1 /*******************************************************************
2  * This file is part of the Emulex RoCE Device Driver for          *
3  * RoCE (RDMA over Converged Ethernet) CNA Adapters.              *
4  * Copyright (C) 2008-2012 Emulex. All rights reserved.            *
5  * EMULEX and SLI are trademarks of Emulex.                        *
6  * www.emulex.com                                                  *
7  *                                                                 *
8  * This program is free software; you can redistribute it and/or   *
9  * modify it under the terms of version 2 of the GNU General       *
10  * Public License as published by the Free Software Foundation.    *
11  * This program is distributed in the hope that it will be useful. *
12  * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
13  * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
14  * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
15  * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
16  * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
17  * more details, a copy of which can be found in the file COPYING  *
18  * included with this package.                                     *
19  *
20  * Contact Information:
21  * linux-drivers@emulex.com
22  *
23  * Emulex
24  * 3333 Susan Street
25  * Costa Mesa, CA 92626
26  *******************************************************************/
27 
28 #include <linux/sched.h>
29 #include <linux/interrupt.h>
30 #include <linux/log2.h>
31 #include <linux/dma-mapping.h>
32 
33 #include <rdma/ib_verbs.h>
34 #include <rdma/ib_user_verbs.h>
35 
36 #include "ocrdma.h"
37 #include "ocrdma_hw.h"
38 #include "ocrdma_verbs.h"
39 #include "ocrdma_ah.h"
40 
41 enum mbx_status {
42 	OCRDMA_MBX_STATUS_FAILED		= 1,
43 	OCRDMA_MBX_STATUS_ILLEGAL_FIELD		= 3,
44 	OCRDMA_MBX_STATUS_OOR			= 100,
45 	OCRDMA_MBX_STATUS_INVALID_PD		= 101,
46 	OCRDMA_MBX_STATUS_PD_INUSE		= 102,
47 	OCRDMA_MBX_STATUS_INVALID_CQ		= 103,
48 	OCRDMA_MBX_STATUS_INVALID_QP		= 104,
49 	OCRDMA_MBX_STATUS_INVALID_LKEY		= 105,
50 	OCRDMA_MBX_STATUS_ORD_EXCEEDS		= 106,
51 	OCRDMA_MBX_STATUS_IRD_EXCEEDS		= 107,
52 	OCRDMA_MBX_STATUS_SENDQ_WQE_EXCEEDS	= 108,
53 	OCRDMA_MBX_STATUS_RECVQ_RQE_EXCEEDS	= 109,
54 	OCRDMA_MBX_STATUS_SGE_SEND_EXCEEDS	= 110,
55 	OCRDMA_MBX_STATUS_SGE_WRITE_EXCEEDS	= 111,
56 	OCRDMA_MBX_STATUS_SGE_RECV_EXCEEDS	= 112,
57 	OCRDMA_MBX_STATUS_INVALID_STATE_CHANGE	= 113,
58 	OCRDMA_MBX_STATUS_MW_BOUND		= 114,
59 	OCRDMA_MBX_STATUS_INVALID_VA		= 115,
60 	OCRDMA_MBX_STATUS_INVALID_LENGTH	= 116,
61 	OCRDMA_MBX_STATUS_INVALID_FBO		= 117,
62 	OCRDMA_MBX_STATUS_INVALID_ACC_RIGHTS	= 118,
63 	OCRDMA_MBX_STATUS_INVALID_PBE_SIZE	= 119,
64 	OCRDMA_MBX_STATUS_INVALID_PBL_ENTRY	= 120,
65 	OCRDMA_MBX_STATUS_INVALID_PBL_SHIFT	= 121,
66 	OCRDMA_MBX_STATUS_INVALID_SRQ_ID	= 129,
67 	OCRDMA_MBX_STATUS_SRQ_ERROR		= 133,
68 	OCRDMA_MBX_STATUS_RQE_EXCEEDS		= 134,
69 	OCRDMA_MBX_STATUS_MTU_EXCEEDS		= 135,
70 	OCRDMA_MBX_STATUS_MAX_QP_EXCEEDS	= 136,
71 	OCRDMA_MBX_STATUS_SRQ_LIMIT_EXCEEDS	= 137,
72 	OCRDMA_MBX_STATUS_SRQ_SIZE_UNDERUNS	= 138,
73 	OCRDMA_MBX_STATUS_QP_BOUND		= 130,
74 	OCRDMA_MBX_STATUS_INVALID_CHANGE	= 139,
75 	OCRDMA_MBX_STATUS_ATOMIC_OPS_UNSUP	= 140,
76 	OCRDMA_MBX_STATUS_INVALID_RNR_NAK_TIMER	= 141,
77 	OCRDMA_MBX_STATUS_MW_STILL_BOUND	= 142,
78 	OCRDMA_MBX_STATUS_PKEY_INDEX_INVALID	= 143,
79 	OCRDMA_MBX_STATUS_PKEY_INDEX_EXCEEDS	= 144
80 };
81 
82 enum additional_status {
83 	OCRDMA_MBX_ADDI_STATUS_INSUFFICIENT_RESOURCES = 22
84 };
85 
86 enum cqe_status {
87 	OCRDMA_MBX_CQE_STATUS_INSUFFICIENT_PRIVILEDGES	= 1,
88 	OCRDMA_MBX_CQE_STATUS_INVALID_PARAMETER		= 2,
89 	OCRDMA_MBX_CQE_STATUS_INSUFFICIENT_RESOURCES	= 3,
90 	OCRDMA_MBX_CQE_STATUS_QUEUE_FLUSHING		= 4,
91 	OCRDMA_MBX_CQE_STATUS_DMA_FAILED		= 5
92 };
93 
94 static inline void *ocrdma_get_eqe(struct ocrdma_eq *eq)
95 {
96 	return eq->q.va + (eq->q.tail * sizeof(struct ocrdma_eqe));
97 }
98 
99 static inline void ocrdma_eq_inc_tail(struct ocrdma_eq *eq)
100 {
101 	eq->q.tail = (eq->q.tail + 1) & (OCRDMA_EQ_LEN - 1);
102 }
103 
104 static inline void *ocrdma_get_mcqe(struct ocrdma_dev *dev)
105 {
106 	struct ocrdma_mcqe *cqe = (struct ocrdma_mcqe *)
107 	    (dev->mq.cq.va + (dev->mq.cq.tail * sizeof(struct ocrdma_mcqe)));
108 
109 	if (!(le32_to_cpu(cqe->valid_ae_cmpl_cons) & OCRDMA_MCQE_VALID_MASK))
110 		return NULL;
111 	return cqe;
112 }
113 
114 static inline void ocrdma_mcq_inc_tail(struct ocrdma_dev *dev)
115 {
116 	dev->mq.cq.tail = (dev->mq.cq.tail + 1) & (OCRDMA_MQ_CQ_LEN - 1);
117 }
118 
119 static inline struct ocrdma_mqe *ocrdma_get_mqe(struct ocrdma_dev *dev)
120 {
121 	return dev->mq.sq.va + (dev->mq.sq.head * sizeof(struct ocrdma_mqe));
122 }
123 
124 static inline void ocrdma_mq_inc_head(struct ocrdma_dev *dev)
125 {
126 	dev->mq.sq.head = (dev->mq.sq.head + 1) & (OCRDMA_MQ_LEN - 1);
127 }
128 
129 static inline void *ocrdma_get_mqe_rsp(struct ocrdma_dev *dev)
130 {
131 	return dev->mq.sq.va + (dev->mqe_ctx.tag * sizeof(struct ocrdma_mqe));
132 }
133 
134 enum ib_qp_state get_ibqp_state(enum ocrdma_qp_state qps)
135 {
136 	switch (qps) {
137 	case OCRDMA_QPS_RST:
138 		return IB_QPS_RESET;
139 	case OCRDMA_QPS_INIT:
140 		return IB_QPS_INIT;
141 	case OCRDMA_QPS_RTR:
142 		return IB_QPS_RTR;
143 	case OCRDMA_QPS_RTS:
144 		return IB_QPS_RTS;
145 	case OCRDMA_QPS_SQD:
146 	case OCRDMA_QPS_SQ_DRAINING:
147 		return IB_QPS_SQD;
148 	case OCRDMA_QPS_SQE:
149 		return IB_QPS_SQE;
150 	case OCRDMA_QPS_ERR:
151 		return IB_QPS_ERR;
152 	}
153 	return IB_QPS_ERR;
154 }
155 
156 static enum ocrdma_qp_state get_ocrdma_qp_state(enum ib_qp_state qps)
157 {
158 	switch (qps) {
159 	case IB_QPS_RESET:
160 		return OCRDMA_QPS_RST;
161 	case IB_QPS_INIT:
162 		return OCRDMA_QPS_INIT;
163 	case IB_QPS_RTR:
164 		return OCRDMA_QPS_RTR;
165 	case IB_QPS_RTS:
166 		return OCRDMA_QPS_RTS;
167 	case IB_QPS_SQD:
168 		return OCRDMA_QPS_SQD;
169 	case IB_QPS_SQE:
170 		return OCRDMA_QPS_SQE;
171 	case IB_QPS_ERR:
172 		return OCRDMA_QPS_ERR;
173 	}
174 	return OCRDMA_QPS_ERR;
175 }
176 
177 static int ocrdma_get_mbx_errno(u32 status)
178 {
179 	int err_num;
180 	u8 mbox_status = (status & OCRDMA_MBX_RSP_STATUS_MASK) >>
181 					OCRDMA_MBX_RSP_STATUS_SHIFT;
182 	u8 add_status = (status & OCRDMA_MBX_RSP_ASTATUS_MASK) >>
183 					OCRDMA_MBX_RSP_ASTATUS_SHIFT;
184 
185 	switch (mbox_status) {
186 	case OCRDMA_MBX_STATUS_OOR:
187 	case OCRDMA_MBX_STATUS_MAX_QP_EXCEEDS:
188 		err_num = -EAGAIN;
189 		break;
190 
191 	case OCRDMA_MBX_STATUS_INVALID_PD:
192 	case OCRDMA_MBX_STATUS_INVALID_CQ:
193 	case OCRDMA_MBX_STATUS_INVALID_SRQ_ID:
194 	case OCRDMA_MBX_STATUS_INVALID_QP:
195 	case OCRDMA_MBX_STATUS_INVALID_CHANGE:
196 	case OCRDMA_MBX_STATUS_MTU_EXCEEDS:
197 	case OCRDMA_MBX_STATUS_INVALID_RNR_NAK_TIMER:
198 	case OCRDMA_MBX_STATUS_PKEY_INDEX_INVALID:
199 	case OCRDMA_MBX_STATUS_PKEY_INDEX_EXCEEDS:
200 	case OCRDMA_MBX_STATUS_ILLEGAL_FIELD:
201 	case OCRDMA_MBX_STATUS_INVALID_PBL_ENTRY:
202 	case OCRDMA_MBX_STATUS_INVALID_LKEY:
203 	case OCRDMA_MBX_STATUS_INVALID_VA:
204 	case OCRDMA_MBX_STATUS_INVALID_LENGTH:
205 	case OCRDMA_MBX_STATUS_INVALID_FBO:
206 	case OCRDMA_MBX_STATUS_INVALID_ACC_RIGHTS:
207 	case OCRDMA_MBX_STATUS_INVALID_PBE_SIZE:
208 	case OCRDMA_MBX_STATUS_ATOMIC_OPS_UNSUP:
209 	case OCRDMA_MBX_STATUS_SRQ_ERROR:
210 	case OCRDMA_MBX_STATUS_SRQ_SIZE_UNDERUNS:
211 		err_num = -EINVAL;
212 		break;
213 
214 	case OCRDMA_MBX_STATUS_PD_INUSE:
215 	case OCRDMA_MBX_STATUS_QP_BOUND:
216 	case OCRDMA_MBX_STATUS_MW_STILL_BOUND:
217 	case OCRDMA_MBX_STATUS_MW_BOUND:
218 		err_num = -EBUSY;
219 		break;
220 
221 	case OCRDMA_MBX_STATUS_RECVQ_RQE_EXCEEDS:
222 	case OCRDMA_MBX_STATUS_SGE_RECV_EXCEEDS:
223 	case OCRDMA_MBX_STATUS_RQE_EXCEEDS:
224 	case OCRDMA_MBX_STATUS_SRQ_LIMIT_EXCEEDS:
225 	case OCRDMA_MBX_STATUS_ORD_EXCEEDS:
226 	case OCRDMA_MBX_STATUS_IRD_EXCEEDS:
227 	case OCRDMA_MBX_STATUS_SENDQ_WQE_EXCEEDS:
228 	case OCRDMA_MBX_STATUS_SGE_SEND_EXCEEDS:
229 	case OCRDMA_MBX_STATUS_SGE_WRITE_EXCEEDS:
230 		err_num = -ENOBUFS;
231 		break;
232 
233 	case OCRDMA_MBX_STATUS_FAILED:
234 		switch (add_status) {
235 		case OCRDMA_MBX_ADDI_STATUS_INSUFFICIENT_RESOURCES:
236 			err_num = -EAGAIN;
237 			break;
238 		}
239 	default:
240 		err_num = -EFAULT;
241 	}
242 	return err_num;
243 }
244 
245 char *port_speed_string(struct ocrdma_dev *dev)
246 {
247 	char *str = "";
248 	u16 speeds_supported;
249 
250 	speeds_supported = dev->phy.fixed_speeds_supported |
251 				dev->phy.auto_speeds_supported;
252 	if (speeds_supported & OCRDMA_PHY_SPEED_40GBPS)
253 		str = "40Gbps ";
254 	else if (speeds_supported & OCRDMA_PHY_SPEED_10GBPS)
255 		str = "10Gbps ";
256 	else if (speeds_supported & OCRDMA_PHY_SPEED_1GBPS)
257 		str = "1Gbps ";
258 
259 	return str;
260 }
261 
262 static int ocrdma_get_mbx_cqe_errno(u16 cqe_status)
263 {
264 	int err_num = -EINVAL;
265 
266 	switch (cqe_status) {
267 	case OCRDMA_MBX_CQE_STATUS_INSUFFICIENT_PRIVILEDGES:
268 		err_num = -EPERM;
269 		break;
270 	case OCRDMA_MBX_CQE_STATUS_INVALID_PARAMETER:
271 		err_num = -EINVAL;
272 		break;
273 	case OCRDMA_MBX_CQE_STATUS_INSUFFICIENT_RESOURCES:
274 	case OCRDMA_MBX_CQE_STATUS_QUEUE_FLUSHING:
275 		err_num = -EINVAL;
276 		break;
277 	case OCRDMA_MBX_CQE_STATUS_DMA_FAILED:
278 	default:
279 		err_num = -EINVAL;
280 		break;
281 	}
282 	return err_num;
283 }
284 
285 void ocrdma_ring_cq_db(struct ocrdma_dev *dev, u16 cq_id, bool armed,
286 		       bool solicited, u16 cqe_popped)
287 {
288 	u32 val = cq_id & OCRDMA_DB_CQ_RING_ID_MASK;
289 
290 	val |= ((cq_id & OCRDMA_DB_CQ_RING_ID_EXT_MASK) <<
291 	     OCRDMA_DB_CQ_RING_ID_EXT_MASK_SHIFT);
292 
293 	if (armed)
294 		val |= (1 << OCRDMA_DB_CQ_REARM_SHIFT);
295 	if (solicited)
296 		val |= (1 << OCRDMA_DB_CQ_SOLICIT_SHIFT);
297 	val |= (cqe_popped << OCRDMA_DB_CQ_NUM_POPPED_SHIFT);
298 	iowrite32(val, dev->nic_info.db + OCRDMA_DB_CQ_OFFSET);
299 }
300 
301 static void ocrdma_ring_mq_db(struct ocrdma_dev *dev)
302 {
303 	u32 val = 0;
304 
305 	val |= dev->mq.sq.id & OCRDMA_MQ_ID_MASK;
306 	val |= 1 << OCRDMA_MQ_NUM_MQE_SHIFT;
307 	iowrite32(val, dev->nic_info.db + OCRDMA_DB_MQ_OFFSET);
308 }
309 
310 static void ocrdma_ring_eq_db(struct ocrdma_dev *dev, u16 eq_id,
311 			      bool arm, bool clear_int, u16 num_eqe)
312 {
313 	u32 val = 0;
314 
315 	val |= eq_id & OCRDMA_EQ_ID_MASK;
316 	val |= ((eq_id & OCRDMA_EQ_ID_EXT_MASK) << OCRDMA_EQ_ID_EXT_MASK_SHIFT);
317 	if (arm)
318 		val |= (1 << OCRDMA_REARM_SHIFT);
319 	if (clear_int)
320 		val |= (1 << OCRDMA_EQ_CLR_SHIFT);
321 	val |= (1 << OCRDMA_EQ_TYPE_SHIFT);
322 	val |= (num_eqe << OCRDMA_NUM_EQE_SHIFT);
323 	iowrite32(val, dev->nic_info.db + OCRDMA_DB_EQ_OFFSET);
324 }
325 
326 static void ocrdma_init_mch(struct ocrdma_mbx_hdr *cmd_hdr,
327 			    u8 opcode, u8 subsys, u32 cmd_len)
328 {
329 	cmd_hdr->subsys_op = (opcode | (subsys << OCRDMA_MCH_SUBSYS_SHIFT));
330 	cmd_hdr->timeout = 20; /* seconds */
331 	cmd_hdr->cmd_len = cmd_len - sizeof(struct ocrdma_mbx_hdr);
332 }
333 
334 static void *ocrdma_init_emb_mqe(u8 opcode, u32 cmd_len)
335 {
336 	struct ocrdma_mqe *mqe;
337 
338 	mqe = kzalloc(sizeof(struct ocrdma_mqe), GFP_KERNEL);
339 	if (!mqe)
340 		return NULL;
341 	mqe->hdr.spcl_sge_cnt_emb |=
342 		(OCRDMA_MQE_EMBEDDED << OCRDMA_MQE_HDR_EMB_SHIFT) &
343 					OCRDMA_MQE_HDR_EMB_MASK;
344 	mqe->hdr.pyld_len = cmd_len - sizeof(struct ocrdma_mqe_hdr);
345 
346 	ocrdma_init_mch(&mqe->u.emb_req.mch, opcode, OCRDMA_SUBSYS_ROCE,
347 			mqe->hdr.pyld_len);
348 	return mqe;
349 }
350 
351 static void ocrdma_free_q(struct ocrdma_dev *dev, struct ocrdma_queue_info *q)
352 {
353 	dma_free_coherent(&dev->nic_info.pdev->dev, q->size, q->va, q->dma);
354 }
355 
356 static int ocrdma_alloc_q(struct ocrdma_dev *dev,
357 			  struct ocrdma_queue_info *q, u16 len, u16 entry_size)
358 {
359 	memset(q, 0, sizeof(*q));
360 	q->len = len;
361 	q->entry_size = entry_size;
362 	q->size = len * entry_size;
363 	q->va = dma_alloc_coherent(&dev->nic_info.pdev->dev, q->size,
364 				   &q->dma, GFP_KERNEL);
365 	if (!q->va)
366 		return -ENOMEM;
367 	memset(q->va, 0, q->size);
368 	return 0;
369 }
370 
371 static void ocrdma_build_q_pages(struct ocrdma_pa *q_pa, int cnt,
372 					dma_addr_t host_pa, int hw_page_size)
373 {
374 	int i;
375 
376 	for (i = 0; i < cnt; i++) {
377 		q_pa[i].lo = (u32) (host_pa & 0xffffffff);
378 		q_pa[i].hi = (u32) upper_32_bits(host_pa);
379 		host_pa += hw_page_size;
380 	}
381 }
382 
383 static int ocrdma_mbx_delete_q(struct ocrdma_dev *dev,
384 			       struct ocrdma_queue_info *q, int queue_type)
385 {
386 	u8 opcode = 0;
387 	int status;
388 	struct ocrdma_delete_q_req *cmd = dev->mbx_cmd;
389 
390 	switch (queue_type) {
391 	case QTYPE_MCCQ:
392 		opcode = OCRDMA_CMD_DELETE_MQ;
393 		break;
394 	case QTYPE_CQ:
395 		opcode = OCRDMA_CMD_DELETE_CQ;
396 		break;
397 	case QTYPE_EQ:
398 		opcode = OCRDMA_CMD_DELETE_EQ;
399 		break;
400 	default:
401 		BUG();
402 	}
403 	memset(cmd, 0, sizeof(*cmd));
404 	ocrdma_init_mch(&cmd->req, opcode, OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
405 	cmd->id = q->id;
406 
407 	status = be_roce_mcc_cmd(dev->nic_info.netdev,
408 				 cmd, sizeof(*cmd), NULL, NULL);
409 	if (!status)
410 		q->created = false;
411 	return status;
412 }
413 
414 static int ocrdma_mbx_create_eq(struct ocrdma_dev *dev, struct ocrdma_eq *eq)
415 {
416 	int status;
417 	struct ocrdma_create_eq_req *cmd = dev->mbx_cmd;
418 	struct ocrdma_create_eq_rsp *rsp = dev->mbx_cmd;
419 
420 	memset(cmd, 0, sizeof(*cmd));
421 	ocrdma_init_mch(&cmd->req, OCRDMA_CMD_CREATE_EQ, OCRDMA_SUBSYS_COMMON,
422 			sizeof(*cmd));
423 
424 	cmd->req.rsvd_version = 2;
425 	cmd->num_pages = 4;
426 	cmd->valid = OCRDMA_CREATE_EQ_VALID;
427 	cmd->cnt = 4 << OCRDMA_CREATE_EQ_CNT_SHIFT;
428 
429 	ocrdma_build_q_pages(&cmd->pa[0], cmd->num_pages, eq->q.dma,
430 			     PAGE_SIZE_4K);
431 	status = be_roce_mcc_cmd(dev->nic_info.netdev, cmd, sizeof(*cmd), NULL,
432 				 NULL);
433 	if (!status) {
434 		eq->q.id = rsp->vector_eqid & 0xffff;
435 		eq->vector = (rsp->vector_eqid >> 16) & 0xffff;
436 		eq->q.created = true;
437 	}
438 	return status;
439 }
440 
441 static int ocrdma_create_eq(struct ocrdma_dev *dev,
442 			    struct ocrdma_eq *eq, u16 q_len)
443 {
444 	int status;
445 
446 	status = ocrdma_alloc_q(dev, &eq->q, OCRDMA_EQ_LEN,
447 				sizeof(struct ocrdma_eqe));
448 	if (status)
449 		return status;
450 
451 	status = ocrdma_mbx_create_eq(dev, eq);
452 	if (status)
453 		goto mbx_err;
454 	eq->dev = dev;
455 	ocrdma_ring_eq_db(dev, eq->q.id, true, true, 0);
456 
457 	return 0;
458 mbx_err:
459 	ocrdma_free_q(dev, &eq->q);
460 	return status;
461 }
462 
463 int ocrdma_get_irq(struct ocrdma_dev *dev, struct ocrdma_eq *eq)
464 {
465 	int irq;
466 
467 	if (dev->nic_info.intr_mode == BE_INTERRUPT_MODE_INTX)
468 		irq = dev->nic_info.pdev->irq;
469 	else
470 		irq = dev->nic_info.msix.vector_list[eq->vector];
471 	return irq;
472 }
473 
474 static void _ocrdma_destroy_eq(struct ocrdma_dev *dev, struct ocrdma_eq *eq)
475 {
476 	if (eq->q.created) {
477 		ocrdma_mbx_delete_q(dev, &eq->q, QTYPE_EQ);
478 		ocrdma_free_q(dev, &eq->q);
479 	}
480 }
481 
482 static void ocrdma_destroy_eq(struct ocrdma_dev *dev, struct ocrdma_eq *eq)
483 {
484 	int irq;
485 
486 	/* disarm EQ so that interrupts are not generated
487 	 * during freeing and EQ delete is in progress.
