xref: /linux/drivers/infiniband/hw/hns/hns_roce_hw_v2.c (revision 172cdcaefea5c297fdb3d20b7d5aff60ae4fbce6)
1 /*
2  * Copyright (c) 2016-2017 Hisilicon Limited.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #include <linux/acpi.h>
34 #include <linux/etherdevice.h>
35 #include <linux/interrupt.h>
36 #include <linux/kernel.h>
37 #include <linux/types.h>
38 #include <net/addrconf.h>
39 #include <rdma/ib_addr.h>
40 #include <rdma/ib_cache.h>
41 #include <rdma/ib_umem.h>
42 #include <rdma/uverbs_ioctl.h>
43 
44 #include "hnae3.h"
45 #include "hns_roce_common.h"
46 #include "hns_roce_device.h"
47 #include "hns_roce_cmd.h"
48 #include "hns_roce_hem.h"
49 #include "hns_roce_hw_v2.h"
50 
51 enum {
52 	CMD_RST_PRC_OTHERS,
53 	CMD_RST_PRC_SUCCESS,
54 	CMD_RST_PRC_EBUSY,
55 };
56 
57 static inline void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
58 				   struct ib_sge *sg)
59 {
60 	dseg->lkey = cpu_to_le32(sg->lkey);
61 	dseg->addr = cpu_to_le64(sg->addr);
62 	dseg->len  = cpu_to_le32(sg->length);
63 }
64 
65 /*
66  * mapped-value = 1 + real-value
67  * The hns wr opcode real value is start from 0, In order to distinguish between
68  * initialized and uninitialized map values, we plus 1 to the actual value when
69  * defining the mapping, so that the validity can be identified by checking the
70  * mapped value is greater than 0.
71  */
72 #define HR_OPC_MAP(ib_key, hr_key) \
73 		[IB_WR_ ## ib_key] = 1 + HNS_ROCE_V2_WQE_OP_ ## hr_key
74 
75 static const u32 hns_roce_op_code[] = {
76 	HR_OPC_MAP(RDMA_WRITE,			RDMA_WRITE),
77 	HR_OPC_MAP(RDMA_WRITE_WITH_IMM,		RDMA_WRITE_WITH_IMM),
78 	HR_OPC_MAP(SEND,			SEND),
79 	HR_OPC_MAP(SEND_WITH_IMM,		SEND_WITH_IMM),
80 	HR_OPC_MAP(RDMA_READ,			RDMA_READ),
81 	HR_OPC_MAP(ATOMIC_CMP_AND_SWP,		ATOM_CMP_AND_SWAP),
82 	HR_OPC_MAP(ATOMIC_FETCH_AND_ADD,	ATOM_FETCH_AND_ADD),
83 	HR_OPC_MAP(SEND_WITH_INV,		SEND_WITH_INV),
84 	HR_OPC_MAP(LOCAL_INV,			LOCAL_INV),
85 	HR_OPC_MAP(MASKED_ATOMIC_CMP_AND_SWP,	ATOM_MSK_CMP_AND_SWAP),
86 	HR_OPC_MAP(MASKED_ATOMIC_FETCH_AND_ADD,	ATOM_MSK_FETCH_AND_ADD),
87 	HR_OPC_MAP(REG_MR,			FAST_REG_PMR),
88 };
89 
90 static u32 to_hr_opcode(u32 ib_opcode)
91 {
92 	if (ib_opcode >= ARRAY_SIZE(hns_roce_op_code))
93 		return HNS_ROCE_V2_WQE_OP_MASK;
94 
95 	return hns_roce_op_code[ib_opcode] ? hns_roce_op_code[ib_opcode] - 1 :
96 					     HNS_ROCE_V2_WQE_OP_MASK;
97 }
98 
99 static void set_frmr_seg(struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
100 			 const struct ib_reg_wr *wr)
101 {
102 	struct hns_roce_wqe_frmr_seg *fseg =
103 		(void *)rc_sq_wqe + sizeof(struct hns_roce_v2_rc_send_wqe);
104 	struct hns_roce_mr *mr = to_hr_mr(wr->mr);
105 	u64 pbl_ba;
106 
107 	/* use ib_access_flags */
108 	roce_set_bit(fseg->byte_40, V2_RC_FRMR_WQE_BYTE_40_BIND_EN_S,
109 		     !!(wr->access & IB_ACCESS_MW_BIND));
110 	roce_set_bit(fseg->byte_40, V2_RC_FRMR_WQE_BYTE_40_ATOMIC_S,
111 		     !!(wr->access & IB_ACCESS_REMOTE_ATOMIC));
112 	roce_set_bit(fseg->byte_40, V2_RC_FRMR_WQE_BYTE_40_RR_S,
113 		     !!(wr->access & IB_ACCESS_REMOTE_READ));
114 	roce_set_bit(fseg->byte_40, V2_RC_FRMR_WQE_BYTE_40_RW_S,
115 		     !!(wr->access & IB_ACCESS_REMOTE_WRITE));
116 	roce_set_bit(fseg->byte_40, V2_RC_FRMR_WQE_BYTE_40_LW_S,
117 		     !!(wr->access & IB_ACCESS_LOCAL_WRITE));
118 
119 	/* Data structure reuse may lead to confusion */
120 	pbl_ba = mr->pbl_mtr.hem_cfg.root_ba;
121 	rc_sq_wqe->msg_len = cpu_to_le32(lower_32_bits(pbl_ba));
122 	rc_sq_wqe->inv_key = cpu_to_le32(upper_32_bits(pbl_ba));
123 
124 	rc_sq_wqe->byte_16 = cpu_to_le32(wr->mr->length & 0xffffffff);
125 	rc_sq_wqe->byte_20 = cpu_to_le32(wr->mr->length >> 32);
126 	rc_sq_wqe->rkey = cpu_to_le32(wr->key);
127 	rc_sq_wqe->va = cpu_to_le64(wr->mr->iova);
128 
129 	fseg->pbl_size = cpu_to_le32(mr->npages);
130 	roce_set_field(fseg->byte_40, V2_RC_FRMR_WQE_BYTE_40_PBL_BUF_PG_SZ_M,
131 		       V2_RC_FRMR_WQE_BYTE_40_PBL_BUF_PG_SZ_S,
132 		       to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.buf_pg_shift));
133 	roce_set_bit(fseg->byte_40, V2_RC_FRMR_WQE_BYTE_40_BLK_MODE_S, 0);
134 }
135 
136 static void set_atomic_seg(const struct ib_send_wr *wr,
137 			   struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
138 			   unsigned int valid_num_sge)
139 {
140 	struct hns_roce_v2_wqe_data_seg *dseg =
141 		(void *)rc_sq_wqe + sizeof(struct hns_roce_v2_rc_send_wqe);
142 	struct hns_roce_wqe_atomic_seg *aseg =
143 		(void *)dseg + sizeof(struct hns_roce_v2_wqe_data_seg);
144 
145 	set_data_seg_v2(dseg, wr->sg_list);
146 
147 	if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
148 		aseg->fetchadd_swap_data = cpu_to_le64(atomic_wr(wr)->swap);
149 		aseg->cmp_data = cpu_to_le64(atomic_wr(wr)->compare_add);
150 	} else {
151 		aseg->fetchadd_swap_data =
152 			cpu_to_le64(atomic_wr(wr)->compare_add);
153 		aseg->cmp_data = 0;
154 	}
155 
156 	roce_set_field(rc_sq_wqe->byte_16, V2_RC_SEND_WQE_BYTE_16_SGE_NUM_M,
157 		       V2_RC_SEND_WQE_BYTE_16_SGE_NUM_S, valid_num_sge);
158 }
159 
160 static int fill_ext_sge_inl_data(struct hns_roce_qp *qp,
161 				 const struct ib_send_wr *wr,
162 				 unsigned int *sge_idx, u32 msg_len)
163 {
164 	struct ib_device *ibdev = &(to_hr_dev(qp->ibqp.device))->ib_dev;
165 	unsigned int dseg_len = sizeof(struct hns_roce_v2_wqe_data_seg);
166 	unsigned int ext_sge_sz = qp->sq.max_gs * dseg_len;
167 	unsigned int left_len_in_pg;
168 	unsigned int idx = *sge_idx;
169 	unsigned int i = 0;
170 	unsigned int len;
171 	void *addr;
172 	void *dseg;
173 
174 	if (msg_len > ext_sge_sz) {
175 		ibdev_err(ibdev,
176 			  "no enough extended sge space for inline data.\n");
177 		return -EINVAL;
178 	}
179 
180 	dseg = hns_roce_get_extend_sge(qp, idx & (qp->sge.sge_cnt - 1));
181 	left_len_in_pg = hr_hw_page_align((uintptr_t)dseg) - (uintptr_t)dseg;
182 	len = wr->sg_list[0].length;
183 	addr = (void *)(unsigned long)(wr->sg_list[0].addr);
184 
185 	/* When copying data to extended sge space, the left length in page may
186 	 * not long enough for current user's sge. So the data should be
187 	 * splited into several parts, one in the first page, and the others in
188 	 * the subsequent pages.
189 	 */
190 	while (1) {
191 		if (len <= left_len_in_pg) {
192 			memcpy(dseg, addr, len);
193 
194 			idx += len / dseg_len;
195 
196 			i++;
197 			if (i >= wr->num_sge)
198 				break;
199 
200 			left_len_in_pg -= len;
201 			len = wr->sg_list[i].length;
202 			addr = (void *)(unsigned long)(wr->sg_list[i].addr);
203 			dseg += len;
204 		} else {
205 			memcpy(dseg, addr, left_len_in_pg);
206 
207 			len -= left_len_in_pg;
208 			addr += left_len_in_pg;
209 			idx += left_len_in_pg / dseg_len;
210 			dseg = hns_roce_get_extend_sge(qp,
211 						idx & (qp->sge.sge_cnt - 1));
212 			left_len_in_pg = 1 << HNS_HW_PAGE_SHIFT;
213 		}
214 	}
215 
216 	*sge_idx = idx;
217 
218 	return 0;
219 }
220 
221 static void set_extend_sge(struct hns_roce_qp *qp, struct ib_sge *sge,
222 			   unsigned int *sge_ind, unsigned int cnt)
223 {
224 	struct hns_roce_v2_wqe_data_seg *dseg;
225 	unsigned int idx = *sge_ind;
226 
227 	while (cnt > 0) {
228 		dseg = hns_roce_get_extend_sge(qp, idx & (qp->sge.sge_cnt - 1));
229 		if (likely(sge->length)) {
230 			set_data_seg_v2(dseg, sge);
231 			idx++;
232 			cnt--;
233 		}
234 		sge++;
235 	}
236 
237 	*sge_ind = idx;
238 }
239 
240 static bool check_inl_data_len(struct hns_roce_qp *qp, unsigned int len)
241 {
242 	struct hns_roce_dev *hr_dev = to_hr_dev(qp->ibqp.device);
243 	int mtu = ib_mtu_enum_to_int(qp->path_mtu);
244 
245 	if (len > qp->max_inline_data || len > mtu) {
246 		ibdev_err(&hr_dev->ib_dev,
247 			  "invalid length of data, data len = %u, max inline len = %u, path mtu = %d.\n",
248 			  len, qp->max_inline_data, mtu);
249 		return false;
250 	}
251 
252 	return true;
253 }
254 
255 static int set_rc_inl(struct hns_roce_qp *qp, const struct ib_send_wr *wr,
256 		      struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
257 		      unsigned int *sge_idx)
258 {
259 	struct hns_roce_dev *hr_dev = to_hr_dev(qp->ibqp.device);
260 	u32 msg_len = le32_to_cpu(rc_sq_wqe->msg_len);
261 	struct ib_device *ibdev = &hr_dev->ib_dev;
262 	unsigned int curr_idx = *sge_idx;
263 	void *dseg = rc_sq_wqe;
264 	unsigned int i;
265 	int ret;
266 
267 	if (unlikely(wr->opcode == IB_WR_RDMA_READ)) {
268 		ibdev_err(ibdev, "invalid inline parameters!\n");
269 		return -EINVAL;
270 	}
271 
272 	if (!check_inl_data_len(qp, msg_len))
273 		return -EINVAL;
274 
275 	dseg += sizeof(struct hns_roce_v2_rc_send_wqe);
276 
277 	roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_INLINE_S, 1);
278 
279 	if (msg_len <= HNS_ROCE_V2_MAX_RC_INL_INN_SZ) {
280 		roce_set_bit(rc_sq_wqe->byte_20,
281 			     V2_RC_SEND_WQE_BYTE_20_INL_TYPE_S, 0);
282 
283 		for (i = 0; i < wr->num_sge; i++) {
284 			memcpy(dseg, ((void *)wr->sg_list[i].addr),
285 			       wr->sg_list[i].length);
286 			dseg += wr->sg_list[i].length;
287 		}
288 	} else {
289 		roce_set_bit(rc_sq_wqe->byte_20,
290 			     V2_RC_SEND_WQE_BYTE_20_INL_TYPE_S, 1);
291 
292 		ret = fill_ext_sge_inl_data(qp, wr, &curr_idx, msg_len);
293 		if (ret)
294 			return ret;
295 
296 		roce_set_field(rc_sq_wqe->byte_16,
297 			       V2_RC_SEND_WQE_BYTE_16_SGE_NUM_M,
298 			       V2_RC_SEND_WQE_BYTE_16_SGE_NUM_S,
299 			       curr_idx - *sge_idx);
300 	}
301 
302 	*sge_idx = curr_idx;
303 
304 	return 0;
305 }
306 
307 static int set_rwqe_data_seg(struct ib_qp *ibqp, const struct ib_send_wr *wr,
308 			     struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
309 			     unsigned int *sge_ind,
310 			     unsigned int valid_num_sge)
311 {
312 	struct hns_roce_v2_wqe_data_seg *dseg =
313 		(void *)rc_sq_wqe + sizeof(struct hns_roce_v2_rc_send_wqe);
314 	struct hns_roce_qp *qp = to_hr_qp(ibqp);
315 	int j = 0;
316 	int i;
317 
318 	roce_set_field(rc_sq_wqe->byte_20,
319 		       V2_RC_SEND_WQE_BYTE_20_MSG_START_SGE_IDX_M,
320 		       V2_RC_SEND_WQE_BYTE_20_MSG_START_SGE_IDX_S,
321 		       (*sge_ind) & (qp->sge.sge_cnt - 1));
322 
323 	if (wr->send_flags & IB_SEND_INLINE)
324 		return set_rc_inl(qp, wr, rc_sq_wqe, sge_ind);
325 
326 	if (valid_num_sge <= HNS_ROCE_SGE_IN_WQE) {
327 		for (i = 0; i < wr->num_sge; i++) {
328 			if (likely(wr->sg_list[i].length)) {
329 				set_data_seg_v2(dseg, wr->sg_list + i);
330 				dseg++;
331 			}
332 		}
333 	} else {
334 		for (i = 0; i < wr->num_sge && j < HNS_ROCE_SGE_IN_WQE; i++) {
335 			if (likely(wr->sg_list[i].length)) {
336 				set_data_seg_v2(dseg, wr->sg_list + i);
337 				dseg++;
338 				j++;
339 			}
340 		}
341 
342 		set_extend_sge(qp, wr->sg_list + i, sge_ind,
343 			       valid_num_sge - HNS_ROCE_SGE_IN_WQE);
344 	}
345 
346 	roce_set_field(rc_sq_wqe->byte_16,
347 		       V2_RC_SEND_WQE_BYTE_16_SGE_NUM_M,
348 		       V2_RC_SEND_WQE_BYTE_16_SGE_NUM_S, valid_num_sge);
349 
350 	return 0;
351 }
352 
353 static int check_send_valid(struct hns_roce_dev *hr_dev,
354 			    struct hns_roce_qp *hr_qp)
355 {
356 	struct ib_device *ibdev = &hr_dev->ib_dev;
357 	struct ib_qp *ibqp = &hr_qp->ibqp;
358 
359 	if (unlikely(ibqp->qp_type != IB_QPT_RC &&
360 		     ibqp->qp_type != IB_QPT_GSI &&
361 		     ibqp->qp_type != IB_QPT_UD)) {
362 		ibdev_err(ibdev, "Not supported QP(0x%x)type!\n",
363 			  ibqp->qp_type);
364 		return -EOPNOTSUPP;
365 	} else if (unlikely(hr_qp->state == IB_QPS_RESET ||
366 		   hr_qp->state == IB_QPS_INIT ||
367 		   hr_qp->state == IB_QPS_RTR)) {
368 		ibdev_err(ibdev, "failed to post WQE, QP state %u!\n",
369 			  hr_qp->state);
370 		return -EINVAL;
371 	} else if (unlikely(hr_dev->state >= HNS_ROCE_DEVICE_STATE_RST_DOWN)) {
372 		ibdev_err(ibdev, "failed to post WQE, dev state %d!\n",
373 			  hr_dev->state);
374 		return -EIO;
375 	}
376 
377 	return 0;
378 }
379 
380 static unsigned int calc_wr_sge_num(const struct ib_send_wr *wr,
381 				    unsigned int *sge_len)
382 {
383 	unsigned int valid_num = 0;
384 	unsigned int len = 0;
385 	int i;
386 
387 	for (i = 0; i < wr->num_sge; i++) {
388 		if (likely(wr->sg_list[i].length)) {
389 			len += wr->sg_list[i].length;
390 			valid_num++;
391 		}
392 	}
393 
394 	*sge_len = len;
395 	return valid_num;
396 }
397 
398 static __le32 get_immtdata(const struct ib_send_wr *wr)
399 {
400 	switch (wr->opcode) {
401 	case IB_WR_SEND_WITH_IMM:
402 	case IB_WR_RDMA_WRITE_WITH_IMM:
403 		return cpu_to_le32(be32_to_cpu(wr->ex.imm_data));
404 	default:
405 		return 0;
406 	}
407 }
408 
409 static int set_ud_opcode(struct hns_roce_v2_ud_send_wqe *ud_sq_wqe,
410 			 const struct ib_send_wr *wr)
411 {
412 	u32 ib_op = wr->opcode;
413 
414 	if (ib_op != IB_WR_SEND && ib_op != IB_WR_SEND_WITH_IMM)
415 		return -EINVAL;
416 
417 	ud_sq_wqe->immtdata = get_immtdata(wr);
418 
419 	roce_set_field(ud_sq_wqe->byte_4, V2_UD_SEND_WQE_BYTE_4_OPCODE_M,
420 		       V2_UD_SEND_WQE_BYTE_4_OPCODE_S, to_hr_opcode(ib_op));
421 
422 	return 0;
423 }
424 
425 static int fill_ud_av(struct hns_roce_v2_ud_send_wqe *ud_sq_wqe,
426 		      struct hns_roce_ah *ah)
427 {
428 	struct ib_device *ib_dev = ah->ibah.device;
429 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
430 
431 	roce_set_field(ud_sq_wqe->byte_24, V2_UD_SEND_WQE_BYTE_24_UDPSPN_M,
432 		       V2_UD_SEND_WQE_BYTE_24_UDPSPN_S, ah->av.udp_sport);
433 
434 	roce_set_field(ud_sq_wqe->byte_36, V2_UD_SEND_WQE_BYTE_36_HOPLIMIT_M,
435 		       V2_UD_SEND_WQE_BYTE_36_HOPLIMIT_S, ah->av.hop_limit);
436 	roce_set_field(ud_sq_wqe->byte_36, V2_UD_SEND_WQE_BYTE_36_TCLASS_M,
437 		       V2_UD_SEND_WQE_BYTE_36_TCLASS_S, ah->av.tclass);
438 	roce_set_field(ud_sq_wqe->byte_40, V2_UD_SEND_WQE_BYTE_40_FLOW_LABEL_M,
439 		       V2_UD_SEND_WQE_BYTE_40_FLOW_LABEL_S, ah->av.flowlabel);
440 
441 	if (WARN_ON(ah->av.sl > MAX_SERVICE_LEVEL))
442 		return -EINVAL;
443 
444 	roce_set_field(ud_sq_wqe->byte_40, V2_UD_SEND_WQE_BYTE_40_SL_M,
445 		       V2_UD_SEND_WQE_BYTE_40_SL_S, ah->av.sl);
446 
447 	ud_sq_wqe->sgid_index = ah->av.gid_index;
448 
449 	memcpy(ud_sq_wqe->dmac, ah->av.mac, ETH_ALEN);
450 	memcpy(ud_sq_wqe->dgid, ah->av.dgid, GID_LEN_V2);
451 
452 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
453 		return 0;
454 
455 	roce_set_bit(ud_sq_wqe->byte_40, V2_UD_SEND_WQE_BYTE_40_UD_VLAN_EN_S,
456 		     ah->av.vlan_en);
457 	roce_set_field(ud_sq_wqe->byte_36, V2_UD_SEND_WQE_BYTE_36_VLAN_M,
458 		       V2_UD_SEND_WQE_BYTE_36_VLAN_S, ah->av.vlan_id);
459 
460 	return 0;
461 }
462 
463 static inline int set_ud_wqe(struct hns_roce_qp *qp,
464 			     const struct ib_send_wr *wr,
465 			     void *wqe, unsigned int *sge_idx,
466 			     unsigned int owner_bit)
467 {
468 	struct hns_roce_ah *ah = to_hr_ah(ud_wr(wr)->ah);
469 	struct hns_roce_v2_ud_send_wqe *ud_sq_wqe = wqe;
470 	unsigned int curr_idx = *sge_idx;
471 	unsigned int valid_num_sge;
472 	u32 msg_len = 0;
473 	int ret;
474 
475 	valid_num_sge = calc_wr_sge_num(wr, &msg_len);
476 
477 	ret = set_ud_opcode(ud_sq_wqe, wr);
478 	if (WARN_ON(ret))
479 		return ret;
480 
481 	ud_sq_wqe->msg_len = cpu_to_le32(msg_len);
482 
483 	roce_set_bit(ud_sq_wqe->byte_4, V2_UD_SEND_WQE_BYTE_4_CQE_S,
484 		     !!(wr->send_flags & IB_SEND_SIGNALED));
485 
486 	roce_set_bit(ud_sq_wqe->byte_4, V2_UD_SEND_WQE_BYTE_4_SE_S,
487 		     !!(wr->send_flags & IB_SEND_SOLICITED));
488 
489 	roce_set_field(ud_sq_wqe->byte_16, V2_UD_SEND_WQE_BYTE_16_PD_M,
490 		       V2_UD_SEND_WQE_BYTE_16_PD_S, to_hr_pd(qp->ibqp.pd)->pdn);
491 
492 	roce_set_field(ud_sq_wqe->byte_16, V2_UD_SEND_WQE_BYTE_16_SGE_NUM_M,
493 		       V2_UD_SEND_WQE_BYTE_16_SGE_NUM_S, valid_num_sge);
494 
495 	roce_set_field(ud_sq_wqe->byte_20,
496 		       V2_UD_SEND_WQE_BYTE_20_MSG_START_SGE_IDX_M,
497 		       V2_UD_SEND_WQE_BYTE_20_MSG_START_SGE_IDX_S,
498 		       curr_idx & (qp->sge.sge_cnt - 1));
499 
500 	ud_sq_wqe->qkey = cpu_to_le32(ud_wr(wr)->remote_qkey & 0x80000000 ?
501 			  qp->qkey : ud_wr(wr)->remote_qkey);
502 	roce_set_field(ud_sq_wqe->byte_32, V2_UD_SEND_WQE_BYTE_32_DQPN_M,
503 		       V2_UD_SEND_WQE_BYTE_32_DQPN_S, ud_wr(wr)->remote_qpn);
504 
505 	ret = fill_ud_av(ud_sq_wqe, ah);
506 	if (ret)
507 		return ret;
508 
509 	qp->sl = to_hr_ah(ud_wr(wr)->ah)->av.sl;
510 
511 	set_extend_sge(qp, wr->sg_list, &curr_idx, valid_num_sge);
512 
513 	/*
514 	 * The pipeline can sequentially post all valid WQEs into WQ buffer,
515 	 * including new WQEs waiting for the doorbell to update the PI again.
516 	 * Therefore, the owner bit of WQE MUST be updated after all fields
517 	 * and extSGEs have been written into DDR instead of cache.
518 	 */
519 	if (qp->en_flags & HNS_ROCE_QP_CAP_OWNER_DB)
520 		dma_wmb();
521 
522 	*sge_idx = curr_idx;
523 	roce_set_bit(ud_sq_wqe->byte_4, V2_UD_SEND_WQE_BYTE_4_OWNER_S,
524 		     owner_bit);
525 
526 	return 0;
527 }
528 
529 static int set_rc_opcode(struct hns_roce_dev *hr_dev,
530 			 struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
531 			 const struct ib_send_wr *wr)
532 {
533 	u32 ib_op = wr->opcode;
534 	int ret = 0;
535 
536 	rc_sq_wqe->immtdata = get_immtdata(wr);
537 
538 	switch (ib_op) {
539 	case IB_WR_RDMA_READ:
540 	case IB_WR_RDMA_WRITE:
541 	case IB_WR_RDMA_WRITE_WITH_IMM:
542 		rc_sq_wqe->rkey = cpu_to_le32(rdma_wr(wr)->rkey);
543 		rc_sq_wqe->va = cpu_to_le64(rdma_wr(wr)->remote_addr);
544 		break;
545 	case IB_WR_SEND:
546 	case IB_WR_SEND_WITH_IMM:
547 		break;
548 	case IB_WR_ATOMIC_CMP_AND_SWP:
549 	case IB_WR_ATOMIC_FETCH_AND_ADD:
550 		rc_sq_wqe->rkey = cpu_to_le32(atomic_wr(wr)->rkey);
551 		rc_sq_wqe->va = cpu_to_le64(atomic_wr(wr)->remote_addr);
552 		break;
553 	case IB_WR_REG_MR:
554 		if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
555 			set_frmr_seg(rc_sq_wqe, reg_wr(wr));
556 		else
557 			ret = -EOPNOTSUPP;
558 		break;
559 	case IB_WR_LOCAL_INV:
560 		roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_SO_S, 1);
561 		fallthrough;
562 	case IB_WR_SEND_WITH_INV:
563 		rc_sq_wqe->inv_key = cpu_to_le32(wr->ex.invalidate_rkey);
564 		break;
565 	default:
566 		ret = -EINVAL;
567 	}
568 
569 	if (unlikely(ret))
570 		return ret;
571 
572 	roce_set_field(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
573 		       V2_RC_SEND_WQE_BYTE_4_OPCODE_S, to_hr_opcode(ib_op));
574 
575 	return ret;
576 }
577 static inline int set_rc_wqe(struct hns_roce_qp *qp,
578 			     const struct ib_send_wr *wr,
579 			     void *wqe, unsigned int *sge_idx,
580 			     unsigned int owner_bit)
581 {
582 	struct hns_roce_dev *hr_dev = to_hr_dev(qp->ibqp.device);
583 	struct hns_roce_v2_rc_send_wqe *rc_sq_wqe = wqe;
584 	unsigned int curr_idx = *sge_idx;
585 	unsigned int valid_num_sge;
586 	u32 msg_len = 0;
587 	int ret;
588 
589 	valid_num_sge = calc_wr_sge_num(wr, &msg_len);
590 
591 	rc_sq_wqe->msg_len = cpu_to_le32(msg_len);
592 
593 	ret = set_rc_opcode(hr_dev, rc_sq_wqe, wr);
594 	if (WARN_ON(ret))
595 		return ret;
596 
597 	roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_FENCE_S,
598 		     (wr->send_flags & IB_SEND_FENCE) ? 1 : 0);
599 
600 	roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_SE_S,
601 		     (wr->send_flags & IB_SEND_SOLICITED) ? 1 : 0);
602 
603 	roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_CQE_S,
604 		     (wr->send_flags & IB_SEND_SIGNALED) ? 1 : 0);
605 
606 	if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
607 	    wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
608 		set_atomic_seg(wr, rc_sq_wqe, valid_num_sge);
609 	else if (wr->opcode != IB_WR_REG_MR)
610 		ret = set_rwqe_data_seg(&qp->ibqp, wr, rc_sq_wqe,
611 					&curr_idx, valid_num_sge);
612 
613 	/*
614 	 * The pipeline can sequentially post all valid WQEs into WQ buffer,
615 	 * including new WQEs waiting for the doorbell to update the PI again.
616 	 * Therefore, the owner bit of WQE MUST be updated after all fields
617 	 * and extSGEs have been written into DDR instead of cache.
618 	 */
619 	if (qp->en_flags & HNS_ROCE_QP_CAP_OWNER_DB)
620 		dma_wmb();
621 
622 	*sge_idx = curr_idx;
623 	roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_OWNER_S,
624 		     owner_bit);
625 
626 	return ret;
627 }
628 
629 static inline void update_sq_db(struct hns_roce_dev *hr_dev,
630 				struct hns_roce_qp *qp)
631 {
632 	/*
633 	 * Hip08 hardware cannot flush the WQEs in SQ if the QP state
634 	 * gets into errored mode. Hence, as a workaround to this
635 	 * hardware limitation, driver needs to assist in flushing. But
636 	 * the flushing operation uses mailbox to convey the QP state to
637 	 * the hardware and which can sleep due to the mutex protection
638 	 * around the mailbox calls. Hence, use the deferred flush for
639 	 * now.
640 	 */
641 	if (unlikely(qp->state == IB_QPS_ERR)) {
642 		if (!test_and_set_bit(HNS_ROCE_FLUSH_FLAG, &qp->flush_flag))
643 			init_flush_work(hr_dev, qp);
644 	} else {
645 		struct hns_roce_v2_db sq_db = {};
646 
647 		roce_set_field(sq_db.byte_4, V2_DB_TAG_M, V2_DB_TAG_S,
648 			       qp->doorbell_qpn);
649 		roce_set_field(sq_db.byte_4, V2_DB_CMD_M, V2_DB_CMD_S,
650 			       HNS_ROCE_V2_SQ_DB);
651 
652 		/* indicates data on new BAR, 0 : SQ doorbell, 1 : DWQE */
653 		roce_set_bit(sq_db.byte_4, V2_DB_FLAG_S, 0);
654 		roce_set_field(sq_db.parameter, V2_DB_PRODUCER_IDX_M,
655 			       V2_DB_PRODUCER_IDX_S, qp->sq.head);
656 		roce_set_field(sq_db.parameter, V2_DB_SL_M, V2_DB_SL_S,
657 			       qp->sl);
658 
659 		hns_roce_write64(hr_dev, (__le32 *)&sq_db, qp->sq.db_reg);
660 	}
661 }
662 
663 static inline void update_rq_db(struct hns_roce_dev *hr_dev,
664 				struct hns_roce_qp *qp)
665 {
666 	/*
667 	 * Hip08 hardware cannot flush the WQEs in RQ if the QP state
668 	 * gets into errored mode. Hence, as a workaround to this
669 	 * hardware limitation, driver needs to assist in flushing. But
670 	 * the flushing operation uses mailbox to convey the QP state to
671 	 * the hardware and which can sleep due to the mutex protection
672 	 * around the mailbox calls. Hence, use the deferred flush for
673 	 * now.
674 	 */
675 	if (unlikely(qp->state == IB_QPS_ERR)) {
676 		if (!test_and_set_bit(HNS_ROCE_FLUSH_FLAG, &qp->flush_flag))
677 			init_flush_work(hr_dev, qp);
678 	} else {
679 		if (likely(qp->en_flags & HNS_ROCE_QP_CAP_RQ_RECORD_DB)) {
680 			*qp->rdb.db_record =
681 					qp->rq.head & V2_DB_PRODUCER_IDX_M;
682 		} else {
683 			struct hns_roce_v2_db rq_db = {};
684 
685 			roce_set_field(rq_db.byte_4, V2_DB_TAG_M, V2_DB_TAG_S,
686 				       qp->qpn);
687 			roce_set_field(rq_db.byte_4, V2_DB_CMD_M, V2_DB_CMD_S,
688 				       HNS_ROCE_V2_RQ_DB);
689 			roce_set_field(rq_db.parameter, V2_DB_PRODUCER_IDX_M,
690 				       V2_DB_PRODUCER_IDX_S, qp->rq.head);
691 
692 			hns_roce_write64(hr_dev, (__le32 *)&rq_db,
693 					 qp->rq.db_reg);
694 		}
695 	}
696 }
697 
698 static void hns_roce_write512(struct hns_roce_dev *hr_dev, u64 *val,
699 			      u64 __iomem *dest)
700 {
701 #define HNS_ROCE_WRITE_TIMES 8
702 	struct hns_roce_v2_priv *priv = (struct hns_roce_v2_priv *)hr_dev->priv;
703 	struct hnae3_handle *handle = priv->handle;
704 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
705 	int i;
706 
707 	if (!hr_dev->dis_db && !ops->get_hw_reset_stat(handle))
708 		for (i = 0; i < HNS_ROCE_WRITE_TIMES; i++)
709 			writeq_relaxed(*(val + i), dest + i);
710 }
711 
712 static void write_dwqe(struct hns_roce_dev *hr_dev, struct hns_roce_qp *qp,
713 		       void *wqe)
714 {
715 	struct hns_roce_v2_rc_send_wqe *rc_sq_wqe = wqe;
716 
717 	/* All kinds of DirectWQE have the same header field layout */
718 	roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_FLAG_S, 1);
719 	roce_set_field(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_DB_SL_L_M,
720 		       V2_RC_SEND_WQE_BYTE_4_DB_SL_L_S, qp->sl);
721 	roce_set_field(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_DB_SL_H_M,
722 		       V2_RC_SEND_WQE_BYTE_4_DB_SL_H_S, qp->sl >> 2);
723 	roce_set_field(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_WQE_INDEX_M,
724 		       V2_RC_SEND_WQE_BYTE_4_WQE_INDEX_S, qp->sq.head);
725 
726 	hns_roce_write512(hr_dev, wqe, qp->sq.db_reg);
727 }
728 
729 static int hns_roce_v2_post_send(struct ib_qp *ibqp,
730 				 const struct ib_send_wr *wr,
731 				 const struct ib_send_wr **bad_wr)
732 {
733 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
734 	struct ib_device *ibdev = &hr_dev->ib_dev;
735 	struct hns_roce_qp *qp = to_hr_qp(ibqp);
736 	unsigned long flags = 0;
737 	unsigned int owner_bit;
738 	unsigned int sge_idx;
739 	unsigned int wqe_idx;
740 	void *wqe = NULL;
741 	u32 nreq;
742 	int ret;
743 
744 	spin_lock_irqsave(&qp->sq.lock, flags);
745 
746 	ret = check_send_valid(hr_dev, qp);
747 	if (unlikely(ret)) {
748 		*bad_wr = wr;
749 		nreq = 0;
750 		goto out;
751 	}
752 
753 	sge_idx = qp->next_sge;
754 
755 	for (nreq = 0; wr; ++nreq, wr = wr->next) {
756 		if (hns_roce_wq_overflow(&qp->sq, nreq, qp->ibqp.send_cq)) {
757 			ret = -ENOMEM;
758 			*bad_wr = wr;
759 			goto out;
760 		}
761 
762 		wqe_idx = (qp->sq.head + nreq) & (qp->sq.wqe_cnt - 1);
763 
764 		if (unlikely(wr->num_sge > qp->sq.max_gs)) {
765 			ibdev_err(ibdev, "num_sge = %d > qp->sq.max_gs = %u.\n",
766 				  wr->num_sge, qp->sq.max_gs);
767 			ret = -EINVAL;
768 			*bad_wr = wr;
769 			goto out;
770 		}
771 
772 		wqe = hns_roce_get_send_wqe(qp, wqe_idx);
773 		qp->sq.wrid[wqe_idx] = wr->wr_id;
774 		owner_bit =
775 		       ~(((qp->sq.head + nreq) >> ilog2(qp->sq.wqe_cnt)) & 0x1);
776 
777 		/* Corresponding to the QP type, wqe process separately */
778 		if (ibqp->qp_type == IB_QPT_GSI || ibqp->qp_type == IB_QPT_UD)
779 			ret = set_ud_wqe(qp, wr, wqe, &sge_idx, owner_bit);
780 		else if (ibqp->qp_type == IB_QPT_RC)
781 			ret = set_rc_wqe(qp, wr, wqe, &sge_idx, owner_bit);
782 
783 		if (unlikely(ret)) {
784 			*bad_wr = wr;
785 			goto out;
786 		}
787 	}
788 
789 out:
790 	if (likely(nreq)) {
791 		qp->sq.head += nreq;
792 		qp->next_sge = sge_idx;
793 
794 		if (nreq == 1 && qp->sq.head == qp->sq.tail + 1 &&
795 		    (qp->en_flags & HNS_ROCE_QP_CAP_DIRECT_WQE))
796 			write_dwqe(hr_dev, qp, wqe);
797 		else
798 			update_sq_db(hr_dev, qp);
799 	}
800 
801 	spin_unlock_irqrestore(&qp->sq.lock, flags);
802 
803 	return ret;
804 }
805 
806 static int check_recv_valid(struct hns_roce_dev *hr_dev,
807 			    struct hns_roce_qp *hr_qp)
808 {
809 	struct ib_device *ibdev = &hr_dev->ib_dev;
810 	struct ib_qp *ibqp = &hr_qp->ibqp;
811 
812 	if (unlikely(ibqp->qp_type != IB_QPT_RC &&
813 		     ibqp->qp_type != IB_QPT_GSI &&
814 		     ibqp->qp_type != IB_QPT_UD)) {
815 		ibdev_err(ibdev, "unsupported qp type, qp_type = %d.\n",
816 			  ibqp->qp_type);
817 		return -EOPNOTSUPP;
818 	}
819 
820 	if (unlikely(hr_dev->state >= HNS_ROCE_DEVICE_STATE_RST_DOWN))
821 		return -EIO;
822 
823 	if (hr_qp->state == IB_QPS_RESET)
824 		return -EINVAL;
825 
826 	return 0;
827 }
828 
829 static void fill_recv_sge_to_wqe(const struct ib_recv_wr *wr, void *wqe,
830 				 u32 max_sge, bool rsv)
831 {
832 	struct hns_roce_v2_wqe_data_seg *dseg = wqe;
833 	u32 i, cnt;
834 
835 	for (i = 0, cnt = 0; i < wr->num_sge; i++) {
836 		/* Skip zero-length sge */
837 		if (!wr->sg_list[i].length)
838 			continue;
839 		set_data_seg_v2(dseg + cnt, wr->sg_list + i);
840 		cnt++;
841 	}
842 
843 	/* Fill a reserved sge to make hw stop reading remaining segments */
844 	if (rsv) {
845 		dseg[cnt].lkey = cpu_to_le32(HNS_ROCE_INVALID_LKEY);
846 		dseg[cnt].addr = 0;
847 		dseg[cnt].len = cpu_to_le32(HNS_ROCE_INVALID_SGE_LENGTH);
848 	} else {
849 		/* Clear remaining segments to make ROCEE ignore sges */
850 		if (cnt < max_sge)
851 			memset(dseg + cnt, 0,
852 			       (max_sge - cnt) * HNS_ROCE_SGE_SIZE);
853 	}
854 }
855 
856 static void fill_rq_wqe(struct hns_roce_qp *hr_qp, const struct ib_recv_wr *wr,
857 			u32 wqe_idx, u32 max_sge)
858 {
859 	struct hns_roce_rinl_sge *sge_list;
860 	void *wqe = NULL;
861 	u32 i;
862 
863 	wqe = hns_roce_get_recv_wqe(hr_qp, wqe_idx);
864 	fill_recv_sge_to_wqe(wr, wqe, max_sge, hr_qp->rq.rsv_sge);
865 
866 	/* rq support inline data */
867 	if (hr_qp->rq_inl_buf.wqe_cnt) {
868 		sge_list = hr_qp->rq_inl_buf.wqe_list[wqe_idx].sg_list;
869 		hr_qp->rq_inl_buf.wqe_list[wqe_idx].sge_cnt = (u32)wr->num_sge;
870 		for (i = 0; i < wr->num_sge; i++) {
871 			sge_list[i].addr = (void *)(u64)wr->sg_list[i].addr;
872 			sge_list[i].len = wr->sg_list[i].length;
873 		}
874 	}
875 }
876 
877 static int hns_roce_v2_post_recv(struct ib_qp *ibqp,
878 				 const struct ib_recv_wr *wr,
879 				 const struct ib_recv_wr **bad_wr)
880 {
881 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
882 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
883 	struct ib_device *ibdev = &hr_dev->ib_dev;
884 	u32 wqe_idx, nreq, max_sge;
885 	unsigned long flags;
886 	int ret;
887 
888 	spin_lock_irqsave(&hr_qp->rq.lock, flags);
889 
890 	ret = check_recv_valid(hr_dev, hr_qp);
891 	if (unlikely(ret)) {
892 		*bad_wr = wr;
893 		nreq = 0;
894 		goto out;
895 	}
896 
897 	max_sge = hr_qp->rq.max_gs - hr_qp->rq.rsv_sge;
898 	for (nreq = 0; wr; ++nreq, wr = wr->next) {
899 		if (unlikely(hns_roce_wq_overflow(&hr_qp->rq, nreq,
900 						  hr_qp->ibqp.recv_cq))) {
901 			ret = -ENOMEM;
902 			*bad_wr = wr;
903 			goto out;
904 		}
905 
906 		if (unlikely(wr->num_sge > max_sge)) {
907 			ibdev_err(ibdev, "num_sge = %d >= max_sge = %u.\n",
908 				  wr->num_sge, max_sge);
909 			ret = -EINVAL;
910 			*bad_wr = wr;
911 			goto out;
912 		}
913 
914 		wqe_idx = (hr_qp->rq.head + nreq) & (hr_qp->rq.wqe_cnt - 1);
915 		fill_rq_wqe(hr_qp, wr, wqe_idx, max_sge);
916 		hr_qp->rq.wrid[wqe_idx] = wr->wr_id;
917 	}
918 
919 out:
920 	if (likely(nreq)) {
921 		hr_qp->rq.head += nreq;
922 
923 		update_rq_db(hr_dev, hr_qp);
924 	}
925 	spin_unlock_irqrestore(&hr_qp->rq.lock, flags);
926 
927 	return ret;
928 }
929 
930 static void *get_srq_wqe_buf(struct hns_roce_srq *srq, u32 n)
931 {
932 	return hns_roce_buf_offset(srq->buf_mtr.kmem, n << srq->wqe_shift);
933 }
934 
935 static void *get_idx_buf(struct hns_roce_idx_que *idx_que, u32 n)
936 {
937 	return hns_roce_buf_offset(idx_que->mtr.kmem,
938 				   n << idx_que->entry_shift);
939 }
940 
941 static void hns_roce_free_srq_wqe(struct hns_roce_srq *srq, u32 wqe_index)
942 {
943 	/* always called with interrupts disabled. */
944 	spin_lock(&srq->lock);
945 
946 	bitmap_clear(srq->idx_que.bitmap, wqe_index, 1);
947 	srq->idx_que.tail++;
948 
949 	spin_unlock(&srq->lock);
950 }
951 
952 static int hns_roce_srqwq_overflow(struct hns_roce_srq *srq)
953 {
954 	struct hns_roce_idx_que *idx_que = &srq->idx_que;
955 
956 	return idx_que->head - idx_que->tail >= srq->wqe_cnt;
957 }
958 
959 static int check_post_srq_valid(struct hns_roce_srq *srq, u32 max_sge,
960 				const struct ib_recv_wr *wr)
961 {
962 	struct ib_device *ib_dev = srq->ibsrq.device;
963 
964 	if (unlikely(wr->num_sge > max_sge)) {
965 		ibdev_err(ib_dev,
966 			  "failed to check sge, wr->num_sge = %d, max_sge = %u.\n",
967 			  wr->num_sge, max_sge);
968 		return -EINVAL;
969 	}
970 
971 	if (unlikely(hns_roce_srqwq_overflow(srq))) {
972 		ibdev_err(ib_dev,
973 			  "failed to check srqwq status, srqwq is full.\n");
974 		return -ENOMEM;
975 	}
976 
977 	return 0;
978 }
979 
980 static int get_srq_wqe_idx(struct hns_roce_srq *srq, u32 *wqe_idx)
981 {
982 	struct hns_roce_idx_que *idx_que = &srq->idx_que;
983 	u32 pos;
984 
985 	pos = find_first_zero_bit(idx_que->bitmap, srq->wqe_cnt);
986 	if (unlikely(pos == srq->wqe_cnt))
987 		return -ENOSPC;
988 
989 	bitmap_set(idx_que->bitmap, pos, 1);
990 	*wqe_idx = pos;
991 	return 0;
992 }
993 
994 static void fill_wqe_idx(struct hns_roce_srq *srq, unsigned int wqe_idx)
995 {
996 	struct hns_roce_idx_que *idx_que = &srq->idx_que;
997 	unsigned int head;
998 	__le32 *buf;
999 
1000 	head = idx_que->head & (srq->wqe_cnt - 1);
1001 
1002 	buf = get_idx_buf(idx_que, head);
1003 	*buf = cpu_to_le32(wqe_idx);
1004 
1005 	idx_que->head++;
1006 }
1007 
1008 static int hns_roce_v2_post_srq_recv(struct ib_srq *ibsrq,
1009 				     const struct ib_recv_wr *wr,
1010 				     const struct ib_recv_wr **bad_wr)
1011 {
1012 	struct hns_roce_dev *hr_dev = to_hr_dev(ibsrq->device);
1013 	struct hns_roce_srq *srq = to_hr_srq(ibsrq);
1014 	struct hns_roce_v2_db srq_db;
1015 	unsigned long flags;
1016 	int ret = 0;
1017 	u32 max_sge;
1018 	u32 wqe_idx;
1019 	void *wqe;
1020 	u32 nreq;
1021 
1022 	spin_lock_irqsave(&srq->lock, flags);
1023 
1024 	max_sge = srq->max_gs - srq->rsv_sge;
1025 	for (nreq = 0; wr; ++nreq, wr = wr->next) {
1026 		ret = check_post_srq_valid(srq, max_sge, wr);
1027 		if (ret) {
1028 			*bad_wr = wr;
1029 			break;
1030 		}
1031 
1032 		ret = get_srq_wqe_idx(srq, &wqe_idx);
1033 		if (unlikely(ret)) {
1034 			*bad_wr = wr;
1035 			break;
1036 		}
1037 
1038 		wqe = get_srq_wqe_buf(srq, wqe_idx);
1039 		fill_recv_sge_to_wqe(wr, wqe, max_sge, srq->rsv_sge);
1040 		fill_wqe_idx(srq, wqe_idx);
1041 		srq->wrid[wqe_idx] = wr->wr_id;
1042 	}
1043 
1044 	if (likely(nreq)) {
1045 		roce_set_field(srq_db.byte_4, V2_DB_TAG_M, V2_DB_TAG_S,
1046 			       srq->srqn);
1047 		roce_set_field(srq_db.byte_4, V2_DB_CMD_M, V2_DB_CMD_S,
1048 			       HNS_ROCE_V2_SRQ_DB);
1049 		roce_set_field(srq_db.parameter, V2_DB_PRODUCER_IDX_M,
1050 			       V2_DB_PRODUCER_IDX_S, srq->idx_que.head);
1051 
1052 		hns_roce_write64(hr_dev, (__le32 *)&srq_db, srq->db_reg);
1053 	}
1054 
1055 	spin_unlock_irqrestore(&srq->lock, flags);
1056 
1057 	return ret;
1058 }
1059 
1060 static u32 hns_roce_v2_cmd_hw_reseted(struct hns_roce_dev *hr_dev,
1061 				      unsigned long instance_stage,
1062 				      unsigned long reset_stage)
1063 {
1064 	/* When hardware reset has been completed once or more, we should stop
1065 	 * sending mailbox&cmq&doorbell to hardware. If now in .init_instance()
1066 	 * function, we should exit with error. If now at HNAE3_INIT_CLIENT
1067 	 * stage of soft reset process, we should exit with error, and then
1068 	 * HNAE3_INIT_CLIENT related process can rollback the operation like
1069 	 * notifing hardware to free resources, HNAE3_INIT_CLIENT related
1070 	 * process will exit with error to notify NIC driver to reschedule soft
1071 	 * reset process once again.
