1 /* This file is part of the Emulex RoCE Device Driver for
2  * RoCE (RDMA over Converged Ethernet) adapters.
3  * Copyright (C) 2012-2015 Emulex. All rights reserved.
4  * EMULEX and SLI are trademarks of Emulex.
5  * www.emulex.com
6  *
7  * This software is available to you under a choice of one of two licenses.
8  * You may choose to be licensed under the terms of the GNU General Public
9  * License (GPL) Version 2, available from the file COPYING in the main
10  * directory of this source tree, or the BSD license below:
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  *
16  * - Redistributions of source code must retain the above copyright notice,
17  *   this list of conditions and the following disclaimer.
18  *
19  * - Redistributions in binary form must reproduce the above copyright
20  *   notice, this list of conditions and the following disclaimer in
21  *   the documentation and/or other materials provided with the distribution.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
24  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
27  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
31  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
32  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
33  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34  *
35  * Contact Information:
36  * linux-drivers@emulex.com
37  *
38  * Emulex
39  * 3333 Susan Street
40  * Costa Mesa, CA 92626
41  */
42 
43 #include <linux/sched.h>
44 #include <linux/interrupt.h>
45 #include <linux/log2.h>
46 #include <linux/dma-mapping.h>
47 
48 #include <rdma/ib_verbs.h>
49 #include <rdma/ib_user_verbs.h>
50 #include <rdma/ib_cache.h>
51 
52 #include "ocrdma.h"
53 #include "ocrdma_hw.h"
54 #include "ocrdma_verbs.h"
55 #include "ocrdma_ah.h"
56 
57 enum mbx_status {
58 	OCRDMA_MBX_STATUS_FAILED		= 1,
59 	OCRDMA_MBX_STATUS_ILLEGAL_FIELD		= 3,
60 	OCRDMA_MBX_STATUS_OOR			= 100,
61 	OCRDMA_MBX_STATUS_INVALID_PD		= 101,
62 	OCRDMA_MBX_STATUS_PD_INUSE		= 102,
63 	OCRDMA_MBX_STATUS_INVALID_CQ		= 103,
64 	OCRDMA_MBX_STATUS_INVALID_QP		= 104,
65 	OCRDMA_MBX_STATUS_INVALID_LKEY		= 105,
66 	OCRDMA_MBX_STATUS_ORD_EXCEEDS		= 106,
67 	OCRDMA_MBX_STATUS_IRD_EXCEEDS		= 107,
68 	OCRDMA_MBX_STATUS_SENDQ_WQE_EXCEEDS	= 108,
69 	OCRDMA_MBX_STATUS_RECVQ_RQE_EXCEEDS	= 109,
70 	OCRDMA_MBX_STATUS_SGE_SEND_EXCEEDS	= 110,
71 	OCRDMA_MBX_STATUS_SGE_WRITE_EXCEEDS	= 111,
72 	OCRDMA_MBX_STATUS_SGE_RECV_EXCEEDS	= 112,
73 	OCRDMA_MBX_STATUS_INVALID_STATE_CHANGE	= 113,
74 	OCRDMA_MBX_STATUS_MW_BOUND		= 114,
75 	OCRDMA_MBX_STATUS_INVALID_VA		= 115,
76 	OCRDMA_MBX_STATUS_INVALID_LENGTH	= 116,
77 	OCRDMA_MBX_STATUS_INVALID_FBO		= 117,
78 	OCRDMA_MBX_STATUS_INVALID_ACC_RIGHTS	= 118,
79 	OCRDMA_MBX_STATUS_INVALID_PBE_SIZE	= 119,
80 	OCRDMA_MBX_STATUS_INVALID_PBL_ENTRY	= 120,
81 	OCRDMA_MBX_STATUS_INVALID_PBL_SHIFT	= 121,
82 	OCRDMA_MBX_STATUS_INVALID_SRQ_ID	= 129,