488 	 */
489 	ocrdma_ring_eq_db(dev, eq->q.id, false, false, 0);
490 
491 	irq = ocrdma_get_irq(dev, eq);
492 	free_irq(irq, eq);
493 	_ocrdma_destroy_eq(dev, eq);
494 }
495 
496 static void ocrdma_destroy_eqs(struct ocrdma_dev *dev)
497 {
498 	int i;
499 
500 	for (i = 0; i < dev->eq_cnt; i++)
501 		ocrdma_destroy_eq(dev, &dev->eq_tbl[i]);
502 }
503 
504 static int ocrdma_mbx_mq_cq_create(struct ocrdma_dev *dev,
505 				   struct ocrdma_queue_info *cq,
506 				   struct ocrdma_queue_info *eq)
507 {
508 	struct ocrdma_create_cq_cmd *cmd = dev->mbx_cmd;
509 	struct ocrdma_create_cq_cmd_rsp *rsp = dev->mbx_cmd;
510 	int status;
511 
512 	memset(cmd, 0, sizeof(*cmd));
513 	ocrdma_init_mch(&cmd->req, OCRDMA_CMD_CREATE_CQ,
514 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
515 
516 	cmd->req.rsvd_version = OCRDMA_CREATE_CQ_VER2;
517 	cmd->pgsz_pgcnt = (cq->size / OCRDMA_MIN_Q_PAGE_SIZE) <<
518 		OCRDMA_CREATE_CQ_PAGE_SIZE_SHIFT;
519 	cmd->pgsz_pgcnt |= PAGES_4K_SPANNED(cq->va, cq->size);
520 
521 	cmd->ev_cnt_flags = OCRDMA_CREATE_CQ_DEF_FLAGS;
522 	cmd->eqn = eq->id;
523 	cmd->pdid_cqecnt = cq->size / sizeof(struct ocrdma_mcqe);
524 
525 	ocrdma_build_q_pages(&cmd->pa[0], cq->size / OCRDMA_MIN_Q_PAGE_SIZE,
526 			     cq->dma, PAGE_SIZE_4K);
527 	status = be_roce_mcc_cmd(dev->nic_info.netdev,
528 				 cmd, sizeof(*cmd), NULL, NULL);
529 	if (!status) {
530 		cq->id = (u16) (rsp->cq_id & OCRDMA_CREATE_CQ_RSP_CQ_ID_MASK);
531 		cq->created = true;
532 	}
533 	return status;
534 }
535 
536 static u32 ocrdma_encoded_q_len(int q_len)
537 {
538 	u32 len_encoded = fls(q_len);	/* log2(len) + 1 */
539 
540 	if (len_encoded == 16)
541 		len_encoded = 0;
542 	return len_encoded;
543 }
544 
545 static int ocrdma_mbx_create_mq(struct ocrdma_dev *dev,
546 				struct ocrdma_queue_info *mq,
547 				struct ocrdma_queue_info *cq)
548 {
549 	int num_pages, status;
550 	struct ocrdma_create_mq_req *cmd = dev->mbx_cmd;
551 	struct ocrdma_create_mq_rsp *rsp = dev->mbx_cmd;
552 	struct ocrdma_pa *pa;
553 
554 	memset(cmd, 0, sizeof(*cmd));
555 	num_pages = PAGES_4K_SPANNED(mq->va, mq->size);
556 
557 	ocrdma_init_mch(&cmd->req, OCRDMA_CMD_CREATE_MQ_EXT,
558 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
559 	cmd->req.rsvd_version = 1;
560 	cmd->cqid_pages = num_pages;
561 	cmd->cqid_pages |= (cq->id << OCRDMA_CREATE_MQ_CQ_ID_SHIFT);
562 	cmd->async_cqid_valid = OCRDMA_CREATE_MQ_ASYNC_CQ_VALID;
563 
564 	cmd->async_event_bitmap = BIT(OCRDMA_ASYNC_GRP5_EVE_CODE);
565 	cmd->async_event_bitmap |= BIT(OCRDMA_ASYNC_RDMA_EVE_CODE);
566 
567 	cmd->async_cqid_ringsize = cq->id;
568 	cmd->async_cqid_ringsize |= (ocrdma_encoded_q_len(mq->len) <<
569 				OCRDMA_CREATE_MQ_RING_SIZE_SHIFT);
570 	cmd->valid = OCRDMA_CREATE_MQ_VALID;
571 	pa = &cmd->pa[0];
572 
573 	ocrdma_build_q_pages(pa, num_pages, mq->dma, PAGE_SIZE_4K);
574 	status = be_roce_mcc_cmd(dev->nic_info.netdev,
575 				 cmd, sizeof(*cmd), NULL, NULL);
576 	if (!status) {
577 		mq->id = rsp->id;
578 		mq->created = true;
579 	}
580 	return status;
581 }
582 
583 static int ocrdma_create_mq(struct ocrdma_dev *dev)
584 {
585 	int status;
586 
587 	/* Alloc completion queue for Mailbox queue */
588 	status = ocrdma_alloc_q(dev, &dev->mq.cq, OCRDMA_MQ_CQ_LEN,
589 				sizeof(struct ocrdma_mcqe));
590 	if (status)
591 		goto alloc_err;
592 
593 	dev->eq_tbl[0].cq_cnt++;
594 	status = ocrdma_mbx_mq_cq_create(dev, &dev->mq.cq, &dev->eq_tbl[0].q);
595 	if (status)
596 		goto mbx_cq_free;
597 
598 	memset(&dev->mqe_ctx, 0, sizeof(dev->mqe_ctx));
599 	init_waitqueue_head(&dev->mqe_ctx.cmd_wait);
600 	mutex_init(&dev->mqe_ctx.lock);
601 
602 	/* Alloc Mailbox queue */
603 	status = ocrdma_alloc_q(dev, &dev->mq.sq, OCRDMA_MQ_LEN,
604 				sizeof(struct ocrdma_mqe));
605 	if (status)
606 		goto mbx_cq_destroy;
607 	status = ocrdma_mbx_create_mq(dev, &dev->mq.sq, &dev->mq.cq);
608 	if (status)
609 		goto mbx_q_free;
610 	ocrdma_ring_cq_db(dev, dev->mq.cq.id, true, false, 0);
611 	return 0;
612 
613 mbx_q_free:
614 	ocrdma_free_q(dev, &dev->mq.sq);
615 mbx_cq_destroy:
616 	ocrdma_mbx_delete_q(dev, &dev->mq.cq, QTYPE_CQ);
617 mbx_cq_free:
618 	ocrdma_free_q(dev, &dev->mq.cq);
619 alloc_err:
620 	return status;
621 }
622 
623 static void ocrdma_destroy_mq(struct ocrdma_dev *dev)
624 {
625 	struct ocrdma_queue_info *mbxq, *cq;
626 
627 	/* mqe_ctx lock synchronizes with any other pending cmds. */
628 	mutex_lock(&dev->mqe_ctx.lock);
629 	mbxq = &dev->mq.sq;
630 	if (mbxq->created) {
631 		ocrdma_mbx_delete_q(dev, mbxq, QTYPE_MCCQ);
632 		ocrdma_free_q(dev, mbxq);
633 	}
634 	mutex_unlock(&dev->mqe_ctx.lock);
635 
636 	cq = &dev->mq.cq;
637 	if (cq->created) {
638 		ocrdma_mbx_delete_q(dev, cq, QTYPE_CQ);
639 		ocrdma_free_q(dev, cq);
640 	}
641 }
642 
643 static void ocrdma_process_qpcat_error(struct ocrdma_dev *dev,
644 				       struct ocrdma_qp *qp)
645 {
646 	enum ib_qp_state new_ib_qps = IB_QPS_ERR;
647 	enum ib_qp_state old_ib_qps;
648 
649 	if (qp == NULL)
650 		BUG();
651 	ocrdma_qp_state_change(qp, new_ib_qps, &old_ib_qps);
652 }
653 
654 static void ocrdma_dispatch_ibevent(struct ocrdma_dev *dev,
655 				    struct ocrdma_ae_mcqe *cqe)
656 {
657 	struct ocrdma_qp *qp = NULL;
658 	struct ocrdma_cq *cq = NULL;
659 	struct ib_event ib_evt;
660 	int cq_event = 0;
661 	int qp_event = 1;
662 	int srq_event = 0;
663 	int dev_event = 0;
664 	int type = (cqe->valid_ae_event & OCRDMA_AE_MCQE_EVENT_TYPE_MASK) >>
665 	    OCRDMA_AE_MCQE_EVENT_TYPE_SHIFT;
666 
667 	if (cqe->qpvalid_qpid & OCRDMA_AE_MCQE_QPVALID)
668 		qp = dev->qp_tbl[cqe->qpvalid_qpid & OCRDMA_AE_MCQE_QPID_MASK];
669 	if (cqe->cqvalid_cqid & OCRDMA_AE_MCQE_CQVALID)
670 		cq = dev->cq_tbl[cqe->cqvalid_cqid & OCRDMA_AE_MCQE_CQID_MASK];
671 
672 	memset(&ib_evt, 0, sizeof(ib_evt));
673 
674 	ib_evt.device = &dev->ibdev;
675 
676 	switch (type) {
677 	case OCRDMA_CQ_ERROR:
678 		ib_evt.element.cq = &cq->ibcq;
679 		ib_evt.event = IB_EVENT_CQ_ERR;
680 		cq_event = 1;
681 		qp_event = 0;
682 		break;
683 	case OCRDMA_CQ_OVERRUN_ERROR:
684 		ib_evt.element.cq = &cq->ibcq;
685 		ib_evt.event = IB_EVENT_CQ_ERR;
686 		cq_event = 1;
687 		qp_event = 0;
688 		break;
689 	case OCRDMA_CQ_QPCAT_ERROR:
690 		ib_evt.element.qp = &qp->ibqp;
691 		ib_evt.event = IB_EVENT_QP_FATAL;
692 		ocrdma_process_qpcat_error(dev, qp);
693 		break;
694 	case OCRDMA_QP_ACCESS_ERROR:
695 		ib_evt.element.qp = &qp->ibqp;
696 		ib_evt.event = IB_EVENT_QP_ACCESS_ERR;
697 		break;
698 	case OCRDMA_QP_COMM_EST_EVENT:
699 		ib_evt.element.qp = &qp->ibqp;
700 		ib_evt.event = IB_EVENT_COMM_EST;
701 		break;
702 	case OCRDMA_SQ_DRAINED_EVENT:
703 		ib_evt.element.qp = &qp->ibqp;
704 		ib_evt.event = IB_EVENT_SQ_DRAINED;
705 		break;
706 	case OCRDMA_DEVICE_FATAL_EVENT:
707 		ib_evt.element.port_num = 1;
708 		ib_evt.event = IB_EVENT_DEVICE_FATAL;
709 		qp_event = 0;
710 		dev_event = 1;
711 		break;
712 	case OCRDMA_SRQCAT_ERROR:
713 		ib_evt.element.srq = &qp->srq->ibsrq;
714 		ib_evt.event = IB_EVENT_SRQ_ERR;
715 		srq_event = 1;
716 		qp_event = 0;
717 		break;
718 	case OCRDMA_SRQ_LIMIT_EVENT:
719 		ib_evt.element.srq = &qp->srq->ibsrq;
720 		ib_evt.event = IB_EVENT_SRQ_LIMIT_REACHED;
721 		srq_event = 1;
722 		qp_event = 0;
723 		break;
724 	case OCRDMA_QP_LAST_WQE_EVENT:
725 		ib_evt.element.qp = &qp->ibqp;
726 		ib_evt.event = IB_EVENT_QP_LAST_WQE_REACHED;
727 		break;
728 	default:
729 		cq_event = 0;
730 		qp_event = 0;
731 		srq_event = 0;
732 		dev_event = 0;
733 		pr_err("%s() unknown type=0x%x\n", __func__, type);
734 		break;
735 	}
736 
737 	if (type < OCRDMA_MAX_ASYNC_ERRORS)
738 		atomic_inc(&dev->async_err_stats[type]);
739 
740 	if (qp_event) {
741 		if (qp->ibqp.event_handler)
742 			qp->ibqp.event_handler(&ib_evt, qp->ibqp.qp_context);
743 	} else if (cq_event) {
744 		if (cq->ibcq.event_handler)
745 			cq->ibcq.event_handler(&ib_evt, cq->ibcq.cq_context);
746 	} else if (srq_event) {
747 		if (qp->srq->ibsrq.event_handler)
748 			qp->srq->ibsrq.event_handler(&ib_evt,
749 						     qp->srq->ibsrq.
750 						     srq_context);
751 	} else if (dev_event) {
752 		pr_err("%s: Fatal event received\n", dev->ibdev.name);
753 		ib_dispatch_event(&ib_evt);
754 	}
755 
756 }
757 
758 static void ocrdma_process_grp5_aync(struct ocrdma_dev *dev,
759 					struct ocrdma_ae_mcqe *cqe)
760 {
761 	struct ocrdma_ae_pvid_mcqe *evt;
762 	int type = (cqe->valid_ae_event & OCRDMA_AE_MCQE_EVENT_TYPE_MASK) >>
763 			OCRDMA_AE_MCQE_EVENT_TYPE_SHIFT;
764 
765 	switch (type) {
766 	case OCRDMA_ASYNC_EVENT_PVID_STATE:
767 		evt = (struct ocrdma_ae_pvid_mcqe *)cqe;
768 		if ((evt->tag_enabled & OCRDMA_AE_PVID_MCQE_ENABLED_MASK) >>
769 			OCRDMA_AE_PVID_MCQE_ENABLED_SHIFT)
770 			dev->pvid = ((evt->tag_enabled &
771 					OCRDMA_AE_PVID_MCQE_TAG_MASK) >>
772 					OCRDMA_AE_PVID_MCQE_TAG_SHIFT);
773 		break;
774 
775 	case OCRDMA_ASYNC_EVENT_COS_VALUE:
776 		atomic_set(&dev->update_sl, 1);
777 		break;
778 	default:
779 		/* Not interested evts. */
780 		break;
781 	}
782 }
783 
784 static void ocrdma_process_acqe(struct ocrdma_dev *dev, void *ae_cqe)
785 {
786 	/* async CQE processing */
787 	struct ocrdma_ae_mcqe *cqe = ae_cqe;
788 	u32 evt_code = (cqe->valid_ae_event & OCRDMA_AE_MCQE_EVENT_CODE_MASK) >>
789 			OCRDMA_AE_MCQE_EVENT_CODE_SHIFT;
790 
791 	if (evt_code == OCRDMA_ASYNC_RDMA_EVE_CODE)
792 		ocrdma_dispatch_ibevent(dev, cqe);
793 	else if (evt_code == OCRDMA_ASYNC_GRP5_EVE_CODE)
794 		ocrdma_process_grp5_aync(dev, cqe);
795 	else
796 		pr_err("%s(%d) invalid evt code=0x%x\n", __func__,
797 		       dev->id, evt_code);
798 }
799 
800 static void ocrdma_process_mcqe(struct ocrdma_dev *dev, struct ocrdma_mcqe *cqe)
801 {
802 	if (dev->mqe_ctx.tag == cqe->tag_lo && dev->mqe_ctx.cmd_done == false) {
803 		dev->mqe_ctx.cqe_status = (cqe->status &
804 		     OCRDMA_MCQE_STATUS_MASK) >> OCRDMA_MCQE_STATUS_SHIFT;
805 		dev->mqe_ctx.ext_status =
806 		    (cqe->status & OCRDMA_MCQE_ESTATUS_MASK)
807 		    >> OCRDMA_MCQE_ESTATUS_SHIFT;
808 		dev->mqe_ctx.cmd_done = true;
809 		wake_up(&dev->mqe_ctx.cmd_wait);
810 	} else
811 		pr_err("%s() cqe for invalid tag0x%x.expected=0x%x\n",
812 		       __func__, cqe->tag_lo, dev->mqe_ctx.tag);
813 }
814 
815 static int ocrdma_mq_cq_handler(struct ocrdma_dev *dev, u16 cq_id)
816 {
817 	u16 cqe_popped = 0;
818 	struct ocrdma_mcqe *cqe;
819 
820 	while (1) {
821 		cqe = ocrdma_get_mcqe(dev);
822 		if (cqe == NULL)
823 			break;
824 		ocrdma_le32_to_cpu(cqe, sizeof(*cqe));
825 		cqe_popped += 1;
826 		if (cqe->valid_ae_cmpl_cons & OCRDMA_MCQE_AE_MASK)
827 			ocrdma_process_acqe(dev, cqe);
828 		else if (cqe->valid_ae_cmpl_cons & OCRDMA_MCQE_CMPL_MASK)
829 			ocrdma_process_mcqe(dev, cqe);
830 		memset(cqe, 0, sizeof(struct ocrdma_mcqe));
831 		ocrdma_mcq_inc_tail(dev);
832 	}
833 	ocrdma_ring_cq_db(dev, dev->mq.cq.id, true, false, cqe_popped);
834 	return 0;
835 }
836 
837 static struct ocrdma_cq *_ocrdma_qp_buddy_cq_handler(struct ocrdma_dev *dev,
838 				struct ocrdma_cq *cq, bool sq)
839 {
840 	struct ocrdma_qp *qp;
841 	struct list_head *cur;
842 	struct ocrdma_cq *bcq = NULL;
843 	struct list_head *head = sq?(&cq->sq_head):(&cq->rq_head);
844 
845 	list_for_each(cur, head) {
846 		if (sq)
847 			qp = list_entry(cur, struct ocrdma_qp, sq_entry);
848 		else
849 			qp = list_entry(cur, struct ocrdma_qp, rq_entry);
850 
851 		if (qp->srq)
852 			continue;
853 		/* if wq and rq share the same cq, than comp_handler
854 		 * is already invoked.