1072 	 */
1073 	hr_dev->is_reset = true;
1074 	hr_dev->dis_db = true;
1075 
1076 	if (reset_stage == HNS_ROCE_STATE_RST_INIT ||
1077 	    instance_stage == HNS_ROCE_STATE_INIT)
1078 		return CMD_RST_PRC_EBUSY;
1079 
1080 	return CMD_RST_PRC_SUCCESS;
1081 }
1082 
1083 static u32 hns_roce_v2_cmd_hw_resetting(struct hns_roce_dev *hr_dev,
1084 					unsigned long instance_stage,
1085 					unsigned long reset_stage)
1086 {
1087 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1088 	struct hnae3_handle *handle = priv->handle;
1089 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1090 
1091 	/* When hardware reset is detected, we should stop sending mailbox&cmq&
1092 	 * doorbell to hardware. If now in .init_instance() function, we should
1093 	 * exit with error. If now at HNAE3_INIT_CLIENT stage of soft reset
1094 	 * process, we should exit with error, and then HNAE3_INIT_CLIENT
1095 	 * related process can rollback the operation like notifing hardware to
1096 	 * free resources, HNAE3_INIT_CLIENT related process will exit with
1097 	 * error to notify NIC driver to reschedule soft reset process once
1098 	 * again.
1099 	 */
1100 	hr_dev->dis_db = true;
1101 	if (!ops->get_hw_reset_stat(handle))
1102 		hr_dev->is_reset = true;
1103 
1104 	if (!hr_dev->is_reset || reset_stage == HNS_ROCE_STATE_RST_INIT ||
1105 	    instance_stage == HNS_ROCE_STATE_INIT)
1106 		return CMD_RST_PRC_EBUSY;
1107 
1108 	return CMD_RST_PRC_SUCCESS;
1109 }
1110 
1111 static u32 hns_roce_v2_cmd_sw_resetting(struct hns_roce_dev *hr_dev)
1112 {
1113 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1114 	struct hnae3_handle *handle = priv->handle;
1115 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1116 
1117 	/* When software reset is detected at .init_instance() function, we
1118 	 * should stop sending mailbox&cmq&doorbell to hardware, and exit
1119 	 * with error.
1120 	 */
1121 	hr_dev->dis_db = true;
1122 	if (ops->ae_dev_reset_cnt(handle) != hr_dev->reset_cnt)
1123 		hr_dev->is_reset = true;
1124 
1125 	return CMD_RST_PRC_EBUSY;
1126 }
1127 
1128 static u32 check_aedev_reset_status(struct hns_roce_dev *hr_dev,
1129 				    struct hnae3_handle *handle)
1130 {
1131 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1132 	unsigned long instance_stage; /* the current instance stage */
1133 	unsigned long reset_stage; /* the current reset stage */
1134 	unsigned long reset_cnt;
1135 	bool sw_resetting;
1136 	bool hw_resetting;
1137 
1138 	/* Get information about reset from NIC driver or RoCE driver itself,
1139 	 * the meaning of the following variables from NIC driver are described
1140 	 * as below:
1141 	 * reset_cnt -- The count value of completed hardware reset.
1142 	 * hw_resetting -- Whether hardware device is resetting now.
1143 	 * sw_resetting -- Whether NIC's software reset process is running now.
1144 	 */
1145 	instance_stage = handle->rinfo.instance_state;
1146 	reset_stage = handle->rinfo.reset_state;
1147 	reset_cnt = ops->ae_dev_reset_cnt(handle);
1148 	if (reset_cnt != hr_dev->reset_cnt)
1149 		return hns_roce_v2_cmd_hw_reseted(hr_dev, instance_stage,
1150 						  reset_stage);
1151 
1152 	hw_resetting = ops->get_cmdq_stat(handle);
1153 	if (hw_resetting)
1154 		return hns_roce_v2_cmd_hw_resetting(hr_dev, instance_stage,
1155 						    reset_stage);
1156 
1157 	sw_resetting = ops->ae_dev_resetting(handle);
1158 	if (sw_resetting && instance_stage == HNS_ROCE_STATE_INIT)
1159 		return hns_roce_v2_cmd_sw_resetting(hr_dev);
1160 
1161 	return CMD_RST_PRC_OTHERS;
1162 }
1163 
1164 static bool check_device_is_in_reset(struct hns_roce_dev *hr_dev)
1165 {
1166 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1167 	struct hnae3_handle *handle = priv->handle;
1168 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1169 
1170 	if (hr_dev->reset_cnt != ops->ae_dev_reset_cnt(handle))
1171 		return true;
1172 
1173 	if (ops->get_hw_reset_stat(handle))
1174 		return true;
1175 
1176 	if (ops->ae_dev_resetting(handle))
1177 		return true;
1178 
1179 	return false;
1180 }
1181 
1182 static bool v2_chk_mbox_is_avail(struct hns_roce_dev *hr_dev, bool *busy)
1183 {
1184 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1185 	u32 status;
1186 
1187 	if (hr_dev->is_reset)
1188 		status = CMD_RST_PRC_SUCCESS;
1189 	else
1190 		status = check_aedev_reset_status(hr_dev, priv->handle);
1191 
1192 	*busy = (status == CMD_RST_PRC_EBUSY);
1193 
1194 	return status == CMD_RST_PRC_OTHERS;
1195 }
1196 
1197 static int hns_roce_alloc_cmq_desc(struct hns_roce_dev *hr_dev,
1198 				   struct hns_roce_v2_cmq_ring *ring)
1199 {
1200 	int size = ring->desc_num * sizeof(struct hns_roce_cmq_desc);
1201 
1202 	ring->desc = kzalloc(size, GFP_KERNEL);
1203 	if (!ring->desc)
1204 		return -ENOMEM;
1205 
1206 	ring->desc_dma_addr = dma_map_single(hr_dev->dev, ring->desc, size,
1207 					     DMA_BIDIRECTIONAL);
1208 	if (dma_mapping_error(hr_dev->dev, ring->desc_dma_addr)) {
1209 		ring->desc_dma_addr = 0;
1210 		kfree(ring->desc);
1211 		ring->desc = NULL;
1212 
1213 		dev_err_ratelimited(hr_dev->dev,
1214 				    "failed to map cmq desc addr.\n");
1215 		return -ENOMEM;
1216 	}
1217 
1218 	return 0;
1219 }
1220 
1221 static void hns_roce_free_cmq_desc(struct hns_roce_dev *hr_dev,
1222 				   struct hns_roce_v2_cmq_ring *ring)
1223 {
1224 	dma_unmap_single(hr_dev->dev, ring->desc_dma_addr,
1225 			 ring->desc_num * sizeof(struct hns_roce_cmq_desc),
1226 			 DMA_BIDIRECTIONAL);
1227 
1228 	ring->desc_dma_addr = 0;
1229 	kfree(ring->desc);
1230 }
1231 
1232 static int hns_roce_init_cmq_ring(struct hns_roce_dev *hr_dev, bool ring_type)
1233 {
1234 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1235 	struct hns_roce_v2_cmq_ring *ring = (ring_type == TYPE_CSQ) ?
1236 					    &priv->cmq.csq : &priv->cmq.crq;
1237 
1238 	ring->flag = ring_type;
1239 	ring->head = 0;
1240 
1241 	return hns_roce_alloc_cmq_desc(hr_dev, ring);
1242 }
1243 
1244 static void hns_roce_cmq_init_regs(struct hns_roce_dev *hr_dev, bool ring_type)
1245 {
1246 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1247 	struct hns_roce_v2_cmq_ring *ring = (ring_type == TYPE_CSQ) ?
1248 					    &priv->cmq.csq : &priv->cmq.crq;
1249 	dma_addr_t dma = ring->desc_dma_addr;
1250 
1251 	if (ring_type == TYPE_CSQ) {
1252 		roce_write(hr_dev, ROCEE_TX_CMQ_BASEADDR_L_REG, (u32)dma);
1253 		roce_write(hr_dev, ROCEE_TX_CMQ_BASEADDR_H_REG,
1254 			   upper_32_bits(dma));
1255 		roce_write(hr_dev, ROCEE_TX_CMQ_DEPTH_REG,
1256 			   (u32)ring->desc_num >> HNS_ROCE_CMQ_DESC_NUM_S);
1257 
1258 		/* Make sure to write tail first and then head */
1259 		roce_write(hr_dev, ROCEE_TX_CMQ_TAIL_REG, 0);
1260 		roce_write(hr_dev, ROCEE_TX_CMQ_HEAD_REG, 0);
1261 	} else {
1262 		roce_write(hr_dev, ROCEE_RX_CMQ_BASEADDR_L_REG, (u32)dma);
1263 		roce_write(hr_dev, ROCEE_RX_CMQ_BASEADDR_H_REG,
1264 			   upper_32_bits(dma));
1265 		roce_write(hr_dev, ROCEE_RX_CMQ_DEPTH_REG,
1266 			   (u32)ring->desc_num >> HNS_ROCE_CMQ_DESC_NUM_S);
1267 		roce_write(hr_dev, ROCEE_RX_CMQ_HEAD_REG, 0);
1268 		roce_write(hr_dev, ROCEE_RX_CMQ_TAIL_REG, 0);
1269 	}
1270 }
1271 
1272 static int hns_roce_v2_cmq_init(struct hns_roce_dev *hr_dev)
1273 {
1274 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1275 	int ret;
1276 
1277 	/* Setup the queue entries for command queue */
1278 	priv->cmq.csq.desc_num = CMD_CSQ_DESC_NUM;
1279 	priv->cmq.crq.desc_num = CMD_CRQ_DESC_NUM;
1280 
1281 	/* Setup the lock for command queue */
1282 	spin_lock_init(&priv->cmq.csq.lock);
1283 	spin_lock_init(&priv->cmq.crq.lock);
1284 
1285 	/* Setup Tx write back timeout */
1286 	priv->cmq.tx_timeout = HNS_ROCE_CMQ_TX_TIMEOUT;
1287 
1288 	/* Init CSQ */
1289 	ret = hns_roce_init_cmq_ring(hr_dev, TYPE_CSQ);
1290 	if (ret) {
1291 		dev_err_ratelimited(hr_dev->dev,
1292 				    "failed to init CSQ, ret = %d.\n", ret);
1293 		return ret;
1294 	}
1295 
1296 	/* Init CRQ */
1297 	ret = hns_roce_init_cmq_ring(hr_dev, TYPE_CRQ);
1298 	if (ret) {
1299 		dev_err_ratelimited(hr_dev->dev,
1300 				    "failed to init CRQ, ret = %d.\n", ret);
1301 		goto err_crq;
1302 	}
1303 
1304 	/* Init CSQ REG */
1305 	hns_roce_cmq_init_regs(hr_dev, TYPE_CSQ);
1306 
1307 	/* Init CRQ REG */
1308 	hns_roce_cmq_init_regs(hr_dev, TYPE_CRQ);
1309 
1310 	return 0;
1311 
1312 err_crq:
1313 	hns_roce_free_cmq_desc(hr_dev, &priv->cmq.csq);
1314 
1315 	return ret;
1316 }
1317 
1318 static void hns_roce_v2_cmq_exit(struct hns_roce_dev *hr_dev)
1319 {
1320 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1321 
1322 	hns_roce_free_cmq_desc(hr_dev, &priv->cmq.csq);
1323 	hns_roce_free_cmq_desc(hr_dev, &priv->cmq.crq);
1324 }
1325 
1326 static void hns_roce_cmq_setup_basic_desc(struct hns_roce_cmq_desc *desc,
1327 					  enum hns_roce_opcode_type opcode,
1328 					  bool is_read)
1329 {
1330 	memset((void *)desc, 0, sizeof(struct hns_roce_cmq_desc));
1331 	desc->opcode = cpu_to_le16(opcode);
1332 	desc->flag =
1333 		cpu_to_le16(HNS_ROCE_CMD_FLAG_NO_INTR | HNS_ROCE_CMD_FLAG_IN);
1334 	if (is_read)
1335 		desc->flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_WR);
1336 	else
1337 		desc->flag &= cpu_to_le16(~HNS_ROCE_CMD_FLAG_WR);
1338 }
1339 
1340 static int hns_roce_cmq_csq_done(struct hns_roce_dev *hr_dev)
1341 {
1342 	u32 tail = roce_read(hr_dev, ROCEE_TX_CMQ_TAIL_REG);
1343 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1344 
1345 	return tail == priv->cmq.csq.head;
1346 }
1347 
1348 static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
1349 			       struct hns_roce_cmq_desc *desc, int num)
1350 {
1351 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1352 	struct hns_roce_v2_cmq_ring *csq = &priv->cmq.csq;
1353 	u32 timeout = 0;
1354 	u16 desc_ret;
1355 	u32 tail;
1356 	int ret;
1357 	int i;
1358 
1359 	spin_lock_bh(&csq->lock);
1360 
1361 	tail = csq->head;
1362 
1363 	for (i = 0; i < num; i++) {
1364 		csq->desc[csq->head++] = desc[i];
1365 		if (csq->head == csq->desc_num)
1366 			csq->head = 0;
1367 	}
1368 
1369 	/* Write to hardware */
1370 	roce_write(hr_dev, ROCEE_TX_CMQ_HEAD_REG, csq->head);
1371 
1372 	/* If the command is sync, wait for the firmware to write back,
1373 	 * if multi descriptors to be sent, use the first one to check
1374 	 */
1375 	if (le16_to_cpu(desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
1376 		do {
1377 			if (hns_roce_cmq_csq_done(hr_dev))
1378 				break;
1379 			udelay(1);
1380 		} while (++timeout < priv->cmq.tx_timeout);
1381 	}
1382 
1383 	if (hns_roce_cmq_csq_done(hr_dev)) {
1384 		for (ret = 0, i = 0; i < num; i++) {
1385 			/* check the result of hardware write back */
1386 			desc[i] = csq->desc[tail++];
1387 			if (tail == csq->desc_num)
1388 				tail = 0;
1389 
1390 			desc_ret = le16_to_cpu(desc[i].retval);
1391 			if (likely(desc_ret == CMD_EXEC_SUCCESS))
1392 				continue;
1393 
1394 			dev_err_ratelimited(hr_dev->dev,
1395 					    "Cmdq IO error, opcode = %x, return = %x\n",
1396 					    desc->opcode, desc_ret);
1397 			ret = -EIO;
1398 		}
1399 	} else {
1400 		/* FW/HW reset or incorrect number of desc */
1401 		tail = roce_read(hr_dev, ROCEE_TX_CMQ_TAIL_REG);
1402 		dev_warn(hr_dev->dev, "CMDQ move tail from %d to %d\n",
1403 			 csq->head, tail);
1404 		csq->head = tail;
1405 
1406 		ret = -EAGAIN;
1407 	}
1408 
1409 	spin_unlock_bh(&csq->lock);
1410 
1411 	return ret;
1412 }
1413 
1414 static int hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
1415 			     struct hns_roce_cmq_desc *desc, int num)
1416 {
1417 	bool busy;
1418 	int ret;
1419 
1420 	if (!v2_chk_mbox_is_avail(hr_dev, &busy))
1421 		return busy ? -EBUSY : 0;
1422 
1423 	ret = __hns_roce_cmq_send(hr_dev, desc, num);
1424 	if (ret) {
1425 		if (!v2_chk_mbox_is_avail(hr_dev, &busy))
1426 			return busy ? -EBUSY : 0;
1427 	}
1428 
1429 	return ret;
1430 }
1431 
1432 static int config_hem_ba_to_hw(struct hns_roce_dev *hr_dev, unsigned long obj,
1433 			       dma_addr_t base_addr, u16 op)
1434 {
1435 	struct hns_roce_cmd_mailbox *mbox = hns_roce_alloc_cmd_mailbox(hr_dev);
1436 	int ret;
1437 
1438 	if (IS_ERR(mbox))
1439 		return PTR_ERR(mbox);
1440 
1441 	ret = hns_roce_cmd_mbox(hr_dev, base_addr, mbox->dma, obj, 0, op,
1442 				HNS_ROCE_CMD_TIMEOUT_MSECS);
1443 	hns_roce_free_cmd_mailbox(hr_dev, mbox);
1444 	return ret;
1445 }
1446 
1447 static int hns_roce_cmq_query_hw_info(struct hns_roce_dev *hr_dev)
1448 {
1449 	struct hns_roce_query_version *resp;
1450 	struct hns_roce_cmq_desc desc;
1451 	int ret;
1452 
1453 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_HW_VER, true);
1454 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1455 	if (ret)
1456 		return ret;
1457 
1458 	resp = (struct hns_roce_query_version *)desc.data;
1459 	hr_dev->hw_rev = le16_to_cpu(resp->rocee_hw_version);
1460 	hr_dev->vendor_id = hr_dev->pci_dev->vendor;
1461 
1462 	return 0;
1463 }
1464 
1465 static void func_clr_hw_resetting_state(struct hns_roce_dev *hr_dev,
1466 					struct hnae3_handle *handle)
1467 {
1468 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1469 	unsigned long end;
1470 
1471 	hr_dev->dis_db = true;
1472 
1473 	dev_warn(hr_dev->dev,
1474 		 "Func clear is pending, device in resetting state.\n");
1475 	end = HNS_ROCE_V2_HW_RST_TIMEOUT;
1476 	while (end) {
1477 		if (!ops->get_hw_reset_stat(handle)) {
1478 			hr_dev->is_reset = true;
1479 			dev_info(hr_dev->dev,
1480 				 "Func clear success after reset.\n");
1481 			return;
1482 		}
1483 		msleep(HNS_ROCE_V2_HW_RST_COMPLETION_WAIT);
1484 		end -= HNS_ROCE_V2_HW_RST_COMPLETION_WAIT;
1485 	}
1486 
1487 	dev_warn(hr_dev->dev, "Func clear failed.\n");
1488 }
1489 
1490 static void func_clr_sw_resetting_state(struct hns_roce_dev *hr_dev,
1491 					struct hnae3_handle *handle)
1492 {
1493 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1494 	unsigned long end;
1495 
1496 	hr_dev->dis_db = true;
1497 
1498 	dev_warn(hr_dev->dev,
1499 		 "Func clear is pending, device in resetting state.\n");
1500 	end = HNS_ROCE_V2_HW_RST_TIMEOUT;
1501 	while (end) {
1502 		if (ops->ae_dev_reset_cnt(handle) !=
1503 		    hr_dev->reset_cnt) {
1504 			hr_dev->is_reset = true;
1505 			dev_info(hr_dev->dev,
1506 				 "Func clear success after sw reset\n");
1507 			return;
1508 		}
1509 		msleep(HNS_ROCE_V2_HW_RST_COMPLETION_WAIT);
1510 		end -= HNS_ROCE_V2_HW_RST_COMPLETION_WAIT;
1511 	}
1512 
1513 	dev_warn(hr_dev->dev, "Func clear failed because of unfinished sw reset\n");
1514 }
1515 
1516 static void hns_roce_func_clr_rst_proc(struct hns_roce_dev *hr_dev, int retval,
1517 				       int flag)
1518 {
1519 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1520 	struct hnae3_handle *handle = priv->handle;
1521 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
1522 
1523 	if (ops->ae_dev_reset_cnt(handle) != hr_dev->reset_cnt) {
1524 		hr_dev->dis_db = true;
1525 		hr_dev->is_reset = true;
1526 		dev_info(hr_dev->dev, "Func clear success after reset.\n");
1527 		return;
1528 	}
1529 
1530 	if (ops->get_hw_reset_stat(handle)) {
1531 		func_clr_hw_resetting_state(hr_dev, handle);
1532 		return;
1533 	}
1534 
1535 	if (ops->ae_dev_resetting(handle) &&
1536 	    handle->rinfo.instance_state == HNS_ROCE_STATE_INIT) {
1537 		func_clr_sw_resetting_state(hr_dev, handle);
1538 		return;
1539 	}
1540 
1541 	if (retval && !flag)
1542 		dev_warn(hr_dev->dev,
1543 			 "Func clear read failed, ret = %d.\n", retval);
1544 
1545 	dev_warn(hr_dev->dev, "Func clear failed.\n");
1546 }
1547 
1548 static void __hns_roce_function_clear(struct hns_roce_dev *hr_dev, int vf_id)
1549 {
1550 	bool fclr_write_fail_flag = false;
1551 	struct hns_roce_func_clear *resp;
1552 	struct hns_roce_cmq_desc desc;
1553 	unsigned long end;
1554 	int ret = 0;
1555 
1556 	if (check_device_is_in_reset(hr_dev))
1557 		goto out;
1558 
1559 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_FUNC_CLEAR, false);
1560 	resp = (struct hns_roce_func_clear *)desc.data;
1561 	resp->rst_funcid_en = cpu_to_le32(vf_id);
1562 
1563 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1564 	if (ret) {
1565 		fclr_write_fail_flag = true;
1566 		dev_err(hr_dev->dev, "Func clear write failed, ret = %d.\n",
1567 			 ret);
1568 		goto out;
1569 	}
1570 
1571 	msleep(HNS_ROCE_V2_READ_FUNC_CLEAR_FLAG_INTERVAL);
1572 	end = HNS_ROCE_V2_FUNC_CLEAR_TIMEOUT_MSECS;
1573 	while (end) {
1574 		if (check_device_is_in_reset(hr_dev))
1575 			goto out;
1576 		msleep(HNS_ROCE_V2_READ_FUNC_CLEAR_FLAG_FAIL_WAIT);
1577 		end -= HNS_ROCE_V2_READ_FUNC_CLEAR_FLAG_FAIL_WAIT;
1578 
1579 		hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_FUNC_CLEAR,
1580 					      true);
1581 
1582 		resp->rst_funcid_en = cpu_to_le32(vf_id);
1583 		ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1584 		if (ret)
1585 			continue;
1586 
1587 		if (roce_get_bit(resp->func_done, FUNC_CLEAR_RST_FUN_DONE_S)) {
1588 			if (vf_id == 0)
1589 				hr_dev->is_reset = true;
1590 			return;
1591 		}
1592 	}
1593 
1594 out:
1595 	hns_roce_func_clr_rst_proc(hr_dev, ret, fclr_write_fail_flag);
1596 }
1597 
1598 static void hns_roce_free_vf_resource(struct hns_roce_dev *hr_dev, int vf_id)
1599 {
1600 	enum hns_roce_opcode_type opcode = HNS_ROCE_OPC_ALLOC_VF_RES;
1601 	struct hns_roce_cmq_desc desc[2];
1602 	struct hns_roce_cmq_req *req_a;
1603 
1604 	req_a = (struct hns_roce_cmq_req *)desc[0].data;
1605 	hns_roce_cmq_setup_basic_desc(&desc[0], opcode, false);
1606 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
1607 	hns_roce_cmq_setup_basic_desc(&desc[1], opcode, false);
1608 	hr_reg_write(req_a, FUNC_RES_A_VF_ID, vf_id);
1609 	hns_roce_cmq_send(hr_dev, desc, 2);
1610 }
1611 
1612 static void hns_roce_function_clear(struct hns_roce_dev *hr_dev)
1613 {
1614 	int i;
1615 
1616 	for (i = hr_dev->func_num - 1; i >= 0; i--) {
1617 		__hns_roce_function_clear(hr_dev, i);
1618 		if (i != 0)
1619 			hns_roce_free_vf_resource(hr_dev, i);
1620 	}
1621 }
1622 
1623 static int hns_roce_query_fw_ver(struct hns_roce_dev *hr_dev)
1624 {
1625 	struct hns_roce_query_fw_info *resp;
1626 	struct hns_roce_cmq_desc desc;
1627 	int ret;
1628 
1629 	hns_roce_cmq_setup_basic_desc(&desc, HNS_QUERY_FW_VER, true);
1630 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1631 	if (ret)
1632 		return ret;
1633 
1634 	resp = (struct hns_roce_query_fw_info *)desc.data;
1635 	hr_dev->caps.fw_ver = (u64)(le32_to_cpu(resp->fw_ver));
1636 
1637 	return 0;
1638 }
1639 
1640 static int hns_roce_query_func_info(struct hns_roce_dev *hr_dev)
1641 {
1642 	struct hns_roce_cmq_desc desc;
1643 	int ret;
1644 
1645 	if (hr_dev->pci_dev->revision < PCI_REVISION_ID_HIP09) {
1646 		hr_dev->func_num = 1;
1647 		return 0;
1648 	}
1649 
1650 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_FUNC_INFO,
1651 				      true);
1652 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1653 	if (ret) {
1654 		hr_dev->func_num = 1;
1655 		return ret;
1656 	}
1657 
1658 	hr_dev->func_num = le32_to_cpu(desc.func_info.own_func_num);
1659 	hr_dev->cong_algo_tmpl_id = le32_to_cpu(desc.func_info.own_mac_id);
1660 
1661 	return 0;
1662 }
1663 
1664 static int hns_roce_config_global_param(struct hns_roce_dev *hr_dev)
1665 {
1666 	struct hns_roce_cmq_desc desc;
1667 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
1668 
1669 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_GLOBAL_PARAM,
1670 				      false);
1671 
1672 	hr_reg_write(req, CFG_GLOBAL_PARAM_1US_CYCLES, 0x3e8);
1673 	hr_reg_write(req, CFG_GLOBAL_PARAM_UDP_PORT, ROCE_V2_UDP_DPORT);
1674 
1675 	return hns_roce_cmq_send(hr_dev, &desc, 1);
1676 }
1677 
1678 static int load_func_res_caps(struct hns_roce_dev *hr_dev, bool is_vf)
1679 {
1680 	struct hns_roce_cmq_desc desc[2];
1681 	struct hns_roce_cmq_req *r_a = (struct hns_roce_cmq_req *)desc[0].data;
1682 	struct hns_roce_cmq_req *r_b = (struct hns_roce_cmq_req *)desc[1].data;
1683 	struct hns_roce_caps *caps = &hr_dev->caps;
1684 	enum hns_roce_opcode_type opcode;
1685 	u32 func_num;
1686 	int ret;
1687 
1688 	if (is_vf) {
1689 		opcode = HNS_ROCE_OPC_QUERY_VF_RES;
1690 		func_num = 1;
1691 	} else {
1692 		opcode = HNS_ROCE_OPC_QUERY_PF_RES;
1693 		func_num = hr_dev->func_num;
1694 	}
1695 
1696 	hns_roce_cmq_setup_basic_desc(&desc[0], opcode, true);
1697 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
1698 	hns_roce_cmq_setup_basic_desc(&desc[1], opcode, true);
1699 
1700 	ret = hns_roce_cmq_send(hr_dev, desc, 2);
1701 	if (ret)
1702 		return ret;
1703 
1704 	caps->qpc_bt_num = hr_reg_read(r_a, FUNC_RES_A_QPC_BT_NUM) / func_num;
1705 	caps->srqc_bt_num = hr_reg_read(r_a, FUNC_RES_A_SRQC_BT_NUM) / func_num;
1706 	caps->cqc_bt_num = hr_reg_read(r_a, FUNC_RES_A_CQC_BT_NUM) / func_num;
1707 	caps->mpt_bt_num = hr_reg_read(r_a, FUNC_RES_A_MPT_BT_NUM) / func_num;
1708 	caps->eqc_bt_num = hr_reg_read(r_a, FUNC_RES_A_EQC_BT_NUM) / func_num;
1709 	caps->smac_bt_num = hr_reg_read(r_b, FUNC_RES_B_SMAC_NUM) / func_num;
1710 	caps->sgid_bt_num = hr_reg_read(r_b, FUNC_RES_B_SGID_NUM) / func_num;
1711 	caps->sccc_bt_num = hr_reg_read(r_b, FUNC_RES_B_SCCC_BT_NUM) / func_num;
1712 
1713 	if (is_vf) {
1714 		caps->sl_num = hr_reg_read(r_b, FUNC_RES_V_QID_NUM) / func_num;
1715 		caps->gmv_bt_num = hr_reg_read(r_b, FUNC_RES_V_GMV_BT_NUM) /
1716 					       func_num;
1717 	} else {
1718 		caps->sl_num = hr_reg_read(r_b, FUNC_RES_B_QID_NUM) / func_num;
1719 		caps->gmv_bt_num = hr_reg_read(r_b, FUNC_RES_B_GMV_BT_NUM) /
1720 					       func_num;
1721 	}
1722 
1723 	return 0;
1724 }
1725 
1726 static int hns_roce_query_pf_resource(struct hns_roce_dev *hr_dev)
1727 {
1728 	return load_func_res_caps(hr_dev, false);
1729 }
1730 
1731 static int hns_roce_query_vf_resource(struct hns_roce_dev *hr_dev)
1732 {
1733 	return load_func_res_caps(hr_dev, true);
1734 }
1735 
1736 static int hns_roce_query_pf_timer_resource(struct hns_roce_dev *hr_dev)
1737 {
1738 	struct hns_roce_cmq_desc desc;
1739 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
1740 	struct hns_roce_caps *caps = &hr_dev->caps;
1741 	int ret;
1742 
1743 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_PF_TIMER_RES,
1744 				      true);
1745 
1746 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1747 	if (ret)
1748 		return ret;
1749 
1750 	caps->qpc_timer_bt_num = hr_reg_read(req, PF_TIMER_RES_QPC_ITEM_NUM);
1751 	caps->cqc_timer_bt_num = hr_reg_read(req, PF_TIMER_RES_CQC_ITEM_NUM);
1752 
1753 	return 0;
1754 }
1755 
1756 static int __hns_roce_set_vf_switch_param(struct hns_roce_dev *hr_dev,
1757 					  u32 vf_id)
1758 {
1759 	struct hns_roce_vf_switch *swt;
1760 	struct hns_roce_cmq_desc desc;
1761 	int ret;
1762 
1763 	swt = (struct hns_roce_vf_switch *)desc.data;
1764 	hns_roce_cmq_setup_basic_desc(&desc, HNS_SWITCH_PARAMETER_CFG, true);
1765 	swt->rocee_sel |= cpu_to_le32(HNS_ICL_SWITCH_CMD_ROCEE_SEL);
1766 	roce_set_field(swt->fun_id, VF_SWITCH_DATA_FUN_ID_VF_ID_M,
1767 		       VF_SWITCH_DATA_FUN_ID_VF_ID_S, vf_id);
1768 	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
1769 	if (ret)
1770 		return ret;
1771 
1772 	desc.flag =
1773 		cpu_to_le16(HNS_ROCE_CMD_FLAG_NO_INTR | HNS_ROCE_CMD_FLAG_IN);
1774 	desc.flag &= cpu_to_le16(~HNS_ROCE_CMD_FLAG_WR);
1775 	roce_set_bit(swt->cfg, VF_SWITCH_DATA_CFG_ALW_LPBK_S, 1);
1776 	roce_set_bit(swt->cfg, VF_SWITCH_DATA_CFG_ALW_LCL_LPBK_S, 0);
1777 	roce_set_bit(swt->cfg, VF_SWITCH_DATA_CFG_ALW_DST_OVRD_S, 1);
1778 
1779 	return hns_roce_cmq_send(hr_dev, &desc, 1);
1780 }
1781 
1782 static int hns_roce_set_vf_switch_param(struct hns_roce_dev *hr_dev)
1783 {
1784 	u32 vf_id;
1785 	int ret;
1786 
1787 	for (vf_id = 0; vf_id < hr_dev->func_num; vf_id++) {
1788 		ret = __hns_roce_set_vf_switch_param(hr_dev, vf_id);
1789 		if (ret)
1790 			return ret;
1791 	}
1792 	return 0;
1793 }
1794 
1795 static int __hns_roce_alloc_vf_resource(struct hns_roce_dev *hr_dev, int vf_id)
1796 {
1797 	struct hns_roce_cmq_desc desc[2];
1798 	struct hns_roce_cmq_req *r_a = (struct hns_roce_cmq_req *)desc[0].data;
1799 	struct hns_roce_cmq_req *r_b = (struct hns_roce_cmq_req *)desc[1].data;
1800 	enum hns_roce_opcode_type opcode = HNS_ROCE_OPC_ALLOC_VF_RES;
1801 	struct hns_roce_caps *caps = &hr_dev->caps;
1802 
1803 	hns_roce_cmq_setup_basic_desc(&desc[0], opcode, false);
1804 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
1805 	hns_roce_cmq_setup_basic_desc(&desc[1], opcode, false);
1806 
1807 	hr_reg_write(r_a, FUNC_RES_A_VF_ID, vf_id);
1808 
1809 	hr_reg_write(r_a, FUNC_RES_A_QPC_BT_NUM, caps->qpc_bt_num);
1810 	hr_reg_write(r_a, FUNC_RES_A_QPC_BT_IDX, vf_id * caps->qpc_bt_num);
1811 	hr_reg_write(r_a, FUNC_RES_A_SRQC_BT_NUM, caps->srqc_bt_num);
1812 	hr_reg_write(r_a, FUNC_RES_A_SRQC_BT_IDX, vf_id * caps->srqc_bt_num);
1813 	hr_reg_write(r_a, FUNC_RES_A_CQC_BT_NUM, caps->cqc_bt_num);
1814 	hr_reg_write(r_a, FUNC_RES_A_CQC_BT_IDX, vf_id * caps->cqc_bt_num);
1815 	hr_reg_write(r_a, FUNC_RES_A_MPT_BT_NUM, caps->mpt_bt_num);
1816 	hr_reg_write(r_a, FUNC_RES_A_MPT_BT_IDX, vf_id * caps->mpt_bt_num);
1817 	hr_reg_write(r_a, FUNC_RES_A_EQC_BT_NUM, caps->eqc_bt_num);
1818 	hr_reg_write(r_a, FUNC_RES_A_EQC_BT_IDX, vf_id * caps->eqc_bt_num);
1819 	hr_reg_write(r_b, FUNC_RES_V_QID_NUM, caps->sl_num);
1820 	hr_reg_write(r_b, FUNC_RES_B_QID_IDX, vf_id * caps->sl_num);
1821 	hr_reg_write(r_b, FUNC_RES_B_SCCC_BT_NUM, caps->sccc_bt_num);
1822 	hr_reg_write(r_b, FUNC_RES_B_SCCC_BT_IDX, vf_id * caps->sccc_bt_num);
1823 
1824 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
1825 		hr_reg_write(r_b, FUNC_RES_V_GMV_BT_NUM, caps->gmv_bt_num);
1826 		hr_reg_write(r_b, FUNC_RES_B_GMV_BT_IDX,
1827 			     vf_id * caps->gmv_bt_num);
1828 	} else {
1829 		hr_reg_write(r_b, FUNC_RES_B_SGID_NUM, caps->sgid_bt_num);
1830 		hr_reg_write(r_b, FUNC_RES_B_SGID_IDX,
1831 			     vf_id * caps->sgid_bt_num);
1832 		hr_reg_write(r_b, FUNC_RES_B_SMAC_NUM, caps->smac_bt_num);
1833 		hr_reg_write(r_b, FUNC_RES_B_SMAC_IDX,
1834 			     vf_id * caps->smac_bt_num);
1835 	}
1836 
1837 	return hns_roce_cmq_send(hr_dev, desc, 2);
1838 }
1839 
1840 static int hns_roce_alloc_vf_resource(struct hns_roce_dev *hr_dev)
1841 {
1842 	int vf_id;
1843 	int ret;
1844 
1845 	for (vf_id = 0; vf_id < hr_dev->func_num; vf_id++) {
1846 		ret = __hns_roce_alloc_vf_resource(hr_dev, vf_id);
1847 		if (ret)
1848 			return ret;
1849 	}
1850 
1851 	return 0;
1852 }
1853 
1854 static int hns_roce_v2_set_bt(struct hns_roce_dev *hr_dev)
1855 {
1856 	struct hns_roce_cmq_desc desc;
1857 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
1858 	struct hns_roce_caps *caps = &hr_dev->caps;
1859 
1860 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_BT_ATTR, false);
1861 
1862 	hr_reg_write(req, CFG_BT_ATTR_QPC_BA_PGSZ,
1863 		     caps->qpc_ba_pg_sz + PG_SHIFT_OFFSET);
1864 	hr_reg_write(req, CFG_BT_ATTR_QPC_BUF_PGSZ,
1865 		     caps->qpc_buf_pg_sz + PG_SHIFT_OFFSET);
1866 	hr_reg_write(req, CFG_BT_ATTR_QPC_HOPNUM,
1867 		     to_hr_hem_hopnum(caps->qpc_hop_num, caps->num_qps));
1868 
1869 	hr_reg_write(req, CFG_BT_ATTR_SRQC_BA_PGSZ,
1870 		     caps->srqc_ba_pg_sz + PG_SHIFT_OFFSET);
1871 	hr_reg_write(req, CFG_BT_ATTR_SRQC_BUF_PGSZ,
1872 		     caps->srqc_buf_pg_sz + PG_SHIFT_OFFSET);
1873 	hr_reg_write(req, CFG_BT_ATTR_SRQC_HOPNUM,
1874 		     to_hr_hem_hopnum(caps->srqc_hop_num, caps->num_srqs));
1875 
1876 	hr_reg_write(req, CFG_BT_ATTR_CQC_BA_PGSZ,
1877 		     caps->cqc_ba_pg_sz + PG_SHIFT_OFFSET);
1878 	hr_reg_write(req, CFG_BT_ATTR_CQC_BUF_PGSZ,
1879 		     caps->cqc_buf_pg_sz + PG_SHIFT_OFFSET);
1880 	hr_reg_write(req, CFG_BT_ATTR_CQC_HOPNUM,
1881 		     to_hr_hem_hopnum(caps->cqc_hop_num, caps->num_cqs));
1882 
1883 	hr_reg_write(req, CFG_BT_ATTR_MPT_BA_PGSZ,
1884 		     caps->mpt_ba_pg_sz + PG_SHIFT_OFFSET);
1885 	hr_reg_write(req, CFG_BT_ATTR_MPT_BUF_PGSZ,
1886 		     caps->mpt_buf_pg_sz + PG_SHIFT_OFFSET);
1887 	hr_reg_write(req, CFG_BT_ATTR_MPT_HOPNUM,
1888 		     to_hr_hem_hopnum(caps->mpt_hop_num, caps->num_mtpts));
1889 
1890 	hr_reg_write(req, CFG_BT_ATTR_SCCC_BA_PGSZ,
1891 		     caps->sccc_ba_pg_sz + PG_SHIFT_OFFSET);
1892 	hr_reg_write(req, CFG_BT_ATTR_SCCC_BUF_PGSZ,
1893 		     caps->sccc_buf_pg_sz + PG_SHIFT_OFFSET);
1894 	hr_reg_write(req, CFG_BT_ATTR_SCCC_HOPNUM,
1895 		     to_hr_hem_hopnum(caps->sccc_hop_num, caps->num_qps));
1896 
1897 	return hns_roce_cmq_send(hr_dev, &desc, 1);
1898 }
1899 
1900 static void set_default_caps(struct hns_roce_dev *hr_dev)
1901 {
1902 	struct hns_roce_v2_priv *priv = hr_dev->priv;
1903 	struct hns_roce_caps *caps = &hr_dev->caps;
1904 
1905 	caps->num_qps		= HNS_ROCE_V2_MAX_QP_NUM;
