err:
2839 	rc = ecore_rdma_destroy_qp(dev->rdma_ctx, qp->ecore_qp, &d_out_params);
2840 
2841 	if (rc)
2842 		QL_DPRINT12(ha, "fatal fault\n");
2843 
2844 err1:
2845 	qlnxr_cleanup_user(dev, qp);
2846 
2847 	QL_DPRINT12(ha, "exit[%d]\n", rc);
2848 	return rc;
2849 }
2850 
2851 static void
2852 qlnxr_set_roce_db_info(struct qlnxr_dev *dev,
2853 	struct qlnxr_qp *qp)
2854 {
2855 	qlnx_host_t	*ha;
2856 
2857 	ha = dev->ha;
2858 
2859 	QL_DPRINT12(ha, "enter qp = %p qp->srq %p\n", qp, qp->srq);
2860 
2861 	qp->sq.db = dev->db_addr +
2862 		DB_ADDR_SHIFT(DQ_PWM_OFFSET_XCM_RDMA_SQ_PROD);
2863 	qp->sq.db_data.data.icid = qp->icid + 1;
2864 
2865 	if (!qp->srq) {
2866 		qp->rq.db = dev->db_addr +
2867 			DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_ROCE_RQ_PROD);
2868 		qp->rq.db_data.data.icid = qp->icid;
2869 	}
2870 
2871 	QL_DPRINT12(ha, "exit\n");
2872 	return;
2873 }
2874 
2875 static void
2876 qlnxr_set_iwarp_db_info(struct qlnxr_dev *dev,
2877 	struct qlnxr_qp *qp)
2878 
2879 {
2880 	qlnx_host_t	*ha;
2881 
2882 	ha = dev->ha;
2883 
2884 	QL_DPRINT12(ha, "enter qp = %p qp->srq %p\n", qp, qp->srq);
2885 
2886 	qp->sq.db = dev->db_addr +
2887 		DB_ADDR_SHIFT(DQ_PWM_OFFSET_XCM_RDMA_SQ_PROD);
2888 	qp->sq.db_data.data.icid = qp->icid;
2889 
2890 	if (!qp->srq) {
2891 		qp->rq.db = dev->db_addr +
2892 			DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_IWARP_RQ_PROD);
2893 		qp->rq.db_data.data.icid = qp->icid;
2894 
2895 		qp->rq.iwarp_db2 = dev->db_addr +
2896 			DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_FLAGS);
2897 		qp->rq.iwarp_db2_data.data.icid = qp->icid;
2898 		qp->rq.iwarp_db2_data.data.value = DQ_TCM_IWARP_POST_RQ_CF_CMD;
2899 	}
2900 
2901 	QL_DPRINT12(ha,
2902 		"qp->sq.db = %p qp->sq.db_data.data.icid =0x%x\n"
2903 		"\t\t\tqp->rq.db = %p qp->rq.db_data.data.icid =0x%x\n"
2904 		"\t\t\tqp->rq.iwarp_db2 = %p qp->rq.iwarp_db2.data.icid =0x%x"
2905 		" qp->rq.iwarp_db2.data.prod_val =0x%x\n",
2906 		qp->sq.db, qp->sq.db_data.data.icid,
2907 		qp->rq.db, qp->rq.db_data.data.icid,
2908 		qp->rq.iwarp_db2, qp->rq.iwarp_db2_data.data.icid,
2909 		qp->rq.iwarp_db2_data.data.value);
2910 
2911 	QL_DPRINT12(ha, "exit\n");
2912 	return;
2913 }
2914 
2915 static int
2916 qlnxr_roce_create_kernel_qp(struct qlnxr_dev *dev,
2917 	struct qlnxr_qp *qp,
2918 	struct ecore_rdma_create_qp_in_params *in_params,
2919 	u32 n_sq_elems,
2920 	u32 n_rq_elems)
2921 {
2922 	struct ecore_rdma_create_qp_out_params out_params;
2923 	int		rc;
2924 	qlnx_host_t	*ha;
2925 
2926 	ha = dev->ha;
2927 
2928 	QL_DPRINT12(ha, "enter\n");
2929 
2930         rc = ecore_chain_alloc(
2931                 dev->cdev,
2932                 ECORE_CHAIN_USE_TO_PRODUCE,
2933                 ECORE_CHAIN_MODE_PBL,
2934                 ECORE_CHAIN_CNT_TYPE_U32,
2935                 n_sq_elems,
2936                 QLNXR_SQE_ELEMENT_SIZE,
2937                 &qp->sq.pbl,
2938                 NULL);
2939 
2940 	if (rc) {
2941 		QL_DPRINT11(ha, "ecore_chain_alloc qp->sq.pbl failed[%d]\n", rc);
2942 		return rc;
2943 	}
2944 
2945 	in_params->sq_num_pages = ecore_chain_get_page_cnt(&qp->sq.pbl);
2946 	in_params->sq_pbl_ptr = ecore_chain_get_pbl_phys(&qp->sq.pbl);
2947 
2948 	if (!qp->srq) {
2949 
2950                 rc = ecore_chain_alloc(
2951                         dev->cdev,
2952                         ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
2953                         ECORE_CHAIN_MODE_PBL,
2954                         ECORE_CHAIN_CNT_TYPE_U32,
2955                         n_rq_elems,
2956                         QLNXR_RQE_ELEMENT_SIZE,
2957                         &qp->rq.pbl,
2958                         NULL);
2959 
2960 		if (rc) {
2961 			QL_DPRINT11(ha,
2962 				"ecore_chain_alloc qp->rq.pbl failed[%d]\n", rc);
2963 			return rc;
2964 		}
2965 
2966 		in_params->rq_num_pages = ecore_chain_get_page_cnt(&qp->rq.pbl);
2967 		in_params->rq_pbl_ptr = ecore_chain_get_pbl_phys(&qp->rq.pbl);
2968 	}
2969 
2970 	qp->ecore_qp = ecore_rdma_create_qp(dev->rdma_ctx, in_params, &out_params);
2971 
2972 	if (!qp->ecore_qp) {
2973 		QL_DPRINT11(ha, "qp->ecore_qp == NULL\n");
2974 		return -EINVAL;
2975 	}
2976 
2977 	qp->qp_id = out_params.qp_id;
2978 	qp->icid = out_params.icid;
2979 
2980 	qlnxr_set_roce_db_info(dev, qp);
2981 
2982 	QL_DPRINT12(ha, "exit\n");
2983 	return 0;
2984 }
2985 
2986 static int
2987 qlnxr_iwarp_create_kernel_qp(struct qlnxr_dev *dev,
2988 	struct qlnxr_qp *qp,
2989 	struct ecore_rdma_create_qp_in_params *in_params,
2990 	u32 n_sq_elems,
2991 	u32 n_rq_elems)
2992 {
2993 	struct ecore_rdma_destroy_qp_out_params d_out_params;
2994 	struct ecore_rdma_create_qp_out_params out_params;
2995 	struct ecore_chain_ext_pbl ext_pbl;
2996 	int rc;
2997 	qlnx_host_t	*ha;
2998 
2999 	ha = dev->ha;
3000 
3001 	QL_DPRINT12(ha, "enter\n");
3002 
3003 	in_params->sq_num_pages = ECORE_CHAIN_PAGE_CNT(n_sq_elems,
3004 						     QLNXR_SQE_ELEMENT_SIZE,
3005 						     ECORE_CHAIN_MODE_PBL);
3006 	in_params->rq_num_pages = ECORE_CHAIN_PAGE_CNT(n_rq_elems,
3007 						     QLNXR_RQE_ELEMENT_SIZE,
3008 						     ECORE_CHAIN_MODE_PBL);
3009 
3010 	QL_DPRINT12(ha, "n_sq_elems = 0x%x"
3011 		" n_rq_elems = 0x%x in_params\n"
3012 		"\t\t\tqp_handle_lo\t\t= 0x%08x\n"
3013 		"\t\t\tqp_handle_hi\t\t= 0x%08x\n"
3014 		"\t\t\tqp_handle_async_lo\t\t= 0x%08x\n"
3015 		"\t\t\tqp_handle_async_hi\t\t= 0x%08x\n"
3016 		"\t\t\tuse_srq\t\t\t= 0x%x\n"
3017 		"\t\t\tsignal_all\t\t= 0x%x\n"
3018 		"\t\t\tfmr_and_reserved_lkey\t= 0x%x\n"
3019 		"\t\t\tpd\t\t\t= 0x%x\n"
3020 		"\t\t\tdpi\t\t\t= 0x%x\n"
3021 		"\t\t\tsq_cq_id\t\t\t= 0x%x\n"
3022 		"\t\t\tsq_num_pages\t\t= 0x%x\n"
3023 		"\t\t\tsq_pbl_ptr\t\t= %p\n"
3024 		"\t\t\tmax_sq_sges\t\t= 0x%x\n"
3025 		"\t\t\trq_cq_id\t\t\t= 0x%x\n"
3026 		"\t\t\trq_num_pages\t\t= 0x%x\n"
3027 		"\t\t\trq_pbl_ptr\t\t= %p\n"
3028 		"\t\t\tsrq_id\t\t\t= 0x%x\n"
3029 		"\t\t\tstats_queue\t\t= 0x%x\n",
3030 		n_sq_elems, n_rq_elems,
3031 		in_params->qp_handle_lo,
3032 		in_params->qp_handle_hi,
3033 		in_params->qp_handle_async_lo,
3034 		in_params->qp_handle_async_hi,
3035 		in_params->use_srq,
3036 		in_params->signal_all,
3037 		in_params->fmr_and_reserved_lkey,
3038 		in_params->pd,
3039 		in_params->dpi,
3040 		in_params->sq_cq_id,
3041 		in_params->sq_num_pages,
3042 		(void *)in_params->sq_pbl_ptr,
3043 		in_params->max_sq_sges,
3044 		in_params->rq_cq_id,
3045 		in_params->rq_num_pages,
3046 		(void *)in_params->rq_pbl_ptr,
3047 		in_params->srq_id,
3048 		in_params->stats_queue );
3049 
3050 	memset(&out_params, 0, sizeof (struct ecore_rdma_create_qp_out_params));
3051 	memset(&ext_pbl, 0, sizeof (struct ecore_chain_ext_pbl));
3052 
3053 	qp->ecore_qp = ecore_rdma_create_qp(dev->rdma_ctx, in_params, &out_params);
3054 
3055 	if (!qp->ecore_qp) {
3056 		QL_DPRINT11(ha, "ecore_rdma_create_qp failed\n");
3057 		return -EINVAL;
3058 	}
3059 
3060 	/* Now we allocate the chain */
3061 	ext_pbl.p_pbl_virt = out_params.sq_pbl_virt;
3062 	ext_pbl.p_pbl_phys = out_params.sq_pbl_phys;
3063 
3064 	QL_DPRINT12(ha, "ext_pbl.p_pbl_virt = %p "
3065 		"ext_pbl.p_pbl_phys = %p\n",
3066 		ext_pbl.p_pbl_virt, ext_pbl.p_pbl_phys);
3067 
3068         rc = ecore_chain_alloc(
3069                 dev->cdev,
3070                 ECORE_CHAIN_USE_TO_PRODUCE,
3071                 ECORE_CHAIN_MODE_PBL,
3072                 ECORE_CHAIN_CNT_TYPE_U32,
3073                 n_sq_elems,
3074                 QLNXR_SQE_ELEMENT_SIZE,
3075                 &qp->sq.pbl,
3076                 &ext_pbl);
3077 
3078 	if (rc) {
3079 		QL_DPRINT11(ha,
3080 			"ecore_chain_alloc qp->sq.pbl failed rc = %d\n", rc);
3081 		goto err;
3082 	}
3083 
3084 	ext_pbl.p_pbl_virt = out_params.rq_pbl_virt;
3085 	ext_pbl.p_pbl_phys = out_params.rq_pbl_phys;
3086 
3087 	QL_DPRINT12(ha, "ext_pbl.p_pbl_virt = %p "
3088 		"ext_pbl.p_pbl_phys = %p\n",
3089 		ext_pbl.p_pbl_virt, ext_pbl.p_pbl_phys);
3090 
3091 	if (!qp->srq) {
3092 
3093                 rc = ecore_chain_alloc(
3094                         dev->cdev,
3095                         ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
3096                         ECORE_CHAIN_MODE_PBL,
3097                         ECORE_CHAIN_CNT_TYPE_U32,
3098                         n_rq_elems,
3099                         QLNXR_RQE_ELEMENT_SIZE,
3100                         &qp->rq.pbl,
3101                         &ext_pbl);
3102 
3103 		if (rc) {
3104 			QL_DPRINT11(ha,, "ecore_chain_alloc qp->rq.pbl"
3105 				" failed rc = %d\n", rc);
3106 			goto err;
3107 		}
3108 	}
3109 
3110 	QL_DPRINT12(ha, "qp_id = 0x%x icid =0x%x\n",
3111 		out_params.qp_id, out_params.icid);
3112 
3113 	qp->qp_id = out_params.qp_id;
3114 	qp->icid = out_params.icid;
3115 
3116 	qlnxr_set_iwarp_db_info(dev, qp);
3117 
3118 	QL_DPRINT12(ha, "exit\n");
3119 	return 0;
3120 
3121 err:
3122 	ecore_rdma_destroy_qp(dev->rdma_ctx, qp->ecore_qp, &d_out_params);
3123 
3124 	QL_DPRINT12(ha, "exit rc = %d\n", rc);
3125 	return rc;
3126 }
3127 
3128 static int
3129 qlnxr_create_kernel_qp(struct qlnxr_dev *dev,
3130 	struct qlnxr_qp *qp,
3131 	struct ib_pd *ibpd,
3132 	struct ib_qp_init_attr *attrs)
3133 {
3134 	struct ecore_rdma_create_qp_in_params in_params;
3135 	struct qlnxr_pd *pd = get_qlnxr_pd(ibpd);
3136 	int rc = -EINVAL;
3137 	u32 n_rq_elems;
3138 	u32 n_sq_elems;
3139 	u32 n_sq_entries;
3140 	struct ecore_rdma_device *qattr = ecore_rdma_query_device(dev->rdma_ctx);
3141 	qlnx_host_t	*ha;
3142 
3143 	ha = dev->ha;
3144 
3145 	QL_DPRINT12(ha, "enter\n");
3146 
3147 	memset(&in_params, 0, sizeof(in_params));
3148 
3149 	/* A single work request may take up to MAX_SQ_WQE_SIZE elements in
3150 	 * the ring. The ring should allow at least a single WR, even if the
3151 	 * user requested none, due to allocation issues.
3152 	 * We should add an extra WR since the prod and cons indices of
3153 	 * wqe_wr_id are managed in such a way that the WQ is considered full
3154 	 * when (prod+1)%max_wr==cons. We currently don't do that because we
3155 	 * double the number of entries due an iSER issue that pushes far more
3156 	 * WRs than indicated. If we decline its ib_post_send() then we get
3157 	 * error prints in the dmesg we'd like to avoid.
3158 	 */
3159 	qp->sq.max_wr = min_t(u32, attrs->cap.max_send_wr * dev->wq_multiplier,
3160 			      qattr->max_wqe);
3161 
3162 	qp->wqe_wr_id = kzalloc(qp->sq.max_wr * sizeof(*qp->wqe_wr_id),
3163 			GFP_KERNEL);
3164 	if (!qp->wqe_wr_id) {
3165 		QL_DPRINT11(ha, "failed SQ shadow memory allocation\n");
3166 		return -ENOMEM;
3167 	}
3168 
3169 	/* QP handle to be written in CQE */
3170 	in_params.qp_handle_lo = lower_32_bits((uintptr_t)qp);
3171 	in_params.qp_handle_hi = upper_32_bits((uintptr_t)qp);
3172 
3173 	/* A single work request may take up to MAX_RQ_WQE_SIZE elements in
3174 	 * the ring. There ring should allow at least a single WR, even if the
3175 	 * user requested none, due to allocation issues.
3176 	 */
3177 	qp->rq.max_wr = (u16)max_t(u32, attrs->cap.max_recv_wr, 1);
3178 
3179 	/* Allocate driver internal RQ array */
3180 	if (!qp->srq) {
3181 		qp->rqe_wr_id = kzalloc(qp->rq.max_wr * sizeof(*qp->rqe_wr_id),
3182 					GFP_KERNEL);
3183 		if (!qp->rqe_wr_id) {
3184 			QL_DPRINT11(ha, "failed RQ shadow memory allocation\n");
3185 			kfree(qp->wqe_wr_id);
3186 			return -ENOMEM;
3187 		}
3188 	}
3189 
3190 	//qlnxr_init_common_qp_in_params(dev, pd, qp, attrs, true, &in_params);
3191 
3192         in_params.qp_handle_async_lo = lower_32_bits((uintptr_t)qp);
3193         in_params.qp_handle_async_hi = upper_32_bits((uintptr_t)qp);
3194 
3195         in_params.signal_all = (attrs->sq_sig_type == IB_SIGNAL_ALL_WR);
3196         in_params.fmr_and_reserved_lkey = true;
3197         in_params.pd = pd->pd_id;
3198         in_params.dpi = pd->uctx ? pd->uctx->dpi : dev->dpi;
3199         in_params.sq_cq_id = get_qlnxr_cq(attrs->send_cq)->icid;
3200         in_params.stats_queue = 0;
3201 
3202         in_params.rq_cq_id = get_qlnxr_cq(attrs->recv_cq)->icid;
3203 
3204         if (qp->srq) {
3205                 /* QP is associated with SRQ instead of RQ */
3206                 in_params.srq_id = qp->srq->srq_id;
3207                 in_params.use_srq = true;
3208                 QL_DPRINT11(ha, "exit srq_id = 0x%x use_srq = 0x%x\n",
3209                         in_params.srq_id, in_params.use_srq);
3210         } else {
3211         	in_params.srq_id = 0;
3212 		in_params.use_srq = false;
3213 	}
3214 
3215 	n_sq_entries = attrs->cap.max_send_wr;
3216 	n_sq_entries = min_t(u32, n_sq_entries, qattr->max_wqe);
3217 	n_sq_entries = max_t(u32, n_sq_entries, 1);
3218 	n_sq_elems = n_sq_entries * QLNXR_MAX_SQE_ELEMENTS_PER_SQE;
3219 
3220 	n_rq_elems = qp->rq.max_wr * QLNXR_MAX_RQE_ELEMENTS_PER_RQE;
3221 
3222 	if (QLNX_IS_ROCE(dev)) {
3223 		rc = qlnxr_roce_create_kernel_qp(dev, qp, &in_params,
3224 						n_sq_elems, n_rq_elems);
3225 	} else {
3226 		rc = qlnxr_iwarp_create_kernel_qp(dev, qp, &in_params,
3227 						 n_sq_elems, n_rq_elems);
3228 	}
3229 
3230 	if (rc)
3231 		qlnxr_cleanup_kernel(dev, qp);
3232 
3233 	QL_DPRINT12(ha, "exit [%d]\n", rc);
3234 	return rc;
3235 }
3236 
3237 struct ib_qp *
3238 qlnxr_create_qp(struct ib_pd *ibpd,
3239 		struct ib_qp_init_attr *attrs,
3240 		struct ib_udata *udata)
3241 {
3242 	struct qlnxr_dev *dev = get_qlnxr_dev(ibpd->device);
3243 	struct qlnxr_pd *pd = get_qlnxr_pd(ibpd);
3244 	struct qlnxr_qp *qp;
3245 	int rc = 0;
3246 	qlnx_host_t	*ha;
3247 
3248 	ha = dev->ha;
3249 
3250 	QL_DPRINT12(ha, "enter\n");
3251 
3252 	rc = qlnxr_check_qp_attrs(ibpd, dev, attrs, udata);
3253 	if (rc) {
3254 		QL_DPRINT11(ha, "qlnxr_check_qp_attrs failed [%d]\n", rc);
3255 		return ERR_PTR(rc);
3256 	}
3257 
3258 	QL_DPRINT12(ha, "called from %s, event_handle=%p,"
3259 		" eepd=%p sq_cq=%p, sq_icid=%d, rq_cq=%p, rq_icid=%d\n",
3260 		(udata ? "user library" : "kernel"),
3261 		attrs->event_handler, pd,
3262 		get_qlnxr_cq(attrs->send_cq),
3263 		get_qlnxr_cq(attrs->send_cq)->icid,
3264 		get_qlnxr_cq(attrs->recv_cq),
3265 		get_qlnxr_cq(attrs->recv_cq)->icid);
3266 
3267 	qp = qlnx_zalloc(sizeof(struct qlnxr_qp));
3268 
3269 	if (!qp) {
3270 		QL_DPRINT11(ha, "kzalloc(qp) failed\n");
3271 		return ERR_PTR(-ENOMEM);
3272 	}
3273 
3274 	qlnxr_set_common_qp_params(dev, qp, pd, attrs);
3275 
3276 	if (attrs->qp_type == IB_QPT_GSI) {
3277 		QL_DPRINT11(ha, "calling qlnxr_create_gsi_qp\n");
3278 		return qlnxr_create_gsi_qp(dev, attrs, qp);
3279 	}
3280 
3281 	if (udata) {
3282 		rc = qlnxr_create_user_qp(dev, qp, ibpd, udata, attrs);
3283 
3284 		if (rc) {
3285 			QL_DPRINT11(ha, "qlnxr_create_user_qp failed\n");
3286 			goto err;
3287 		}
3288 	} else {
3289 		rc = qlnxr_create_kernel_qp(dev, qp, ibpd, attrs);
3290 
3291 		if (rc) {
3292 			QL_DPRINT11(ha, "qlnxr_create_kernel_qp failed\n");
3293 			goto err;
3294 		}
3295 	}
3296 
3297 	qp->ibqp.qp_num = qp->qp_id;
3298 
3299 	rc = qlnxr_idr_add(dev, qp, qp->qp_id);
3300 
3301 	if (rc) {
3302 		QL_DPRINT11(ha, "qlnxr_idr_add failed\n");