855 		 */
856 		if (qp->sq_cq == qp->rq_cq)
857 			continue;
858 		/* if completion came on sq, rq's cq is buddy cq.
859 		 * if completion came on rq, sq's cq is buddy cq.
860 		 */
861 		if (qp->sq_cq == cq)
862 			bcq = qp->rq_cq;
863 		else
864 			bcq = qp->sq_cq;
865 		return bcq;
866 	}
867 	return NULL;
868 }
869 
870 static void ocrdma_qp_buddy_cq_handler(struct ocrdma_dev *dev,
871 				       struct ocrdma_cq *cq)
872 {
873 	unsigned long flags;
874 	struct ocrdma_cq *bcq = NULL;
875 
876 	/* Go through list of QPs in error state which are using this CQ
877 	 * and invoke its callback handler to trigger CQE processing for
878 	 * error/flushed CQE. It is rare to find more than few entries in
879 	 * this list as most consumers stops after getting error CQE.
880 	 * List is traversed only once when a matching buddy cq found for a QP.
881 	 */
882 	spin_lock_irqsave(&dev->flush_q_lock, flags);
883 	/* Check if buddy CQ is present.
884 	 * true - Check for  SQ CQ
885 	 * false - Check for RQ CQ
886 	 */
887 	bcq = _ocrdma_qp_buddy_cq_handler(dev, cq, true);
888 	if (bcq == NULL)
889 		bcq = _ocrdma_qp_buddy_cq_handler(dev, cq, false);
890 	spin_unlock_irqrestore(&dev->flush_q_lock, flags);
891 
892 	/* if there is valid buddy cq, look for its completion handler */
893 	if (bcq && bcq->ibcq.comp_handler) {
894 		spin_lock_irqsave(&bcq->comp_handler_lock, flags);
895 		(*bcq->ibcq.comp_handler) (&bcq->ibcq, bcq->ibcq.cq_context);
896 		spin_unlock_irqrestore(&bcq->comp_handler_lock, flags);
897 	}
898 }
899 
900 static void ocrdma_qp_cq_handler(struct ocrdma_dev *dev, u16 cq_idx)
901 {
902 	unsigned long flags;
903 	struct ocrdma_cq *cq;
904 
905 	if (cq_idx >= OCRDMA_MAX_CQ)
906 		BUG();
907 
908 	cq = dev->cq_tbl[cq_idx];
909 	if (cq == NULL)
910 		return;
911 
912 	if (cq->ibcq.comp_handler) {
913 		spin_lock_irqsave(&cq->comp_handler_lock, flags);
914 		(*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context);
915 		spin_unlock_irqrestore(&cq->comp_handler_lock, flags);
916 	}
917 	ocrdma_qp_buddy_cq_handler(dev, cq);
918 }
919 
920 static void ocrdma_cq_handler(struct ocrdma_dev *dev, u16 cq_id)
921 {
922 	/* process the MQ-CQE. */
923 	if (cq_id == dev->mq.cq.id)
924 		ocrdma_mq_cq_handler(dev, cq_id);
925 	else
926 		ocrdma_qp_cq_handler(dev, cq_id);
927 }
928 
929 static irqreturn_t ocrdma_irq_handler(int irq, void *handle)
930 {
931 	struct ocrdma_eq *eq = handle;
932 	struct ocrdma_dev *dev = eq->dev;
933 	struct ocrdma_eqe eqe;
934 	struct ocrdma_eqe *ptr;
935 	u16 cq_id;
936 	int budget = eq->cq_cnt;
937 
938 	do {
939 		ptr = ocrdma_get_eqe(eq);
940 		eqe = *ptr;
941 		ocrdma_le32_to_cpu(&eqe, sizeof(eqe));
942 		if ((eqe.id_valid & OCRDMA_EQE_VALID_MASK) == 0)
943 			break;
944 
945 		ptr->id_valid = 0;
946 		/* ring eq doorbell as soon as its consumed. */
947 		ocrdma_ring_eq_db(dev, eq->q.id, false, true, 1);
948 		/* check whether its CQE or not. */
949 		if ((eqe.id_valid & OCRDMA_EQE_FOR_CQE_MASK) == 0) {
950 			cq_id = eqe.id_valid >> OCRDMA_EQE_RESOURCE_ID_SHIFT;
951 			ocrdma_cq_handler(dev, cq_id);
952 		}
953 		ocrdma_eq_inc_tail(eq);
954 
955 		/* There can be a stale EQE after the last bound CQ is
956 		 * destroyed. EQE valid and budget == 0 implies this.
957 		 */
958 		if (budget)
959 			budget--;
960 
961 	} while (budget);
962 
963 	eq->aic_obj.eq_intr_cnt++;
964 	ocrdma_ring_eq_db(dev, eq->q.id, true, true, 0);
965 	return IRQ_HANDLED;
966 }
967 
968 static void ocrdma_post_mqe(struct ocrdma_dev *dev, struct ocrdma_mqe *cmd)
969 {
970 	struct ocrdma_mqe *mqe;
971 
972 	dev->mqe_ctx.tag = dev->mq.sq.head;
973 	dev->mqe_ctx.cmd_done = false;
974 	mqe = ocrdma_get_mqe(dev);
975 	cmd->hdr.tag_lo = dev->mq.sq.head;
976 	ocrdma_copy_cpu_to_le32(mqe, cmd, sizeof(*mqe));
977 	/* make sure descriptor is written before ringing doorbell */
978 	wmb();
979 	ocrdma_mq_inc_head(dev);
980 	ocrdma_ring_mq_db(dev);
981 }
982 
983 static int ocrdma_wait_mqe_cmpl(struct ocrdma_dev *dev)
984 {
985 	long status;
986 	/* 30 sec timeout */
987 	status = wait_event_timeout(dev->mqe_ctx.cmd_wait,
988 				    (dev->mqe_ctx.cmd_done != false),
989 				    msecs_to_jiffies(30000));
990 	if (status)
991 		return 0;
992 	else {
993 		dev->mqe_ctx.fw_error_state = true;
994 		pr_err("%s(%d) mailbox timeout: fw not responding\n",
995 		       __func__, dev->id);
996 		return -1;
997 	}
998 }
999 
1000 /* issue a mailbox command on the MQ */
1001 static int ocrdma_mbx_cmd(struct ocrdma_dev *dev, struct ocrdma_mqe *mqe)
1002 {
1003 	int status = 0;
1004 	u16 cqe_status, ext_status;
1005 	struct ocrdma_mqe *rsp_mqe;
1006 	struct ocrdma_mbx_rsp *rsp = NULL;
1007 
1008 	mutex_lock(&dev->mqe_ctx.lock);
1009 	if (dev->mqe_ctx.fw_error_state)
1010 		goto mbx_err;
1011 	ocrdma_post_mqe(dev, mqe);
1012 	status = ocrdma_wait_mqe_cmpl(dev);
1013 	if (status)
1014 		goto mbx_err;
1015 	cqe_status = dev->mqe_ctx.cqe_status;
1016 	ext_status = dev->mqe_ctx.ext_status;
1017 	rsp_mqe = ocrdma_get_mqe_rsp(dev);
1018 	ocrdma_copy_le32_to_cpu(mqe, rsp_mqe, (sizeof(*mqe)));
1019 	if ((mqe->hdr.spcl_sge_cnt_emb & OCRDMA_MQE_HDR_EMB_MASK) >>
1020 				OCRDMA_MQE_HDR_EMB_SHIFT)
1021 		rsp = &mqe->u.rsp;
1022 
1023 	if (cqe_status || ext_status) {
1024 		pr_err("%s() cqe_status=0x%x, ext_status=0x%x,",
1025 		       __func__, cqe_status, ext_status);
1026 		if (rsp) {
1027 			/* This is for embedded cmds. */
1028 			pr_err("opcode=0x%x, subsystem=0x%x\n",
1029 			       (rsp->subsys_op & OCRDMA_MBX_RSP_OPCODE_MASK) >>
1030 				OCRDMA_MBX_RSP_OPCODE_SHIFT,
1031 				(rsp->subsys_op & OCRDMA_MBX_RSP_SUBSYS_MASK) >>
1032 				OCRDMA_MBX_RSP_SUBSYS_SHIFT);
1033 		}
1034 		status = ocrdma_get_mbx_cqe_errno(cqe_status);
1035 		goto mbx_err;
1036 	}
1037 	/* For non embedded, rsp errors are handled in ocrdma_nonemb_mbx_cmd */
1038 	if (rsp && (mqe->u.rsp.status & OCRDMA_MBX_RSP_STATUS_MASK))
1039 		status = ocrdma_get_mbx_errno(mqe->u.rsp.status);
1040 mbx_err:
1041 	mutex_unlock(&dev->mqe_ctx.lock);
1042 	return status;
1043 }
1044 
1045 static int ocrdma_nonemb_mbx_cmd(struct ocrdma_dev *dev, struct ocrdma_mqe *mqe,
1046 				 void *payload_va)
1047 {
1048 	int status = 0;
1049 	struct ocrdma_mbx_rsp *rsp = payload_va;
1050 
1051 	if ((mqe->hdr.spcl_sge_cnt_emb & OCRDMA_MQE_HDR_EMB_MASK) >>
1052 				OCRDMA_MQE_HDR_EMB_SHIFT)
1053 		BUG();
1054 
1055 	status = ocrdma_mbx_cmd(dev, mqe);
1056 	if (!status)
1057 		/* For non embedded, only CQE failures are handled in
1058 		 * ocrdma_mbx_cmd. We need to check for RSP errors.
1059 		 */
1060 		if (rsp->status & OCRDMA_MBX_RSP_STATUS_MASK)
1061 			status = ocrdma_get_mbx_errno(rsp->status);
1062 
1063 	if (status)
1064 		pr_err("opcode=0x%x, subsystem=0x%x\n",
1065 		       (rsp->subsys_op & OCRDMA_MBX_RSP_OPCODE_MASK) >>
1066 			OCRDMA_MBX_RSP_OPCODE_SHIFT,
1067 			(rsp->subsys_op & OCRDMA_MBX_RSP_SUBSYS_MASK) >>
1068 			OCRDMA_MBX_RSP_SUBSYS_SHIFT);
1069 	return status;
1070 }
1071 
1072 static void ocrdma_get_attr(struct ocrdma_dev *dev,
1073 			      struct ocrdma_dev_attr *attr,
1074 			      struct ocrdma_mbx_query_config *rsp)
1075 {
1076 	attr->max_pd =
1077 	    (rsp->max_pd_ca_ack_delay & OCRDMA_MBX_QUERY_CFG_MAX_PD_MASK) >>
1078 	    OCRDMA_MBX_QUERY_CFG_MAX_PD_SHIFT;
1079 	attr->max_dpp_pds =
1080 	   (rsp->max_dpp_pds_credits & OCRDMA_MBX_QUERY_CFG_MAX_DPP_PDS_MASK) >>
1081 	    OCRDMA_MBX_QUERY_CFG_MAX_DPP_PDS_OFFSET;
1082 	attr->max_qp =
1083 	    (rsp->qp_srq_cq_ird_ord & OCRDMA_MBX_QUERY_CFG_MAX_QP_MASK) >>
1084 	    OCRDMA_MBX_QUERY_CFG_MAX_QP_SHIFT;
1085 	attr->max_srq =
1086 		(rsp->max_srq_rpir_qps & OCRDMA_MBX_QUERY_CFG_MAX_SRQ_MASK) >>
1087 		OCRDMA_MBX_QUERY_CFG_MAX_SRQ_OFFSET;
1088 	attr->max_send_sge = ((rsp->max_write_send_sge &
1089 			       OCRDMA_MBX_QUERY_CFG_MAX_SEND_SGE_MASK) >>
1090 			      OCRDMA_MBX_QUERY_CFG_MAX_SEND_SGE_SHIFT);
1091 	attr->max_recv_sge = (rsp->max_write_send_sge &
1092 			      OCRDMA_MBX_QUERY_CFG_MAX_SEND_SGE_MASK) >>
1093 	    OCRDMA_MBX_QUERY_CFG_MAX_SEND_SGE_SHIFT;
1094 	attr->max_srq_sge = (rsp->max_srq_rqe_sge &
1095 			      OCRDMA_MBX_QUERY_CFG_MAX_SRQ_SGE_MASK) >>
1096 	    OCRDMA_MBX_QUERY_CFG_MAX_SRQ_SGE_OFFSET;
1097 	attr->max_rdma_sge = (rsp->max_write_send_sge &
1098 			      OCRDMA_MBX_QUERY_CFG_MAX_WRITE_SGE_MASK) >>
1099 	    OCRDMA_MBX_QUERY_CFG_MAX_WRITE_SGE_SHIFT;
1100 	attr->max_ord_per_qp = (rsp->max_ird_ord_per_qp &
1101 				OCRDMA_MBX_QUERY_CFG_MAX_ORD_PER_QP_MASK) >>
1102 	    OCRDMA_MBX_QUERY_CFG_MAX_ORD_PER_QP_SHIFT;
1103 	attr->max_ird_per_qp = (rsp->max_ird_ord_per_qp &
1104 				OCRDMA_MBX_QUERY_CFG_MAX_IRD_PER_QP_MASK) >>
1105 	    OCRDMA_MBX_QUERY_CFG_MAX_IRD_PER_QP_SHIFT;
1106 	attr->cq_overflow_detect = (rsp->qp_srq_cq_ird_ord &
1107 				    OCRDMA_MBX_QUERY_CFG_CQ_OVERFLOW_MASK) >>
1108 	    OCRDMA_MBX_QUERY_CFG_CQ_OVERFLOW_SHIFT;
1109 	attr->srq_supported = (rsp->qp_srq_cq_ird_ord &
1110 			       OCRDMA_MBX_QUERY_CFG_SRQ_SUPPORTED_MASK) >>
1111 	    OCRDMA_MBX_QUERY_CFG_SRQ_SUPPORTED_SHIFT;
1112 	attr->local_ca_ack_delay = (rsp->max_pd_ca_ack_delay &
1113 				    OCRDMA_MBX_QUERY_CFG_CA_ACK_DELAY_MASK) >>
1114 	    OCRDMA_MBX_QUERY_CFG_CA_ACK_DELAY_SHIFT;
1115 	attr->max_mw = rsp->max_mw;
1116 	attr->max_mr = rsp->max_mr;
1117 	attr->max_mr_size = ((u64)rsp->max_mr_size_hi << 32) |
1118 			      rsp->max_mr_size_lo;
1119 	attr->max_fmr = 0;
1120 	attr->max_pages_per_frmr = rsp->max_pages_per_frmr;
1121 	attr->max_num_mr_pbl = rsp->max_num_mr_pbl;
1122 	attr->max_cqe = rsp->max_cq_cqes_per_cq &
1123 			OCRDMA_MBX_QUERY_CFG_MAX_CQES_PER_CQ_MASK;
1124 	attr->max_cq = (rsp->max_cq_cqes_per_cq &
1125 			OCRDMA_MBX_QUERY_CFG_MAX_CQ_MASK) >>
1126 			OCRDMA_MBX_QUERY_CFG_MAX_CQ_OFFSET;
1127 	attr->wqe_size = ((rsp->wqe_rqe_stride_max_dpp_cqs &
1128 		OCRDMA_MBX_QUERY_CFG_MAX_WQE_SIZE_MASK) >>
1129 		OCRDMA_MBX_QUERY_CFG_MAX_WQE_SIZE_OFFSET) *
1130 		OCRDMA_WQE_STRIDE;
1131 	attr->rqe_size = ((rsp->wqe_rqe_stride_max_dpp_cqs &
1132 		OCRDMA_MBX_QUERY_CFG_MAX_RQE_SIZE_MASK) >>
1133 		OCRDMA_MBX_QUERY_CFG_MAX_RQE_SIZE_OFFSET) *
1134 		OCRDMA_WQE_STRIDE;
1135 	attr->max_inline_data =
1136 	    attr->wqe_size - (sizeof(struct ocrdma_hdr_wqe) +
1137 			      sizeof(struct ocrdma_sge));
1138 	if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) {
1139 		attr->ird = 1;
1140 		attr->ird_page_size = OCRDMA_MIN_Q_PAGE_SIZE;
1141 		attr->num_ird_pages = MAX_OCRDMA_IRD_PAGES;
1142 	}
1143 	dev->attr.max_wqe = rsp->max_wqes_rqes_per_q >>
1144 		 OCRDMA_MBX_QUERY_CFG_MAX_WQES_PER_WQ_OFFSET;
1145 	dev->attr.max_rqe = rsp->max_wqes_rqes_per_q &
1146 		OCRDMA_MBX_QUERY_CFG_MAX_RQES_PER_RQ_MASK;
1147 }
1148 
1149 static int ocrdma_check_fw_config(struct ocrdma_dev *dev,
1150 				   struct ocrdma_fw_conf_rsp *conf)
1151 {
1152 	u32 fn_mode;
1153 
1154 	fn_mode = conf->fn_mode & OCRDMA_FN_MODE_RDMA;
1155 	if (fn_mode != OCRDMA_FN_MODE_RDMA)
1156 		return -EINVAL;
1157 	dev->base_eqid = conf->base_eqid;
1158 	dev->max_eq = conf->max_eq;
1159 	return 0;
1160 }
1161 
1162 /* can be issued only during init time. */
1163 static int ocrdma_mbx_query_fw_ver(struct ocrdma_dev *dev)
1164 {
1165 	int status = -ENOMEM;
1166 	struct ocrdma_mqe *cmd;
1167 	struct ocrdma_fw_ver_rsp *rsp;
1168 
1169 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_GET_FW_VER, sizeof(*cmd));
1170 	if (!cmd)
1171 		return -ENOMEM;
1172 	ocrdma_init_mch((struct ocrdma_mbx_hdr *)&cmd->u.cmd[0],
1173 			OCRDMA_CMD_GET_FW_VER,
1174 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
1175 
1176 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1177 	if (status)
1178 		goto mbx_err;
1179 	rsp = (struct ocrdma_fw_ver_rsp *)cmd;
1180 	memset(&dev->attr.fw_ver[0], 0, sizeof(dev->attr.fw_ver));
1181 	memcpy(&dev->attr.fw_ver[0], &rsp->running_ver[0],
1182 	       sizeof(rsp->running_ver));
1183 	ocrdma_le32_to_cpu(dev->attr.fw_ver, sizeof(rsp->running_ver));
1184 mbx_err:
1185 	kfree(cmd);
1186 	return status;
1187 }
1188 
1189 /* can be issued only during init time. */
1190 static int ocrdma_mbx_query_fw_config(struct ocrdma_dev *dev)
1191 {
1192 	int status = -ENOMEM;
1193 	struct ocrdma_mqe *cmd;
1194 	struct ocrdma_fw_conf_rsp *rsp;
1195 
1196 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_GET_FW_CONFIG, sizeof(*cmd));
1197 	if (!cmd)
1198 		return -ENOMEM;
1199 	ocrdma_init_mch((struct ocrdma_mbx_hdr *)&cmd->u.cmd[0],
1200 			OCRDMA_CMD_GET_FW_CONFIG,
1201 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
1202 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1203 	if (status)
1204 		goto mbx_err;
1205 	rsp = (struct ocrdma_fw_conf_rsp *)cmd;
1206 	status = ocrdma_check_fw_config(dev, rsp);
1207 mbx_err:
1208 	kfree(cmd);
1209 	return status;
1210 }
1211 
1212 int ocrdma_mbx_rdma_stats(struct ocrdma_dev *dev, bool reset)
1213 {
1214 	struct ocrdma_rdma_stats_req *req = dev->stats_mem.va;
1215 	struct ocrdma_mqe *mqe = &dev->stats_mem.mqe;
1216 	struct ocrdma_rdma_stats_resp *old_stats;
1217 	int status;
1218 
1219 	old_stats = kmalloc(sizeof(*old_stats), GFP_KERNEL);
1220 	if (old_stats == NULL)
1221 		return -ENOMEM;
1222 
1223 	memset(mqe, 0, sizeof(*mqe));
1224 	mqe->hdr.pyld_len = dev->stats_mem.size;
1225 	mqe->hdr.spcl_sge_cnt_emb |=
1226 			(1 << OCRDMA_MQE_HDR_SGE_CNT_SHIFT) &
1227 				OCRDMA_MQE_HDR_SGE_CNT_MASK;
1228 	mqe->u.nonemb_req.sge[0].pa_lo = (u32) (dev->stats_mem.pa & 0xffffffff);
1229 	mqe->u.nonemb_req.sge[0].pa_hi = (u32) upper_32_bits(dev->stats_mem.pa);
1230 	mqe->u.nonemb_req.sge[0].len = dev->stats_mem.size;
1231 
1232 	/* Cache the old stats */
1233 	memcpy(old_stats, req, sizeof(struct ocrdma_rdma_stats_resp));
1234 	memset(req, 0, dev->stats_mem.size);
1235 
1236 	ocrdma_init_mch((struct ocrdma_mbx_hdr *)req,
1237 			OCRDMA_CMD_GET_RDMA_STATS,
1238 			OCRDMA_SUBSYS_ROCE,
1239 			dev->stats_mem.size);
1240 	if (reset)
1241 		req->reset_stats = reset;
1242 
1243 	status = ocrdma_nonemb_mbx_cmd(dev, mqe, dev->stats_mem.va);
1244 	if (status)
1245 		/* Copy from cache, if mbox fails */
1246 		memcpy(req, old_stats, sizeof(struct ocrdma_rdma_stats_resp));
1247 	else