1906 	caps->max_wqes		= HNS_ROCE_V2_MAX_WQE_NUM;
1907 	caps->num_cqs		= HNS_ROCE_V2_MAX_CQ_NUM;
1908 	caps->num_srqs		= HNS_ROCE_V2_MAX_SRQ_NUM;
1909 	caps->min_cqes		= HNS_ROCE_MIN_CQE_NUM;
1910 	caps->max_cqes		= HNS_ROCE_V2_MAX_CQE_NUM;
1911 	caps->max_sq_sg		= HNS_ROCE_V2_MAX_SQ_SGE_NUM;
1912 	caps->max_extend_sg	= HNS_ROCE_V2_MAX_EXTEND_SGE_NUM;
1913 	caps->max_rq_sg		= HNS_ROCE_V2_MAX_RQ_SGE_NUM;
1914 	caps->num_uars		= HNS_ROCE_V2_UAR_NUM;
1915 	caps->phy_num_uars	= HNS_ROCE_V2_PHY_UAR_NUM;
1916 	caps->num_aeq_vectors	= HNS_ROCE_V2_AEQE_VEC_NUM;
1917 	caps->num_comp_vectors	=
1918 			min_t(u32, caps->eqc_bt_num - 1,
1919 			      (u32)priv->handle->rinfo.num_vectors - 2);
1920 	caps->num_other_vectors = HNS_ROCE_V2_ABNORMAL_VEC_NUM;
1921 	caps->num_mtpts		= HNS_ROCE_V2_MAX_MTPT_NUM;
1922 	caps->num_mtt_segs	= HNS_ROCE_V2_MAX_MTT_SEGS;
1923 	caps->num_srqwqe_segs	= HNS_ROCE_V2_MAX_SRQWQE_SEGS;
1924 	caps->num_idx_segs	= HNS_ROCE_V2_MAX_IDX_SEGS;
1925 	caps->num_pds		= HNS_ROCE_V2_MAX_PD_NUM;
1926 	caps->num_xrcds		= HNS_ROCE_V2_MAX_XRCD_NUM;
1927 	caps->max_qp_init_rdma	= HNS_ROCE_V2_MAX_QP_INIT_RDMA;
1928 	caps->max_qp_dest_rdma	= HNS_ROCE_V2_MAX_QP_DEST_RDMA;
1929 	caps->max_sq_desc_sz	= HNS_ROCE_V2_MAX_SQ_DESC_SZ;
1930 	caps->max_rq_desc_sz	= HNS_ROCE_V2_MAX_RQ_DESC_SZ;
1931 	caps->max_srq_desc_sz	= HNS_ROCE_V2_MAX_SRQ_DESC_SZ;
1932 	caps->irrl_entry_sz	= HNS_ROCE_V2_IRRL_ENTRY_SZ;
1933 	caps->trrl_entry_sz	= HNS_ROCE_V2_EXT_ATOMIC_TRRL_ENTRY_SZ;
1934 	caps->cqc_entry_sz	= HNS_ROCE_V2_CQC_ENTRY_SZ;
1935 	caps->srqc_entry_sz	= HNS_ROCE_V2_SRQC_ENTRY_SZ;
1936 	caps->mtpt_entry_sz	= HNS_ROCE_V2_MTPT_ENTRY_SZ;
1937 	caps->mtt_entry_sz	= HNS_ROCE_V2_MTT_ENTRY_SZ;
1938 	caps->idx_entry_sz	= HNS_ROCE_V2_IDX_ENTRY_SZ;
1939 	caps->page_size_cap	= HNS_ROCE_V2_PAGE_SIZE_SUPPORTED;
1940 	caps->reserved_lkey	= 0;
1941 	caps->reserved_pds	= 0;
1942 	caps->reserved_xrcds	= HNS_ROCE_V2_RSV_XRCD_NUM;
1943 	caps->reserved_mrws	= 1;
1944 	caps->reserved_uars	= 0;
1945 	caps->reserved_cqs	= 0;
1946 	caps->reserved_srqs	= 0;
1947 	caps->reserved_qps	= HNS_ROCE_V2_RSV_QPS;
1948 
1949 	caps->qpc_hop_num	= HNS_ROCE_CONTEXT_HOP_NUM;
1950 	caps->srqc_hop_num	= HNS_ROCE_CONTEXT_HOP_NUM;
1951 	caps->cqc_hop_num	= HNS_ROCE_CONTEXT_HOP_NUM;
1952 	caps->mpt_hop_num	= HNS_ROCE_CONTEXT_HOP_NUM;
1953 	caps->mtt_hop_num	= HNS_ROCE_MTT_HOP_NUM;
1954 	caps->pbl_hop_num       = HNS_ROCE_PBL_HOP_NUM;
1955 	caps->wqe_sq_hop_num	= HNS_ROCE_SQWQE_HOP_NUM;
1956 	caps->wqe_sge_hop_num	= HNS_ROCE_EXT_SGE_HOP_NUM;
1957 	caps->wqe_rq_hop_num	= HNS_ROCE_RQWQE_HOP_NUM;
1958 	caps->cqe_hop_num	= HNS_ROCE_CQE_HOP_NUM;
1959 	caps->srqwqe_hop_num	= HNS_ROCE_SRQWQE_HOP_NUM;
1960 	caps->idx_hop_num	= HNS_ROCE_IDX_HOP_NUM;
1961 	caps->eqe_hop_num       = HNS_ROCE_EQE_HOP_NUM;
1962 	caps->chunk_sz          = HNS_ROCE_V2_TABLE_CHUNK_SIZE;
1963 
1964 	caps->flags		= HNS_ROCE_CAP_FLAG_REREG_MR |
1965 				  HNS_ROCE_CAP_FLAG_ROCE_V1_V2 |
1966 				  HNS_ROCE_CAP_FLAG_CQ_RECORD_DB |
1967 				  HNS_ROCE_CAP_FLAG_QP_RECORD_DB;
1968 
1969 	caps->pkey_table_len[0] = 1;
1970 	caps->ceqe_depth	= HNS_ROCE_V2_COMP_EQE_NUM;
1971 	caps->aeqe_depth	= HNS_ROCE_V2_ASYNC_EQE_NUM;
1972 	caps->local_ca_ack_delay = 0;
1973 	caps->max_mtu = IB_MTU_4096;
1974 
1975 	caps->max_srq_wrs	= HNS_ROCE_V2_MAX_SRQ_WR;
1976 	caps->max_srq_sges	= HNS_ROCE_V2_MAX_SRQ_SGE;
1977 
1978 	caps->flags |= HNS_ROCE_CAP_FLAG_ATOMIC | HNS_ROCE_CAP_FLAG_MW |
1979 		       HNS_ROCE_CAP_FLAG_SRQ | HNS_ROCE_CAP_FLAG_FRMR |
1980 		       HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL | HNS_ROCE_CAP_FLAG_XRC;
1981 
1982 	caps->num_qpc_timer	  = HNS_ROCE_V2_MAX_QPC_TIMER_NUM;
1983 	caps->qpc_timer_entry_sz  = HNS_ROCE_V2_QPC_TIMER_ENTRY_SZ;
1984 	caps->qpc_timer_hop_num   = HNS_ROCE_HOP_NUM_0;
1985 	caps->num_cqc_timer	  = HNS_ROCE_V2_MAX_CQC_TIMER_NUM;
1986 	caps->cqc_timer_entry_sz  = HNS_ROCE_V2_CQC_TIMER_ENTRY_SZ;
1987 	caps->cqc_timer_hop_num   = HNS_ROCE_HOP_NUM_0;
1988 
1989 	caps->sccc_hop_num	  = HNS_ROCE_SCCC_HOP_NUM;
1990 
1991 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
1992 		caps->aeqe_size = HNS_ROCE_V3_EQE_SIZE;
1993 		caps->ceqe_size = HNS_ROCE_V3_EQE_SIZE;
1994 		caps->cqe_sz = HNS_ROCE_V3_CQE_SIZE;
1995 		caps->qpc_sz = HNS_ROCE_V3_QPC_SZ;
1996 		caps->sccc_sz = HNS_ROCE_V3_SCCC_SZ;
1997 		caps->gmv_entry_sz = HNS_ROCE_V3_GMV_ENTRY_SZ;
1998 		caps->gmv_entry_num = caps->gmv_bt_num * (PAGE_SIZE /
1999 							  caps->gmv_entry_sz);
2000 		caps->gmv_hop_num = HNS_ROCE_HOP_NUM_0;
2001 		caps->gid_table_len[0] = caps->gmv_bt_num * (HNS_HW_PAGE_SIZE /
2002 					 caps->gmv_entry_sz);
2003 		caps->max_sq_inline = HNS_ROCE_V2_MAX_SQ_INL_EXT;
2004 	} else {
2005 		caps->aeqe_size = HNS_ROCE_AEQE_SIZE;
2006 		caps->ceqe_size = HNS_ROCE_CEQE_SIZE;
2007 		caps->cqe_sz = HNS_ROCE_V2_CQE_SIZE;
2008 		caps->qpc_sz = HNS_ROCE_V2_QPC_SZ;
2009 		caps->sccc_sz = HNS_ROCE_V2_SCCC_SZ;
2010 		caps->gid_table_len[0] = HNS_ROCE_V2_GID_INDEX_NUM;
2011 		caps->max_sq_inline = HNS_ROCE_V2_MAX_SQ_INLINE;
2012 	}
2013 }
2014 
2015 static void calc_pg_sz(u32 obj_num, u32 obj_size, u32 hop_num, u32 ctx_bt_num,
2016 		       u32 *buf_page_size, u32 *bt_page_size, u32 hem_type)
2017 {
2018 	u64 obj_per_chunk;
2019 	u64 bt_chunk_size = PAGE_SIZE;
2020 	u64 buf_chunk_size = PAGE_SIZE;
2021 	u64 obj_per_chunk_default = buf_chunk_size / obj_size;
2022 
2023 	*buf_page_size = 0;
2024 	*bt_page_size = 0;
2025 
2026 	switch (hop_num) {
2027 	case 3:
2028 		obj_per_chunk = ctx_bt_num * (bt_chunk_size / BA_BYTE_LEN) *
2029 				(bt_chunk_size / BA_BYTE_LEN) *
2030 				(bt_chunk_size / BA_BYTE_LEN) *
2031 				 obj_per_chunk_default;
2032 		break;
2033 	case 2:
2034 		obj_per_chunk = ctx_bt_num * (bt_chunk_size / BA_BYTE_LEN) *
2035 				(bt_chunk_size / BA_BYTE_LEN) *
2036 				 obj_per_chunk_default;
2037 		break;
2038 	case 1:
2039 		obj_per_chunk = ctx_bt_num * (bt_chunk_size / BA_BYTE_LEN) *
2040 				obj_per_chunk_default;
2041 		break;
2042 	case HNS_ROCE_HOP_NUM_0:
2043 		obj_per_chunk = ctx_bt_num * obj_per_chunk_default;
2044 		break;
2045 	default:
2046 		pr_err("table %u not support hop_num = %u!\n", hem_type,
2047 		       hop_num);
2048 		return;
2049 	}
2050 
2051 	if (hem_type >= HEM_TYPE_MTT)
2052 		*bt_page_size = ilog2(DIV_ROUND_UP(obj_num, obj_per_chunk));
2053 	else
2054 		*buf_page_size = ilog2(DIV_ROUND_UP(obj_num, obj_per_chunk));
2055 }
2056 
2057 static void set_hem_page_size(struct hns_roce_dev *hr_dev)
2058 {
2059 	struct hns_roce_caps *caps = &hr_dev->caps;
2060 
2061 	/* EQ */
2062 	caps->eqe_ba_pg_sz = 0;
2063 	caps->eqe_buf_pg_sz = 0;
2064 
2065 	/* Link Table */
2066 	caps->tsq_buf_pg_sz = 0;
2067 
2068 	/* MR */
2069 	caps->pbl_ba_pg_sz = HNS_ROCE_BA_PG_SZ_SUPPORTED_16K;
2070 	caps->pbl_buf_pg_sz = 0;
2071 	calc_pg_sz(caps->num_mtpts, caps->mtpt_entry_sz, caps->mpt_hop_num,
2072 		   caps->mpt_bt_num, &caps->mpt_buf_pg_sz, &caps->mpt_ba_pg_sz,
2073 		   HEM_TYPE_MTPT);
2074 
2075 	/* QP */
2076 	caps->qpc_timer_ba_pg_sz  = 0;
2077 	caps->qpc_timer_buf_pg_sz = 0;
2078 	caps->mtt_ba_pg_sz = 0;
2079 	caps->mtt_buf_pg_sz = 0;
2080 	calc_pg_sz(caps->num_qps, caps->qpc_sz, caps->qpc_hop_num,
2081 		   caps->qpc_bt_num, &caps->qpc_buf_pg_sz, &caps->qpc_ba_pg_sz,
2082 		   HEM_TYPE_QPC);
2083 
2084 	if (caps->flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL)
2085 		calc_pg_sz(caps->num_qps, caps->sccc_sz, caps->sccc_hop_num,
2086 			   caps->sccc_bt_num, &caps->sccc_buf_pg_sz,
2087 			   &caps->sccc_ba_pg_sz, HEM_TYPE_SCCC);
2088 
2089 	/* CQ */
2090 	calc_pg_sz(caps->num_cqs, caps->cqc_entry_sz, caps->cqc_hop_num,
2091 		   caps->cqc_bt_num, &caps->cqc_buf_pg_sz, &caps->cqc_ba_pg_sz,
2092 		   HEM_TYPE_CQC);
2093 	calc_pg_sz(caps->max_cqes, caps->cqe_sz, caps->cqe_hop_num,
2094 		   1, &caps->cqe_buf_pg_sz, &caps->cqe_ba_pg_sz, HEM_TYPE_CQE);
2095 
2096 	if (caps->cqc_timer_entry_sz)
2097 		calc_pg_sz(caps->num_cqc_timer, caps->cqc_timer_entry_sz,
2098 			   caps->cqc_timer_hop_num, caps->cqc_timer_bt_num,
2099 			   &caps->cqc_timer_buf_pg_sz,
2100 			   &caps->cqc_timer_ba_pg_sz, HEM_TYPE_CQC_TIMER);
2101 
2102 	/* SRQ */
2103 	if (caps->flags & HNS_ROCE_CAP_FLAG_SRQ) {
2104 		calc_pg_sz(caps->num_srqs, caps->srqc_entry_sz,
2105 			   caps->srqc_hop_num, caps->srqc_bt_num,
2106 			   &caps->srqc_buf_pg_sz, &caps->srqc_ba_pg_sz,
2107 			   HEM_TYPE_SRQC);
2108 		calc_pg_sz(caps->num_srqwqe_segs, caps->mtt_entry_sz,
2109 			   caps->srqwqe_hop_num, 1, &caps->srqwqe_buf_pg_sz,
2110 			   &caps->srqwqe_ba_pg_sz, HEM_TYPE_SRQWQE);
2111 		calc_pg_sz(caps->num_idx_segs, caps->idx_entry_sz,
2112 			   caps->idx_hop_num, 1, &caps->idx_buf_pg_sz,
2113 			   &caps->idx_ba_pg_sz, HEM_TYPE_IDX);
2114 	}
2115 
2116 	/* GMV */
2117 	caps->gmv_ba_pg_sz = 0;
2118 	caps->gmv_buf_pg_sz = 0;
2119 }
2120 
2121 static int hns_roce_query_pf_caps(struct hns_roce_dev *hr_dev)
2122 {
2123 	struct hns_roce_cmq_desc desc[HNS_ROCE_QUERY_PF_CAPS_CMD_NUM];
2124 	struct hns_roce_caps *caps = &hr_dev->caps;
2125 	struct hns_roce_query_pf_caps_a *resp_a;
2126 	struct hns_roce_query_pf_caps_b *resp_b;
2127 	struct hns_roce_query_pf_caps_c *resp_c;
2128 	struct hns_roce_query_pf_caps_d *resp_d;
2129 	struct hns_roce_query_pf_caps_e *resp_e;
2130 	int ctx_hop_num;
2131 	int pbl_hop_num;
2132 	int ret;
2133 	int i;
2134 
2135 	for (i = 0; i < HNS_ROCE_QUERY_PF_CAPS_CMD_NUM; i++) {
2136 		hns_roce_cmq_setup_basic_desc(&desc[i],
2137 					      HNS_ROCE_OPC_QUERY_PF_CAPS_NUM,
2138 					      true);
2139 		if (i < (HNS_ROCE_QUERY_PF_CAPS_CMD_NUM - 1))
2140 			desc[i].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
2141 		else
2142 			desc[i].flag &= ~cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
2143 	}
2144 
2145 	ret = hns_roce_cmq_send(hr_dev, desc, HNS_ROCE_QUERY_PF_CAPS_CMD_NUM);
2146 	if (ret)
2147 		return ret;
2148 
2149 	resp_a = (struct hns_roce_query_pf_caps_a *)desc[0].data;
2150 	resp_b = (struct hns_roce_query_pf_caps_b *)desc[1].data;
2151 	resp_c = (struct hns_roce_query_pf_caps_c *)desc[2].data;
2152 	resp_d = (struct hns_roce_query_pf_caps_d *)desc[3].data;
2153 	resp_e = (struct hns_roce_query_pf_caps_e *)desc[4].data;
2154 
2155 	caps->local_ca_ack_delay     = resp_a->local_ca_ack_delay;
2156 	caps->max_sq_sg		     = le16_to_cpu(resp_a->max_sq_sg);
2157 	caps->max_sq_inline	     = le16_to_cpu(resp_a->max_sq_inline);
2158 	caps->max_rq_sg		     = le16_to_cpu(resp_a->max_rq_sg);
2159 	caps->max_rq_sg = roundup_pow_of_two(caps->max_rq_sg);
2160 	caps->max_extend_sg	     = le32_to_cpu(resp_a->max_extend_sg);
2161 	caps->num_qpc_timer	     = le16_to_cpu(resp_a->num_qpc_timer);
2162 	caps->num_cqc_timer	     = le16_to_cpu(resp_a->num_cqc_timer);
2163 	caps->max_srq_sges	     = le16_to_cpu(resp_a->max_srq_sges);
2164 	caps->max_srq_sges = roundup_pow_of_two(caps->max_srq_sges);
2165 	caps->num_aeq_vectors	     = resp_a->num_aeq_vectors;
2166 	caps->num_other_vectors	     = resp_a->num_other_vectors;
2167 	caps->max_sq_desc_sz	     = resp_a->max_sq_desc_sz;
2168 	caps->max_rq_desc_sz	     = resp_a->max_rq_desc_sz;
2169 	caps->max_srq_desc_sz	     = resp_a->max_srq_desc_sz;
2170 	caps->cqe_sz		     = HNS_ROCE_V2_CQE_SIZE;
2171 
2172 	caps->mtpt_entry_sz	     = resp_b->mtpt_entry_sz;
2173 	caps->irrl_entry_sz	     = resp_b->irrl_entry_sz;
2174 	caps->trrl_entry_sz	     = resp_b->trrl_entry_sz;
2175 	caps->cqc_entry_sz	     = resp_b->cqc_entry_sz;
2176 	caps->srqc_entry_sz	     = resp_b->srqc_entry_sz;
2177 	caps->idx_entry_sz	     = resp_b->idx_entry_sz;
2178 	caps->sccc_sz		     = resp_b->sccc_sz;
2179 	caps->max_mtu		     = resp_b->max_mtu;
2180 	caps->qpc_sz		     = HNS_ROCE_V2_QPC_SZ;
2181 	caps->min_cqes		     = resp_b->min_cqes;
2182 	caps->min_wqes		     = resp_b->min_wqes;
2183 	caps->page_size_cap	     = le32_to_cpu(resp_b->page_size_cap);
2184 	caps->pkey_table_len[0]	     = resp_b->pkey_table_len;
2185 	caps->phy_num_uars	     = resp_b->phy_num_uars;
2186 	ctx_hop_num		     = resp_b->ctx_hop_num;
2187 	pbl_hop_num		     = resp_b->pbl_hop_num;
2188 
2189 	caps->num_pds = 1 << roce_get_field(resp_c->cap_flags_num_pds,
2190 					    V2_QUERY_PF_CAPS_C_NUM_PDS_M,
2191 					    V2_QUERY_PF_CAPS_C_NUM_PDS_S);
2192 	caps->flags = roce_get_field(resp_c->cap_flags_num_pds,
2193 				     V2_QUERY_PF_CAPS_C_CAP_FLAGS_M,
2194 				     V2_QUERY_PF_CAPS_C_CAP_FLAGS_S);
2195 	caps->flags |= le16_to_cpu(resp_d->cap_flags_ex) <<
2196 		       HNS_ROCE_CAP_FLAGS_EX_SHIFT;
2197 
2198 	caps->num_cqs = 1 << roce_get_field(resp_c->max_gid_num_cqs,
2199 					    V2_QUERY_PF_CAPS_C_NUM_CQS_M,
2200 					    V2_QUERY_PF_CAPS_C_NUM_CQS_S);
2201 	caps->gid_table_len[0] = roce_get_field(resp_c->max_gid_num_cqs,
2202 						V2_QUERY_PF_CAPS_C_MAX_GID_M,
2203 						V2_QUERY_PF_CAPS_C_MAX_GID_S);
2204 
2205 	caps->gid_table_len[0] /= hr_dev->func_num;
2206 
2207 	caps->max_cqes = 1 << roce_get_field(resp_c->cq_depth,
2208 					     V2_QUERY_PF_CAPS_C_CQ_DEPTH_M,
2209 					     V2_QUERY_PF_CAPS_C_CQ_DEPTH_S);
2210 	caps->num_mtpts = 1 << roce_get_field(resp_c->num_mrws,
2211 					      V2_QUERY_PF_CAPS_C_NUM_MRWS_M,
2212 					      V2_QUERY_PF_CAPS_C_NUM_MRWS_S);
2213 	caps->num_qps = 1 << roce_get_field(resp_c->ord_num_qps,
2214 					    V2_QUERY_PF_CAPS_C_NUM_QPS_M,
2215 					    V2_QUERY_PF_CAPS_C_NUM_QPS_S);
2216 	caps->max_qp_init_rdma = roce_get_field(resp_c->ord_num_qps,
2217 						V2_QUERY_PF_CAPS_C_MAX_ORD_M,
2218 						V2_QUERY_PF_CAPS_C_MAX_ORD_S);
2219 	caps->max_qp_dest_rdma = caps->max_qp_init_rdma;
2220 	caps->max_wqes = 1 << le16_to_cpu(resp_c->sq_depth);
2221 	caps->num_srqs = 1 << roce_get_field(resp_d->wq_hop_num_max_srqs,
2222 					     V2_QUERY_PF_CAPS_D_NUM_SRQS_M,
2223 					     V2_QUERY_PF_CAPS_D_NUM_SRQS_S);
2224 	caps->cong_type = roce_get_field(resp_d->wq_hop_num_max_srqs,
2225 					 V2_QUERY_PF_CAPS_D_CONG_TYPE_M,
2226 					 V2_QUERY_PF_CAPS_D_CONG_TYPE_S);
2227 	caps->max_srq_wrs = 1 << le16_to_cpu(resp_d->srq_depth);
2228 
2229 	caps->ceqe_depth = 1 << roce_get_field(resp_d->num_ceqs_ceq_depth,
2230 					       V2_QUERY_PF_CAPS_D_CEQ_DEPTH_M,
2231 					       V2_QUERY_PF_CAPS_D_CEQ_DEPTH_S);
2232 	caps->num_comp_vectors = roce_get_field(resp_d->num_ceqs_ceq_depth,
2233 						V2_QUERY_PF_CAPS_D_NUM_CEQS_M,
2234 						V2_QUERY_PF_CAPS_D_NUM_CEQS_S);
2235 
2236 	caps->aeqe_depth = 1 << roce_get_field(resp_d->arm_st_aeq_depth,
2237 					       V2_QUERY_PF_CAPS_D_AEQ_DEPTH_M,
2238 					       V2_QUERY_PF_CAPS_D_AEQ_DEPTH_S);
2239 	caps->default_aeq_arm_st = roce_get_field(resp_d->arm_st_aeq_depth,
2240 					    V2_QUERY_PF_CAPS_D_AEQ_ARM_ST_M,
2241 					    V2_QUERY_PF_CAPS_D_AEQ_ARM_ST_S);
2242 	caps->default_ceq_arm_st = roce_get_field(resp_d->arm_st_aeq_depth,
2243 					    V2_QUERY_PF_CAPS_D_CEQ_ARM_ST_M,
2244 					    V2_QUERY_PF_CAPS_D_CEQ_ARM_ST_S);
2245 	caps->reserved_pds = roce_get_field(resp_d->num_uars_rsv_pds,
2246 					    V2_QUERY_PF_CAPS_D_RSV_PDS_M,
2247 					    V2_QUERY_PF_CAPS_D_RSV_PDS_S);
2248 	caps->num_uars = 1 << roce_get_field(resp_d->num_uars_rsv_pds,
2249 					     V2_QUERY_PF_CAPS_D_NUM_UARS_M,
2250 					     V2_QUERY_PF_CAPS_D_NUM_UARS_S);
2251 	caps->reserved_qps = roce_get_field(resp_d->rsv_uars_rsv_qps,
2252 					    V2_QUERY_PF_CAPS_D_RSV_QPS_M,
2253 					    V2_QUERY_PF_CAPS_D_RSV_QPS_S);
2254 	caps->reserved_uars = roce_get_field(resp_d->rsv_uars_rsv_qps,
2255 					     V2_QUERY_PF_CAPS_D_RSV_UARS_M,
2256 					     V2_QUERY_PF_CAPS_D_RSV_UARS_S);
2257 	caps->reserved_mrws = roce_get_field(resp_e->chunk_size_shift_rsv_mrws,
2258 					     V2_QUERY_PF_CAPS_E_RSV_MRWS_M,
2259 					     V2_QUERY_PF_CAPS_E_RSV_MRWS_S);
2260 	caps->chunk_sz = 1 << roce_get_field(resp_e->chunk_size_shift_rsv_mrws,
2261 					 V2_QUERY_PF_CAPS_E_CHUNK_SIZE_SHIFT_M,
2262 					 V2_QUERY_PF_CAPS_E_CHUNK_SIZE_SHIFT_S);
2263 	caps->reserved_cqs = roce_get_field(resp_e->rsv_cqs,
2264 					    V2_QUERY_PF_CAPS_E_RSV_CQS_M,
2265 					    V2_QUERY_PF_CAPS_E_RSV_CQS_S);
2266 	caps->reserved_srqs = roce_get_field(resp_e->rsv_srqs,
2267 					     V2_QUERY_PF_CAPS_E_RSV_SRQS_M,
2268 					     V2_QUERY_PF_CAPS_E_RSV_SRQS_S);
2269 	caps->reserved_lkey = roce_get_field(resp_e->rsv_lkey,
2270 					     V2_QUERY_PF_CAPS_E_RSV_LKEYS_M,
2271 					     V2_QUERY_PF_CAPS_E_RSV_LKEYS_S);
2272 	caps->default_ceq_max_cnt = le16_to_cpu(resp_e->ceq_max_cnt);
2273 	caps->default_ceq_period = le16_to_cpu(resp_e->ceq_period);
2274 	caps->default_aeq_max_cnt = le16_to_cpu(resp_e->aeq_max_cnt);
2275 	caps->default_aeq_period = le16_to_cpu(resp_e->aeq_period);
2276 
2277 	caps->qpc_timer_entry_sz = HNS_ROCE_V2_QPC_TIMER_ENTRY_SZ;
2278 	caps->cqc_timer_entry_sz = HNS_ROCE_V2_CQC_TIMER_ENTRY_SZ;
2279 	caps->mtt_entry_sz = HNS_ROCE_V2_MTT_ENTRY_SZ;
2280 	caps->num_mtt_segs = HNS_ROCE_V2_MAX_MTT_SEGS;
2281 	caps->ceqe_size = HNS_ROCE_CEQE_SIZE;
2282 	caps->aeqe_size = HNS_ROCE_AEQE_SIZE;
2283 	caps->num_xrcds = HNS_ROCE_V2_MAX_XRCD_NUM;
2284 	caps->reserved_xrcds = HNS_ROCE_V2_RSV_XRCD_NUM;
2285 	caps->num_srqwqe_segs = HNS_ROCE_V2_MAX_SRQWQE_SEGS;
2286 	caps->num_idx_segs = HNS_ROCE_V2_MAX_IDX_SEGS;
2287 
2288 	caps->qpc_hop_num = ctx_hop_num;
2289 	caps->srqc_hop_num = ctx_hop_num;
2290 	caps->cqc_hop_num = ctx_hop_num;
2291 	caps->mpt_hop_num = ctx_hop_num;
2292 	caps->mtt_hop_num = pbl_hop_num;
2293 	caps->cqe_hop_num = pbl_hop_num;
2294 	caps->srqwqe_hop_num = pbl_hop_num;
2295 	caps->idx_hop_num = pbl_hop_num;
2296 	caps->wqe_sq_hop_num = roce_get_field(resp_d->wq_hop_num_max_srqs,
2297 					  V2_QUERY_PF_CAPS_D_SQWQE_HOP_NUM_M,
2298 					  V2_QUERY_PF_CAPS_D_SQWQE_HOP_NUM_S);
2299 	caps->wqe_sge_hop_num = roce_get_field(resp_d->wq_hop_num_max_srqs,
2300 					  V2_QUERY_PF_CAPS_D_EX_SGE_HOP_NUM_M,
2301 					  V2_QUERY_PF_CAPS_D_EX_SGE_HOP_NUM_S);
2302 	caps->wqe_rq_hop_num = roce_get_field(resp_d->wq_hop_num_max_srqs,
2303 					  V2_QUERY_PF_CAPS_D_RQWQE_HOP_NUM_M,
2304 					  V2_QUERY_PF_CAPS_D_RQWQE_HOP_NUM_S);
2305 
2306 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
2307 		caps->ceqe_size = HNS_ROCE_V3_EQE_SIZE;
2308 		caps->aeqe_size = HNS_ROCE_V3_EQE_SIZE;
2309 		caps->cqe_sz = HNS_ROCE_V3_CQE_SIZE;
2310 		caps->qpc_sz = HNS_ROCE_V3_QPC_SZ;
2311 		caps->sccc_sz = HNS_ROCE_V3_SCCC_SZ;
2312 		caps->gmv_entry_sz = HNS_ROCE_V3_GMV_ENTRY_SZ;
2313 		caps->gmv_entry_num = caps->gmv_bt_num * (PAGE_SIZE /
2314 						    caps->gmv_entry_sz);
2315 		caps->gmv_hop_num = HNS_ROCE_HOP_NUM_0;
2316 		caps->gid_table_len[0] = caps->gmv_bt_num *
2317 				(HNS_HW_PAGE_SIZE / caps->gmv_entry_sz);
2318 	}
2319 
2320 	caps->qpc_timer_hop_num = HNS_ROCE_HOP_NUM_0;
2321 	caps->cqc_timer_hop_num = HNS_ROCE_HOP_NUM_0;
2322 
2323 	return 0;
2324 }
2325 
2326 static int config_hem_entry_size(struct hns_roce_dev *hr_dev, u32 type, u32 val)
2327 {
2328 	struct hns_roce_cmq_desc desc;
2329 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
2330 
2331 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_ENTRY_SIZE,
2332 				      false);
2333 
2334 	hr_reg_write(req, CFG_HEM_ENTRY_SIZE_TYPE, type);
2335 	hr_reg_write(req, CFG_HEM_ENTRY_SIZE_VALUE, val);
2336 
2337 	return hns_roce_cmq_send(hr_dev, &desc, 1);
2338 }
2339 
2340 static int hns_roce_config_entry_size(struct hns_roce_dev *hr_dev)
2341 {
2342 	struct hns_roce_caps *caps = &hr_dev->caps;
2343 	int ret;
2344 
2345 	if (hr_dev->pci_dev->revision < PCI_REVISION_ID_HIP09)
2346 		return 0;
2347 
2348 	ret = config_hem_entry_size(hr_dev, HNS_ROCE_CFG_QPC_SIZE,
2349 				    caps->qpc_sz);
2350 	if (ret) {
2351 		dev_err(hr_dev->dev, "failed to cfg qpc sz, ret = %d.\n", ret);
2352 		return ret;
2353 	}
2354 
2355 	ret = config_hem_entry_size(hr_dev, HNS_ROCE_CFG_SCCC_SIZE,
2356 				    caps->sccc_sz);
2357 	if (ret)
2358 		dev_err(hr_dev->dev, "failed to cfg sccc sz, ret = %d.\n", ret);
2359 
2360 	return ret;
2361 }
2362 
2363 static int hns_roce_v2_vf_profile(struct hns_roce_dev *hr_dev)
2364 {
2365 	int ret;
2366 
2367 	hr_dev->vendor_part_id = hr_dev->pci_dev->device;
2368 	hr_dev->sys_image_guid = be64_to_cpu(hr_dev->ib_dev.node_guid);
2369 	hr_dev->func_num = 1;
2370 
2371 	ret = hns_roce_query_vf_resource(hr_dev);
2372 	if (ret) {
2373 		dev_err(hr_dev->dev,
2374 			"Query the VF resource fail, ret = %d.\n", ret);
2375 		return ret;
2376 	}
2377 
2378 	set_default_caps(hr_dev);
2379 	set_hem_page_size(hr_dev);
2380 
2381 	ret = hns_roce_v2_set_bt(hr_dev);
2382 	if (ret) {
2383 		dev_err(hr_dev->dev,
2384 			"Configure the VF bt attribute fail, ret = %d.\n",
2385 			ret);
2386 		return ret;
2387 	}
2388 
2389 	return 0;
2390 }
2391 
2392 static int hns_roce_v2_profile(struct hns_roce_dev *hr_dev)
2393 {
2394 	struct hns_roce_caps *caps = &hr_dev->caps;
2395 	int ret;
2396 
2397 	ret = hns_roce_cmq_query_hw_info(hr_dev);
2398 	if (ret) {
2399 		dev_err(hr_dev->dev, "Query hardware version fail, ret = %d.\n",
2400 			ret);
2401 		return ret;
2402 	}
2403 
2404 	ret = hns_roce_query_fw_ver(hr_dev);
2405 	if (ret) {
2406 		dev_err(hr_dev->dev, "Query firmware version fail, ret = %d.\n",
2407 			ret);
2408 		return ret;
2409 	}
2410 
2411 	if (hr_dev->is_vf)
2412 		return hns_roce_v2_vf_profile(hr_dev);
2413 
2414 	ret = hns_roce_query_func_info(hr_dev);
2415 	if (ret) {
2416 		dev_err(hr_dev->dev, "Query function info fail, ret = %d.\n",
2417 			ret);
2418 		return ret;
2419 	}
2420 
2421 	ret = hns_roce_config_global_param(hr_dev);
2422 	if (ret) {
2423 		dev_err(hr_dev->dev, "Configure global param fail, ret = %d.\n",
2424 			ret);
2425 		return ret;
2426 	}
2427 
2428 	/* Get pf resource owned by every pf */
2429 	ret = hns_roce_query_pf_resource(hr_dev);
2430 	if (ret) {
2431 		dev_err(hr_dev->dev, "Query pf resource fail, ret = %d.\n",
2432 			ret);
2433 		return ret;
2434 	}
2435 
2436 	ret = hns_roce_query_pf_timer_resource(hr_dev);
2437 	if (ret) {
2438 		dev_err(hr_dev->dev,
2439 			"failed to query pf timer resource, ret = %d.\n", ret);
2440 		return ret;
2441 	}
2442 
2443 	ret = hns_roce_set_vf_switch_param(hr_dev);
2444 	if (ret) {
2445 		dev_err(hr_dev->dev,
2446 			"failed to set function switch param, ret = %d.\n",
2447 			ret);
2448 		return ret;
2449 	}
2450 
2451 	hr_dev->vendor_part_id = hr_dev->pci_dev->device;
2452 	hr_dev->sys_image_guid = be64_to_cpu(hr_dev->ib_dev.node_guid);
2453 
2454 	caps->pbl_hop_num	= HNS_ROCE_PBL_HOP_NUM;
2455 	caps->eqe_hop_num	= HNS_ROCE_EQE_HOP_NUM;
2456 
2457 	ret = hns_roce_query_pf_caps(hr_dev);
2458 	if (ret)
2459 		set_default_caps(hr_dev);
2460 
2461 	ret = hns_roce_alloc_vf_resource(hr_dev);
2462 	if (ret) {
2463 		dev_err(hr_dev->dev, "Allocate vf resource fail, ret = %d.\n",
2464 			ret);
2465 		return ret;
2466 	}
2467 
2468 	set_hem_page_size(hr_dev);
2469 	ret = hns_roce_v2_set_bt(hr_dev);
2470 	if (ret) {
2471 		dev_err(hr_dev->dev,
2472 			"Configure bt attribute fail, ret = %d.\n", ret);
2473 		return ret;
2474 	}
2475 
2476 	/* Configure the size of QPC, SCCC, etc. */
2477 	ret = hns_roce_config_entry_size(hr_dev);
2478 
2479 	return ret;
2480 }
2481 
2482 static int hns_roce_config_link_table(struct hns_roce_dev *hr_dev,
2483 				      enum hns_roce_link_table_type type)
2484 {
2485 	struct hns_roce_cmq_desc desc[2];
2486 	struct hns_roce_cfg_llm_a *req_a =
2487 				(struct hns_roce_cfg_llm_a *)desc[0].data;
2488 	struct hns_roce_cfg_llm_b *req_b =
2489 				(struct hns_roce_cfg_llm_b *)desc[1].data;
2490 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2491 	struct hns_roce_link_table *link_tbl;
2492 	struct hns_roce_link_table_entry *entry;
2493 	enum hns_roce_opcode_type opcode;
2494 	u32 page_num;
2495 
2496 	switch (type) {
2497 	case TSQ_LINK_TABLE:
2498 		link_tbl = &priv->tsq;
2499 		opcode = HNS_ROCE_OPC_CFG_EXT_LLM;
2500 		break;
2501 	case TPQ_LINK_TABLE:
2502 		link_tbl = &priv->tpq;
2503 		opcode = HNS_ROCE_OPC_CFG_TMOUT_LLM;
2504 		break;
2505 	default:
2506 		return -EINVAL;
2507 	}
2508 
2509 	page_num = link_tbl->npages;
2510 	entry = link_tbl->table.buf;
2511 
2512 	hns_roce_cmq_setup_basic_desc(&desc[0], opcode, false);
2513 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
2514 
2515 	hns_roce_cmq_setup_basic_desc(&desc[1], opcode, false);
2516 
2517 	req_a->base_addr_l = cpu_to_le32(link_tbl->table.map & 0xffffffff);
2518 	req_a->base_addr_h = cpu_to_le32(link_tbl->table.map >> 32);
2519 	roce_set_field(req_a->depth_pgsz_init_en, CFG_LLM_QUE_DEPTH_M,
2520 		       CFG_LLM_QUE_DEPTH_S, link_tbl->npages);
2521 	roce_set_field(req_a->depth_pgsz_init_en, CFG_LLM_QUE_PGSZ_M,
2522 		       CFG_LLM_QUE_PGSZ_S, link_tbl->pg_sz);
2523 	roce_set_field(req_a->depth_pgsz_init_en, CFG_LLM_INIT_EN_M,
2524 		       CFG_LLM_INIT_EN_S, 1);
2525 	req_a->head_ba_l = cpu_to_le32(entry[0].blk_ba0);
2526 	req_a->head_ba_h_nxtptr = cpu_to_le32(entry[0].blk_ba1_nxt_ptr);
2527 	roce_set_field(req_a->head_ptr, CFG_LLM_HEAD_PTR_M, CFG_LLM_HEAD_PTR_S,
2528 		       0);
2529 
2530 	req_b->tail_ba_l = cpu_to_le32(entry[page_num - 1].blk_ba0);
2531 	roce_set_field(req_b->tail_ba_h, CFG_LLM_TAIL_BA_H_M,
2532 		       CFG_LLM_TAIL_BA_H_S,
2533 		       entry[page_num - 1].blk_ba1_nxt_ptr &
2534 		       HNS_ROCE_LINK_TABLE_BA1_M);
2535 	roce_set_field(req_b->tail_ptr, CFG_LLM_TAIL_PTR_M, CFG_LLM_TAIL_PTR_S,
2536 		       (entry[page_num - 2].blk_ba1_nxt_ptr &
2537 			HNS_ROCE_LINK_TABLE_NXT_PTR_M) >>
2538 			HNS_ROCE_LINK_TABLE_NXT_PTR_S);
2539 
2540 	return hns_roce_cmq_send(hr_dev, desc, 2);
2541 }
2542 
2543 static int hns_roce_init_link_table(struct hns_roce_dev *hr_dev,
2544 				    enum hns_roce_link_table_type type)
2545 {
2546 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2547 	struct hns_roce_link_table *link_tbl;
2548 	struct hns_roce_link_table_entry *entry;
2549 	struct device *dev = hr_dev->dev;
2550 	u32 buf_chk_sz;
2551 	dma_addr_t t;
2552 	int func_num = 1;
2553 	u32 pg_num_a;
2554 	u32 pg_num_b;
2555 	u32 pg_num;
2556 	u32 size;
2557 	int i;
2558 
2559 	switch (type) {
2560 	case TSQ_LINK_TABLE:
2561 		link_tbl = &priv->tsq;
2562 		buf_chk_sz = 1 << (hr_dev->caps.tsq_buf_pg_sz + PAGE_SHIFT);
2563 		pg_num_a = hr_dev->caps.num_qps * 8 / buf_chk_sz;
2564 		pg_num_b = hr_dev->caps.sl_num * 4 + 2;
2565 		break;
2566 	case TPQ_LINK_TABLE:
2567 		link_tbl = &priv->tpq;
2568 		buf_chk_sz = 1 << (hr_dev->caps.tpq_buf_pg_sz +	PAGE_SHIFT);
2569 		pg_num_a = hr_dev->caps.num_cqs * 4 / buf_chk_sz;
2570 		pg_num_b = 2 * 4 * func_num + 2;
2571 		break;
2572 	default:
2573 		return -EINVAL;
2574 	}
2575 
2576 	pg_num = max(pg_num_a, pg_num_b);
2577 	size = pg_num * sizeof(struct hns_roce_link_table_entry);
2578 
2579 	link_tbl->table.buf = dma_alloc_coherent(dev, size,
2580 						 &link_tbl->table.map,
2581 						 GFP_KERNEL);
2582 	if (!link_tbl->table.buf)
2583 		goto out;
2584 
2585 	link_tbl->pg_list = kcalloc(pg_num, sizeof(*link_tbl->pg_list),
2586 				    GFP_KERNEL);
2587 	if (!link_tbl->pg_list)
2588 		goto err_kcalloc_failed;
2589 
2590 	entry = link_tbl->table.buf;
2591 	for (i = 0; i < pg_num; ++i) {
2592 		link_tbl->pg_list[i].buf = dma_alloc_coherent(dev, buf_chk_sz,
2593 							      &t, GFP_KERNEL);
2594 		if (!link_tbl->pg_list[i].buf)
2595 			goto err_alloc_buf_failed;
2596 
2597 		link_tbl->pg_list[i].map = t;
2598 
2599 		entry[i].blk_ba0 = (u32)(t >> 12);
2600 		entry[i].blk_ba1_nxt_ptr = (u32)(t >> 44);
2601 
2602 		if (i < (pg_num - 1))
2603 			entry[i].blk_ba1_nxt_ptr |=
2604 				(i + 1) << HNS_ROCE_LINK_TABLE_NXT_PTR_S;
2605 	}
2606 	link_tbl->npages = pg_num;
2607 	link_tbl->pg_sz = buf_chk_sz;
2608 
2609 	return hns_roce_config_link_table(hr_dev, type);
2610 
2611 err_alloc_buf_failed:
2612 	for (i -= 1; i >= 0; i--)
2613 		dma_free_coherent(dev, buf_chk_sz,
2614 				  link_tbl->pg_list[i].buf,
2615 				  link_tbl->pg_list[i].map);
2616 	kfree(link_tbl->pg_list);
2617 
2618 err_kcalloc_failed:
2619 	dma_free_coherent(dev, size, link_tbl->table.buf,
2620 			  link_tbl->table.map);
2621 
2622 out:
2623 	return -ENOMEM;
2624 }
2625 
2626 static void hns_roce_free_link_table(struct hns_roce_dev *hr_dev,
2627 				     struct hns_roce_link_table *link_tbl)
2628 {
2629 	struct device *dev = hr_dev->dev;
2630 	int size;
2631 	int i;
2632 
2633 	size = link_tbl->npages * sizeof(struct hns_roce_link_table_entry);
2634 
2635 	for (i = 0; i < link_tbl->npages; ++i)
2636 		if (link_tbl->pg_list[i].buf)
2637 			dma_free_coherent(dev, link_tbl->pg_sz,
2638 					  link_tbl->pg_list[i].buf,
2639 					  link_tbl->pg_list[i].map);
2640 	kfree(link_tbl->pg_list);
2641 
2642 	dma_free_coherent(dev, size, link_tbl->table.buf,