3303 		goto err;
3304 	}
3305 
3306 	QL_DPRINT12(ha, "exit [%p]\n", &qp->ibqp);
3307 
3308 	return &qp->ibqp;
3309 err:
3310 	kfree(qp);
3311 
3312 	QL_DPRINT12(ha, "failed exit\n");
3313 	return ERR_PTR(-EFAULT);
3314 }
3315 
3316 
3317 static enum ib_qp_state
3318 qlnxr_get_ibqp_state(enum ecore_roce_qp_state qp_state)
3319 {
3320 	enum ib_qp_state state = IB_QPS_ERR;
3321 
3322 	switch (qp_state) {
3323 	case ECORE_ROCE_QP_STATE_RESET:
3324 		state = IB_QPS_RESET;
3325 		break;
3326 
3327 	case ECORE_ROCE_QP_STATE_INIT:
3328 		state = IB_QPS_INIT;
3329 		break;
3330 
3331 	case ECORE_ROCE_QP_STATE_RTR:
3332 		state = IB_QPS_RTR;
3333 		break;
3334 
3335 	case ECORE_ROCE_QP_STATE_RTS:
3336 		state = IB_QPS_RTS;
3337 		break;
3338 
3339 	case ECORE_ROCE_QP_STATE_SQD:
3340 		state = IB_QPS_SQD;
3341 		break;
3342 
3343 	case ECORE_ROCE_QP_STATE_ERR:
3344 		state = IB_QPS_ERR;
3345 		break;
3346 
3347 	case ECORE_ROCE_QP_STATE_SQE:
3348 		state = IB_QPS_SQE;
3349 		break;
3350 	}
3351 	return state;
3352 }
3353 
3354 static enum ecore_roce_qp_state
3355 qlnxr_get_state_from_ibqp( enum ib_qp_state qp_state)
3356 {
3357 	enum ecore_roce_qp_state ecore_qp_state;
3358 
3359 	ecore_qp_state = ECORE_ROCE_QP_STATE_ERR;
3360 
3361 	switch (qp_state) {
3362 	case IB_QPS_RESET:
3363 		ecore_qp_state =  ECORE_ROCE_QP_STATE_RESET;
3364 		break;
3365 
3366 	case IB_QPS_INIT:
3367 		ecore_qp_state =  ECORE_ROCE_QP_STATE_INIT;
3368 		break;
3369 
3370 	case IB_QPS_RTR:
3371 		ecore_qp_state =  ECORE_ROCE_QP_STATE_RTR;
3372 		break;
3373 
3374 	case IB_QPS_RTS:
3375 		ecore_qp_state =  ECORE_ROCE_QP_STATE_RTS;
3376 		break;
3377 
3378 	case IB_QPS_SQD:
3379 		ecore_qp_state =  ECORE_ROCE_QP_STATE_SQD;
3380 		break;
3381 
3382 	case IB_QPS_ERR:
3383 		ecore_qp_state =  ECORE_ROCE_QP_STATE_ERR;
3384 		break;
3385 
3386 	default:
3387 		ecore_qp_state =  ECORE_ROCE_QP_STATE_ERR;
3388 		break;
3389 	}
3390 
3391 	return (ecore_qp_state);
3392 }
3393 
3394 static void
3395 qlnxr_reset_qp_hwq_info(struct qlnxr_qp_hwq_info *qph)
3396 {
3397 	ecore_chain_reset(&qph->pbl);
3398 	qph->prod = qph->cons = 0;
3399 	qph->wqe_cons = 0;
3400 	qph->db_data.data.value = cpu_to_le16(0);
3401 
3402 	return;
3403 }
3404 
3405 static int
3406 qlnxr_update_qp_state(struct qlnxr_dev *dev,
3407 	struct qlnxr_qp *qp,
3408 	enum ecore_roce_qp_state new_state)
3409 {
3410 	int		status = 0;
3411 	uint32_t	reg_addr;
3412 	struct ecore_dev *cdev;
3413 	qlnx_host_t	*ha;
3414 
3415 	ha = dev->ha;
3416 	cdev = &ha->cdev;
3417 
3418 	QL_DPRINT12(ha, "enter qp = %p new_state = 0x%x qp->state = 0x%x\n",
3419 		qp, new_state, qp->state);
3420 
3421 	if (new_state == qp->state) {
3422 		return 0;
3423 	}
3424 
3425 	switch (qp->state) {
3426 	case ECORE_ROCE_QP_STATE_RESET:
3427 		switch (new_state) {
3428 		case ECORE_ROCE_QP_STATE_INIT:
3429 			qp->prev_wqe_size = 0;
3430 			qlnxr_reset_qp_hwq_info(&qp->sq);
3431 			if (!(qp->srq))
3432 				qlnxr_reset_qp_hwq_info(&qp->rq);
3433 			break;
3434 		default:
3435 			status = -EINVAL;
3436 			break;
3437 		};
3438 		break;
3439 	case ECORE_ROCE_QP_STATE_INIT:
3440 		/* INIT->XXX */
3441 		switch (new_state) {
3442 		case ECORE_ROCE_QP_STATE_RTR:
3443 		/* Update doorbell (in case post_recv was done before move to RTR) */
3444 			if (qp->srq)
3445 				break;
3446 			wmb();
3447 			//writel(qp->rq.db_data.raw, qp->rq.db);
3448 			//if (QLNX_IS_IWARP(dev))
3449 			//	writel(qp->rq.iwarp_db2_data.raw,
3450 			//	       qp->rq.iwarp_db2);
3451 
3452 			reg_addr = (uint32_t)((uint8_t *)qp->rq.db -
3453 					(uint8_t *)cdev->doorbells);
3454 
3455 			bus_write_4(ha->pci_dbells, reg_addr, qp->rq.db_data.raw);
3456 			bus_barrier(ha->pci_dbells,  0, 0, BUS_SPACE_BARRIER_READ);
3457 
3458 			if (QLNX_IS_IWARP(dev)) {
3459 				reg_addr = (uint32_t)((uint8_t *)qp->rq.iwarp_db2 -
3460 					(uint8_t *)cdev->doorbells);
3461 				bus_write_4(ha->pci_dbells, reg_addr,\
3462 					qp->rq.iwarp_db2_data.raw);
3463 				bus_barrier(ha->pci_dbells,  0, 0,\
3464 					BUS_SPACE_BARRIER_READ);
3465 			}
3466 
3467 
3468 			mmiowb();
3469 			break;
3470 		case ECORE_ROCE_QP_STATE_ERR:
3471 			/* TBD:flush qps... */
3472 			break;
3473 		default:
3474 			/* invalid state change. */
3475 			status = -EINVAL;
3476 			break;
3477 		};
3478 		break;
3479 	case ECORE_ROCE_QP_STATE_RTR:
3480 		/* RTR->XXX */
3481 		switch (new_state) {
3482 		case ECORE_ROCE_QP_STATE_RTS:
3483 			break;
3484 		case ECORE_ROCE_QP_STATE_ERR:
3485 			break;
3486 		default:
3487 			/* invalid state change. */
3488 			status = -EINVAL;
3489 			break;
3490 		};
3491 		break;
3492 	case ECORE_ROCE_QP_STATE_RTS:
3493 		/* RTS->XXX */
3494 		switch (new_state) {
3495 		case ECORE_ROCE_QP_STATE_SQD:
3496 			break;
3497 		case ECORE_ROCE_QP_STATE_ERR:
3498 			break;
3499 		default:
3500 			/* invalid state change. */
3501 			status = -EINVAL;
3502 			break;
3503 		};
3504 		break;
3505 	case ECORE_ROCE_QP_STATE_SQD:
3506 		/* SQD->XXX */
3507 		switch (new_state) {
3508 		case ECORE_ROCE_QP_STATE_RTS:
3509 		case ECORE_ROCE_QP_STATE_ERR:
3510 			break;
3511 		default:
3512 			/* invalid state change. */
3513 			status = -EINVAL;
3514 			break;
3515 		};
3516 		break;
3517 	case ECORE_ROCE_QP_STATE_ERR:
3518 		/* ERR->XXX */
3519 		switch (new_state) {
3520 		case ECORE_ROCE_QP_STATE_RESET:
3521 			if ((qp->rq.prod != qp->rq.cons) ||
3522 			    (qp->sq.prod != qp->sq.cons)) {
3523 				QL_DPRINT11(ha,
3524 					"Error->Reset with rq/sq "
3525 					"not empty rq.prod=0x%x rq.cons=0x%x"
3526 					" sq.prod=0x%x sq.cons=0x%x\n",
3527 					qp->rq.prod, qp->rq.cons,
3528 					qp->sq.prod, qp->sq.cons);
3529 				status = -EINVAL;
3530 			}
3531 			break;
3532 		default:
3533 			status = -EINVAL;
3534 			break;
3535 		};
3536 		break;
3537 	default:
3538 		status = -EINVAL;
3539 		break;
3540 	};
3541 
3542 	QL_DPRINT12(ha, "exit\n");
3543 	return status;
3544 }
3545 
3546 int
3547 qlnxr_modify_qp(struct ib_qp	*ibqp,
3548 	struct ib_qp_attr	*attr,
3549 	int			attr_mask,
3550 	struct ib_udata		*udata)
3551 {
3552 	int rc = 0;
3553 	struct qlnxr_qp *qp = get_qlnxr_qp(ibqp);
3554 	struct qlnxr_dev *dev = get_qlnxr_dev(&qp->dev->ibdev);
3555 	struct ecore_rdma_modify_qp_in_params qp_params = { 0 };
3556 	enum ib_qp_state old_qp_state, new_qp_state;
3557 	struct ecore_rdma_device *qattr = ecore_rdma_query_device(dev->rdma_ctx);
3558 	qlnx_host_t	*ha;
3559 
3560 	ha = dev->ha;
3561 
3562 	QL_DPRINT12(ha,
3563 		"enter qp = %p attr_mask = 0x%x, state = %d udata = %p\n",
3564 		qp, attr_mask, attr->qp_state, udata);
3565 
3566 	old_qp_state = qlnxr_get_ibqp_state(qp->state);
3567 	if (attr_mask & IB_QP_STATE)
3568 		new_qp_state = attr->qp_state;
3569 	else
3570 		new_qp_state = old_qp_state;
3571 
3572 	if (QLNX_IS_ROCE(dev)) {
3573 #if __FreeBSD_version >= 1100000
3574 		if (!ib_modify_qp_is_ok(old_qp_state,
3575 					new_qp_state,
3576 					ibqp->qp_type,
3577 					attr_mask,
3578 					IB_LINK_LAYER_ETHERNET)) {
3579 			QL_DPRINT12(ha,
3580 				"invalid attribute mask=0x%x"
3581 				" specified for qpn=0x%x of type=0x%x \n"
3582 				" old_qp_state=0x%x, new_qp_state=0x%x\n",
3583 				attr_mask, qp->qp_id, ibqp->qp_type,
3584 				old_qp_state, new_qp_state);
3585 			rc = -EINVAL;
3586 			goto err;
3587 		}
3588 #else
3589 		if (!ib_modify_qp_is_ok(old_qp_state,
3590 					new_qp_state,
3591 					ibqp->qp_type,
3592 					attr_mask )) {
3593 			QL_DPRINT12(ha,
3594 				"invalid attribute mask=0x%x"
3595 				" specified for qpn=0x%x of type=0x%x \n"
3596 				" old_qp_state=0x%x, new_qp_state=0x%x\n",
3597 				attr_mask, qp->qp_id, ibqp->qp_type,
3598 				old_qp_state, new_qp_state);
3599 			rc = -EINVAL;
3600 			goto err;
3601 		}
3602 
3603 #endif /* #if __FreeBSD_version >= 1100000 */
3604 	}
3605 	/* translate the masks... */
3606 	if (attr_mask & IB_QP_STATE) {
3607 		SET_FIELD(qp_params.modify_flags,
3608 			  ECORE_RDMA_MODIFY_QP_VALID_NEW_STATE, 1);
3609 		qp_params.new_state = qlnxr_get_state_from_ibqp(attr->qp_state);
3610 	}
3611 
3612 	// TBD consider changing ecore to be a flag as well...
3613 	if (attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY)
3614 		qp_params.sqd_async = true;
3615 
3616 	if (attr_mask & IB_QP_PKEY_INDEX) {
3617 		SET_FIELD(qp_params.modify_flags,
3618 			  ECORE_ROCE_MODIFY_QP_VALID_PKEY,
3619 			  1);
3620 		if (attr->pkey_index >= QLNXR_ROCE_PKEY_TABLE_LEN) {
3621 			rc = -EINVAL;
3622 			goto err;
3623 		}
3624 
3625 		qp_params.pkey = QLNXR_ROCE_PKEY_DEFAULT;
3626 	}
3627 
3628 	if (attr_mask & IB_QP_QKEY) {
3629 		qp->qkey = attr->qkey;
3630 	}
3631 
3632 	/* tbd consider splitting in ecore.. */
3633 	if (attr_mask & IB_QP_ACCESS_FLAGS) {
3634 		SET_FIELD(qp_params.modify_flags,
3635 			  ECORE_RDMA_MODIFY_QP_VALID_RDMA_OPS_EN, 1);
3636 		qp_params.incoming_rdma_read_en =
3637 			attr->qp_access_flags & IB_ACCESS_REMOTE_READ;
3638 		qp_params.incoming_rdma_write_en =
3639 			attr->qp_access_flags & IB_ACCESS_REMOTE_WRITE;
3640 		qp_params.incoming_atomic_en =
3641 			attr->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC;
3642 	}
3643 
3644 	if (attr_mask & (IB_QP_AV | IB_QP_PATH_MTU)) {
3645 		if (attr_mask & IB_QP_PATH_MTU) {
3646 			if (attr->path_mtu < IB_MTU_256 ||
3647 			    attr->path_mtu > IB_MTU_4096) {
3648 
3649 				QL_DPRINT12(ha,
3650 					"Only MTU sizes of 256, 512, 1024,"
3651 					" 2048 and 4096 are supported "
3652 					" attr->path_mtu = [%d]\n",
3653 					attr->path_mtu);
3654 
3655 				rc = -EINVAL;
3656 				goto err;
3657 			}
3658 			qp->mtu = min(ib_mtu_enum_to_int(attr->path_mtu),
3659 				      ib_mtu_enum_to_int(
3660 						iboe_get_mtu(dev->ha->ifp->if_mtu)));
3661 		}
3662 
3663 		if (qp->mtu == 0) {
3664 			qp->mtu = ib_mtu_enum_to_int(
3665 					iboe_get_mtu(dev->ha->ifp->if_mtu));
3666 			QL_DPRINT12(ha, "fixing zetoed MTU to qp->mtu = %d\n",
3667 				qp->mtu);
3668 		}
3669 
3670 		SET_FIELD(qp_params.modify_flags,
3671 			  ECORE_ROCE_MODIFY_QP_VALID_ADDRESS_VECTOR,
3672 			  1);
3673 
3674 		qp_params.traffic_class_tos = attr->ah_attr.grh.traffic_class;
3675 		qp_params.flow_label = attr->ah_attr.grh.flow_label;
3676 		qp_params.hop_limit_ttl = attr->ah_attr.grh.hop_limit;
3677 
3678 		qp->sgid_idx = attr->ah_attr.grh.sgid_index;
3679 
3680 		get_gid_info(ibqp, attr, attr_mask, dev, qp, &qp_params);
3681 
3682 		rc = qlnxr_get_dmac(dev, &attr->ah_attr, qp_params.remote_mac_addr);
3683 		if (rc)
3684 			return rc;
3685 
3686 		qp_params.use_local_mac = true;
3687 		memcpy(qp_params.local_mac_addr, dev->ha->primary_mac, ETH_ALEN);
3688 
3689 		QL_DPRINT12(ha, "dgid=0x%x:0x%x:0x%x:0x%x\n",
3690 		       qp_params.dgid.dwords[0], qp_params.dgid.dwords[1],
3691 		       qp_params.dgid.dwords[2], qp_params.dgid.dwords[3]);
3692 		QL_DPRINT12(ha, "sgid=0x%x:0x%x:0x%x:0x%x\n",
3693 		       qp_params.sgid.dwords[0], qp_params.sgid.dwords[1],
3694 		       qp_params.sgid.dwords[2], qp_params.sgid.dwords[3]);
3695 		QL_DPRINT12(ha,
3696 			"remote_mac=[0x%x:0x%x:0x%x:0x%x:0x%x:0x%x]\n",
3697 			qp_params.remote_mac_addr[0],
3698 			qp_params.remote_mac_addr[1],
3699 			qp_params.remote_mac_addr[2],
3700 			qp_params.remote_mac_addr[3],
3701 			qp_params.remote_mac_addr[4],
3702 			qp_params.remote_mac_addr[5]);
3703 
3704 		qp_params.mtu = qp->mtu;
3705 	}
3706 
3707 	if (qp_params.mtu == 0) {
3708 		/* stay with current MTU */
3709 		if (qp->mtu) {
3710 			qp_params.mtu = qp->mtu;
3711 		} else {
3712 			qp_params.mtu = ib_mtu_enum_to_int(
3713 						iboe_get_mtu(dev->ha->ifp->if_mtu));
3714 		}
3715 	}
3716 
3717 	if (attr_mask & IB_QP_TIMEOUT) {
3718 		SET_FIELD(qp_params.modify_flags, \
3719 			ECORE_ROCE_MODIFY_QP_VALID_ACK_TIMEOUT, 1);
3720 
3721 		qp_params.ack_timeout = attr->timeout;
3722 		if (attr->timeout) {
3723 			u32 temp;
3724 
3725 			/* 12.7.34 LOCAL ACK TIMEOUT
3726 			 * Value representing the transport (ACK) timeout for
3727 			 * use by the remote, expressed as (4.096 μS*2Local ACK
3728 			 * Timeout)
3729 			 */
3730 			/* We use 1UL since the temporal value may be  overflow
3731 			 * 32 bits
3732 			 */
3733 			temp = 4096 * (1UL << attr->timeout) / 1000 / 1000;
3734 			qp_params.ack_timeout = temp; /* FW requires [msec] */
3735 		}
3736 		else
3737 			qp_params.ack_timeout = 0; /* infinite */
3738 	}
3739 	if (attr_mask & IB_QP_RETRY_CNT) {
3740 		SET_FIELD(qp_params.modify_flags,\
3741 			 ECORE_ROCE_MODIFY_QP_VALID_RETRY_CNT, 1);
3742 		qp_params.retry_cnt = attr->retry_cnt;
3743 	}
3744 
3745 	if (attr_mask & IB_QP_RNR_RETRY) {
3746 		SET_FIELD(qp_params.modify_flags,
3747 			  ECORE_ROCE_MODIFY_QP_VALID_RNR_RETRY_CNT,
3748 			  1);
3749 		qp_params.rnr_retry_cnt = attr->rnr_retry;
3750 	}
3751 
3752 	if (attr_mask & IB_QP_RQ_PSN) {
3753 		SET_FIELD(qp_params.modify_flags,
3754 			  ECORE_ROCE_MODIFY_QP_VALID_RQ_PSN,
3755 			  1);
3756 		qp_params.rq_psn = attr->rq_psn;
3757 		qp->rq_psn = attr->rq_psn;
3758 	}
3759 
3760 	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
3761 		if (attr->max_rd_atomic > qattr->max_qp_req_rd_atomic_resc) {
3762 			rc = -EINVAL;
3763 			QL_DPRINT12(ha,
3764 				"unsupported  max_rd_atomic=%d, supported=%d\n",
3765 				attr->max_rd_atomic,
3766 				qattr->max_qp_req_rd_atomic_resc);
3767 			goto err;
3768 		}
3769 
3770 		SET_FIELD(qp_params.modify_flags,
3771 			  ECORE_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_REQ,
3772 			  1);
3773 		qp_params.max_rd_atomic_req = attr->max_rd_atomic;
3774 	}
3775 
3776 	if (attr_mask & IB_QP_MIN_RNR_TIMER) {
3777 		SET_FIELD(qp_params.modify_flags,
3778 			  ECORE_ROCE_MODIFY_QP_VALID_MIN_RNR_NAK_TIMER,
3779 			  1);
3780 		qp_params.min_rnr_nak_timer = attr->min_rnr_timer;
3781 	}
3782 
3783 	if (attr_mask & IB_QP_SQ_PSN) {
3784 		SET_FIELD(qp_params.modify_flags,
3785 			  ECORE_ROCE_MODIFY_QP_VALID_SQ_PSN,
3786 			  1);
3787 		qp_params.sq_psn = attr->sq_psn;
3788 		qp->sq_psn = attr->sq_psn;
3789 	}
3790 
3791 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
3792 		if (attr->max_dest_rd_atomic >
3793 		    qattr->max_qp_resp_rd_atomic_resc) {
3794 			QL_DPRINT12(ha,
3795 				"unsupported max_dest_rd_atomic=%d, "
3796 				"supported=%d\n",
3797 				attr->max_dest_rd_atomic,
3798 				qattr->max_qp_resp_rd_atomic_resc);
3799 
3800 			rc = -EINVAL;
3801 			goto err;
3802 		}
3803 
3804 		SET_FIELD(qp_params.modify_flags,
3805 			  ECORE_RDMA_MODIFY_QP_VALID_MAX_RD_ATOMIC_RESP,
3806 			  1);
3807 		qp_params.max_rd_atomic_resp = attr->max_dest_rd_atomic;
3808 	}
3809 
3810  	if (attr_mask & IB_QP_DEST_QPN) {
3811 		SET_FIELD(qp_params.modify_flags,
3812 			  ECORE_ROCE_MODIFY_QP_VALID_DEST_QP,
3813 			  1);
3814 
3815 		qp_params.dest_qp = attr->dest_qp_num;
3816 		qp->dest_qp_num = attr->dest_qp_num;
3817 	}
3818 
3819 	/*
3820 	 * Update the QP state before the actual ramrod to prevent a race with
3821 	 * fast path. Modifying the QP state to error will cause the device to
3822 	 * flush the CQEs and while polling the flushed CQEs will considered as
3823 	 * a potential issue if the QP isn't in error state.