83 	OCRDMA_MBX_STATUS_SRQ_ERROR		= 133,
84 	OCRDMA_MBX_STATUS_RQE_EXCEEDS		= 134,
85 	OCRDMA_MBX_STATUS_MTU_EXCEEDS		= 135,
86 	OCRDMA_MBX_STATUS_MAX_QP_EXCEEDS	= 136,
87 	OCRDMA_MBX_STATUS_SRQ_LIMIT_EXCEEDS	= 137,
88 	OCRDMA_MBX_STATUS_SRQ_SIZE_UNDERUNS	= 138,
89 	OCRDMA_MBX_STATUS_QP_BOUND		= 130,
90 	OCRDMA_MBX_STATUS_INVALID_CHANGE	= 139,
91 	OCRDMA_MBX_STATUS_ATOMIC_OPS_UNSUP	= 140,
92 	OCRDMA_MBX_STATUS_INVALID_RNR_NAK_TIMER	= 141,
93 	OCRDMA_MBX_STATUS_MW_STILL_BOUND	= 142,
94 	OCRDMA_MBX_STATUS_PKEY_INDEX_INVALID	= 143,
95 	OCRDMA_MBX_STATUS_PKEY_INDEX_EXCEEDS	= 144
96 };
97 
98 enum additional_status {
99 	OCRDMA_MBX_ADDI_STATUS_INSUFFICIENT_RESOURCES = 22
100 };
101 
102 enum cqe_status {
103 	OCRDMA_MBX_CQE_STATUS_INSUFFICIENT_PRIVILEDGES	= 1,
104 	OCRDMA_MBX_CQE_STATUS_INVALID_PARAMETER		= 2,
105 	OCRDMA_MBX_CQE_STATUS_INSUFFICIENT_RESOURCES	= 3,
106 	OCRDMA_MBX_CQE_STATUS_QUEUE_FLUSHING		= 4,
107 	OCRDMA_MBX_CQE_STATUS_DMA_FAILED		= 5
108 };
109 
110 static inline void *ocrdma_get_eqe(struct ocrdma_eq *eq)
111 {
112 	return eq->q.va + (eq->q.tail * sizeof(struct ocrdma_eqe));
113 }
114 
115 static inline void ocrdma_eq_inc_tail(struct ocrdma_eq *eq)
116 {
117 	eq->q.tail = (eq->q.tail + 1) & (OCRDMA_EQ_LEN - 1);
118 }
119 
120 static inline void *ocrdma_get_mcqe(struct ocrdma_dev *dev)
121 {
122 	struct ocrdma_mcqe *cqe = (struct ocrdma_mcqe *)
123 	    (dev->mq.cq.va + (dev->mq.cq.tail * sizeof(struct ocrdma_mcqe)));
124 
125 	if (!(le32_to_cpu(cqe->valid_ae_cmpl_cons) & OCRDMA_MCQE_VALID_MASK))
126 		return NULL;
127 	return cqe;
128 }
129 
130 static inline void ocrdma_mcq_inc_tail(struct ocrdma_dev *dev)
131 {
132 	dev->mq.cq.tail = (dev->mq.cq.tail + 1) & (OCRDMA_MQ_CQ_LEN - 1);
133 }
134 
135 static inline struct ocrdma_mqe *ocrdma_get_mqe(struct ocrdma_dev *dev)
136 {
137 	return dev->mq.sq.va + (dev->mq.sq.head * sizeof(struct ocrdma_mqe));
138 }
139 
140 static inline void ocrdma_mq_inc_head(struct ocrdma_dev *dev)
141 {
142 	dev->mq.sq.head = (dev->mq.sq.head + 1) & (OCRDMA_MQ_LEN - 1);
143 }
144 
145 static inline void *ocrdma_get_mqe_rsp(struct ocrdma_dev *dev)
146 {
147 	return dev->mq.sq.va + (dev->mqe_ctx.tag * sizeof(struct ocrdma_mqe));
148 }
149 
150 enum ib_qp_state get_ibqp_state(enum ocrdma_qp_state qps)
151 {
152 	switch (qps) {
153 	case OCRDMA_QPS_RST:
154 		return IB_QPS_RESET;
155 	case OCRDMA_QPS_INIT:
156 		return IB_QPS_INIT;
157 	case OCRDMA_QPS_RTR:
158 		return IB_QPS_RTR;
159 	case OCRDMA_QPS_RTS:
160 		return IB_QPS_RTS;
161 	case OCRDMA_QPS_SQD:
162 	case OCRDMA_QPS_SQ_DRAINING:
163 		return IB_QPS_SQD;
164 	case OCRDMA_QPS_SQE:
165 		return IB_QPS_SQE;
166 	case OCRDMA_QPS_ERR:
167 		return IB_QPS_ERR;
168 	}
169 	return IB_QPS_ERR;
170 }
171 
172 static enum ocrdma_qp_state get_ocrdma_qp_state(enum ib_qp_state qps)
173 {
174 	switch (qps) {
175 	case IB_QPS_RESET:
176 		return OCRDMA_QPS_RST;
177 	case IB_QPS_INIT:
178 		return OCRDMA_QPS_INIT;
179 	case IB_QPS_RTR:
180 		return OCRDMA_QPS_RTR;
181 	case IB_QPS_RTS:
182 		return OCRDMA_QPS_RTS;
183 	case IB_QPS_SQD:
184 		return OCRDMA_QPS_SQD;
185 	case IB_QPS_SQE:
186 		return OCRDMA_QPS_SQE;
187 	case IB_QPS_ERR:
188 		return OCRDMA_QPS_ERR;
189 	}
190 	return OCRDMA_QPS_ERR;
191 }
192 
193 static int ocrdma_get_mbx_errno(u32 status)
194 {
195 	int err_num;
196 	u8 mbox_status = (status & OCRDMA_MBX_RSP_STATUS_MASK) >>
197 					OCRDMA_MBX_RSP_STATUS_SHIFT;
198 	u8 add_status = (status & OCRDMA_MBX_RSP_ASTATUS_MASK) >>
199 					OCRDMA_MBX_RSP_ASTATUS_SHIFT;
200 
201 	switch (mbox_status) {
202 	case OCRDMA_MBX_STATUS_OOR:
203 	case OCRDMA_MBX_STATUS_MAX_QP_EXCEEDS:
204 		err_num = -EAGAIN;
205 		break;
206 
207 	case OCRDMA_MBX_STATUS_INVALID_PD:
208 	case OCRDMA_MBX_STATUS_INVALID_CQ:
209 	case OCRDMA_MBX_STATUS_INVALID_SRQ_ID:
210 	case OCRDMA_MBX_STATUS_INVALID_QP:
211 	case OCRDMA_MBX_STATUS_INVALID_CHANGE:
212 	case OCRDMA_MBX_STATUS_MTU_EXCEEDS:
213 	case OCRDMA_MBX_STATUS_INVALID_RNR_NAK_TIMER:
214 	case OCRDMA_MBX_STATUS_PKEY_INDEX_INVALID:
215 	case OCRDMA_MBX_STATUS_PKEY_INDEX_EXCEEDS:
216 	case OCRDMA_MBX_STATUS_ILLEGAL_FIELD:
217 	case OCRDMA_MBX_STATUS_INVALID_PBL_ENTRY:
218 	case OCRDMA_MBX_STATUS_INVALID_LKEY:
219 	case OCRDMA_MBX_STATUS_INVALID_VA:
220 	case OCRDMA_MBX_STATUS_INVALID_LENGTH:
221 	case OCRDMA_MBX_STATUS_INVALID_FBO:
222 	case OCRDMA_MBX_STATUS_INVALID_ACC_RIGHTS:
223 	case OCRDMA_MBX_STATUS_INVALID_PBE_SIZE:
224 	case OCRDMA_MBX_STATUS_ATOMIC_OPS_UNSUP:
225 	case OCRDMA_MBX_STATUS_SRQ_ERROR:
226 	case OCRDMA_MBX_STATUS_SRQ_SIZE_UNDERUNS:
227 		err_num = -EINVAL;
228 		break;
229 
230 	case OCRDMA_MBX_STATUS_PD_INUSE:
231 	case OCRDMA_MBX_STATUS_QP_BOUND:
232 	case OCRDMA_MBX_STATUS_MW_STILL_BOUND:
233 	case OCRDMA_MBX_STATUS_MW_BOUND:
234 		err_num = -EBUSY;
235 		break;
236 
237 	case OCRDMA_MBX_STATUS_RECVQ_RQE_EXCEEDS:
238 	case OCRDMA_MBX_STATUS_SGE_RECV_EXCEEDS:
239 	case OCRDMA_MBX_STATUS_RQE_EXCEEDS:
240 	case OCRDMA_MBX_STATUS_SRQ_LIMIT_EXCEEDS:
241 	case OCRDMA_MBX_STATUS_ORD_EXCEEDS:
242 	case OCRDMA_MBX_STATUS_IRD_EXCEEDS:
243 	case OCRDMA_MBX_STATUS_SENDQ_WQE_EXCEEDS:
244 	case OCRDMA_MBX_STATUS_SGE_SEND_EXCEEDS:
245 	case OCRDMA_MBX_STATUS_SGE_WRITE_EXCEEDS:
246 		err_num = -ENOBUFS;
247 		break;
248 
249 	case OCRDMA_MBX_STATUS_FAILED:
250 		switch (add_status) {
251 		case OCRDMA_MBX_ADDI_STATUS_INSUFFICIENT_RESOURCES:
252 			err_num = -EAGAIN;
253 			break;
254 		}
255 	default:
256 		err_num = -EFAULT;
257 	}
258 	return err_num;
259 }
260 
261 char *port_speed_string(struct ocrdma_dev *dev)
262 {
263 	char *str = "";
264 	u16 speeds_supported;
265 
266 	speeds_supported = dev->phy.fixed_speeds_supported |
267 				dev->phy.auto_speeds_supported;