2643 			  link_tbl->table.map);
2644 }
2645 
2646 static void free_dip_list(struct hns_roce_dev *hr_dev)
2647 {
2648 	struct hns_roce_dip *hr_dip;
2649 	struct hns_roce_dip *tmp;
2650 	unsigned long flags;
2651 
2652 	spin_lock_irqsave(&hr_dev->dip_list_lock, flags);
2653 
2654 	list_for_each_entry_safe(hr_dip, tmp, &hr_dev->dip_list, node) {
2655 		list_del(&hr_dip->node);
2656 		kfree(hr_dip);
2657 	}
2658 
2659 	spin_unlock_irqrestore(&hr_dev->dip_list_lock, flags);
2660 }
2661 
2662 static int get_hem_table(struct hns_roce_dev *hr_dev)
2663 {
2664 	unsigned int qpc_count;
2665 	unsigned int cqc_count;
2666 	unsigned int gmv_count;
2667 	int ret;
2668 	int i;
2669 
2670 	/* Alloc memory for source address table buffer space chunk */
2671 	for (gmv_count = 0; gmv_count < hr_dev->caps.gmv_entry_num;
2672 	     gmv_count++) {
2673 		ret = hns_roce_table_get(hr_dev, &hr_dev->gmv_table, gmv_count);
2674 		if (ret)
2675 			goto err_gmv_failed;
2676 	}
2677 
2678 	if (hr_dev->is_vf)
2679 		return 0;
2680 
2681 	/* Alloc memory for QPC Timer buffer space chunk */
2682 	for (qpc_count = 0; qpc_count < hr_dev->caps.qpc_timer_bt_num;
2683 	     qpc_count++) {
2684 		ret = hns_roce_table_get(hr_dev, &hr_dev->qpc_timer_table,
2685 					 qpc_count);
2686 		if (ret) {
2687 			dev_err(hr_dev->dev, "QPC Timer get failed\n");
2688 			goto err_qpc_timer_failed;
2689 		}
2690 	}
2691 
2692 	/* Alloc memory for CQC Timer buffer space chunk */
2693 	for (cqc_count = 0; cqc_count < hr_dev->caps.cqc_timer_bt_num;
2694 	     cqc_count++) {
2695 		ret = hns_roce_table_get(hr_dev, &hr_dev->cqc_timer_table,
2696 					 cqc_count);
2697 		if (ret) {
2698 			dev_err(hr_dev->dev, "CQC Timer get failed\n");
2699 			goto err_cqc_timer_failed;
2700 		}
2701 	}
2702 
2703 	return 0;
2704 
2705 err_cqc_timer_failed:
2706 	for (i = 0; i < cqc_count; i++)
2707 		hns_roce_table_put(hr_dev, &hr_dev->cqc_timer_table, i);
2708 
2709 err_qpc_timer_failed:
2710 	for (i = 0; i < qpc_count; i++)
2711 		hns_roce_table_put(hr_dev, &hr_dev->qpc_timer_table, i);
2712 
2713 err_gmv_failed:
2714 	for (i = 0; i < gmv_count; i++)
2715 		hns_roce_table_put(hr_dev, &hr_dev->gmv_table, i);
2716 
2717 	return ret;
2718 }
2719 
2720 static void put_hem_table(struct hns_roce_dev *hr_dev)
2721 {
2722 	int i;
2723 
2724 	for (i = 0; i < hr_dev->caps.gmv_entry_num; i++)
2725 		hns_roce_table_put(hr_dev, &hr_dev->gmv_table, i);
2726 
2727 	if (hr_dev->is_vf)
2728 		return;
2729 
2730 	for (i = 0; i < hr_dev->caps.qpc_timer_bt_num; i++)
2731 		hns_roce_table_put(hr_dev, &hr_dev->qpc_timer_table, i);
2732 
2733 	for (i = 0; i < hr_dev->caps.cqc_timer_bt_num; i++)
2734 		hns_roce_table_put(hr_dev, &hr_dev->cqc_timer_table, i);
2735 }
2736 
2737 static int hns_roce_v2_init(struct hns_roce_dev *hr_dev)
2738 {
2739 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2740 	int ret;
2741 
2742 	ret = get_hem_table(hr_dev);
2743 	if (ret)
2744 		return ret;
2745 
2746 	if (hr_dev->is_vf)
2747 		return 0;
2748 
2749 	/* TSQ includes SQ doorbell and ack doorbell */
2750 	ret = hns_roce_init_link_table(hr_dev, TSQ_LINK_TABLE);
2751 	if (ret) {
2752 		dev_err(hr_dev->dev, "failed to init TSQ, ret = %d.\n", ret);
2753 		goto err_tsq_init_failed;
2754 	}
2755 
2756 	ret = hns_roce_init_link_table(hr_dev, TPQ_LINK_TABLE);
2757 	if (ret) {
2758 		dev_err(hr_dev->dev, "failed to init TPQ, ret = %d.\n", ret);
2759 		goto err_tpq_init_failed;
2760 	}
2761 
2762 	return 0;
2763 
2764 err_tsq_init_failed:
2765 	put_hem_table(hr_dev);
2766 
2767 err_tpq_init_failed:
2768 	hns_roce_free_link_table(hr_dev, &priv->tpq);
2769 
2770 	return ret;
2771 }
2772 
2773 static void hns_roce_v2_exit(struct hns_roce_dev *hr_dev)
2774 {
2775 	struct hns_roce_v2_priv *priv = hr_dev->priv;
2776 
2777 	hns_roce_function_clear(hr_dev);
2778 
2779 	if (!hr_dev->is_vf) {
2780 		hns_roce_free_link_table(hr_dev, &priv->tpq);
2781 		hns_roce_free_link_table(hr_dev, &priv->tsq);
2782 	}
2783 
2784 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP09)
2785 		free_dip_list(hr_dev);
2786 }
2787 
2788 static int hns_roce_mbox_post(struct hns_roce_dev *hr_dev, u64 in_param,
2789 			      u64 out_param, u32 in_modifier, u8 op_modifier,
2790 			      u16 op, u16 token, int event)
2791 {
2792 	struct hns_roce_cmq_desc desc;
2793 	struct hns_roce_post_mbox *mb = (struct hns_roce_post_mbox *)desc.data;
2794 
2795 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_POST_MB, false);
2796 
2797 	mb->in_param_l = cpu_to_le32(in_param);
2798 	mb->in_param_h = cpu_to_le32(in_param >> 32);
2799 	mb->out_param_l = cpu_to_le32(out_param);
2800 	mb->out_param_h = cpu_to_le32(out_param >> 32);
2801 	mb->cmd_tag = cpu_to_le32(in_modifier << 8 | op);
2802 	mb->token_event_en = cpu_to_le32(event << 16 | token);
2803 
2804 	return hns_roce_cmq_send(hr_dev, &desc, 1);
2805 }
2806 
2807 static int v2_wait_mbox_complete(struct hns_roce_dev *hr_dev, u32 timeout,
2808 				 u8 *complete_status)
2809 {
2810 	struct hns_roce_mbox_status *mb_st;
2811 	struct hns_roce_cmq_desc desc;
2812 	unsigned long end;
2813 	int ret = -EBUSY;
2814 	u32 status;
2815 	bool busy;
2816 
2817 	mb_st = (struct hns_roce_mbox_status *)desc.data;
2818 	end = msecs_to_jiffies(timeout) + jiffies;
2819 	while (v2_chk_mbox_is_avail(hr_dev, &busy)) {
2820 		status = 0;
2821 		hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_MB_ST,
2822 					      true);
2823 		ret = __hns_roce_cmq_send(hr_dev, &desc, 1);
2824 		if (!ret) {
2825 			status = le32_to_cpu(mb_st->mb_status_hw_run);
2826 			/* No pending message exists in ROCEE mbox. */
2827 			if (!(status & MB_ST_HW_RUN_M))
2828 				break;
2829 		} else if (!v2_chk_mbox_is_avail(hr_dev, &busy)) {
2830 			break;
2831 		}
2832 
2833 		if (time_after(jiffies, end)) {
2834 			dev_err_ratelimited(hr_dev->dev,
2835 					    "failed to wait mbox status 0x%x\n",
2836 					    status);
2837 			return -ETIMEDOUT;
2838 		}
2839 
2840 		cond_resched();
2841 		ret = -EBUSY;
2842 	}
2843 
2844 	if (!ret) {
2845 		*complete_status = (u8)(status & MB_ST_COMPLETE_M);
2846 	} else if (!v2_chk_mbox_is_avail(hr_dev, &busy)) {
2847 		/* Ignore all errors if the mbox is unavailable. */
2848 		ret = 0;
2849 		*complete_status = MB_ST_COMPLETE_M;
2850 	}
2851 
2852 	return ret;
2853 }
2854 
2855 static int v2_post_mbox(struct hns_roce_dev *hr_dev, u64 in_param,
2856 			u64 out_param, u32 in_modifier, u8 op_modifier,
2857 			u16 op, u16 token, int event)
2858 {
2859 	u8 status = 0;
2860 	int ret;
2861 
2862 	/* Waiting for the mbox to be idle */
2863 	ret = v2_wait_mbox_complete(hr_dev, HNS_ROCE_V2_GO_BIT_TIMEOUT_MSECS,
2864 				    &status);
2865 	if (unlikely(ret)) {
2866 		dev_err_ratelimited(hr_dev->dev,
2867 				    "failed to check post mbox status = 0x%x, ret = %d.\n",
2868 				    status, ret);
2869 		return ret;
2870 	}
2871 
2872 	/* Post new message to mbox */
2873 	ret = hns_roce_mbox_post(hr_dev, in_param, out_param, in_modifier,
2874 				 op_modifier, op, token, event);
2875 	if (ret)
2876 		dev_err_ratelimited(hr_dev->dev,
2877 				    "failed to post mailbox, ret = %d.\n", ret);
2878 
2879 	return ret;
2880 }
2881 
2882 static int v2_poll_mbox_done(struct hns_roce_dev *hr_dev, unsigned int timeout)
2883 {
2884 	u8 status = 0;
2885 	int ret;
2886 
2887 	ret = v2_wait_mbox_complete(hr_dev, timeout, &status);
2888 	if (!ret) {
2889 		if (status != MB_ST_COMPLETE_SUCC)
2890 			return -EBUSY;
2891 	} else {
2892 		dev_err_ratelimited(hr_dev->dev,
2893 				    "failed to check mbox status = 0x%x, ret = %d.\n",
2894 				    status, ret);
2895 	}
2896 
2897 	return ret;
2898 }
2899 
2900 static void copy_gid(void *dest, const union ib_gid *gid)
2901 {
2902 #define GID_SIZE 4
2903 	const union ib_gid *src = gid;
2904 	__le32 (*p)[GID_SIZE] = dest;
2905 	int i;
2906 
2907 	if (!gid)
2908 		src = &zgid;
2909 
2910 	for (i = 0; i < GID_SIZE; i++)
2911 		(*p)[i] = cpu_to_le32(*(u32 *)&src->raw[i * sizeof(u32)]);
2912 }
2913 
2914 static int config_sgid_table(struct hns_roce_dev *hr_dev,
2915 			     int gid_index, const union ib_gid *gid,
2916 			     enum hns_roce_sgid_type sgid_type)
2917 {
2918 	struct hns_roce_cmq_desc desc;
2919 	struct hns_roce_cfg_sgid_tb *sgid_tb =
2920 				    (struct hns_roce_cfg_sgid_tb *)desc.data;
2921 
2922 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_SGID_TB, false);
2923 
2924 	roce_set_field(sgid_tb->table_idx_rsv, CFG_SGID_TB_TABLE_IDX_M,
2925 		       CFG_SGID_TB_TABLE_IDX_S, gid_index);
2926 	roce_set_field(sgid_tb->vf_sgid_type_rsv, CFG_SGID_TB_VF_SGID_TYPE_M,
2927 		       CFG_SGID_TB_VF_SGID_TYPE_S, sgid_type);
2928 
2929 	copy_gid(&sgid_tb->vf_sgid_l, gid);
2930 
2931 	return hns_roce_cmq_send(hr_dev, &desc, 1);
2932 }
2933 
2934 static int config_gmv_table(struct hns_roce_dev *hr_dev,
2935 			    int gid_index, const union ib_gid *gid,
2936 			    enum hns_roce_sgid_type sgid_type,
2937 			    const struct ib_gid_attr *attr)
2938 {
2939 	struct hns_roce_cmq_desc desc[2];
2940 	struct hns_roce_cfg_gmv_tb_a *tb_a =
2941 				(struct hns_roce_cfg_gmv_tb_a *)desc[0].data;
2942 	struct hns_roce_cfg_gmv_tb_b *tb_b =
2943 				(struct hns_roce_cfg_gmv_tb_b *)desc[1].data;
2944 
2945 	u16 vlan_id = VLAN_CFI_MASK;
2946 	u8 mac[ETH_ALEN] = {};
2947 	int ret;
2948 
2949 	if (gid) {
2950 		ret = rdma_read_gid_l2_fields(attr, &vlan_id, mac);
2951 		if (ret)
2952 			return ret;
2953 	}
2954 
2955 	hns_roce_cmq_setup_basic_desc(&desc[0], HNS_ROCE_OPC_CFG_GMV_TBL, false);
2956 	desc[0].flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_NEXT);
2957 
2958 	hns_roce_cmq_setup_basic_desc(&desc[1], HNS_ROCE_OPC_CFG_GMV_TBL, false);
2959 
2960 	copy_gid(&tb_a->vf_sgid_l, gid);
2961 
2962 	roce_set_field(tb_a->vf_sgid_type_vlan, CFG_GMV_TB_VF_SGID_TYPE_M,
2963 		       CFG_GMV_TB_VF_SGID_TYPE_S, sgid_type);
2964 	roce_set_bit(tb_a->vf_sgid_type_vlan, CFG_GMV_TB_VF_VLAN_EN_S,
2965 		     vlan_id < VLAN_CFI_MASK);
2966 	roce_set_field(tb_a->vf_sgid_type_vlan, CFG_GMV_TB_VF_VLAN_ID_M,
2967 		       CFG_GMV_TB_VF_VLAN_ID_S, vlan_id);
2968 
2969 	tb_b->vf_smac_l = cpu_to_le32(*(u32 *)mac);
2970 	roce_set_field(tb_b->vf_smac_h, CFG_GMV_TB_SMAC_H_M,
2971 		       CFG_GMV_TB_SMAC_H_S, *(u16 *)&mac[4]);
2972 
2973 	roce_set_field(tb_b->table_idx_rsv, CFG_GMV_TB_SGID_IDX_M,
2974 		       CFG_GMV_TB_SGID_IDX_S, gid_index);
2975 
2976 	return hns_roce_cmq_send(hr_dev, desc, 2);
2977 }
2978 
2979 static int hns_roce_v2_set_gid(struct hns_roce_dev *hr_dev, u32 port,
2980 			       int gid_index, const union ib_gid *gid,
2981 			       const struct ib_gid_attr *attr)
2982 {
2983 	enum hns_roce_sgid_type sgid_type = GID_TYPE_FLAG_ROCE_V1;
2984 	int ret;
2985 
2986 	if (gid) {
2987 		if (attr->gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
2988 			if (ipv6_addr_v4mapped((void *)gid))
2989 				sgid_type = GID_TYPE_FLAG_ROCE_V2_IPV4;
2990 			else
2991 				sgid_type = GID_TYPE_FLAG_ROCE_V2_IPV6;
2992 		} else if (attr->gid_type == IB_GID_TYPE_ROCE) {
2993 			sgid_type = GID_TYPE_FLAG_ROCE_V1;
2994 		}
2995 	}
2996 
2997 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
2998 		ret = config_gmv_table(hr_dev, gid_index, gid, sgid_type, attr);
2999 	else
3000 		ret = config_sgid_table(hr_dev, gid_index, gid, sgid_type);
3001 
3002 	if (ret)
3003 		ibdev_err(&hr_dev->ib_dev, "failed to set gid, ret = %d!\n",
3004 			  ret);
3005 
3006 	return ret;
3007 }
3008 
3009 static int hns_roce_v2_set_mac(struct hns_roce_dev *hr_dev, u8 phy_port,
3010 			       u8 *addr)
3011 {
3012 	struct hns_roce_cmq_desc desc;
3013 	struct hns_roce_cfg_smac_tb *smac_tb =
3014 				    (struct hns_roce_cfg_smac_tb *)desc.data;
3015 	u16 reg_smac_h;
3016 	u32 reg_smac_l;
3017 
3018 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_SMAC_TB, false);
3019 
3020 	reg_smac_l = *(u32 *)(&addr[0]);
3021 	reg_smac_h = *(u16 *)(&addr[4]);
3022 
3023 	roce_set_field(smac_tb->tb_idx_rsv, CFG_SMAC_TB_IDX_M,
3024 		       CFG_SMAC_TB_IDX_S, phy_port);
3025 	roce_set_field(smac_tb->vf_smac_h_rsv, CFG_SMAC_TB_VF_SMAC_H_M,
3026 		       CFG_SMAC_TB_VF_SMAC_H_S, reg_smac_h);
3027 	smac_tb->vf_smac_l = cpu_to_le32(reg_smac_l);
3028 
3029 	return hns_roce_cmq_send(hr_dev, &desc, 1);
3030 }
3031 
3032 static int set_mtpt_pbl(struct hns_roce_dev *hr_dev,
3033 			struct hns_roce_v2_mpt_entry *mpt_entry,
3034 			struct hns_roce_mr *mr)
3035 {
3036 	u64 pages[HNS_ROCE_V2_MAX_INNER_MTPT_NUM] = { 0 };
3037 	struct ib_device *ibdev = &hr_dev->ib_dev;
3038 	dma_addr_t pbl_ba;
3039 	int i, count;
3040 
3041 	count = hns_roce_mtr_find(hr_dev, &mr->pbl_mtr, 0, pages,
3042 				  ARRAY_SIZE(pages), &pbl_ba);
3043 	if (count < 1) {
3044 		ibdev_err(ibdev, "failed to find PBL mtr, count = %d.\n",
3045 			  count);
3046 		return -ENOBUFS;
3047 	}
3048 
3049 	/* Aligned to the hardware address access unit */
3050 	for (i = 0; i < count; i++)
3051 		pages[i] >>= 6;
3052 
3053 	mpt_entry->pbl_size = cpu_to_le32(mr->npages);
3054 	mpt_entry->pbl_ba_l = cpu_to_le32(pbl_ba >> 3);
3055 	roce_set_field(mpt_entry->byte_48_mode_ba,
3056 		       V2_MPT_BYTE_48_PBL_BA_H_M, V2_MPT_BYTE_48_PBL_BA_H_S,
3057 		       upper_32_bits(pbl_ba >> 3));
3058 
3059 	mpt_entry->pa0_l = cpu_to_le32(lower_32_bits(pages[0]));
3060 	roce_set_field(mpt_entry->byte_56_pa0_h, V2_MPT_BYTE_56_PA0_H_M,
3061 		       V2_MPT_BYTE_56_PA0_H_S, upper_32_bits(pages[0]));
3062 
3063 	mpt_entry->pa1_l = cpu_to_le32(lower_32_bits(pages[1]));
3064 	roce_set_field(mpt_entry->byte_64_buf_pa1, V2_MPT_BYTE_64_PA1_H_M,
3065 		       V2_MPT_BYTE_64_PA1_H_S, upper_32_bits(pages[1]));
3066 	roce_set_field(mpt_entry->byte_64_buf_pa1,
3067 		       V2_MPT_BYTE_64_PBL_BUF_PG_SZ_M,
3068 		       V2_MPT_BYTE_64_PBL_BUF_PG_SZ_S,
3069 		       to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.buf_pg_shift));
3070 
3071 	return 0;
3072 }
3073 
3074 static int hns_roce_v2_write_mtpt(struct hns_roce_dev *hr_dev,
3075 				  void *mb_buf, struct hns_roce_mr *mr,
3076 				  unsigned long mtpt_idx)
3077 {
3078 	struct hns_roce_v2_mpt_entry *mpt_entry;
3079 	int ret;
3080 
3081 	mpt_entry = mb_buf;
3082 	memset(mpt_entry, 0, sizeof(*mpt_entry));
3083 
3084 	hr_reg_write(mpt_entry, MPT_ST, V2_MPT_ST_VALID);
3085 	hr_reg_write(mpt_entry, MPT_PD, mr->pd);
3086 	hr_reg_enable(mpt_entry, MPT_L_INV_EN);
3087 
3088 	hr_reg_write(mpt_entry, MPT_BIND_EN,
3089 		     !!(mr->access & IB_ACCESS_MW_BIND));
3090 	hr_reg_write(mpt_entry, MPT_ATOMIC_EN,
3091 		     !!(mr->access & IB_ACCESS_REMOTE_ATOMIC));
3092 	hr_reg_write(mpt_entry, MPT_RR_EN,
3093 		     !!(mr->access & IB_ACCESS_REMOTE_READ));
3094 	hr_reg_write(mpt_entry, MPT_RW_EN,
3095 		     !!(mr->access & IB_ACCESS_REMOTE_WRITE));
3096 	hr_reg_write(mpt_entry, MPT_LW_EN,
3097 		     !!((mr->access & IB_ACCESS_LOCAL_WRITE)));
3098 
3099 	mpt_entry->len_l = cpu_to_le32(lower_32_bits(mr->size));
3100 	mpt_entry->len_h = cpu_to_le32(upper_32_bits(mr->size));
3101 	mpt_entry->lkey = cpu_to_le32(mr->key);
3102 	mpt_entry->va_l = cpu_to_le32(lower_32_bits(mr->iova));
3103 	mpt_entry->va_h = cpu_to_le32(upper_32_bits(mr->iova));
3104 
3105 	if (mr->type != MR_TYPE_MR)
3106 		hr_reg_enable(mpt_entry, MPT_PA);
3107 
3108 	if (mr->type == MR_TYPE_DMA)
3109 		return 0;
3110 
3111 	if (mr->pbl_hop_num != HNS_ROCE_HOP_NUM_0)
3112 		hr_reg_write(mpt_entry, MPT_PBL_HOP_NUM, mr->pbl_hop_num);
3113 
3114 	hr_reg_write(mpt_entry, MPT_PBL_BA_PG_SZ,
3115 		     to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.ba_pg_shift));
3116 	hr_reg_enable(mpt_entry, MPT_INNER_PA_VLD);
3117 
3118 	ret = set_mtpt_pbl(hr_dev, mpt_entry, mr);
3119 
3120 	return ret;
3121 }
3122 
3123 static int hns_roce_v2_rereg_write_mtpt(struct hns_roce_dev *hr_dev,
3124 					struct hns_roce_mr *mr, int flags,
3125 					void *mb_buf)
3126 {
3127 	struct hns_roce_v2_mpt_entry *mpt_entry = mb_buf;
3128 	u32 mr_access_flags = mr->access;
3129 	int ret = 0;
3130 
3131 	roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_MPT_ST_M,
3132 		       V2_MPT_BYTE_4_MPT_ST_S, V2_MPT_ST_VALID);
3133 
3134 	roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_PD_M,
3135 		       V2_MPT_BYTE_4_PD_S, mr->pd);
3136 
3137 	if (flags & IB_MR_REREG_ACCESS) {
3138 		roce_set_bit(mpt_entry->byte_8_mw_cnt_en,
3139 			     V2_MPT_BYTE_8_BIND_EN_S,
3140 			     (mr_access_flags & IB_ACCESS_MW_BIND ? 1 : 0));
3141 		roce_set_bit(mpt_entry->byte_8_mw_cnt_en,
3142 			     V2_MPT_BYTE_8_ATOMIC_EN_S,
3143 			     mr_access_flags & IB_ACCESS_REMOTE_ATOMIC ? 1 : 0);
3144 		roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_RR_EN_S,
3145 			     mr_access_flags & IB_ACCESS_REMOTE_READ ? 1 : 0);
3146 		roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_RW_EN_S,
3147 			     mr_access_flags & IB_ACCESS_REMOTE_WRITE ? 1 : 0);
3148 		roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_LW_EN_S,
3149 			     mr_access_flags & IB_ACCESS_LOCAL_WRITE ? 1 : 0);
3150 	}
3151 
3152 	if (flags & IB_MR_REREG_TRANS) {
3153 		mpt_entry->va_l = cpu_to_le32(lower_32_bits(mr->iova));
3154 		mpt_entry->va_h = cpu_to_le32(upper_32_bits(mr->iova));
3155 		mpt_entry->len_l = cpu_to_le32(lower_32_bits(mr->size));
3156 		mpt_entry->len_h = cpu_to_le32(upper_32_bits(mr->size));
3157 
3158 		ret = set_mtpt_pbl(hr_dev, mpt_entry, mr);
3159 	}
3160 
3161 	return ret;
3162 }
3163 
3164 static int hns_roce_v2_frmr_write_mtpt(struct hns_roce_dev *hr_dev,
3165 				       void *mb_buf, struct hns_roce_mr *mr)
3166 {
3167 	struct ib_device *ibdev = &hr_dev->ib_dev;
3168 	struct hns_roce_v2_mpt_entry *mpt_entry;
3169 	dma_addr_t pbl_ba = 0;
3170 
3171 	mpt_entry = mb_buf;
3172 	memset(mpt_entry, 0, sizeof(*mpt_entry));
3173 
3174 	if (hns_roce_mtr_find(hr_dev, &mr->pbl_mtr, 0, NULL, 0, &pbl_ba) < 0) {
3175 		ibdev_err(ibdev, "failed to find frmr mtr.\n");
3176 		return -ENOBUFS;
3177 	}
3178 
3179 	roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_MPT_ST_M,
3180 		       V2_MPT_BYTE_4_MPT_ST_S, V2_MPT_ST_FREE);
3181 	roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_PBL_HOP_NUM_M,
3182 		       V2_MPT_BYTE_4_PBL_HOP_NUM_S, 1);
3183 	roce_set_field(mpt_entry->byte_4_pd_hop_st,
3184 		       V2_MPT_BYTE_4_PBL_BA_PG_SZ_M,
3185 		       V2_MPT_BYTE_4_PBL_BA_PG_SZ_S,
3186 		       to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.ba_pg_shift));
3187 	roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_PD_M,
3188 		       V2_MPT_BYTE_4_PD_S, mr->pd);
3189 
3190 	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_RA_EN_S, 1);
3191 	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_R_INV_EN_S, 1);
3192 	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_L_INV_EN_S, 1);
3193 
3194 	roce_set_bit(mpt_entry->byte_12_mw_pa, V2_MPT_BYTE_12_FRE_S, 1);
3195 	roce_set_bit(mpt_entry->byte_12_mw_pa, V2_MPT_BYTE_12_PA_S, 0);
3196 	roce_set_bit(mpt_entry->byte_12_mw_pa, V2_MPT_BYTE_12_MR_MW_S, 0);
3197 	roce_set_bit(mpt_entry->byte_12_mw_pa, V2_MPT_BYTE_12_BPD_S, 1);
3198 
3199 	mpt_entry->pbl_size = cpu_to_le32(mr->npages);
3200 
3201 	mpt_entry->pbl_ba_l = cpu_to_le32(lower_32_bits(pbl_ba >> 3));
3202 	roce_set_field(mpt_entry->byte_48_mode_ba, V2_MPT_BYTE_48_PBL_BA_H_M,
3203 		       V2_MPT_BYTE_48_PBL_BA_H_S,
3204 		       upper_32_bits(pbl_ba >> 3));
3205 
3206 	roce_set_field(mpt_entry->byte_64_buf_pa1,
3207 		       V2_MPT_BYTE_64_PBL_BUF_PG_SZ_M,
3208 		       V2_MPT_BYTE_64_PBL_BUF_PG_SZ_S,
3209 		       to_hr_hw_page_shift(mr->pbl_mtr.hem_cfg.buf_pg_shift));
3210 
3211 	return 0;
3212 }
3213 
3214 static int hns_roce_v2_mw_write_mtpt(void *mb_buf, struct hns_roce_mw *mw)
3215 {
3216 	struct hns_roce_v2_mpt_entry *mpt_entry;
3217 
3218 	mpt_entry = mb_buf;
3219 	memset(mpt_entry, 0, sizeof(*mpt_entry));
3220 
3221 	roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_MPT_ST_M,
3222 		       V2_MPT_BYTE_4_MPT_ST_S, V2_MPT_ST_FREE);
3223 	roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_PD_M,
3224 		       V2_MPT_BYTE_4_PD_S, mw->pdn);
3225 	roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_PBL_HOP_NUM_M,
3226 		       V2_MPT_BYTE_4_PBL_HOP_NUM_S,
3227 		       mw->pbl_hop_num == HNS_ROCE_HOP_NUM_0 ? 0 :
3228 							       mw->pbl_hop_num);
3229 	roce_set_field(mpt_entry->byte_4_pd_hop_st,
3230 		       V2_MPT_BYTE_4_PBL_BA_PG_SZ_M,
3231 		       V2_MPT_BYTE_4_PBL_BA_PG_SZ_S,
3232 		       mw->pbl_ba_pg_sz + PG_SHIFT_OFFSET);
3233 
3234 	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_R_INV_EN_S, 1);
3235 	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_L_INV_EN_S, 1);
3236 	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_LW_EN_S, 1);
3237 
3238 	roce_set_bit(mpt_entry->byte_12_mw_pa, V2_MPT_BYTE_12_PA_S, 0);
3239 	roce_set_bit(mpt_entry->byte_12_mw_pa, V2_MPT_BYTE_12_MR_MW_S, 1);
3240 	roce_set_bit(mpt_entry->byte_12_mw_pa, V2_MPT_BYTE_12_BPD_S, 1);
3241 	roce_set_bit(mpt_entry->byte_12_mw_pa, V2_MPT_BYTE_12_BQP_S,
3242 		     mw->ibmw.type == IB_MW_TYPE_1 ? 0 : 1);
3243 
3244 	roce_set_field(mpt_entry->byte_64_buf_pa1,
3245 		       V2_MPT_BYTE_64_PBL_BUF_PG_SZ_M,
3246 		       V2_MPT_BYTE_64_PBL_BUF_PG_SZ_S,
3247 		       mw->pbl_buf_pg_sz + PG_SHIFT_OFFSET);
3248 
3249 	mpt_entry->lkey = cpu_to_le32(mw->rkey);
3250 
3251 	return 0;
3252 }
3253 
3254 static void *get_cqe_v2(struct hns_roce_cq *hr_cq, int n)
3255 {
3256 	return hns_roce_buf_offset(hr_cq->mtr.kmem, n * hr_cq->cqe_size);
3257 }
3258 
3259 static void *get_sw_cqe_v2(struct hns_roce_cq *hr_cq, unsigned int n)
3260 {
3261 	struct hns_roce_v2_cqe *cqe = get_cqe_v2(hr_cq, n & hr_cq->ib_cq.cqe);
3262 
3263 	/* Get cqe when Owner bit is Conversely with the MSB of cons_idx */
3264 	return (roce_get_bit(cqe->byte_4, V2_CQE_BYTE_4_OWNER_S) ^
3265 		!!(n & hr_cq->cq_depth)) ? cqe : NULL;
3266 }
3267 
3268 static inline void update_cq_db(struct hns_roce_dev *hr_dev,
3269 				struct hns_roce_cq *hr_cq)
3270 {
3271 	if (likely(hr_cq->flags & HNS_ROCE_CQ_FLAG_RECORD_DB)) {
3272 		*hr_cq->set_ci_db = hr_cq->cons_index & V2_CQ_DB_CONS_IDX_M;
3273 	} else {
3274 		struct hns_roce_v2_db cq_db = {};
3275 
3276 		roce_set_field(cq_db.byte_4, V2_DB_TAG_M, V2_DB_TAG_S,
3277 			       hr_cq->cqn);
3278 		roce_set_field(cq_db.byte_4, V2_DB_CMD_M, V2_DB_CMD_S,
3279 			       HNS_ROCE_V2_CQ_DB);
3280 		roce_set_field(cq_db.parameter, V2_CQ_DB_CONS_IDX_M,
3281 			       V2_CQ_DB_CONS_IDX_S, hr_cq->cons_index);
3282 		roce_set_field(cq_db.parameter, V2_CQ_DB_CMD_SN_M,
3283 			       V2_CQ_DB_CMD_SN_S, 1);
3284 
3285 		hns_roce_write64(hr_dev, (__le32 *)&cq_db, hr_cq->db_reg);
3286 	}
3287 }
3288 
3289 static void __hns_roce_v2_cq_clean(struct hns_roce_cq *hr_cq, u32 qpn,
3290 				   struct hns_roce_srq *srq)
3291 {
3292 	struct hns_roce_dev *hr_dev = to_hr_dev(hr_cq->ib_cq.device);
3293 	struct hns_roce_v2_cqe *cqe, *dest;
3294 	u32 prod_index;
3295 	int nfreed = 0;
3296 	int wqe_index;
3297 	u8 owner_bit;
3298 
3299 	for (prod_index = hr_cq->cons_index; get_sw_cqe_v2(hr_cq, prod_index);
3300 	     ++prod_index) {
3301 		if (prod_index > hr_cq->cons_index + hr_cq->ib_cq.cqe)
3302 			break;
3303 	}
3304 
3305 	/*
3306 	 * Now backwards through the CQ, removing CQ entries
3307 	 * that match our QP by overwriting them with next entries.
3308 	 */
3309 	while ((int) --prod_index - (int) hr_cq->cons_index >= 0) {
3310 		cqe = get_cqe_v2(hr_cq, prod_index & hr_cq->ib_cq.cqe);
3311 		if ((roce_get_field(cqe->byte_16, V2_CQE_BYTE_16_LCL_QPN_M,
3312 				    V2_CQE_BYTE_16_LCL_QPN_S) &
3313 				    HNS_ROCE_V2_CQE_QPN_MASK) == qpn) {
3314 			if (srq &&
3315 			    roce_get_bit(cqe->byte_4, V2_CQE_BYTE_4_S_R_S)) {
3316 				wqe_index = roce_get_field(cqe->byte_4,
3317 						     V2_CQE_BYTE_4_WQE_INDX_M,
3318 						     V2_CQE_BYTE_4_WQE_INDX_S);
3319 				hns_roce_free_srq_wqe(srq, wqe_index);
3320 			}
3321 			++nfreed;
3322 		} else if (nfreed) {
3323 			dest = get_cqe_v2(hr_cq, (prod_index + nfreed) &
3324 					  hr_cq->ib_cq.cqe);
3325 			owner_bit = roce_get_bit(dest->byte_4,
3326 						 V2_CQE_BYTE_4_OWNER_S);
3327 			memcpy(dest, cqe, sizeof(*cqe));
3328 			roce_set_bit(dest->byte_4, V2_CQE_BYTE_4_OWNER_S,
3329 				     owner_bit);
3330 		}
3331 	}
3332 
3333 	if (nfreed) {
3334 		hr_cq->cons_index += nfreed;
3335 		update_cq_db(hr_dev, hr_cq);
3336 	}
3337 }
3338 
3339 static void hns_roce_v2_cq_clean(struct hns_roce_cq *hr_cq, u32 qpn,
3340 				 struct hns_roce_srq *srq)
3341 {
3342 	spin_lock_irq(&hr_cq->lock);
3343 	__hns_roce_v2_cq_clean(hr_cq, qpn, srq);
3344 	spin_unlock_irq(&hr_cq->lock);
3345 }
3346 
3347 static void hns_roce_v2_write_cqc(struct hns_roce_dev *hr_dev,
3348 				  struct hns_roce_cq *hr_cq, void *mb_buf,
3349 				  u64 *mtts, dma_addr_t dma_handle)
3350 {
3351 	struct hns_roce_v2_cq_context *cq_context;
3352 
3353 	cq_context = mb_buf;
3354 	memset(cq_context, 0, sizeof(*cq_context));
3355 
3356 	roce_set_field(cq_context->byte_4_pg_ceqn, V2_CQC_BYTE_4_CQ_ST_M,
3357 		       V2_CQC_BYTE_4_CQ_ST_S, V2_CQ_STATE_VALID);
3358 	roce_set_field(cq_context->byte_4_pg_ceqn, V2_CQC_BYTE_4_ARM_ST_M,
3359 		       V2_CQC_BYTE_4_ARM_ST_S, REG_NXT_CEQE);
3360 	roce_set_field(cq_context->byte_4_pg_ceqn, V2_CQC_BYTE_4_SHIFT_M,
3361 		       V2_CQC_BYTE_4_SHIFT_S, ilog2(hr_cq->cq_depth));
3362 	roce_set_field(cq_context->byte_4_pg_ceqn, V2_CQC_BYTE_4_CEQN_M,
3363 		       V2_CQC_BYTE_4_CEQN_S, hr_cq->vector);
3364 
3365 	roce_set_field(cq_context->byte_8_cqn, V2_CQC_BYTE_8_CQN_M,
3366 		       V2_CQC_BYTE_8_CQN_S, hr_cq->cqn);
3367 
3368 	roce_set_field(cq_context->byte_8_cqn, V2_CQC_BYTE_8_CQE_SIZE_M,
3369 		       V2_CQC_BYTE_8_CQE_SIZE_S, hr_cq->cqe_size ==
3370 		       HNS_ROCE_V3_CQE_SIZE ? 1 : 0);
3371 
3372 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_STASH)
3373 		hr_reg_enable(cq_context, CQC_STASH);
3374 
3375 	cq_context->cqe_cur_blk_addr = cpu_to_le32(to_hr_hw_page_addr(mtts[0]));
3376 
3377 	roce_set_field(cq_context->byte_16_hop_addr,
3378 		       V2_CQC_BYTE_16_CQE_CUR_BLK_ADDR_M,
3379 		       V2_CQC_BYTE_16_CQE_CUR_BLK_ADDR_S,
3380 		       upper_32_bits(to_hr_hw_page_addr(mtts[0])));
3381 	roce_set_field(cq_context->byte_16_hop_addr,
3382 		       V2_CQC_BYTE_16_CQE_HOP_NUM_M,
3383 		       V2_CQC_BYTE_16_CQE_HOP_NUM_S, hr_dev->caps.cqe_hop_num ==
3384 		       HNS_ROCE_HOP_NUM_0 ? 0 : hr_dev->caps.cqe_hop_num);
3385 
3386 	cq_context->cqe_nxt_blk_addr = cpu_to_le32(to_hr_hw_page_addr(mtts[1]));
3387 	roce_set_field(cq_context->byte_24_pgsz_addr,
3388 		       V2_CQC_BYTE_24_CQE_NXT_BLK_ADDR_M,
3389 		       V2_CQC_BYTE_24_CQE_NXT_BLK_ADDR_S,
3390 		       upper_32_bits(to_hr_hw_page_addr(mtts[1])));
3391 	roce_set_field(cq_context->byte_24_pgsz_addr,
3392 		       V2_CQC_BYTE_24_CQE_BA_PG_SZ_M,
3393 		       V2_CQC_BYTE_24_CQE_BA_PG_SZ_S,
3394 		       to_hr_hw_page_shift(hr_cq->mtr.hem_cfg.ba_pg_shift));
3395 	roce_set_field(cq_context->byte_24_pgsz_addr,
3396 		       V2_CQC_BYTE_24_CQE_BUF_PG_SZ_M,
3397 		       V2_CQC_BYTE_24_CQE_BUF_PG_SZ_S,
3398 		       to_hr_hw_page_shift(hr_cq->mtr.hem_cfg.buf_pg_shift));
3399 
3400 	cq_context->cqe_ba = cpu_to_le32(dma_handle >> 3);
3401 
3402 	roce_set_field(cq_context->byte_40_cqe_ba, V2_CQC_BYTE_40_CQE_BA_M,
3403 		       V2_CQC_BYTE_40_CQE_BA_S, (dma_handle >> (32 + 3)));
3404 
3405 	roce_set_bit(cq_context->byte_44_db_record,
3406 		     V2_CQC_BYTE_44_DB_RECORD_EN_S,
3407 		     (hr_cq->flags & HNS_ROCE_CQ_FLAG_RECORD_DB) ? 1 : 0);
3408 
3409 	roce_set_field(cq_context->byte_44_db_record,
3410 		       V2_CQC_BYTE_44_DB_RECORD_ADDR_M,
3411 		       V2_CQC_BYTE_44_DB_RECORD_ADDR_S,
3412 		       ((u32)hr_cq->db.dma) >> 1);
3413 	cq_context->db_record_addr = cpu_to_le32(hr_cq->db.dma >> 32);
3414 
3415 	roce_set_field(cq_context->byte_56_cqe_period_maxcnt,
3416 		       V2_CQC_BYTE_56_CQ_MAX_CNT_M,
3417 		       V2_CQC_BYTE_56_CQ_MAX_CNT_S,
3418 		       HNS_ROCE_V2_CQ_DEFAULT_BURST_NUM);
3419 	roce_set_field(cq_context->byte_56_cqe_period_maxcnt,
3420 		       V2_CQC_BYTE_56_CQ_PERIOD_M,
3421 		       V2_CQC_BYTE_56_CQ_PERIOD_S,
3422 		       HNS_ROCE_V2_CQ_DEFAULT_INTERVAL);
3423 }
3424 
3425 static int hns_roce_v2_req_notify_cq(struct ib_cq *ibcq,
3426 				     enum ib_cq_notify_flags flags)
3427 {
3428 	struct hns_roce_dev *hr_dev = to_hr_dev(ibcq->device);
3429 	struct hns_roce_cq *hr_cq = to_hr_cq(ibcq);
3430 	struct hns_roce_v2_db cq_db = {};
3431 	u32 notify_flag;
3432 
3433 	/*
3434 	 * flags = 0, then notify_flag : next
3435 	 * flags = 1, then notify flag : solocited
3436 	 */
3437 	notify_flag = (flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED ?