3824 	 */
3825 	if ((attr_mask & IB_QP_STATE) && (qp->qp_type != IB_QPT_GSI) &&
3826 		(!udata) && (qp_params.new_state == ECORE_ROCE_QP_STATE_ERR))
3827 		qp->state = ECORE_ROCE_QP_STATE_ERR;
3828 
3829 	if (qp->qp_type != IB_QPT_GSI)
3830 		rc = ecore_rdma_modify_qp(dev->rdma_ctx, qp->ecore_qp, &qp_params);
3831 
3832 	if (attr_mask & IB_QP_STATE) {
3833 		if ((qp->qp_type != IB_QPT_GSI) && (!udata))
3834 			rc = qlnxr_update_qp_state(dev, qp, qp_params.new_state);
3835 		qp->state = qp_params.new_state;
3836 	}
3837 
3838 err:
3839 	QL_DPRINT12(ha, "exit\n");
3840 	return rc;
3841 }
3842 
3843 static int
3844 qlnxr_to_ib_qp_acc_flags(struct ecore_rdma_query_qp_out_params *params)
3845 {
3846 	int ib_qp_acc_flags = 0;
3847 
3848 	if (params->incoming_rdma_write_en)
3849 		ib_qp_acc_flags |= IB_ACCESS_REMOTE_WRITE;
3850 	if (params->incoming_rdma_read_en)
3851 		ib_qp_acc_flags |= IB_ACCESS_REMOTE_READ;
3852 	if (params->incoming_atomic_en)
3853 		ib_qp_acc_flags |= IB_ACCESS_REMOTE_ATOMIC;
3854 	if (true) /* FIXME -> local write ?? */
3855 		ib_qp_acc_flags |= IB_ACCESS_LOCAL_WRITE;
3856 
3857 	return ib_qp_acc_flags;
3858 }
3859 
3860 static enum ib_mtu
3861 qlnxr_mtu_int_to_enum(u16 mtu)
3862 {
3863 	enum ib_mtu ib_mtu_size;
3864 
3865 	switch (mtu) {
3866 	case 256:
3867 		ib_mtu_size = IB_MTU_256;
3868 		break;
3869 
3870 	case 512:
3871 		ib_mtu_size = IB_MTU_512;
3872 		break;
3873 
3874 	case 1024:
3875 		ib_mtu_size = IB_MTU_1024;
3876 		break;
3877 
3878 	case 2048:
3879 		ib_mtu_size = IB_MTU_2048;
3880 		break;
3881 
3882 	case 4096:
3883 		ib_mtu_size = IB_MTU_4096;
3884 		break;
3885 
3886 	default:
3887 		ib_mtu_size = IB_MTU_1024;
3888 		break;
3889 	}
3890 	return (ib_mtu_size);
3891 }
3892 
3893 int
3894 qlnxr_query_qp(struct ib_qp *ibqp,
3895 	struct ib_qp_attr *qp_attr,
3896 	int attr_mask,
3897 	struct ib_qp_init_attr *qp_init_attr)
3898 {
3899 	int rc = 0;
3900 	struct ecore_rdma_query_qp_out_params params;
3901 	struct qlnxr_qp *qp = get_qlnxr_qp(ibqp);
3902 	struct qlnxr_dev *dev = qp->dev;
3903 	qlnx_host_t	*ha;
3904 
3905 	ha = dev->ha;
3906 
3907 	QL_DPRINT12(ha, "enter\n");
3908 
3909 	memset(&params, 0, sizeof(params));
3910 
3911 	rc = ecore_rdma_query_qp(dev->rdma_ctx, qp->ecore_qp, &params);
3912 	if (rc)
3913 		goto err;
3914 
3915 	memset(qp_attr, 0, sizeof(*qp_attr));
3916 	memset(qp_init_attr, 0, sizeof(*qp_init_attr));
3917 
3918 	qp_attr->qp_state = qlnxr_get_ibqp_state(params.state);
3919 	qp_attr->cur_qp_state = qlnxr_get_ibqp_state(params.state);
3920 
3921 	/* In some cases in iWARP qelr will ask for the state only */
3922 	if (QLNX_IS_IWARP(dev) && (attr_mask == IB_QP_STATE)) {
3923 		QL_DPRINT11(ha, "only state requested\n");
3924 		return 0;
3925 	}
3926 
3927 	qp_attr->path_mtu = qlnxr_mtu_int_to_enum(params.mtu);
3928 	qp_attr->path_mig_state = IB_MIG_MIGRATED;
3929 	qp_attr->rq_psn = params.rq_psn;
3930 	qp_attr->sq_psn = params.sq_psn;
3931 	qp_attr->dest_qp_num = params.dest_qp;
3932 
3933 	qp_attr->qp_access_flags = qlnxr_to_ib_qp_acc_flags(&params);
3934 
3935 	QL_DPRINT12(ha, "qp_state = 0x%x cur_qp_state = 0x%x "
3936 		"path_mtu = %d qp_access_flags = 0x%x\n",
3937 		qp_attr->qp_state, qp_attr->cur_qp_state, qp_attr->path_mtu,
3938 		qp_attr->qp_access_flags);
3939 
3940 	qp_attr->cap.max_send_wr = qp->sq.max_wr;
3941 	qp_attr->cap.max_recv_wr = qp->rq.max_wr;
3942 	qp_attr->cap.max_send_sge = qp->sq.max_sges;
3943 	qp_attr->cap.max_recv_sge = qp->rq.max_sges;
3944 	qp_attr->cap.max_inline_data = qp->max_inline_data;
3945 	qp_init_attr->cap = qp_attr->cap;
3946 
3947 	memcpy(&qp_attr->ah_attr.grh.dgid.raw[0], &params.dgid.bytes[0],
3948 	       sizeof(qp_attr->ah_attr.grh.dgid.raw));
3949 
3950 	qp_attr->ah_attr.grh.flow_label = params.flow_label;
3951 	qp_attr->ah_attr.grh.sgid_index = qp->sgid_idx;
3952 	qp_attr->ah_attr.grh.hop_limit = params.hop_limit_ttl;
3953 	qp_attr->ah_attr.grh.traffic_class = params.traffic_class_tos;
3954 
3955 	qp_attr->ah_attr.ah_flags = IB_AH_GRH;
3956 	qp_attr->ah_attr.port_num = 1; /* FIXME -> check this */
3957 	qp_attr->ah_attr.sl = 0;/* FIXME -> check this */
3958 	qp_attr->timeout = params.timeout;
3959 	qp_attr->rnr_retry = params.rnr_retry;
3960 	qp_attr->retry_cnt = params.retry_cnt;
3961 	qp_attr->min_rnr_timer = params.min_rnr_nak_timer;
3962 	qp_attr->pkey_index = params.pkey_index;
3963 	qp_attr->port_num = 1; /* FIXME -> check this */
3964 	qp_attr->ah_attr.src_path_bits = 0;
3965 	qp_attr->ah_attr.static_rate = 0;
3966 	qp_attr->alt_pkey_index = 0;
3967 	qp_attr->alt_port_num = 0;
3968 	qp_attr->alt_timeout = 0;
3969 	memset(&qp_attr->alt_ah_attr, 0, sizeof(qp_attr->alt_ah_attr));
3970 
3971 	qp_attr->sq_draining = (params.state == ECORE_ROCE_QP_STATE_SQD) ? 1 : 0;
3972 	qp_attr->max_dest_rd_atomic = params.max_dest_rd_atomic;
3973 	qp_attr->max_rd_atomic = params.max_rd_atomic;
3974 	qp_attr->en_sqd_async_notify = (params.sqd_async)? 1 : 0;
3975 
3976 	QL_DPRINT12(ha, "max_inline_data=%d\n",
3977 		qp_attr->cap.max_inline_data);
3978 
3979 err:
3980 	QL_DPRINT12(ha, "exit\n");
3981 	return rc;
3982 }
3983 
3984 
3985 static void
3986 qlnxr_cleanup_user(struct qlnxr_dev *dev, struct qlnxr_qp *qp)
3987 {
3988 	qlnx_host_t	*ha;
3989 
3990 	ha = dev->ha;
3991 
3992 	QL_DPRINT12(ha, "enter\n");
3993 
3994 	if (qp->usq.umem)
3995 		ib_umem_release(qp->usq.umem);
3996 
3997 	qp->usq.umem = NULL;
3998 
3999 	if (qp->urq.umem)
4000 		ib_umem_release(qp->urq.umem);
4001 
4002 	qp->urq.umem = NULL;
4003 
4004 	QL_DPRINT12(ha, "exit\n");
4005 	return;
4006 }
4007 
4008 static void
4009 qlnxr_cleanup_kernel(struct qlnxr_dev *dev, struct qlnxr_qp *qp)
4010 {
4011 	qlnx_host_t	*ha;
4012 
4013 	ha = dev->ha;
4014 
4015 	QL_DPRINT12(ha, "enter\n");
4016 
4017 	if (qlnxr_qp_has_sq(qp)) {
4018 		QL_DPRINT12(ha, "freeing SQ\n");
4019 		ha->qlnxr_debug = 1;
4020 //		ecore_chain_free(dev->cdev, &qp->sq.pbl);
4021 		ha->qlnxr_debug = 0;
4022 		kfree(qp->wqe_wr_id);
4023 	}
4024 
4025 	if (qlnxr_qp_has_rq(qp)) {
4026 		QL_DPRINT12(ha, "freeing RQ\n");
4027 		ha->qlnxr_debug = 1;
4028 	//	ecore_chain_free(dev->cdev, &qp->rq.pbl);
4029 		ha->qlnxr_debug = 0;
4030 		kfree(qp->rqe_wr_id);
4031 	}
4032 
4033 	QL_DPRINT12(ha, "exit\n");
4034 	return;
4035 }
4036 
4037 int
4038 qlnxr_free_qp_resources(struct qlnxr_dev *dev,
4039 	struct qlnxr_qp *qp)
4040 {
4041 	int		rc = 0;
4042 	qlnx_host_t	*ha;
4043 	struct ecore_rdma_destroy_qp_out_params d_out_params;
4044 
4045 	ha = dev->ha;
4046 
4047 	QL_DPRINT12(ha, "enter\n");
4048 
4049 #if 0
4050 	if (qp->qp_type != IB_QPT_GSI) {
4051 		rc = ecore_rdma_destroy_qp(dev->rdma_ctx, qp->ecore_qp,
4052 				&d_out_params);
4053 		if (rc)
4054 			return rc;
4055 	}
4056 
4057 	if (qp->ibqp.uobject && qp->ibqp.uobject->context)
4058 		qlnxr_cleanup_user(dev, qp);
4059 	else
4060 		qlnxr_cleanup_kernel(dev, qp);
4061 #endif
4062 
4063 	if (qp->ibqp.uobject && qp->ibqp.uobject->context)
4064 		qlnxr_cleanup_user(dev, qp);
4065 	else
4066 		qlnxr_cleanup_kernel(dev, qp);
4067 
4068 	if (qp->qp_type != IB_QPT_GSI) {
4069 		rc = ecore_rdma_destroy_qp(dev->rdma_ctx, qp->ecore_qp,
4070 				&d_out_params);
4071 		if (rc)
4072 			return rc;
4073 	}
4074 
4075 	QL_DPRINT12(ha, "exit\n");
4076 	return 0;
4077 }
4078 
4079 int
4080 qlnxr_destroy_qp(struct ib_qp *ibqp)
4081 {
4082 	struct qlnxr_qp *qp = get_qlnxr_qp(ibqp);
4083 	struct qlnxr_dev *dev = qp->dev;
4084 	int rc = 0;
4085 	struct ib_qp_attr attr;
4086 	int attr_mask = 0;
4087 	qlnx_host_t	*ha;
4088 
4089 	ha = dev->ha;
4090 
4091 	QL_DPRINT12(ha, "enter qp = %p, qp_type=%d\n", qp, qp->qp_type);
4092 
4093 	qp->destroyed = 1;
4094 
4095 	if (QLNX_IS_ROCE(dev) && (qp->state != (ECORE_ROCE_QP_STATE_RESET |
4096 				  ECORE_ROCE_QP_STATE_ERR |
4097 				  ECORE_ROCE_QP_STATE_INIT))) {
4098 
4099 		attr.qp_state = IB_QPS_ERR;
4100 		attr_mask |= IB_QP_STATE;
4101 
4102 		/* change the QP state to ERROR */
4103 		qlnxr_modify_qp(ibqp, &attr, attr_mask, NULL);
4104 	}
4105 
4106 	if (qp->qp_type == IB_QPT_GSI)
4107 		qlnxr_destroy_gsi_qp(dev);
4108 
4109 	qp->sig = ~qp->sig;
4110 
4111 	qlnxr_free_qp_resources(dev, qp);
4112 
4113 	if (atomic_dec_and_test(&qp->refcnt)) {
4114 		/* TODO: only for iWARP? */
4115 		qlnxr_idr_remove(dev, qp->qp_id);
4116 		kfree(qp);
4117 	}
4118 
4119 	QL_DPRINT12(ha, "exit\n");
4120 	return rc;
4121 }
4122 
4123 static inline int
4124 qlnxr_wq_is_full(struct qlnxr_qp_hwq_info *wq)
4125 {
4126 	return (((wq->prod + 1) % wq->max_wr) == wq->cons);
4127 }
4128 
4129 static int
4130 sge_data_len(struct ib_sge *sg_list, int num_sge)
4131 {
4132 	int i, len = 0;
4133 	for (i = 0; i < num_sge; i++)
4134 		len += sg_list[i].length;
4135 	return len;
4136 }
4137 
4138 static void
4139 swap_wqe_data64(u64 *p)
4140 {
4141 	int i;
4142 
4143 	for (i = 0; i < QLNXR_SQE_ELEMENT_SIZE / sizeof(u64); i++, p++)
4144 		*p = cpu_to_be64(cpu_to_le64(*p));
4145 }
4146 
4147 
4148 static u32
4149 qlnxr_prepare_sq_inline_data(struct qlnxr_dev *dev,
4150 	struct qlnxr_qp		*qp,
4151 	u8			*wqe_size,
4152 	struct ib_send_wr	*wr,
4153 	struct ib_send_wr	**bad_wr,
4154 	u8			*bits,
4155 	u8			bit)
4156 {
4157 	int i, seg_siz;
4158 	char *seg_prt, *wqe;
4159 	u32 data_size = sge_data_len(wr->sg_list, wr->num_sge);
4160 	qlnx_host_t	*ha;
4161 
4162 	ha = dev->ha;
4163 
4164 	QL_DPRINT12(ha, "enter[%d]\n", data_size);
4165 
4166 	if (data_size > ROCE_REQ_MAX_INLINE_DATA_SIZE) {
4167 		QL_DPRINT12(ha,
4168 			"Too much inline data in WR:[%d, %d]\n",
4169 			data_size, ROCE_REQ_MAX_INLINE_DATA_SIZE);
4170 		*bad_wr = wr;
4171 		return 0;
4172 	}
4173 
4174 	if (!data_size)
4175 		return data_size;
4176 
4177 	/* set the bit */
4178 	*bits |= bit;
4179 
4180 	seg_prt = wqe = NULL;
4181 	seg_siz = 0;
4182 
4183 	/* copy data inline */
4184 	for (i = 0; i < wr->num_sge; i++) {
4185 		u32 len = wr->sg_list[i].length;
4186 		void *src = (void *)(uintptr_t)wr->sg_list[i].addr;
4187 
4188 		while (len > 0) {
4189 			u32 cur;
4190 
4191 			/* new segment required */
4192 			if (!seg_siz) {
4193 				wqe = (char *)ecore_chain_produce(&qp->sq.pbl);
4194 				seg_prt = wqe;
4195 				seg_siz = sizeof(struct rdma_sq_common_wqe);
4196 				(*wqe_size)++;
4197 			}
4198 
4199 			/* calculate currently allowed length */
4200 			cur = MIN(len, seg_siz);
4201 
4202 			memcpy(seg_prt, src, cur);
4203 
4204 			/* update segment variables */
4205 			seg_prt += cur;
4206 			seg_siz -= cur;
4207 			/* update sge variables */
4208 			src += cur;
4209 			len -= cur;
4210 
4211 			/* swap fully-completed segments */
4212 			if (!seg_siz)
4213 				swap_wqe_data64((u64 *)wqe);
4214 		}
4215 	}
4216 
4217 	/* swap last not completed segment */
4218 	if (seg_siz)
4219 		swap_wqe_data64((u64 *)wqe);
4220 
4221 	QL_DPRINT12(ha, "exit\n");
4222 	return data_size;
4223 }
4224 
4225 static u32
4226 qlnxr_prepare_sq_sges(struct qlnxr_dev *dev, struct qlnxr_qp *qp,
4227 	u8 *wqe_size, struct ib_send_wr *wr)
4228 {
4229 	int i;
4230 	u32 data_size = 0;
4231 	qlnx_host_t	*ha;
4232 
4233 	ha = dev->ha;
4234 
4235 	QL_DPRINT12(ha, "enter wr->num_sge = %d \n", wr->num_sge);
4236 
4237 	for (i = 0; i < wr->num_sge; i++) {
4238 		struct rdma_sq_sge *sge = ecore_chain_produce(&qp->sq.pbl);
4239 
4240 		TYPEPTR_ADDR_SET(sge, addr, wr->sg_list[i].addr);
4241 		sge->l_key = cpu_to_le32(wr->sg_list[i].lkey);
4242 		sge->length = cpu_to_le32(wr->sg_list[i].length);
4243 		data_size += wr->sg_list[i].length;
4244 	}
4245 
4246 	if (wqe_size)
4247 		*wqe_size += wr->num_sge;
4248 
4249 	QL_DPRINT12(ha, "exit data_size = %d\n", data_size);
4250 	return data_size;
4251 }
4252 
4253 static u32
4254 qlnxr_prepare_sq_rdma_data(struct qlnxr_dev *dev,
4255 	struct qlnxr_qp *qp,
4256 	struct rdma_sq_rdma_wqe_1st *rwqe,
4257 	struct rdma_sq_rdma_wqe_2nd *rwqe2,
4258 	struct ib_send_wr *wr,
4259 	struct ib_send_wr **bad_wr)
4260 {
4261 	qlnx_host_t	*ha;
4262 	u32             ret = 0;
4263 
4264 	ha = dev->ha;
4265 
4266 	QL_DPRINT12(ha, "enter\n");
4267 
4268 	rwqe2->r_key = cpu_to_le32(rdma_wr(wr)->rkey);
4269 	TYPEPTR_ADDR_SET(rwqe2, remote_va, rdma_wr(wr)->remote_addr);
4270 
4271 	if (wr->send_flags & IB_SEND_INLINE) {
4272 		u8 flags = 0;
4273 		SET_FIELD2(flags, RDMA_SQ_RDMA_WQE_1ST_INLINE_FLG, 1);
4274 		return qlnxr_prepare_sq_inline_data(dev, qp, &rwqe->wqe_size,
4275 				wr, bad_wr, &rwqe->flags, flags);
4276 	}
4277 
4278 	ret = qlnxr_prepare_sq_sges(dev, qp, &rwqe->wqe_size, wr);
4279 
4280 	QL_DPRINT12(ha, "exit ret = 0x%x\n", ret);
4281 
4282 	return (ret);
4283 }
4284 
4285 static u32
4286 qlnxr_prepare_sq_send_data(struct qlnxr_dev *dev,
4287 	struct qlnxr_qp *qp,
4288 	struct rdma_sq_send_wqe *swqe,
4289 	struct rdma_sq_send_wqe *swqe2,
4290 	struct ib_send_wr *wr,
4291 	struct ib_send_wr **bad_wr)
4292 {
4293 	qlnx_host_t	*ha;
4294 	u32             ret = 0;
4295 
4296 	ha = dev->ha;
4297 
4298 	QL_DPRINT12(ha, "enter\n");
4299 
4300 	memset(swqe2, 0, sizeof(*swqe2));
4301 
4302 	if (wr->send_flags & IB_SEND_INLINE) {
4303 		u8 flags = 0;
4304 		SET_FIELD2(flags, RDMA_SQ_SEND_WQE_INLINE_FLG, 1);
4305 		return qlnxr_prepare_sq_inline_data(dev, qp, &swqe->wqe_size,
4306 				wr, bad_wr, &swqe->flags, flags);
4307 	}
4308 
4309 	ret = qlnxr_prepare_sq_sges(dev, qp, &swqe->wqe_size, wr);
4310 
4311 	QL_DPRINT12(ha, "exit ret = 0x%x\n", ret);
4312 
4313 	return (ret);
4314 }
4315 
4316 static void
4317 qlnx_handle_completed_mrs(struct qlnxr_dev *dev, struct mr_info *info)
4318 {
4319 	qlnx_host_t	*ha;
4320 
4321 	ha = dev->ha;
4322 
4323 	int work = info->completed - info->completed_handled - 1;
4324 
4325 	QL_DPRINT12(ha, "enter [%d]\n", work);
4326 
4327 	while (work-- > 0 && !list_empty(&info->inuse_pbl_list)) {
4328 		struct qlnxr_pbl *pbl;
4329 
4330 		/* Free all the page list that are possible to be freed
4331 		 * (all the ones that were invalidated), under the assumption
4332 		 * that if an FMR was completed successfully that means that
4333 		 * if there was an invalidate operation before it also ended
4334 		 */
4335 		pbl = list_first_entry(&info->inuse_pbl_list,
4336 				       struct qlnxr_pbl,
4337 				       list_entry);
4338 		list_del(&pbl->list_entry);
4339 		list_add_tail(&pbl->list_entry, &info->free_pbl_list);
4340 		info->completed_handled++;
4341 	}
4342 
4343 	QL_DPRINT12(ha, "exit\n");
4344 	return;
4345 }
4346 
4347 #if __FreeBSD_version >= 1102000
4348 
4349 static int qlnxr_prepare_reg(struct qlnxr_qp *qp,
4350 		struct rdma_sq_fmr_wqe_1st *fwqe1,
4351 		struct ib_reg_wr *wr)
4352 {
4353 	struct qlnxr_mr *mr = get_qlnxr_mr(wr->mr);
4354 	struct rdma_sq_fmr_wqe_2nd *fwqe2;
4355 
4356 	fwqe2 = (struct rdma_sq_fmr_wqe_2nd *)ecore_chain_produce(&qp->sq.pbl);
4357 	fwqe1->addr.hi = upper_32_bits(mr->ibmr.iova);
4358 	fwqe1->addr.lo = lower_32_bits(mr->ibmr.iova);
4359 	fwqe1->l_key = wr->key;
4360 
4361 	fwqe2->access_ctrl = 0;
4362 
4363 	SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_REMOTE_READ,
4364 		!!(wr->access & IB_ACCESS_REMOTE_READ));
4365 	SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_REMOTE_WRITE,
4366 		!!(wr->access & IB_ACCESS_REMOTE_WRITE));
4367 	SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_ENABLE_ATOMIC,
4368 		!!(wr->access & IB_ACCESS_REMOTE_ATOMIC));
4369 	SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_LOCAL_READ, 1);
4370 	SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_2ND_LOCAL_WRITE,
4371 		!!(wr->access & IB_ACCESS_LOCAL_WRITE));
4372 	fwqe2->fmr_ctrl = 0;
4373 
4374 	SET_FIELD2(fwqe2->fmr_ctrl, RDMA_SQ_FMR_WQE_2ND_PAGE_SIZE_LOG,
4375 		ilog2(mr->ibmr.page_size) - 12);
4376 
4377 	fwqe2->length_hi = 0; /* TODO - figure out why length is only 32bit.. */
4378 	fwqe2->length_lo = mr->ibmr.length;
4379 	fwqe2->pbl_addr.hi = upper_32_bits(mr->info.pbl_table->pa);
4380 	fwqe2->pbl_addr.lo = lower_32_bits(mr->info.pbl_table->pa);
4381 
4382 	qp->wqe_wr_id[qp->sq.prod].mr = mr;
4383 
4384 	return 0;
4385 }
4386 
4387 #else
4388 
4389 static void
4390 build_frmr_pbes(struct qlnxr_dev *dev, struct ib_send_wr *wr,
4391 	struct mr_info *info)
4392 {
4393 	int i;
4394 	u64 buf_addr = 0;
4395 	int num_pbes, total_num_pbes = 0;
4396 	struct regpair *pbe;
4397 	struct qlnxr_pbl *pbl_tbl = info->pbl_table;
4398 	struct qlnxr_pbl_info *pbl_info = &info->pbl_info;
4399 	qlnx_host_t	*ha;
4400 
4401 	ha = dev->ha;
4402 
4403 	QL_DPRINT12(ha, "enter\n");
4404 
4405 	pbe = (struct regpair *)pbl_tbl->va;
4406 	num_pbes = 0;
4407 
4408 	for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
4409 		buf_addr = wr->wr.fast_reg.page_list->page_list[i];
4410 		pbe->lo = cpu_to_le32((u32)buf_addr);
4411 		pbe->hi = cpu_to_le32((u32)upper_32_bits(buf_addr));
4412 
4413 		num_pbes += 1;
4414 		pbe++;
4415 		total_num_pbes++;
4416 
4417 		if (total_num_pbes == pbl_info->num_pbes)
4418 			return;
4419 
4420 		/* if the given pbl is full storing the pbes,
4421 		 * move to next pbl.