268 	if (speeds_supported & OCRDMA_PHY_SPEED_40GBPS)
269 		str = "40Gbps ";
270 	else if (speeds_supported & OCRDMA_PHY_SPEED_10GBPS)
271 		str = "10Gbps ";
272 	else if (speeds_supported & OCRDMA_PHY_SPEED_1GBPS)
273 		str = "1Gbps ";
274 
275 	return str;
276 }
277 
278 static int ocrdma_get_mbx_cqe_errno(u16 cqe_status)
279 {
280 	int err_num = -EINVAL;
281 
282 	switch (cqe_status) {
283 	case OCRDMA_MBX_CQE_STATUS_INSUFFICIENT_PRIVILEDGES:
284 		err_num = -EPERM;
285 		break;
286 	case OCRDMA_MBX_CQE_STATUS_INVALID_PARAMETER:
287 		err_num = -EINVAL;
288 		break;
289 	case OCRDMA_MBX_CQE_STATUS_INSUFFICIENT_RESOURCES:
290 	case OCRDMA_MBX_CQE_STATUS_QUEUE_FLUSHING:
291 		err_num = -EINVAL;
292 		break;
293 	case OCRDMA_MBX_CQE_STATUS_DMA_FAILED:
294 	default:
295 		err_num = -EINVAL;
296 		break;
297 	}
298 	return err_num;
299 }
300 
301 void ocrdma_ring_cq_db(struct ocrdma_dev *dev, u16 cq_id, bool armed,
302 		       bool solicited, u16 cqe_popped)
303 {
304 	u32 val = cq_id & OCRDMA_DB_CQ_RING_ID_MASK;
305 
306 	val |= ((cq_id & OCRDMA_DB_CQ_RING_ID_EXT_MASK) <<
307 	     OCRDMA_DB_CQ_RING_ID_EXT_MASK_SHIFT);
308 
309 	if (armed)
310 		val |= (1 << OCRDMA_DB_CQ_REARM_SHIFT);
311 	if (solicited)
312 		val |= (1 << OCRDMA_DB_CQ_SOLICIT_SHIFT);
313 	val |= (cqe_popped << OCRDMA_DB_CQ_NUM_POPPED_SHIFT);
314 	iowrite32(val, dev->nic_info.db + OCRDMA_DB_CQ_OFFSET);
315 }
316 
317 static void ocrdma_ring_mq_db(struct ocrdma_dev *dev)
318 {
319 	u32 val = 0;
320 
321 	val |= dev->mq.sq.id & OCRDMA_MQ_ID_MASK;
322 	val |= 1 << OCRDMA_MQ_NUM_MQE_SHIFT;
323 	iowrite32(val, dev->nic_info.db + OCRDMA_DB_MQ_OFFSET);
324 }
325 
326 static void ocrdma_ring_eq_db(struct ocrdma_dev *dev, u16 eq_id,
327 			      bool arm, bool clear_int, u16 num_eqe)
328 {
329 	u32 val = 0;
330 
331 	val |= eq_id & OCRDMA_EQ_ID_MASK;
332 	val |= ((eq_id & OCRDMA_EQ_ID_EXT_MASK) << OCRDMA_EQ_ID_EXT_MASK_SHIFT);
333 	if (arm)
334 		val |= (1 << OCRDMA_REARM_SHIFT);
335 	if (clear_int)
336 		val |= (1 << OCRDMA_EQ_CLR_SHIFT);
337 	val |= (1 << OCRDMA_EQ_TYPE_SHIFT);
338 	val |= (num_eqe << OCRDMA_NUM_EQE_SHIFT);
339 	iowrite32(val, dev->nic_info.db + OCRDMA_DB_EQ_OFFSET);
340 }
341 
342 static void ocrdma_init_mch(struct ocrdma_mbx_hdr *cmd_hdr,
343 			    u8 opcode, u8 subsys, u32 cmd_len)
344 {
345 	cmd_hdr->subsys_op = (opcode | (subsys << OCRDMA_MCH_SUBSYS_SHIFT));
346 	cmd_hdr->timeout = 20; /* seconds */
347 	cmd_hdr->cmd_len = cmd_len - sizeof(struct ocrdma_mbx_hdr);
348 }
349 
350 static void *ocrdma_init_emb_mqe(u8 opcode, u32 cmd_len)
351 {
352 	struct ocrdma_mqe *mqe;
353 
354 	mqe = kzalloc(sizeof(struct ocrdma_mqe), GFP_KERNEL);
355 	if (!mqe)
356 		return NULL;
357 	mqe->hdr.spcl_sge_cnt_emb |=
358 		(OCRDMA_MQE_EMBEDDED << OCRDMA_MQE_HDR_EMB_SHIFT) &
359 					OCRDMA_MQE_HDR_EMB_MASK;
360 	mqe->hdr.pyld_len = cmd_len - sizeof(struct ocrdma_mqe_hdr);
361 