3438 		      V2_CQ_DB_REQ_NOT : V2_CQ_DB_REQ_NOT_SOL;
3439 
3440 	roce_set_field(cq_db.byte_4, V2_DB_TAG_M, V2_DB_TAG_S, hr_cq->cqn);
3441 	roce_set_field(cq_db.byte_4, V2_DB_CMD_M, V2_DB_CMD_S,
3442 		       HNS_ROCE_V2_CQ_DB_NOTIFY);
3443 	roce_set_field(cq_db.parameter, V2_CQ_DB_CONS_IDX_M,
3444 		       V2_CQ_DB_CONS_IDX_S, hr_cq->cons_index);
3445 	roce_set_field(cq_db.parameter, V2_CQ_DB_CMD_SN_M,
3446 		       V2_CQ_DB_CMD_SN_S, hr_cq->arm_sn);
3447 	roce_set_bit(cq_db.parameter, V2_CQ_DB_NOTIFY_TYPE_S, notify_flag);
3448 
3449 	hns_roce_write64(hr_dev, (__le32 *)&cq_db, hr_cq->db_reg);
3450 
3451 	return 0;
3452 }
3453 
3454 static int hns_roce_handle_recv_inl_wqe(struct hns_roce_v2_cqe *cqe,
3455 					struct hns_roce_qp *qp,
3456 					struct ib_wc *wc)
3457 {
3458 	struct hns_roce_rinl_sge *sge_list;
3459 	u32 wr_num, wr_cnt, sge_num;
3460 	u32 sge_cnt, data_len, size;
3461 	void *wqe_buf;
3462 
3463 	wr_num = roce_get_field(cqe->byte_4, V2_CQE_BYTE_4_WQE_INDX_M,
3464 				V2_CQE_BYTE_4_WQE_INDX_S) & 0xffff;
3465 	wr_cnt = wr_num & (qp->rq.wqe_cnt - 1);
3466 
3467 	sge_list = qp->rq_inl_buf.wqe_list[wr_cnt].sg_list;
3468 	sge_num = qp->rq_inl_buf.wqe_list[wr_cnt].sge_cnt;
3469 	wqe_buf = hns_roce_get_recv_wqe(qp, wr_cnt);
3470 	data_len = wc->byte_len;
3471 
3472 	for (sge_cnt = 0; (sge_cnt < sge_num) && (data_len); sge_cnt++) {
3473 		size = min(sge_list[sge_cnt].len, data_len);
3474 		memcpy((void *)sge_list[sge_cnt].addr, wqe_buf, size);
3475 
3476 		data_len -= size;
3477 		wqe_buf += size;
3478 	}
3479 
3480 	if (unlikely(data_len)) {
3481 		wc->status = IB_WC_LOC_LEN_ERR;
3482 		return -EAGAIN;
3483 	}
3484 
3485 	return 0;
3486 }
3487 
3488 static int sw_comp(struct hns_roce_qp *hr_qp, struct hns_roce_wq *wq,
3489 		   int num_entries, struct ib_wc *wc)
3490 {
3491 	unsigned int left;
3492 	int npolled = 0;
3493 
3494 	left = wq->head - wq->tail;
3495 	if (left == 0)
3496 		return 0;
3497 
3498 	left = min_t(unsigned int, (unsigned int)num_entries, left);
3499 	while (npolled < left) {
3500 		wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
3501 		wc->status = IB_WC_WR_FLUSH_ERR;
3502 		wc->vendor_err = 0;
3503 		wc->qp = &hr_qp->ibqp;
3504 
3505 		wq->tail++;
3506 		wc++;
3507 		npolled++;
3508 	}
3509 
3510 	return npolled;
3511 }
3512 
3513 static int hns_roce_v2_sw_poll_cq(struct hns_roce_cq *hr_cq, int num_entries,
3514 				  struct ib_wc *wc)
3515 {
3516 	struct hns_roce_qp *hr_qp;
3517 	int npolled = 0;
3518 
3519 	list_for_each_entry(hr_qp, &hr_cq->sq_list, sq_node) {
3520 		npolled += sw_comp(hr_qp, &hr_qp->sq,
3521 				   num_entries - npolled, wc + npolled);
3522 		if (npolled >= num_entries)
3523 			goto out;
3524 	}
3525 
3526 	list_for_each_entry(hr_qp, &hr_cq->rq_list, rq_node) {
3527 		npolled += sw_comp(hr_qp, &hr_qp->rq,
3528 				   num_entries - npolled, wc + npolled);
3529 		if (npolled >= num_entries)
3530 			goto out;
3531 	}
3532 
3533 out:
3534 	return npolled;
3535 }
3536 
3537 static void get_cqe_status(struct hns_roce_dev *hr_dev, struct hns_roce_qp *qp,
3538 			   struct hns_roce_cq *cq, struct hns_roce_v2_cqe *cqe,
3539 			   struct ib_wc *wc)
3540 {
3541 	static const struct {
3542 		u32 cqe_status;
3543 		enum ib_wc_status wc_status;
3544 	} map[] = {
3545 		{ HNS_ROCE_CQE_V2_SUCCESS, IB_WC_SUCCESS },
3546 		{ HNS_ROCE_CQE_V2_LOCAL_LENGTH_ERR, IB_WC_LOC_LEN_ERR },
3547 		{ HNS_ROCE_CQE_V2_LOCAL_QP_OP_ERR, IB_WC_LOC_QP_OP_ERR },
3548 		{ HNS_ROCE_CQE_V2_LOCAL_PROT_ERR, IB_WC_LOC_PROT_ERR },
3549 		{ HNS_ROCE_CQE_V2_WR_FLUSH_ERR, IB_WC_WR_FLUSH_ERR },
3550 		{ HNS_ROCE_CQE_V2_MW_BIND_ERR, IB_WC_MW_BIND_ERR },
3551 		{ HNS_ROCE_CQE_V2_BAD_RESP_ERR, IB_WC_BAD_RESP_ERR },
3552 		{ HNS_ROCE_CQE_V2_LOCAL_ACCESS_ERR, IB_WC_LOC_ACCESS_ERR },
3553 		{ HNS_ROCE_CQE_V2_REMOTE_INVAL_REQ_ERR, IB_WC_REM_INV_REQ_ERR },
3554 		{ HNS_ROCE_CQE_V2_REMOTE_ACCESS_ERR, IB_WC_REM_ACCESS_ERR },
3555 		{ HNS_ROCE_CQE_V2_REMOTE_OP_ERR, IB_WC_REM_OP_ERR },
3556 		{ HNS_ROCE_CQE_V2_TRANSPORT_RETRY_EXC_ERR,
3557 		  IB_WC_RETRY_EXC_ERR },
3558 		{ HNS_ROCE_CQE_V2_RNR_RETRY_EXC_ERR, IB_WC_RNR_RETRY_EXC_ERR },
3559 		{ HNS_ROCE_CQE_V2_REMOTE_ABORT_ERR, IB_WC_REM_ABORT_ERR },
3560 		{ HNS_ROCE_CQE_V2_GENERAL_ERR, IB_WC_GENERAL_ERR}
3561 	};
3562 
3563 	u32 cqe_status = roce_get_field(cqe->byte_4, V2_CQE_BYTE_4_STATUS_M,
3564 					V2_CQE_BYTE_4_STATUS_S);
3565 	int i;
3566 
3567 	wc->status = IB_WC_GENERAL_ERR;
3568 	for (i = 0; i < ARRAY_SIZE(map); i++)
3569 		if (cqe_status == map[i].cqe_status) {
3570 			wc->status = map[i].wc_status;
3571 			break;
3572 		}
3573 
3574 	if (likely(wc->status == IB_WC_SUCCESS ||
3575 		   wc->status == IB_WC_WR_FLUSH_ERR))
3576 		return;
3577 
3578 	ibdev_err(&hr_dev->ib_dev, "error cqe status 0x%x:\n", cqe_status);
3579 	print_hex_dump(KERN_ERR, "", DUMP_PREFIX_NONE, 16, 4, cqe,
3580 		       cq->cqe_size, false);
3581 
3582 	/*
3583 	 * For hns ROCEE, GENERAL_ERR is an error type that is not defined in
3584 	 * the standard protocol, the driver must ignore it and needn't to set
3585 	 * the QP to an error state.
3586 	 */
3587 	if (cqe_status == HNS_ROCE_CQE_V2_GENERAL_ERR)
3588 		return;
3589 
3590 	/*
3591 	 * Hip08 hardware cannot flush the WQEs in SQ/RQ if the QP state gets
3592 	 * into errored mode. Hence, as a workaround to this hardware
3593 	 * limitation, driver needs to assist in flushing. But the flushing
3594 	 * operation uses mailbox to convey the QP state to the hardware and
3595 	 * which can sleep due to the mutex protection around the mailbox calls.
3596 	 * Hence, use the deferred flush for now. Once wc error detected, the
3597 	 * flushing operation is needed.
3598 	 */
3599 	if (!test_and_set_bit(HNS_ROCE_FLUSH_FLAG, &qp->flush_flag))
3600 		init_flush_work(hr_dev, qp);
3601 }
3602 
3603 static int get_cur_qp(struct hns_roce_cq *hr_cq, struct hns_roce_v2_cqe *cqe,
3604 		      struct hns_roce_qp **cur_qp)
3605 {
3606 	struct hns_roce_dev *hr_dev = to_hr_dev(hr_cq->ib_cq.device);
3607 	struct hns_roce_qp *hr_qp = *cur_qp;
3608 	u32 qpn;
3609 
3610 	qpn = roce_get_field(cqe->byte_16, V2_CQE_BYTE_16_LCL_QPN_M,
3611 			     V2_CQE_BYTE_16_LCL_QPN_S) &
3612 	      HNS_ROCE_V2_CQE_QPN_MASK;
3613 
3614 	if (!hr_qp || qpn != hr_qp->qpn) {
3615 		hr_qp = __hns_roce_qp_lookup(hr_dev, qpn);
3616 		if (unlikely(!hr_qp)) {
3617 			ibdev_err(&hr_dev->ib_dev,
3618 				  "CQ %06lx with entry for unknown QPN %06x\n",
3619 				  hr_cq->cqn, qpn);
3620 			return -EINVAL;
3621 		}
3622 		*cur_qp = hr_qp;
3623 	}
3624 
3625 	return 0;
3626 }
3627 
3628 /*
3629  * mapped-value = 1 + real-value
3630  * The ib wc opcode's real value is start from 0, In order to distinguish
3631  * between initialized and uninitialized map values, we plus 1 to the actual
3632  * value when defining the mapping, so that the validity can be identified by
3633  * checking whether the mapped value is greater than 0.
3634  */
3635 #define HR_WC_OP_MAP(hr_key, ib_key) \
3636 		[HNS_ROCE_V2_WQE_OP_ ## hr_key] = 1 + IB_WC_ ## ib_key
3637 
3638 static const u32 wc_send_op_map[] = {
3639 	HR_WC_OP_MAP(SEND,			SEND),
3640 	HR_WC_OP_MAP(SEND_WITH_INV,		SEND),
3641 	HR_WC_OP_MAP(SEND_WITH_IMM,		SEND),
3642 	HR_WC_OP_MAP(RDMA_READ,			RDMA_READ),
3643 	HR_WC_OP_MAP(RDMA_WRITE,		RDMA_WRITE),
3644 	HR_WC_OP_MAP(RDMA_WRITE_WITH_IMM,	RDMA_WRITE),
3645 	HR_WC_OP_MAP(LOCAL_INV,			LOCAL_INV),
3646 	HR_WC_OP_MAP(ATOM_CMP_AND_SWAP,		COMP_SWAP),
3647 	HR_WC_OP_MAP(ATOM_FETCH_AND_ADD,	FETCH_ADD),
3648 	HR_WC_OP_MAP(ATOM_MSK_CMP_AND_SWAP,	MASKED_COMP_SWAP),
3649 	HR_WC_OP_MAP(ATOM_MSK_FETCH_AND_ADD,	MASKED_FETCH_ADD),
3650 	HR_WC_OP_MAP(FAST_REG_PMR,		REG_MR),
3651 	HR_WC_OP_MAP(BIND_MW,			REG_MR),
3652 };
3653 
3654 static int to_ib_wc_send_op(u32 hr_opcode)
3655 {
3656 	if (hr_opcode >= ARRAY_SIZE(wc_send_op_map))
3657 		return -EINVAL;
3658 
3659 	return wc_send_op_map[hr_opcode] ? wc_send_op_map[hr_opcode] - 1 :
3660 					   -EINVAL;
3661 }
3662 
3663 static const u32 wc_recv_op_map[] = {
3664 	HR_WC_OP_MAP(RDMA_WRITE_WITH_IMM,		WITH_IMM),
3665 	HR_WC_OP_MAP(SEND,				RECV),
3666 	HR_WC_OP_MAP(SEND_WITH_IMM,			WITH_IMM),
3667 	HR_WC_OP_MAP(SEND_WITH_INV,			RECV),
3668 };
3669 
3670 static int to_ib_wc_recv_op(u32 hr_opcode)
3671 {
3672 	if (hr_opcode >= ARRAY_SIZE(wc_recv_op_map))
3673 		return -EINVAL;
3674 
3675 	return wc_recv_op_map[hr_opcode] ? wc_recv_op_map[hr_opcode] - 1 :
3676 					   -EINVAL;
3677 }
3678 
3679 static void fill_send_wc(struct ib_wc *wc, struct hns_roce_v2_cqe *cqe)
3680 {
3681 	u32 hr_opcode;
3682 	int ib_opcode;
3683 
3684 	wc->wc_flags = 0;
3685 
3686 	hr_opcode = roce_get_field(cqe->byte_4, V2_CQE_BYTE_4_OPCODE_M,
3687 				   V2_CQE_BYTE_4_OPCODE_S) & 0x1f;
3688 	switch (hr_opcode) {
3689 	case HNS_ROCE_V2_WQE_OP_RDMA_READ:
3690 		wc->byte_len = le32_to_cpu(cqe->byte_cnt);
3691 		break;
3692 	case HNS_ROCE_V2_WQE_OP_SEND_WITH_IMM:
3693 	case HNS_ROCE_V2_WQE_OP_RDMA_WRITE_WITH_IMM:
3694 		wc->wc_flags |= IB_WC_WITH_IMM;
3695 		break;
3696 	case HNS_ROCE_V2_WQE_OP_LOCAL_INV:
3697 		wc->wc_flags |= IB_WC_WITH_INVALIDATE;
3698 		break;
3699 	case HNS_ROCE_V2_WQE_OP_ATOM_CMP_AND_SWAP:
3700 	case HNS_ROCE_V2_WQE_OP_ATOM_FETCH_AND_ADD:
3701 	case HNS_ROCE_V2_WQE_OP_ATOM_MSK_CMP_AND_SWAP:
3702 	case HNS_ROCE_V2_WQE_OP_ATOM_MSK_FETCH_AND_ADD:
3703 		wc->byte_len  = 8;
3704 		break;
3705 	default:
3706 		break;
3707 	}
3708 
3709 	ib_opcode = to_ib_wc_send_op(hr_opcode);
3710 	if (ib_opcode < 0)
3711 		wc->status = IB_WC_GENERAL_ERR;
3712 	else
3713 		wc->opcode = ib_opcode;
3714 }
3715 
3716 static inline bool is_rq_inl_enabled(struct ib_wc *wc, u32 hr_opcode,
3717 				     struct hns_roce_v2_cqe *cqe)
3718 {
3719 	return wc->qp->qp_type != IB_QPT_UD &&
3720 	       wc->qp->qp_type != IB_QPT_GSI &&
3721 	       (hr_opcode == HNS_ROCE_V2_OPCODE_SEND ||
3722 		hr_opcode == HNS_ROCE_V2_OPCODE_SEND_WITH_IMM ||
3723 		hr_opcode == HNS_ROCE_V2_OPCODE_SEND_WITH_INV) &&
3724 	       roce_get_bit(cqe->byte_4, V2_CQE_BYTE_4_RQ_INLINE_S);
3725 }
3726 
3727 static int fill_recv_wc(struct ib_wc *wc, struct hns_roce_v2_cqe *cqe)
3728 {
3729 	struct hns_roce_qp *qp = to_hr_qp(wc->qp);
3730 	u32 hr_opcode;
3731 	int ib_opcode;
3732 	int ret;
3733 
3734 	wc->byte_len = le32_to_cpu(cqe->byte_cnt);
3735 
3736 	hr_opcode = roce_get_field(cqe->byte_4, V2_CQE_BYTE_4_OPCODE_M,
3737 				   V2_CQE_BYTE_4_OPCODE_S) & 0x1f;
3738 	switch (hr_opcode) {
3739 	case HNS_ROCE_V2_OPCODE_RDMA_WRITE_IMM:
3740 	case HNS_ROCE_V2_OPCODE_SEND_WITH_IMM:
3741 		wc->wc_flags = IB_WC_WITH_IMM;
3742 		wc->ex.imm_data = cpu_to_be32(le32_to_cpu(cqe->immtdata));
3743 		break;
3744 	case HNS_ROCE_V2_OPCODE_SEND_WITH_INV:
3745 		wc->wc_flags = IB_WC_WITH_INVALIDATE;
3746 		wc->ex.invalidate_rkey = le32_to_cpu(cqe->rkey);
3747 		break;
3748 	default:
3749 		wc->wc_flags = 0;
3750 	}
3751 
3752 	ib_opcode = to_ib_wc_recv_op(hr_opcode);
3753 	if (ib_opcode < 0)
3754 		wc->status = IB_WC_GENERAL_ERR;
3755 	else
3756 		wc->opcode = ib_opcode;
3757 
3758 	if (is_rq_inl_enabled(wc, hr_opcode, cqe)) {
3759 		ret = hns_roce_handle_recv_inl_wqe(cqe, qp, wc);
3760 		if (unlikely(ret))
3761 			return ret;
3762 	}
3763 
3764 	wc->sl = roce_get_field(cqe->byte_32, V2_CQE_BYTE_32_SL_M,
3765 				V2_CQE_BYTE_32_SL_S);
3766 	wc->src_qp = roce_get_field(cqe->byte_32, V2_CQE_BYTE_32_RMT_QPN_M,
3767 				    V2_CQE_BYTE_32_RMT_QPN_S);
3768 	wc->slid = 0;
3769 	wc->wc_flags |= roce_get_bit(cqe->byte_32, V2_CQE_BYTE_32_GRH_S) ?
3770 				     IB_WC_GRH : 0;
3771 	wc->port_num = roce_get_field(cqe->byte_32, V2_CQE_BYTE_32_PORTN_M,
3772 				      V2_CQE_BYTE_32_PORTN_S);
3773 	wc->pkey_index = 0;
3774 
3775 	if (roce_get_bit(cqe->byte_28, V2_CQE_BYTE_28_VID_VLD_S)) {
3776 		wc->vlan_id = roce_get_field(cqe->byte_28, V2_CQE_BYTE_28_VID_M,
3777 					     V2_CQE_BYTE_28_VID_S);
3778 		wc->wc_flags |= IB_WC_WITH_VLAN;
3779 	} else {
3780 		wc->vlan_id = 0xffff;
3781 	}
3782 
3783 	wc->network_hdr_type = roce_get_field(cqe->byte_28,
3784 					      V2_CQE_BYTE_28_PORT_TYPE_M,
3785 					      V2_CQE_BYTE_28_PORT_TYPE_S);
3786 
3787 	return 0;
3788 }
3789 
3790 static int hns_roce_v2_poll_one(struct hns_roce_cq *hr_cq,
3791 				struct hns_roce_qp **cur_qp, struct ib_wc *wc)
3792 {
3793 	struct hns_roce_dev *hr_dev = to_hr_dev(hr_cq->ib_cq.device);
3794 	struct hns_roce_qp *qp = *cur_qp;
3795 	struct hns_roce_srq *srq = NULL;
3796 	struct hns_roce_v2_cqe *cqe;
3797 	struct hns_roce_wq *wq;
3798 	int is_send;
3799 	u16 wqe_idx;
3800 	int ret;
3801 
3802 	cqe = get_sw_cqe_v2(hr_cq, hr_cq->cons_index);
3803 	if (!cqe)
3804 		return -EAGAIN;
3805 
3806 	++hr_cq->cons_index;
3807 	/* Memory barrier */
3808 	rmb();
3809 
3810 	ret = get_cur_qp(hr_cq, cqe, &qp);
3811 	if (ret)
3812 		return ret;
3813 
3814 	wc->qp = &qp->ibqp;
3815 	wc->vendor_err = 0;
3816 
3817 	wqe_idx = roce_get_field(cqe->byte_4, V2_CQE_BYTE_4_WQE_INDX_M,
3818 				 V2_CQE_BYTE_4_WQE_INDX_S);
3819 
3820 	is_send = !roce_get_bit(cqe->byte_4, V2_CQE_BYTE_4_S_R_S);
3821 	if (is_send) {
3822 		wq = &qp->sq;
3823 
3824 		/* If sg_signal_bit is set, tail pointer will be updated to
3825 		 * the WQE corresponding to the current CQE.
3826 		 */
3827 		if (qp->sq_signal_bits)
3828 			wq->tail += (wqe_idx - (u16)wq->tail) &
3829 				    (wq->wqe_cnt - 1);
3830 
3831 		wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
3832 		++wq->tail;
3833 
3834 		fill_send_wc(wc, cqe);
3835 	} else {
3836 		if (qp->ibqp.srq) {
3837 			srq = to_hr_srq(qp->ibqp.srq);
3838 			wc->wr_id = srq->wrid[wqe_idx];
3839 			hns_roce_free_srq_wqe(srq, wqe_idx);
3840 		} else {
3841 			wq = &qp->rq;
3842 			wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
3843 			++wq->tail;
3844 		}
3845 
3846 		ret = fill_recv_wc(wc, cqe);
3847 	}
3848 
3849 	get_cqe_status(hr_dev, qp, hr_cq, cqe, wc);
3850 	if (unlikely(wc->status != IB_WC_SUCCESS))
3851 		return 0;
3852 
3853 	return ret;
3854 }
3855 
3856 static int hns_roce_v2_poll_cq(struct ib_cq *ibcq, int num_entries,
3857 			       struct ib_wc *wc)
3858 {
3859 	struct hns_roce_dev *hr_dev = to_hr_dev(ibcq->device);
3860 	struct hns_roce_cq *hr_cq = to_hr_cq(ibcq);
3861 	struct hns_roce_qp *cur_qp = NULL;
3862 	unsigned long flags;
3863 	int npolled;
3864 
3865 	spin_lock_irqsave(&hr_cq->lock, flags);
3866 
3867 	/*
3868 	 * When the device starts to reset, the state is RST_DOWN. At this time,
3869 	 * there may still be some valid CQEs in the hardware that are not
3870 	 * polled. Therefore, it is not allowed to switch to the software mode
3871 	 * immediately. When the state changes to UNINIT, CQE no longer exists
3872 	 * in the hardware, and then switch to software mode.
3873 	 */
3874 	if (hr_dev->state == HNS_ROCE_DEVICE_STATE_UNINIT) {
3875 		npolled = hns_roce_v2_sw_poll_cq(hr_cq, num_entries, wc);
3876 		goto out;
3877 	}
3878 
3879 	for (npolled = 0; npolled < num_entries; ++npolled) {
3880 		if (hns_roce_v2_poll_one(hr_cq, &cur_qp, wc + npolled))
3881 			break;
3882 	}
3883 
3884 	if (npolled)
3885 		update_cq_db(hr_dev, hr_cq);
3886 
3887 out:
3888 	spin_unlock_irqrestore(&hr_cq->lock, flags);
3889 
3890 	return npolled;
3891 }
3892 
3893 static int get_op_for_set_hem(struct hns_roce_dev *hr_dev, u32 type,
3894 			      int step_idx, u16 *mbox_op)
3895 {
3896 	u16 op;
3897 
3898 	switch (type) {
3899 	case HEM_TYPE_QPC:
3900 		op = HNS_ROCE_CMD_WRITE_QPC_BT0;
3901 		break;
3902 	case HEM_TYPE_MTPT:
3903 		op = HNS_ROCE_CMD_WRITE_MPT_BT0;
3904 		break;
3905 	case HEM_TYPE_CQC:
3906 		op = HNS_ROCE_CMD_WRITE_CQC_BT0;
3907 		break;
3908 	case HEM_TYPE_SRQC:
3909 		op = HNS_ROCE_CMD_WRITE_SRQC_BT0;
3910 		break;
3911 	case HEM_TYPE_SCCC:
3912 		op = HNS_ROCE_CMD_WRITE_SCCC_BT0;
3913 		break;
3914 	case HEM_TYPE_QPC_TIMER:
3915 		op = HNS_ROCE_CMD_WRITE_QPC_TIMER_BT0;
3916 		break;
3917 	case HEM_TYPE_CQC_TIMER:
3918 		op = HNS_ROCE_CMD_WRITE_CQC_TIMER_BT0;
3919 		break;
3920 	default:
3921 		dev_warn(hr_dev->dev, "failed to check hem type %u.\n", type);
3922 		return -EINVAL;
3923 	}
3924 
3925 	*mbox_op = op + step_idx;
3926 
3927 	return 0;
3928 }
3929 
3930 static int config_gmv_ba_to_hw(struct hns_roce_dev *hr_dev, unsigned long obj,
3931 			       dma_addr_t base_addr)
3932 {
3933 	struct hns_roce_cmq_desc desc;
3934 	struct hns_roce_cmq_req *req = (struct hns_roce_cmq_req *)desc.data;
3935 	u32 idx = obj / (HNS_HW_PAGE_SIZE / hr_dev->caps.gmv_entry_sz);
3936 	u64 addr = to_hr_hw_page_addr(base_addr);
3937 
3938 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CFG_GMV_BT, false);
3939 
3940 	hr_reg_write(req, CFG_GMV_BT_BA_L, lower_32_bits(addr));
3941 	hr_reg_write(req, CFG_GMV_BT_BA_H, upper_32_bits(addr));
3942 	hr_reg_write(req, CFG_GMV_BT_IDX, idx);
3943 
3944 	return hns_roce_cmq_send(hr_dev, &desc, 1);
3945 }
3946 
3947 static int set_hem_to_hw(struct hns_roce_dev *hr_dev, int obj,
3948 			 dma_addr_t base_addr, u32 hem_type, int step_idx)
3949 {
3950 	int ret;
3951 	u16 op;
3952 
3953 	if (unlikely(hem_type == HEM_TYPE_GMV))
3954 		return config_gmv_ba_to_hw(hr_dev, obj, base_addr);
3955 
3956 	if (unlikely(hem_type == HEM_TYPE_SCCC && step_idx))
3957 		return 0;
3958 
3959 	ret = get_op_for_set_hem(hr_dev, hem_type, step_idx, &op);
3960 	if (ret < 0)
3961 		return ret;
3962 
3963 	return config_hem_ba_to_hw(hr_dev, obj, base_addr, op);
3964 }
3965 
3966 static int hns_roce_v2_set_hem(struct hns_roce_dev *hr_dev,
3967 			       struct hns_roce_hem_table *table, int obj,
3968 			       int step_idx)
3969 {
3970 	struct hns_roce_hem_iter iter;
3971 	struct hns_roce_hem_mhop mhop;
3972 	struct hns_roce_hem *hem;
3973 	unsigned long mhop_obj = obj;
3974 	int i, j, k;
3975 	int ret = 0;
3976 	u64 hem_idx = 0;
3977 	u64 l1_idx = 0;
3978 	u64 bt_ba = 0;
3979 	u32 chunk_ba_num;
3980 	u32 hop_num;
3981 
3982 	if (!hns_roce_check_whether_mhop(hr_dev, table->type))
3983 		return 0;
3984 
3985 	hns_roce_calc_hem_mhop(hr_dev, table, &mhop_obj, &mhop);
3986 	i = mhop.l0_idx;
3987 	j = mhop.l1_idx;
3988 	k = mhop.l2_idx;
3989 	hop_num = mhop.hop_num;
3990 	chunk_ba_num = mhop.bt_chunk_size / 8;
3991 
3992 	if (hop_num == 2) {
3993 		hem_idx = i * chunk_ba_num * chunk_ba_num + j * chunk_ba_num +
3994 			  k;
3995 		l1_idx = i * chunk_ba_num + j;
3996 	} else if (hop_num == 1) {
3997 		hem_idx = i * chunk_ba_num + j;
3998 	} else if (hop_num == HNS_ROCE_HOP_NUM_0) {
3999 		hem_idx = i;
4000 	}
4001 
4002 	if (table->type == HEM_TYPE_SCCC)
4003 		obj = mhop.l0_idx;
4004 
4005 	if (check_whether_last_step(hop_num, step_idx)) {
4006 		hem = table->hem[hem_idx];
4007 		for (hns_roce_hem_first(hem, &iter);
4008 		     !hns_roce_hem_last(&iter); hns_roce_hem_next(&iter)) {
4009 			bt_ba = hns_roce_hem_addr(&iter);
4010 			ret = set_hem_to_hw(hr_dev, obj, bt_ba, table->type,
4011 					    step_idx);
4012 		}
4013 	} else {
4014 		if (step_idx == 0)
4015 			bt_ba = table->bt_l0_dma_addr[i];
4016 		else if (step_idx == 1 && hop_num == 2)
4017 			bt_ba = table->bt_l1_dma_addr[l1_idx];
4018 
4019 		ret = set_hem_to_hw(hr_dev, obj, bt_ba, table->type, step_idx);
4020 	}
4021 
4022 	return ret;
4023 }
4024 
4025 static int hns_roce_v2_clear_hem(struct hns_roce_dev *hr_dev,
4026 				 struct hns_roce_hem_table *table, int obj,
4027 				 int step_idx)
4028 {
4029 	struct device *dev = hr_dev->dev;
4030 	struct hns_roce_cmd_mailbox *mailbox;
4031 	int ret;
4032 	u16 op = 0xff;
4033 
4034 	if (!hns_roce_check_whether_mhop(hr_dev, table->type))
4035 		return 0;
4036 
4037 	switch (table->type) {
4038 	case HEM_TYPE_QPC:
4039 		op = HNS_ROCE_CMD_DESTROY_QPC_BT0;
4040 		break;
4041 	case HEM_TYPE_MTPT:
4042 		op = HNS_ROCE_CMD_DESTROY_MPT_BT0;
4043 		break;
4044 	case HEM_TYPE_CQC:
4045 		op = HNS_ROCE_CMD_DESTROY_CQC_BT0;
4046 		break;
4047 	case HEM_TYPE_SRQC:
4048 		op = HNS_ROCE_CMD_DESTROY_SRQC_BT0;
4049 		break;
4050 	case HEM_TYPE_SCCC:
4051 	case HEM_TYPE_QPC_TIMER:
4052 	case HEM_TYPE_CQC_TIMER:
4053 	case HEM_TYPE_GMV:
4054 		return 0;
4055 	default:
4056 		dev_warn(dev, "table %u not to be destroyed by mailbox!\n",
4057 			 table->type);
4058 		return 0;
4059 	}
4060 
4061 	op += step_idx;
4062 
4063 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
4064 	if (IS_ERR(mailbox))
4065 		return PTR_ERR(mailbox);
4066 
4067 	/* configure the tag and op */
4068 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, obj, 0, op,
4069 				HNS_ROCE_CMD_TIMEOUT_MSECS);
4070 
4071 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
4072 	return ret;
4073 }
4074 
4075 static int hns_roce_v2_qp_modify(struct hns_roce_dev *hr_dev,
4076 				 struct hns_roce_v2_qp_context *context,
4077 				 struct hns_roce_v2_qp_context *qpc_mask,
4078 				 struct hns_roce_qp *hr_qp)
4079 {
4080 	struct hns_roce_cmd_mailbox *mailbox;
4081 	int qpc_size;
4082 	int ret;
4083 
4084 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
4085 	if (IS_ERR(mailbox))
4086 		return PTR_ERR(mailbox);
4087 
4088 	/* The qpc size of HIP08 is only 256B, which is half of HIP09 */
4089 	qpc_size = hr_dev->caps.qpc_sz;
4090 	memcpy(mailbox->buf, context, qpc_size);
4091 	memcpy(mailbox->buf + qpc_size, qpc_mask, qpc_size);
4092 
4093 	ret = hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0, hr_qp->qpn, 0,
4094 				HNS_ROCE_CMD_MODIFY_QPC,
4095 				HNS_ROCE_CMD_TIMEOUT_MSECS);
4096 
4097 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
4098 
4099 	return ret;
4100 }
4101 
4102 static void set_access_flags(struct hns_roce_qp *hr_qp,
4103 			     struct hns_roce_v2_qp_context *context,
4104 			     struct hns_roce_v2_qp_context *qpc_mask,
4105 			     const struct ib_qp_attr *attr, int attr_mask)
4106 {
4107 	u8 dest_rd_atomic;
4108 	u32 access_flags;
4109 
4110 	dest_rd_atomic = (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) ?
4111 			 attr->max_dest_rd_atomic : hr_qp->resp_depth;
4112 
4113 	access_flags = (attr_mask & IB_QP_ACCESS_FLAGS) ?
4114 		       attr->qp_access_flags : hr_qp->atomic_rd_en;
4115 
4116 	if (!dest_rd_atomic)
4117 		access_flags &= IB_ACCESS_REMOTE_WRITE;
4118 
4119 	roce_set_bit(context->byte_76_srqn_op_en, V2_QPC_BYTE_76_RRE_S,
4120 		     !!(access_flags & IB_ACCESS_REMOTE_READ));
4121 	roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_RRE_S, 0);
4122 
4123 	roce_set_bit(context->byte_76_srqn_op_en, V2_QPC_BYTE_76_RWE_S,
4124 		     !!(access_flags & IB_ACCESS_REMOTE_WRITE));
4125 	roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_RWE_S, 0);
4126 
4127 	roce_set_bit(context->byte_76_srqn_op_en, V2_QPC_BYTE_76_ATE_S,
4128 		     !!(access_flags & IB_ACCESS_REMOTE_ATOMIC));
4129 	roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_ATE_S, 0);
4130 	roce_set_bit(context->byte_76_srqn_op_en, V2_QPC_BYTE_76_EXT_ATE_S,
4131 		     !!(access_flags & IB_ACCESS_REMOTE_ATOMIC));
4132 	roce_set_bit(qpc_mask->byte_76_srqn_op_en, V2_QPC_BYTE_76_EXT_ATE_S, 0);
4133 }
4134 
4135 static void set_qpc_wqe_cnt(struct hns_roce_qp *hr_qp,
4136 			    struct hns_roce_v2_qp_context *context,
4137 			    struct hns_roce_v2_qp_context *qpc_mask)
4138 {
4139 	roce_set_field(context->byte_4_sqpn_tst,
4140 		       V2_QPC_BYTE_4_SGE_SHIFT_M, V2_QPC_BYTE_4_SGE_SHIFT_S,
4141 		       to_hr_hem_entries_shift(hr_qp->sge.sge_cnt,
4142 					       hr_qp->sge.sge_shift));
4143 
4144 	roce_set_field(context->byte_20_smac_sgid_idx,
4145 		       V2_QPC_BYTE_20_SQ_SHIFT_M, V2_QPC_BYTE_20_SQ_SHIFT_S,
4146 		       ilog2(hr_qp->sq.wqe_cnt));
4147 
4148 	roce_set_field(context->byte_20_smac_sgid_idx,
4149 		       V2_QPC_BYTE_20_RQ_SHIFT_M, V2_QPC_BYTE_20_RQ_SHIFT_S,
4150 		       ilog2(hr_qp->rq.wqe_cnt));
4151 }
4152 
4153 static inline int get_cqn(struct ib_cq *ib_cq)
4154 {
4155 	return ib_cq ? to_hr_cq(ib_cq)->cqn : 0;
4156 }
4157 
4158 static inline int get_pdn(struct ib_pd *ib_pd)
4159 {
4160 	return ib_pd ? to_hr_pd(ib_pd)->pdn : 0;
4161 }
4162 
4163 static void modify_qp_reset_to_init(struct ib_qp *ibqp,
4164 				    const struct ib_qp_attr *attr,
4165 				    int attr_mask,
4166 				    struct hns_roce_v2_qp_context *context,
4167 				    struct hns_roce_v2_qp_context *qpc_mask)
4168 {
4169 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4170 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4171 
4172 	/*
4173 	 * In v2 engine, software pass context and context mask to hardware
4174 	 * when modifying qp. If software need modify some fields in context,
4175 	 * we should set all bits of the relevant fields in context mask to
4176 	 * 0 at the same time, else set them to 0x1.
4177 	 */
4178 	roce_set_field(context->byte_4_sqpn_tst, V2_QPC_BYTE_4_TST_M,
4179 		       V2_QPC_BYTE_4_TST_S, to_hr_qp_type(ibqp->qp_type));
4180 
4181 	roce_set_field(context->byte_4_sqpn_tst, V2_QPC_BYTE_4_SQPN_M,
4182 		       V2_QPC_BYTE_4_SQPN_S, hr_qp->qpn);
4183 
4184 	roce_set_field(context->byte_16_buf_ba_pg_sz, V2_QPC_BYTE_16_PD_M,
4185 		       V2_QPC_BYTE_16_PD_S, get_pdn(ibqp->pd));
4186 
4187 	roce_set_field(context->byte_20_smac_sgid_idx, V2_QPC_BYTE_20_RQWS_M,
4188 		       V2_QPC_BYTE_20_RQWS_S, ilog2(hr_qp->rq.max_gs));
4189 
4190 	set_qpc_wqe_cnt(hr_qp, context, qpc_mask);
4191 
4192 	/* No VLAN need to set 0xFFF */
4193 	roce_set_field(context->byte_24_mtu_tc, V2_QPC_BYTE_24_VLAN_ID_M,
4194 		       V2_QPC_BYTE_24_VLAN_ID_S, 0xfff);
4195 
4196 	if (ibqp->qp_type == IB_QPT_XRC_TGT) {
4197 		context->qkey_xrcd = cpu_to_le32(hr_qp->xrcdn);
4198 
4199 		roce_set_bit(context->byte_80_rnr_rx_cqn,
4200 			     V2_QPC_BYTE_80_XRC_QP_TYPE_S, 1);
4201 	}
4202 
4203 	if (hr_qp->en_flags & HNS_ROCE_QP_CAP_RQ_RECORD_DB)
4204 		roce_set_bit(context->byte_68_rq_db,
4205 			     V2_QPC_BYTE_68_RQ_RECORD_EN_S, 1);
4206 
4207 	roce_set_field(context->byte_68_rq_db,
4208 		       V2_QPC_BYTE_68_RQ_DB_RECORD_ADDR_M,
4209 		       V2_QPC_BYTE_68_RQ_DB_RECORD_ADDR_S,
4210 		       ((u32)hr_qp->rdb.dma) >> 1);
4211 	context->rq_db_record_addr = cpu_to_le32(hr_qp->rdb.dma >> 32);
4212 
4213 	if (ibqp->qp_type != IB_QPT_UD && ibqp->qp_type != IB_QPT_GSI)
4214 		roce_set_bit(context->byte_76_srqn_op_en,
4215 			     V2_QPC_BYTE_76_RQIE_S,
4216 			     !!(hr_dev->caps.flags &
4217 				HNS_ROCE_CAP_FLAG_RQ_INLINE));
4218 
4219 	roce_set_field(context->byte_80_rnr_rx_cqn, V2_QPC_BYTE_80_RX_CQN_M,
4220 		       V2_QPC_BYTE_80_RX_CQN_S, get_cqn(ibqp->recv_cq));
4221 
4222 	if (ibqp->srq) {
4223 		roce_set_bit(context->byte_76_srqn_op_en,
4224 			     V2_QPC_BYTE_76_SRQ_EN_S, 1);
4225 		roce_set_field(context->byte_76_srqn_op_en,
4226 			       V2_QPC_BYTE_76_SRQN_M, V2_QPC_BYTE_76_SRQN_S,
4227 			       to_hr_srq(ibqp->srq)->srqn);
4228 	}
4229 
4230 	roce_set_bit(context->byte_172_sq_psn, V2_QPC_BYTE_172_FRE_S, 1);
4231 
4232 	roce_set_field(context->byte_252_err_txcqn, V2_QPC_BYTE_252_TX_CQN_M,
4233 		       V2_QPC_BYTE_252_TX_CQN_S, get_cqn(ibqp->send_cq));
4234 
4235 	if (hr_dev->caps.qpc_sz < HNS_ROCE_V3_QPC_SZ)
4236 		return;
4237 
4238 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_STASH)
4239 		hr_reg_enable(&context->ext, QPCEX_STASH);
4240 }
4241 
4242 static void modify_qp_init_to_init(struct ib_qp *ibqp,
4243 				   const struct ib_qp_attr *attr, int attr_mask,
4244 				   struct hns_roce_v2_qp_context *context,
4245 				   struct hns_roce_v2_qp_context *qpc_mask)
4246 {
4247 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4248 
4249 	/*
4250 	 * In v2 engine, software pass context and context mask to hardware
4251 	 * when modifying qp. If software need modify some fields in context,
4252 	 * we should set all bits of the relevant fields in context mask to
4253 	 * 0 at the same time, else set them to 0x1.
4254 	 */
4255 	roce_set_field(context->byte_4_sqpn_tst, V2_QPC_BYTE_4_TST_M,
4256 		       V2_QPC_BYTE_4_TST_S, to_hr_qp_type(ibqp->qp_type));
4257 	roce_set_field(qpc_mask->byte_4_sqpn_tst, V2_QPC_BYTE_4_TST_M,
4258 		       V2_QPC_BYTE_4_TST_S, 0);
4259 
4260 	roce_set_field(context->byte_16_buf_ba_pg_sz, V2_QPC_BYTE_16_PD_M,
4261 		       V2_QPC_BYTE_16_PD_S, get_pdn(ibqp->pd));
4262 
4263 	roce_set_field(qpc_mask->byte_16_buf_ba_pg_sz, V2_QPC_BYTE_16_PD_M,
4264 		       V2_QPC_BYTE_16_PD_S, 0);
4265 
4266 	roce_set_field(context->byte_80_rnr_rx_cqn, V2_QPC_BYTE_80_RX_CQN_M,
4267 		       V2_QPC_BYTE_80_RX_CQN_S, get_cqn(ibqp->recv_cq));
4268 	roce_set_field(qpc_mask->byte_80_rnr_rx_cqn, V2_QPC_BYTE_80_RX_CQN_M,
4269 		       V2_QPC_BYTE_80_RX_CQN_S, 0);
4270 
4271 	roce_set_field(context->byte_252_err_txcqn, V2_QPC_BYTE_252_TX_CQN_M,
4272 		       V2_QPC_BYTE_252_TX_CQN_S, get_cqn(ibqp->send_cq));
4273 	roce_set_field(qpc_mask->byte_252_err_txcqn, V2_QPC_BYTE_252_TX_CQN_M,
4274 		       V2_QPC_BYTE_252_TX_CQN_S, 0);
4275 
4276 	if (ibqp->srq) {
4277 		roce_set_bit(context->byte_76_srqn_op_en,
4278 			     V2_QPC_BYTE_76_SRQ_EN_S, 1);
4279 		roce_set_bit(qpc_mask->byte_76_srqn_op_en,
4280 			     V2_QPC_BYTE_76_SRQ_EN_S, 0);
4281 		roce_set_field(context->byte_76_srqn_op_en,
4282 			       V2_QPC_BYTE_76_SRQN_M, V2_QPC_BYTE_76_SRQN_S,
4283 			       to_hr_srq(ibqp->srq)->srqn);
4284 		roce_set_field(qpc_mask->byte_76_srqn_op_en,
4285 			       V2_QPC_BYTE_76_SRQN_M, V2_QPC_BYTE_76_SRQN_S, 0);
4286 	}
4287 
4288 	roce_set_field(context->byte_4_sqpn_tst, V2_QPC_BYTE_4_SQPN_M,
4289 		       V2_QPC_BYTE_4_SQPN_S, hr_qp->qpn);
4290 	roce_set_field(qpc_mask->byte_4_sqpn_tst, V2_QPC_BYTE_4_SQPN_M,
4291 		       V2_QPC_BYTE_4_SQPN_S, 0);
4292 
4293 	if (attr_mask & IB_QP_DEST_QPN) {
4294 		roce_set_field(context->byte_56_dqpn_err, V2_QPC_BYTE_56_DQPN_M,
4295 			       V2_QPC_BYTE_56_DQPN_S, hr_qp->qpn);
4296 		roce_set_field(qpc_mask->byte_56_dqpn_err,
4297 			       V2_QPC_BYTE_56_DQPN_M, V2_QPC_BYTE_56_DQPN_S, 0);
4298 	}
4299 }
4300 
4301 static int config_qp_rq_buf(struct hns_roce_dev *hr_dev,
4302 			    struct hns_roce_qp *hr_qp,
4303 			    struct hns_roce_v2_qp_context *context,
4304 			    struct hns_roce_v2_qp_context *qpc_mask)
4305 {
4306 	u64 mtts[MTT_MIN_COUNT] = { 0 };
4307 	u64 wqe_sge_ba;
4308 	int count;
4309 
4310 	/* Search qp buf's mtts */
4311 	count = hns_roce_mtr_find(hr_dev, &hr_qp->mtr, hr_qp->rq.offset, mtts,
4312 				  MTT_MIN_COUNT, &wqe_sge_ba);
4313 	if (hr_qp->rq.wqe_cnt && count < 1) {
4314 		ibdev_err(&hr_dev->ib_dev,
4315 			  "failed to find RQ WQE, QPN = 0x%lx.\n", hr_qp->qpn);
4316 		return -EINVAL;
4317 	}
4318 
4319 	context->wqe_sge_ba = cpu_to_le32(wqe_sge_ba >> 3);
4320 	qpc_mask->wqe_sge_ba = 0;
4321 
4322 	/*
4323 	 * In v2 engine, software pass context and context mask to hardware
4324 	 * when modifying qp. If software need modify some fields in context,
4325 	 * we should set all bits of the relevant fields in context mask to
4326 	 * 0 at the same time, else set them to 0x1.