4422 		 */
4423 		if (num_pbes ==
4424 		    (pbl_info->pbl_size / sizeof(u64))) {
4425 			pbl_tbl++;
4426 			pbe = (struct regpair *)pbl_tbl->va;
4427 			num_pbes = 0;
4428 		}
4429 	}
4430 	QL_DPRINT12(ha, "exit\n");
4431 
4432 	return;
4433 }
4434 
4435 static int
4436 qlnxr_prepare_safe_pbl(struct qlnxr_dev *dev, struct mr_info *info)
4437 {
4438 	int rc = 0;
4439 	qlnx_host_t	*ha;
4440 
4441 	ha = dev->ha;
4442 
4443 	QL_DPRINT12(ha, "enter\n");
4444 
4445 	if (info->completed == 0) {
4446 		//DP_VERBOSE(dev, QLNXR_MSG_MR, "First FMR\n");
4447 		/* first fmr */
4448 		return 0;
4449 	}
4450 
4451 	qlnx_handle_completed_mrs(dev, info);
4452 
4453 	list_add_tail(&info->pbl_table->list_entry, &info->inuse_pbl_list);
4454 
4455 	if (list_empty(&info->free_pbl_list)) {
4456 		info->pbl_table = qlnxr_alloc_pbl_tbl(dev, &info->pbl_info,
4457 							  GFP_ATOMIC);
4458 	} else {
4459 		info->pbl_table = list_first_entry(&info->free_pbl_list,
4460 					struct qlnxr_pbl,
4461 					list_entry);
4462 		list_del(&info->pbl_table->list_entry);
4463 	}
4464 
4465 	if (!info->pbl_table)
4466 		rc = -ENOMEM;
4467 
4468 	QL_DPRINT12(ha, "exit\n");
4469 	return rc;
4470 }
4471 
4472 static inline int
4473 qlnxr_prepare_fmr(struct qlnxr_qp *qp,
4474 	struct rdma_sq_fmr_wqe_1st *fwqe1,
4475 	struct ib_send_wr *wr)
4476 {
4477 	struct qlnxr_dev *dev = qp->dev;
4478 	u64 fbo;
4479 	struct qlnxr_fast_reg_page_list *frmr_list =
4480 		get_qlnxr_frmr_list(wr->wr.fast_reg.page_list);
4481 	struct rdma_sq_fmr_wqe *fwqe2 =
4482 		(struct rdma_sq_fmr_wqe *)ecore_chain_produce(&qp->sq.pbl);
4483 	int rc = 0;
4484 	qlnx_host_t	*ha;
4485 
4486 	ha = dev->ha;
4487 
4488 	QL_DPRINT12(ha, "enter\n");
4489 
4490 	if (wr->wr.fast_reg.page_list_len == 0)
4491 		BUG();
4492 
4493 	rc = qlnxr_prepare_safe_pbl(dev, &frmr_list->info);
4494 	if (rc)
4495 		return rc;
4496 
4497 	fwqe1->addr.hi = upper_32_bits(wr->wr.fast_reg.iova_start);
4498 	fwqe1->addr.lo = lower_32_bits(wr->wr.fast_reg.iova_start);
4499 	fwqe1->l_key = wr->wr.fast_reg.rkey;
4500 
4501 	SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_REMOTE_READ,
4502 		   !!(wr->wr.fast_reg.access_flags & IB_ACCESS_REMOTE_READ));
4503 	SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_REMOTE_WRITE,
4504 		   !!(wr->wr.fast_reg.access_flags & IB_ACCESS_REMOTE_WRITE));
4505 	SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_ENABLE_ATOMIC,
4506 		   !!(wr->wr.fast_reg.access_flags & IB_ACCESS_REMOTE_ATOMIC));
4507 	SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_LOCAL_READ, 1);
4508 	SET_FIELD2(fwqe2->access_ctrl, RDMA_SQ_FMR_WQE_LOCAL_WRITE,
4509 		   !!(wr->wr.fast_reg.access_flags & IB_ACCESS_LOCAL_WRITE));
4510 
4511 	fwqe2->fmr_ctrl = 0;
4512 
4513 	SET_FIELD2(fwqe2->fmr_ctrl, RDMA_SQ_FMR_WQE_2ND_PAGE_SIZE_LOG,
4514 		   ilog2(1 << wr->wr.fast_reg.page_shift) - 12);
4515 	SET_FIELD2(fwqe2->fmr_ctrl, RDMA_SQ_FMR_WQE_2ND_ZERO_BASED, 0);
4516 
4517 	fwqe2->length_hi = 0; /* Todo - figure this out... why length is only 32bit.. */
4518 	fwqe2->length_lo = wr->wr.fast_reg.length;
4519 	fwqe2->pbl_addr.hi = upper_32_bits(frmr_list->info.pbl_table->pa);
4520 	fwqe2->pbl_addr.lo = lower_32_bits(frmr_list->info.pbl_table->pa);
4521 
4522 	/* produce another wqe for fwqe3 */
4523 	ecore_chain_produce(&qp->sq.pbl);
4524 
4525 	fbo = wr->wr.fast_reg.iova_start -
4526 	    (wr->wr.fast_reg.page_list->page_list[0] & PAGE_MASK);
4527 
4528 	QL_DPRINT12(ha, "wr.fast_reg.iova_start = %p rkey=%x addr=%x:%x"
4529 		" length = %x pbl_addr %x:%x\n",
4530 		wr->wr.fast_reg.iova_start, wr->wr.fast_reg.rkey,
4531 		fwqe1->addr.hi, fwqe1->addr.lo, fwqe2->length_lo,
4532 		fwqe2->pbl_addr.hi, fwqe2->pbl_addr.lo);
4533 
4534 	build_frmr_pbes(dev, wr, &frmr_list->info);
4535 
4536 	qp->wqe_wr_id[qp->sq.prod].frmr = frmr_list;
4537 
4538 	QL_DPRINT12(ha, "exit\n");
4539 	return 0;
4540 }
4541 
4542 #endif /* #if __FreeBSD_version >= 1102000 */
4543 
4544 static enum ib_wc_opcode
4545 qlnxr_ib_to_wc_opcode(enum ib_wr_opcode opcode)
4546 {
4547 	switch (opcode) {
4548 	case IB_WR_RDMA_WRITE:
4549 	case IB_WR_RDMA_WRITE_WITH_IMM:
4550 		return IB_WC_RDMA_WRITE;
4551 	case IB_WR_SEND_WITH_IMM:
4552 	case IB_WR_SEND:
4553 	case IB_WR_SEND_WITH_INV:
4554 		return IB_WC_SEND;
4555 	case IB_WR_RDMA_READ:
4556 		return IB_WC_RDMA_READ;
4557 	case IB_WR_ATOMIC_CMP_AND_SWP:
4558 		return IB_WC_COMP_SWAP;
4559 	case IB_WR_ATOMIC_FETCH_AND_ADD:
4560 		return IB_WC_FETCH_ADD;
4561 
4562 #if __FreeBSD_version >= 1102000
4563 	case IB_WR_REG_MR:
4564 		return IB_WC_REG_MR;
4565 #else
4566 	case IB_WR_FAST_REG_MR:
4567 		return IB_WC_FAST_REG_MR;
4568 #endif /* #if __FreeBSD_version >= 1102000 */
4569 
4570 	case IB_WR_LOCAL_INV:
4571 		return IB_WC_LOCAL_INV;
4572 	default:
4573 		return IB_WC_SEND;
4574 	}
4575 }
4576 static inline bool
4577 qlnxr_can_post_send(struct qlnxr_qp *qp, struct ib_send_wr *wr)
4578 {
4579 	int wq_is_full, err_wr, pbl_is_full;
4580 	struct qlnxr_dev *dev = qp->dev;
4581 	qlnx_host_t	*ha;
4582 
4583 	ha = dev->ha;
4584 
4585 	QL_DPRINT12(ha, "enter[qp, wr] = [%p,%p]\n", qp, wr);
4586 
4587 	/* prevent SQ overflow and/or processing of a bad WR */
4588 	err_wr = wr->num_sge > qp->sq.max_sges;
4589 	wq_is_full = qlnxr_wq_is_full(&qp->sq);
4590 	pbl_is_full = ecore_chain_get_elem_left_u32(&qp->sq.pbl) <
4591 		      QLNXR_MAX_SQE_ELEMENTS_PER_SQE;
4592 	if (wq_is_full || err_wr || pbl_is_full) {
4593 		if (wq_is_full &&
4594 		    !(qp->err_bitmap & QLNXR_QP_ERR_SQ_FULL)) {
4595 
4596 			qp->err_bitmap |= QLNXR_QP_ERR_SQ_FULL;
4597 
4598 			QL_DPRINT12(ha,
4599 				"error: WQ is full. Post send on QP failed"
4600 				" (this error appears only once) "
4601 				"[qp, wr, qp->err_bitmap]=[%p, %p, 0x%x]\n",
4602 				qp, wr, qp->err_bitmap);
4603 		}
4604 
4605 		if (err_wr &&
4606 		    !(qp->err_bitmap & QLNXR_QP_ERR_BAD_SR)) {
4607 
4608 			qp->err_bitmap |= QLNXR_QP_ERR_BAD_SR;
4609 
4610 			QL_DPRINT12(ha,
4611 				"error: WQ is bad. Post send on QP failed"
4612 				" (this error appears only once) "
4613 				"[qp, wr, qp->err_bitmap]=[%p, %p, 0x%x]\n",
4614 				qp, wr, qp->err_bitmap);
4615 		}
4616 
4617 		if (pbl_is_full &&
4618 		    !(qp->err_bitmap & QLNXR_QP_ERR_SQ_PBL_FULL)) {
4619 
4620 			qp->err_bitmap |= QLNXR_QP_ERR_SQ_PBL_FULL;
4621 
4622 			QL_DPRINT12(ha,
4623 				"error: WQ PBL is full. Post send on QP failed"
4624 				" (this error appears only once) "
4625 				"[qp, wr, qp->err_bitmap]=[%p, %p, 0x%x]\n",
4626 				qp, wr, qp->err_bitmap);
4627 		}
4628 		return false;
4629 	}
4630 	QL_DPRINT12(ha, "exit[qp, wr] = [%p,%p]\n", qp, wr);
4631 	return true;
4632 }
4633 
4634 int
4635 qlnxr_post_send(struct ib_qp *ibqp,
4636 	struct ib_send_wr *wr,
4637 	struct ib_send_wr **bad_wr)
4638 {
4639 	struct qlnxr_dev	*dev = get_qlnxr_dev(ibqp->device);
4640 	struct qlnxr_qp		*qp = get_qlnxr_qp(ibqp);
4641 	unsigned long 		flags;
4642 	int 			status = 0, rc = 0;
4643 	bool			comp;
4644 	qlnx_host_t		*ha;
4645 	uint32_t		reg_addr;
4646 
4647 	*bad_wr = NULL;
4648 	ha = dev->ha;
4649 
4650 	QL_DPRINT12(ha, "exit[ibqp, wr, bad_wr] = [%p, %p, %p]\n",
4651 		ibqp, wr, bad_wr);
4652 
4653 	if (!(ha->ifp->if_drv_flags & IFF_DRV_RUNNING))
4654 		return -EINVAL;
4655 
4656 	if (qp->qp_type == IB_QPT_GSI)
4657 		return qlnxr_gsi_post_send(ibqp, wr, bad_wr);
4658 
4659 	spin_lock_irqsave(&qp->q_lock, flags);
4660 
4661 	if (QLNX_IS_ROCE(dev) && (qp->state != ECORE_ROCE_QP_STATE_RTS) &&
4662 	    (qp->state != ECORE_ROCE_QP_STATE_ERR) &&
4663 	    (qp->state != ECORE_ROCE_QP_STATE_SQD)) {
4664 		spin_unlock_irqrestore(&qp->q_lock, flags);
4665 		*bad_wr = wr;
4666 		QL_DPRINT11(ha, "QP in wrong state! QP icid=0x%x state %d\n",
4667 			qp->icid, qp->state);
4668 		return -EINVAL;
4669 	}
4670 
4671 	if (!wr) {
4672 		QL_DPRINT11(ha, "Got an empty post send???\n");
4673 	}
4674 
4675 	while (wr) {
4676 		struct rdma_sq_common_wqe	*wqe;
4677 		struct rdma_sq_send_wqe		*swqe;
4678 		struct rdma_sq_send_wqe		*swqe2;
4679 		struct rdma_sq_rdma_wqe_1st	*rwqe;
4680 		struct rdma_sq_rdma_wqe_2nd	*rwqe2;
4681 		struct rdma_sq_local_inv_wqe	*iwqe;
4682 		struct rdma_sq_atomic_wqe	*awqe1;
4683 		struct rdma_sq_atomic_wqe	*awqe2;
4684 		struct rdma_sq_atomic_wqe	*awqe3;
4685 		struct rdma_sq_fmr_wqe_1st	*fwqe1;
4686 
4687 		if (!qlnxr_can_post_send(qp, wr)) {
4688 			status = -ENOMEM;
4689 			*bad_wr = wr;
4690 			break;
4691 		}
4692 
4693 		wqe = ecore_chain_produce(&qp->sq.pbl);
4694 
4695 		qp->wqe_wr_id[qp->sq.prod].signaled =
4696 			!!(wr->send_flags & IB_SEND_SIGNALED) || qp->signaled;
4697 
4698 		/* common fields */
4699 		wqe->flags = 0;
4700 		wqe->flags |= (RDMA_SQ_SEND_WQE_COMP_FLG_MASK <<
4701 				RDMA_SQ_SEND_WQE_COMP_FLG_SHIFT);
4702 
4703 		SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_SE_FLG, \
4704 			!!(wr->send_flags & IB_SEND_SOLICITED));
4705 
4706 		comp = (!!(wr->send_flags & IB_SEND_SIGNALED)) ||
4707 				(qp->signaled);
4708 
4709 		SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_COMP_FLG, comp);
4710 		SET_FIELD2(wqe->flags, RDMA_SQ_SEND_WQE_RD_FENCE_FLG,  \
4711 			!!(wr->send_flags & IB_SEND_FENCE));
4712 
4713 		wqe->prev_wqe_size = qp->prev_wqe_size;
4714 
4715 		qp->wqe_wr_id[qp->sq.prod].opcode = qlnxr_ib_to_wc_opcode(wr->opcode);
4716 
4717 
4718 		switch (wr->opcode) {
4719 
4720 		case IB_WR_SEND_WITH_IMM:
4721 
4722 			wqe->req_type = RDMA_SQ_REQ_TYPE_SEND_WITH_IMM;
4723 			swqe = (struct rdma_sq_send_wqe *)wqe;
4724 			swqe->wqe_size = 2;
4725 			swqe2 = (struct rdma_sq_send_wqe *)
4726 					ecore_chain_produce(&qp->sq.pbl);
4727 			swqe->inv_key_or_imm_data =
4728 				cpu_to_le32(wr->ex.imm_data);
4729 			swqe->length = cpu_to_le32(
4730 						qlnxr_prepare_sq_send_data(dev,
4731 							qp, swqe, swqe2, wr,
4732 							bad_wr));
4733 
4734 			qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size;
4735 			qp->prev_wqe_size = swqe->wqe_size;
4736 			qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length;
4737 
4738 			QL_DPRINT12(ha, "SEND w/ IMM length = %d imm data=%x\n",
4739 				swqe->length, wr->ex.imm_data);
4740 
4741 			break;
4742 
4743 		case IB_WR_SEND:
4744 
4745 			wqe->req_type = RDMA_SQ_REQ_TYPE_SEND;
4746 			swqe = (struct rdma_sq_send_wqe *)wqe;
4747 
4748 			swqe->wqe_size = 2;
4749 			swqe2 = (struct rdma_sq_send_wqe *)
4750 					ecore_chain_produce(&qp->sq.pbl);
4751 			swqe->length = cpu_to_le32(
4752 						qlnxr_prepare_sq_send_data(dev,
4753 							qp, swqe, swqe2, wr,
4754 							bad_wr));
4755 			qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size;
4756 			qp->prev_wqe_size = swqe->wqe_size;
4757 			qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length;
4758 
4759 			QL_DPRINT12(ha, "SEND w/o IMM length = %d\n",
4760 				swqe->length);
4761 
4762 			break;
4763 
4764 		case IB_WR_SEND_WITH_INV:
4765 
4766 			wqe->req_type = RDMA_SQ_REQ_TYPE_SEND_WITH_INVALIDATE;
4767 			swqe = (struct rdma_sq_send_wqe *)wqe;
4768 			swqe2 = (struct rdma_sq_send_wqe *)
4769 					ecore_chain_produce(&qp->sq.pbl);
4770 			swqe->wqe_size = 2;
4771 			swqe->inv_key_or_imm_data =
4772 				cpu_to_le32(wr->ex.invalidate_rkey);
4773 			swqe->length = cpu_to_le32(qlnxr_prepare_sq_send_data(dev,
4774 						qp, swqe, swqe2, wr, bad_wr));
4775 			qp->wqe_wr_id[qp->sq.prod].wqe_size = swqe->wqe_size;
4776 			qp->prev_wqe_size = swqe->wqe_size;
4777 			qp->wqe_wr_id[qp->sq.prod].bytes_len = swqe->length;
4778 
4779 			QL_DPRINT12(ha, "SEND w INVALIDATE length = %d\n",
4780 				swqe->length);
4781 			break;
4782 
4783 		case IB_WR_RDMA_WRITE_WITH_IMM:
4784 
4785 			wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_WR_WITH_IMM;
4786 			rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe;
4787 
4788 			rwqe->wqe_size = 2;
4789 			rwqe->imm_data = htonl(cpu_to_le32(wr->ex.imm_data));
4790 			rwqe2 = (struct rdma_sq_rdma_wqe_2nd *)
4791 					ecore_chain_produce(&qp->sq.pbl);
4792 			rwqe->length = cpu_to_le32(qlnxr_prepare_sq_rdma_data(dev,
4793 						qp, rwqe, rwqe2, wr, bad_wr));
4794 			qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size;
4795 			qp->prev_wqe_size = rwqe->wqe_size;
4796 			qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length;
4797 
4798 			QL_DPRINT12(ha,
4799 				"RDMA WRITE w/ IMM length = %d imm data=%x\n",
4800 				rwqe->length, rwqe->imm_data);
4801 
4802 			break;
4803 
4804 		case IB_WR_RDMA_WRITE:
4805 
4806 			wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_WR;
4807 			rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe;
4808 
4809 			rwqe->wqe_size = 2;
4810 			rwqe2 = (struct rdma_sq_rdma_wqe_2nd *)
4811 					ecore_chain_produce(&qp->sq.pbl);
4812 			rwqe->length = cpu_to_le32(qlnxr_prepare_sq_rdma_data(dev,
4813 						qp, rwqe, rwqe2, wr, bad_wr));
4814 			qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size;
4815 			qp->prev_wqe_size = rwqe->wqe_size;
4816 			qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length;
4817 
4818 			QL_DPRINT12(ha,
4819 				"RDMA WRITE w/o IMM length = %d\n",
4820 				rwqe->length);
4821 
4822 			break;
4823 
4824 		case IB_WR_RDMA_READ_WITH_INV:
4825 
4826 			QL_DPRINT12(ha,
4827 				"RDMA READ WITH INVALIDATE not supported\n");
4828 
4829 			*bad_wr = wr;
4830 			rc = -EINVAL;
4831 
4832 			break;
4833 
4834 		case IB_WR_RDMA_READ:
4835 
4836 			wqe->req_type = RDMA_SQ_REQ_TYPE_RDMA_RD;
4837 			rwqe = (struct rdma_sq_rdma_wqe_1st *)wqe;
4838 
4839 			rwqe->wqe_size = 2;
4840 			rwqe2 = (struct rdma_sq_rdma_wqe_2nd *)
4841 					ecore_chain_produce(&qp->sq.pbl);
4842 			rwqe->length = cpu_to_le32(qlnxr_prepare_sq_rdma_data(dev,
4843 						qp, rwqe, rwqe2, wr, bad_wr));
4844 
4845 			qp->wqe_wr_id[qp->sq.prod].wqe_size = rwqe->wqe_size;
4846 			qp->prev_wqe_size = rwqe->wqe_size;
4847 			qp->wqe_wr_id[qp->sq.prod].bytes_len = rwqe->length;
4848 
4849 			QL_DPRINT12(ha, "RDMA READ length = %d\n",
4850 				rwqe->length);
4851 
4852 			break;
4853 
4854 		case IB_WR_ATOMIC_CMP_AND_SWP:
4855 		case IB_WR_ATOMIC_FETCH_AND_ADD:
4856 
4857 			QL_DPRINT12(ha,
4858 				"ATOMIC operation = %s\n",
4859 				((wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) ?
4860 					"IB_WR_ATOMIC_CMP_AND_SWP" :
4861 					"IB_WR_ATOMIC_FETCH_AND_ADD"));
4862 
4863 			awqe1 = (struct rdma_sq_atomic_wqe *)wqe;
4864 			awqe1->prev_wqe_size = 4;
4865 
4866 			awqe2 = (struct rdma_sq_atomic_wqe *)
4867 					ecore_chain_produce(&qp->sq.pbl);
4868 
4869 			TYPEPTR_ADDR_SET(awqe2, remote_va, \
4870 				atomic_wr(wr)->remote_addr);
4871 
4872 			awqe2->r_key = cpu_to_le32(atomic_wr(wr)->rkey);
4873 
4874 			awqe3 = (struct rdma_sq_atomic_wqe *)
4875 					ecore_chain_produce(&qp->sq.pbl);
4876 
4877 			if (wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
4878 				wqe->req_type = RDMA_SQ_REQ_TYPE_ATOMIC_ADD;
4879 				TYPEPTR_ADDR_SET(awqe3, swap_data,
4880 						 atomic_wr(wr)->compare_add);
4881 			} else {
4882 				wqe->req_type = RDMA_SQ_REQ_TYPE_ATOMIC_CMP_AND_SWAP;
4883 				TYPEPTR_ADDR_SET(awqe3, swap_data,
4884 						 atomic_wr(wr)->swap);
4885 				TYPEPTR_ADDR_SET(awqe3, cmp_data,
4886 						 atomic_wr(wr)->compare_add);
4887 			}
4888 
4889 			qlnxr_prepare_sq_sges(dev, qp, NULL, wr);
4890 
4891 			qp->wqe_wr_id[qp->sq.prod].wqe_size = awqe1->prev_wqe_size;
4892 			qp->prev_wqe_size = awqe1->prev_wqe_size;
4893 
4894 			break;
4895 
4896 		case IB_WR_LOCAL_INV:
4897 
4898 			QL_DPRINT12(ha,
4899 				"INVALIDATE length (IB_WR_LOCAL_INV)\n");
4900 
4901 			iwqe = (struct rdma_sq_local_inv_wqe *)wqe;
4902 			iwqe->prev_wqe_size = 1;
4903 
4904 			iwqe->req_type = RDMA_SQ_REQ_TYPE_LOCAL_INVALIDATE;
4905 			iwqe->inv_l_key = wr->ex.invalidate_rkey;
4906 			qp->wqe_wr_id[qp->sq.prod].wqe_size = iwqe->prev_wqe_size;
4907 			qp->prev_wqe_size = iwqe->prev_wqe_size;
4908 
4909 			break;
4910 
4911 #if __FreeBSD_version >= 1102000
4912 
4913 		case IB_WR_REG_MR:
4914 
4915 			QL_DPRINT12(ha, "IB_WR_REG_MR\n");
4916 
4917 			wqe->req_type = RDMA_SQ_REQ_TYPE_FAST_MR;
4918 			fwqe1 = (struct rdma_sq_fmr_wqe_1st *)wqe;
4919 			fwqe1->wqe_size = 2;
4920 
4921 			rc = qlnxr_prepare_reg(qp, fwqe1, reg_wr(wr));
4922 			if (rc) {
4923 				QL_DPRINT11(ha, "IB_WR_REG_MR failed rc=%d\n", rc);
4924 				*bad_wr = wr;
4925 				break;
4926 			}
4927 
4928 			qp->wqe_wr_id[qp->sq.prod].wqe_size = fwqe1->wqe_size;
4929 			qp->prev_wqe_size = fwqe1->wqe_size;
4930 
4931 			break;
4932 #else
4933 		case IB_WR_FAST_REG_MR:
4934 
4935 			QL_DPRINT12(ha, "FAST_MR (IB_WR_FAST_REG_MR)\n");
4936 
4937 			wqe->req_type = RDMA_SQ_REQ_TYPE_FAST_MR;
4938 			fwqe1 = (struct rdma_sq_fmr_wqe_1st *)wqe;
4939 			fwqe1->prev_wqe_size = 3;
4940 
4941 			rc = qlnxr_prepare_fmr(qp, fwqe1, wr);
4942 
4943 			if (rc) {
4944 				QL_DPRINT12(ha,
4945 					"FAST_MR (IB_WR_FAST_REG_MR) failed"
4946 					" rc = %d\n", rc);
4947 				*bad_wr = wr;
4948 				break;
4949 			}
4950 
4951 			qp->wqe_wr_id[qp->sq.prod].wqe_size = fwqe1->prev_wqe_size;
4952 			qp->prev_wqe_size = fwqe1->prev_wqe_size;
4953 
4954 			break;
4955 #endif /* #if __FreeBSD_version >= 1102000 */
4956 
4957 		default:
4958 
4959 			QL_DPRINT12(ha, "Invalid Opcode 0x%x!\n", wr->opcode);
4960 
4961 			rc = -EINVAL;
4962 			*bad_wr = wr;
4963 			break;
4964 		}
4965 
4966 		if (*bad_wr) {
4967 			/*
4968 			 * restore prod to its position before this WR was processed
4969 			 */
4970 			ecore_chain_set_prod(&qp->sq.pbl,
4971 			     le16_to_cpu(qp->sq.db_data.data.value),
4972 			     wqe);
4973 			/* restore prev_wqe_size */
4974 			qp->prev_wqe_size = wqe->prev_wqe_size;
4975 			status = rc;
4976 
4977 			QL_DPRINT12(ha, "failed *bad_wr = %p\n", *bad_wr);
4978 			break; /* out of the loop */
4979 		}
4980 
4981 		qp->wqe_wr_id[qp->sq.prod].wr_id = wr->wr_id;
4982 
4983 		qlnxr_inc_sw_prod(&qp->sq);
4984 
4985 		qp->sq.db_data.data.value++;
4986 
4987 		wr = wr->next;
4988 	}
4989 
4990 	/* Trigger doorbell
4991 	 * If there was a failure in the first WR then it will be triggered in
4992 	 * vane. However this is not harmful (as long as the producer value is
4993 	 * unchanged). For performance reasons we avoid checking for this
4994 	 * redundant doorbell.