1248 		ocrdma_le32_to_cpu(req, dev->stats_mem.size);
1249 
1250 	kfree(old_stats);
1251 	return status;
1252 }
1253 
1254 static int ocrdma_mbx_get_ctrl_attribs(struct ocrdma_dev *dev)
1255 {
1256 	int status = -ENOMEM;
1257 	struct ocrdma_dma_mem dma;
1258 	struct ocrdma_mqe *mqe;
1259 	struct ocrdma_get_ctrl_attribs_rsp *ctrl_attr_rsp;
1260 	struct mgmt_hba_attribs *hba_attribs;
1261 
1262 	mqe = kzalloc(sizeof(struct ocrdma_mqe), GFP_KERNEL);
1263 	if (!mqe)
1264 		return status;
1265 
1266 	dma.size = sizeof(struct ocrdma_get_ctrl_attribs_rsp);
1267 	dma.va	 = dma_alloc_coherent(&dev->nic_info.pdev->dev,
1268 					dma.size, &dma.pa, GFP_KERNEL);
1269 	if (!dma.va)
1270 		goto free_mqe;
1271 
1272 	mqe->hdr.pyld_len = dma.size;
1273 	mqe->hdr.spcl_sge_cnt_emb |=
1274 			(1 << OCRDMA_MQE_HDR_SGE_CNT_SHIFT) &
1275 			OCRDMA_MQE_HDR_SGE_CNT_MASK;
1276 	mqe->u.nonemb_req.sge[0].pa_lo = (u32) (dma.pa & 0xffffffff);
1277 	mqe->u.nonemb_req.sge[0].pa_hi = (u32) upper_32_bits(dma.pa);
1278 	mqe->u.nonemb_req.sge[0].len = dma.size;
1279 
1280 	memset(dma.va, 0, dma.size);
1281 	ocrdma_init_mch((struct ocrdma_mbx_hdr *)dma.va,
1282 			OCRDMA_CMD_GET_CTRL_ATTRIBUTES,
1283 			OCRDMA_SUBSYS_COMMON,
1284 			dma.size);
1285 
1286 	status = ocrdma_nonemb_mbx_cmd(dev, mqe, dma.va);
1287 	if (!status) {
1288 		ctrl_attr_rsp = (struct ocrdma_get_ctrl_attribs_rsp *)dma.va;
1289 		hba_attribs = &ctrl_attr_rsp->ctrl_attribs.hba_attribs;
1290 
1291 		dev->hba_port_num = (hba_attribs->ptpnum_maxdoms_hbast_cv &
1292 					OCRDMA_HBA_ATTRB_PTNUM_MASK)
1293 					>> OCRDMA_HBA_ATTRB_PTNUM_SHIFT;
1294 		strncpy(dev->model_number,
1295 			hba_attribs->controller_model_number, 31);
1296 	}
1297 	dma_free_coherent(&dev->nic_info.pdev->dev, dma.size, dma.va, dma.pa);
1298 free_mqe:
1299 	kfree(mqe);
1300 	return status;
1301 }
1302 
1303 static int ocrdma_mbx_query_dev(struct ocrdma_dev *dev)
1304 {
1305 	int status = -ENOMEM;
1306 	struct ocrdma_mbx_query_config *rsp;
1307 	struct ocrdma_mqe *cmd;
1308 
1309 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_CONFIG, sizeof(*cmd));
1310 	if (!cmd)
1311 		return status;
1312 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1313 	if (status)
1314 		goto mbx_err;
1315 	rsp = (struct ocrdma_mbx_query_config *)cmd;
1316 	ocrdma_get_attr(dev, &dev->attr, rsp);
1317 mbx_err:
1318 	kfree(cmd);
1319 	return status;
1320 }
1321 
1322 int ocrdma_mbx_get_link_speed(struct ocrdma_dev *dev, u8 *lnk_speed)
1323 {
1324 	int status = -ENOMEM;
1325 	struct ocrdma_get_link_speed_rsp *rsp;
1326 	struct ocrdma_mqe *cmd;
1327 
1328 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_NTWK_LINK_CONFIG_V1,
1329 				  sizeof(*cmd));
1330 	if (!cmd)
1331 		return status;
1332 	ocrdma_init_mch((struct ocrdma_mbx_hdr *)&cmd->u.cmd[0],
1333 			OCRDMA_CMD_QUERY_NTWK_LINK_CONFIG_V1,
1334 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
1335 
1336 	((struct ocrdma_mbx_hdr *)cmd->u.cmd)->rsvd_version = 0x1;
1337 
1338 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1339 	if (status)
1340 		goto mbx_err;
1341 
1342 	rsp = (struct ocrdma_get_link_speed_rsp *)cmd;
1343 	*lnk_speed = (rsp->pflt_pps_ld_pnum & OCRDMA_PHY_PS_MASK)
1344 			>> OCRDMA_PHY_PS_SHIFT;
1345 
1346 mbx_err:
1347 	kfree(cmd);
1348 	return status;
1349 }
1350 
1351 static int ocrdma_mbx_get_phy_info(struct ocrdma_dev *dev)
1352 {
1353 	int status = -ENOMEM;
1354 	struct ocrdma_mqe *cmd;
1355 	struct ocrdma_get_phy_info_rsp *rsp;
1356 
1357 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_PHY_DETAILS, sizeof(*cmd));
1358 	if (!cmd)
1359 		return status;
1360 
1361 	ocrdma_init_mch((struct ocrdma_mbx_hdr *)&cmd->u.cmd[0],
1362 			OCRDMA_CMD_PHY_DETAILS, OCRDMA_SUBSYS_COMMON,
1363 			sizeof(*cmd));
1364 
1365 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1366 	if (status)
1367 		goto mbx_err;
1368 
1369 	rsp = (struct ocrdma_get_phy_info_rsp *)cmd;
1370 	dev->phy.phy_type =
1371 			(rsp->ityp_ptyp & OCRDMA_PHY_TYPE_MASK);
1372 	dev->phy.interface_type =
1373 			(rsp->ityp_ptyp & OCRDMA_IF_TYPE_MASK)
1374 				>> OCRDMA_IF_TYPE_SHIFT;
1375 	dev->phy.auto_speeds_supported  =
1376 			(rsp->fspeed_aspeed & OCRDMA_ASPEED_SUPP_MASK);
1377 	dev->phy.fixed_speeds_supported =
1378 			(rsp->fspeed_aspeed & OCRDMA_FSPEED_SUPP_MASK)
1379 				>> OCRDMA_FSPEED_SUPP_SHIFT;
1380 mbx_err:
1381 	kfree(cmd);
1382 	return status;
1383 }
1384 
1385 int ocrdma_mbx_alloc_pd(struct ocrdma_dev *dev, struct ocrdma_pd *pd)
1386 {
1387 	int status = -ENOMEM;
1388 	struct ocrdma_alloc_pd *cmd;
1389 	struct ocrdma_alloc_pd_rsp *rsp;
1390 
1391 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_ALLOC_PD, sizeof(*cmd));
1392 	if (!cmd)
1393 		return status;
1394 	if (pd->dpp_enabled)
1395 		cmd->enable_dpp_rsvd |= OCRDMA_ALLOC_PD_ENABLE_DPP;
1396 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1397 	if (status)
1398 		goto mbx_err;
1399 	rsp = (struct ocrdma_alloc_pd_rsp *)cmd;
1400 	pd->id = rsp->dpp_page_pdid & OCRDMA_ALLOC_PD_RSP_PDID_MASK;
1401 	if (rsp->dpp_page_pdid & OCRDMA_ALLOC_PD_RSP_DPP) {
1402 		pd->dpp_enabled = true;
1403 		pd->dpp_page = rsp->dpp_page_pdid >>
1404 				OCRDMA_ALLOC_PD_RSP_DPP_PAGE_SHIFT;
1405 	} else {
1406 		pd->dpp_enabled = false;
1407 		pd->num_dpp_qp = 0;
1408 	}
1409 mbx_err:
1410 	kfree(cmd);
1411 	return status;
1412 }
1413 
1414 int ocrdma_mbx_dealloc_pd(struct ocrdma_dev *dev, struct ocrdma_pd *pd)
1415 {
1416 	int status = -ENOMEM;
1417 	struct ocrdma_dealloc_pd *cmd;
1418 
1419 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DEALLOC_PD, sizeof(*cmd));
1420 	if (!cmd)
1421 		return status;
1422 	cmd->id = pd->id;
1423 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1424 	kfree(cmd);
1425 	return status;
1426 }
1427 
1428 
1429 static int ocrdma_mbx_alloc_pd_range(struct ocrdma_dev *dev)
1430 {
1431 	int status = -ENOMEM;
1432 	size_t pd_bitmap_size;
1433 	struct ocrdma_alloc_pd_range *cmd;
1434 	struct ocrdma_alloc_pd_range_rsp *rsp;
1435 
1436 	/* Pre allocate the DPP PDs */
1437 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_ALLOC_PD_RANGE, sizeof(*cmd));
1438 	if (!cmd)
1439 		return -ENOMEM;
1440 	cmd->pd_count = dev->attr.max_dpp_pds;
1441 	cmd->enable_dpp_rsvd |= OCRDMA_ALLOC_PD_ENABLE_DPP;
1442 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1443 	if (status)
1444 		goto mbx_err;
1445 	rsp = (struct ocrdma_alloc_pd_range_rsp *)cmd;
1446 
1447 	if ((rsp->dpp_page_pdid & OCRDMA_ALLOC_PD_RSP_DPP) && rsp->pd_count) {
1448 		dev->pd_mgr->dpp_page_index = rsp->dpp_page_pdid >>
1449 				OCRDMA_ALLOC_PD_RSP_DPP_PAGE_SHIFT;
1450 		dev->pd_mgr->pd_dpp_start = rsp->dpp_page_pdid &
1451 				OCRDMA_ALLOC_PD_RNG_RSP_START_PDID_MASK;
1452 		dev->pd_mgr->max_dpp_pd = rsp->pd_count;
1453 		pd_bitmap_size = BITS_TO_LONGS(rsp->pd_count) * sizeof(long);
1454 		dev->pd_mgr->pd_dpp_bitmap = kzalloc(pd_bitmap_size,
1455 						     GFP_KERNEL);
1456 	}
1457 	kfree(cmd);
1458 
1459 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_ALLOC_PD_RANGE, sizeof(*cmd));
1460 	if (!cmd)
1461 		return -ENOMEM;
1462 
1463 	cmd->pd_count = dev->attr.max_pd - dev->attr.max_dpp_pds;
1464 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1465 	if (status)
1466 		goto mbx_err;
1467 	rsp = (struct ocrdma_alloc_pd_range_rsp *)cmd;
1468 	if (rsp->pd_count) {
1469 		dev->pd_mgr->pd_norm_start = rsp->dpp_page_pdid &
1470 					OCRDMA_ALLOC_PD_RNG_RSP_START_PDID_MASK;
1471 		dev->pd_mgr->max_normal_pd = rsp->pd_count;
1472 		pd_bitmap_size = BITS_TO_LONGS(rsp->pd_count) * sizeof(long);
1473 		dev->pd_mgr->pd_norm_bitmap = kzalloc(pd_bitmap_size,
1474 						      GFP_KERNEL);
1475 	}
1476 
1477 	if (dev->pd_mgr->pd_norm_bitmap || dev->pd_mgr->pd_dpp_bitmap) {
1478 		/* Enable PD resource manager */
1479 		dev->pd_mgr->pd_prealloc_valid = true;
1480 	} else {
1481 		return -ENOMEM;
1482 	}
1483 mbx_err:
1484 	kfree(cmd);
1485 	return status;
1486 }
1487 
1488 static void ocrdma_mbx_dealloc_pd_range(struct ocrdma_dev *dev)
1489 {
1490 	struct ocrdma_dealloc_pd_range *cmd;
1491 
1492 	/* return normal PDs to firmware */
1493 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DEALLOC_PD_RANGE, sizeof(*cmd));
1494 	if (!cmd)
1495 		goto mbx_err;
1496 
1497 	if (dev->pd_mgr->max_normal_pd) {
1498 		cmd->start_pd_id = dev->pd_mgr->pd_norm_start;
1499 		cmd->pd_count = dev->pd_mgr->max_normal_pd;
1500 		ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1501 	}
1502 
1503 	if (dev->pd_mgr->max_dpp_pd) {
1504 		kfree(cmd);
1505 		/* return DPP PDs to firmware */
1506 		cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DEALLOC_PD_RANGE,
1507 					  sizeof(*cmd));
1508 		if (!cmd)
1509 			goto mbx_err;
1510 
1511 		cmd->start_pd_id = dev->pd_mgr->pd_dpp_start;
1512 		cmd->pd_count = dev->pd_mgr->max_dpp_pd;
1513 		ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1514 	}
1515 mbx_err:
1516 	kfree(cmd);
1517 }
1518 
1519 void ocrdma_alloc_pd_pool(struct ocrdma_dev *dev)
1520 {
1521 	int status;
1522 
1523 	dev->pd_mgr = kzalloc(sizeof(struct ocrdma_pd_resource_mgr),
1524 			      GFP_KERNEL);
1525 	if (!dev->pd_mgr) {
1526 		pr_err("%s(%d)Memory allocation failure.\n", __func__, dev->id);
1527 		return;
1528 	}
1529 	status = ocrdma_mbx_alloc_pd_range(dev);
1530 	if (status) {
1531 		pr_err("%s(%d) Unable to initialize PD pool, using default.\n",
1532 			 __func__, dev->id);
1533 	}
1534 }
1535 
1536 static void ocrdma_free_pd_pool(struct ocrdma_dev *dev)
1537 {
1538 	ocrdma_mbx_dealloc_pd_range(dev);
1539 	kfree(dev->pd_mgr->pd_norm_bitmap);
1540 	kfree(dev->pd_mgr->pd_dpp_bitmap);
1541 	kfree(dev->pd_mgr);
1542 }
1543 
1544 static int ocrdma_build_q_conf(u32 *num_entries, int entry_size,
1545 			       int *num_pages, int *page_size)
1546 {
1547 	int i;
1548 	int mem_size;
1549 
1550 	*num_entries = roundup_pow_of_two(*num_entries);
1551 	mem_size = *num_entries * entry_size;
1552 	/* find the possible lowest possible multiplier */
1553 	for (i = 0; i < OCRDMA_MAX_Q_PAGE_SIZE_CNT; i++) {
1554 		if (mem_size <= (OCRDMA_Q_PAGE_BASE_SIZE << i))
1555 			break;
1556 	}
1557 	if (i >= OCRDMA_MAX_Q_PAGE_SIZE_CNT)
1558 		return -EINVAL;
1559 	mem_size = roundup(mem_size,
1560 		       ((OCRDMA_Q_PAGE_BASE_SIZE << i) / OCRDMA_MAX_Q_PAGES));
1561 	*num_pages =
1562 	    mem_size / ((OCRDMA_Q_PAGE_BASE_SIZE << i) / OCRDMA_MAX_Q_PAGES);
1563 	*page_size = ((OCRDMA_Q_PAGE_BASE_SIZE << i) / OCRDMA_MAX_Q_PAGES);
1564 	*num_entries = mem_size / entry_size;
1565 	return 0;
1566 }
1567 
1568 static int ocrdma_mbx_create_ah_tbl(struct ocrdma_dev *dev)
1569 {
1570 	int i;
1571 	int status = 0;
1572 	int max_ah;
1573 	struct ocrdma_create_ah_tbl *cmd;
1574 	struct ocrdma_create_ah_tbl_rsp *rsp;
1575 	struct pci_dev *pdev = dev->nic_info.pdev;
1576 	dma_addr_t pa;
1577 	struct ocrdma_pbe *pbes;
1578 
1579 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_CREATE_AH_TBL, sizeof(*cmd));
1580 	if (!cmd)
1581 		return status;
1582 
1583 	max_ah = OCRDMA_MAX_AH;
1584 	dev->av_tbl.size = sizeof(struct ocrdma_av) * max_ah;
1585 
1586 	/* number of PBEs in PBL */
1587 	cmd->ah_conf = (OCRDMA_AH_TBL_PAGES <<
1588 				OCRDMA_CREATE_AH_NUM_PAGES_SHIFT) &
1589 				OCRDMA_CREATE_AH_NUM_PAGES_MASK;
1590 
1591 	/* page size */
1592 	for (i = 0; i < OCRDMA_MAX_Q_PAGE_SIZE_CNT; i++) {
1593 		if (PAGE_SIZE == (OCRDMA_MIN_Q_PAGE_SIZE << i))
1594 			break;
1595 	}
1596 	cmd->ah_conf |= (i << OCRDMA_CREATE_AH_PAGE_SIZE_SHIFT) &
1597 				OCRDMA_CREATE_AH_PAGE_SIZE_MASK;
1598 
1599 	/* ah_entry size */
1600 	cmd->ah_conf |= (sizeof(struct ocrdma_av) <<
1601 				OCRDMA_CREATE_AH_ENTRY_SIZE_SHIFT) &
1602 				OCRDMA_CREATE_AH_ENTRY_SIZE_MASK;
1603 
1604 	dev->av_tbl.pbl.va = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
1605 						&dev->av_tbl.pbl.pa,
1606 						GFP_KERNEL);
1607 	if (dev->av_tbl.pbl.va == NULL)
1608 		goto mem_err;
1609 
1610 	dev->av_tbl.va = dma_alloc_coherent(&pdev->dev, dev->av_tbl.size,
1611 					    &pa, GFP_KERNEL);
1612 	if (dev->av_tbl.va == NULL)
1613 		goto mem_err_ah;
1614 	dev->av_tbl.pa = pa;
1615 	dev->av_tbl.num_ah = max_ah;
1616 	memset(dev->av_tbl.va, 0, dev->av_tbl.size);
1617 
1618 	pbes = (struct ocrdma_pbe *)dev->av_tbl.pbl.va;
1619 	for (i = 0; i < dev->av_tbl.size / OCRDMA_MIN_Q_PAGE_SIZE; i++) {
1620 		pbes[i].pa_lo = (u32)cpu_to_le32(pa & 0xffffffff);
1621 		pbes[i].pa_hi = (u32)cpu_to_le32(upper_32_bits(pa));
1622 		pa += PAGE_SIZE;
1623 	}
1624 	cmd->tbl_addr[0].lo = (u32)(dev->av_tbl.pbl.pa & 0xFFFFFFFF);
1625 	cmd->tbl_addr[0].hi = (u32)upper_32_bits(dev->av_tbl.pbl.pa);
1626 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1627 	if (status)
1628 		goto mbx_err;
1629 	rsp = (struct ocrdma_create_ah_tbl_rsp *)cmd;
1630 	dev->av_tbl.ahid = rsp->ahid & 0xFFFF;
1631 	kfree(cmd);
1632 	return 0;
1633 
1634 mbx_err:
1635 	dma_free_coherent(&pdev->dev, dev->av_tbl.size, dev->av_tbl.va,
1636 			  dev->av_tbl.pa);
1637 	dev->av_tbl.va = NULL;
1638 mem_err_ah:
1639 	dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->av_tbl.pbl.va,
1640 			  dev->av_tbl.pbl.pa);
1641 	dev->av_tbl.pbl.va = NULL;
1642 	dev->av_tbl.size = 0;
1643 mem_err:
1644 	kfree(cmd);
1645 	return status;
1646 }
1647 
1648 static void ocrdma_mbx_delete_ah_tbl(struct ocrdma_dev *dev)
1649 {
1650 	struct ocrdma_delete_ah_tbl *cmd;
1651 	struct pci_dev *pdev = dev->nic_info.pdev;
1652 
1653 	if (dev->av_tbl.va == NULL)
1654 		return;
1655 
1656 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DELETE_AH_TBL, sizeof(*cmd));
1657 	if (!cmd)
1658 		return;
1659 	cmd->ahid = dev->av_tbl.ahid;
1660 
1661 	ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1662 	dma_free_coherent(&pdev->dev, dev->av_tbl.size, dev->av_tbl.va,
1663 			  dev->av_tbl.pa);
1664 	dev->av_tbl.va = NULL;
1665 	dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->av_tbl.pbl.va,
1666 			  dev->av_tbl.pbl.pa);
1667 	kfree(cmd);
1668 }
1669 
1670 /* Multiple CQs uses the EQ. This routine returns least used
1671  * EQ to associate with CQ. This will distributes the interrupt
1672  * processing and CPU load to associated EQ, vector and so to that CPU.