4327 	 */
4328 	roce_set_field(context->byte_12_sq_hop, V2_QPC_BYTE_12_WQE_SGE_BA_M,
4329 		       V2_QPC_BYTE_12_WQE_SGE_BA_S, wqe_sge_ba >> (32 + 3));
4330 	roce_set_field(qpc_mask->byte_12_sq_hop, V2_QPC_BYTE_12_WQE_SGE_BA_M,
4331 		       V2_QPC_BYTE_12_WQE_SGE_BA_S, 0);
4332 
4333 	roce_set_field(context->byte_12_sq_hop, V2_QPC_BYTE_12_SQ_HOP_NUM_M,
4334 		       V2_QPC_BYTE_12_SQ_HOP_NUM_S,
4335 		       to_hr_hem_hopnum(hr_dev->caps.wqe_sq_hop_num,
4336 					hr_qp->sq.wqe_cnt));
4337 	roce_set_field(qpc_mask->byte_12_sq_hop, V2_QPC_BYTE_12_SQ_HOP_NUM_M,
4338 		       V2_QPC_BYTE_12_SQ_HOP_NUM_S, 0);
4339 
4340 	roce_set_field(context->byte_20_smac_sgid_idx,
4341 		       V2_QPC_BYTE_20_SGE_HOP_NUM_M,
4342 		       V2_QPC_BYTE_20_SGE_HOP_NUM_S,
4343 		       to_hr_hem_hopnum(hr_dev->caps.wqe_sge_hop_num,
4344 					hr_qp->sge.sge_cnt));
4345 	roce_set_field(qpc_mask->byte_20_smac_sgid_idx,
4346 		       V2_QPC_BYTE_20_SGE_HOP_NUM_M,
4347 		       V2_QPC_BYTE_20_SGE_HOP_NUM_S, 0);
4348 
4349 	roce_set_field(context->byte_20_smac_sgid_idx,
4350 		       V2_QPC_BYTE_20_RQ_HOP_NUM_M,
4351 		       V2_QPC_BYTE_20_RQ_HOP_NUM_S,
4352 		       to_hr_hem_hopnum(hr_dev->caps.wqe_rq_hop_num,
4353 					hr_qp->rq.wqe_cnt));
4354 
4355 	roce_set_field(qpc_mask->byte_20_smac_sgid_idx,
4356 		       V2_QPC_BYTE_20_RQ_HOP_NUM_M,
4357 		       V2_QPC_BYTE_20_RQ_HOP_NUM_S, 0);
4358 
4359 	roce_set_field(context->byte_16_buf_ba_pg_sz,
4360 		       V2_QPC_BYTE_16_WQE_SGE_BA_PG_SZ_M,
4361 		       V2_QPC_BYTE_16_WQE_SGE_BA_PG_SZ_S,
4362 		       to_hr_hw_page_shift(hr_qp->mtr.hem_cfg.ba_pg_shift));
4363 	roce_set_field(qpc_mask->byte_16_buf_ba_pg_sz,
4364 		       V2_QPC_BYTE_16_WQE_SGE_BA_PG_SZ_M,
4365 		       V2_QPC_BYTE_16_WQE_SGE_BA_PG_SZ_S, 0);
4366 
4367 	roce_set_field(context->byte_16_buf_ba_pg_sz,
4368 		       V2_QPC_BYTE_16_WQE_SGE_BUF_PG_SZ_M,
4369 		       V2_QPC_BYTE_16_WQE_SGE_BUF_PG_SZ_S,
4370 		       to_hr_hw_page_shift(hr_qp->mtr.hem_cfg.buf_pg_shift));
4371 	roce_set_field(qpc_mask->byte_16_buf_ba_pg_sz,
4372 		       V2_QPC_BYTE_16_WQE_SGE_BUF_PG_SZ_M,
4373 		       V2_QPC_BYTE_16_WQE_SGE_BUF_PG_SZ_S, 0);
4374 
4375 	context->rq_cur_blk_addr = cpu_to_le32(to_hr_hw_page_addr(mtts[0]));
4376 	qpc_mask->rq_cur_blk_addr = 0;
4377 
4378 	roce_set_field(context->byte_92_srq_info,
4379 		       V2_QPC_BYTE_92_RQ_CUR_BLK_ADDR_M,
4380 		       V2_QPC_BYTE_92_RQ_CUR_BLK_ADDR_S,
4381 		       upper_32_bits(to_hr_hw_page_addr(mtts[0])));
4382 	roce_set_field(qpc_mask->byte_92_srq_info,
4383 		       V2_QPC_BYTE_92_RQ_CUR_BLK_ADDR_M,
4384 		       V2_QPC_BYTE_92_RQ_CUR_BLK_ADDR_S, 0);
4385 
4386 	context->rq_nxt_blk_addr = cpu_to_le32(to_hr_hw_page_addr(mtts[1]));
4387 	qpc_mask->rq_nxt_blk_addr = 0;
4388 
4389 	roce_set_field(context->byte_104_rq_sge,
4390 		       V2_QPC_BYTE_104_RQ_NXT_BLK_ADDR_M,
4391 		       V2_QPC_BYTE_104_RQ_NXT_BLK_ADDR_S,
4392 		       upper_32_bits(to_hr_hw_page_addr(mtts[1])));
4393 	roce_set_field(qpc_mask->byte_104_rq_sge,
4394 		       V2_QPC_BYTE_104_RQ_NXT_BLK_ADDR_M,
4395 		       V2_QPC_BYTE_104_RQ_NXT_BLK_ADDR_S, 0);
4396 
4397 	return 0;
4398 }
4399 
4400 static int config_qp_sq_buf(struct hns_roce_dev *hr_dev,
4401 			    struct hns_roce_qp *hr_qp,
4402 			    struct hns_roce_v2_qp_context *context,
4403 			    struct hns_roce_v2_qp_context *qpc_mask)
4404 {
4405 	struct ib_device *ibdev = &hr_dev->ib_dev;
4406 	u64 sge_cur_blk = 0;
4407 	u64 sq_cur_blk = 0;
4408 	int count;
4409 
4410 	/* search qp buf's mtts */
4411 	count = hns_roce_mtr_find(hr_dev, &hr_qp->mtr, 0, &sq_cur_blk, 1, NULL);
4412 	if (count < 1) {
4413 		ibdev_err(ibdev, "failed to find QP(0x%lx) SQ buf.\n",
4414 			  hr_qp->qpn);
4415 		return -EINVAL;
4416 	}
4417 	if (hr_qp->sge.sge_cnt > 0) {
4418 		count = hns_roce_mtr_find(hr_dev, &hr_qp->mtr,
4419 					  hr_qp->sge.offset,
4420 					  &sge_cur_blk, 1, NULL);
4421 		if (count < 1) {
4422 			ibdev_err(ibdev, "failed to find QP(0x%lx) SGE buf.\n",
4423 				  hr_qp->qpn);
4424 			return -EINVAL;
4425 		}
4426 	}
4427 
4428 	/*
4429 	 * In v2 engine, software pass context and context mask to hardware
4430 	 * when modifying qp. If software need modify some fields in context,
4431 	 * we should set all bits of the relevant fields in context mask to
4432 	 * 0 at the same time, else set them to 0x1.
4433 	 */
4434 	context->sq_cur_blk_addr = cpu_to_le32(to_hr_hw_page_addr(sq_cur_blk));
4435 	roce_set_field(context->byte_168_irrl_idx,
4436 		       V2_QPC_BYTE_168_SQ_CUR_BLK_ADDR_M,
4437 		       V2_QPC_BYTE_168_SQ_CUR_BLK_ADDR_S,
4438 		       upper_32_bits(to_hr_hw_page_addr(sq_cur_blk)));
4439 	qpc_mask->sq_cur_blk_addr = 0;
4440 	roce_set_field(qpc_mask->byte_168_irrl_idx,
4441 		       V2_QPC_BYTE_168_SQ_CUR_BLK_ADDR_M,
4442 		       V2_QPC_BYTE_168_SQ_CUR_BLK_ADDR_S, 0);
4443 
4444 	context->sq_cur_sge_blk_addr =
4445 		cpu_to_le32(to_hr_hw_page_addr(sge_cur_blk));
4446 	roce_set_field(context->byte_184_irrl_idx,
4447 		       V2_QPC_BYTE_184_SQ_CUR_SGE_BLK_ADDR_M,
4448 		       V2_QPC_BYTE_184_SQ_CUR_SGE_BLK_ADDR_S,
4449 		       upper_32_bits(to_hr_hw_page_addr(sge_cur_blk)));
4450 	qpc_mask->sq_cur_sge_blk_addr = 0;
4451 	roce_set_field(qpc_mask->byte_184_irrl_idx,
4452 		       V2_QPC_BYTE_184_SQ_CUR_SGE_BLK_ADDR_M,
4453 		       V2_QPC_BYTE_184_SQ_CUR_SGE_BLK_ADDR_S, 0);
4454 
4455 	context->rx_sq_cur_blk_addr =
4456 		cpu_to_le32(to_hr_hw_page_addr(sq_cur_blk));
4457 	roce_set_field(context->byte_232_irrl_sge,
4458 		       V2_QPC_BYTE_232_RX_SQ_CUR_BLK_ADDR_M,
4459 		       V2_QPC_BYTE_232_RX_SQ_CUR_BLK_ADDR_S,
4460 		       upper_32_bits(to_hr_hw_page_addr(sq_cur_blk)));
4461 	qpc_mask->rx_sq_cur_blk_addr = 0;
4462 	roce_set_field(qpc_mask->byte_232_irrl_sge,
4463 		       V2_QPC_BYTE_232_RX_SQ_CUR_BLK_ADDR_M,
4464 		       V2_QPC_BYTE_232_RX_SQ_CUR_BLK_ADDR_S, 0);
4465 
4466 	return 0;
4467 }
4468 
4469 static inline enum ib_mtu get_mtu(struct ib_qp *ibqp,
4470 				  const struct ib_qp_attr *attr)
4471 {
4472 	if (ibqp->qp_type == IB_QPT_GSI || ibqp->qp_type == IB_QPT_UD)
4473 		return IB_MTU_4096;
4474 
4475 	return attr->path_mtu;
4476 }
4477 
4478 static int modify_qp_init_to_rtr(struct ib_qp *ibqp,
4479 				 const struct ib_qp_attr *attr, int attr_mask,
4480 				 struct hns_roce_v2_qp_context *context,
4481 				 struct hns_roce_v2_qp_context *qpc_mask)
4482 {
4483 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4484 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4485 	struct ib_device *ibdev = &hr_dev->ib_dev;
4486 	dma_addr_t trrl_ba;
4487 	dma_addr_t irrl_ba;
4488 	enum ib_mtu mtu;
4489 	u8 lp_pktn_ini;
4490 	u64 *mtts;
4491 	u8 *dmac;
4492 	u8 *smac;
4493 	u32 port;
4494 	int ret;
4495 
4496 	ret = config_qp_rq_buf(hr_dev, hr_qp, context, qpc_mask);
4497 	if (ret) {
4498 		ibdev_err(ibdev, "failed to config rq buf, ret = %d.\n", ret);
4499 		return ret;
4500 	}
4501 
4502 	/* Search IRRL's mtts */
4503 	mtts = hns_roce_table_find(hr_dev, &hr_dev->qp_table.irrl_table,
4504 				   hr_qp->qpn, &irrl_ba);
4505 	if (!mtts) {
4506 		ibdev_err(ibdev, "failed to find qp irrl_table.\n");
4507 		return -EINVAL;
4508 	}
4509 
4510 	/* Search TRRL's mtts */
4511 	mtts = hns_roce_table_find(hr_dev, &hr_dev->qp_table.trrl_table,
4512 				   hr_qp->qpn, &trrl_ba);
4513 	if (!mtts) {
4514 		ibdev_err(ibdev, "failed to find qp trrl_table.\n");
4515 		return -EINVAL;
4516 	}
4517 
4518 	if (attr_mask & IB_QP_ALT_PATH) {
4519 		ibdev_err(ibdev, "INIT2RTR attr_mask (0x%x) error.\n",
4520 			  attr_mask);
4521 		return -EINVAL;
4522 	}
4523 
4524 	roce_set_field(context->byte_132_trrl, V2_QPC_BYTE_132_TRRL_BA_M,
4525 		       V2_QPC_BYTE_132_TRRL_BA_S, trrl_ba >> 4);
4526 	roce_set_field(qpc_mask->byte_132_trrl, V2_QPC_BYTE_132_TRRL_BA_M,
4527 		       V2_QPC_BYTE_132_TRRL_BA_S, 0);
4528 	context->trrl_ba = cpu_to_le32(trrl_ba >> (16 + 4));
4529 	qpc_mask->trrl_ba = 0;
4530 	roce_set_field(context->byte_140_raq, V2_QPC_BYTE_140_TRRL_BA_M,
4531 		       V2_QPC_BYTE_140_TRRL_BA_S,
4532 		       (u32)(trrl_ba >> (32 + 16 + 4)));
4533 	roce_set_field(qpc_mask->byte_140_raq, V2_QPC_BYTE_140_TRRL_BA_M,
4534 		       V2_QPC_BYTE_140_TRRL_BA_S, 0);
4535 
4536 	context->irrl_ba = cpu_to_le32(irrl_ba >> 6);
4537 	qpc_mask->irrl_ba = 0;
4538 	roce_set_field(context->byte_208_irrl, V2_QPC_BYTE_208_IRRL_BA_M,
4539 		       V2_QPC_BYTE_208_IRRL_BA_S,
4540 		       irrl_ba >> (32 + 6));
4541 	roce_set_field(qpc_mask->byte_208_irrl, V2_QPC_BYTE_208_IRRL_BA_M,
4542 		       V2_QPC_BYTE_208_IRRL_BA_S, 0);
4543 
4544 	roce_set_bit(context->byte_208_irrl, V2_QPC_BYTE_208_RMT_E2E_S, 1);
4545 	roce_set_bit(qpc_mask->byte_208_irrl, V2_QPC_BYTE_208_RMT_E2E_S, 0);
4546 
4547 	roce_set_bit(context->byte_252_err_txcqn, V2_QPC_BYTE_252_SIG_TYPE_S,
4548 		     hr_qp->sq_signal_bits);
4549 	roce_set_bit(qpc_mask->byte_252_err_txcqn, V2_QPC_BYTE_252_SIG_TYPE_S,
4550 		     0);
4551 
4552 	port = (attr_mask & IB_QP_PORT) ? (attr->port_num - 1) : hr_qp->port;
4553 
4554 	smac = (u8 *)hr_dev->dev_addr[port];
4555 	dmac = (u8 *)attr->ah_attr.roce.dmac;
4556 	/* when dmac equals smac or loop_idc is 1, it should loopback */
4557 	if (ether_addr_equal_unaligned(dmac, smac) ||
4558 	    hr_dev->loop_idc == 0x1) {
4559 		roce_set_bit(context->byte_28_at_fl, V2_QPC_BYTE_28_LBI_S, 1);
4560 		roce_set_bit(qpc_mask->byte_28_at_fl, V2_QPC_BYTE_28_LBI_S, 0);
4561 	}
4562 
4563 	if (attr_mask & IB_QP_DEST_QPN) {
4564 		roce_set_field(context->byte_56_dqpn_err, V2_QPC_BYTE_56_DQPN_M,
4565 			       V2_QPC_BYTE_56_DQPN_S, attr->dest_qp_num);
4566 		roce_set_field(qpc_mask->byte_56_dqpn_err,
4567 			       V2_QPC_BYTE_56_DQPN_M, V2_QPC_BYTE_56_DQPN_S, 0);
4568 	}
4569 
4570 	memcpy(&(context->dmac), dmac, sizeof(u32));
4571 	roce_set_field(context->byte_52_udpspn_dmac, V2_QPC_BYTE_52_DMAC_M,
4572 		       V2_QPC_BYTE_52_DMAC_S, *((u16 *)(&dmac[4])));
4573 	qpc_mask->dmac = 0;
4574 	roce_set_field(qpc_mask->byte_52_udpspn_dmac, V2_QPC_BYTE_52_DMAC_M,
4575 		       V2_QPC_BYTE_52_DMAC_S, 0);
4576 
4577 	mtu = get_mtu(ibqp, attr);
4578 	hr_qp->path_mtu = mtu;
4579 
4580 	if (attr_mask & IB_QP_PATH_MTU) {
4581 		roce_set_field(context->byte_24_mtu_tc, V2_QPC_BYTE_24_MTU_M,
4582 			       V2_QPC_BYTE_24_MTU_S, mtu);
4583 		roce_set_field(qpc_mask->byte_24_mtu_tc, V2_QPC_BYTE_24_MTU_M,
4584 			       V2_QPC_BYTE_24_MTU_S, 0);
4585 	}
4586 
4587 #define MAX_LP_MSG_LEN 65536
4588 	/* MTU * (2 ^ LP_PKTN_INI) shouldn't be bigger than 64KB */
4589 	lp_pktn_ini = ilog2(MAX_LP_MSG_LEN / ib_mtu_enum_to_int(mtu));
4590 
4591 	roce_set_field(context->byte_56_dqpn_err, V2_QPC_BYTE_56_LP_PKTN_INI_M,
4592 		       V2_QPC_BYTE_56_LP_PKTN_INI_S, lp_pktn_ini);
4593 	roce_set_field(qpc_mask->byte_56_dqpn_err, V2_QPC_BYTE_56_LP_PKTN_INI_M,
4594 		       V2_QPC_BYTE_56_LP_PKTN_INI_S, 0);
4595 
4596 	/* ACK_REQ_FREQ should be larger than or equal to LP_PKTN_INI */
4597 	roce_set_field(context->byte_172_sq_psn, V2_QPC_BYTE_172_ACK_REQ_FREQ_M,
4598 		       V2_QPC_BYTE_172_ACK_REQ_FREQ_S, lp_pktn_ini);
4599 	roce_set_field(qpc_mask->byte_172_sq_psn,
4600 		       V2_QPC_BYTE_172_ACK_REQ_FREQ_M,
4601 		       V2_QPC_BYTE_172_ACK_REQ_FREQ_S, 0);
4602 
4603 	roce_set_bit(qpc_mask->byte_108_rx_reqepsn,
4604 		     V2_QPC_BYTE_108_RX_REQ_PSN_ERR_S, 0);
4605 	roce_set_field(qpc_mask->byte_96_rx_reqmsn, V2_QPC_BYTE_96_RX_REQ_MSN_M,
4606 		       V2_QPC_BYTE_96_RX_REQ_MSN_S, 0);
4607 	roce_set_field(qpc_mask->byte_108_rx_reqepsn,
4608 		       V2_QPC_BYTE_108_RX_REQ_LAST_OPTYPE_M,
4609 		       V2_QPC_BYTE_108_RX_REQ_LAST_OPTYPE_S, 0);
4610 
4611 	context->rq_rnr_timer = 0;
4612 	qpc_mask->rq_rnr_timer = 0;
4613 
4614 	roce_set_field(qpc_mask->byte_132_trrl, V2_QPC_BYTE_132_TRRL_HEAD_MAX_M,
4615 		       V2_QPC_BYTE_132_TRRL_HEAD_MAX_S, 0);
4616 	roce_set_field(qpc_mask->byte_132_trrl, V2_QPC_BYTE_132_TRRL_TAIL_MAX_M,
4617 		       V2_QPC_BYTE_132_TRRL_TAIL_MAX_S, 0);
4618 
4619 	/* rocee send 2^lp_sgen_ini segs every time */
4620 	roce_set_field(context->byte_168_irrl_idx,
4621 		       V2_QPC_BYTE_168_LP_SGEN_INI_M,
4622 		       V2_QPC_BYTE_168_LP_SGEN_INI_S, 3);
4623 	roce_set_field(qpc_mask->byte_168_irrl_idx,
4624 		       V2_QPC_BYTE_168_LP_SGEN_INI_M,
4625 		       V2_QPC_BYTE_168_LP_SGEN_INI_S, 0);
4626 
4627 	return 0;
4628 }
4629 
4630 static int modify_qp_rtr_to_rts(struct ib_qp *ibqp,
4631 				const struct ib_qp_attr *attr, int attr_mask,
4632 				struct hns_roce_v2_qp_context *context,
4633 				struct hns_roce_v2_qp_context *qpc_mask)
4634 {
4635 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4636 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4637 	struct ib_device *ibdev = &hr_dev->ib_dev;
4638 	int ret;
4639 
4640 	/* Not support alternate path and path migration */
4641 	if (attr_mask & (IB_QP_ALT_PATH | IB_QP_PATH_MIG_STATE)) {
4642 		ibdev_err(ibdev, "RTR2RTS attr_mask (0x%x)error\n", attr_mask);
4643 		return -EINVAL;
4644 	}
4645 
4646 	ret = config_qp_sq_buf(hr_dev, hr_qp, context, qpc_mask);
4647 	if (ret) {
4648 		ibdev_err(ibdev, "failed to config sq buf, ret = %d.\n", ret);
4649 		return ret;
4650 	}
4651 
4652 	/*
4653 	 * Set some fields in context to zero, Because the default values
4654 	 * of all fields in context are zero, we need not set them to 0 again.
4655 	 * but we should set the relevant fields of context mask to 0.
4656 	 */
4657 	roce_set_field(qpc_mask->byte_232_irrl_sge,
4658 		       V2_QPC_BYTE_232_IRRL_SGE_IDX_M,
4659 		       V2_QPC_BYTE_232_IRRL_SGE_IDX_S, 0);
4660 
4661 	roce_set_field(qpc_mask->byte_240_irrl_tail,
4662 		       V2_QPC_BYTE_240_RX_ACK_MSN_M,
4663 		       V2_QPC_BYTE_240_RX_ACK_MSN_S, 0);
4664 
4665 	roce_set_field(qpc_mask->byte_248_ack_psn,
4666 		       V2_QPC_BYTE_248_ACK_LAST_OPTYPE_M,
4667 		       V2_QPC_BYTE_248_ACK_LAST_OPTYPE_S, 0);
4668 	roce_set_bit(qpc_mask->byte_248_ack_psn,
4669 		     V2_QPC_BYTE_248_IRRL_PSN_VLD_S, 0);
4670 	roce_set_field(qpc_mask->byte_248_ack_psn,
4671 		       V2_QPC_BYTE_248_IRRL_PSN_M,
4672 		       V2_QPC_BYTE_248_IRRL_PSN_S, 0);
4673 
4674 	roce_set_field(qpc_mask->byte_240_irrl_tail,
4675 		       V2_QPC_BYTE_240_IRRL_TAIL_REAL_M,
4676 		       V2_QPC_BYTE_240_IRRL_TAIL_REAL_S, 0);
4677 
4678 	roce_set_field(qpc_mask->byte_220_retry_psn_msn,
4679 		       V2_QPC_BYTE_220_RETRY_MSG_MSN_M,
4680 		       V2_QPC_BYTE_220_RETRY_MSG_MSN_S, 0);
4681 
4682 	roce_set_bit(qpc_mask->byte_248_ack_psn,
4683 		     V2_QPC_BYTE_248_RNR_RETRY_FLAG_S, 0);
4684 
4685 	roce_set_field(qpc_mask->byte_212_lsn, V2_QPC_BYTE_212_CHECK_FLG_M,
4686 		       V2_QPC_BYTE_212_CHECK_FLG_S, 0);
4687 
4688 	roce_set_field(context->byte_212_lsn, V2_QPC_BYTE_212_LSN_M,
4689 		       V2_QPC_BYTE_212_LSN_S, 0x100);
4690 	roce_set_field(qpc_mask->byte_212_lsn, V2_QPC_BYTE_212_LSN_M,
4691 		       V2_QPC_BYTE_212_LSN_S, 0);
4692 
4693 	roce_set_field(qpc_mask->byte_196_sq_psn, V2_QPC_BYTE_196_IRRL_HEAD_M,
4694 		       V2_QPC_BYTE_196_IRRL_HEAD_S, 0);
4695 
4696 	return 0;
4697 }
4698 
4699 static inline u16 get_udp_sport(u32 fl, u32 lqpn, u32 rqpn)
4700 {
4701 	if (!fl)
4702 		fl = rdma_calc_flow_label(lqpn, rqpn);
4703 
4704 	return rdma_flow_label_to_udp_sport(fl);
4705 }
4706 
4707 static int get_dip_ctx_idx(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
4708 			   u32 *dip_idx)
4709 {
4710 	const struct ib_global_route *grh = rdma_ah_read_grh(&attr->ah_attr);
4711 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4712 	struct hns_roce_dip *hr_dip;
4713 	unsigned long flags;
4714 	int ret = 0;
4715 
4716 	spin_lock_irqsave(&hr_dev->dip_list_lock, flags);
4717 
4718 	list_for_each_entry(hr_dip, &hr_dev->dip_list, node) {
4719 		if (!memcmp(grh->dgid.raw, hr_dip->dgid, 16))
4720 			goto out;
4721 	}
4722 
4723 	/* If no dgid is found, a new dip and a mapping between dgid and
4724 	 * dip_idx will be created.
4725 	 */
4726 	hr_dip = kzalloc(sizeof(*hr_dip), GFP_ATOMIC);
4727 	if (!hr_dip) {
4728 		ret = -ENOMEM;
4729 		goto out;
4730 	}
4731 
4732 	memcpy(hr_dip->dgid, grh->dgid.raw, sizeof(grh->dgid.raw));
4733 	hr_dip->dip_idx = *dip_idx = ibqp->qp_num;
4734 	list_add_tail(&hr_dip->node, &hr_dev->dip_list);
4735 
4736 out:
4737 	spin_unlock_irqrestore(&hr_dev->dip_list_lock, flags);
4738 	return ret;
4739 }
4740 
4741 enum {
4742 	CONG_DCQCN,
4743 	CONG_WINDOW,
4744 };
4745 
4746 enum {
4747 	UNSUPPORT_CONG_LEVEL,
4748 	SUPPORT_CONG_LEVEL,
4749 };
4750 
4751 enum {
4752 	CONG_LDCP,
4753 	CONG_HC3,
4754 };
4755 
4756 enum {
4757 	DIP_INVALID,
4758 	DIP_VALID,
4759 };
4760 
4761 static int check_cong_type(struct ib_qp *ibqp,
4762 			   struct hns_roce_congestion_algorithm *cong_alg)
4763 {
4764 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4765 
4766 	/* different congestion types match different configurations */
4767 	switch (hr_dev->caps.cong_type) {
4768 	case CONG_TYPE_DCQCN:
4769 		cong_alg->alg_sel = CONG_DCQCN;
4770 		cong_alg->alg_sub_sel = UNSUPPORT_CONG_LEVEL;
4771 		cong_alg->dip_vld = DIP_INVALID;
4772 		break;
4773 	case CONG_TYPE_LDCP:
4774 		cong_alg->alg_sel = CONG_WINDOW;
4775 		cong_alg->alg_sub_sel = CONG_LDCP;
4776 		cong_alg->dip_vld = DIP_INVALID;
4777 		break;
4778 	case CONG_TYPE_HC3:
4779 		cong_alg->alg_sel = CONG_WINDOW;
4780 		cong_alg->alg_sub_sel = CONG_HC3;
4781 		cong_alg->dip_vld = DIP_INVALID;
4782 		break;
4783 	case CONG_TYPE_DIP:
4784 		cong_alg->alg_sel = CONG_DCQCN;
4785 		cong_alg->alg_sub_sel = UNSUPPORT_CONG_LEVEL;
4786 		cong_alg->dip_vld = DIP_VALID;
4787 		break;
4788 	default:
4789 		ibdev_err(&hr_dev->ib_dev,
4790 			  "error type(%u) for congestion selection.\n",
4791 			  hr_dev->caps.cong_type);
4792 		return -EINVAL;
4793 	}
4794 
4795 	return 0;
4796 }
4797 
4798 static int fill_cong_field(struct ib_qp *ibqp, const struct ib_qp_attr *attr,
4799 			   struct hns_roce_v2_qp_context *context,
4800 			   struct hns_roce_v2_qp_context *qpc_mask)
4801 {
4802 	const struct ib_global_route *grh = rdma_ah_read_grh(&attr->ah_attr);
4803 	struct hns_roce_congestion_algorithm cong_field;
4804 	struct ib_device *ibdev = ibqp->device;
4805 	struct hns_roce_dev *hr_dev = to_hr_dev(ibdev);
4806 	u32 dip_idx = 0;
4807 	int ret;
4808 
4809 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08 ||
4810 	    grh->sgid_attr->gid_type == IB_GID_TYPE_ROCE)
4811 		return 0;
4812 
4813 	ret = check_cong_type(ibqp, &cong_field);
4814 	if (ret)
4815 		return ret;
4816 
4817 	hr_reg_write(context, QPC_CONG_ALGO_TMPL_ID, hr_dev->cong_algo_tmpl_id +
4818 		     hr_dev->caps.cong_type * HNS_ROCE_CONG_SIZE);
4819 	hr_reg_write(qpc_mask, QPC_CONG_ALGO_TMPL_ID, 0);
4820 	hr_reg_write(&context->ext, QPCEX_CONG_ALG_SEL, cong_field.alg_sel);
4821 	hr_reg_write(&qpc_mask->ext, QPCEX_CONG_ALG_SEL, 0);
4822 	hr_reg_write(&context->ext, QPCEX_CONG_ALG_SUB_SEL,
4823 		     cong_field.alg_sub_sel);
4824 	hr_reg_write(&qpc_mask->ext, QPCEX_CONG_ALG_SUB_SEL, 0);
4825 	hr_reg_write(&context->ext, QPCEX_DIP_CTX_IDX_VLD, cong_field.dip_vld);
4826 	hr_reg_write(&qpc_mask->ext, QPCEX_DIP_CTX_IDX_VLD, 0);
4827 
4828 	/* if dip is disabled, there is no need to set dip idx */
4829 	if (cong_field.dip_vld == 0)
4830 		return 0;
4831 
4832 	ret = get_dip_ctx_idx(ibqp, attr, &dip_idx);
4833 	if (ret) {
4834 		ibdev_err(ibdev, "failed to fill cong field, ret = %d.\n", ret);
4835 		return ret;
4836 	}
4837 
4838 	hr_reg_write(&context->ext, QPCEX_DIP_CTX_IDX, dip_idx);
4839 	hr_reg_write(&qpc_mask->ext, QPCEX_DIP_CTX_IDX, 0);
4840 
4841 	return 0;
4842 }
4843 
4844 static int hns_roce_v2_set_path(struct ib_qp *ibqp,
4845 				const struct ib_qp_attr *attr,
4846 				int attr_mask,
4847 				struct hns_roce_v2_qp_context *context,
4848 				struct hns_roce_v2_qp_context *qpc_mask)
4849 {
4850 	const struct ib_global_route *grh = rdma_ah_read_grh(&attr->ah_attr);
4851 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4852 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
4853 	struct ib_device *ibdev = &hr_dev->ib_dev;
4854 	const struct ib_gid_attr *gid_attr = NULL;
4855 	int is_roce_protocol;
4856 	u16 vlan_id = 0xffff;
4857 	bool is_udp = false;
4858 	u8 ib_port;
4859 	u8 hr_port;
4860 	int ret;
4861 
4862 	ib_port = (attr_mask & IB_QP_PORT) ? attr->port_num : hr_qp->port + 1;
4863 	hr_port = ib_port - 1;
4864 	is_roce_protocol = rdma_cap_eth_ah(&hr_dev->ib_dev, ib_port) &&
4865 			   rdma_ah_get_ah_flags(&attr->ah_attr) & IB_AH_GRH;
4866 
4867 	if (is_roce_protocol) {
4868 		gid_attr = attr->ah_attr.grh.sgid_attr;
4869 		ret = rdma_read_gid_l2_fields(gid_attr, &vlan_id, NULL);
4870 		if (ret)
4871 			return ret;
4872 
4873 		if (gid_attr)
4874 			is_udp = (gid_attr->gid_type ==
4875 				 IB_GID_TYPE_ROCE_UDP_ENCAP);
4876 	}
4877 
4878 	/* Only HIP08 needs to set the vlan_en bits in QPC */
4879 	if (vlan_id < VLAN_N_VID &&
4880 	    hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08) {
4881 		roce_set_bit(context->byte_76_srqn_op_en,
4882 			     V2_QPC_BYTE_76_RQ_VLAN_EN_S, 1);
4883 		roce_set_bit(qpc_mask->byte_76_srqn_op_en,
4884 			     V2_QPC_BYTE_76_RQ_VLAN_EN_S, 0);
4885 		roce_set_bit(context->byte_168_irrl_idx,
4886 			     V2_QPC_BYTE_168_SQ_VLAN_EN_S, 1);
4887 		roce_set_bit(qpc_mask->byte_168_irrl_idx,
4888 			     V2_QPC_BYTE_168_SQ_VLAN_EN_S, 0);
4889 	}
4890 
4891 	roce_set_field(context->byte_24_mtu_tc, V2_QPC_BYTE_24_VLAN_ID_M,
4892 		       V2_QPC_BYTE_24_VLAN_ID_S, vlan_id);
4893 	roce_set_field(qpc_mask->byte_24_mtu_tc, V2_QPC_BYTE_24_VLAN_ID_M,
4894 		       V2_QPC_BYTE_24_VLAN_ID_S, 0);
4895 
4896 	if (grh->sgid_index >= hr_dev->caps.gid_table_len[hr_port]) {
4897 		ibdev_err(ibdev, "sgid_index(%u) too large. max is %d\n",
4898 			  grh->sgid_index, hr_dev->caps.gid_table_len[hr_port]);
4899 		return -EINVAL;
4900 	}
4901 
4902 	if (attr->ah_attr.type != RDMA_AH_ATTR_TYPE_ROCE) {
4903 		ibdev_err(ibdev, "ah attr is not RDMA roce type\n");
4904 		return -EINVAL;
4905 	}
4906 
4907 	roce_set_field(context->byte_52_udpspn_dmac, V2_QPC_BYTE_52_UDPSPN_M,
4908 		       V2_QPC_BYTE_52_UDPSPN_S,
4909 		       is_udp ? get_udp_sport(grh->flow_label, ibqp->qp_num,
4910 					      attr->dest_qp_num) : 0);
4911 
4912 	roce_set_field(qpc_mask->byte_52_udpspn_dmac, V2_QPC_BYTE_52_UDPSPN_M,
4913 		       V2_QPC_BYTE_52_UDPSPN_S, 0);
4914 
4915 	roce_set_field(context->byte_20_smac_sgid_idx,
4916 		       V2_QPC_BYTE_20_SGID_IDX_M, V2_QPC_BYTE_20_SGID_IDX_S,
4917 		       grh->sgid_index);
4918 
4919 	roce_set_field(qpc_mask->byte_20_smac_sgid_idx,
4920 		       V2_QPC_BYTE_20_SGID_IDX_M, V2_QPC_BYTE_20_SGID_IDX_S, 0);
4921 
4922 	roce_set_field(context->byte_24_mtu_tc, V2_QPC_BYTE_24_HOP_LIMIT_M,
4923 		       V2_QPC_BYTE_24_HOP_LIMIT_S, grh->hop_limit);
4924 	roce_set_field(qpc_mask->byte_24_mtu_tc, V2_QPC_BYTE_24_HOP_LIMIT_M,
4925 		       V2_QPC_BYTE_24_HOP_LIMIT_S, 0);
4926 
4927 	ret = fill_cong_field(ibqp, attr, context, qpc_mask);
4928 	if (ret)
4929 		return ret;
4930 
4931 	roce_set_field(context->byte_24_mtu_tc, V2_QPC_BYTE_24_TC_M,
4932 		       V2_QPC_BYTE_24_TC_S, get_tclass(&attr->ah_attr.grh));
4933 	roce_set_field(qpc_mask->byte_24_mtu_tc, V2_QPC_BYTE_24_TC_M,
4934 		       V2_QPC_BYTE_24_TC_S, 0);
4935 
4936 	roce_set_field(context->byte_28_at_fl, V2_QPC_BYTE_28_FL_M,
4937 		       V2_QPC_BYTE_28_FL_S, grh->flow_label);
4938 	roce_set_field(qpc_mask->byte_28_at_fl, V2_QPC_BYTE_28_FL_M,
4939 		       V2_QPC_BYTE_28_FL_S, 0);
4940 	memcpy(context->dgid, grh->dgid.raw, sizeof(grh->dgid.raw));
4941 	memset(qpc_mask->dgid, 0, sizeof(grh->dgid.raw));
4942 
4943 	hr_qp->sl = rdma_ah_get_sl(&attr->ah_attr);
4944 	if (unlikely(hr_qp->sl > MAX_SERVICE_LEVEL)) {
4945 		ibdev_err(ibdev,
4946 			  "failed to fill QPC, sl (%d) shouldn't be larger than %d.\n",
4947 			  hr_qp->sl, MAX_SERVICE_LEVEL);
4948 		return -EINVAL;
4949 	}
4950 
4951 	roce_set_field(context->byte_28_at_fl, V2_QPC_BYTE_28_SL_M,
4952 		       V2_QPC_BYTE_28_SL_S, hr_qp->sl);
4953 	roce_set_field(qpc_mask->byte_28_at_fl, V2_QPC_BYTE_28_SL_M,
4954 		       V2_QPC_BYTE_28_SL_S, 0);
4955 
4956 	return 0;
4957 }
4958 
4959 static bool check_qp_state(enum ib_qp_state cur_state,
4960 			   enum ib_qp_state new_state)
4961 {
4962 	static const bool sm[][IB_QPS_ERR + 1] = {
4963 		[IB_QPS_RESET] = { [IB_QPS_RESET] = true,
4964 				   [IB_QPS_INIT] = true },
4965 		[IB_QPS_INIT] = { [IB_QPS_RESET] = true,
4966 				  [IB_QPS_INIT] = true,
4967 				  [IB_QPS_RTR] = true,
4968 				  [IB_QPS_ERR] = true },
4969 		[IB_QPS_RTR] = { [IB_QPS_RESET] = true,
4970 				 [IB_QPS_RTS] = true,
4971 				 [IB_QPS_ERR] = true },
4972 		[IB_QPS_RTS] = { [IB_QPS_RESET] = true,
4973 				 [IB_QPS_RTS] = true,
4974 				 [IB_QPS_ERR] = true },
4975 		[IB_QPS_SQD] = {},
4976 		[IB_QPS_SQE] = {},
4977 		[IB_QPS_ERR] = { [IB_QPS_RESET] = true, [IB_QPS_ERR] = true }
4978 	};
4979 
4980 	return sm[cur_state][new_state];
4981 }
4982 
4983 static int hns_roce_v2_set_abs_fields(struct ib_qp *ibqp,
4984 				      const struct ib_qp_attr *attr,
4985 				      int attr_mask,
4986 				      enum ib_qp_state cur_state,
4987 				      enum ib_qp_state new_state,
4988 				      struct hns_roce_v2_qp_context *context,
4989 				      struct hns_roce_v2_qp_context *qpc_mask)
4990 {
4991 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
4992 	int ret = 0;
4993 
4994 	if (!check_qp_state(cur_state, new_state)) {
4995 		ibdev_err(&hr_dev->ib_dev, "Illegal state for QP!\n");
4996 		return -EINVAL;
4997 	}
4998 
4999 	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) {
5000 		memset(qpc_mask, 0, hr_dev->caps.qpc_sz);
5001 		modify_qp_reset_to_init(ibqp, attr, attr_mask, context,
5002 					qpc_mask);
5003 	} else if (cur_state == IB_QPS_INIT && new_state == IB_QPS_INIT) {
5004 		modify_qp_init_to_init(ibqp, attr, attr_mask, context,
5005 				       qpc_mask);
5006 	} else if (cur_state == IB_QPS_INIT && new_state == IB_QPS_RTR) {
5007 		ret = modify_qp_init_to_rtr(ibqp, attr, attr_mask, context,
5008 					    qpc_mask);
5009 	} else if (cur_state == IB_QPS_RTR && new_state == IB_QPS_RTS) {
5010 		ret = modify_qp_rtr_to_rts(ibqp, attr, attr_mask, context,
5011 					   qpc_mask);
5012 	}
5013 
5014 	return ret;
5015 }
5016 
5017 static int hns_roce_v2_set_opt_fields(struct ib_qp *ibqp,
5018 				      const struct ib_qp_attr *attr,
5019 				      int attr_mask,
5020 				      struct hns_roce_v2_qp_context *context,
5021 				      struct hns_roce_v2_qp_context *qpc_mask)
5022 {
5023 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5024 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5025 	int ret = 0;
5026 
5027 	if (attr_mask & IB_QP_AV) {
5028 		ret = hns_roce_v2_set_path(ibqp, attr, attr_mask, context,
5029 					   qpc_mask);
5030 		if (ret)
5031 			return ret;
5032 	}
5033 
5034 	if (attr_mask & IB_QP_TIMEOUT) {
5035 		if (attr->timeout < 31) {
5036 			roce_set_field(context->byte_28_at_fl,
5037 				       V2_QPC_BYTE_28_AT_M, V2_QPC_BYTE_28_AT_S,
5038 				       attr->timeout);
5039 			roce_set_field(qpc_mask->byte_28_at_fl,
5040 				       V2_QPC_BYTE_28_AT_M, V2_QPC_BYTE_28_AT_S,
5041 				       0);
5042 		} else {
5043 			ibdev_warn(&hr_dev->ib_dev,
5044 				   "Local ACK timeout shall be 0 to 30.\n");
5045 		}
5046 	}
5047 
5048 	if (attr_mask & IB_QP_RETRY_CNT) {
5049 		roce_set_field(context->byte_212_lsn,
5050 			       V2_QPC_BYTE_212_RETRY_NUM_INIT_M,
5051 			       V2_QPC_BYTE_212_RETRY_NUM_INIT_S,
5052 			       attr->retry_cnt);
5053 		roce_set_field(qpc_mask->byte_212_lsn,
5054 			       V2_QPC_BYTE_212_RETRY_NUM_INIT_M,
5055 			       V2_QPC_BYTE_212_RETRY_NUM_INIT_S, 0);
5056 
5057 		roce_set_field(context->byte_212_lsn,
5058 			       V2_QPC_BYTE_212_RETRY_CNT_M,
5059 			       V2_QPC_BYTE_212_RETRY_CNT_S, attr->retry_cnt);
5060 		roce_set_field(qpc_mask->byte_212_lsn,
5061 			       V2_QPC_BYTE_212_RETRY_CNT_M,
5062 			       V2_QPC_BYTE_212_RETRY_CNT_S, 0);
5063 	}
5064 
5065 	if (attr_mask & IB_QP_RNR_RETRY) {
5066 		roce_set_field(context->byte_244_rnr_rxack,
5067 			       V2_QPC_BYTE_244_RNR_NUM_INIT_M,
5068 			       V2_QPC_BYTE_244_RNR_NUM_INIT_S, attr->rnr_retry);
5069 		roce_set_field(qpc_mask->byte_244_rnr_rxack,
5070 			       V2_QPC_BYTE_244_RNR_NUM_INIT_M,
5071 			       V2_QPC_BYTE_244_RNR_NUM_INIT_S, 0);
5072 
5073 		roce_set_field(context->byte_244_rnr_rxack,
5074 			       V2_QPC_BYTE_244_RNR_CNT_M,
5075 			       V2_QPC_BYTE_244_RNR_CNT_S, attr->rnr_retry);
5076 		roce_set_field(qpc_mask->byte_244_rnr_rxack,
5077 			       V2_QPC_BYTE_244_RNR_CNT_M,
5078 			       V2_QPC_BYTE_244_RNR_CNT_S, 0);
5079 	}
5080 
5081 	if (attr_mask & IB_QP_SQ_PSN) {
5082 		roce_set_field(context->byte_172_sq_psn,
5083 			       V2_QPC_BYTE_172_SQ_CUR_PSN_M,
5084 			       V2_QPC_BYTE_172_SQ_CUR_PSN_S, attr->sq_psn);
5085 		roce_set_field(qpc_mask->byte_172_sq_psn,
5086 			       V2_QPC_BYTE_172_SQ_CUR_PSN_M,
5087 			       V2_QPC_BYTE_172_SQ_CUR_PSN_S, 0);
5088 
5089 		roce_set_field(context->byte_196_sq_psn,
5090 			       V2_QPC_BYTE_196_SQ_MAX_PSN_M,
5091 			       V2_QPC_BYTE_196_SQ_MAX_PSN_S, attr->sq_psn);
5092 		roce_set_field(qpc_mask->byte_196_sq_psn,
5093 			       V2_QPC_BYTE_196_SQ_MAX_PSN_M,
5094 			       V2_QPC_BYTE_196_SQ_MAX_PSN_S, 0);
5095 
5096 		roce_set_field(context->byte_220_retry_psn_msn,
5097 			       V2_QPC_BYTE_220_RETRY_MSG_PSN_M,
5098 			       V2_QPC_BYTE_220_RETRY_MSG_PSN_S, attr->sq_psn);
5099 		roce_set_field(qpc_mask->byte_220_retry_psn_msn,
5100 			       V2_QPC_BYTE_220_RETRY_MSG_PSN_M,
5101 			       V2_QPC_BYTE_220_RETRY_MSG_PSN_S, 0);
5102 
5103 		roce_set_field(context->byte_224_retry_msg,
5104 			       V2_QPC_BYTE_224_RETRY_MSG_PSN_M,
5105 			       V2_QPC_BYTE_224_RETRY_MSG_PSN_S,
5106 			       attr->sq_psn >> V2_QPC_BYTE_220_RETRY_MSG_PSN_S);
5107 		roce_set_field(qpc_mask->byte_224_retry_msg,
5108 			       V2_QPC_BYTE_224_RETRY_MSG_PSN_M,
5109 			       V2_QPC_BYTE_224_RETRY_MSG_PSN_S, 0);
5110 
5111 		roce_set_field(context->byte_224_retry_msg,
5112 			       V2_QPC_BYTE_224_RETRY_MSG_FPKT_PSN_M,
5113 			       V2_QPC_BYTE_224_RETRY_MSG_FPKT_PSN_S,
5114 			       attr->sq_psn);
5115 		roce_set_field(qpc_mask->byte_224_retry_msg,
5116 			       V2_QPC_BYTE_224_RETRY_MSG_FPKT_PSN_M,
5117 			       V2_QPC_BYTE_224_RETRY_MSG_FPKT_PSN_S, 0);
5118 
5119 		roce_set_field(context->byte_244_rnr_rxack,
5120 			       V2_QPC_BYTE_244_RX_ACK_EPSN_M,
5121 			       V2_QPC_BYTE_244_RX_ACK_EPSN_S, attr->sq_psn);
5122 		roce_set_field(qpc_mask->byte_244_rnr_rxack,
5123 			       V2_QPC_BYTE_244_RX_ACK_EPSN_M,
5124 			       V2_QPC_BYTE_244_RX_ACK_EPSN_S, 0);
5125 	}
5126 
5127 	if ((attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) &&
5128 	     attr->max_dest_rd_atomic) {
5129 		roce_set_field(context->byte_140_raq, V2_QPC_BYTE_140_RR_MAX_M,
5130 			       V2_QPC_BYTE_140_RR_MAX_S,
5131 			       fls(attr->max_dest_rd_atomic - 1));
5132 		roce_set_field(qpc_mask->byte_140_raq, V2_QPC_BYTE_140_RR_MAX_M,
5133 			       V2_QPC_BYTE_140_RR_MAX_S, 0);
5134 	}
5135 
5136 	if ((attr_mask & IB_QP_MAX_QP_RD_ATOMIC) && attr->max_rd_atomic) {
5137 		roce_set_field(context->byte_208_irrl, V2_QPC_BYTE_208_SR_MAX_M,
5138 			       V2_QPC_BYTE_208_SR_MAX_S,
5139 			       fls(attr->max_rd_atomic - 1));
5140 		roce_set_field(qpc_mask->byte_208_irrl,
5141 			       V2_QPC_BYTE_208_SR_MAX_M,
5142 			       V2_QPC_BYTE_208_SR_MAX_S, 0);
5143 	}
5144 
5145 	if (attr_mask & (IB_QP_ACCESS_FLAGS | IB_QP_MAX_DEST_RD_ATOMIC))
5146 		set_access_flags(hr_qp, context, qpc_mask, attr, attr_mask);
5147 
5148 	if (attr_mask & IB_QP_MIN_RNR_TIMER) {
5149 		roce_set_field(context->byte_80_rnr_rx_cqn,
5150 			       V2_QPC_BYTE_80_MIN_RNR_TIME_M,
5151 			       V2_QPC_BYTE_80_MIN_RNR_TIME_S,
5152 			       attr->min_rnr_timer);
5153 		roce_set_field(qpc_mask->byte_80_rnr_rx_cqn,
5154 			       V2_QPC_BYTE_80_MIN_RNR_TIME_M,
5155 			       V2_QPC_BYTE_80_MIN_RNR_TIME_S, 0);
5156 	}
5157 
5158 	if (attr_mask & IB_QP_RQ_PSN) {
5159 		roce_set_field(context->byte_108_rx_reqepsn,
5160 			       V2_QPC_BYTE_108_RX_REQ_EPSN_M,
5161 			       V2_QPC_BYTE_108_RX_REQ_EPSN_S, attr->rq_psn);
5162 		roce_set_field(qpc_mask->byte_108_rx_reqepsn,
5163 			       V2_QPC_BYTE_108_RX_REQ_EPSN_M,
5164 			       V2_QPC_BYTE_108_RX_REQ_EPSN_S, 0);
5165 
5166 		roce_set_field(context->byte_152_raq, V2_QPC_BYTE_152_RAQ_PSN_M,
5167 			       V2_QPC_BYTE_152_RAQ_PSN_S, attr->rq_psn - 1);
5168 		roce_set_field(qpc_mask->byte_152_raq,
5169 			       V2_QPC_BYTE_152_RAQ_PSN_M,
5170 			       V2_QPC_BYTE_152_RAQ_PSN_S, 0);
5171 	}
5172 
5173 	if (attr_mask & IB_QP_QKEY) {
5174 		context->qkey_xrcd = cpu_to_le32(attr->qkey);
5175 		qpc_mask->qkey_xrcd = 0;
5176 		hr_qp->qkey = attr->qkey;
5177 	}
5178 
5179 	return ret;
5180 }
5181 
5182 static void hns_roce_v2_record_opt_fields(struct ib_qp *ibqp,
5183 					  const struct ib_qp_attr *attr,
5184 					  int attr_mask)
5185 {
5186 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5187 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5188 
5189 	if (attr_mask & IB_QP_ACCESS_FLAGS)
5190 		hr_qp->atomic_rd_en = attr->qp_access_flags;
5191 
5192 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
5193 		hr_qp->resp_depth = attr->max_dest_rd_atomic;
5194 	if (attr_mask & IB_QP_PORT) {
5195 		hr_qp->port = attr->port_num - 1;
5196 		hr_qp->phy_port = hr_dev->iboe.phy_port[hr_qp->port];
5197 	}
5198 }
5199 
5200 static void clear_qp(struct hns_roce_qp *hr_qp)
5201 {
5202 	struct ib_qp *ibqp = &hr_qp->ibqp;
5203 
5204 	if (ibqp->send_cq)
5205 		hns_roce_v2_cq_clean(to_hr_cq(ibqp->send_cq),
5206 				     hr_qp->qpn, NULL);
5207 
5208 	if (ibqp->recv_cq  && ibqp->recv_cq != ibqp->send_cq)
5209 		hns_roce_v2_cq_clean(to_hr_cq(ibqp->recv_cq),
5210 				     hr_qp->qpn, ibqp->srq ?