4995 	 */
4996 	wmb();
4997 	//writel(qp->sq.db_data.raw, qp->sq.db);
4998 
4999 	reg_addr = (uint32_t)((uint8_t *)qp->sq.db - (uint8_t *)ha->cdev.doorbells);
5000         bus_write_4(ha->pci_dbells, reg_addr, qp->sq.db_data.raw);
5001         bus_barrier(ha->pci_dbells,  0, 0, BUS_SPACE_BARRIER_READ);
5002 
5003 	mmiowb();
5004 
5005 	spin_unlock_irqrestore(&qp->q_lock, flags);
5006 
5007 	QL_DPRINT12(ha, "exit[ibqp, wr, bad_wr] = [%p, %p, %p]\n",
5008 		ibqp, wr, bad_wr);
5009 
5010 	return status;
5011 }
5012 
5013 static u32
5014 qlnxr_srq_elem_left(struct qlnxr_srq_hwq_info *hw_srq)
5015 {
5016 	u32 used;
5017 
5018 	/* Calculate number of elements used based on producer
5019 	 * count and consumer count and subtract it from max
5020 	 * work request supported so that we get elements left.
5021 	 */
5022 	used = hw_srq->wr_prod_cnt - hw_srq->wr_cons_cnt;
5023 
5024 	return hw_srq->max_wr - used;
5025 }
5026 
5027 
5028 int
5029 qlnxr_post_recv(struct ib_qp *ibqp,
5030 	struct ib_recv_wr *wr,
5031 	struct ib_recv_wr **bad_wr)
5032 {
5033  	struct qlnxr_qp		*qp = get_qlnxr_qp(ibqp);
5034 	struct qlnxr_dev	*dev = qp->dev;
5035 	unsigned long		flags;
5036 	int			status = 0;
5037 	qlnx_host_t		*ha;
5038 	uint32_t		reg_addr;
5039 
5040 	ha = dev->ha;
5041 
5042 	if (!(ha->ifp->if_drv_flags & IFF_DRV_RUNNING))
5043 		return -EINVAL;
5044 
5045 	QL_DPRINT12(ha, "enter\n");
5046 
5047 	if (qp->qp_type == IB_QPT_GSI) {
5048 		QL_DPRINT12(ha, "(qp->qp_type = IB_QPT_GSI)\n");
5049 		return qlnxr_gsi_post_recv(ibqp, wr, bad_wr);
5050 	}
5051 
5052 	if (qp->srq) {
5053 		QL_DPRINT11(ha, "qp->srq [%p]"
5054 			" QP is associated with SRQ, cannot post RQ buffers\n",
5055 			qp->srq);
5056 		return -EINVAL;
5057 	}
5058 
5059 	spin_lock_irqsave(&qp->q_lock, flags);
5060 
5061 	if (qp->state == ECORE_ROCE_QP_STATE_RESET) {
5062 		spin_unlock_irqrestore(&qp->q_lock, flags);
5063 		*bad_wr = wr;
5064 
5065 		QL_DPRINT11(ha, "qp->qp_type = ECORE_ROCE_QP_STATE_RESET\n");
5066 
5067 		return -EINVAL;
5068 	}
5069 
5070 	while (wr) {
5071 		int i;
5072 
5073 		if ((ecore_chain_get_elem_left_u32(&qp->rq.pbl) <
5074 			QLNXR_MAX_RQE_ELEMENTS_PER_RQE) ||
5075 			(wr->num_sge > qp->rq.max_sges)) {
5076 			status = -ENOMEM;
5077 			*bad_wr = wr;
5078 			break;
5079 		}
5080 		for (i = 0; i < wr->num_sge; i++) {
5081 			u32 flags = 0;
5082 			struct rdma_rq_sge *rqe = ecore_chain_produce(&qp->rq.pbl);
5083 
5084 			/* first one must include the number of SGE in the list */
5085 			if (!i)
5086 				SET_FIELD(flags, RDMA_RQ_SGE_NUM_SGES, wr->num_sge);
5087 
5088 			SET_FIELD(flags, RDMA_RQ_SGE_L_KEY, wr->sg_list[i].lkey);
5089 
5090 			RQ_SGE_SET(rqe, wr->sg_list[i].addr, \
5091 				wr->sg_list[i].length, flags);
5092 		}
5093 		/* Special case of no sges. FW requires between 1-4 sges...
5094 		 * in this case we need to post 1 sge with length zero. this is
5095 		 * because rdma write with immediate consumes an RQ. */
5096 		if (!wr->num_sge) {
5097 			u32 flags = 0;
5098 			struct rdma_rq_sge *rqe = ecore_chain_produce(&qp->rq.pbl);
5099 
5100 			/* first one must include the number of SGE in the list */
5101 			SET_FIELD(flags, RDMA_RQ_SGE_L_KEY, 0);
5102 			SET_FIELD(flags, RDMA_RQ_SGE_NUM_SGES, 1);
5103 
5104 			//RQ_SGE_SET(rqe, 0, 0, flags);
5105 			rqe->addr.hi = 0;
5106 			rqe->addr.lo = 0;
5107 
5108 			rqe->length = 0;
5109 			rqe->flags = cpu_to_le32(flags);
5110 
5111 			i = 1;
5112 		}
5113 
5114 		qp->rqe_wr_id[qp->rq.prod].wr_id = wr->wr_id;
5115 		qp->rqe_wr_id[qp->rq.prod].wqe_size = i;
5116 
5117 		qlnxr_inc_sw_prod(&qp->rq);
5118 
5119 		wmb();
5120 
5121 		qp->rq.db_data.data.value++;
5122 
5123 	//	writel(qp->rq.db_data.raw, qp->rq.db);
5124 		mmiowb();
5125 	//	if (QLNX_IS_IWARP(dev)) {
5126 	//		writel(qp->rq.iwarp_db2_data.raw, qp->rq.iwarp_db2);
5127 	//		mmiowb(); /* for second doorbell */
5128 	//	}
5129 
5130 		reg_addr = (uint32_t)((uint8_t *)qp->rq.db -
5131 				(uint8_t *)ha->cdev.doorbells);
5132 
5133 		bus_write_4(ha->pci_dbells, reg_addr, qp->rq.db_data.raw);
5134 		bus_barrier(ha->pci_dbells,  0, 0, BUS_SPACE_BARRIER_READ);
5135 
5136 		if (QLNX_IS_IWARP(dev)) {
5137 			reg_addr = (uint32_t)((uint8_t *)qp->rq.iwarp_db2 -
5138 						(uint8_t *)ha->cdev.doorbells);
5139 			bus_write_4(ha->pci_dbells, reg_addr, \
5140 				qp->rq.iwarp_db2_data.raw);
5141 			bus_barrier(ha->pci_dbells,  0, 0, \
5142 				BUS_SPACE_BARRIER_READ);
5143 		}
5144 
5145 		wr = wr->next;
5146 	}
5147 
5148 	spin_unlock_irqrestore(&qp->q_lock, flags);
5149 
5150 	QL_DPRINT12(ha, "exit status = 0x%x\n", status);
5151 
5152 	return status;
5153 }
5154 
5155 /* In fmr we need to increase the number of fmr completed counter for the fmr
5156  * algorithm determining whether we can free a pbl or not.
5157  * we need to perform this whether the work request was signaled or not. for
5158  * this purpose we call this function from the condition that checks if a wr
5159  * should be skipped, to make sure we don't miss it ( possibly this fmr
5160  * operation was not signalted)
5161  */
5162 static inline void
5163 qlnxr_chk_if_fmr(struct qlnxr_qp *qp)
5164 {
5165 #if __FreeBSD_version >= 1102000
5166 
5167 	if (qp->wqe_wr_id[qp->sq.cons].opcode == IB_WC_REG_MR)
5168 		qp->wqe_wr_id[qp->sq.cons].mr->info.completed++;
5169 #else
5170 	if (qp->wqe_wr_id[qp->sq.cons].opcode == IB_WC_FAST_REG_MR)
5171 		qp->wqe_wr_id[qp->sq.cons].frmr->info.completed++;
5172 
5173 #endif /* #if __FreeBSD_version >= 1102000 */
5174 }
5175 
5176 static int
5177 process_req(struct qlnxr_dev *dev,
5178 	struct qlnxr_qp *qp,
5179 	struct qlnxr_cq *cq,
5180 	int num_entries,
5181 	struct ib_wc *wc,
5182 	u16 hw_cons,
5183 	enum ib_wc_status status,
5184 	int force)
5185 {
5186 	u16		cnt = 0;
5187 	qlnx_host_t	*ha = dev->ha;
5188 
5189 	QL_DPRINT12(ha, "enter\n");
5190 
5191 	while (num_entries && qp->sq.wqe_cons != hw_cons) {
5192 		if (!qp->wqe_wr_id[qp->sq.cons].signaled && !force) {
5193 			qlnxr_chk_if_fmr(qp);
5194 			/* skip WC */
5195 			goto next_cqe;
5196 		}
5197 
5198 		/* fill WC */
5199 		wc->status = status;
5200 		wc->vendor_err = 0;
5201 		wc->wc_flags = 0;
5202 		wc->src_qp = qp->id;
5203 		wc->qp = &qp->ibqp;
5204 
5205 		// common section
5206 		wc->wr_id = qp->wqe_wr_id[qp->sq.cons].wr_id;
5207 		wc->opcode = qp->wqe_wr_id[qp->sq.cons].opcode;
5208 
5209 		switch (wc->opcode) {
5210 
5211 		case IB_WC_RDMA_WRITE:
5212 
5213 			wc->byte_len = qp->wqe_wr_id[qp->sq.cons].bytes_len;
5214 
5215 			QL_DPRINT12(ha,
5216 				"opcode = IB_WC_RDMA_WRITE bytes = %d\n",
5217 				qp->wqe_wr_id[qp->sq.cons].bytes_len);
5218 			break;
5219 
5220 		case IB_WC_COMP_SWAP:
5221 		case IB_WC_FETCH_ADD:
5222 			wc->byte_len = 8;
5223 			break;
5224 
5225 #if __FreeBSD_version >= 1102000
5226 		case IB_WC_REG_MR:
5227 			qp->wqe_wr_id[qp->sq.cons].mr->info.completed++;
5228 			break;
5229 #else
5230 		case IB_WC_FAST_REG_MR:
5231 			qp->wqe_wr_id[qp->sq.cons].frmr->info.completed++;
5232 			break;
5233 #endif /* #if __FreeBSD_version >= 1102000 */
5234 
5235 		case IB_WC_RDMA_READ:
5236 		case IB_WC_SEND:
5237 
5238 			QL_DPRINT12(ha, "opcode = 0x%x \n", wc->opcode);
5239 			break;
5240 		default:
5241 			;//DP_ERR("TBD ERROR");
5242 		}
5243 
5244 		num_entries--;
5245 		wc++;
5246 		cnt++;
5247 next_cqe:
5248 		while (qp->wqe_wr_id[qp->sq.cons].wqe_size--)
5249 			ecore_chain_consume(&qp->sq.pbl);
5250 		qlnxr_inc_sw_cons(&qp->sq);
5251 	}
5252 
5253 	QL_DPRINT12(ha, "exit cnt = 0x%x\n", cnt);
5254 	return cnt;
5255 }
5256 
5257 static int
5258 qlnxr_poll_cq_req(struct qlnxr_dev *dev,
5259 	struct qlnxr_qp *qp,
5260 	struct qlnxr_cq *cq,
5261 	int num_entries,
5262 	struct ib_wc *wc,
5263 	struct rdma_cqe_requester *req)
5264 {
5265 	int		cnt = 0;
5266 	qlnx_host_t	*ha = dev->ha;
5267 
5268 	QL_DPRINT12(ha, "enter req->status = 0x%x\n", req->status);
5269 
5270 	switch (req->status) {
5271 
5272 	case RDMA_CQE_REQ_STS_OK:
5273 
5274 		cnt = process_req(dev, qp, cq, num_entries, wc, req->sq_cons,
5275 			IB_WC_SUCCESS, 0);
5276 		break;
5277 
5278 	case RDMA_CQE_REQ_STS_WORK_REQUEST_FLUSHED_ERR:
5279 
5280 		if (qp->state != ECORE_ROCE_QP_STATE_ERR)
5281 		cnt = process_req(dev, qp, cq, num_entries, wc, req->sq_cons,
5282 				  IB_WC_WR_FLUSH_ERR, 1);
5283 		break;
5284 
5285 	default: /* other errors case */
5286 
5287 		/* process all WQE before the cosumer */
5288 		qp->state = ECORE_ROCE_QP_STATE_ERR;
5289 		cnt = process_req(dev, qp, cq, num_entries, wc,
5290 				req->sq_cons - 1, IB_WC_SUCCESS, 0);
5291 		wc += cnt;
5292 		/* if we have extra WC fill it with actual error info */
5293 
5294 		if (cnt < num_entries) {
5295 			enum ib_wc_status wc_status;
5296 
5297 			switch (req->status) {
5298 			case 	RDMA_CQE_REQ_STS_BAD_RESPONSE_ERR:
5299 				wc_status = IB_WC_BAD_RESP_ERR;
5300 				break;
5301 			case 	RDMA_CQE_REQ_STS_LOCAL_LENGTH_ERR:
5302 				wc_status = IB_WC_LOC_LEN_ERR;
5303 				break;
5304 			case    RDMA_CQE_REQ_STS_LOCAL_QP_OPERATION_ERR:
5305 				wc_status = IB_WC_LOC_QP_OP_ERR;
5306 				break;
5307 			case    RDMA_CQE_REQ_STS_LOCAL_PROTECTION_ERR:
5308 				wc_status = IB_WC_LOC_PROT_ERR;
5309 				break;
5310 			case    RDMA_CQE_REQ_STS_MEMORY_MGT_OPERATION_ERR:
5311 				wc_status = IB_WC_MW_BIND_ERR;
5312 				break;
5313 			case    RDMA_CQE_REQ_STS_REMOTE_INVALID_REQUEST_ERR:
5314 				wc_status = IB_WC_REM_INV_REQ_ERR;
5315 				break;
5316 			case    RDMA_CQE_REQ_STS_REMOTE_ACCESS_ERR:
5317 				wc_status = IB_WC_REM_ACCESS_ERR;
5318 				break;
5319 			case    RDMA_CQE_REQ_STS_REMOTE_OPERATION_ERR:
5320 				wc_status = IB_WC_REM_OP_ERR;
5321 				break;
5322 			case    RDMA_CQE_REQ_STS_RNR_NAK_RETRY_CNT_ERR:
5323 				wc_status = IB_WC_RNR_RETRY_EXC_ERR;
5324 				break;
5325 			case    RDMA_CQE_REQ_STS_TRANSPORT_RETRY_CNT_ERR:
5326 				wc_status = IB_WC_RETRY_EXC_ERR;
5327 				break;
5328 			default:
5329 				wc_status = IB_WC_GENERAL_ERR;
5330 			}
5331 
5332 			cnt += process_req(dev, qp, cq, 1, wc, req->sq_cons,
5333 					wc_status, 1 /* force use of WC */);
5334 		}
5335 	}
5336 
5337 	QL_DPRINT12(ha, "exit cnt = %d\n", cnt);
5338 	return cnt;
5339 }
5340 
5341 static void
5342 __process_resp_one(struct qlnxr_dev *dev,
5343 	struct qlnxr_qp *qp,
5344 	struct qlnxr_cq *cq,
5345 	struct ib_wc *wc,
5346 	struct rdma_cqe_responder *resp,
5347 	u64 wr_id)
5348 {
5349 	enum ib_wc_status	wc_status = IB_WC_SUCCESS;
5350 #if __FreeBSD_version < 1102000
5351 	u8			flags;
5352 #endif
5353 	qlnx_host_t		*ha = dev->ha;
5354 
5355 	QL_DPRINT12(ha, "enter qp = %p resp->status = 0x%x\n",
5356 		qp, resp->status);
5357 
5358 	wc->opcode = IB_WC_RECV;
5359 	wc->wc_flags = 0;
5360 
5361 	switch (resp->status) {
5362 
5363 	case RDMA_CQE_RESP_STS_LOCAL_ACCESS_ERR:
5364 		wc_status = IB_WC_LOC_ACCESS_ERR;
5365 		break;
5366 
5367 	case RDMA_CQE_RESP_STS_LOCAL_LENGTH_ERR:
5368 		wc_status = IB_WC_LOC_LEN_ERR;
5369 		break;
5370 
5371 	case RDMA_CQE_RESP_STS_LOCAL_QP_OPERATION_ERR:
5372 		wc_status = IB_WC_LOC_QP_OP_ERR;
5373 		break;
5374 
5375 	case RDMA_CQE_RESP_STS_LOCAL_PROTECTION_ERR:
5376 		wc_status = IB_WC_LOC_PROT_ERR;
5377 		break;
5378 
5379 	case RDMA_CQE_RESP_STS_MEMORY_MGT_OPERATION_ERR:
5380 		wc_status = IB_WC_MW_BIND_ERR;
5381 		break;
5382 
5383 	case RDMA_CQE_RESP_STS_REMOTE_INVALID_REQUEST_ERR:
5384 		wc_status = IB_WC_REM_INV_RD_REQ_ERR;
5385 		break;
5386 
5387 	case RDMA_CQE_RESP_STS_OK:
5388 
5389 #if __FreeBSD_version >= 1102000
5390 		if (resp->flags & QLNXR_RESP_IMM) {
5391 			wc->ex.imm_data =
5392 				le32_to_cpu(resp->imm_data_or_inv_r_Key);
5393 			wc->wc_flags |= IB_WC_WITH_IMM;
5394 
5395 			if (resp->flags & QLNXR_RESP_RDMA)
5396 				wc->opcode = IB_WC_RECV_RDMA_WITH_IMM;
5397 
5398 			if (resp->flags & QLNXR_RESP_INV) {
5399 				QL_DPRINT11(ha,
5400 					"Invalid flags QLNXR_RESP_INV [0x%x]"
5401 					"qp = %p qp->id = 0x%x cq = %p"
5402 					" cq->icid = 0x%x\n",
5403 					resp->flags, qp, qp->id, cq, cq->icid );
5404 			}
5405 		} else if (resp->flags & QLNXR_RESP_INV) {
5406 			wc->ex.imm_data =
5407 				le32_to_cpu(resp->imm_data_or_inv_r_Key);
5408 			wc->wc_flags |= IB_WC_WITH_INVALIDATE;
5409 
5410 			if (resp->flags & QLNXR_RESP_RDMA) {
5411 				QL_DPRINT11(ha,
5412 					"Invalid flags QLNXR_RESP_RDMA [0x%x]"
5413 					"qp = %p qp->id = 0x%x cq = %p"
5414 					" cq->icid = 0x%x\n",
5415 					resp->flags, qp, qp->id, cq, cq->icid );
5416 			}
5417 		} else if (resp->flags & QLNXR_RESP_RDMA) {
5418 			QL_DPRINT11(ha, "Invalid flags QLNXR_RESP_RDMA [0x%x]"
5419 				"qp = %p qp->id = 0x%x cq = %p cq->icid = 0x%x\n",
5420 				resp->flags, qp, qp->id, cq, cq->icid );
5421 		}
5422 #else
5423 		wc_status = IB_WC_SUCCESS;
5424 		wc->byte_len = le32_to_cpu(resp->length);
5425 
5426 		flags = resp->flags & QLNXR_RESP_RDMA_IMM;
5427 
5428 		switch (flags) {
5429 
5430 		case QLNXR_RESP_RDMA_IMM:
5431 			/* update opcode */
5432 			wc->opcode = IB_WC_RECV_RDMA_WITH_IMM;
5433 			/* fall to set imm data */
5434 		case QLNXR_RESP_IMM:
5435 			wc->ex.imm_data =
5436 				le32_to_cpu(resp->imm_data_or_inv_r_Key);
5437 			wc->wc_flags |= IB_WC_WITH_IMM;
5438 			break;
5439 		case QLNXR_RESP_RDMA:
5440 			QL_DPRINT11(ha, "Invalid flags QLNXR_RESP_RDMA [0x%x]"
5441 				"qp = %p qp->id = 0x%x cq = %p cq->icid = 0x%x\n",
5442 				resp->flags, qp, qp->id, cq, cq->icid );
5443 			break;
5444 		default:
5445 			/* valid configuration, but nothing todo here */
5446 			;
5447 		}
5448 #endif /* #if __FreeBSD_version >= 1102000 */
5449 
5450 		break;
5451 	default:
5452 		wc_status = IB_WC_GENERAL_ERR;
5453 	}
5454 
5455 	/* fill WC */
5456 	wc->status = wc_status;
5457 	wc->vendor_err = 0;
5458 	wc->src_qp = qp->id;
5459 	wc->qp = &qp->ibqp;
5460 	wc->wr_id = wr_id;
5461 
5462 	QL_DPRINT12(ha, "exit status = 0x%x\n", wc_status);
5463 
5464 	return;
5465 }
5466 
5467 static int
5468 process_resp_one_srq(struct qlnxr_dev *dev,
5469 	struct qlnxr_qp *qp,
5470 	struct qlnxr_cq *cq,
5471 	struct ib_wc *wc,
5472 	struct rdma_cqe_responder *resp)
5473 {
5474 	struct qlnxr_srq	*srq = qp->srq;
5475 	u64			wr_id;
5476 	qlnx_host_t		*ha = dev->ha;
5477 
5478 	QL_DPRINT12(ha, "enter\n");
5479 
5480 	wr_id = HILO_U64(resp->srq_wr_id.hi, resp->srq_wr_id.lo);
5481 
5482 	if (resp->status == RDMA_CQE_RESP_STS_WORK_REQUEST_FLUSHED_ERR) {
5483 		wc->status = IB_WC_WR_FLUSH_ERR;
5484 		wc->vendor_err = 0;
5485 		wc->wr_id = wr_id;
5486 		wc->byte_len = 0;
5487 		wc->src_qp = qp->id;
5488 		wc->qp = &qp->ibqp;
5489 		wc->wr_id = wr_id;
5490 	} else {
5491 		__process_resp_one(dev, qp, cq, wc, resp, wr_id);
5492 	}
5493 
5494 	/* PBL is maintained in case of WR granularity.