362 	ocrdma_init_mch(&mqe->u.emb_req.mch, opcode, OCRDMA_SUBSYS_ROCE,
363 			mqe->hdr.pyld_len);
364 	return mqe;
365 }
366 
367 static void ocrdma_free_q(struct ocrdma_dev *dev, struct ocrdma_queue_info *q)
368 {
369 	dma_free_coherent(&dev->nic_info.pdev->dev, q->size, q->va, q->dma);
370 }
371 
372 static int ocrdma_alloc_q(struct ocrdma_dev *dev,
373 			  struct ocrdma_queue_info *q, u16 len, u16 entry_size)
374 {
375 	memset(q, 0, sizeof(*q));
376 	q->len = len;
377 	q->entry_size = entry_size;
378 	q->size = len * entry_size;
379 	q->va = dma_alloc_coherent(&dev->nic_info.pdev->dev, q->size,
380 				   &q->dma, GFP_KERNEL);
381 	if (!q->va)
382 		return -ENOMEM;
383 	memset(q->va, 0, q->size);
384 	return 0;
385 }
386 
387 static void ocrdma_build_q_pages(struct ocrdma_pa *q_pa, int cnt,
388 					dma_addr_t host_pa, int hw_page_size)
389 {
390 	int i;
391 
392 	for (i = 0; i < cnt; i++) {
393 		q_pa[i].lo = (u32) (host_pa & 0xffffffff);
394 		q_pa[i].hi = (u32) upper_32_bits(host_pa);
395 		host_pa += hw_page_size;
396 	}
397 }
398 
399 static int ocrdma_mbx_delete_q(struct ocrdma_dev *dev,
400 			       struct ocrdma_queue_info *q, int queue_type)
401 {
402 	u8 opcode = 0;
403 	int status;
404 	struct ocrdma_delete_q_req *cmd = dev->mbx_cmd;
405 
406 	switch (queue_type) {
407 	case QTYPE_MCCQ:
408 		opcode = OCRDMA_CMD_DELETE_MQ;
409 		break;
410 	case QTYPE_CQ:
411 		opcode = OCRDMA_CMD_DELETE_CQ;
412 		break;
413 	case QTYPE_EQ:
414 		opcode = OCRDMA_CMD_DELETE_EQ;
415 		break;
416 	default:
417 		BUG();
418 	}
419 	memset(cmd, 0, sizeof(*cmd));
420 	ocrdma_init_mch(&cmd->req, opcode, OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
421 	cmd->id = q->id;
422 
423 	status = be_roce_mcc_cmd(dev->nic_info.netdev,
424 				 cmd, sizeof(*cmd), NULL, NULL);
425 	if (!status)
426 		q->created = false;
427 	return status;
428 }
429 
430 static int ocrdma_mbx_create_eq(struct ocrdma_dev *dev, struct ocrdma_eq *eq)
431 {
432 	int status;
433 	struct ocrdma_create_eq_req *cmd = dev->mbx_cmd;
434 	struct ocrdma_create_eq_rsp *rsp = dev->mbx_cmd;
435 
436 	memset(cmd, 0, sizeof(*cmd));
437 	ocrdma_init_mch(&cmd->req, OCRDMA_CMD_CREATE_EQ, OCRDMA_SUBSYS_COMMON,
438 			sizeof(*cmd));
439 
440 	cmd->req.rsvd_version = 2;
441 	cmd->num_pages = 4;
442 	cmd->valid = OCRDMA_CREATE_EQ_VALID;
443 	cmd->cnt = 4 << OCRDMA_CREATE_EQ_CNT_SHIFT;
444 
445 	ocrdma_build_q_pages(&cmd->pa[0], cmd->num_pages, eq->q.dma,
446 			     PAGE_SIZE_4K);
447 	status = be_roce_mcc_cmd(dev->nic_info.netdev, cmd, sizeof(*cmd), NULL,
448 				 NULL);
449 	if (!status) {
450 		eq->q.id = rsp->vector_eqid & 0xffff;
451 		eq->vector = (rsp->vector_eqid >> 16) & 0xffff;
452 		eq->q.created = true;
453 	}
454 	return status;
455 }
456 
457 static int ocrdma_create_eq(struct ocrdma_dev *dev,
458 			    struct ocrdma_eq *eq, u16 q_len)
459 {
460 	int status;
461 