5211 				     to_hr_srq(ibqp->srq) : NULL);
5212 
5213 	if (hr_qp->en_flags & HNS_ROCE_QP_CAP_RQ_RECORD_DB)
5214 		*hr_qp->rdb.db_record = 0;
5215 
5216 	hr_qp->rq.head = 0;
5217 	hr_qp->rq.tail = 0;
5218 	hr_qp->sq.head = 0;
5219 	hr_qp->sq.tail = 0;
5220 	hr_qp->next_sge = 0;
5221 }
5222 
5223 static int hns_roce_v2_modify_qp(struct ib_qp *ibqp,
5224 				 const struct ib_qp_attr *attr,
5225 				 int attr_mask, enum ib_qp_state cur_state,
5226 				 enum ib_qp_state new_state)
5227 {
5228 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5229 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5230 	struct hns_roce_v2_qp_context ctx[2];
5231 	struct hns_roce_v2_qp_context *context = ctx;
5232 	struct hns_roce_v2_qp_context *qpc_mask = ctx + 1;
5233 	struct ib_device *ibdev = &hr_dev->ib_dev;
5234 	unsigned long sq_flag = 0;
5235 	unsigned long rq_flag = 0;
5236 	int ret;
5237 
5238 	if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
5239 		return -EOPNOTSUPP;
5240 
5241 	/*
5242 	 * In v2 engine, software pass context and context mask to hardware
5243 	 * when modifying qp. If software need modify some fields in context,
5244 	 * we should set all bits of the relevant fields in context mask to
5245 	 * 0 at the same time, else set them to 0x1.
5246 	 */
5247 	memset(context, 0, hr_dev->caps.qpc_sz);
5248 	memset(qpc_mask, 0xff, hr_dev->caps.qpc_sz);
5249 
5250 	ret = hns_roce_v2_set_abs_fields(ibqp, attr, attr_mask, cur_state,
5251 					 new_state, context, qpc_mask);
5252 	if (ret)
5253 		goto out;
5254 
5255 	/* When QP state is err, SQ and RQ WQE should be flushed */
5256 	if (new_state == IB_QPS_ERR) {
5257 		if (ibqp->qp_type != IB_QPT_XRC_TGT) {
5258 			spin_lock_irqsave(&hr_qp->sq.lock, sq_flag);
5259 			hr_qp->state = IB_QPS_ERR;
5260 			roce_set_field(context->byte_160_sq_ci_pi,
5261 				       V2_QPC_BYTE_160_SQ_PRODUCER_IDX_M,
5262 				       V2_QPC_BYTE_160_SQ_PRODUCER_IDX_S,
5263 				       hr_qp->sq.head);
5264 			roce_set_field(qpc_mask->byte_160_sq_ci_pi,
5265 				       V2_QPC_BYTE_160_SQ_PRODUCER_IDX_M,
5266 				       V2_QPC_BYTE_160_SQ_PRODUCER_IDX_S, 0);
5267 			spin_unlock_irqrestore(&hr_qp->sq.lock, sq_flag);
5268 		}
5269 
5270 		if (!ibqp->srq && ibqp->qp_type != IB_QPT_XRC_INI &&
5271 		    ibqp->qp_type != IB_QPT_XRC_TGT) {
5272 			spin_lock_irqsave(&hr_qp->rq.lock, rq_flag);
5273 			hr_qp->state = IB_QPS_ERR;
5274 			roce_set_field(context->byte_84_rq_ci_pi,
5275 			       V2_QPC_BYTE_84_RQ_PRODUCER_IDX_M,
5276 			       V2_QPC_BYTE_84_RQ_PRODUCER_IDX_S,
5277 			       hr_qp->rq.head);
5278 			roce_set_field(qpc_mask->byte_84_rq_ci_pi,
5279 			       V2_QPC_BYTE_84_RQ_PRODUCER_IDX_M,
5280 			       V2_QPC_BYTE_84_RQ_PRODUCER_IDX_S, 0);
5281 			spin_unlock_irqrestore(&hr_qp->rq.lock, rq_flag);
5282 		}
5283 	}
5284 
5285 	/* Configure the optional fields */
5286 	ret = hns_roce_v2_set_opt_fields(ibqp, attr, attr_mask, context,
5287 					 qpc_mask);
5288 	if (ret)
5289 		goto out;
5290 
5291 	roce_set_bit(context->byte_108_rx_reqepsn, V2_QPC_BYTE_108_INV_CREDIT_S,
5292 		     ((to_hr_qp_type(hr_qp->ibqp.qp_type) == SERV_TYPE_XRC) ||
5293 		     ibqp->srq) ? 1 : 0);
5294 	roce_set_bit(qpc_mask->byte_108_rx_reqepsn,
5295 		     V2_QPC_BYTE_108_INV_CREDIT_S, 0);
5296 
5297 	/* Every status migrate must change state */
5298 	roce_set_field(context->byte_60_qpst_tempid, V2_QPC_BYTE_60_QP_ST_M,
5299 		       V2_QPC_BYTE_60_QP_ST_S, new_state);
5300 	roce_set_field(qpc_mask->byte_60_qpst_tempid, V2_QPC_BYTE_60_QP_ST_M,
5301 		       V2_QPC_BYTE_60_QP_ST_S, 0);
5302 
5303 	/* SW pass context to HW */
5304 	ret = hns_roce_v2_qp_modify(hr_dev, context, qpc_mask, hr_qp);
5305 	if (ret) {
5306 		ibdev_err(ibdev, "failed to modify QP, ret = %d.\n", ret);
5307 		goto out;
5308 	}
5309 
5310 	hr_qp->state = new_state;
5311 
5312 	hns_roce_v2_record_opt_fields(ibqp, attr, attr_mask);
5313 
5314 	if (new_state == IB_QPS_RESET && !ibqp->uobject)
5315 		clear_qp(hr_qp);
5316 
5317 out:
5318 	return ret;
5319 }
5320 
5321 static int to_ib_qp_st(enum hns_roce_v2_qp_state state)
5322 {
5323 	static const enum ib_qp_state map[] = {
5324 		[HNS_ROCE_QP_ST_RST] = IB_QPS_RESET,
5325 		[HNS_ROCE_QP_ST_INIT] = IB_QPS_INIT,
5326 		[HNS_ROCE_QP_ST_RTR] = IB_QPS_RTR,
5327 		[HNS_ROCE_QP_ST_RTS] = IB_QPS_RTS,
5328 		[HNS_ROCE_QP_ST_SQD] = IB_QPS_SQD,
5329 		[HNS_ROCE_QP_ST_SQER] = IB_QPS_SQE,
5330 		[HNS_ROCE_QP_ST_ERR] = IB_QPS_ERR,
5331 		[HNS_ROCE_QP_ST_SQ_DRAINING] = IB_QPS_SQD
5332 	};
5333 
5334 	return (state < ARRAY_SIZE(map)) ? map[state] : -1;
5335 }
5336 
5337 static int hns_roce_v2_query_qpc(struct hns_roce_dev *hr_dev,
5338 				 struct hns_roce_qp *hr_qp,
5339 				 struct hns_roce_v2_qp_context *hr_context)
5340 {
5341 	struct hns_roce_cmd_mailbox *mailbox;
5342 	int ret;
5343 
5344 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5345 	if (IS_ERR(mailbox))
5346 		return PTR_ERR(mailbox);
5347 
5348 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, hr_qp->qpn, 0,
5349 				HNS_ROCE_CMD_QUERY_QPC,
5350 				HNS_ROCE_CMD_TIMEOUT_MSECS);
5351 	if (ret)
5352 		goto out;
5353 
5354 	memcpy(hr_context, mailbox->buf, hr_dev->caps.qpc_sz);
5355 
5356 out:
5357 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5358 	return ret;
5359 }
5360 
5361 static int hns_roce_v2_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr,
5362 				int qp_attr_mask,
5363 				struct ib_qp_init_attr *qp_init_attr)
5364 {
5365 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5366 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5367 	struct hns_roce_v2_qp_context context = {};
5368 	struct ib_device *ibdev = &hr_dev->ib_dev;
5369 	int tmp_qp_state;
5370 	int state;
5371 	int ret;
5372 
5373 	memset(qp_attr, 0, sizeof(*qp_attr));
5374 	memset(qp_init_attr, 0, sizeof(*qp_init_attr));
5375 
5376 	mutex_lock(&hr_qp->mutex);
5377 
5378 	if (hr_qp->state == IB_QPS_RESET) {
5379 		qp_attr->qp_state = IB_QPS_RESET;
5380 		ret = 0;
5381 		goto done;
5382 	}
5383 
5384 	ret = hns_roce_v2_query_qpc(hr_dev, hr_qp, &context);
5385 	if (ret) {
5386 		ibdev_err(ibdev, "failed to query QPC, ret = %d.\n", ret);
5387 		ret = -EINVAL;
5388 		goto out;
5389 	}
5390 
5391 	state = roce_get_field(context.byte_60_qpst_tempid,
5392 			       V2_QPC_BYTE_60_QP_ST_M, V2_QPC_BYTE_60_QP_ST_S);
5393 	tmp_qp_state = to_ib_qp_st((enum hns_roce_v2_qp_state)state);
5394 	if (tmp_qp_state == -1) {
5395 		ibdev_err(ibdev, "Illegal ib_qp_state\n");
5396 		ret = -EINVAL;
5397 		goto out;
5398 	}
5399 	hr_qp->state = (u8)tmp_qp_state;
5400 	qp_attr->qp_state = (enum ib_qp_state)hr_qp->state;
5401 	qp_attr->path_mtu = (enum ib_mtu)roce_get_field(context.byte_24_mtu_tc,
5402 							V2_QPC_BYTE_24_MTU_M,
5403 							V2_QPC_BYTE_24_MTU_S);
5404 	qp_attr->path_mig_state = IB_MIG_ARMED;
5405 	qp_attr->ah_attr.type   = RDMA_AH_ATTR_TYPE_ROCE;
5406 	if (hr_qp->ibqp.qp_type == IB_QPT_UD)
5407 		qp_attr->qkey = le32_to_cpu(context.qkey_xrcd);
5408 
5409 	qp_attr->rq_psn = roce_get_field(context.byte_108_rx_reqepsn,
5410 					 V2_QPC_BYTE_108_RX_REQ_EPSN_M,
5411 					 V2_QPC_BYTE_108_RX_REQ_EPSN_S);
5412 	qp_attr->sq_psn = (u32)roce_get_field(context.byte_172_sq_psn,
5413 					      V2_QPC_BYTE_172_SQ_CUR_PSN_M,
5414 					      V2_QPC_BYTE_172_SQ_CUR_PSN_S);
5415 	qp_attr->dest_qp_num = (u8)roce_get_field(context.byte_56_dqpn_err,
5416 						  V2_QPC_BYTE_56_DQPN_M,
5417 						  V2_QPC_BYTE_56_DQPN_S);
5418 	qp_attr->qp_access_flags = ((roce_get_bit(context.byte_76_srqn_op_en,
5419 				    V2_QPC_BYTE_76_RRE_S)) << V2_QP_RRE_S) |
5420 				    ((roce_get_bit(context.byte_76_srqn_op_en,
5421 				    V2_QPC_BYTE_76_RWE_S)) << V2_QP_RWE_S) |
5422 				    ((roce_get_bit(context.byte_76_srqn_op_en,
5423 				    V2_QPC_BYTE_76_ATE_S)) << V2_QP_ATE_S);
5424 
5425 	if (hr_qp->ibqp.qp_type == IB_QPT_RC ||
5426 	    hr_qp->ibqp.qp_type == IB_QPT_XRC_INI ||
5427 	    hr_qp->ibqp.qp_type == IB_QPT_XRC_TGT) {
5428 		struct ib_global_route *grh =
5429 				rdma_ah_retrieve_grh(&qp_attr->ah_attr);
5430 
5431 		rdma_ah_set_sl(&qp_attr->ah_attr,
5432 			       roce_get_field(context.byte_28_at_fl,
5433 					      V2_QPC_BYTE_28_SL_M,
5434 					      V2_QPC_BYTE_28_SL_S));
5435 		grh->flow_label = roce_get_field(context.byte_28_at_fl,
5436 						 V2_QPC_BYTE_28_FL_M,
5437 						 V2_QPC_BYTE_28_FL_S);
5438 		grh->sgid_index = roce_get_field(context.byte_20_smac_sgid_idx,
5439 						 V2_QPC_BYTE_20_SGID_IDX_M,
5440 						 V2_QPC_BYTE_20_SGID_IDX_S);
5441 		grh->hop_limit = roce_get_field(context.byte_24_mtu_tc,
5442 						V2_QPC_BYTE_24_HOP_LIMIT_M,
5443 						V2_QPC_BYTE_24_HOP_LIMIT_S);
5444 		grh->traffic_class = roce_get_field(context.byte_24_mtu_tc,
5445 						    V2_QPC_BYTE_24_TC_M,
5446 						    V2_QPC_BYTE_24_TC_S);
5447 
5448 		memcpy(grh->dgid.raw, context.dgid, sizeof(grh->dgid.raw));
5449 	}
5450 
5451 	qp_attr->port_num = hr_qp->port + 1;
5452 	qp_attr->sq_draining = 0;
5453 	qp_attr->max_rd_atomic = 1 << roce_get_field(context.byte_208_irrl,
5454 						     V2_QPC_BYTE_208_SR_MAX_M,
5455 						     V2_QPC_BYTE_208_SR_MAX_S);
5456 	qp_attr->max_dest_rd_atomic = 1 << roce_get_field(context.byte_140_raq,
5457 						     V2_QPC_BYTE_140_RR_MAX_M,
5458 						     V2_QPC_BYTE_140_RR_MAX_S);
5459 
5460 	qp_attr->min_rnr_timer = (u8)roce_get_field(context.byte_80_rnr_rx_cqn,
5461 						 V2_QPC_BYTE_80_MIN_RNR_TIME_M,
5462 						 V2_QPC_BYTE_80_MIN_RNR_TIME_S);
5463 	qp_attr->timeout = (u8)roce_get_field(context.byte_28_at_fl,
5464 					      V2_QPC_BYTE_28_AT_M,
5465 					      V2_QPC_BYTE_28_AT_S);
5466 	qp_attr->retry_cnt = roce_get_field(context.byte_212_lsn,
5467 					    V2_QPC_BYTE_212_RETRY_NUM_INIT_M,
5468 					    V2_QPC_BYTE_212_RETRY_NUM_INIT_S);
5469 	qp_attr->rnr_retry = roce_get_field(context.byte_244_rnr_rxack,
5470 					    V2_QPC_BYTE_244_RNR_NUM_INIT_M,
5471 					    V2_QPC_BYTE_244_RNR_NUM_INIT_S);
5472 
5473 done:
5474 	qp_attr->cur_qp_state = qp_attr->qp_state;
5475 	qp_attr->cap.max_recv_wr = hr_qp->rq.wqe_cnt;
5476 	qp_attr->cap.max_recv_sge = hr_qp->rq.max_gs - hr_qp->rq.rsv_sge;
5477 	qp_attr->cap.max_inline_data = hr_qp->max_inline_data;
5478 
5479 	if (!ibqp->uobject) {
5480 		qp_attr->cap.max_send_wr = hr_qp->sq.wqe_cnt;
5481 		qp_attr->cap.max_send_sge = hr_qp->sq.max_gs;
5482 	} else {
5483 		qp_attr->cap.max_send_wr = 0;
5484 		qp_attr->cap.max_send_sge = 0;
5485 	}
5486 
5487 	qp_init_attr->cap = qp_attr->cap;
5488 	qp_init_attr->sq_sig_type = hr_qp->sq_signal_bits;
5489 
5490 out:
5491 	mutex_unlock(&hr_qp->mutex);
5492 	return ret;
5493 }
5494 
5495 static inline int modify_qp_is_ok(struct hns_roce_qp *hr_qp)
5496 {
5497 	return ((hr_qp->ibqp.qp_type == IB_QPT_RC ||
5498 		 hr_qp->ibqp.qp_type == IB_QPT_UD ||
5499 		 hr_qp->ibqp.qp_type == IB_QPT_XRC_INI ||
5500 		 hr_qp->ibqp.qp_type == IB_QPT_XRC_TGT) &&
5501 		hr_qp->state != IB_QPS_RESET);
5502 }
5503 
5504 static int hns_roce_v2_destroy_qp_common(struct hns_roce_dev *hr_dev,
5505 					 struct hns_roce_qp *hr_qp,
5506 					 struct ib_udata *udata)
5507 {
5508 	struct ib_device *ibdev = &hr_dev->ib_dev;
5509 	struct hns_roce_cq *send_cq, *recv_cq;
5510 	unsigned long flags;
5511 	int ret = 0;
5512 
5513 	if (modify_qp_is_ok(hr_qp)) {
5514 		/* Modify qp to reset before destroying qp */
5515 		ret = hns_roce_v2_modify_qp(&hr_qp->ibqp, NULL, 0,
5516 					    hr_qp->state, IB_QPS_RESET);
5517 		if (ret)
5518 			ibdev_err(ibdev,
5519 				  "failed to modify QP to RST, ret = %d.\n",
5520 				  ret);
5521 	}
5522 
5523 	send_cq = hr_qp->ibqp.send_cq ? to_hr_cq(hr_qp->ibqp.send_cq) : NULL;
5524 	recv_cq = hr_qp->ibqp.recv_cq ? to_hr_cq(hr_qp->ibqp.recv_cq) : NULL;
5525 
5526 	spin_lock_irqsave(&hr_dev->qp_list_lock, flags);
5527 	hns_roce_lock_cqs(send_cq, recv_cq);
5528 
5529 	if (!udata) {
5530 		if (recv_cq)
5531 			__hns_roce_v2_cq_clean(recv_cq, hr_qp->qpn,
5532 					       (hr_qp->ibqp.srq ?
5533 						to_hr_srq(hr_qp->ibqp.srq) :
5534 						NULL));
5535 
5536 		if (send_cq && send_cq != recv_cq)
5537 			__hns_roce_v2_cq_clean(send_cq, hr_qp->qpn, NULL);
5538 
5539 	}
5540 
5541 	hns_roce_qp_remove(hr_dev, hr_qp);
5542 
5543 	hns_roce_unlock_cqs(send_cq, recv_cq);
5544 	spin_unlock_irqrestore(&hr_dev->qp_list_lock, flags);
5545 
5546 	return ret;
5547 }
5548 
5549 static int hns_roce_v2_destroy_qp(struct ib_qp *ibqp, struct ib_udata *udata)
5550 {
5551 	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
5552 	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
5553 	int ret;
5554 
5555 	ret = hns_roce_v2_destroy_qp_common(hr_dev, hr_qp, udata);
5556 	if (ret)
5557 		ibdev_err(&hr_dev->ib_dev,
5558 			  "failed to destroy QP, QPN = 0x%06lx, ret = %d.\n",
5559 			  hr_qp->qpn, ret);
5560 
5561 	hns_roce_qp_destroy(hr_dev, hr_qp, udata);
5562 
5563 	return 0;
5564 }
5565 
5566 static int hns_roce_v2_qp_flow_control_init(struct hns_roce_dev *hr_dev,
5567 					    struct hns_roce_qp *hr_qp)
5568 {
5569 	struct ib_device *ibdev = &hr_dev->ib_dev;
5570 	struct hns_roce_sccc_clr_done *resp;
5571 	struct hns_roce_sccc_clr *clr;
5572 	struct hns_roce_cmq_desc desc;
5573 	int ret, i;
5574 
5575 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
5576 		return 0;
5577 
5578 	mutex_lock(&hr_dev->qp_table.scc_mutex);
5579 
5580 	/* set scc ctx clear done flag */
5581 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_RESET_SCCC, false);
5582 	ret =  hns_roce_cmq_send(hr_dev, &desc, 1);
5583 	if (ret) {
5584 		ibdev_err(ibdev, "failed to reset SCC ctx, ret = %d.\n", ret);
5585 		goto out;
5586 	}
5587 
5588 	/* clear scc context */
5589 	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_CLR_SCCC, false);
5590 	clr = (struct hns_roce_sccc_clr *)desc.data;
5591 	clr->qpn = cpu_to_le32(hr_qp->qpn);
5592 	ret =  hns_roce_cmq_send(hr_dev, &desc, 1);
5593 	if (ret) {
5594 		ibdev_err(ibdev, "failed to clear SCC ctx, ret = %d.\n", ret);
5595 		goto out;
5596 	}
5597 
5598 	/* query scc context clear is done or not */
5599 	resp = (struct hns_roce_sccc_clr_done *)desc.data;
5600 	for (i = 0; i <= HNS_ROCE_CMQ_SCC_CLR_DONE_CNT; i++) {
5601 		hns_roce_cmq_setup_basic_desc(&desc,
5602 					      HNS_ROCE_OPC_QUERY_SCCC, true);
5603 		ret = hns_roce_cmq_send(hr_dev, &desc, 1);
5604 		if (ret) {
5605 			ibdev_err(ibdev, "failed to query clr cmq, ret = %d\n",
5606 				  ret);
5607 			goto out;
5608 		}
5609 
5610 		if (resp->clr_done)
5611 			goto out;
5612 
5613 		msleep(20);
5614 	}
5615 
5616 	ibdev_err(ibdev, "Query SCC clr done flag overtime.\n");
5617 	ret = -ETIMEDOUT;
5618 
5619 out:
5620 	mutex_unlock(&hr_dev->qp_table.scc_mutex);
5621 	return ret;
5622 }
5623 
5624 #define DMA_IDX_SHIFT 3
5625 #define DMA_WQE_SHIFT 3
5626 
5627 static int hns_roce_v2_write_srqc_index_queue(struct hns_roce_srq *srq,
5628 					      struct hns_roce_srq_context *ctx)
5629 {
5630 	struct hns_roce_idx_que *idx_que = &srq->idx_que;
5631 	struct ib_device *ibdev = srq->ibsrq.device;
5632 	struct hns_roce_dev *hr_dev = to_hr_dev(ibdev);
5633 	u64 mtts_idx[MTT_MIN_COUNT] = {};
5634 	dma_addr_t dma_handle_idx = 0;
5635 	int ret;
5636 
5637 	/* Get physical address of idx que buf */
5638 	ret = hns_roce_mtr_find(hr_dev, &idx_que->mtr, 0, mtts_idx,
5639 				ARRAY_SIZE(mtts_idx), &dma_handle_idx);
5640 	if (ret < 1) {
5641 		ibdev_err(ibdev, "failed to find mtr for SRQ idx, ret = %d.\n",
5642 			  ret);
5643 		return -ENOBUFS;
5644 	}
5645 
5646 	hr_reg_write(ctx, SRQC_IDX_HOP_NUM,
5647 		     to_hr_hem_hopnum(hr_dev->caps.idx_hop_num, srq->wqe_cnt));
5648 
5649 	hr_reg_write(ctx, SRQC_IDX_BT_BA_L, dma_handle_idx >> DMA_IDX_SHIFT);
5650 	hr_reg_write(ctx, SRQC_IDX_BT_BA_H,
5651 		     upper_32_bits(dma_handle_idx >> DMA_IDX_SHIFT));
5652 
5653 	hr_reg_write(ctx, SRQC_IDX_BA_PG_SZ,
5654 		     to_hr_hw_page_shift(idx_que->mtr.hem_cfg.ba_pg_shift));
5655 	hr_reg_write(ctx, SRQC_IDX_BUF_PG_SZ,
5656 		     to_hr_hw_page_shift(idx_que->mtr.hem_cfg.buf_pg_shift));
5657 
5658 	hr_reg_write(ctx, SRQC_IDX_CUR_BLK_ADDR_L,
5659 		     to_hr_hw_page_addr(mtts_idx[0]));
5660 	hr_reg_write(ctx, SRQC_IDX_CUR_BLK_ADDR_H,
5661 		     upper_32_bits(to_hr_hw_page_addr(mtts_idx[0])));
5662 
5663 	hr_reg_write(ctx, SRQC_IDX_NXT_BLK_ADDR_L,
5664 		     to_hr_hw_page_addr(mtts_idx[1]));
5665 	hr_reg_write(ctx, SRQC_IDX_NXT_BLK_ADDR_H,
5666 		     upper_32_bits(to_hr_hw_page_addr(mtts_idx[1])));
5667 
5668 	return 0;
5669 }
5670 
5671 static int hns_roce_v2_write_srqc(struct hns_roce_srq *srq, void *mb_buf)
5672 {
5673 	struct ib_device *ibdev = srq->ibsrq.device;
5674 	struct hns_roce_dev *hr_dev = to_hr_dev(ibdev);
5675 	struct hns_roce_srq_context *ctx = mb_buf;
5676 	u64 mtts_wqe[MTT_MIN_COUNT] = {};
5677 	dma_addr_t dma_handle_wqe = 0;
5678 	int ret;
5679 
5680 	memset(ctx, 0, sizeof(*ctx));
5681 
5682 	/* Get the physical address of srq buf */
5683 	ret = hns_roce_mtr_find(hr_dev, &srq->buf_mtr, 0, mtts_wqe,
5684 				ARRAY_SIZE(mtts_wqe), &dma_handle_wqe);
5685 	if (ret < 1) {
5686 		ibdev_err(ibdev, "failed to find mtr for SRQ WQE, ret = %d.\n",
5687 			  ret);
5688 		return -ENOBUFS;
5689 	}
5690 
5691 	hr_reg_write(ctx, SRQC_SRQ_ST, 1);
5692 	hr_reg_write(ctx, SRQC_SRQ_TYPE,
5693 		     !!(srq->ibsrq.srq_type == IB_SRQT_XRC));
5694 	hr_reg_write(ctx, SRQC_PD, to_hr_pd(srq->ibsrq.pd)->pdn);
5695 	hr_reg_write(ctx, SRQC_SRQN, srq->srqn);
5696 	hr_reg_write(ctx, SRQC_XRCD, srq->xrcdn);
5697 	hr_reg_write(ctx, SRQC_XRC_CQN, srq->cqn);
5698 	hr_reg_write(ctx, SRQC_SHIFT, ilog2(srq->wqe_cnt));
5699 	hr_reg_write(ctx, SRQC_RQWS,
5700 		     srq->max_gs <= 0 ? 0 : fls(srq->max_gs - 1));
5701 
5702 	hr_reg_write(ctx, SRQC_WQE_HOP_NUM,
5703 		     to_hr_hem_hopnum(hr_dev->caps.srqwqe_hop_num,
5704 				      srq->wqe_cnt));
5705 
5706 	hr_reg_write(ctx, SRQC_WQE_BT_BA_L, dma_handle_wqe >> DMA_WQE_SHIFT);
5707 	hr_reg_write(ctx, SRQC_WQE_BT_BA_H,
5708 		     upper_32_bits(dma_handle_wqe >> DMA_WQE_SHIFT));
5709 
5710 	hr_reg_write(ctx, SRQC_WQE_BA_PG_SZ,
5711 		     to_hr_hw_page_shift(srq->buf_mtr.hem_cfg.ba_pg_shift));
5712 	hr_reg_write(ctx, SRQC_WQE_BUF_PG_SZ,
5713 		     to_hr_hw_page_shift(srq->buf_mtr.hem_cfg.buf_pg_shift));
5714 
5715 	return hns_roce_v2_write_srqc_index_queue(srq, ctx);
5716 }
5717 
5718 static int hns_roce_v2_modify_srq(struct ib_srq *ibsrq,
5719 				  struct ib_srq_attr *srq_attr,
5720 				  enum ib_srq_attr_mask srq_attr_mask,
5721 				  struct ib_udata *udata)
5722 {
5723 	struct hns_roce_dev *hr_dev = to_hr_dev(ibsrq->device);
5724 	struct hns_roce_srq *srq = to_hr_srq(ibsrq);
5725 	struct hns_roce_srq_context *srq_context;
5726 	struct hns_roce_srq_context *srqc_mask;
5727 	struct hns_roce_cmd_mailbox *mailbox;
5728 	int ret;
5729 
5730 	/* Resizing SRQs is not supported yet */
5731 	if (srq_attr_mask & IB_SRQ_MAX_WR)
5732 		return -EINVAL;
5733 
5734 	if (srq_attr_mask & IB_SRQ_LIMIT) {
5735 		if (srq_attr->srq_limit > srq->wqe_cnt)
5736 			return -EINVAL;
5737 
5738 		mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5739 		if (IS_ERR(mailbox))
5740 			return PTR_ERR(mailbox);
5741 
5742 		srq_context = mailbox->buf;
5743 		srqc_mask = (struct hns_roce_srq_context *)mailbox->buf + 1;
5744 
5745 		memset(srqc_mask, 0xff, sizeof(*srqc_mask));
5746 
5747 		roce_set_field(srq_context->byte_8_limit_wl,
5748 			       SRQC_BYTE_8_SRQ_LIMIT_WL_M,
5749 			       SRQC_BYTE_8_SRQ_LIMIT_WL_S, srq_attr->srq_limit);
5750 		roce_set_field(srqc_mask->byte_8_limit_wl,
5751 			       SRQC_BYTE_8_SRQ_LIMIT_WL_M,
5752 			       SRQC_BYTE_8_SRQ_LIMIT_WL_S, 0);
5753 
5754 		ret = hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0, srq->srqn, 0,
5755 					HNS_ROCE_CMD_MODIFY_SRQC,
5756 					HNS_ROCE_CMD_TIMEOUT_MSECS);
5757 		hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5758 		if (ret) {
5759 			ibdev_err(&hr_dev->ib_dev,
5760 				  "failed to handle cmd of modifying SRQ, ret = %d.\n",
5761 				  ret);
5762 			return ret;
5763 		}
5764 	}
5765 
5766 	return 0;
5767 }
5768 
5769 static int hns_roce_v2_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr)
5770 {
5771 	struct hns_roce_dev *hr_dev = to_hr_dev(ibsrq->device);
5772 	struct hns_roce_srq *srq = to_hr_srq(ibsrq);
5773 	struct hns_roce_srq_context *srq_context;
5774 	struct hns_roce_cmd_mailbox *mailbox;
5775 	int limit_wl;
5776 	int ret;
5777 
5778 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5779 	if (IS_ERR(mailbox))
5780 		return PTR_ERR(mailbox);
5781 
5782 	srq_context = mailbox->buf;
5783 	ret = hns_roce_cmd_mbox(hr_dev, 0, mailbox->dma, srq->srqn, 0,
5784 				HNS_ROCE_CMD_QUERY_SRQC,
5785 				HNS_ROCE_CMD_TIMEOUT_MSECS);
5786 	if (ret) {
5787 		ibdev_err(&hr_dev->ib_dev,
5788 			  "failed to process cmd of querying SRQ, ret = %d.\n",
5789 			  ret);
5790 		goto out;
5791 	}
5792 
5793 	limit_wl = roce_get_field(srq_context->byte_8_limit_wl,
5794 				  SRQC_BYTE_8_SRQ_LIMIT_WL_M,
5795 				  SRQC_BYTE_8_SRQ_LIMIT_WL_S);
5796 
5797 	attr->srq_limit = limit_wl;
5798 	attr->max_wr = srq->wqe_cnt;
5799 	attr->max_sge = srq->max_gs - srq->rsv_sge;
5800 
5801 out:
5802 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5803 	return ret;
5804 }
5805 
5806 static int hns_roce_v2_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
5807 {
5808 	struct hns_roce_dev *hr_dev = to_hr_dev(cq->device);
5809 	struct hns_roce_v2_cq_context *cq_context;
5810 	struct hns_roce_cq *hr_cq = to_hr_cq(cq);
5811 	struct hns_roce_v2_cq_context *cqc_mask;
5812 	struct hns_roce_cmd_mailbox *mailbox;
5813 	int ret;
5814 
5815 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
5816 	if (IS_ERR(mailbox))
5817 		return PTR_ERR(mailbox);
5818 
5819 	cq_context = mailbox->buf;
5820 	cqc_mask = (struct hns_roce_v2_cq_context *)mailbox->buf + 1;
5821 
5822 	memset(cqc_mask, 0xff, sizeof(*cqc_mask));
5823 
5824 	roce_set_field(cq_context->byte_56_cqe_period_maxcnt,
5825 		       V2_CQC_BYTE_56_CQ_MAX_CNT_M, V2_CQC_BYTE_56_CQ_MAX_CNT_S,
5826 		       cq_count);
5827 	roce_set_field(cqc_mask->byte_56_cqe_period_maxcnt,
5828 		       V2_CQC_BYTE_56_CQ_MAX_CNT_M, V2_CQC_BYTE_56_CQ_MAX_CNT_S,
5829 		       0);
5830 	roce_set_field(cq_context->byte_56_cqe_period_maxcnt,
5831 		       V2_CQC_BYTE_56_CQ_PERIOD_M, V2_CQC_BYTE_56_CQ_PERIOD_S,
5832 		       cq_period);
5833 	roce_set_field(cqc_mask->byte_56_cqe_period_maxcnt,
5834 		       V2_CQC_BYTE_56_CQ_PERIOD_M, V2_CQC_BYTE_56_CQ_PERIOD_S,
5835 		       0);
5836 
5837 	ret = hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0, hr_cq->cqn, 1,
5838 				HNS_ROCE_CMD_MODIFY_CQC,
5839 				HNS_ROCE_CMD_TIMEOUT_MSECS);
5840 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
5841 	if (ret)
5842 		ibdev_err(&hr_dev->ib_dev,
5843 			  "failed to process cmd when modifying CQ, ret = %d.\n",
5844 			  ret);
5845 
5846 	return ret;
5847 }
5848 
5849 static void hns_roce_irq_work_handle(struct work_struct *work)
5850 {
5851 	struct hns_roce_work *irq_work =
5852 				container_of(work, struct hns_roce_work, work);
5853 	struct ib_device *ibdev = &irq_work->hr_dev->ib_dev;
5854 
5855 	switch (irq_work->event_type) {
5856 	case HNS_ROCE_EVENT_TYPE_PATH_MIG:
5857 		ibdev_info(ibdev, "Path migrated succeeded.\n");
5858 		break;
5859 	case HNS_ROCE_EVENT_TYPE_PATH_MIG_FAILED:
5860 		ibdev_warn(ibdev, "Path migration failed.\n");
5861 		break;
5862 	case HNS_ROCE_EVENT_TYPE_COMM_EST:
5863 		break;
5864 	case HNS_ROCE_EVENT_TYPE_SQ_DRAINED:
5865 		ibdev_warn(ibdev, "Send queue drained.\n");
5866 		break;
5867 	case HNS_ROCE_EVENT_TYPE_WQ_CATAS_ERROR:
5868 		ibdev_err(ibdev, "Local work queue 0x%x catast error, sub_event type is: %d\n",
5869 			  irq_work->queue_num, irq_work->sub_type);
5870 		break;
5871 	case HNS_ROCE_EVENT_TYPE_INV_REQ_LOCAL_WQ_ERROR:
5872 		ibdev_err(ibdev, "Invalid request local work queue 0x%x error.\n",
5873 			  irq_work->queue_num);
5874 		break;
5875 	case HNS_ROCE_EVENT_TYPE_LOCAL_WQ_ACCESS_ERROR:
5876 		ibdev_err(ibdev, "Local access violation work queue 0x%x error, sub_event type is: %d\n",
5877 			  irq_work->queue_num, irq_work->sub_type);
5878 		break;
5879 	case HNS_ROCE_EVENT_TYPE_SRQ_LIMIT_REACH:
5880 		ibdev_warn(ibdev, "SRQ limit reach.\n");
5881 		break;
5882 	case HNS_ROCE_EVENT_TYPE_SRQ_LAST_WQE_REACH:
5883 		ibdev_warn(ibdev, "SRQ last wqe reach.\n");
5884 		break;
5885 	case HNS_ROCE_EVENT_TYPE_SRQ_CATAS_ERROR:
5886 		ibdev_err(ibdev, "SRQ catas error.\n");
5887 		break;
5888 	case HNS_ROCE_EVENT_TYPE_CQ_ACCESS_ERROR:
5889 		ibdev_err(ibdev, "CQ 0x%x access err.\n", irq_work->queue_num);
5890 		break;
5891 	case HNS_ROCE_EVENT_TYPE_CQ_OVERFLOW:
5892 		ibdev_warn(ibdev, "CQ 0x%x overflow\n", irq_work->queue_num);
5893 		break;
5894 	case HNS_ROCE_EVENT_TYPE_DB_OVERFLOW:
5895 		ibdev_warn(ibdev, "DB overflow.\n");
5896 		break;
5897 	case HNS_ROCE_EVENT_TYPE_FLR:
5898 		ibdev_warn(ibdev, "Function level reset.\n");
5899 		break;
5900 	case HNS_ROCE_EVENT_TYPE_XRCD_VIOLATION:
5901 		ibdev_err(ibdev, "xrc domain violation error.\n");
5902 		break;
5903 	case HNS_ROCE_EVENT_TYPE_INVALID_XRCETH:
5904 		ibdev_err(ibdev, "invalid xrceth error.\n");
5905 		break;
5906 	default:
5907 		break;
5908 	}
5909 
5910 	kfree(irq_work);
5911 }
5912 
5913 static void hns_roce_v2_init_irq_work(struct hns_roce_dev *hr_dev,
5914 				      struct hns_roce_eq *eq, u32 queue_num)
5915 {
5916 	struct hns_roce_work *irq_work;
5917 
5918 	irq_work = kzalloc(sizeof(struct hns_roce_work), GFP_ATOMIC);
5919 	if (!irq_work)
5920 		return;
5921 
5922 	INIT_WORK(&(irq_work->work), hns_roce_irq_work_handle);
5923 	irq_work->hr_dev = hr_dev;
5924 	irq_work->event_type = eq->event_type;
5925 	irq_work->sub_type = eq->sub_type;
5926 	irq_work->queue_num = queue_num;
5927 	queue_work(hr_dev->irq_workq, &(irq_work->work));
5928 }
5929 
5930 static void update_eq_db(struct hns_roce_eq *eq)
5931 {
5932 	struct hns_roce_dev *hr_dev = eq->hr_dev;
5933 	struct hns_roce_v2_db eq_db = {};
5934 
5935 	if (eq->type_flag == HNS_ROCE_AEQ) {
5936 		roce_set_field(eq_db.byte_4, V2_EQ_DB_CMD_M, V2_EQ_DB_CMD_S,
5937 			       eq->arm_st == HNS_ROCE_V2_EQ_ALWAYS_ARMED ?