5495 	 * So increment WR consumer after consuming WR
5496 	 */
5497 	srq->hw_srq.wr_cons_cnt++;
5498 
5499 	QL_DPRINT12(ha, "exit\n");
5500 	return 1;
5501 }
5502 
5503 static int
5504 process_resp_one(struct qlnxr_dev *dev,
5505 	struct qlnxr_qp *qp,
5506 	struct qlnxr_cq *cq,
5507 	struct ib_wc *wc,
5508 	struct rdma_cqe_responder *resp)
5509 {
5510 	qlnx_host_t	*ha = dev->ha;
5511 	u64		wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id;
5512 
5513 	QL_DPRINT12(ha, "enter\n");
5514 
5515 	__process_resp_one(dev, qp, cq, wc, resp, wr_id);
5516 
5517 	while (qp->rqe_wr_id[qp->rq.cons].wqe_size--)
5518 		ecore_chain_consume(&qp->rq.pbl);
5519 	qlnxr_inc_sw_cons(&qp->rq);
5520 
5521 	QL_DPRINT12(ha, "exit\n");
5522 	return 1;
5523 }
5524 
5525 static int
5526 process_resp_flush(struct qlnxr_qp *qp,
5527 	int num_entries,
5528 	struct ib_wc *wc,
5529 	u16 hw_cons)
5530 {
5531 	u16		cnt = 0;
5532 	qlnx_host_t	*ha = qp->dev->ha;
5533 
5534 	QL_DPRINT12(ha, "enter\n");
5535 
5536 	while (num_entries && qp->rq.wqe_cons != hw_cons) {
5537 		/* fill WC */
5538 		wc->status = IB_WC_WR_FLUSH_ERR;
5539 		wc->vendor_err = 0;
5540 		wc->wc_flags = 0;
5541 		wc->src_qp = qp->id;
5542 		wc->byte_len = 0;
5543 		wc->wr_id = qp->rqe_wr_id[qp->rq.cons].wr_id;
5544 		wc->qp = &qp->ibqp;
5545 		num_entries--;
5546 		wc++;
5547 		cnt++;
5548 		while (qp->rqe_wr_id[qp->rq.cons].wqe_size--)
5549 			ecore_chain_consume(&qp->rq.pbl);
5550 		qlnxr_inc_sw_cons(&qp->rq);
5551 	}
5552 
5553 	QL_DPRINT12(ha, "exit cnt = 0x%x\n", cnt);
5554 	return cnt;
5555 }
5556 
5557 static void
5558 try_consume_resp_cqe(struct qlnxr_cq *cq,
5559 	struct qlnxr_qp *qp,
5560 	struct rdma_cqe_responder *resp,
5561 	int *update)
5562 {
5563 	if (le16_to_cpu(resp->rq_cons) == qp->rq.wqe_cons) {
5564 		consume_cqe(cq);
5565 		*update |= 1;
5566 	}
5567 }
5568 
5569 static int
5570 qlnxr_poll_cq_resp_srq(struct qlnxr_dev *dev,
5571 	struct qlnxr_qp *qp,
5572 	struct qlnxr_cq *cq,
5573 	int num_entries,
5574 	struct ib_wc *wc,
5575 	struct rdma_cqe_responder *resp,
5576 	int *update)
5577 {
5578 	int		cnt;
5579 	qlnx_host_t	*ha = dev->ha;
5580 
5581 	QL_DPRINT12(ha, "enter\n");
5582 
5583 	cnt = process_resp_one_srq(dev, qp, cq, wc, resp);
5584 	consume_cqe(cq);
5585 	*update |= 1;
5586 
5587 	QL_DPRINT12(ha, "exit cnt = 0x%x\n", cnt);
5588 	return cnt;
5589 }
5590 
5591 static int
5592 qlnxr_poll_cq_resp(struct qlnxr_dev *dev,
5593 	struct qlnxr_qp *qp,
5594 	struct qlnxr_cq *cq,
5595 	int num_entries,
5596 	struct ib_wc *wc,
5597 	struct rdma_cqe_responder *resp,
5598 	int *update)
5599 {
5600 	int		cnt;
5601 	qlnx_host_t	*ha = dev->ha;
5602 
5603 	QL_DPRINT12(ha, "enter\n");
5604 
5605 	if (resp->status == RDMA_CQE_RESP_STS_WORK_REQUEST_FLUSHED_ERR) {
5606 		cnt = process_resp_flush(qp, num_entries, wc,
5607 				resp->rq_cons);
5608 		try_consume_resp_cqe(cq, qp, resp, update);
5609 	} else {
5610 		cnt = process_resp_one(dev, qp, cq, wc, resp);
5611 		consume_cqe(cq);
5612 		*update |= 1;
5613 	}
5614 
5615 	QL_DPRINT12(ha, "exit cnt = 0x%x\n", cnt);
5616 	return cnt;
5617 }
5618 
5619 static void
5620 try_consume_req_cqe(struct qlnxr_cq *cq, struct qlnxr_qp *qp,
5621 	struct rdma_cqe_requester *req, int *update)
5622 {
5623 	if (le16_to_cpu(req->sq_cons) == qp->sq.wqe_cons) {
5624 		consume_cqe(cq);
5625 		*update |= 1;
5626 	}
5627 }
5628 
5629 static void
5630 doorbell_cq(struct qlnxr_dev *dev, struct qlnxr_cq *cq, u32 cons, u8 flags)
5631 {
5632 	uint64_t	reg_addr;
5633 	qlnx_host_t	*ha = dev->ha;
5634 
5635 	QL_DPRINT12(ha, "enter\n");
5636 
5637 	wmb();
5638 	cq->db.data.agg_flags = flags;
5639 	cq->db.data.value = cpu_to_le32(cons);
5640 
5641 	reg_addr = (uint64_t)((uint8_t *)cq->db_addr -
5642 				(uint8_t *)(ha->cdev.doorbells));
5643 
5644 	bus_write_8(ha->pci_dbells, reg_addr, cq->db.raw);
5645 	bus_barrier(ha->pci_dbells,  0, 0, BUS_SPACE_BARRIER_READ);
5646 
5647 	QL_DPRINT12(ha, "exit\n");
5648 	return;
5649 
5650 //#ifdef __LP64__
5651 //	writeq(cq->db.raw, cq->db_addr);
5652 //#else
5653 	/* Note that since the FW allows 64 bit write only, in 32bit systems
5654 	 * the value of db_addr must be low enough. This is currently not
5655 	 * enforced.
5656 	 */
5657 //	writel(cq->db.raw & 0xffffffff, cq->db_addr);
5658 //	mmiowb();
5659 //#endif
5660 }
5661 
5662 
5663 static int
5664 is_valid_cqe(struct qlnxr_cq *cq, union rdma_cqe *cqe)
5665 {
5666 	struct rdma_cqe_requester *resp_cqe = &cqe->req;
5667 	return (resp_cqe->flags & RDMA_RESIZE_CQ_RAMROD_DATA_TOGGLE_BIT_MASK) ==
5668 			cq->pbl_toggle;
5669 }
5670 
5671 int
5672 qlnxr_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
5673 {
5674 	struct qlnxr_cq	*cq = get_qlnxr_cq(ibcq);
5675 	struct qlnxr_dev *dev = get_qlnxr_dev((ibcq->device));
5676 	int		done = 0;
5677 	union rdma_cqe	*cqe = cq->latest_cqe;
5678 	int 		update = 0;
5679 	u32		old_cons, new_cons;
5680 	unsigned long	flags;
5681 	qlnx_host_t	*ha = dev->ha;
5682 
5683 	QL_DPRINT12(ha, "enter\n");
5684 
5685 	if (!(ha->ifp->if_drv_flags & IFF_DRV_RUNNING))
5686 		return -EINVAL;
5687 
5688 	if (cq->destroyed) {
5689 		QL_DPRINT11(ha, "called after destroy for cq %p (icid=%d)\n",
5690 			cq, cq->icid);
5691 		return 0;
5692 	}
5693 
5694 	if (cq->cq_type == QLNXR_CQ_TYPE_GSI)
5695 		return qlnxr_gsi_poll_cq(ibcq, num_entries, wc);
5696 
5697 	spin_lock_irqsave(&cq->cq_lock, flags);
5698 
5699 	old_cons = ecore_chain_get_cons_idx_u32(&cq->pbl);
5700 
5701 	while (num_entries && is_valid_cqe(cq, cqe)) {
5702 		int cnt = 0;
5703 		struct qlnxr_qp *qp;
5704 		struct rdma_cqe_requester *resp_cqe;
5705 		enum rdma_cqe_type cqe_type;
5706 
5707 		/* prevent speculative reads of any field of CQE */
5708 		rmb();
5709 
5710 		resp_cqe = &cqe->req;
5711 		qp = (struct qlnxr_qp *)(uintptr_t)HILO_U64(resp_cqe->qp_handle.hi,
5712 						resp_cqe->qp_handle.lo);
5713 
5714 		if (!qp) {
5715 			QL_DPRINT11(ha, "qp = NULL\n");
5716 			break;
5717 		}
5718 
5719 		wc->qp = &qp->ibqp;
5720 
5721 		cqe_type = GET_FIELD(resp_cqe->flags, RDMA_CQE_REQUESTER_TYPE);
5722 
5723 		switch (cqe_type) {
5724 		case RDMA_CQE_TYPE_REQUESTER:
5725 			cnt = qlnxr_poll_cq_req(dev, qp, cq, num_entries,
5726 					wc, &cqe->req);
5727 			try_consume_req_cqe(cq, qp, &cqe->req, &update);
5728 			break;
5729 		case RDMA_CQE_TYPE_RESPONDER_RQ:
5730 			cnt = qlnxr_poll_cq_resp(dev, qp, cq, num_entries,
5731 					wc, &cqe->resp, &update);
5732 			break;
5733 		case RDMA_CQE_TYPE_RESPONDER_SRQ:
5734 			cnt = qlnxr_poll_cq_resp_srq(dev, qp, cq, num_entries,
5735 					wc, &cqe->resp, &update);
5736 			break;
5737 		case RDMA_CQE_TYPE_INVALID:
5738 		default:
5739 			QL_DPRINT11(ha, "cqe type [0x%x] invalid\n", cqe_type);
5740 			break;
5741 		}
5742 		num_entries -= cnt;
5743 		wc += cnt;
5744 		done += cnt;
5745 
5746 		cqe = cq->latest_cqe;
5747 	}
5748 	new_cons = ecore_chain_get_cons_idx_u32(&cq->pbl);
5749 
5750 	cq->cq_cons += new_cons - old_cons;
5751 
5752 	if (update) {
5753 		/* doorbell notifies abount latest VALID entry,
5754 		 * but chain already point to the next INVALID one
5755 		 */
5756 		doorbell_cq(dev, cq, cq->cq_cons - 1, cq->arm_flags);
5757 		QL_DPRINT12(ha, "cq = %p cons = 0x%x "
5758 			"arm_flags = 0x%x db.icid = 0x%x\n", cq,
5759 			(cq->cq_cons - 1), cq->arm_flags, cq->db.data.icid);
5760 	}
5761 
5762 	spin_unlock_irqrestore(&cq->cq_lock, flags);
5763 
5764 	QL_DPRINT12(ha, "exit\n");
5765 
5766 	return done;
5767 }
5768 
5769 
5770 int
5771 qlnxr_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
5772 {
5773         struct qlnxr_cq *cq = get_qlnxr_cq(ibcq);
5774         unsigned long sflags;
5775         struct qlnxr_dev *dev;
5776 	qlnx_host_t	*ha;
5777 
5778 	dev = get_qlnxr_dev((ibcq->device));
5779 	ha = dev->ha;
5780 
5781 	QL_DPRINT12(ha, "enter ibcq = %p flags = 0x%x "
5782 		"cp = %p cons = 0x%x cq_type = 0x%x\n", ibcq,
5783 		flags, cq, cq->cq_cons, cq->cq_type);
5784 
5785 	if (!(ha->ifp->if_drv_flags & IFF_DRV_RUNNING))
5786 		return -EINVAL;
5787 
5788 	if (cq->destroyed) {
5789 		QL_DPRINT11(ha, "cq was already destroyed cq = %p icid=%d\n",
5790 			cq, cq->icid);
5791 		return -EINVAL;
5792 	}
5793 
5794         if (cq->cq_type == QLNXR_CQ_TYPE_GSI) {
5795                 return 0;
5796         }
5797 
5798         spin_lock_irqsave(&cq->cq_lock, sflags);
5799 
5800         cq->arm_flags = 0;
5801 
5802         if (flags & IB_CQ_SOLICITED) {
5803                 cq->arm_flags |= DQ_UCM_ROCE_CQ_ARM_SE_CF_CMD;
5804         }
5805         if (flags & IB_CQ_NEXT_COMP) {
5806                 cq->arm_flags |= DQ_UCM_ROCE_CQ_ARM_CF_CMD;
5807         }
5808 
5809         doorbell_cq(dev, cq, (cq->cq_cons - 1), cq->arm_flags);
5810 
5811         spin_unlock_irqrestore(&cq->cq_lock, sflags);
5812 
5813 	QL_DPRINT12(ha, "exit ibcq = %p flags = 0x%x\n", ibcq, flags);
5814         return 0;
5815 }
5816 
5817 
5818 static struct qlnxr_mr *
5819 __qlnxr_alloc_mr(struct ib_pd *ibpd, int max_page_list_len)
5820 {
5821 	struct qlnxr_pd *pd = get_qlnxr_pd(ibpd);
5822 	struct qlnxr_dev *dev = get_qlnxr_dev((ibpd->device));
5823 	struct qlnxr_mr *mr;
5824 	int		rc = -ENOMEM;
5825 	qlnx_host_t	*ha;
5826 
5827 	ha = dev->ha;
5828 
5829 	QL_DPRINT12(ha, "enter ibpd = %p pd = %p "
5830 		" pd_id = %d max_page_list_len = %d\n",
5831 		ibpd, pd, pd->pd_id, max_page_list_len);
5832 
5833 	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
5834 	if (!mr) {
5835 		QL_DPRINT11(ha, "kzalloc(mr) failed\n");
5836 		return ERR_PTR(rc);
5837 	}
5838 
5839 	mr->dev = dev;
5840 	mr->type = QLNXR_MR_FRMR;
5841 
5842 	rc = qlnxr_init_mr_info(dev, &mr->info, max_page_list_len,
5843 				  1 /* allow dual layer pbl */);
5844 	if (rc) {
5845 		QL_DPRINT11(ha, "qlnxr_init_mr_info failed\n");
5846 		goto err0;
5847 	}
5848 
5849 	rc = ecore_rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
5850 	if (rc) {
5851 		QL_DPRINT11(ha, "ecore_rdma_alloc_tid failed\n");
5852 		goto err0;
5853 	}
5854 
5855 	/* index only, 18 bit long, lkey = itid << 8 | key */
5856 	mr->hw_mr.tid_type = ECORE_RDMA_TID_FMR;
5857 	mr->hw_mr.key = 0;
5858 	mr->hw_mr.pd = pd->pd_id;
5859 	mr->hw_mr.local_read = 1;
5860 	mr->hw_mr.local_write = 0;
5861 	mr->hw_mr.remote_read = 0;
5862 	mr->hw_mr.remote_write = 0;
5863 	mr->hw_mr.remote_atomic = 0;
5864 	mr->hw_mr.mw_bind = false; /* TBD MW BIND */
5865 	mr->hw_mr.pbl_ptr = 0; /* Will be supplied during post */
5866 	mr->hw_mr.pbl_two_level = mr->info.pbl_info.two_layered;
5867 	mr->hw_mr.pbl_page_size_log = ilog2(mr->info.pbl_info.pbl_size);
5868 	mr->hw_mr.fbo = 0;
5869 	mr->hw_mr.length = 0;
5870 	mr->hw_mr.vaddr = 0;
5871 	mr->hw_mr.zbva = false; /* TBD figure when this should be true */
5872 	mr->hw_mr.phy_mr = true; /* Fast MR - True, Regular Register False */
5873 	mr->hw_mr.dma_mr = false;
5874 
5875 	rc = ecore_rdma_register_tid(dev->rdma_ctx, &mr->hw_mr);
5876 	if (rc) {
5877 		QL_DPRINT11(ha, "ecore_rdma_register_tid failed\n");
5878 		goto err1;
5879 	}
5880 
5881 	mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
5882 	mr->ibmr.rkey = mr->ibmr.lkey;
5883 
5884 	QL_DPRINT12(ha, "exit mr = %p mr->ibmr.lkey = 0x%x\n",
5885 		mr, mr->ibmr.lkey);
5886 
5887 	return mr;
5888 
5889 err1:
5890 	ecore_rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
5891 err0:
5892 	kfree(mr);
5893 
5894 	QL_DPRINT12(ha, "exit\n");
5895 
5896 	return ERR_PTR(rc);
5897 }
5898 
5899 #if __FreeBSD_version >= 1102000
5900 
5901 struct ib_mr *
5902 qlnxr_alloc_mr(struct ib_pd *ibpd, enum ib_mr_type mr_type, u32 max_num_sg)
5903 {
5904 	struct qlnxr_dev *dev;
5905 	struct qlnxr_mr *mr;
5906 	qlnx_host_t     *ha;
5907 
5908 	dev = get_qlnxr_dev(ibpd->device);
5909 	ha = dev->ha;
5910 
5911 	QL_DPRINT12(ha, "enter\n");
5912 
5913 	if (mr_type != IB_MR_TYPE_MEM_REG)
5914 		return ERR_PTR(-EINVAL);
5915 
5916 	mr = __qlnxr_alloc_mr(ibpd, max_num_sg);
5917 
5918 	if (IS_ERR(mr))
5919 		return ERR_PTR(-EINVAL);
5920 
5921 	QL_DPRINT12(ha, "exit mr = %p &mr->ibmr = %p\n", mr, &mr->ibmr);
5922 
5923 	return &mr->ibmr;
5924 }
5925 
5926 static int
5927 qlnxr_set_page(struct ib_mr *ibmr, u64 addr)
5928 {
5929 	struct qlnxr_mr *mr = get_qlnxr_mr(ibmr);
5930 	struct qlnxr_pbl *pbl_table;
5931 	struct regpair *pbe;
5932 	struct qlnxr_dev *dev;
5933 	qlnx_host_t     *ha;
5934 	u32 pbes_in_page;
5935 
5936 	dev = mr->dev;
5937 	ha = dev->ha;
5938 
5939 	if (unlikely(mr->npages == mr->info.pbl_info.num_pbes)) {
5940 		QL_DPRINT12(ha, "fails mr->npages %d\n", mr->npages);
5941 		return -ENOMEM;
5942 	}
5943 
5944 	QL_DPRINT12(ha, "mr->npages %d addr = %p enter\n", mr->npages,
5945 		((void *)addr));
5946 
5947 	pbes_in_page = mr->info.pbl_info.pbl_size / sizeof(u64);
5948 	pbl_table = mr->info.pbl_table + (mr->npages / pbes_in_page);
5949 	pbe = (struct regpair *)pbl_table->va;
5950 	pbe +=  mr->npages % pbes_in_page;
5951 	pbe->lo = cpu_to_le32((u32)addr);
5952 	pbe->hi = cpu_to_le32((u32)upper_32_bits(addr));
5953 
5954 	mr->npages++;
5955 
5956 	QL_DPRINT12(ha, "mr->npages %d addr = %p exit \n", mr->npages,
5957 		((void *)addr));
5958 	return 0;
5959 }
5960 
5961 int
5962 qlnxr_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
5963 	int sg_nents, unsigned int *sg_offset)
5964 {
5965 	int             ret;
5966 	struct qlnxr_mr *mr = get_qlnxr_mr(ibmr);
5967 	qlnx_host_t     *ha;
5968 
5969 	if (mr == NULL)
5970 		return (-1);
5971 
5972 	if (mr->dev == NULL)
5973 		return (-1);
5974 
5975 	ha = mr->dev->ha;
5976 
5977 	QL_DPRINT12(ha, "enter\n");
5978 
5979 	mr->npages = 0;
5980 	qlnx_handle_completed_mrs(mr->dev, &mr->info);
5981 
5982 	ret = ib_sg_to_pages(ibmr, sg, sg_nents, NULL, qlnxr_set_page);
5983 
5984 	QL_DPRINT12(ha, "exit ret = %d\n", ret);
5985 
5986 	return (ret);
5987 }
5988 
5989 #else
5990 
5991 struct ib_mr *
5992 qlnxr_alloc_frmr(struct ib_pd *ibpd, int max_page_list_len)
5993 {
5994 	struct qlnxr_dev *dev;
5995 	struct qlnxr_mr *mr;
5996 	qlnx_host_t	*ha;
5997 	struct ib_mr *ibmr = NULL;
5998 
5999 	dev = get_qlnxr_dev((ibpd->device));
6000 	ha = dev->ha;
6001 
6002 	QL_DPRINT12(ha, "enter\n");
6003 
6004 	mr = __qlnxr_alloc_mr(ibpd, max_page_list_len);
6005 
6006 	if (IS_ERR(mr)) {
6007 		ibmr = ERR_PTR(-EINVAL);
6008 	} else {
6009 		ibmr = &mr->ibmr;
6010 	}
6011 
6012 	QL_DPRINT12(ha, "exit %p\n", ibmr);
6013 	return (ibmr);
6014 }
6015 
6016 void
6017 qlnxr_free_frmr_page_list(struct ib_fast_reg_page_list *page_list)
6018 {
6019 	struct qlnxr_fast_reg_page_list *frmr_list;
6020 
6021 	frmr_list = get_qlnxr_frmr_list(page_list);
6022 
6023 	free_mr_info(frmr_list->dev, &frmr_list->info);
6024 
6025 	kfree(frmr_list->ibfrpl.page_list);
6026 	kfree(frmr_list);
6027 
6028 	return;
6029 }
6030 
6031 struct ib_fast_reg_page_list *
6032 qlnxr_alloc_frmr_page_list(struct ib_device *ibdev, int page_list_len)
6033 {
6034 	struct qlnxr_fast_reg_page_list *frmr_list = NULL;
6035 	struct qlnxr_dev		*dev;
6036 	int				size = page_list_len * sizeof(u64);
6037 	int				rc = -ENOMEM;
6038 	qlnx_host_t			*ha;
6039 
6040 	dev = get_qlnxr_dev(ibdev);
6041 	ha = dev->ha;
6042 
6043 	QL_DPRINT12(ha, "enter\n");
6044 
6045 	frmr_list = kzalloc(sizeof(*frmr_list), GFP_KERNEL);
6046 	if (!frmr_list) {
6047 		QL_DPRINT11(ha, "kzalloc(frmr_list) failed\n");
6048 		goto err;
6049 	}
6050 
6051 	frmr_list->dev = dev;
6052 	frmr_list->ibfrpl.page_list = kzalloc(size, GFP_KERNEL);
6053 	if (!frmr_list->ibfrpl.page_list) {
6054 		QL_DPRINT11(ha, "frmr_list->ibfrpl.page_list = NULL failed\n");
6055 		goto err0;
6056 	}
6057 
6058 	rc = qlnxr_init_mr_info(dev, &frmr_list->info, page_list_len,
6059 			  1 /* allow dual layer pbl */);
6060 	if (rc)
6061 		goto err1;
6062 
6063 	QL_DPRINT12(ha, "exit %p\n", &frmr_list->ibfrpl);
6064 
6065 	return &frmr_list->ibfrpl;
6066 
6067 err1:
6068 	kfree(frmr_list->ibfrpl.page_list);
6069 err0:
6070 	kfree(frmr_list);
6071 err:
6072 	QL_DPRINT12(ha, "exit with error\n");
6073 
6074 	return ERR_PTR(rc);
6075 }
6076 
6077 static int
6078 qlnxr_validate_phys_buf_list(qlnx_host_t *ha, struct ib_phys_buf *buf_list,
6079 	int buf_cnt, uint64_t *total_size)
6080 {
6081 	u64 size = 0;
6082 
6083 	*total_size = 0;
6084 
6085 	if (!buf_cnt || buf_list == NULL) {
6086 		QL_DPRINT11(ha,
6087 			"failed buf_list = %p buf_cnt = %d\n", buf_list, buf_cnt);