1673  */
1674 static u16 ocrdma_bind_eq(struct ocrdma_dev *dev)
1675 {
1676 	int i, selected_eq = 0, cq_cnt = 0;
1677 	u16 eq_id;
1678 
1679 	mutex_lock(&dev->dev_lock);
1680 	cq_cnt = dev->eq_tbl[0].cq_cnt;
1681 	eq_id = dev->eq_tbl[0].q.id;
1682 	/* find the EQ which is has the least number of
1683 	 * CQs associated with it.
1684 	 */
1685 	for (i = 0; i < dev->eq_cnt; i++) {
1686 		if (dev->eq_tbl[i].cq_cnt < cq_cnt) {
1687 			cq_cnt = dev->eq_tbl[i].cq_cnt;
1688 			eq_id = dev->eq_tbl[i].q.id;
1689 			selected_eq = i;
1690 		}
1691 	}
1692 	dev->eq_tbl[selected_eq].cq_cnt += 1;
1693 	mutex_unlock(&dev->dev_lock);
1694 	return eq_id;
1695 }
1696 
1697 static void ocrdma_unbind_eq(struct ocrdma_dev *dev, u16 eq_id)
1698 {
1699 	int i;
1700 
1701 	mutex_lock(&dev->dev_lock);
1702 	i = ocrdma_get_eq_table_index(dev, eq_id);
1703 	if (i == -EINVAL)
1704 		BUG();
1705 	dev->eq_tbl[i].cq_cnt -= 1;
1706 	mutex_unlock(&dev->dev_lock);
1707 }
1708 
1709 int ocrdma_mbx_create_cq(struct ocrdma_dev *dev, struct ocrdma_cq *cq,
1710 			 int entries, int dpp_cq, u16 pd_id)
1711 {
1712 	int status = -ENOMEM; int max_hw_cqe;
1713 	struct pci_dev *pdev = dev->nic_info.pdev;
1714 	struct ocrdma_create_cq *cmd;
1715 	struct ocrdma_create_cq_rsp *rsp;
1716 	u32 hw_pages, cqe_size, page_size, cqe_count;
1717 
1718 	if (entries > dev->attr.max_cqe) {
1719 		pr_err("%s(%d) max_cqe=0x%x, requester_cqe=0x%x\n",
1720 		       __func__, dev->id, dev->attr.max_cqe, entries);
1721 		return -EINVAL;
1722 	}
1723 	if (dpp_cq && (ocrdma_get_asic_type(dev) != OCRDMA_ASIC_GEN_SKH_R))
1724 		return -EINVAL;
1725 
1726 	if (dpp_cq) {
1727 		cq->max_hw_cqe = 1;
1728 		max_hw_cqe = 1;
1729 		cqe_size = OCRDMA_DPP_CQE_SIZE;
1730 		hw_pages = 1;
1731 	} else {
1732 		cq->max_hw_cqe = dev->attr.max_cqe;
1733 		max_hw_cqe = dev->attr.max_cqe;
1734 		cqe_size = sizeof(struct ocrdma_cqe);
1735 		hw_pages = OCRDMA_CREATE_CQ_MAX_PAGES;
1736 	}
1737 
1738 	cq->len = roundup(max_hw_cqe * cqe_size, OCRDMA_MIN_Q_PAGE_SIZE);
1739 
1740 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_CREATE_CQ, sizeof(*cmd));
1741 	if (!cmd)
1742 		return -ENOMEM;
1743 	ocrdma_init_mch(&cmd->cmd.req, OCRDMA_CMD_CREATE_CQ,
1744 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
1745 	cq->va = dma_alloc_coherent(&pdev->dev, cq->len, &cq->pa, GFP_KERNEL);
1746 	if (!cq->va) {
1747 		status = -ENOMEM;
1748 		goto mem_err;
1749 	}
1750 	memset(cq->va, 0, cq->len);
1751 	page_size = cq->len / hw_pages;
1752 	cmd->cmd.pgsz_pgcnt = (page_size / OCRDMA_MIN_Q_PAGE_SIZE) <<
1753 					OCRDMA_CREATE_CQ_PAGE_SIZE_SHIFT;
1754 	cmd->cmd.pgsz_pgcnt |= hw_pages;
1755 	cmd->cmd.ev_cnt_flags = OCRDMA_CREATE_CQ_DEF_FLAGS;
1756 
1757 	cq->eqn = ocrdma_bind_eq(dev);
1758 	cmd->cmd.req.rsvd_version = OCRDMA_CREATE_CQ_VER3;
1759 	cqe_count = cq->len / cqe_size;
1760 	cq->cqe_cnt = cqe_count;
1761 	if (cqe_count > 1024) {
1762 		/* Set cnt to 3 to indicate more than 1024 cq entries */
1763 		cmd->cmd.ev_cnt_flags |= (0x3 << OCRDMA_CREATE_CQ_CNT_SHIFT);
1764 	} else {
1765 		u8 count = 0;
1766 		switch (cqe_count) {
1767 		case 256:
1768 			count = 0;
1769 			break;
1770 		case 512:
1771 			count = 1;
1772 			break;
1773 		case 1024:
1774 			count = 2;
1775 			break;
1776 		default:
1777 			goto mbx_err;
1778 		}
1779 		cmd->cmd.ev_cnt_flags |= (count << OCRDMA_CREATE_CQ_CNT_SHIFT);
1780 	}
1781 	/* shared eq between all the consumer cqs. */
1782 	cmd->cmd.eqn = cq->eqn;
1783 	if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) {
1784 		if (dpp_cq)
1785 			cmd->cmd.pgsz_pgcnt |= OCRDMA_CREATE_CQ_DPP <<
1786 				OCRDMA_CREATE_CQ_TYPE_SHIFT;
1787 		cq->phase_change = false;
1788 		cmd->cmd.pdid_cqecnt = (cq->len / cqe_size);
1789 	} else {
1790 		cmd->cmd.pdid_cqecnt = (cq->len / cqe_size) - 1;
1791 		cmd->cmd.ev_cnt_flags |= OCRDMA_CREATE_CQ_FLAGS_AUTO_VALID;
1792 		cq->phase_change = true;
1793 	}
1794 
1795 	/* pd_id valid only for v3 */
1796 	cmd->cmd.pdid_cqecnt |= (pd_id <<
1797 		OCRDMA_CREATE_CQ_CMD_PDID_SHIFT);
1798 	ocrdma_build_q_pages(&cmd->cmd.pa[0], hw_pages, cq->pa, page_size);
1799 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1800 	if (status)
1801 		goto mbx_err;
1802 
1803 	rsp = (struct ocrdma_create_cq_rsp *)cmd;
1804 	cq->id = (u16) (rsp->rsp.cq_id & OCRDMA_CREATE_CQ_RSP_CQ_ID_MASK);
1805 	kfree(cmd);
1806 	return 0;
1807 mbx_err:
1808 	ocrdma_unbind_eq(dev, cq->eqn);
1809 	dma_free_coherent(&pdev->dev, cq->len, cq->va, cq->pa);
1810 mem_err:
1811 	kfree(cmd);
1812 	return status;
1813 }
1814 
1815 int ocrdma_mbx_destroy_cq(struct ocrdma_dev *dev, struct ocrdma_cq *cq)
1816 {
1817 	int status = -ENOMEM;
1818 	struct ocrdma_destroy_cq *cmd;
1819 
1820 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DELETE_CQ, sizeof(*cmd));
1821 	if (!cmd)
1822 		return status;
1823 	ocrdma_init_mch(&cmd->req, OCRDMA_CMD_DELETE_CQ,
1824 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
1825 
1826 	cmd->bypass_flush_qid |=
1827 	    (cq->id << OCRDMA_DESTROY_CQ_QID_SHIFT) &
1828 	    OCRDMA_DESTROY_CQ_QID_MASK;
1829 
1830 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1831 	ocrdma_unbind_eq(dev, cq->eqn);
1832 	dma_free_coherent(&dev->nic_info.pdev->dev, cq->len, cq->va, cq->pa);
1833 	kfree(cmd);
1834 	return status;
1835 }
1836 
1837 int ocrdma_mbx_alloc_lkey(struct ocrdma_dev *dev, struct ocrdma_hw_mr *hwmr,
1838 			  u32 pdid, int addr_check)
1839 {
1840 	int status = -ENOMEM;
1841 	struct ocrdma_alloc_lkey *cmd;
1842 	struct ocrdma_alloc_lkey_rsp *rsp;
1843 
1844 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_ALLOC_LKEY, sizeof(*cmd));
1845 	if (!cmd)
1846 		return status;
1847 	cmd->pdid = pdid;
1848 	cmd->pbl_sz_flags |= addr_check;
1849 	cmd->pbl_sz_flags |= (hwmr->fr_mr << OCRDMA_ALLOC_LKEY_FMR_SHIFT);
1850 	cmd->pbl_sz_flags |=
1851 	    (hwmr->remote_wr << OCRDMA_ALLOC_LKEY_REMOTE_WR_SHIFT);
1852 	cmd->pbl_sz_flags |=
1853 	    (hwmr->remote_rd << OCRDMA_ALLOC_LKEY_REMOTE_RD_SHIFT);
1854 	cmd->pbl_sz_flags |=
1855 	    (hwmr->local_wr << OCRDMA_ALLOC_LKEY_LOCAL_WR_SHIFT);
1856 	cmd->pbl_sz_flags |=
1857 	    (hwmr->remote_atomic << OCRDMA_ALLOC_LKEY_REMOTE_ATOMIC_SHIFT);
1858 	cmd->pbl_sz_flags |=
1859 	    (hwmr->num_pbls << OCRDMA_ALLOC_LKEY_PBL_SIZE_SHIFT);
1860 
1861 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1862 	if (status)
1863 		goto mbx_err;
1864 	rsp = (struct ocrdma_alloc_lkey_rsp *)cmd;
1865 	hwmr->lkey = rsp->lrkey;
1866 mbx_err:
1867 	kfree(cmd);
1868 	return status;
1869 }
1870 
1871 int ocrdma_mbx_dealloc_lkey(struct ocrdma_dev *dev, int fr_mr, u32 lkey)
1872 {
1873 	int status = -ENOMEM;
1874 	struct ocrdma_dealloc_lkey *cmd;
1875 
1876 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DEALLOC_LKEY, sizeof(*cmd));
1877 	if (!cmd)
1878 		return -ENOMEM;
1879 	cmd->lkey = lkey;
1880 	cmd->rsvd_frmr = fr_mr ? 1 : 0;
1881 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1882 	if (status)
1883 		goto mbx_err;
1884 mbx_err:
1885 	kfree(cmd);
1886 	return status;
1887 }
1888 
1889 static int ocrdma_mbx_reg_mr(struct ocrdma_dev *dev, struct ocrdma_hw_mr *hwmr,
1890 			     u32 pdid, u32 pbl_cnt, u32 pbe_size, u32 last)
1891 {
1892 	int status = -ENOMEM;
1893 	int i;
1894 	struct ocrdma_reg_nsmr *cmd;
1895 	struct ocrdma_reg_nsmr_rsp *rsp;
1896 
1897 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_REGISTER_NSMR, sizeof(*cmd));
1898 	if (!cmd)
1899 		return -ENOMEM;
1900 	cmd->num_pbl_pdid =
1901 	    pdid | (hwmr->num_pbls << OCRDMA_REG_NSMR_NUM_PBL_SHIFT);
1902 	cmd->fr_mr = hwmr->fr_mr;
1903 
1904 	cmd->flags_hpage_pbe_sz |= (hwmr->remote_wr <<
1905 				    OCRDMA_REG_NSMR_REMOTE_WR_SHIFT);
1906 	cmd->flags_hpage_pbe_sz |= (hwmr->remote_rd <<
1907 				    OCRDMA_REG_NSMR_REMOTE_RD_SHIFT);
1908 	cmd->flags_hpage_pbe_sz |= (hwmr->local_wr <<
1909 				    OCRDMA_REG_NSMR_LOCAL_WR_SHIFT);
1910 	cmd->flags_hpage_pbe_sz |= (hwmr->remote_atomic <<
1911 				    OCRDMA_REG_NSMR_REMOTE_ATOMIC_SHIFT);
1912 	cmd->flags_hpage_pbe_sz |= (hwmr->mw_bind <<
1913 				    OCRDMA_REG_NSMR_BIND_MEMWIN_SHIFT);
1914 	cmd->flags_hpage_pbe_sz |= (last << OCRDMA_REG_NSMR_LAST_SHIFT);
1915 
1916 	cmd->flags_hpage_pbe_sz |= (hwmr->pbe_size / OCRDMA_MIN_HPAGE_SIZE);
1917 	cmd->flags_hpage_pbe_sz |= (hwmr->pbl_size / OCRDMA_MIN_HPAGE_SIZE) <<
1918 					OCRDMA_REG_NSMR_HPAGE_SIZE_SHIFT;
1919 	cmd->totlen_low = hwmr->len;
1920 	cmd->totlen_high = upper_32_bits(hwmr->len);
1921 	cmd->fbo_low = (u32) (hwmr->fbo & 0xffffffff);
1922 	cmd->fbo_high = (u32) upper_32_bits(hwmr->fbo);
1923 	cmd->va_loaddr = (u32) hwmr->va;
1924 	cmd->va_hiaddr = (u32) upper_32_bits(hwmr->va);
1925 
1926 	for (i = 0; i < pbl_cnt; i++) {
1927 		cmd->pbl[i].lo = (u32) (hwmr->pbl_table[i].pa & 0xffffffff);
1928 		cmd->pbl[i].hi = upper_32_bits(hwmr->pbl_table[i].pa);
1929 	}
1930 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1931 	if (status)
1932 		goto mbx_err;
1933 	rsp = (struct ocrdma_reg_nsmr_rsp *)cmd;
1934 	hwmr->lkey = rsp->lrkey;
1935 mbx_err:
1936 	kfree(cmd);
1937 	return status;
1938 }
1939 
1940 static int ocrdma_mbx_reg_mr_cont(struct ocrdma_dev *dev,
1941 				  struct ocrdma_hw_mr *hwmr, u32 pbl_cnt,
1942 				  u32 pbl_offset, u32 last)
1943 {
1944 	int status = -ENOMEM;
1945 	int i;
1946 	struct ocrdma_reg_nsmr_cont *cmd;
1947 
1948 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_REGISTER_NSMR_CONT, sizeof(*cmd));
1949 	if (!cmd)
1950 		return -ENOMEM;
1951 	cmd->lrkey = hwmr->lkey;
1952 	cmd->num_pbl_offset = (pbl_cnt << OCRDMA_REG_NSMR_CONT_NUM_PBL_SHIFT) |
1953 	    (pbl_offset & OCRDMA_REG_NSMR_CONT_PBL_SHIFT_MASK);
1954 	cmd->last = last << OCRDMA_REG_NSMR_CONT_LAST_SHIFT;
1955 
1956 	for (i = 0; i < pbl_cnt; i++) {
1957 		cmd->pbl[i].lo =
1958 		    (u32) (hwmr->pbl_table[i + pbl_offset].pa & 0xffffffff);
1959 		cmd->pbl[i].hi =
1960 		    upper_32_bits(hwmr->pbl_table[i + pbl_offset].pa);
1961 	}
1962 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1963 	if (status)
1964 		goto mbx_err;
1965 mbx_err:
1966 	kfree(cmd);
1967 	return status;
1968 }
1969 
1970 int ocrdma_reg_mr(struct ocrdma_dev *dev,
1971 		  struct ocrdma_hw_mr *hwmr, u32 pdid, int acc)
1972 {
1973 	int status;
1974 	u32 last = 0;
1975 	u32 cur_pbl_cnt, pbl_offset;
1976 	u32 pending_pbl_cnt = hwmr->num_pbls;
1977 
1978 	pbl_offset = 0;
1979 	cur_pbl_cnt = min(pending_pbl_cnt, MAX_OCRDMA_NSMR_PBL);
1980 	if (cur_pbl_cnt == pending_pbl_cnt)
1981 		last = 1;
1982 
1983 	status = ocrdma_mbx_reg_mr(dev, hwmr, pdid,
1984 				   cur_pbl_cnt, hwmr->pbe_size, last);
1985 	if (status) {
1986 		pr_err("%s() status=%d\n", __func__, status);
1987 		return status;
1988 	}
1989 	/* if there is no more pbls to register then exit. */
1990 	if (last)
1991 		return 0;
1992 
1993 	while (!last) {
1994 		pbl_offset += cur_pbl_cnt;
1995 		pending_pbl_cnt -= cur_pbl_cnt;
1996 		cur_pbl_cnt = min(pending_pbl_cnt, MAX_OCRDMA_NSMR_PBL);
1997 		/* if we reach the end of the pbls, then need to set the last
1998 		 * bit, indicating no more pbls to register for this memory key.