462 	status = ocrdma_alloc_q(dev, &eq->q, OCRDMA_EQ_LEN,
463 				sizeof(struct ocrdma_eqe));
464 	if (status)
465 		return status;
466 
467 	status = ocrdma_mbx_create_eq(dev, eq);
468 	if (status)
469 		goto mbx_err;
470 	eq->dev = dev;
471 	ocrdma_ring_eq_db(dev, eq->q.id, true, true, 0);
472 
473 	return 0;
474 mbx_err:
475 	ocrdma_free_q(dev, &eq->q);
476 	return status;
477 }
478 
479 int ocrdma_get_irq(struct ocrdma_dev *dev, struct ocrdma_eq *eq)
480 {
481 	int irq;
482 
483 	if (dev->nic_info.intr_mode == BE_INTERRUPT_MODE_INTX)
484 		irq = dev->nic_info.pdev->irq;
485 	else
486 		irq = dev->nic_info.msix.vector_list[eq->vector];
487 	return irq;
488 }
489 
490 static void _ocrdma_destroy_eq(struct ocrdma_dev *dev, struct ocrdma_eq *eq)
491 {
492 	if (eq->q.created) {
493 		ocrdma_mbx_delete_q(dev, &eq->q, QTYPE_EQ);
494 		ocrdma_free_q(dev, &eq->q);
495 	}
496 }
497 
498 static void ocrdma_destroy_eq(struct ocrdma_dev *dev, struct ocrdma_eq *eq)
499 {
500 	int irq;
501 
502 	/* disarm EQ so that interrupts are not generated
503 	 * during freeing and EQ delete is in progress.
504 	 */
505 	ocrdma_ring_eq_db(dev, eq->q.id, false, false, 0);
506 
507 	irq = ocrdma_get_irq(dev, eq);
508 	free_irq(irq, eq);
509 	_ocrdma_destroy_eq(dev, eq);
510 }
511 
512 static void ocrdma_destroy_eqs(struct ocrdma_dev *dev)
513 {
514 	int i;
515 
516 	for (i = 0; i < dev->eq_cnt; i++)
517 		ocrdma_destroy_eq(dev, &dev->eq_tbl[i]);
518 }
519 
520 static int ocrdma_mbx_mq_cq_create(struct ocrdma_dev *dev,
521 				   struct ocrdma_queue_info *cq,
522 				   struct ocrdma_queue_info *eq)
523 {
524 	struct ocrdma_create_cq_cmd *cmd = dev->mbx_cmd;
525 	struct ocrdma_create_cq_cmd_rsp *rsp = dev->mbx_cmd;
526 	int status;
527 
528 	memset(cmd, 0, sizeof(*cmd));
529 	ocrdma_init_mch(&cmd->req, OCRDMA_CMD_CREATE_CQ,
530 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
531 
532 	cmd->req.rsvd_version = OCRDMA_CREATE_CQ_VER2;
533 	cmd->pgsz_pgcnt = (cq->size / OCRDMA_MIN_Q_PAGE_SIZE) <<
534 		OCRDMA_CREATE_CQ_PAGE_SIZE_SHIFT;
535 	cmd->pgsz_pgcnt |= PAGES_4K_SPANNED(cq->va, cq->size);
536 
537 	cmd->ev_cnt_flags = OCRDMA_CREATE_CQ_DEF_FLAGS;
538 	cmd->eqn = eq->id;
539 	cmd->pdid_cqecnt = cq->size / sizeof(struct ocrdma_mcqe);
540 
541 	ocrdma_build_q_pages(&cmd->pa[0], cq->size / OCRDMA_MIN_Q_PAGE_SIZE,
542 			     cq->dma, PAGE_SIZE_4K);
543 	status = be_roce_mcc_cmd(dev->nic_info.netdev,
544 				 cmd, sizeof(*cmd), NULL, NULL);
545 	if (!status) {
546 		cq->id = (u16) (rsp->cq_id & OCRDMA_CREATE_CQ_RSP_CQ_ID_MASK);
547 		cq->created = true;
548 	}
549 	return status;
550 }
551 
552 static u32 ocrdma_encoded_q_len(int q_len)
553 {
554 	u32 len_encoded = fls(q_len);	/* log2(len) + 1 */
555 
556 	if (len_encoded == 16)
557 		len_encoded = 0;
558 	return len_encoded;
559 }
560 
561 static int ocrdma_mbx_create_mq(struct ocrdma_dev *dev,