6088 		return (-1);
6089 	}
6090 
6091 	size = buf_list->size;
6092 
6093 	if (!size) {
6094 		QL_DPRINT11(ha,
6095 			"failed buf_list = %p buf_cnt = %d"
6096 			" buf_list->size = 0\n", buf_list, buf_cnt);
6097 		return (-1);
6098 	}
6099 
6100 	while (buf_cnt) {
6101 
6102 		*total_size += buf_list->size;
6103 
6104 		if (buf_list->size != size) {
6105 			QL_DPRINT11(ha,
6106 				"failed buf_list = %p buf_cnt = %d"
6107 				" all buffers should have same size\n",
6108 				buf_list, buf_cnt);
6109 			return (-1);
6110 		}
6111 
6112 		buf_list++;
6113 		buf_cnt--;
6114 	}
6115 	return (0);
6116 }
6117 
6118 static size_t
6119 qlnxr_get_num_pages(qlnx_host_t *ha, struct ib_phys_buf *buf_list,
6120 	int buf_cnt)
6121 {
6122 	int	i;
6123 	size_t	num_pages = 0;
6124 	u64	size;
6125 
6126 	for (i = 0; i < buf_cnt; i++) {
6127 
6128 		size = 0;
6129 		while (size < buf_list->size) {
6130 			size += PAGE_SIZE;
6131 			num_pages++;
6132 		}
6133 		buf_list++;
6134 	}
6135 	return (num_pages);
6136 }
6137 
6138 static void
6139 qlnxr_populate_phys_mem_pbls(struct qlnxr_dev *dev,
6140 	struct ib_phys_buf *buf_list, int buf_cnt,
6141 	struct qlnxr_pbl *pbl, struct qlnxr_pbl_info *pbl_info)
6142 {
6143 	struct regpair		*pbe;
6144 	struct qlnxr_pbl	*pbl_tbl;
6145 	int			pg_cnt, pages, pbe_cnt, total_num_pbes = 0;
6146 	qlnx_host_t		*ha;
6147         int                     i;
6148 	u64			pbe_addr;
6149 
6150 	ha = dev->ha;
6151 
6152 	QL_DPRINT12(ha, "enter\n");
6153 
6154 	if (!pbl_info) {
6155 		QL_DPRINT11(ha, "PBL_INFO not initialized\n");
6156 		return;
6157 	}
6158 
6159 	if (!pbl_info->num_pbes) {
6160 		QL_DPRINT11(ha, "pbl_info->num_pbes == 0\n");
6161 		return;
6162 	}
6163 
6164 	/* If we have a two layered pbl, the first pbl points to the rest
6165 	 * of the pbls and the first entry lays on the second pbl in the table
6166 	 */
6167 	if (pbl_info->two_layered)
6168 		pbl_tbl = &pbl[1];
6169 	else
6170 		pbl_tbl = pbl;
6171 
6172 	pbe = (struct regpair *)pbl_tbl->va;
6173 	if (!pbe) {
6174 		QL_DPRINT12(ha, "pbe is NULL\n");
6175 		return;
6176 	}
6177 
6178 	pbe_cnt = 0;
6179 
6180 	for (i = 0; i < buf_cnt; i++) {
6181 
6182 		pages = buf_list->size >> PAGE_SHIFT;
6183 
6184 		for (pg_cnt = 0; pg_cnt < pages; pg_cnt++) {
6185 			/* store the page address in pbe */
6186 
6187 			pbe_addr = buf_list->addr + (PAGE_SIZE * pg_cnt);
6188 
6189 			pbe->lo = cpu_to_le32((u32)pbe_addr);
6190 			pbe->hi = cpu_to_le32(((u32)(pbe_addr >> 32)));
6191 
6192 			QL_DPRINT12(ha, "Populate pbl table:"
6193 				" pbe->addr=0x%x:0x%x "
6194 				" pbe_cnt = %d total_num_pbes=%d"
6195 				" pbe=%p\n", pbe->lo, pbe->hi, pbe_cnt,
6196 				total_num_pbes, pbe);
6197 
6198 			pbe_cnt ++;
6199 			total_num_pbes ++;
6200 			pbe++;
6201 
6202 			if (total_num_pbes == pbl_info->num_pbes)
6203 				return;
6204 
6205 			/* if the given pbl is full storing the pbes,
6206 			 * move to next pbl.  */
6207 
6208 			if (pbe_cnt == (pbl_info->pbl_size / sizeof(u64))) {
6209 				pbl_tbl++;
6210 				pbe = (struct regpair *)pbl_tbl->va;
6211 				pbe_cnt = 0;
6212 			}
6213 		}
6214 		buf_list++;
6215 	}
6216 	QL_DPRINT12(ha, "exit\n");
6217 	return;
6218 }
6219 
6220 struct ib_mr *
6221 qlnxr_reg_kernel_mr(struct ib_pd *ibpd,
6222 	struct ib_phys_buf *buf_list,
6223 	int buf_cnt, int acc, u64 *iova_start)
6224 {
6225 	int		rc = -ENOMEM;
6226 	struct qlnxr_dev *dev = get_qlnxr_dev((ibpd->device));
6227 	struct qlnxr_mr *mr;
6228 	struct qlnxr_pd *pd;
6229 	qlnx_host_t	*ha;
6230 	size_t		num_pages = 0;
6231 	uint64_t	length;
6232 
6233 	ha = dev->ha;
6234 
6235 	QL_DPRINT12(ha, "enter\n");
6236 
6237 	pd = get_qlnxr_pd(ibpd);
6238 
6239 	QL_DPRINT12(ha, "pd = %d buf_list = %p, buf_cnt = %d,"
6240 		" iova_start = %p, acc = %d\n",
6241 		pd->pd_id, buf_list, buf_cnt, iova_start, acc);
6242 
6243 	//if (acc & IB_ACCESS_REMOTE_WRITE && !(acc & IB_ACCESS_LOCAL_WRITE)) {
6244 	//	QL_DPRINT11(ha, "(acc & IB_ACCESS_REMOTE_WRITE &&"
6245 	//		" !(acc & IB_ACCESS_LOCAL_WRITE))\n");
6246 	//	return ERR_PTR(-EINVAL);
6247 	//}
6248 
6249 	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
6250 	if (!mr) {
6251 		QL_DPRINT11(ha, "kzalloc(mr) failed\n");
6252 		return ERR_PTR(rc);
6253 	}
6254 
6255 	mr->type = QLNXR_MR_KERNEL;
6256 	mr->iova_start = iova_start;
6257 
6258 	rc = qlnxr_validate_phys_buf_list(ha, buf_list, buf_cnt, &length);
6259 	if (rc)
6260 		goto err0;
6261 
6262 	num_pages = qlnxr_get_num_pages(ha, buf_list, buf_cnt);
6263 	if (!num_pages)
6264 		goto err0;
6265 
6266 	rc = qlnxr_init_mr_info(dev, &mr->info, num_pages, 1);
6267 	if (rc) {
6268 		QL_DPRINT11(ha,
6269 			"qlnxr_init_mr_info failed [%d]\n", rc);
6270 		goto err1;
6271 	}
6272 
6273 	qlnxr_populate_phys_mem_pbls(dev, buf_list, buf_cnt, mr->info.pbl_table,
6274 		   &mr->info.pbl_info);
6275 
6276 	rc = ecore_rdma_alloc_tid(dev->rdma_ctx, &mr->hw_mr.itid);
6277 
6278 	if (rc) {
6279 		QL_DPRINT11(ha, "roce alloc tid returned an error %d\n", rc);
6280 		goto err1;
6281 	}
6282 
6283 	/* index only, 18 bit long, lkey = itid << 8 | key */
6284 	mr->hw_mr.tid_type = ECORE_RDMA_TID_REGISTERED_MR;
6285 	mr->hw_mr.key = 0;
6286 	mr->hw_mr.pd = pd->pd_id;
6287 	mr->hw_mr.local_read = 1;
6288 	mr->hw_mr.local_write = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0;
6289 	mr->hw_mr.remote_read = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0;
6290 	mr->hw_mr.remote_write = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
6291 	mr->hw_mr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0;
6292 	mr->hw_mr.mw_bind = false; /* TBD MW BIND */
6293 	mr->hw_mr.pbl_ptr = mr->info.pbl_table[0].pa;
6294 	mr->hw_mr.pbl_two_level = mr->info.pbl_info.two_layered;
6295 	mr->hw_mr.pbl_page_size_log = ilog2(mr->info.pbl_info.pbl_size);
6296 	mr->hw_mr.page_size_log = ilog2(PAGE_SIZE); /* for the MR pages */
6297 
6298 	mr->hw_mr.fbo = 0;
6299 
6300 	mr->hw_mr.length = length;
6301 	mr->hw_mr.vaddr = (uint64_t)iova_start;
6302 	mr->hw_mr.zbva = false; /* TBD figure when this should be true */
6303 	mr->hw_mr.phy_mr = false; /* Fast MR - True, Regular Register False */
6304 	mr->hw_mr.dma_mr = false;
6305 
6306 	rc = ecore_rdma_register_tid(dev->rdma_ctx, &mr->hw_mr);
6307 	if (rc) {
6308 		QL_DPRINT11(ha, "roce register tid returned an error %d\n", rc);
6309 		goto err2;
6310 	}
6311 
6312 	mr->ibmr.lkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
6313 	if (mr->hw_mr.remote_write || mr->hw_mr.remote_read ||
6314 		mr->hw_mr.remote_atomic)
6315 		mr->ibmr.rkey = mr->hw_mr.itid << 8 | mr->hw_mr.key;
6316 
6317 	QL_DPRINT12(ha, "lkey: %x\n", mr->ibmr.lkey);
6318 
6319 	return (&mr->ibmr);
6320 
6321 err2:
6322 	ecore_rdma_free_tid(dev->rdma_ctx, mr->hw_mr.itid);
6323 err1:
6324 	qlnxr_free_pbl(dev, &mr->info.pbl_info, mr->info.pbl_table);
6325 err0:
6326 	kfree(mr);
6327 
6328 	QL_DPRINT12(ha, "exit [%d]\n", rc);
6329 	return (ERR_PTR(rc));
6330 }
6331 
6332 #endif /* #if __FreeBSD_version >= 1102000 */
6333 
6334 struct ib_ah *
6335 #if __FreeBSD_version >= 1102000
6336 qlnxr_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr,
6337 	struct ib_udata *udata)
6338 #else
6339 qlnxr_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr)
6340 #endif /* #if __FreeBSD_version >= 1102000 */
6341 {
6342 	struct qlnxr_dev *dev;
6343 	qlnx_host_t	*ha;
6344 	struct qlnxr_ah *ah;
6345 
6346 	dev = get_qlnxr_dev((ibpd->device));
6347 	ha = dev->ha;
6348 
6349 	QL_DPRINT12(ha, "in create_ah\n");
6350 
6351 	ah = kzalloc(sizeof(*ah), GFP_ATOMIC);
6352 	if (!ah) {
6353 		QL_DPRINT12(ha, "no address handle can be allocated\n");
6354 		return ERR_PTR(-ENOMEM);
6355 	}
6356 
6357 	ah->attr = *attr;
6358 
6359 	return &ah->ibah;
6360 }
6361 
6362 int
6363 qlnxr_destroy_ah(struct ib_ah *ibah)
6364 {
6365 	struct qlnxr_dev *dev;
6366 	qlnx_host_t     *ha;
6367 	struct qlnxr_ah *ah = get_qlnxr_ah(ibah);
6368 
6369 	dev = get_qlnxr_dev((ibah->device));
6370 	ha = dev->ha;
6371 
6372 	QL_DPRINT12(ha, "in destroy_ah\n");
6373 
6374 	kfree(ah);
6375 	return 0;
6376 }
6377 
6378 int
6379 qlnxr_query_ah(struct ib_ah *ibah, struct ib_ah_attr *attr)
6380 {
6381 	struct qlnxr_dev *dev;
6382 	qlnx_host_t     *ha;
6383 
6384 	dev = get_qlnxr_dev((ibah->device));
6385 	ha = dev->ha;
6386 	QL_DPRINT12(ha, "Query AH not supported\n");
6387 	return -EINVAL;
6388 }
6389 
6390 int
6391 qlnxr_modify_ah(struct ib_ah *ibah, struct ib_ah_attr *attr)
6392 {
6393 	struct qlnxr_dev *dev;
6394 	qlnx_host_t     *ha;
6395 
6396 	dev = get_qlnxr_dev((ibah->device));
6397 	ha = dev->ha;
6398 	QL_DPRINT12(ha, "Modify AH not supported\n");
6399 	return -ENOSYS;
6400 }
6401 
6402 #if __FreeBSD_version >= 1102000
6403 int
6404 qlnxr_process_mad(struct ib_device *ibdev,
6405 		int process_mad_flags,
6406 		u8 port_num,
6407 		const struct ib_wc *in_wc,
6408 		const struct ib_grh *in_grh,
6409 		const struct ib_mad_hdr *mad_hdr,
6410 		size_t in_mad_size,
6411 		struct ib_mad_hdr *out_mad,
6412 		size_t *out_mad_size,
6413 		u16 *out_mad_pkey_index)
6414 
6415 #else
6416 
6417 int
6418 qlnxr_process_mad(struct ib_device *ibdev,
6419                         int process_mad_flags,
6420                         u8 port_num,
6421                         struct ib_wc *in_wc,
6422                         struct ib_grh *in_grh,
6423                         struct ib_mad *in_mad,
6424                         struct ib_mad *out_mad)
6425 
6426 #endif /* #if __FreeBSD_version >= 1102000 */
6427 {
6428 	struct qlnxr_dev *dev;
6429 	qlnx_host_t	*ha;
6430 
6431 	dev = get_qlnxr_dev(ibdev);
6432 	ha = dev->ha;
6433 	QL_DPRINT12(ha, "process mad not supported\n");
6434 
6435 	return -ENOSYS;
6436 //	QL_DPRINT12(ha, "qlnxr_process_mad in_mad %x %x %x %x %x %x %x %x\n",
6437 //               in_mad->mad_hdr.attr_id, in_mad->mad_hdr.base_version,
6438 //               in_mad->mad_hdr.attr_mod, in_mad->mad_hdr.class_specific,
6439 //               in_mad->mad_hdr.class_version, in_mad->mad_hdr.method,
6440 //               in_mad->mad_hdr.mgmt_class, in_mad->mad_hdr.status);
6441 
6442 //	return IB_MAD_RESULT_SUCCESS;
6443 }
6444 
6445 
6446 #if __FreeBSD_version >= 1102000
6447 int
6448 qlnxr_get_port_immutable(struct ib_device *ibdev, u8 port_num,
6449 	struct ib_port_immutable *immutable)
6450 {
6451 	struct qlnxr_dev        *dev;
6452 	qlnx_host_t             *ha;
6453 	struct ib_port_attr     attr;
6454 	int                     err;
6455 
6456 	dev = get_qlnxr_dev(ibdev);
6457 	ha = dev->ha;
6458 
6459 	QL_DPRINT12(ha, "enter\n");
6460 
6461 	err = qlnxr_query_port(ibdev, port_num, &attr);
6462 	if (err)
6463 		return err;
6464 
6465 	if (QLNX_IS_IWARP(dev)) {
6466 		immutable->pkey_tbl_len = 1;
6467 		immutable->gid_tbl_len = 1;
6468 		immutable->core_cap_flags = RDMA_CORE_PORT_IWARP;
6469 		immutable->max_mad_size = 0;
6470 	} else {
6471 		immutable->pkey_tbl_len = attr.pkey_tbl_len;
6472 		immutable->gid_tbl_len = attr.gid_tbl_len;
6473 		immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
6474 		immutable->max_mad_size = IB_MGMT_MAD_SIZE;
6475 	}
6476 
6477 	QL_DPRINT12(ha, "exit\n");
6478 	return 0;
6479 }
6480 #endif /* #if __FreeBSD_version > 1102000 */
6481 
6482 
6483 /***** iWARP related functions *************/
6484 
6485 
6486 static void
6487 qlnxr_iw_mpa_request(void *context,
6488 	struct ecore_iwarp_cm_event_params *params)
6489 {
6490 	struct qlnxr_iw_listener *listener = (struct qlnxr_iw_listener *)context;
6491 	struct qlnxr_dev *dev = listener->dev;
6492 	struct qlnxr_iw_ep *ep;
6493 	struct iw_cm_event event;
6494 	struct sockaddr_in *laddr;
6495 	struct sockaddr_in *raddr;
6496 	qlnx_host_t	*ha;
6497 
6498 	ha = dev->ha;
6499 
6500 	QL_DPRINT12(ha, "enter\n");
6501 
6502 	if (params->cm_info->ip_version != ECORE_TCP_IPV4) {
6503 		QL_DPRINT11(ha, "only IPv4 supported [0x%x]\n",
6504 			params->cm_info->ip_version);
6505 		return;
6506 	}
6507 
6508 	ep = kzalloc(sizeof(*ep), GFP_ATOMIC);
6509 
6510 	if (!ep) {
6511 		QL_DPRINT11(ha, "kzalloc{ep) failed\n");
6512 		return;
6513 	}
6514 
6515 	ep->dev = dev;
6516 	ep->ecore_context = params->ep_context;
6517 
6518 	memset(&event, 0, sizeof(event));
6519 
6520 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
6521 	event.status = params->status;
6522 
6523 	laddr = (struct sockaddr_in *)&event.local_addr;
6524 	raddr = (struct sockaddr_in *)&event.remote_addr;
6525 
6526 	laddr->sin_family = AF_INET;
6527 	raddr->sin_family = AF_INET;
6528 
6529 	laddr->sin_port = htons(params->cm_info->local_port);
6530 	raddr->sin_port = htons(params->cm_info->remote_port);
6531 
6532 	laddr->sin_addr.s_addr = htonl(params->cm_info->local_ip[0]);
6533 	raddr->sin_addr.s_addr = htonl(params->cm_info->remote_ip[0]);
6534 
6535 	event.provider_data = (void *)ep;
6536 	event.private_data = (void *)params->cm_info->private_data;
6537 	event.private_data_len = (u8)params->cm_info->private_data_len;
6538 
6539 #if __FreeBSD_version >= 1100000
6540 	event.ord = params->cm_info->ord;
6541 	event.ird = params->cm_info->ird;
6542 #endif /* #if __FreeBSD_version >= 1100000 */
6543 
6544 	listener->cm_id->event_handler(listener->cm_id, &event);
6545 
6546 	QL_DPRINT12(ha, "exit\n");
6547 
6548 	return;
6549 }
6550 
6551 static void
6552 qlnxr_iw_issue_event(void *context,
6553 	 struct ecore_iwarp_cm_event_params *params,
6554 	 enum iw_cm_event_type event_type,
6555 	 char *str)
6556 {
6557 	struct qlnxr_iw_ep *ep = (struct qlnxr_iw_ep *)context;
6558 	struct qlnxr_dev *dev = ep->dev;
6559 	struct iw_cm_event event;
6560 	qlnx_host_t	*ha;
6561 
6562 	ha = dev->ha;
6563 
6564 	QL_DPRINT12(ha, "enter\n");
6565 
6566 	memset(&event, 0, sizeof(event));
6567 	event.status = params->status;
6568 	event.event = event_type;
6569 
6570 	if (params->cm_info != NULL) {
6571 #if __FreeBSD_version >= 1100000
6572 		event.ird = params->cm_info->ird;
6573 		event.ord = params->cm_info->ord;
6574 		QL_DPRINT12(ha, "ord=[%d] \n", event.ord);
6575 		QL_DPRINT12(ha, "ird=[%d] \n", event.ird);
6576 #endif /* #if __FreeBSD_version >= 1100000 */
6577 
6578 		event.private_data_len = params->cm_info->private_data_len;
6579 		event.private_data = (void *)params->cm_info->private_data;
6580 		QL_DPRINT12(ha, "private_data_len=[%d] \n",
6581 			event.private_data_len);
6582 	}
6583 
6584 	QL_DPRINT12(ha, "event=[%d] %s\n", event.event, str);
6585 	QL_DPRINT12(ha, "status=[%d] \n", event.status);
6586 
6587 	if (ep) {
6588 		if (ep->cm_id)
6589 			ep->cm_id->event_handler(ep->cm_id, &event);
6590 		else
6591 			QL_DPRINT11(ha, "ep->cm_id == NULL \n");
6592 	} else {
6593 		QL_DPRINT11(ha, "ep == NULL \n");
6594 	}
6595 
6596 	QL_DPRINT12(ha, "exit\n");
6597 
6598 	return;
6599 }
6600 
6601 static void
6602 qlnxr_iw_close_event(void *context,
6603 	 struct ecore_iwarp_cm_event_params *params)
6604 {
6605 	struct qlnxr_iw_ep *ep = (struct qlnxr_iw_ep *)context;
6606 	struct qlnxr_dev *dev = ep->dev;
6607 	qlnx_host_t	*ha;
6608 
6609 	ha = dev->ha;
6610 
6611 	QL_DPRINT12(ha, "enter\n");
6612 
6613 	if (ep->cm_id) {
6614 		qlnxr_iw_issue_event(context,
6615 				    params,
6616 				    IW_CM_EVENT_CLOSE,
6617 				    "IW_CM_EVENT_EVENT_CLOSE");
6618 		ep->cm_id->rem_ref(ep->cm_id);
6619 		ep->cm_id = NULL;
6620 	}
6621 
6622 	QL_DPRINT12(ha, "exit\n");
6623 
6624 	return;
6625 }
6626 
6627 #if __FreeBSD_version >= 1102000
6628 
6629 static void
6630 qlnxr_iw_passive_complete(void *context,
6631         struct ecore_iwarp_cm_event_params *params)
6632 {
6633         struct qlnxr_iw_ep      *ep = (struct qlnxr_iw_ep *)context;
6634         struct qlnxr_dev        *dev = ep->dev;
6635         qlnx_host_t             *ha;
6636 
6637         ha = dev->ha;
6638 
6639         /* We will only reach the following state if MPA_REJECT was called on
6640          * passive. In this case there will be no associated QP.
6641          */
6642         if ((params->status == -ECONNREFUSED) && (ep->qp == NULL)) {
6643                 QL_DPRINT11(ha, "PASSIVE connection refused releasing ep...\n");
6644                 kfree(ep);
6645                 return;
6646         }
6647 
6648         /* We always issue an established event, however, ofed does not look
6649          * at event code for established. So if there was a failure, we follow
6650          * with close...