1999 		 */
2000 		if (cur_pbl_cnt == pending_pbl_cnt)
2001 			last = 1;
2002 
2003 		status = ocrdma_mbx_reg_mr_cont(dev, hwmr, cur_pbl_cnt,
2004 						pbl_offset, last);
2005 		if (status)
2006 			break;
2007 	}
2008 	if (status)
2009 		pr_err("%s() err. status=%d\n", __func__, status);
2010 
2011 	return status;
2012 }
2013 
2014 bool ocrdma_is_qp_in_sq_flushlist(struct ocrdma_cq *cq, struct ocrdma_qp *qp)
2015 {
2016 	struct ocrdma_qp *tmp;
2017 	bool found = false;
2018 	list_for_each_entry(tmp, &cq->sq_head, sq_entry) {
2019 		if (qp == tmp) {
2020 			found = true;
2021 			break;
2022 		}
2023 	}
2024 	return found;
2025 }
2026 
2027 bool ocrdma_is_qp_in_rq_flushlist(struct ocrdma_cq *cq, struct ocrdma_qp *qp)
2028 {
2029 	struct ocrdma_qp *tmp;
2030 	bool found = false;
2031 	list_for_each_entry(tmp, &cq->rq_head, rq_entry) {
2032 		if (qp == tmp) {
2033 			found = true;
2034 			break;
2035 		}
2036 	}
2037 	return found;
2038 }
2039 
2040 void ocrdma_flush_qp(struct ocrdma_qp *qp)
2041 {
2042 	bool found;
2043 	unsigned long flags;
2044 	struct ocrdma_dev *dev = get_ocrdma_dev(qp->ibqp.device);
2045 
2046 	spin_lock_irqsave(&dev->flush_q_lock, flags);
2047 	found = ocrdma_is_qp_in_sq_flushlist(qp->sq_cq, qp);
2048 	if (!found)
2049 		list_add_tail(&qp->sq_entry, &qp->sq_cq->sq_head);
2050 	if (!qp->srq) {
2051 		found = ocrdma_is_qp_in_rq_flushlist(qp->rq_cq, qp);
2052 		if (!found)
2053 			list_add_tail(&qp->rq_entry, &qp->rq_cq->rq_head);
2054 	}
2055 	spin_unlock_irqrestore(&dev->flush_q_lock, flags);
2056 }
2057 
2058 static void ocrdma_init_hwq_ptr(struct ocrdma_qp *qp)
2059 {
2060 	qp->sq.head = 0;
2061 	qp->sq.tail = 0;
2062 	qp->rq.head = 0;
2063 	qp->rq.tail = 0;
2064 }
2065 
2066 int ocrdma_qp_state_change(struct ocrdma_qp *qp, enum ib_qp_state new_ib_state,
2067 			   enum ib_qp_state *old_ib_state)
2068 {
2069 	unsigned long flags;
2070 	int status = 0;
2071 	enum ocrdma_qp_state new_state;
2072 	new_state = get_ocrdma_qp_state(new_ib_state);
2073 
2074 	/* sync with wqe and rqe posting */
2075 	spin_lock_irqsave(&qp->q_lock, flags);
2076 
2077 	if (old_ib_state)
2078 		*old_ib_state = get_ibqp_state(qp->state);
2079 	if (new_state == qp->state) {
2080 		spin_unlock_irqrestore(&qp->q_lock, flags);
2081 		return 1;
2082 	}
2083 
2084 
2085 	if (new_state == OCRDMA_QPS_INIT) {
2086 		ocrdma_init_hwq_ptr(qp);
2087 		ocrdma_del_flush_qp(qp);
2088 	} else if (new_state == OCRDMA_QPS_ERR) {
2089 		ocrdma_flush_qp(qp);
2090 	}
2091 
2092 	qp->state = new_state;
2093 
2094 	spin_unlock_irqrestore(&qp->q_lock, flags);
2095 	return status;
2096 }
2097 
2098 static u32 ocrdma_set_create_qp_mbx_access_flags(struct ocrdma_qp *qp)
2099 {
2100 	u32 flags = 0;
2101 	if (qp->cap_flags & OCRDMA_QP_INB_RD)
2102 		flags |= OCRDMA_CREATE_QP_REQ_INB_RDEN_MASK;
2103 	if (qp->cap_flags & OCRDMA_QP_INB_WR)
2104 		flags |= OCRDMA_CREATE_QP_REQ_INB_WREN_MASK;
2105 	if (qp->cap_flags & OCRDMA_QP_MW_BIND)
2106 		flags |= OCRDMA_CREATE_QP_REQ_BIND_MEMWIN_MASK;
2107 	if (qp->cap_flags & OCRDMA_QP_LKEY0)
2108 		flags |= OCRDMA_CREATE_QP_REQ_ZERO_LKEYEN_MASK;
2109 	if (qp->cap_flags & OCRDMA_QP_FAST_REG)
2110 		flags |= OCRDMA_CREATE_QP_REQ_FMR_EN_MASK;
2111 	return flags;
2112 }
2113 
2114 static int ocrdma_set_create_qp_sq_cmd(struct ocrdma_create_qp_req *cmd,
2115 					struct ib_qp_init_attr *attrs,
2116 					struct ocrdma_qp *qp)
2117 {
2118 	int status;
2119 	u32 len, hw_pages, hw_page_size;
2120 	dma_addr_t pa;
2121 	struct ocrdma_pd *pd = qp->pd;
2122 	struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device);
2123 	struct pci_dev *pdev = dev->nic_info.pdev;
2124 	u32 max_wqe_allocated;
2125 	u32 max_sges = attrs->cap.max_send_sge;
2126 
2127 	/* QP1 may exceed 127 */
2128 	max_wqe_allocated = min_t(u32, attrs->cap.max_send_wr + 1,
2129 				dev->attr.max_wqe);
2130 
2131 	status = ocrdma_build_q_conf(&max_wqe_allocated,
2132 		dev->attr.wqe_size, &hw_pages, &hw_page_size);
2133 	if (status) {
2134 		pr_err("%s() req. max_send_wr=0x%x\n", __func__,
2135 		       max_wqe_allocated);
2136 		return -EINVAL;
2137 	}
2138 	qp->sq.max_cnt = max_wqe_allocated;
2139 	len = (hw_pages * hw_page_size);
2140 
2141 	qp->sq.va = dma_alloc_coherent(&pdev->dev, len, &pa, GFP_KERNEL);
2142 	if (!qp->sq.va)
2143 		return -EINVAL;
2144 	memset(qp->sq.va, 0, len);
2145 	qp->sq.len = len;
2146 	qp->sq.pa = pa;
2147 	qp->sq.entry_size = dev->attr.wqe_size;
2148 	ocrdma_build_q_pages(&cmd->wq_addr[0], hw_pages, pa, hw_page_size);
2149 
2150 	cmd->type_pgsz_pdn |= (ilog2(hw_page_size / OCRDMA_MIN_Q_PAGE_SIZE)
2151 				<< OCRDMA_CREATE_QP_REQ_SQ_PAGE_SIZE_SHIFT);
2152 	cmd->num_wq_rq_pages |= (hw_pages <<
2153 				 OCRDMA_CREATE_QP_REQ_NUM_WQ_PAGES_SHIFT) &
2154 	    OCRDMA_CREATE_QP_REQ_NUM_WQ_PAGES_MASK;
2155 	cmd->max_sge_send_write |= (max_sges <<
2156 				    OCRDMA_CREATE_QP_REQ_MAX_SGE_SEND_SHIFT) &
2157 	    OCRDMA_CREATE_QP_REQ_MAX_SGE_SEND_MASK;
2158 	cmd->max_sge_send_write |= (max_sges <<
2159 				    OCRDMA_CREATE_QP_REQ_MAX_SGE_WRITE_SHIFT) &
2160 					OCRDMA_CREATE_QP_REQ_MAX_SGE_WRITE_MASK;
2161 	cmd->max_wqe_rqe |= (ilog2(qp->sq.max_cnt) <<
2162 			     OCRDMA_CREATE_QP_REQ_MAX_WQE_SHIFT) &
2163 				OCRDMA_CREATE_QP_REQ_MAX_WQE_MASK;
2164 	cmd->wqe_rqe_size |= (dev->attr.wqe_size <<
2165 			      OCRDMA_CREATE_QP_REQ_WQE_SIZE_SHIFT) &
2166 				OCRDMA_CREATE_QP_REQ_WQE_SIZE_MASK;
2167 	return 0;
2168 }
2169 
2170 static int ocrdma_set_create_qp_rq_cmd(struct ocrdma_create_qp_req *cmd,
2171 					struct ib_qp_init_attr *attrs,
2172 					struct ocrdma_qp *qp)
2173 {
2174 	int status;
2175 	u32 len, hw_pages, hw_page_size;
2176 	dma_addr_t pa = 0;
2177 	struct ocrdma_pd *pd = qp->pd;
2178 	struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device);
2179 	struct pci_dev *pdev = dev->nic_info.pdev;
2180 	u32 max_rqe_allocated = attrs->cap.max_recv_wr + 1;
2181 
2182 	status = ocrdma_build_q_conf(&max_rqe_allocated, dev->attr.rqe_size,
2183 				     &hw_pages, &hw_page_size);
2184 	if (status) {
2185 		pr_err("%s() req. max_recv_wr=0x%x\n", __func__,
2186 		       attrs->cap.max_recv_wr + 1);
2187 		return status;
2188 	}
2189 	qp->rq.max_cnt = max_rqe_allocated;
2190 	len = (hw_pages * hw_page_size);
2191 
2192 	qp->rq.va = dma_alloc_coherent(&pdev->dev, len, &pa, GFP_KERNEL);
2193 	if (!qp->rq.va)
2194 		return -ENOMEM;
2195 	memset(qp->rq.va, 0, len);
2196 	qp->rq.pa = pa;
2197 	qp->rq.len = len;
2198 	qp->rq.entry_size = dev->attr.rqe_size;
2199 
2200 	ocrdma_build_q_pages(&cmd->rq_addr[0], hw_pages, pa, hw_page_size);
2201 	cmd->type_pgsz_pdn |= (ilog2(hw_page_size / OCRDMA_MIN_Q_PAGE_SIZE) <<
2202 		OCRDMA_CREATE_QP_REQ_RQ_PAGE_SIZE_SHIFT);
2203 	cmd->num_wq_rq_pages |=
2204 	    (hw_pages << OCRDMA_CREATE_QP_REQ_NUM_RQ_PAGES_SHIFT) &
2205 	    OCRDMA_CREATE_QP_REQ_NUM_RQ_PAGES_MASK;
2206 	cmd->max_sge_recv_flags |= (attrs->cap.max_recv_sge <<
2207 				OCRDMA_CREATE_QP_REQ_MAX_SGE_RECV_SHIFT) &
2208 				OCRDMA_CREATE_QP_REQ_MAX_SGE_RECV_MASK;
2209 	cmd->max_wqe_rqe |= (ilog2(qp->rq.max_cnt) <<
2210 				OCRDMA_CREATE_QP_REQ_MAX_RQE_SHIFT) &
2211 				OCRDMA_CREATE_QP_REQ_MAX_RQE_MASK;
2212 	cmd->wqe_rqe_size |= (dev->attr.rqe_size <<
2213 			OCRDMA_CREATE_QP_REQ_RQE_SIZE_SHIFT) &
2214 			OCRDMA_CREATE_QP_REQ_RQE_SIZE_MASK;
2215 	return 0;
2216 }
2217 
2218 static void ocrdma_set_create_qp_dpp_cmd(struct ocrdma_create_qp_req *cmd,
2219 					 struct ocrdma_pd *pd,
2220 					 struct ocrdma_qp *qp,
2221 					 u8 enable_dpp_cq, u16 dpp_cq_id)
2222 {
2223 	pd->num_dpp_qp--;
2224 	qp->dpp_enabled = true;
2225 	cmd->max_sge_recv_flags |= OCRDMA_CREATE_QP_REQ_ENABLE_DPP_MASK;
2226 	if (!enable_dpp_cq)
2227 		return;
2228 	cmd->max_sge_recv_flags |= OCRDMA_CREATE_QP_REQ_ENABLE_DPP_MASK;
2229 	cmd->dpp_credits_cqid = dpp_cq_id;
2230 	cmd->dpp_credits_cqid |= OCRDMA_CREATE_QP_REQ_DPP_CREDIT_LIMIT <<
2231 					OCRDMA_CREATE_QP_REQ_DPP_CREDIT_SHIFT;
2232 }
2233 
2234 static int ocrdma_set_create_qp_ird_cmd(struct ocrdma_create_qp_req *cmd,
2235 					struct ocrdma_qp *qp)
2236 {
2237 	struct ocrdma_pd *pd = qp->pd;
2238 	struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device);
2239 	struct pci_dev *pdev = dev->nic_info.pdev;
2240 	dma_addr_t pa = 0;
2241 	int ird_page_size = dev->attr.ird_page_size;
2242 	int ird_q_len = dev->attr.num_ird_pages * ird_page_size;
2243 	struct ocrdma_hdr_wqe *rqe;
2244 	int i  = 0;
2245 
2246 	if (dev->attr.ird == 0)
2247 		return 0;
2248 
2249 	qp->ird_q_va = dma_alloc_coherent(&pdev->dev, ird_q_len,
2250 					&pa, GFP_KERNEL);
2251 	if (!qp->ird_q_va)
2252 		return -ENOMEM;
2253 	memset(qp->ird_q_va, 0, ird_q_len);
2254 	ocrdma_build_q_pages(&cmd->ird_addr[0], dev->attr.num_ird_pages,
2255 			     pa, ird_page_size);
2256 	for (; i < ird_q_len / dev->attr.rqe_size; i++) {
2257 		rqe = (struct ocrdma_hdr_wqe *)(qp->ird_q_va +
2258 			(i * dev->attr.rqe_size));
2259 		rqe->cw = 0;
2260 		rqe->cw |= 2;
2261 		rqe->cw |= (OCRDMA_TYPE_LKEY << OCRDMA_WQE_TYPE_SHIFT);
2262 		rqe->cw |= (8 << OCRDMA_WQE_SIZE_SHIFT);
2263 		rqe->cw |= (8 << OCRDMA_WQE_NXT_WQE_SIZE_SHIFT);
2264 	}
2265 	return 0;
2266 }
2267 
2268 static void ocrdma_get_create_qp_rsp(struct ocrdma_create_qp_rsp *rsp,
2269 				     struct ocrdma_qp *qp,
2270 				     struct ib_qp_init_attr *attrs,
2271 				     u16 *dpp_offset, u16 *dpp_credit_lmt)
2272 {
2273 	u32 max_wqe_allocated, max_rqe_allocated;
2274 	qp->id = rsp->qp_id & OCRDMA_CREATE_QP_RSP_QP_ID_MASK;
2275 	qp->rq.dbid = rsp->sq_rq_id & OCRDMA_CREATE_QP_RSP_RQ_ID_MASK;
2276 	qp->sq.dbid = rsp->sq_rq_id >> OCRDMA_CREATE_QP_RSP_SQ_ID_SHIFT;
2277 	qp->max_ird = rsp->max_ord_ird & OCRDMA_CREATE_QP_RSP_MAX_IRD_MASK;
2278 	qp->max_ord = (rsp->max_ord_ird >> OCRDMA_CREATE_QP_RSP_MAX_ORD_SHIFT);
2279 	qp->dpp_enabled = false;
2280 	if (rsp->dpp_response & OCRDMA_CREATE_QP_RSP_DPP_ENABLED_MASK) {
2281 		qp->dpp_enabled = true;
2282 		*dpp_credit_lmt = (rsp->dpp_response &
2283 				OCRDMA_CREATE_QP_RSP_DPP_CREDITS_MASK) >>
2284 				OCRDMA_CREATE_QP_RSP_DPP_CREDITS_SHIFT;
2285 		*dpp_offset = (rsp->dpp_response &
2286 				OCRDMA_CREATE_QP_RSP_DPP_PAGE_OFFSET_MASK) >>
2287 				OCRDMA_CREATE_QP_RSP_DPP_PAGE_OFFSET_SHIFT;
2288 	}
2289 	max_wqe_allocated =
2290 		rsp->max_wqe_rqe >> OCRDMA_CREATE_QP_RSP_MAX_WQE_SHIFT;
2291 	max_wqe_allocated = 1 << max_wqe_allocated;
2292 	max_rqe_allocated = 1 << ((u16)rsp->max_wqe_rqe);
2293 
2294 	qp->sq.max_cnt = max_wqe_allocated;
2295 	qp->sq.max_wqe_idx = max_wqe_allocated - 1;
2296 
2297 	if (!attrs->srq) {
2298 		qp->rq.max_cnt = max_rqe_allocated;
2299 		qp->rq.max_wqe_idx = max_rqe_allocated - 1;
2300 	}
2301 }
2302 
2303 int ocrdma_mbx_create_qp(struct ocrdma_qp *qp, struct ib_qp_init_attr *attrs,
2304 			 u8 enable_dpp_cq, u16 dpp_cq_id, u16 *dpp_offset,
2305 			 u16 *dpp_credit_lmt)
2306 {
2307 	int status = -ENOMEM;
2308 	u32 flags = 0;
2309 	struct ocrdma_pd *pd = qp->pd;
2310 	struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device);
2311 	struct pci_dev *pdev = dev->nic_info.pdev;
2312 	struct ocrdma_cq *cq;
2313 	struct ocrdma_create_qp_req *cmd;
2314 	struct ocrdma_create_qp_rsp *rsp;
2315 	int qptype;
2316 
2317 	switch (attrs->qp_type) {
2318 	case IB_QPT_GSI:
2319 		qptype = OCRDMA_QPT_GSI;
2320 		break;
2321 	case IB_QPT_RC:
2322 		qptype = OCRDMA_QPT_RC;
2323 		break;
2324 	case IB_QPT_UD:
2325 		qptype = OCRDMA_QPT_UD;
2326 		break;
2327 	default:
2328 		return -EINVAL;
2329 	}
2330 
2331 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_CREATE_QP, sizeof(*cmd));
2332 	if (!cmd)
2333 		return status;
2334 	cmd->type_pgsz_pdn |= (qptype << OCRDMA_CREATE_QP_REQ_QPT_SHIFT) &
2335 						OCRDMA_CREATE_QP_REQ_QPT_MASK;
2336 	status = ocrdma_set_create_qp_sq_cmd(cmd, attrs, qp);
2337 	if (status)
2338 		goto sq_err;
2339 
2340 	if (attrs->srq) {
2341 		struct ocrdma_srq *srq = get_ocrdma_srq(attrs->srq);
2342 		cmd->max_sge_recv_flags |= OCRDMA_CREATE_QP_REQ_USE_SRQ_MASK;
2343 		cmd->rq_addr[0].lo = srq->id;
2344 		qp->srq = srq;
2345 	} else {
2346 		status = ocrdma_set_create_qp_rq_cmd(cmd, attrs, qp);
2347 		if (status)
2348 			goto rq_err;
2349 	}
2350 
2351 	status = ocrdma_set_create_qp_ird_cmd(cmd, qp);
2352 	if (status)
2353 		goto mbx_err;
2354 
2355 	cmd->type_pgsz_pdn |= (pd->id << OCRDMA_CREATE_QP_REQ_PD_ID_SHIFT) &
2356 				OCRDMA_CREATE_QP_REQ_PD_ID_MASK;
2357 
2358 	flags = ocrdma_set_create_qp_mbx_access_flags(qp);
2359 
2360 	cmd->max_sge_recv_flags |= flags;
2361 	cmd->max_ord_ird |= (dev->attr.max_ord_per_qp <<
2362 			     OCRDMA_CREATE_QP_REQ_MAX_ORD_SHIFT) &
2363 				OCRDMA_CREATE_QP_REQ_MAX_ORD_MASK;
2364 	cmd->max_ord_ird |= (dev->attr.max_ird_per_qp <<
2365 			     OCRDMA_CREATE_QP_REQ_MAX_IRD_SHIFT) &
2366 				OCRDMA_CREATE_QP_REQ_MAX_IRD_MASK;
2367 	cq = get_ocrdma_cq(attrs->send_cq);
2368 	cmd->wq_rq_cqid |= (cq->id << OCRDMA_CREATE_QP_REQ_WQ_CQID_SHIFT) &
2369 				OCRDMA_CREATE_QP_REQ_WQ_CQID_MASK;
2370 	qp->sq_cq = cq;
2371 	cq = get_ocrdma_cq(attrs->recv_cq);
2372 	cmd->wq_rq_cqid |= (cq->id << OCRDMA_CREATE_QP_REQ_RQ_CQID_SHIFT) &