5938 			       HNS_ROCE_EQ_DB_CMD_AEQ :
5939 			       HNS_ROCE_EQ_DB_CMD_AEQ_ARMED);
5940 	} else {
5941 		roce_set_field(eq_db.byte_4, V2_EQ_DB_TAG_M, V2_EQ_DB_TAG_S,
5942 			       eq->eqn);
5943 
5944 		roce_set_field(eq_db.byte_4, V2_EQ_DB_CMD_M, V2_EQ_DB_CMD_S,
5945 			       eq->arm_st == HNS_ROCE_V2_EQ_ALWAYS_ARMED ?
5946 			       HNS_ROCE_EQ_DB_CMD_CEQ :
5947 			       HNS_ROCE_EQ_DB_CMD_CEQ_ARMED);
5948 	}
5949 
5950 	roce_set_field(eq_db.parameter, V2_EQ_DB_CONS_IDX_M,
5951 		       V2_EQ_DB_CONS_IDX_S, eq->cons_index);
5952 
5953 	hns_roce_write64(hr_dev, (__le32 *)&eq_db, eq->db_reg);
5954 }
5955 
5956 static struct hns_roce_aeqe *next_aeqe_sw_v2(struct hns_roce_eq *eq)
5957 {
5958 	struct hns_roce_aeqe *aeqe;
5959 
5960 	aeqe = hns_roce_buf_offset(eq->mtr.kmem,
5961 				   (eq->cons_index & (eq->entries - 1)) *
5962 				   eq->eqe_size);
5963 
5964 	return (roce_get_bit(aeqe->asyn, HNS_ROCE_V2_AEQ_AEQE_OWNER_S) ^
5965 		!!(eq->cons_index & eq->entries)) ? aeqe : NULL;
5966 }
5967 
5968 static int hns_roce_v2_aeq_int(struct hns_roce_dev *hr_dev,
5969 			       struct hns_roce_eq *eq)
5970 {
5971 	struct device *dev = hr_dev->dev;
5972 	struct hns_roce_aeqe *aeqe = next_aeqe_sw_v2(eq);
5973 	int aeqe_found = 0;
5974 	int event_type;
5975 	u32 queue_num;
5976 	int sub_type;
5977 
5978 	while (aeqe) {
5979 		/* Make sure we read AEQ entry after we have checked the
5980 		 * ownership bit
5981 		 */
5982 		dma_rmb();
5983 
5984 		event_type = roce_get_field(aeqe->asyn,
5985 					    HNS_ROCE_V2_AEQE_EVENT_TYPE_M,
5986 					    HNS_ROCE_V2_AEQE_EVENT_TYPE_S);
5987 		sub_type = roce_get_field(aeqe->asyn,
5988 					  HNS_ROCE_V2_AEQE_SUB_TYPE_M,
5989 					  HNS_ROCE_V2_AEQE_SUB_TYPE_S);
5990 		queue_num = roce_get_field(aeqe->event.queue_event.num,
5991 					   HNS_ROCE_V2_AEQE_EVENT_QUEUE_NUM_M,
5992 					   HNS_ROCE_V2_AEQE_EVENT_QUEUE_NUM_S);
5993 
5994 		switch (event_type) {
5995 		case HNS_ROCE_EVENT_TYPE_PATH_MIG:
5996 		case HNS_ROCE_EVENT_TYPE_PATH_MIG_FAILED:
5997 		case HNS_ROCE_EVENT_TYPE_COMM_EST:
5998 		case HNS_ROCE_EVENT_TYPE_SQ_DRAINED:
5999 		case HNS_ROCE_EVENT_TYPE_WQ_CATAS_ERROR:
6000 		case HNS_ROCE_EVENT_TYPE_SRQ_LAST_WQE_REACH:
6001 		case HNS_ROCE_EVENT_TYPE_INV_REQ_LOCAL_WQ_ERROR:
6002 		case HNS_ROCE_EVENT_TYPE_LOCAL_WQ_ACCESS_ERROR:
6003 		case HNS_ROCE_EVENT_TYPE_XRCD_VIOLATION:
6004 		case HNS_ROCE_EVENT_TYPE_INVALID_XRCETH:
6005 			hns_roce_qp_event(hr_dev, queue_num, event_type);
6006 			break;
6007 		case HNS_ROCE_EVENT_TYPE_SRQ_LIMIT_REACH:
6008 		case HNS_ROCE_EVENT_TYPE_SRQ_CATAS_ERROR:
6009 			hns_roce_srq_event(hr_dev, queue_num, event_type);
6010 			break;
6011 		case HNS_ROCE_EVENT_TYPE_CQ_ACCESS_ERROR:
6012 		case HNS_ROCE_EVENT_TYPE_CQ_OVERFLOW:
6013 			hns_roce_cq_event(hr_dev, queue_num, event_type);
6014 			break;
6015 		case HNS_ROCE_EVENT_TYPE_MB:
6016 			hns_roce_cmd_event(hr_dev,
6017 					le16_to_cpu(aeqe->event.cmd.token),
6018 					aeqe->event.cmd.status,
6019 					le64_to_cpu(aeqe->event.cmd.out_param));
6020 			break;
6021 		case HNS_ROCE_EVENT_TYPE_DB_OVERFLOW:
6022 		case HNS_ROCE_EVENT_TYPE_FLR:
6023 			break;
6024 		default:
6025 			dev_err(dev, "Unhandled event %d on EQ %d at idx %u.\n",
6026 				event_type, eq->eqn, eq->cons_index);
6027 			break;
6028 		}
6029 
6030 		eq->event_type = event_type;
6031 		eq->sub_type = sub_type;
6032 		++eq->cons_index;
6033 		aeqe_found = 1;
6034 
6035 		hns_roce_v2_init_irq_work(hr_dev, eq, queue_num);
6036 
6037 		aeqe = next_aeqe_sw_v2(eq);
6038 	}
6039 
6040 	update_eq_db(eq);
6041 	return aeqe_found;
6042 }
6043 
6044 static struct hns_roce_ceqe *next_ceqe_sw_v2(struct hns_roce_eq *eq)
6045 {
6046 	struct hns_roce_ceqe *ceqe;
6047 
6048 	ceqe = hns_roce_buf_offset(eq->mtr.kmem,
6049 				   (eq->cons_index & (eq->entries - 1)) *
6050 				   eq->eqe_size);
6051 
6052 	return (!!(roce_get_bit(ceqe->comp, HNS_ROCE_V2_CEQ_CEQE_OWNER_S))) ^
6053 		(!!(eq->cons_index & eq->entries)) ? ceqe : NULL;
6054 }
6055 
6056 static int hns_roce_v2_ceq_int(struct hns_roce_dev *hr_dev,
6057 			       struct hns_roce_eq *eq)
6058 {
6059 	struct hns_roce_ceqe *ceqe = next_ceqe_sw_v2(eq);
6060 	int ceqe_found = 0;
6061 	u32 cqn;
6062 
6063 	while (ceqe) {
6064 		/* Make sure we read CEQ entry after we have checked the
6065 		 * ownership bit
6066 		 */
6067 		dma_rmb();
6068 
6069 		cqn = roce_get_field(ceqe->comp, HNS_ROCE_V2_CEQE_COMP_CQN_M,
6070 				     HNS_ROCE_V2_CEQE_COMP_CQN_S);
6071 
6072 		hns_roce_cq_completion(hr_dev, cqn);
6073 
6074 		++eq->cons_index;
6075 		ceqe_found = 1;
6076 
6077 		ceqe = next_ceqe_sw_v2(eq);
6078 	}
6079 
6080 	update_eq_db(eq);
6081 
6082 	return ceqe_found;
6083 }
6084 
6085 static irqreturn_t hns_roce_v2_msix_interrupt_eq(int irq, void *eq_ptr)
6086 {
6087 	struct hns_roce_eq *eq = eq_ptr;
6088 	struct hns_roce_dev *hr_dev = eq->hr_dev;
6089 	int int_work;
6090 
6091 	if (eq->type_flag == HNS_ROCE_CEQ)
6092 		/* Completion event interrupt */
6093 		int_work = hns_roce_v2_ceq_int(hr_dev, eq);
6094 	else
6095 		/* Asychronous event interrupt */
6096 		int_work = hns_roce_v2_aeq_int(hr_dev, eq);
6097 
6098 	return IRQ_RETVAL(int_work);
6099 }
6100 
6101 static irqreturn_t hns_roce_v2_msix_interrupt_abn(int irq, void *dev_id)
6102 {
6103 	struct hns_roce_dev *hr_dev = dev_id;
6104 	struct device *dev = hr_dev->dev;
6105 	int int_work = 0;
6106 	u32 int_st;
6107 	u32 int_en;
6108 
6109 	/* Abnormal interrupt */
6110 	int_st = roce_read(hr_dev, ROCEE_VF_ABN_INT_ST_REG);
6111 	int_en = roce_read(hr_dev, ROCEE_VF_ABN_INT_EN_REG);
6112 
6113 	if (int_st & BIT(HNS_ROCE_V2_VF_INT_ST_AEQ_OVERFLOW_S)) {
6114 		struct pci_dev *pdev = hr_dev->pci_dev;
6115 		struct hnae3_ae_dev *ae_dev = pci_get_drvdata(pdev);
6116 		const struct hnae3_ae_ops *ops = ae_dev->ops;
6117 
6118 		dev_err(dev, "AEQ overflow!\n");
6119 
6120 		int_st |= 1 << HNS_ROCE_V2_VF_INT_ST_AEQ_OVERFLOW_S;
6121 		roce_write(hr_dev, ROCEE_VF_ABN_INT_ST_REG, int_st);
6122 
6123 		/* Set reset level for reset_event() */
6124 		if (ops->set_default_reset_request)
6125 			ops->set_default_reset_request(ae_dev,
6126 						       HNAE3_FUNC_RESET);
6127 		if (ops->reset_event)
6128 			ops->reset_event(pdev, NULL);
6129 
6130 		int_en |= 1 << HNS_ROCE_V2_VF_ABN_INT_EN_S;
6131 		roce_write(hr_dev, ROCEE_VF_ABN_INT_EN_REG, int_en);
6132 
6133 		int_work = 1;
6134 	} else if (int_st & BIT(HNS_ROCE_V2_VF_INT_ST_RAS_INT_S)) {
6135 		dev_err(dev, "RAS interrupt!\n");
6136 
6137 		int_st |= 1 << HNS_ROCE_V2_VF_INT_ST_RAS_INT_S;
6138 		roce_write(hr_dev, ROCEE_VF_ABN_INT_ST_REG, int_st);
6139 
6140 		int_en |= 1 << HNS_ROCE_V2_VF_ABN_INT_EN_S;
6141 		roce_write(hr_dev, ROCEE_VF_ABN_INT_EN_REG, int_en);
6142 
6143 		int_work = 1;
6144 	} else {
6145 		dev_err(dev, "There is no abnormal irq found!\n");
6146 	}
6147 
6148 	return IRQ_RETVAL(int_work);
6149 }
6150 
6151 static void hns_roce_v2_int_mask_enable(struct hns_roce_dev *hr_dev,
6152 					int eq_num, u32 enable_flag)
6153 {
6154 	int i;
6155 
6156 	for (i = 0; i < eq_num; i++)
6157 		roce_write(hr_dev, ROCEE_VF_EVENT_INT_EN_REG +
6158 			   i * EQ_REG_OFFSET, enable_flag);
6159 
6160 	roce_write(hr_dev, ROCEE_VF_ABN_INT_EN_REG, enable_flag);
6161 	roce_write(hr_dev, ROCEE_VF_ABN_INT_CFG_REG, enable_flag);
6162 }
6163 
6164 static void hns_roce_v2_destroy_eqc(struct hns_roce_dev *hr_dev, int eqn)
6165 {
6166 	struct device *dev = hr_dev->dev;
6167 	int ret;
6168 
6169 	if (eqn < hr_dev->caps.num_comp_vectors)
6170 		ret = hns_roce_cmd_mbox(hr_dev, 0, 0, eqn & HNS_ROCE_V2_EQN_M,
6171 					0, HNS_ROCE_CMD_DESTROY_CEQC,
6172 					HNS_ROCE_CMD_TIMEOUT_MSECS);
6173 	else
6174 		ret = hns_roce_cmd_mbox(hr_dev, 0, 0, eqn & HNS_ROCE_V2_EQN_M,
6175 					0, HNS_ROCE_CMD_DESTROY_AEQC,
6176 					HNS_ROCE_CMD_TIMEOUT_MSECS);
6177 	if (ret)
6178 		dev_err(dev, "[mailbox cmd] destroy eqc(%d) failed.\n", eqn);
6179 }
6180 
6181 static void free_eq_buf(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq)
6182 {
6183 	hns_roce_mtr_destroy(hr_dev, &eq->mtr);
6184 }
6185 
6186 static void init_eq_config(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq)
6187 {
6188 	eq->db_reg = hr_dev->reg_base + ROCEE_VF_EQ_DB_CFG0_REG;
6189 	eq->cons_index = 0;
6190 	eq->over_ignore = HNS_ROCE_V2_EQ_OVER_IGNORE_0;
6191 	eq->coalesce = HNS_ROCE_V2_EQ_COALESCE_0;
6192 	eq->arm_st = HNS_ROCE_V2_EQ_ALWAYS_ARMED;
6193 	eq->shift = ilog2((unsigned int)eq->entries);
6194 }
6195 
6196 static int config_eqc(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq,
6197 		      void *mb_buf)
6198 {
6199 	u64 eqe_ba[MTT_MIN_COUNT] = { 0 };
6200 	struct hns_roce_eq_context *eqc;
6201 	u64 bt_ba = 0;
6202 	int count;
6203 
6204 	eqc = mb_buf;
6205 	memset(eqc, 0, sizeof(struct hns_roce_eq_context));
6206 
6207 	init_eq_config(hr_dev, eq);
6208 
6209 	/* if not multi-hop, eqe buffer only use one trunk */
6210 	count = hns_roce_mtr_find(hr_dev, &eq->mtr, 0, eqe_ba, MTT_MIN_COUNT,
6211 				  &bt_ba);
6212 	if (count < 1) {
6213 		dev_err(hr_dev->dev, "failed to find EQE mtr\n");
6214 		return -ENOBUFS;
6215 	}
6216 
6217 	hr_reg_write(eqc, EQC_EQ_ST, HNS_ROCE_V2_EQ_STATE_VALID);
6218 	hr_reg_write(eqc, EQC_EQE_HOP_NUM, eq->hop_num);
6219 	hr_reg_write(eqc, EQC_OVER_IGNORE, eq->over_ignore);
6220 	hr_reg_write(eqc, EQC_COALESCE, eq->coalesce);
6221 	hr_reg_write(eqc, EQC_ARM_ST, eq->arm_st);
6222 	hr_reg_write(eqc, EQC_EQN, eq->eqn);
6223 	hr_reg_write(eqc, EQC_EQE_CNT, HNS_ROCE_EQ_INIT_EQE_CNT);
6224 	hr_reg_write(eqc, EQC_EQE_BA_PG_SZ,
6225 		     to_hr_hw_page_shift(eq->mtr.hem_cfg.ba_pg_shift));
6226 	hr_reg_write(eqc, EQC_EQE_BUF_PG_SZ,
6227 		     to_hr_hw_page_shift(eq->mtr.hem_cfg.buf_pg_shift));
6228 	hr_reg_write(eqc, EQC_EQ_PROD_INDX, HNS_ROCE_EQ_INIT_PROD_IDX);
6229 	hr_reg_write(eqc, EQC_EQ_MAX_CNT, eq->eq_max_cnt);
6230 
6231 	hr_reg_write(eqc, EQC_EQ_PERIOD, eq->eq_period);
6232 	hr_reg_write(eqc, EQC_EQE_REPORT_TIMER, HNS_ROCE_EQ_INIT_REPORT_TIMER);
6233 	hr_reg_write(eqc, EQC_EQE_BA_L, bt_ba >> 3);
6234 	hr_reg_write(eqc, EQC_EQE_BA_H, bt_ba >> 35);
6235 	hr_reg_write(eqc, EQC_SHIFT, eq->shift);
6236 	hr_reg_write(eqc, EQC_MSI_INDX, HNS_ROCE_EQ_INIT_MSI_IDX);
6237 	hr_reg_write(eqc, EQC_CUR_EQE_BA_L, eqe_ba[0] >> 12);
6238 	hr_reg_write(eqc, EQC_CUR_EQE_BA_M, eqe_ba[0] >> 28);
6239 	hr_reg_write(eqc, EQC_CUR_EQE_BA_H, eqe_ba[0] >> 60);
6240 	hr_reg_write(eqc, EQC_EQ_CONS_INDX, HNS_ROCE_EQ_INIT_CONS_IDX);
6241 	hr_reg_write(eqc, EQC_NEX_EQE_BA_L, eqe_ba[1] >> 12);
6242 	hr_reg_write(eqc, EQC_NEX_EQE_BA_H, eqe_ba[1] >> 44);
6243 	hr_reg_write(eqc, EQC_EQE_SIZE,
6244 		     !!(eq->eqe_size == HNS_ROCE_V3_EQE_SIZE));
6245 
6246 	return 0;
6247 }
6248 
6249 static int alloc_eq_buf(struct hns_roce_dev *hr_dev, struct hns_roce_eq *eq)
6250 {
6251 	struct hns_roce_buf_attr buf_attr = {};
6252 	int err;
6253 
6254 	if (hr_dev->caps.eqe_hop_num == HNS_ROCE_HOP_NUM_0)
6255 		eq->hop_num = 0;
6256 	else
6257 		eq->hop_num = hr_dev->caps.eqe_hop_num;
6258 
6259 	buf_attr.page_shift = hr_dev->caps.eqe_buf_pg_sz + HNS_HW_PAGE_SHIFT;
6260 	buf_attr.region[0].size = eq->entries * eq->eqe_size;
6261 	buf_attr.region[0].hopnum = eq->hop_num;
6262 	buf_attr.region_count = 1;
6263 
6264 	err = hns_roce_mtr_create(hr_dev, &eq->mtr, &buf_attr,
6265 				  hr_dev->caps.eqe_ba_pg_sz +
6266 				  HNS_HW_PAGE_SHIFT, NULL, 0);
6267 	if (err)
6268 		dev_err(hr_dev->dev, "Failed to alloc EQE mtr, err %d\n", err);
6269 
6270 	return err;
6271 }
6272 
6273 static int hns_roce_v2_create_eq(struct hns_roce_dev *hr_dev,
6274 				 struct hns_roce_eq *eq,
6275 				 unsigned int eq_cmd)
6276 {
6277 	struct hns_roce_cmd_mailbox *mailbox;
6278 	int ret;
6279 
6280 	/* Allocate mailbox memory */
6281 	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
6282 	if (IS_ERR_OR_NULL(mailbox))
6283 		return -ENOMEM;
6284 
6285 	ret = alloc_eq_buf(hr_dev, eq);
6286 	if (ret)
6287 		goto free_cmd_mbox;
6288 
6289 	ret = config_eqc(hr_dev, eq, mailbox->buf);
6290 	if (ret)
6291 		goto err_cmd_mbox;
6292 
6293 	ret = hns_roce_cmd_mbox(hr_dev, mailbox->dma, 0, eq->eqn, 0,
6294 				eq_cmd, HNS_ROCE_CMD_TIMEOUT_MSECS);
6295 	if (ret) {
6296 		dev_err(hr_dev->dev, "[mailbox cmd] create eqc failed.\n");
6297 		goto err_cmd_mbox;
6298 	}
6299 
6300 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
6301 
6302 	return 0;
6303 
6304 err_cmd_mbox:
6305 	free_eq_buf(hr_dev, eq);
6306 
6307 free_cmd_mbox:
6308 	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
6309 
6310 	return ret;
6311 }
6312 
6313 static int __hns_roce_request_irq(struct hns_roce_dev *hr_dev, int irq_num,
6314 				  int comp_num, int aeq_num, int other_num)
6315 {
6316 	struct hns_roce_eq_table *eq_table = &hr_dev->eq_table;
6317 	int i, j;
6318 	int ret;
6319 
6320 	for (i = 0; i < irq_num; i++) {
6321 		hr_dev->irq_names[i] = kzalloc(HNS_ROCE_INT_NAME_LEN,
6322 					       GFP_KERNEL);
6323 		if (!hr_dev->irq_names[i]) {
6324 			ret = -ENOMEM;
6325 			goto err_kzalloc_failed;
6326 		}
6327 	}
6328 
6329 	/* irq contains: abnormal + AEQ + CEQ */
6330 	for (j = 0; j < other_num; j++)
6331 		snprintf((char *)hr_dev->irq_names[j], HNS_ROCE_INT_NAME_LEN,
6332 			 "hns-abn-%d", j);
6333 
6334 	for (j = other_num; j < (other_num + aeq_num); j++)
6335 		snprintf((char *)hr_dev->irq_names[j], HNS_ROCE_INT_NAME_LEN,
6336 			 "hns-aeq-%d", j - other_num);
6337 
6338 	for (j = (other_num + aeq_num); j < irq_num; j++)
6339 		snprintf((char *)hr_dev->irq_names[j], HNS_ROCE_INT_NAME_LEN,
6340 			 "hns-ceq-%d", j - other_num - aeq_num);
6341 
6342 	for (j = 0; j < irq_num; j++) {
6343 		if (j < other_num)
6344 			ret = request_irq(hr_dev->irq[j],
6345 					  hns_roce_v2_msix_interrupt_abn,
6346 					  0, hr_dev->irq_names[j], hr_dev);
6347 
6348 		else if (j < (other_num + comp_num))
6349 			ret = request_irq(eq_table->eq[j - other_num].irq,
6350 					  hns_roce_v2_msix_interrupt_eq,
6351 					  0, hr_dev->irq_names[j + aeq_num],
6352 					  &eq_table->eq[j - other_num]);
6353 		else
6354 			ret = request_irq(eq_table->eq[j - other_num].irq,
6355 					  hns_roce_v2_msix_interrupt_eq,
6356 					  0, hr_dev->irq_names[j - comp_num],
6357 					  &eq_table->eq[j - other_num]);
6358 		if (ret) {
6359 			dev_err(hr_dev->dev, "Request irq error!\n");
6360 			goto err_request_failed;
6361 		}
6362 	}
6363 
6364 	return 0;
6365 
6366 err_request_failed:
6367 	for (j -= 1; j >= 0; j--)
6368 		if (j < other_num)
6369 			free_irq(hr_dev->irq[j], hr_dev);
6370 		else
6371 			free_irq(eq_table->eq[j - other_num].irq,
6372 				 &eq_table->eq[j - other_num]);
6373 
6374 err_kzalloc_failed:
6375 	for (i -= 1; i >= 0; i--)
6376 		kfree(hr_dev->irq_names[i]);
6377 
6378 	return ret;
6379 }
6380 
6381 static void __hns_roce_free_irq(struct hns_roce_dev *hr_dev)
6382 {
6383 	int irq_num;
6384 	int eq_num;
6385 	int i;
6386 
6387 	eq_num = hr_dev->caps.num_comp_vectors + hr_dev->caps.num_aeq_vectors;
6388 	irq_num = eq_num + hr_dev->caps.num_other_vectors;
6389 
6390 	for (i = 0; i < hr_dev->caps.num_other_vectors; i++)
6391 		free_irq(hr_dev->irq[i], hr_dev);
6392 
6393 	for (i = 0; i < eq_num; i++)
6394 		free_irq(hr_dev->eq_table.eq[i].irq, &hr_dev->eq_table.eq[i]);
6395 
6396 	for (i = 0; i < irq_num; i++)
6397 		kfree(hr_dev->irq_names[i]);
6398 }
6399 
6400 static int hns_roce_v2_init_eq_table(struct hns_roce_dev *hr_dev)
6401 {
6402 	struct hns_roce_eq_table *eq_table = &hr_dev->eq_table;
6403 	struct device *dev = hr_dev->dev;
6404 	struct hns_roce_eq *eq;
6405 	unsigned int eq_cmd;
6406 	int irq_num;
6407 	int eq_num;
6408 	int other_num;
6409 	int comp_num;
6410 	int aeq_num;
6411 	int i;
6412 	int ret;
6413 
6414 	other_num = hr_dev->caps.num_other_vectors;
6415 	comp_num = hr_dev->caps.num_comp_vectors;
6416 	aeq_num = hr_dev->caps.num_aeq_vectors;
6417 
6418 	eq_num = comp_num + aeq_num;
6419 	irq_num = eq_num + other_num;
6420 
6421 	eq_table->eq = kcalloc(eq_num, sizeof(*eq_table->eq), GFP_KERNEL);
6422 	if (!eq_table->eq)
6423 		return -ENOMEM;
6424 
6425 	/* create eq */
6426 	for (i = 0; i < eq_num; i++) {
6427 		eq = &eq_table->eq[i];
6428 		eq->hr_dev = hr_dev;
6429 		eq->eqn = i;
6430 		if (i < comp_num) {
6431 			/* CEQ */
6432 			eq_cmd = HNS_ROCE_CMD_CREATE_CEQC;
6433 			eq->type_flag = HNS_ROCE_CEQ;
6434 			eq->entries = hr_dev->caps.ceqe_depth;
6435 			eq->eqe_size = hr_dev->caps.ceqe_size;
6436 			eq->irq = hr_dev->irq[i + other_num + aeq_num];
6437 			eq->eq_max_cnt = HNS_ROCE_CEQ_DEFAULT_BURST_NUM;
6438 			eq->eq_period = HNS_ROCE_CEQ_DEFAULT_INTERVAL;
6439 		} else {
6440 			/* AEQ */
6441 			eq_cmd = HNS_ROCE_CMD_CREATE_AEQC;
6442 			eq->type_flag = HNS_ROCE_AEQ;
6443 			eq->entries = hr_dev->caps.aeqe_depth;
6444 			eq->eqe_size = hr_dev->caps.aeqe_size;
6445 			eq->irq = hr_dev->irq[i - comp_num + other_num];
6446 			eq->eq_max_cnt = HNS_ROCE_AEQ_DEFAULT_BURST_NUM;
6447 			eq->eq_period = HNS_ROCE_AEQ_DEFAULT_INTERVAL;
6448 		}
6449 
6450 		ret = hns_roce_v2_create_eq(hr_dev, eq, eq_cmd);
6451 		if (ret) {
6452 			dev_err(dev, "eq create failed.\n");
6453 			goto err_create_eq_fail;
6454 		}
6455 	}
6456 
6457 	/* enable irq */
6458 	hns_roce_v2_int_mask_enable(hr_dev, eq_num, EQ_ENABLE);
6459 
6460 	ret = __hns_roce_request_irq(hr_dev, irq_num, comp_num,
6461 				     aeq_num, other_num);
6462 	if (ret) {
6463 		dev_err(dev, "Request irq failed.\n");
6464 		goto err_request_irq_fail;
6465 	}
6466 
6467 	hr_dev->irq_workq = alloc_ordered_workqueue("hns_roce_irq_workq", 0);
6468 	if (!hr_dev->irq_workq) {
6469 		dev_err(dev, "Create irq workqueue failed!\n");
6470 		ret = -ENOMEM;
6471 		goto err_create_wq_fail;
6472 	}
6473 
6474 	return 0;
6475 
6476 err_create_wq_fail:
6477 	__hns_roce_free_irq(hr_dev);
6478 
6479 err_request_irq_fail:
6480 	hns_roce_v2_int_mask_enable(hr_dev, eq_num, EQ_DISABLE);
6481 
6482 err_create_eq_fail:
6483 	for (i -= 1; i >= 0; i--)
6484 		free_eq_buf(hr_dev, &eq_table->eq[i]);
6485 	kfree(eq_table->eq);
6486 
6487 	return ret;
6488 }
6489 
6490 static void hns_roce_v2_cleanup_eq_table(struct hns_roce_dev *hr_dev)
6491 {
6492 	struct hns_roce_eq_table *eq_table = &hr_dev->eq_table;
6493 	int eq_num;
6494 	int i;
6495 
6496 	eq_num = hr_dev->caps.num_comp_vectors + hr_dev->caps.num_aeq_vectors;
6497 
6498 	/* Disable irq */
6499 	hns_roce_v2_int_mask_enable(hr_dev, eq_num, EQ_DISABLE);
6500 
6501 	__hns_roce_free_irq(hr_dev);
6502 	destroy_workqueue(hr_dev->irq_workq);
6503 
6504 	for (i = 0; i < eq_num; i++) {
6505 		hns_roce_v2_destroy_eqc(hr_dev, i);
6506 
6507 		free_eq_buf(hr_dev, &eq_table->eq[i]);
6508 	}
6509 
6510 	kfree(eq_table->eq);
6511 }
6512 
6513 static const struct hns_roce_dfx_hw hns_roce_dfx_hw_v2 = {
6514 	.query_cqc_info = hns_roce_v2_query_cqc_info,
6515 };
6516 
6517 static const struct ib_device_ops hns_roce_v2_dev_ops = {
6518 	.destroy_qp = hns_roce_v2_destroy_qp,
6519 	.modify_cq = hns_roce_v2_modify_cq,
6520 	.poll_cq = hns_roce_v2_poll_cq,
6521 	.post_recv = hns_roce_v2_post_recv,
6522 	.post_send = hns_roce_v2_post_send,
6523 	.query_qp = hns_roce_v2_query_qp,
6524 	.req_notify_cq = hns_roce_v2_req_notify_cq,
6525 };
6526 
6527 static const struct ib_device_ops hns_roce_v2_dev_srq_ops = {
6528 	.modify_srq = hns_roce_v2_modify_srq,
6529 	.post_srq_recv = hns_roce_v2_post_srq_recv,
6530 	.query_srq = hns_roce_v2_query_srq,
6531 };
6532 
6533 static const struct hns_roce_hw hns_roce_hw_v2 = {
6534 	.cmq_init = hns_roce_v2_cmq_init,
6535 	.cmq_exit = hns_roce_v2_cmq_exit,
6536 	.hw_profile = hns_roce_v2_profile,
6537 	.hw_init = hns_roce_v2_init,
6538 	.hw_exit = hns_roce_v2_exit,
6539 	.post_mbox = v2_post_mbox,
6540 	.poll_mbox_done = v2_poll_mbox_done,
6541 	.chk_mbox_avail = v2_chk_mbox_is_avail,
6542 	.set_gid = hns_roce_v2_set_gid,
6543 	.set_mac = hns_roce_v2_set_mac,
6544 	.write_mtpt = hns_roce_v2_write_mtpt,
6545 	.rereg_write_mtpt = hns_roce_v2_rereg_write_mtpt,
6546 	.frmr_write_mtpt = hns_roce_v2_frmr_write_mtpt,
6547 	.mw_write_mtpt = hns_roce_v2_mw_write_mtpt,
6548 	.write_cqc = hns_roce_v2_write_cqc,
6549 	.set_hem = hns_roce_v2_set_hem,
6550 	.clear_hem = hns_roce_v2_clear_hem,
6551 	.modify_qp = hns_roce_v2_modify_qp,
6552 	.qp_flow_control_init = hns_roce_v2_qp_flow_control_init,
6553 	.init_eq = hns_roce_v2_init_eq_table,
6554 	.cleanup_eq = hns_roce_v2_cleanup_eq_table,
6555 	.write_srqc = hns_roce_v2_write_srqc,
6556 	.hns_roce_dev_ops = &hns_roce_v2_dev_ops,
6557 	.hns_roce_dev_srq_ops = &hns_roce_v2_dev_srq_ops,
6558 };
6559 
6560 static const struct pci_device_id hns_roce_hw_v2_pci_tbl[] = {
6561 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA), 0},
6562 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE_RDMA_MACSEC), 0},
6563 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA), 0},
6564 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_50GE_RDMA_MACSEC), 0},
6565 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_MACSEC), 0},
6566 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_200G_RDMA), 0},
6567 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_RDMA_DCB_PFC_VF),
6568 	 HNAE3_DEV_SUPPORT_ROCE_DCB_BITS},
6569 	/* required last entry */
6570 	{0, }
6571 };
6572 
6573 MODULE_DEVICE_TABLE(pci, hns_roce_hw_v2_pci_tbl);
6574 
6575 static void hns_roce_hw_v2_get_cfg(struct hns_roce_dev *hr_dev,
6576 				  struct hnae3_handle *handle)
6577 {
6578 	struct hns_roce_v2_priv *priv = hr_dev->priv;
6579 	const struct pci_device_id *id;
6580 	int i;
6581 
6582 	hr_dev->pci_dev = handle->pdev;
6583 	id = pci_match_id(hns_roce_hw_v2_pci_tbl, hr_dev->pci_dev);
6584 	hr_dev->is_vf = id->driver_data;
6585 	hr_dev->dev = &handle->pdev->dev;
6586 	hr_dev->hw = &hns_roce_hw_v2;
6587 	hr_dev->dfx = &hns_roce_dfx_hw_v2;
6588 	hr_dev->sdb_offset = ROCEE_DB_SQ_L_0_REG;
6589 	hr_dev->odb_offset = hr_dev->sdb_offset;
6590 
6591 	/* Get info from NIC driver. */
6592 	hr_dev->reg_base = handle->rinfo.roce_io_base;
6593 	hr_dev->mem_base = handle->rinfo.roce_mem_base;
6594 	hr_dev->caps.num_ports = 1;
6595 	hr_dev->iboe.netdevs[0] = handle->rinfo.netdev;
6596 	hr_dev->iboe.phy_port[0] = 0;
6597 
6598 	addrconf_addr_eui48((u8 *)&hr_dev->ib_dev.node_guid,
6599 			    hr_dev->iboe.netdevs[0]->dev_addr);
6600 
6601 	for (i = 0; i < handle->rinfo.num_vectors; i++)
6602 		hr_dev->irq[i] = pci_irq_vector(handle->pdev,
6603 						i + handle->rinfo.base_vector);
6604 
6605 	/* cmd issue mode: 0 is poll, 1 is event */
6606 	hr_dev->cmd_mod = 1;
6607 	hr_dev->loop_idc = 0;
6608 
6609 	hr_dev->reset_cnt = handle->ae_algo->ops->ae_dev_reset_cnt(handle);
6610 	priv->handle = handle;
6611 }
6612 
6613 static int __hns_roce_hw_v2_init_instance(struct hnae3_handle *handle)
6614 {
6615 	struct hns_roce_dev *hr_dev;
6616 	int ret;
6617 
6618 	hr_dev = ib_alloc_device(hns_roce_dev, ib_dev);
6619 	if (!hr_dev)
6620 		return -ENOMEM;
6621 
6622 	hr_dev->priv = kzalloc(sizeof(struct hns_roce_v2_priv), GFP_KERNEL);
6623 	if (!hr_dev->priv) {
6624 		ret = -ENOMEM;
6625 		goto error_failed_kzalloc;
6626 	}
6627 
6628 	hns_roce_hw_v2_get_cfg(hr_dev, handle);
6629 
6630 	ret = hns_roce_init(hr_dev);
6631 	if (ret) {
6632 		dev_err(hr_dev->dev, "RoCE Engine init failed!\n");
6633 		goto error_failed_get_cfg;
6634 	}
6635 
6636 	handle->priv = hr_dev;
6637 
6638 	return 0;
6639 
6640 error_failed_get_cfg:
6641 	kfree(hr_dev->priv);
6642 
6643 error_failed_kzalloc:
6644 	ib_dealloc_device(&hr_dev->ib_dev);
6645 
6646 	return ret;
6647 }
6648 
6649 static void __hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle,
6650 					   bool reset)
6651 {
6652 	struct hns_roce_dev *hr_dev = handle->priv;
6653 
6654 	if (!hr_dev)
6655 		return;
6656 
6657 	handle->priv = NULL;
6658 
6659 	hr_dev->state = HNS_ROCE_DEVICE_STATE_UNINIT;
6660 	hns_roce_handle_device_err(hr_dev);
6661 
6662 	hns_roce_exit(hr_dev);
6663 	kfree(hr_dev->priv);
6664 	ib_dealloc_device(&hr_dev->ib_dev);
6665 }
6666 
6667 static int hns_roce_hw_v2_init_instance(struct hnae3_handle *handle)
6668 {
6669 	const struct hnae3_ae_ops *ops = handle->ae_algo->ops;
6670 	const struct pci_device_id *id;
6671 	struct device *dev = &handle->pdev->dev;
6672 	int ret;
6673 
6674 	handle->rinfo.instance_state = HNS_ROCE_STATE_INIT;
6675 
6676 	if (ops->ae_dev_resetting(handle) || ops->get_hw_reset_stat(handle)) {
6677 		handle->rinfo.instance_state = HNS_ROCE_STATE_NON_INIT;
6678 		goto reset_chk_err;
6679 	}
6680 
6681 	id = pci_match_id(hns_roce_hw_v2_pci_tbl, handle->pdev);
6682 	if (!id)
6683 		return 0;
6684 
6685 	if (id->driver_data && handle->pdev->revision < PCI_REVISION_ID_HIP09)
6686 		return 0;
6687 
6688 	ret = __hns_roce_hw_v2_init_instance(handle);
6689 	if (ret) {
6690 		handle->rinfo.instance_state = HNS_ROCE_STATE_NON_INIT;
6691 		dev_err(dev, "RoCE instance init failed! ret = %d\n", ret);
6692 		if (ops->ae_dev_resetting(handle) ||
6693 		    ops->get_hw_reset_stat(handle))
6694 			goto reset_chk_err;
6695 		else
6696 			return ret;
6697 	}
6698 
6699 	handle->rinfo.instance_state = HNS_ROCE_STATE_INITED;
6700 
6701 
6702 	return 0;
6703 
6704 reset_chk_err:
6705 	dev_err(dev, "Device is busy in resetting state.\n"
6706 		     "please retry later.\n");
6707 
6708 	return -EBUSY;
6709 }
6710 
6711 static void hns_roce_hw_v2_uninit_instance(struct hnae3_handle *handle,
6712 					   bool reset)
6713 {
6714 	if (handle->rinfo.instance_state != HNS_ROCE_STATE_INITED)
6715 		return;
6716 
6717 	handle->rinfo.instance_state = HNS_ROCE_STATE_UNINIT;
6718 
6719 	__hns_roce_hw_v2_uninit_instance(handle, reset);
6720 
6721 	handle->rinfo.instance_state = HNS_ROCE_STATE_NON_INIT;
6722 }
6723 static int hns_roce_hw_v2_reset_notify_down(struct hnae3_handle *handle)
6724 {
6725 	struct hns_roce_dev *hr_dev;
6726 
6727 	if (handle->rinfo.instance_state != HNS_ROCE_STATE_INITED) {
6728 		set_bit(HNS_ROCE_RST_DIRECT_RETURN, &handle->rinfo.state);
6729 		return 0;
6730 	}
6731 
6732 	handle->rinfo.reset_state = HNS_ROCE_STATE_RST_DOWN;
6733 	clear_bit(HNS_ROCE_RST_DIRECT_RETURN, &handle->rinfo.state);
6734 
6735 	hr_dev = handle->priv;
6736 	if (!hr_dev)
6737 		return 0;
6738 
6739 	hr_dev->is_reset = true;
6740 	hr_dev->active = false;
6741 	hr_dev->dis_db = true;
6742 
6743 	hr_dev->state = HNS_ROCE_DEVICE_STATE_RST_DOWN;
6744 
6745 	return 0;
6746 }
6747 
6748 static int hns_roce_hw_v2_reset_notify_init(struct hnae3_handle *handle)
6749 {
6750 	struct device *dev = &handle->pdev->dev;
6751 	int ret;
6752 
6753 	if (test_and_clear_bit(HNS_ROCE_RST_DIRECT_RETURN,
6754 			       &handle->rinfo.state)) {
6755 		handle->rinfo.reset_state = HNS_ROCE_STATE_RST_INITED;
6756 		return 0;
6757 	}
6758 
6759 	handle->rinfo.reset_state = HNS_ROCE_STATE_RST_INIT;
6760 
6761 	dev_info(&handle->pdev->dev, "In reset process RoCE client reinit.\n");
6762 	ret = __hns_roce_hw_v2_init_instance(handle);
6763 	if (ret) {
6764 		/* when reset notify type is HNAE3_INIT_CLIENT In reset notify
6765 		 * callback function, RoCE Engine reinitialize. If RoCE reinit
6766 		 * failed, we should inform NIC driver.
6767 		 */
6768 		handle->priv = NULL;
6769 		dev_err(dev, "In reset process RoCE reinit failed %d.\n", ret);
6770 	} else {
6771 		handle->rinfo.reset_state = HNS_ROCE_STATE_RST_INITED;
6772 		dev_info(dev, "Reset done, RoCE client reinit finished.\n");
6773 	}
6774 
6775 	return ret;
6776 }
6777 
6778 static int hns_roce_hw_v2_reset_notify_uninit(struct hnae3_handle *handle)
6779 {
6780 	if (test_bit(HNS_ROCE_RST_DIRECT_RETURN, &handle->rinfo.state))
6781 		return 0;
6782 
6783 	handle->rinfo.reset_state = HNS_ROCE_STATE_RST_UNINIT;
6784 	dev_info(&handle->pdev->dev, "In reset process RoCE client uninit.\n");
6785 	msleep(HNS_ROCE_V2_HW_RST_UNINT_DELAY);
6786 	__hns_roce_hw_v2_uninit_instance(handle, false);
6787 
6788 	return 0;
6789 }
6790 
6791 static int hns_roce_hw_v2_reset_notify(struct hnae3_handle *handle,
6792 				       enum hnae3_reset_notify_type type)
6793 {
6794 	int ret = 0;
6795 
6796 	switch (type) {
6797 	case HNAE3_DOWN_CLIENT:
6798 		ret = hns_roce_hw_v2_reset_notify_down(handle);
6799 		break;
6800 	case HNAE3_INIT_CLIENT:
6801 		ret = hns_roce_hw_v2_reset_notify_init(handle);
6802 		break;
6803 	case HNAE3_UNINIT_CLIENT:
6804 		ret = hns_roce_hw_v2_reset_notify_uninit(handle);
6805 		break;
6806 	default:
6807 		break;
6808 	}
6809 
6810 	return ret;
6811 }
6812 
6813 static const struct hnae3_client_ops hns_roce_hw_v2_ops = {
6814 	.init_instance = hns_roce_hw_v2_init_instance,
6815 	.uninit_instance = hns_roce_hw_v2_uninit_instance,
6816 	.reset_notify = hns_roce_hw_v2_reset_notify,
6817 };
6818 
6819 static struct hnae3_client hns_roce_hw_v2_client = {
6820 	.name = "hns_roce_hw_v2",
6821 	.type = HNAE3_CLIENT_ROCE,
6822 	.ops = &hns_roce_hw_v2_ops,
6823 };
6824 
6825 static int __init hns_roce_hw_v2_init(void)
6826 {
6827 	return hnae3_register_client(&hns_roce_hw_v2_client);
6828 }
6829 
6830 static void __exit hns_roce_hw_v2_exit(void)
6831 {
6832 	hnae3_unregister_client(&hns_roce_hw_v2_client);
6833 }
6834 
6835 module_init(hns_roce_hw_v2_init);
6836 module_exit(hns_roce_hw_v2_exit);
6837 
6838 MODULE_LICENSE("Dual BSD/GPL");
6839 MODULE_AUTHOR("Wei Hu <xavier.huwei@huawei.com>");
6840 MODULE_AUTHOR("Lijun Ou <oulijun@huawei.com>");
6841 MODULE_AUTHOR("Shaobo Xu <xushaobo2@huawei.com>");
6842 MODULE_DESCRIPTION("Hisilicon Hip08 Family RoCE Driver");
6843