6651          */
6652         qlnxr_iw_issue_event(context,
6653                 params,
6654                 IW_CM_EVENT_ESTABLISHED,
6655                 "IW_CM_EVENT_ESTABLISHED");
6656 
6657         if (params->status < 0) {
6658                 qlnxr_iw_close_event(context, params);
6659         }
6660 
6661         return;
6662 }
6663 
6664 struct qlnxr_discon_work {
6665         struct work_struct work;
6666         struct qlnxr_iw_ep *ep;
6667         enum ecore_iwarp_event_type event;
6668         int status;
6669 };
6670 
6671 static void
6672 qlnxr_iw_disconnect_worker(struct work_struct *work)
6673 {
6674         struct qlnxr_discon_work *dwork =
6675                 container_of(work, struct qlnxr_discon_work, work);
6676         struct ecore_rdma_modify_qp_in_params qp_params = { 0 };
6677         struct qlnxr_iw_ep *ep = dwork->ep;
6678         struct qlnxr_dev *dev = ep->dev;
6679         struct qlnxr_qp *qp = ep->qp;
6680         struct iw_cm_event event;
6681 
6682         if (qp->destroyed) {
6683                 kfree(dwork);
6684                 qlnxr_iw_qp_rem_ref(&qp->ibqp);
6685                 return;
6686         }
6687 
6688         memset(&event, 0, sizeof(event));
6689         event.status = dwork->status;
6690         event.event = IW_CM_EVENT_DISCONNECT;
6691 
6692         /* Success means graceful disconnect was requested. modifying
6693          * to SQD is translated to graceful disconnect. O/w reset is sent
6694          */
6695         if (dwork->status)
6696                 qp_params.new_state = ECORE_ROCE_QP_STATE_ERR;
6697         else
6698                 qp_params.new_state = ECORE_ROCE_QP_STATE_SQD;
6699 
6700         kfree(dwork);
6701 
6702         if (ep->cm_id)
6703                 ep->cm_id->event_handler(ep->cm_id, &event);
6704 
6705         SET_FIELD(qp_params.modify_flags,
6706                   ECORE_RDMA_MODIFY_QP_VALID_NEW_STATE, 1);
6707 
6708         ecore_rdma_modify_qp(dev->rdma_ctx, qp->ecore_qp, &qp_params);
6709 
6710         qlnxr_iw_qp_rem_ref(&qp->ibqp);
6711 
6712         return;
6713 }
6714 
6715 void
6716 qlnxr_iw_disconnect_event(void *context,
6717         struct ecore_iwarp_cm_event_params *params)
6718 {
6719         struct qlnxr_discon_work *work;
6720         struct qlnxr_iw_ep *ep = (struct qlnxr_iw_ep *)context;
6721         struct qlnxr_dev *dev = ep->dev;
6722         struct qlnxr_qp *qp = ep->qp;
6723 
6724         work = kzalloc(sizeof(*work), GFP_ATOMIC);
6725         if (!work)
6726                 return;
6727 
6728         qlnxr_iw_qp_add_ref(&qp->ibqp);
6729         work->ep = ep;
6730         work->event = params->event;
6731         work->status = params->status;
6732 
6733         INIT_WORK(&work->work, qlnxr_iw_disconnect_worker);
6734         queue_work(dev->iwarp_wq, &work->work);
6735 
6736         return;
6737 }
6738 
6739 #endif /* #if __FreeBSD_version >= 1102000 */
6740 
6741 static int
6742 qlnxr_iw_mpa_reply(void *context,
6743 	struct ecore_iwarp_cm_event_params *params)
6744 {
6745         struct qlnxr_iw_ep	*ep = (struct qlnxr_iw_ep *)context;
6746         struct qlnxr_dev	*dev = ep->dev;
6747         struct ecore_iwarp_send_rtr_in rtr_in;
6748         int			rc;
6749 	qlnx_host_t		*ha;
6750 
6751 	ha = dev->ha;
6752 
6753 	QL_DPRINT12(ha, "enter\n");
6754 
6755 	if (!(ha->ifp->if_drv_flags & IFF_DRV_RUNNING))
6756 		return -EINVAL;
6757 
6758 	bzero(&rtr_in, sizeof(struct ecore_iwarp_send_rtr_in));
6759         rtr_in.ep_context = params->ep_context;
6760 
6761         rc = ecore_iwarp_send_rtr(dev->rdma_ctx, &rtr_in);
6762 
6763 	QL_DPRINT12(ha, "exit rc = %d\n", rc);
6764         return rc;
6765 }
6766 
6767 
6768 void
6769 qlnxr_iw_qp_event(void *context,
6770 	struct ecore_iwarp_cm_event_params *params,
6771 	enum ib_event_type ib_event,
6772 	char *str)
6773 {
6774         struct qlnxr_iw_ep *ep = (struct qlnxr_iw_ep *)context;
6775         struct qlnxr_dev *dev = ep->dev;
6776         struct ib_qp *ibqp = &(ep->qp->ibqp);
6777         struct ib_event event;
6778 	qlnx_host_t	*ha;
6779 
6780 	ha = dev->ha;
6781 
6782 	QL_DPRINT12(ha,
6783 		"[context, event, event_handler] = [%p, 0x%x, %s, %p] enter\n",
6784 		context, params->event, str, ibqp->event_handler);
6785 
6786         if (ibqp->event_handler) {
6787                 event.event = ib_event;
6788                 event.device = ibqp->device;
6789                 event.element.qp = ibqp;
6790                 ibqp->event_handler(&event, ibqp->qp_context);
6791         }
6792 
6793 	return;
6794 }
6795 
6796 int
6797 qlnxr_iw_event_handler(void *context,
6798 	struct ecore_iwarp_cm_event_params *params)
6799 {
6800 	struct qlnxr_iw_ep *ep = (struct qlnxr_iw_ep *)context;
6801 	struct qlnxr_dev *dev = ep->dev;
6802 	qlnx_host_t	*ha;
6803 
6804 	ha = dev->ha;
6805 
6806 	QL_DPRINT12(ha, "[context, event] = [%p, 0x%x] "
6807 		"enter\n", context, params->event);
6808 
6809 	switch (params->event) {
6810 
6811 	/* Passive side request received */
6812 	case ECORE_IWARP_EVENT_MPA_REQUEST:
6813 		qlnxr_iw_mpa_request(context, params);
6814 		break;
6815 
6816         case ECORE_IWARP_EVENT_ACTIVE_MPA_REPLY:
6817                 qlnxr_iw_mpa_reply(context, params);
6818                 break;
6819 
6820 	/* Passive side established ( ack on mpa response ) */
6821 	case ECORE_IWARP_EVENT_PASSIVE_COMPLETE:
6822 
6823 #if __FreeBSD_version >= 1102000
6824 
6825 		ep->during_connect = 0;
6826 		qlnxr_iw_passive_complete(context, params);
6827 
6828 #else
6829 		qlnxr_iw_issue_event(context,
6830 				    params,
6831 				    IW_CM_EVENT_ESTABLISHED,
6832 				    "IW_CM_EVENT_ESTABLISHED");
6833 #endif /* #if __FreeBSD_version >= 1102000 */
6834 		break;
6835 
6836 	/* Active side reply received */
6837 	case ECORE_IWARP_EVENT_ACTIVE_COMPLETE:
6838 		ep->during_connect = 0;
6839 		qlnxr_iw_issue_event(context,
6840 				    params,
6841 				    IW_CM_EVENT_CONNECT_REPLY,
6842 				    "IW_CM_EVENT_CONNECT_REPLY");
6843 		if (params->status < 0) {
6844 			struct qlnxr_iw_ep *ep = (struct qlnxr_iw_ep *)context;
6845 
6846 			ep->cm_id->rem_ref(ep->cm_id);
6847 			ep->cm_id = NULL;
6848 		}
6849 		break;
6850 
6851 	case ECORE_IWARP_EVENT_DISCONNECT:
6852 
6853 #if __FreeBSD_version >= 1102000
6854 		qlnxr_iw_disconnect_event(context, params);
6855 #else
6856 		qlnxr_iw_issue_event(context,
6857 				    params,
6858 				    IW_CM_EVENT_DISCONNECT,
6859 				    "IW_CM_EVENT_DISCONNECT");
6860 		qlnxr_iw_close_event(context, params);
6861 #endif /* #if __FreeBSD_version >= 1102000 */
6862 		break;
6863 
6864 	case ECORE_IWARP_EVENT_CLOSE:
6865 		ep->during_connect = 0;
6866 		qlnxr_iw_close_event(context, params);
6867 		break;
6868 
6869         case ECORE_IWARP_EVENT_RQ_EMPTY:
6870                 qlnxr_iw_qp_event(context, params, IB_EVENT_QP_FATAL,
6871                                  "IWARP_EVENT_RQ_EMPTY");
6872                 break;
6873 
6874         case ECORE_IWARP_EVENT_IRQ_FULL:
6875                 qlnxr_iw_qp_event(context, params, IB_EVENT_QP_FATAL,
6876                                  "IWARP_EVENT_IRQ_FULL");
6877                 break;
6878 
6879         case ECORE_IWARP_EVENT_LLP_TIMEOUT:
6880                 qlnxr_iw_qp_event(context, params, IB_EVENT_QP_FATAL,
6881                                  "IWARP_EVENT_LLP_TIMEOUT");
6882                 break;
6883 
6884         case ECORE_IWARP_EVENT_REMOTE_PROTECTION_ERROR:
6885                 qlnxr_iw_qp_event(context, params, IB_EVENT_QP_ACCESS_ERR,
6886                                  "IWARP_EVENT_REMOTE_PROTECTION_ERROR");
6887                 break;
6888 
6889         case ECORE_IWARP_EVENT_CQ_OVERFLOW:
6890                 qlnxr_iw_qp_event(context, params, IB_EVENT_QP_FATAL,
6891                                  "QED_IWARP_EVENT_CQ_OVERFLOW");
6892                 break;
6893 
6894         case ECORE_IWARP_EVENT_QP_CATASTROPHIC:
6895                 qlnxr_iw_qp_event(context, params, IB_EVENT_QP_FATAL,
6896                                  "QED_IWARP_EVENT_QP_CATASTROPHIC");
6897                 break;
6898 
6899         case ECORE_IWARP_EVENT_LOCAL_ACCESS_ERROR:
6900                 qlnxr_iw_qp_event(context, params, IB_EVENT_QP_ACCESS_ERR,
6901                                  "IWARP_EVENT_LOCAL_ACCESS_ERROR");
6902                 break;
6903 
6904         case ECORE_IWARP_EVENT_REMOTE_OPERATION_ERROR:
6905                 qlnxr_iw_qp_event(context, params, IB_EVENT_QP_FATAL,
6906                                  "IWARP_EVENT_REMOTE_OPERATION_ERROR");
6907                 break;
6908 
6909         case ECORE_IWARP_EVENT_TERMINATE_RECEIVED:
6910 		QL_DPRINT12(ha, "Got terminate message"
6911 			" ECORE_IWARP_EVENT_TERMINATE_RECEIVED\n");
6912                 break;
6913 
6914 	default:
6915 		QL_DPRINT12(ha,
6916 			"Unknown event [0x%x] received \n", params->event);
6917 		break;
6918 	};
6919 
6920 	QL_DPRINT12(ha, "[context, event] = [%p, 0x%x] "
6921 		"exit\n", context, params->event);
6922 	return 0;
6923 }
6924 
6925 static int
6926 qlnxr_addr4_resolve(struct qlnxr_dev *dev,
6927 			      struct sockaddr_in *src_in,
6928 			      struct sockaddr_in *dst_in,
6929 			      u8 *dst_mac)
6930 {
6931 	int rc;
6932 
6933 #if __FreeBSD_version >= 1100000
6934 	rc = arpresolve(dev->ha->ifp, 0, NULL, (struct sockaddr *)dst_in,
6935 			dst_mac, NULL, NULL);
6936 #else
6937 	struct llentry *lle;
6938 
6939 	rc = arpresolve(dev->ha->ifp, NULL, NULL, (struct sockaddr *)dst_in,
6940 			dst_mac, &lle);
6941 #endif
6942 
6943 	QL_DPRINT12(dev->ha, "rc = %d "
6944 		"sa_len = 0x%x sa_family = 0x%x IP Address = %d.%d.%d.%d "
6945 		"Dest MAC %02x:%02x:%02x:%02x:%02x:%02x\n", rc,
6946 		dst_in->sin_len, dst_in->sin_family,
6947 		NIPQUAD((dst_in->sin_addr.s_addr)),
6948 		dst_mac[0], dst_mac[1], dst_mac[2],
6949 		dst_mac[3], dst_mac[4], dst_mac[5]);
6950 
6951 	return rc;
6952 }
6953 
6954 int
6955 qlnxr_iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
6956 {
6957 	struct qlnxr_dev *dev;
6958 	struct ecore_iwarp_connect_out out_params;
6959 	struct ecore_iwarp_connect_in in_params;
6960 	struct qlnxr_iw_ep *ep;
6961 	struct qlnxr_qp *qp;
6962 	struct sockaddr_in *laddr;
6963 	struct sockaddr_in *raddr;
6964 	int rc = 0;
6965 	qlnx_host_t	*ha;
6966 
6967 	dev = get_qlnxr_dev((cm_id->device));
6968 	ha = dev->ha;
6969 
6970 	QL_DPRINT12(ha, "[cm_id, conn_param] = [%p, %p] "
6971 		"enter \n", cm_id, conn_param);
6972 
6973 	if (!(ha->ifp->if_drv_flags & IFF_DRV_RUNNING))
6974 		return -EINVAL;
6975 
6976 	qp = idr_find(&dev->qpidr, conn_param->qpn);
6977 
6978 	laddr = (struct sockaddr_in *)&cm_id->local_addr;
6979 	raddr = (struct sockaddr_in *)&cm_id->remote_addr;
6980 
6981 	QL_DPRINT12(ha,
6982 		"local = [%d.%d.%d.%d, %d] remote = [%d.%d.%d.%d, %d]\n",
6983 		NIPQUAD((laddr->sin_addr.s_addr)), laddr->sin_port,
6984 		NIPQUAD((raddr->sin_addr.s_addr)), raddr->sin_port);
6985 
6986 	ep = kzalloc(sizeof(*ep), GFP_KERNEL);
6987 	if (!ep) {
6988 		QL_DPRINT11(ha, "struct qlnxr_iw_ep "
6989 			"alloc memory failed\n");
6990 		return -ENOMEM;
6991 	}
6992 
6993 	ep->dev = dev;
6994 	ep->qp = qp;
6995 	cm_id->add_ref(cm_id);
6996 	ep->cm_id = cm_id;
6997 
6998 	memset(&in_params, 0, sizeof (struct ecore_iwarp_connect_in));
6999 	memset(&out_params, 0, sizeof (struct ecore_iwarp_connect_out));
7000 
7001 	in_params.event_cb = qlnxr_iw_event_handler;
7002 	in_params.cb_context = ep;
7003 
7004 	in_params.cm_info.ip_version = ECORE_TCP_IPV4;
7005 
7006 	in_params.cm_info.remote_ip[0] = ntohl(raddr->sin_addr.s_addr);
7007 	in_params.cm_info.local_ip[0] = ntohl(laddr->sin_addr.s_addr);
7008 	in_params.cm_info.remote_port = ntohs(raddr->sin_port);
7009 	in_params.cm_info.local_port = ntohs(laddr->sin_port);
7010 	in_params.cm_info.vlan = 0;
7011 	in_params.mss = dev->ha->ifp->if_mtu - 40;
7012 
7013 	QL_DPRINT12(ha, "remote_ip = [%d.%d.%d.%d] "
7014 		"local_ip = [%d.%d.%d.%d] remote_port = %d local_port = %d "
7015 		"vlan = %d\n",
7016 		NIPQUAD((in_params.cm_info.remote_ip[0])),
7017 		NIPQUAD((in_params.cm_info.local_ip[0])),
7018 		in_params.cm_info.remote_port, in_params.cm_info.local_port,
7019 		in_params.cm_info.vlan);
7020 
7021 	rc = qlnxr_addr4_resolve(dev, laddr, raddr, (u8 *)in_params.remote_mac_addr);
7022 
7023 	if (rc) {
7024 		QL_DPRINT11(ha, "qlnxr_addr4_resolve failed\n");
7025 		goto err;
7026 	}
7027 
7028 	QL_DPRINT12(ha, "ord = %d ird=%d private_data=%p"
7029 		" private_data_len=%d rq_psn=%d\n",
7030 		conn_param->ord, conn_param->ird, conn_param->private_data,
7031 		conn_param->private_data_len, qp->rq_psn);
7032 
7033 	in_params.cm_info.ord = conn_param->ord;
7034 	in_params.cm_info.ird = conn_param->ird;
7035 	in_params.cm_info.private_data = conn_param->private_data;
7036 	in_params.cm_info.private_data_len = conn_param->private_data_len;
7037 	in_params.qp = qp->ecore_qp;
7038 
7039 	memcpy(in_params.local_mac_addr, dev->ha->primary_mac, ETH_ALEN);
7040 
7041 	rc = ecore_iwarp_connect(dev->rdma_ctx, &in_params, &out_params);
7042 
7043 	if (rc) {
7044 		QL_DPRINT12(ha, "ecore_iwarp_connect failed\n");
7045 		goto err;
7046 	}
7047 
7048 	QL_DPRINT12(ha, "exit\n");
7049 
7050 	return rc;
7051 
7052 err:
7053 	cm_id->rem_ref(cm_id);
7054 	kfree(ep);
7055 
7056 	QL_DPRINT12(ha, "exit [%d]\n", rc);
7057 	return rc;
7058 }
7059 
7060 int
7061 qlnxr_iw_create_listen(struct iw_cm_id *cm_id, int backlog)
7062 {
7063 	struct qlnxr_dev *dev;
7064 	struct qlnxr_iw_listener *listener;
7065 	struct ecore_iwarp_listen_in iparams;
7066 	struct ecore_iwarp_listen_out oparams;
7067 	struct sockaddr_in *laddr;
7068 	qlnx_host_t	*ha;
7069 	int rc;
7070 
7071 	dev = get_qlnxr_dev((cm_id->device));
7072 	ha = dev->ha;
7073 
7074 	QL_DPRINT12(ha, "enter\n");
7075 
7076 	if (!(ha->ifp->if_drv_flags & IFF_DRV_RUNNING))
7077 		return -EINVAL;
7078 
7079 	laddr = (struct sockaddr_in *)&cm_id->local_addr;
7080 
7081 	listener = kzalloc(sizeof(*listener), GFP_KERNEL);
7082 
7083 	if (listener == NULL) {
7084 		QL_DPRINT11(ha, "listener memory alloc failed\n");
7085 		return -ENOMEM;
7086 	}
7087 
7088 	listener->dev = dev;
7089 	cm_id->add_ref(cm_id);
7090 	listener->cm_id = cm_id;
7091 	listener->backlog = backlog;
7092 
7093 	memset(&iparams, 0, sizeof (struct ecore_iwarp_listen_in));
7094 	memset(&oparams, 0, sizeof (struct ecore_iwarp_listen_out));
7095 
7096 	iparams.cb_context = listener;
7097 	iparams.event_cb = qlnxr_iw_event_handler;
7098 	iparams.max_backlog = backlog;
7099 
7100 	iparams.ip_version = ECORE_TCP_IPV4;
7101 
7102 	iparams.ip_addr[0] = ntohl(laddr->sin_addr.s_addr);
7103 	iparams.port = ntohs(laddr->sin_port);
7104 	iparams.vlan = 0;
7105 
7106 	QL_DPRINT12(ha, "[%d.%d.%d.%d, %d] iparamsport=%d\n",
7107 		NIPQUAD((laddr->sin_addr.s_addr)),
7108 		laddr->sin_port, iparams.port);
7109 
7110 	rc = ecore_iwarp_create_listen(dev->rdma_ctx, &iparams, &oparams);
7111 	if (rc) {
7112 		QL_DPRINT11(ha,
7113 			"ecore_iwarp_create_listen failed rc = %d\n", rc);
7114 		goto err;
7115 	}
7116 
7117 	listener->ecore_handle = oparams.handle;
7118 	cm_id->provider_data = listener;
7119 
7120 	QL_DPRINT12(ha, "exit\n");
7121 	return rc;
7122 
7123 err:
7124 	cm_id->rem_ref(cm_id);
7125 	kfree(listener);
7126 
7127 	QL_DPRINT12(ha, "exit [%d]\n", rc);
7128 	return rc;
7129 }
7130 
7131 void
7132 qlnxr_iw_destroy_listen(struct iw_cm_id *cm_id)
7133 {
7134 	struct qlnxr_iw_listener *listener = cm_id->provider_data;
7135 	struct qlnxr_dev *dev = get_qlnxr_dev((cm_id->device));
7136 	int rc = 0;
7137 	qlnx_host_t	*ha;
7138 
7139 	ha = dev->ha;
7140 
7141 	QL_DPRINT12(ha, "enter\n");
7142 
7143 	if (listener->ecore_handle)
7144 		rc = ecore_iwarp_destroy_listen(dev->rdma_ctx,
7145 				listener->ecore_handle);
7146 
7147 	cm_id->rem_ref(cm_id);
7148 
7149 	QL_DPRINT12(ha, "exit [%d]\n", rc);
7150 	return;
7151 }
7152 
7153 int
7154 qlnxr_iw_accept(struct iw_cm_id *cm_id,
7155 	struct iw_cm_conn_param *conn_param)
7156 {
7157 	struct qlnxr_iw_ep *ep = (struct qlnxr_iw_ep *)cm_id->provider_data;
7158 	struct qlnxr_dev *dev = ep->dev;
7159 	struct qlnxr_qp *qp;
7160 	struct ecore_iwarp_accept_in params;
7161 	int rc;
7162 	qlnx_host_t	*ha;
7163 
7164 	ha = dev->ha;
7165 
7166 	QL_DPRINT12(ha, "enter  qpid=%d\n", conn_param->qpn);
7167 
7168 	if (!(ha->ifp->if_drv_flags & IFF_DRV_RUNNING))
7169 		return -EINVAL;
7170 
7171 	qp = idr_find(&dev->qpidr, conn_param->qpn);
7172 	if (!qp) {
7173 		QL_DPRINT11(ha, "idr_find failed invalid qpn = %d\n",
7174 			conn_param->qpn);
7175 		return -EINVAL;
7176 	}
7177 	ep->qp = qp;
7178 	qp->ep = ep;
7179 	cm_id->add_ref(cm_id);
7180 	ep->cm_id = cm_id;
7181 
7182 	params.ep_context = ep->ecore_context;
7183 	params.cb_context = ep;
7184 	params.qp = ep->qp->ecore_qp;
7185 	params.private_data = conn_param->private_data;
7186 	params.private_data_len = conn_param->private_data_len;
7187 	params.ird = conn_param->ird;
7188 	params.ord = conn_param->ord;
7189 
7190 	rc = ecore_iwarp_accept(dev->rdma_ctx, &params);
7191 	if (rc) {
7192 		QL_DPRINT11(ha, "ecore_iwarp_accept failed %d\n", rc);
7193 		goto err;
7194 	}
7195 
7196 	QL_DPRINT12(ha, "exit\n");
7197 	return 0;
7198 err:
7199 	cm_id->rem_ref(cm_id);
7200 	QL_DPRINT12(ha, "exit rc = %d\n", rc);
7201 	return rc;
7202 }
7203 
7204 int
7205 qlnxr_iw_reject(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
7206 {
7207 #if __FreeBSD_version >= 1102000
7208 
7209         struct qlnxr_iw_ep *ep = (struct qlnxr_iw_ep *)cm_id->provider_data;
7210         struct qlnxr_dev *dev = ep->dev;
7211         struct ecore_iwarp_reject_in params;
7212         int rc;
7213 
7214         params.ep_context = ep->ecore_context;
7215         params.cb_context = ep;
7216         params.private_data = pdata;
7217         params.private_data_len = pdata_len;
7218         ep->qp = NULL;
7219 
7220         rc = ecore_iwarp_reject(dev->rdma_ctx, &params);
7221 
7222         return rc;
7223 
7224 #else
7225 
7226 	printf("iWARP reject_cr not implemented\n");
7227 	return -EINVAL;
7228 
7229 #endif /* #if __FreeBSD_version >= 1102000 */
7230 }
7231 
7232 void
7233 qlnxr_iw_qp_add_ref(struct ib_qp *ibqp)
7234 {
7235 	struct qlnxr_qp *qp = get_qlnxr_qp(ibqp);
7236 	qlnx_host_t	*ha;
7237 
7238 	ha = qp->dev->ha;
7239 
7240 	QL_DPRINT12(ha, "enter ibqp = %p\n", ibqp);
7241 
7242 	atomic_inc(&qp->refcnt);
7243 
7244 	QL_DPRINT12(ha, "exit \n");
7245 	return;
7246 }
7247 
7248 void
7249 qlnxr_iw_qp_rem_ref(struct ib_qp *ibqp)
7250 {
7251 	struct qlnxr_qp *qp = get_qlnxr_qp(ibqp);
7252 	qlnx_host_t	*ha;
7253 
7254 	ha = qp->dev->ha;
7255 
7256 	QL_DPRINT12(ha, "enter ibqp = %p qp = %p\n", ibqp, qp);
7257 
7258 	if (atomic_dec_and_test(&qp->refcnt)) {
7259 		qlnxr_idr_remove(qp->dev, qp->qp_id);
7260 		kfree(qp);
7261 	}
7262 
7263 	QL_DPRINT12(ha, "exit \n");
7264 	return;
7265 }
7266 
7267 struct ib_qp *
7268 qlnxr_iw_get_qp(struct ib_device *ibdev, int qpn)
7269 {
7270 	struct qlnxr_dev *dev = get_qlnxr_dev(ibdev);
7271 	struct ib_qp *qp;
7272 	qlnx_host_t	*ha;
7273 
7274 	ha = dev->ha;
7275 
7276 	QL_DPRINT12(ha, "enter dev = %p ibdev = %p qpn = %d\n", dev, ibdev, qpn);
7277 
7278 	qp = idr_find(&dev->qpidr, qpn);
7279 
7280 	QL_DPRINT12(ha, "exit qp = %p\n", qp);
7281 
7282 	return (qp);
7283 }
7284