2373 				OCRDMA_CREATE_QP_REQ_RQ_CQID_MASK;
2374 	qp->rq_cq = cq;
2375 
2376 	if (pd->dpp_enabled && attrs->cap.max_inline_data && pd->num_dpp_qp &&
2377 	    (attrs->cap.max_inline_data <= dev->attr.max_inline_data)) {
2378 		ocrdma_set_create_qp_dpp_cmd(cmd, pd, qp, enable_dpp_cq,
2379 					     dpp_cq_id);
2380 	}
2381 
2382 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2383 	if (status)
2384 		goto mbx_err;
2385 	rsp = (struct ocrdma_create_qp_rsp *)cmd;
2386 	ocrdma_get_create_qp_rsp(rsp, qp, attrs, dpp_offset, dpp_credit_lmt);
2387 	qp->state = OCRDMA_QPS_RST;
2388 	kfree(cmd);
2389 	return 0;
2390 mbx_err:
2391 	if (qp->rq.va)
2392 		dma_free_coherent(&pdev->dev, qp->rq.len, qp->rq.va, qp->rq.pa);
2393 rq_err:
2394 	pr_err("%s(%d) rq_err\n", __func__, dev->id);
2395 	dma_free_coherent(&pdev->dev, qp->sq.len, qp->sq.va, qp->sq.pa);
2396 sq_err:
2397 	pr_err("%s(%d) sq_err\n", __func__, dev->id);
2398 	kfree(cmd);
2399 	return status;
2400 }
2401 
2402 int ocrdma_mbx_query_qp(struct ocrdma_dev *dev, struct ocrdma_qp *qp,
2403 			struct ocrdma_qp_params *param)
2404 {
2405 	int status = -ENOMEM;
2406 	struct ocrdma_query_qp *cmd;
2407 	struct ocrdma_query_qp_rsp *rsp;
2408 
2409 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_QP, sizeof(*cmd));
2410 	if (!cmd)
2411 		return status;
2412 	cmd->qp_id = qp->id;
2413 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2414 	if (status)
2415 		goto mbx_err;
2416 	rsp = (struct ocrdma_query_qp_rsp *)cmd;
2417 	memcpy(param, &rsp->params, sizeof(struct ocrdma_qp_params));
2418 mbx_err:
2419 	kfree(cmd);
2420 	return status;
2421 }
2422 
2423 static int ocrdma_set_av_params(struct ocrdma_qp *qp,
2424 				struct ocrdma_modify_qp *cmd,
2425 				struct ib_qp_attr *attrs,
2426 				int attr_mask)
2427 {
2428 	int status;
2429 	struct ib_ah_attr *ah_attr = &attrs->ah_attr;
2430 	union ib_gid sgid, zgid;
2431 	u32 vlan_id;
2432 	u8 mac_addr[6];
2433 	struct ocrdma_dev *dev = get_ocrdma_dev(qp->ibqp.device);
2434 
2435 	if ((ah_attr->ah_flags & IB_AH_GRH) == 0)
2436 		return -EINVAL;
2437 	if (atomic_cmpxchg(&dev->update_sl, 1, 0))
2438 		ocrdma_init_service_level(dev);
2439 	cmd->params.tclass_sq_psn |=
2440 	    (ah_attr->grh.traffic_class << OCRDMA_QP_PARAMS_TCLASS_SHIFT);
2441 	cmd->params.rnt_rc_sl_fl |=
2442 	    (ah_attr->grh.flow_label & OCRDMA_QP_PARAMS_FLOW_LABEL_MASK);
2443 	cmd->params.rnt_rc_sl_fl |= (ah_attr->sl << OCRDMA_QP_PARAMS_SL_SHIFT);
2444 	cmd->params.hop_lmt_rq_psn |=
2445 	    (ah_attr->grh.hop_limit << OCRDMA_QP_PARAMS_HOP_LMT_SHIFT);
2446 	cmd->flags |= OCRDMA_QP_PARA_FLOW_LBL_VALID;
2447 	memcpy(&cmd->params.dgid[0], &ah_attr->grh.dgid.raw[0],
2448 	       sizeof(cmd->params.dgid));
2449 	status = ocrdma_query_gid(&dev->ibdev, 1,
2450 			ah_attr->grh.sgid_index, &sgid);
2451 	if (status)
2452 		return status;
2453 
2454 	memset(&zgid, 0, sizeof(zgid));
2455 	if (!memcmp(&sgid, &zgid, sizeof(zgid)))
2456 		return -EINVAL;
2457 
2458 	qp->sgid_idx = ah_attr->grh.sgid_index;
2459 	memcpy(&cmd->params.sgid[0], &sgid.raw[0], sizeof(cmd->params.sgid));
2460 	status = ocrdma_resolve_dmac(dev, ah_attr, &mac_addr[0]);
2461 	if (status)
2462 		return status;
2463 	cmd->params.dmac_b0_to_b3 = mac_addr[0] | (mac_addr[1] << 8) |
2464 				(mac_addr[2] << 16) | (mac_addr[3] << 24);
2465 	/* convert them to LE format. */
2466 	ocrdma_cpu_to_le32(&cmd->params.dgid[0], sizeof(cmd->params.dgid));
2467 	ocrdma_cpu_to_le32(&cmd->params.sgid[0], sizeof(cmd->params.sgid));
2468 	cmd->params.vlan_dmac_b4_to_b5 = mac_addr[4] | (mac_addr[5] << 8);
2469 	if (attr_mask & IB_QP_VID) {
2470 		vlan_id = attrs->vlan_id;
2471 		cmd->params.vlan_dmac_b4_to_b5 |=
2472 		    vlan_id << OCRDMA_QP_PARAMS_VLAN_SHIFT;
2473 		cmd->flags |= OCRDMA_QP_PARA_VLAN_EN_VALID;
2474 		cmd->params.rnt_rc_sl_fl |=
2475 			(dev->sl & 0x07) << OCRDMA_QP_PARAMS_SL_SHIFT;
2476 	}
2477 	return 0;
2478 }
2479 
2480 static int ocrdma_set_qp_params(struct ocrdma_qp *qp,
2481 				struct ocrdma_modify_qp *cmd,
2482 				struct ib_qp_attr *attrs, int attr_mask)
2483 {
2484 	int status = 0;
2485 	struct ocrdma_dev *dev = get_ocrdma_dev(qp->ibqp.device);
2486 
2487 	if (attr_mask & IB_QP_PKEY_INDEX) {
2488 		cmd->params.path_mtu_pkey_indx |= (attrs->pkey_index &
2489 					    OCRDMA_QP_PARAMS_PKEY_INDEX_MASK);
2490 		cmd->flags |= OCRDMA_QP_PARA_PKEY_VALID;
2491 	}
2492 	if (attr_mask & IB_QP_QKEY) {
2493 		qp->qkey = attrs->qkey;
2494 		cmd->params.qkey = attrs->qkey;
2495 		cmd->flags |= OCRDMA_QP_PARA_QKEY_VALID;
2496 	}
2497 	if (attr_mask & IB_QP_AV) {
2498 		status = ocrdma_set_av_params(qp, cmd, attrs, attr_mask);
2499 		if (status)
2500 			return status;
2501 	} else if (qp->qp_type == IB_QPT_GSI || qp->qp_type == IB_QPT_UD) {
2502 		/* set the default mac address for UD, GSI QPs */
2503 		cmd->params.dmac_b0_to_b3 = dev->nic_info.mac_addr[0] |
2504 			(dev->nic_info.mac_addr[1] << 8) |
2505 			(dev->nic_info.mac_addr[2] << 16) |
2506 			(dev->nic_info.mac_addr[3] << 24);
2507 		cmd->params.vlan_dmac_b4_to_b5 = dev->nic_info.mac_addr[4] |
2508 					(dev->nic_info.mac_addr[5] << 8);
2509 	}
2510 	if ((attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY) &&
2511 	    attrs->en_sqd_async_notify) {
2512 		cmd->params.max_sge_recv_flags |=
2513 			OCRDMA_QP_PARAMS_FLAGS_SQD_ASYNC;
2514 		cmd->flags |= OCRDMA_QP_PARA_DST_QPN_VALID;
2515 	}
2516 	if (attr_mask & IB_QP_DEST_QPN) {
2517 		cmd->params.ack_to_rnr_rtc_dest_qpn |= (attrs->dest_qp_num &
2518 				OCRDMA_QP_PARAMS_DEST_QPN_MASK);
2519 		cmd->flags |= OCRDMA_QP_PARA_DST_QPN_VALID;
2520 	}
2521 	if (attr_mask & IB_QP_PATH_MTU) {
2522 		if (attrs->path_mtu < IB_MTU_256 ||
2523 		    attrs->path_mtu > IB_MTU_4096) {
2524 			status = -EINVAL;
2525 			goto pmtu_err;
2526 		}
2527 		cmd->params.path_mtu_pkey_indx |=
2528 		    (ib_mtu_enum_to_int(attrs->path_mtu) <<
2529 		     OCRDMA_QP_PARAMS_PATH_MTU_SHIFT) &
2530 		    OCRDMA_QP_PARAMS_PATH_MTU_MASK;
2531 		cmd->flags |= OCRDMA_QP_PARA_PMTU_VALID;
2532 	}
2533 	if (attr_mask & IB_QP_TIMEOUT) {
2534 		cmd->params.ack_to_rnr_rtc_dest_qpn |= attrs->timeout <<
2535 		    OCRDMA_QP_PARAMS_ACK_TIMEOUT_SHIFT;
2536 		cmd->flags |= OCRDMA_QP_PARA_ACK_TO_VALID;
2537 	}
2538 	if (attr_mask & IB_QP_RETRY_CNT) {
2539 		cmd->params.rnt_rc_sl_fl |= (attrs->retry_cnt <<
2540 				      OCRDMA_QP_PARAMS_RETRY_CNT_SHIFT) &
2541 		    OCRDMA_QP_PARAMS_RETRY_CNT_MASK;
2542 		cmd->flags |= OCRDMA_QP_PARA_RETRY_CNT_VALID;
2543 	}
2544 	if (attr_mask & IB_QP_MIN_RNR_TIMER) {
2545 		cmd->params.rnt_rc_sl_fl |= (attrs->min_rnr_timer <<
2546 				      OCRDMA_QP_PARAMS_RNR_NAK_TIMER_SHIFT) &
2547 		    OCRDMA_QP_PARAMS_RNR_NAK_TIMER_MASK;
2548 		cmd->flags |= OCRDMA_QP_PARA_RNT_VALID;
2549 	}
2550 	if (attr_mask & IB_QP_RNR_RETRY) {
2551 		cmd->params.ack_to_rnr_rtc_dest_qpn |= (attrs->rnr_retry <<
2552 			OCRDMA_QP_PARAMS_RNR_RETRY_CNT_SHIFT)
2553 			& OCRDMA_QP_PARAMS_RNR_RETRY_CNT_MASK;
2554 		cmd->flags |= OCRDMA_QP_PARA_RRC_VALID;
2555 	}
2556 	if (attr_mask & IB_QP_SQ_PSN) {
2557 		cmd->params.tclass_sq_psn |= (attrs->sq_psn & 0x00ffffff);
2558 		cmd->flags |= OCRDMA_QP_PARA_SQPSN_VALID;
2559 	}
2560 	if (attr_mask & IB_QP_RQ_PSN) {
2561 		cmd->params.hop_lmt_rq_psn |= (attrs->rq_psn & 0x00ffffff);
2562 		cmd->flags |= OCRDMA_QP_PARA_RQPSN_VALID;
2563 	}
2564 	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
2565 		if (attrs->max_rd_atomic > dev->attr.max_ord_per_qp) {
2566 			status = -EINVAL;
2567 			goto pmtu_err;
2568 		}
2569 		qp->max_ord = attrs->max_rd_atomic;
2570 		cmd->flags |= OCRDMA_QP_PARA_MAX_ORD_VALID;
2571 	}
2572 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
2573 		if (attrs->max_dest_rd_atomic > dev->attr.max_ird_per_qp) {
2574 			status = -EINVAL;
2575 			goto pmtu_err;
2576 		}
2577 		qp->max_ird = attrs->max_dest_rd_atomic;
2578 		cmd->flags |= OCRDMA_QP_PARA_MAX_IRD_VALID;
2579 	}
2580 	cmd->params.max_ord_ird = (qp->max_ord <<
2581 				OCRDMA_QP_PARAMS_MAX_ORD_SHIFT) |
2582 				(qp->max_ird & OCRDMA_QP_PARAMS_MAX_IRD_MASK);
2583 pmtu_err:
2584 	return status;
2585 }
2586 
2587 int ocrdma_mbx_modify_qp(struct ocrdma_dev *dev, struct ocrdma_qp *qp,
2588 			 struct ib_qp_attr *attrs, int attr_mask)
2589 {
2590 	int status = -ENOMEM;
2591 	struct ocrdma_modify_qp *cmd;
2592 
2593 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_MODIFY_QP, sizeof(*cmd));
2594 	if (!cmd)
2595 		return status;
2596 
2597 	cmd->params.id = qp->id;
2598 	cmd->flags = 0;
2599 	if (attr_mask & IB_QP_STATE) {
2600 		cmd->params.max_sge_recv_flags |=
2601 		    (get_ocrdma_qp_state(attrs->qp_state) <<
2602 		     OCRDMA_QP_PARAMS_STATE_SHIFT) &
2603 		    OCRDMA_QP_PARAMS_STATE_MASK;
2604 		cmd->flags |= OCRDMA_QP_PARA_QPS_VALID;
2605 	} else {
2606 		cmd->params.max_sge_recv_flags |=
2607 		    (qp->state << OCRDMA_QP_PARAMS_STATE_SHIFT) &
2608 		    OCRDMA_QP_PARAMS_STATE_MASK;
2609 	}
2610 
2611 	status = ocrdma_set_qp_params(qp, cmd, attrs, attr_mask);
2612 	if (status)
2613 		goto mbx_err;
2614 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2615 	if (status)
2616 		goto mbx_err;
2617 
2618 mbx_err:
2619 	kfree(cmd);
2620 	return status;
2621 }
2622 
2623 int ocrdma_mbx_destroy_qp(struct ocrdma_dev *dev, struct ocrdma_qp *qp)
2624 {
2625 	int status = -ENOMEM;
2626 	struct ocrdma_destroy_qp *cmd;
2627 	struct pci_dev *pdev = dev->nic_info.pdev;
2628 
2629 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DELETE_QP, sizeof(*cmd));
2630 	if (!cmd)
2631 		return status;
2632 	cmd->qp_id = qp->id;
2633 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2634 	if (status)
2635 		goto mbx_err;
2636 
2637 mbx_err:
2638 	kfree(cmd);
2639 	if (qp->sq.va)
2640 		dma_free_coherent(&pdev->dev, qp->sq.len, qp->sq.va, qp->sq.pa);
2641 	if (!qp->srq && qp->rq.va)
2642 		dma_free_coherent(&pdev->dev, qp->rq.len, qp->rq.va, qp->rq.pa);
2643 	if (qp->dpp_enabled)
2644 		qp->pd->num_dpp_qp++;
2645 	return status;
2646 }
2647 
2648 int ocrdma_mbx_create_srq(struct ocrdma_dev *dev, struct ocrdma_srq *srq,
2649 			  struct ib_srq_init_attr *srq_attr,
2650 			  struct ocrdma_pd *pd)
2651 {
2652 	int status = -ENOMEM;
2653 	int hw_pages, hw_page_size;
2654 	int len;
2655 	struct ocrdma_create_srq_rsp *rsp;
2656 	struct ocrdma_create_srq *cmd;
2657 	dma_addr_t pa;
2658 	struct pci_dev *pdev = dev->nic_info.pdev;
2659 	u32 max_rqe_allocated;
2660 
2661 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_CREATE_SRQ, sizeof(*cmd));
2662 	if (!cmd)
2663 		return status;
2664 
2665 	cmd->pgsz_pdid = pd->id & OCRDMA_CREATE_SRQ_PD_ID_MASK;
2666 	max_rqe_allocated = srq_attr->attr.max_wr + 1;
2667 	status = ocrdma_build_q_conf(&max_rqe_allocated,
2668 				dev->attr.rqe_size,
2669 				&hw_pages, &hw_page_size);
2670 	if (status) {
2671 		pr_err("%s() req. max_wr=0x%x\n", __func__,
2672 		       srq_attr->attr.max_wr);
2673 		status = -EINVAL;
2674 		goto ret;
2675 	}
2676 	len = hw_pages * hw_page_size;
2677 	srq->rq.va = dma_alloc_coherent(&pdev->dev, len, &pa, GFP_KERNEL);
2678 	if (!srq->rq.va) {
2679 		status = -ENOMEM;
2680 		goto ret;
2681 	}
2682 	ocrdma_build_q_pages(&cmd->rq_addr[0], hw_pages, pa, hw_page_size);
2683 
2684 	srq->rq.entry_size = dev->attr.rqe_size;
2685 	srq->rq.pa = pa;
2686 	srq->rq.len = len;
2687 	srq->rq.max_cnt = max_rqe_allocated;
2688 
2689 	cmd->max_sge_rqe = ilog2(max_rqe_allocated);
2690 	cmd->max_sge_rqe |= srq_attr->attr.max_sge <<
2691 				OCRDMA_CREATE_SRQ_MAX_SGE_RECV_SHIFT;
2692 
2693 	cmd->pgsz_pdid |= (ilog2(hw_page_size / OCRDMA_MIN_Q_PAGE_SIZE)
2694 		<< OCRDMA_CREATE_SRQ_PG_SZ_SHIFT);
2695 	cmd->pages_rqe_sz |= (dev->attr.rqe_size
2696 		<< OCRDMA_CREATE_SRQ_RQE_SIZE_SHIFT)
2697 		& OCRDMA_CREATE_SRQ_RQE_SIZE_MASK;
2698 	cmd->pages_rqe_sz |= hw_pages << OCRDMA_CREATE_SRQ_NUM_RQ_PAGES_SHIFT;
2699 
2700 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2701 	if (status)
2702 		goto mbx_err;
2703 	rsp = (struct ocrdma_create_srq_rsp *)cmd;
2704 	srq->id = rsp->id;
2705 	srq->rq.dbid = rsp->id;
2706 	max_rqe_allocated = ((rsp->max_sge_rqe_allocated &
2707 		OCRDMA_CREATE_SRQ_RSP_MAX_RQE_ALLOCATED_MASK) >>
2708 		OCRDMA_CREATE_SRQ_RSP_MAX_RQE_ALLOCATED_SHIFT);
2709 	max_rqe_allocated = (1 << max_rqe_allocated);
2710 	srq->rq.max_cnt = max_rqe_allocated;
2711 	srq->rq.max_wqe_idx = max_rqe_allocated - 1;
2712 	srq->rq.max_sges = (rsp->max_sge_rqe_allocated &
2713 		OCRDMA_CREATE_SRQ_RSP_MAX_SGE_RECV_ALLOCATED_MASK) >>
2714 		OCRDMA_CREATE_SRQ_RSP_MAX_SGE_RECV_ALLOCATED_SHIFT;
2715 	goto ret;
2716 mbx_err:
2717 	dma_free_coherent(&pdev->dev, srq->rq.len, srq->rq.va, pa);
2718 ret:
2719 	kfree(cmd);
2720 	return status;
2721 }
2722 
2723 int ocrdma_mbx_modify_srq(struct ocrdma_srq *srq, struct ib_srq_attr *srq_attr)
2724 {
2725 	int status = -ENOMEM;
2726 	struct ocrdma_modify_srq *cmd;
2727 	struct ocrdma_pd *pd = srq->pd;
2728 	struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device);
2729 
2730 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_MODIFY_SRQ, sizeof(*cmd));