562 				struct ocrdma_queue_info *mq,
563 				struct ocrdma_queue_info *cq)
564 {
565 	int num_pages, status;
566 	struct ocrdma_create_mq_req *cmd = dev->mbx_cmd;
567 	struct ocrdma_create_mq_rsp *rsp = dev->mbx_cmd;
568 	struct ocrdma_pa *pa;
569 
570 	memset(cmd, 0, sizeof(*cmd));
571 	num_pages = PAGES_4K_SPANNED(mq->va, mq->size);
572 
573 	ocrdma_init_mch(&cmd->req, OCRDMA_CMD_CREATE_MQ_EXT,
574 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
575 	cmd->req.rsvd_version = 1;
576 	cmd->cqid_pages = num_pages;
577 	cmd->cqid_pages |= (cq->id << OCRDMA_CREATE_MQ_CQ_ID_SHIFT);
578 	cmd->async_cqid_valid = OCRDMA_CREATE_MQ_ASYNC_CQ_VALID;
579 
580 	cmd->async_event_bitmap = BIT(OCRDMA_ASYNC_GRP5_EVE_CODE);
581 	cmd->async_event_bitmap |= BIT(OCRDMA_ASYNC_RDMA_EVE_CODE);
582 	/* Request link events on this  MQ. */
583 	cmd->async_event_bitmap |= BIT(OCRDMA_ASYNC_LINK_EVE_CODE);
584 
585 	cmd->async_cqid_ringsize = cq->id;
586 	cmd->async_cqid_ringsize |= (ocrdma_encoded_q_len(mq->len) <<
587 				OCRDMA_CREATE_MQ_RING_SIZE_SHIFT);
588 	cmd->valid = OCRDMA_CREATE_MQ_VALID;
589 	pa = &cmd->pa[0];
590 
591 	ocrdma_build_q_pages(pa, num_pages, mq->dma, PAGE_SIZE_4K);
592 	status = be_roce_mcc_cmd(dev->nic_info.netdev,
593 				 cmd, sizeof(*cmd), NULL, NULL);
594 	if (!status) {
595 		mq->id = rsp->id;
596 		mq->created = true;
597 	}
598 	return status;
599 }
600 
601 static int ocrdma_create_mq(struct ocrdma_dev *dev)
602 {
603 	int status;
604 
605 	/* Alloc completion queue for Mailbox queue */
606 	status = ocrdma_alloc_q(dev, &dev->mq.cq, OCRDMA_MQ_CQ_LEN,
607 				sizeof(struct ocrdma_mcqe));
608 	if (status)
609 		goto alloc_err;
610 
611 	dev->eq_tbl[0].cq_cnt++;
612 	status = ocrdma_mbx_mq_cq_create(dev, &dev->mq.cq, &dev->eq_tbl[0].q);
613 	if (status)
614 		goto mbx_cq_free;
615 
616 	memset(&dev->mqe_ctx, 0, sizeof(dev->mqe_ctx));
617 	init_waitqueue_head(&dev->mqe_ctx.cmd_wait);
618 	mutex_init(&dev->mqe_ctx.lock);
619 
620 	/* Alloc Mailbox queue */
621 	status = ocrdma_alloc_q(dev, &dev->mq.sq, OCRDMA_MQ_LEN,
622 				sizeof(struct ocrdma_mqe));
623 	if (status)
624 		goto mbx_cq_destroy;
625 	status = ocrdma_mbx_create_mq(dev, &dev->mq.sq, &dev->mq.cq);
626 	if (status)
627 		goto mbx_q_free;
628 	ocrdma_ring_cq_db(dev, dev->mq.cq.id, true, false, 0);
629 	return 0;
630 
631 mbx_q_free:
632 	ocrdma_free_q(dev, &dev->mq.sq);
633 mbx_cq_destroy:
634 	ocrdma_mbx_delete_q(dev, &dev->mq.cq, QTYPE_CQ);
635 mbx_cq_free:
636 	ocrdma_free_q(dev, &dev->mq.cq);
637 alloc_err:
638 	return status;
639 }
640 
641 static void ocrdma_destroy_mq(struct ocrdma_dev *dev)
642 {
643 	struct ocrdma_queue_info *mbxq, *cq;
644 
645 	/* mqe_ctx lock synchronizes with any other pending cmds. */
646 	mutex_lock(&dev->mqe_ctx.lock);
647 	mbxq = &dev->mq.sq;
648 	if (mbxq->created) {
649 		ocrdma_mbx_delete_q(dev, mbxq, QTYPE_MCCQ);