2731 	if (!cmd)
2732 		return status;
2733 	cmd->id = srq->id;
2734 	cmd->limit_max_rqe |= srq_attr->srq_limit <<
2735 	    OCRDMA_MODIFY_SRQ_LIMIT_SHIFT;
2736 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2737 	kfree(cmd);
2738 	return status;
2739 }
2740 
2741 int ocrdma_mbx_query_srq(struct ocrdma_srq *srq, struct ib_srq_attr *srq_attr)
2742 {
2743 	int status = -ENOMEM;
2744 	struct ocrdma_query_srq *cmd;
2745 	struct ocrdma_dev *dev = get_ocrdma_dev(srq->ibsrq.device);
2746 
2747 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_SRQ, sizeof(*cmd));
2748 	if (!cmd)
2749 		return status;
2750 	cmd->id = srq->rq.dbid;
2751 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2752 	if (status == 0) {
2753 		struct ocrdma_query_srq_rsp *rsp =
2754 		    (struct ocrdma_query_srq_rsp *)cmd;
2755 		srq_attr->max_sge =
2756 		    rsp->srq_lmt_max_sge &
2757 		    OCRDMA_QUERY_SRQ_RSP_MAX_SGE_RECV_MASK;
2758 		srq_attr->max_wr =
2759 		    rsp->max_rqe_pdid >> OCRDMA_QUERY_SRQ_RSP_MAX_RQE_SHIFT;
2760 		srq_attr->srq_limit = rsp->srq_lmt_max_sge >>
2761 		    OCRDMA_QUERY_SRQ_RSP_SRQ_LIMIT_SHIFT;
2762 	}
2763 	kfree(cmd);
2764 	return status;
2765 }
2766 
2767 int ocrdma_mbx_destroy_srq(struct ocrdma_dev *dev, struct ocrdma_srq *srq)
2768 {
2769 	int status = -ENOMEM;
2770 	struct ocrdma_destroy_srq *cmd;
2771 	struct pci_dev *pdev = dev->nic_info.pdev;
2772 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DELETE_SRQ, sizeof(*cmd));
2773 	if (!cmd)
2774 		return status;
2775 	cmd->id = srq->id;
2776 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2777 	if (srq->rq.va)
2778 		dma_free_coherent(&pdev->dev, srq->rq.len,
2779 				  srq->rq.va, srq->rq.pa);
2780 	kfree(cmd);
2781 	return status;
2782 }
2783 
2784 static int ocrdma_mbx_get_dcbx_config(struct ocrdma_dev *dev, u32 ptype,
2785 				      struct ocrdma_dcbx_cfg *dcbxcfg)
2786 {
2787 	int status = 0;
2788 	dma_addr_t pa;
2789 	struct ocrdma_mqe cmd;
2790 
2791 	struct ocrdma_get_dcbx_cfg_req *req = NULL;
2792 	struct ocrdma_get_dcbx_cfg_rsp *rsp = NULL;
2793 	struct pci_dev *pdev = dev->nic_info.pdev;
2794 	struct ocrdma_mqe_sge *mqe_sge = cmd.u.nonemb_req.sge;
2795 
2796 	memset(&cmd, 0, sizeof(struct ocrdma_mqe));
2797 	cmd.hdr.pyld_len = max_t (u32, sizeof(struct ocrdma_get_dcbx_cfg_rsp),
2798 					sizeof(struct ocrdma_get_dcbx_cfg_req));
2799 	req = dma_alloc_coherent(&pdev->dev, cmd.hdr.pyld_len, &pa, GFP_KERNEL);
2800 	if (!req) {
2801 		status = -ENOMEM;
2802 		goto mem_err;
2803 	}
2804 
2805 	cmd.hdr.spcl_sge_cnt_emb |= (1 << OCRDMA_MQE_HDR_SGE_CNT_SHIFT) &
2806 					OCRDMA_MQE_HDR_SGE_CNT_MASK;
2807 	mqe_sge->pa_lo = (u32) (pa & 0xFFFFFFFFUL);
2808 	mqe_sge->pa_hi = (u32) upper_32_bits(pa);
2809 	mqe_sge->len = cmd.hdr.pyld_len;
2810 
2811 	memset(req, 0, sizeof(struct ocrdma_get_dcbx_cfg_req));
2812 	ocrdma_init_mch(&req->hdr, OCRDMA_CMD_GET_DCBX_CONFIG,
2813 			OCRDMA_SUBSYS_DCBX, cmd.hdr.pyld_len);
2814 	req->param_type = ptype;
2815 
2816 	status = ocrdma_mbx_cmd(dev, &cmd);
2817 	if (status)
2818 		goto mbx_err;
2819 
2820 	rsp = (struct ocrdma_get_dcbx_cfg_rsp *)req;
2821 	ocrdma_le32_to_cpu(rsp, sizeof(struct ocrdma_get_dcbx_cfg_rsp));
2822 	memcpy(dcbxcfg, &rsp->cfg, sizeof(struct ocrdma_dcbx_cfg));
2823 
2824 mbx_err:
2825 	dma_free_coherent(&pdev->dev, cmd.hdr.pyld_len, req, pa);
2826 mem_err:
2827 	return status;
2828 }
2829 
2830 #define OCRDMA_MAX_SERVICE_LEVEL_INDEX	0x08
2831 #define OCRDMA_DEFAULT_SERVICE_LEVEL	0x05
2832 
2833 static int ocrdma_parse_dcbxcfg_rsp(struct ocrdma_dev *dev, int ptype,
2834 				    struct ocrdma_dcbx_cfg *dcbxcfg,
2835 				    u8 *srvc_lvl)
2836 {
2837 	int status = -EINVAL, indx, slindx;
2838 	int ventry_cnt;
2839 	struct ocrdma_app_parameter *app_param;
2840 	u8 valid, proto_sel;
2841 	u8 app_prio, pfc_prio;
2842 	u16 proto;
2843 
2844 	if (!(dcbxcfg->tcv_aev_opv_st & OCRDMA_DCBX_STATE_MASK)) {
2845 		pr_info("%s ocrdma%d DCBX is disabled\n",
2846 			dev_name(&dev->nic_info.pdev->dev), dev->id);
2847 		goto out;
2848 	}
2849 
2850 	if (!ocrdma_is_enabled_and_synced(dcbxcfg->pfc_state)) {
2851 		pr_info("%s ocrdma%d priority flow control(%s) is %s%s\n",
2852 			dev_name(&dev->nic_info.pdev->dev), dev->id,
2853 			(ptype > 0 ? "operational" : "admin"),
2854 			(dcbxcfg->pfc_state & OCRDMA_STATE_FLAG_ENABLED) ?
2855 			"enabled" : "disabled",
2856 			(dcbxcfg->pfc_state & OCRDMA_STATE_FLAG_SYNC) ?
2857 			"" : ", not sync'ed");
2858 		goto out;
2859 	} else {
2860 		pr_info("%s ocrdma%d priority flow control is enabled and sync'ed\n",
2861 			dev_name(&dev->nic_info.pdev->dev), dev->id);
2862 	}
2863 
2864 	ventry_cnt = (dcbxcfg->tcv_aev_opv_st >>
2865 				OCRDMA_DCBX_APP_ENTRY_SHIFT)
2866 				& OCRDMA_DCBX_STATE_MASK;
2867 
2868 	for (indx = 0; indx < ventry_cnt; indx++) {
2869 		app_param = &dcbxcfg->app_param[indx];
2870 		valid = (app_param->valid_proto_app >>
2871 				OCRDMA_APP_PARAM_VALID_SHIFT)
2872 				& OCRDMA_APP_PARAM_VALID_MASK;
2873 		proto_sel = (app_param->valid_proto_app
2874 				>>  OCRDMA_APP_PARAM_PROTO_SEL_SHIFT)
2875 				& OCRDMA_APP_PARAM_PROTO_SEL_MASK;
2876 		proto = app_param->valid_proto_app &
2877 				OCRDMA_APP_PARAM_APP_PROTO_MASK;
2878 
2879 		if (
2880 			valid && proto == OCRDMA_APP_PROTO_ROCE &&
2881 			proto_sel == OCRDMA_PROTO_SELECT_L2) {
2882 			for (slindx = 0; slindx <
2883 				OCRDMA_MAX_SERVICE_LEVEL_INDEX; slindx++) {
2884 				app_prio = ocrdma_get_app_prio(
2885 						(u8 *)app_param->app_prio,
2886 						slindx);
2887 				pfc_prio = ocrdma_get_pfc_prio(
2888 						(u8 *)dcbxcfg->pfc_prio,
2889 						slindx);
2890 
2891 				if (app_prio && pfc_prio) {
2892 					*srvc_lvl = slindx;
2893 					status = 0;
2894 					goto out;
2895 				}
2896 			}
2897 			if (slindx == OCRDMA_MAX_SERVICE_LEVEL_INDEX) {
2898 				pr_info("%s ocrdma%d application priority not set for 0x%x protocol\n",
2899 					dev_name(&dev->nic_info.pdev->dev),
2900 					dev->id, proto);
2901 			}
2902 		}
2903 	}
2904 
2905 out:
2906 	return status;
2907 }
2908 
2909 void ocrdma_init_service_level(struct ocrdma_dev *dev)
2910 {
2911 	int status = 0, indx;
2912 	struct ocrdma_dcbx_cfg dcbxcfg;
2913 	u8 srvc_lvl = OCRDMA_DEFAULT_SERVICE_LEVEL;
2914 	int ptype = OCRDMA_PARAMETER_TYPE_OPER;
2915 
2916 	for (indx = 0; indx < 2; indx++) {
2917 		status = ocrdma_mbx_get_dcbx_config(dev, ptype, &dcbxcfg);
2918 		if (status) {
2919 			pr_err("%s(): status=%d\n", __func__, status);
2920 			ptype = OCRDMA_PARAMETER_TYPE_ADMIN;
2921 			continue;
2922 		}
2923 
2924 		status = ocrdma_parse_dcbxcfg_rsp(dev, ptype,
2925 						  &dcbxcfg, &srvc_lvl);
2926 		if (status) {
2927 			ptype = OCRDMA_PARAMETER_TYPE_ADMIN;
2928 			continue;
2929 		}
2930 
2931 		break;
2932 	}
2933 
2934 	if (status)
2935 		pr_info("%s ocrdma%d service level default\n",
2936 			dev_name(&dev->nic_info.pdev->dev), dev->id);
2937 	else
2938 		pr_info("%s ocrdma%d service level %d\n",
2939 			dev_name(&dev->nic_info.pdev->dev), dev->id,
2940 			srvc_lvl);
2941 
2942 	dev->pfc_state = ocrdma_is_enabled_and_synced(dcbxcfg.pfc_state);
2943 	dev->sl = srvc_lvl;
2944 }
2945 
2946 int ocrdma_alloc_av(struct ocrdma_dev *dev, struct ocrdma_ah *ah)
2947 {
2948 	int i;
2949 	int status = -EINVAL;
2950 	struct ocrdma_av *av;
2951 	unsigned long flags;
2952 
2953 	av = dev->av_tbl.va;
2954 	spin_lock_irqsave(&dev->av_tbl.lock, flags);
2955 	for (i = 0; i < dev->av_tbl.num_ah; i++) {
2956 		if (av->valid == 0) {
2957 			av->valid = OCRDMA_AV_VALID;
2958 			ah->av = av;
2959 			ah->id = i;
2960 			status = 0;
2961 			break;
2962 		}
2963 		av++;
2964 	}
2965 	if (i == dev->av_tbl.num_ah)
2966 		status = -EAGAIN;
2967 	spin_unlock_irqrestore(&dev->av_tbl.lock, flags);
2968 	return status;
2969 }
2970 
2971 int ocrdma_free_av(struct ocrdma_dev *dev, struct ocrdma_ah *ah)
2972 {
2973 	unsigned long flags;
2974 	spin_lock_irqsave(&dev->av_tbl.lock, flags);
2975 	ah->av->valid = 0;
2976 	spin_unlock_irqrestore(&dev->av_tbl.lock, flags);
2977 	return 0;
2978 }
2979 
2980 static int ocrdma_create_eqs(struct ocrdma_dev *dev)
2981 {
2982 	int num_eq, i, status = 0;
2983 	int irq;
2984 	unsigned long flags = 0;
2985 
2986 	num_eq = dev->nic_info.msix.num_vectors -
2987 			dev->nic_info.msix.start_vector;
2988 	if (dev->nic_info.intr_mode == BE_INTERRUPT_MODE_INTX) {
2989 		num_eq = 1;
2990 		flags = IRQF_SHARED;
2991 	} else {
2992 		num_eq = min_t(u32, num_eq, num_online_cpus());
2993 	}
2994 
2995 	if (!num_eq)
2996 		return -EINVAL;
2997 
2998 	dev->eq_tbl = kzalloc(sizeof(struct ocrdma_eq) * num_eq, GFP_KERNEL);
2999 	if (!dev->eq_tbl)
3000 		return -ENOMEM;
3001 
3002 	for (i = 0; i < num_eq; i++) {
3003 		status = ocrdma_create_eq(dev, &dev->eq_tbl[i],
3004 					OCRDMA_EQ_LEN);
3005 		if (status) {
3006 			status = -EINVAL;
3007 			break;
3008 		}
3009 		sprintf(dev->eq_tbl[i].irq_name, "ocrdma%d-%d",
3010 			dev->id, i);
3011 		irq = ocrdma_get_irq(dev, &dev->eq_tbl[i]);
3012 		status = request_irq(irq, ocrdma_irq_handler, flags,
3013 				     dev->eq_tbl[i].irq_name,
3014 				     &dev->eq_tbl[i]);
3015 		if (status)
3016 			goto done;
3017 		dev->eq_cnt += 1;
3018 	}
3019 	/* one eq is sufficient for data path to work */
3020 	return 0;
3021 done:
3022 	ocrdma_destroy_eqs(dev);
3023 	return status;
3024 }
3025 
3026 static int ocrdma_mbx_modify_eqd(struct ocrdma_dev *dev, struct ocrdma_eq *eq,
3027 				 int num)
3028 {
3029 	int i, status = -ENOMEM;
3030 	struct ocrdma_modify_eqd_req *cmd;
3031 
3032 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_MODIFY_EQ_DELAY, sizeof(*cmd));
3033 	if (!cmd)
3034 		return status;
3035 
3036 	ocrdma_init_mch(&cmd->cmd.req, OCRDMA_CMD_MODIFY_EQ_DELAY,
3037 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
3038 
3039 	cmd->cmd.num_eq = num;
3040 	for (i = 0; i < num; i++) {
3041 		cmd->cmd.set_eqd[i].eq_id = eq[i].q.id;
3042 		cmd->cmd.set_eqd[i].phase = 0;
3043 		cmd->cmd.set_eqd[i].delay_multiplier =
3044 				(eq[i].aic_obj.prev_eqd * 65)/100;
3045 	}
3046 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
3047 	if (status)
3048 		goto mbx_err;
3049 mbx_err:
3050 	kfree(cmd);
3051 	return status;
3052 }
3053 
3054 static int ocrdma_modify_eqd(struct ocrdma_dev *dev, struct ocrdma_eq *eq,
3055 			     int num)
3056 {
3057 	int num_eqs, i = 0;
3058 	if (num > 8) {
3059 		while (num) {
3060 			num_eqs = min(num, 8);
3061 			ocrdma_mbx_modify_eqd(dev, &eq[i], num_eqs);
3062 			i += num_eqs;
3063 			num -= num_eqs;
3064 		}
3065 	} else {
3066 		ocrdma_mbx_modify_eqd(dev, eq, num);
3067 	}
3068 	return 0;
3069 }
3070 
3071 void ocrdma_eqd_set_task(struct work_struct *work)
3072 {
3073 	struct ocrdma_dev *dev =
3074 		container_of(work, struct ocrdma_dev, eqd_work.work);
3075 	struct ocrdma_eq *eq = 0;
3076 	int i, num = 0, status = -EINVAL;
3077 	u64 eq_intr;
3078 
3079 	for (i = 0; i < dev->eq_cnt; i++) {
3080 		eq = &dev->eq_tbl[i];
3081 		if (eq->aic_obj.eq_intr_cnt > eq->aic_obj.prev_eq_intr_cnt) {
3082 			eq_intr = eq->aic_obj.eq_intr_cnt -
3083 				  eq->aic_obj.prev_eq_intr_cnt;
3084 			if ((eq_intr > EQ_INTR_PER_SEC_THRSH_HI) &&
3085 			    (eq->aic_obj.prev_eqd == EQ_AIC_MIN_EQD)) {
3086 				eq->aic_obj.prev_eqd = EQ_AIC_MAX_EQD;
3087 				num++;
3088 			} else if ((eq_intr < EQ_INTR_PER_SEC_THRSH_LOW) &&
3089 				   (eq->aic_obj.prev_eqd == EQ_AIC_MAX_EQD)) {
3090 				eq->aic_obj.prev_eqd = EQ_AIC_MIN_EQD;
3091 				num++;
3092 			}
3093 		}
3094 		eq->aic_obj.prev_eq_intr_cnt = eq->aic_obj.eq_intr_cnt;
3095 	}
3096 
3097 	if (num)
3098 		status = ocrdma_modify_eqd(dev, &dev->eq_tbl[0], num);
3099 	schedule_delayed_work(&dev->eqd_work, msecs_to_jiffies(1000));
3100 }
3101 
3102 int ocrdma_init_hw(struct ocrdma_dev *dev)
3103 {
3104 	int status;
3105 
3106 	/* create the eqs  */
3107 	status = ocrdma_create_eqs(dev);
3108 	if (status)
3109 		goto qpeq_err;
3110 	status = ocrdma_create_mq(dev);
3111 	if (status)
3112 		goto mq_err;
3113 	status = ocrdma_mbx_query_fw_config(dev);
3114 	if (status)
3115 		goto conf_err;
3116 	status = ocrdma_mbx_query_dev(dev);
3117 	if (status)
3118 		goto conf_err;
3119 	status = ocrdma_mbx_query_fw_ver(dev);
3120 	if (status)
3121 		goto conf_err;
3122 	status = ocrdma_mbx_create_ah_tbl(dev);
3123 	if (status)
3124 		goto conf_err;
3125 	status = ocrdma_mbx_get_phy_info(dev);
3126 	if (status)
3127 		goto info_attrb_err;
3128 	status = ocrdma_mbx_get_ctrl_attribs(dev);
3129 	if (status)
3130 		goto info_attrb_err;
3131 
3132 	return 0;
3133 
3134 info_attrb_err:
3135 	ocrdma_mbx_delete_ah_tbl(dev);
3136 conf_err:
3137 	ocrdma_destroy_mq(dev);
3138 mq_err:
3139 	ocrdma_destroy_eqs(dev);
3140 qpeq_err:
3141 	pr_err("%s() status=%d\n", __func__, status);
3142 	return status;
3143 }
3144 
3145 void ocrdma_cleanup_hw(struct ocrdma_dev *dev)
3146 {
3147 	ocrdma_free_pd_pool(dev);
3148 	ocrdma_mbx_delete_ah_tbl(dev);
3149 
3150 	/* cleanup the eqs */
3151 	ocrdma_destroy_eqs(dev);
3152 
3153 	/* cleanup the control path */
3154 	ocrdma_destroy_mq(dev);
3155 }
3156