650 		ocrdma_free_q(dev, mbxq);
651 	}
652 	mutex_unlock(&dev->mqe_ctx.lock);
653 
654 	cq = &dev->mq.cq;
655 	if (cq->created) {
656 		ocrdma_mbx_delete_q(dev, cq, QTYPE_CQ);
657 		ocrdma_free_q(dev, cq);
658 	}
659 }
660 
661 static void ocrdma_process_qpcat_error(struct ocrdma_dev *dev,
662 				       struct ocrdma_qp *qp)
663 {
664 	enum ib_qp_state new_ib_qps = IB_QPS_ERR;
665 	enum ib_qp_state old_ib_qps;
666 
667 	if (qp == NULL)
668 		BUG();
669 	ocrdma_qp_state_change(qp, new_ib_qps, &old_ib_qps);
670 }
671 
672 static void ocrdma_dispatch_ibevent(struct ocrdma_dev *dev,
673 				    struct ocrdma_ae_mcqe *cqe)
674 {
675 	struct ocrdma_qp *qp = NULL;
676 	struct ocrdma_cq *cq = NULL;
677 	struct ib_event ib_evt;
678 	int cq_event = 0;
679 	int qp_event = 1;
680 	int srq_event = 0;
681 	int dev_event = 0;
682 	int type = (cqe->valid_ae_event & OCRDMA_AE_MCQE_EVENT_TYPE_MASK) >>
683 	    OCRDMA_AE_MCQE_EVENT_TYPE_SHIFT;
684 	u16 qpid = cqe->qpvalid_qpid & OCRDMA_AE_MCQE_QPID_MASK;
685 	u16 cqid = cqe->cqvalid_cqid & OCRDMA_AE_MCQE_CQID_MASK;
686 
687 	/*
688 	 * Some FW version returns wrong qp or cq ids in CQEs.
689 	 * Checking whether the IDs are valid
690 	 */
691 
692 	if (cqe->qpvalid_qpid & OCRDMA_AE_MCQE_QPVALID) {
693 		if (qpid < dev->attr.max_qp)
694 			qp = dev->qp_tbl[qpid];
695 		if (qp == NULL) {
696 			pr_err("ocrdma%d:Async event - qpid %u is not valid\n",
697 			       dev->id, qpid);
698 			return;
699 		}
700 	}
701 
702 	if (cqe->cqvalid_cqid & OCRDMA_AE_MCQE_CQVALID) {
703 		if (cqid < dev->attr.max_cq)
704 			cq = dev->cq_tbl[cqid];
705 		if (cq == NULL) {
706 			pr_err("ocrdma%d:Async event - cqid %u is not valid\n",
707 			       dev->id, cqid);
708 			return;
709 		}
710 	}
711 
712 	memset(&ib_evt, 0, sizeof(ib_evt));
713 
714 	ib_evt.device = &dev->ibdev;
715 
716 	switch (type) {
717 	case OCRDMA_CQ_ERROR:
718 		ib_evt.element.cq = &cq->ibcq;
719 		ib_evt.event = IB_EVENT_CQ_ERR;
720 		cq_event = 1;
721 		qp_event = 0;
722 		break;
723 	case OCRDMA_CQ_OVERRUN_ERROR:
724 		ib_evt.element.cq = &cq->ibcq;
725 		ib_evt.event = IB_EVENT_CQ_ERR;
726 		cq_event = 1;
727 		qp_event = 0;
728 		break;
729 	case OCRDMA_CQ_QPCAT_ERROR:
730 		ib_evt.element.qp = &qp->ibqp;
731 		ib_evt.event = IB_EVENT_QP_FATAL;
732 		ocrdma_process_qpcat_error(dev, qp);
733 		break;
734 	case OCRDMA_QP_ACCESS_ERROR:
735 		ib_evt.element.qp = &qp->ibqp;
736 		ib_evt.event = IB_EVENT_QP_ACCESS_ERR;
737 		break;
738 	case OCRDMA_QP_COMM_EST_EVENT:
739 		ib_evt.element.qp = &qp->ibqp;
740 		ib_evt.event = IB_EVENT_COMM_EST;
741 		break;
742 	case OCRDMA_SQ_DRAINED_EVENT:
743 		ib_evt.element.qp = &qp->ibqp;
744 		ib_evt.event = IB_EVENT_SQ_DRAINED;
745 		break;
746 	case OCRDMA_DEVICE_FATAL_EVENT:
747 		ib_evt.element.port_num = 1;
748 		ib_evt.event = IB_EVENT_DEVICE_FATAL;
749 		qp_event = 0;
750 		dev_event = 1;
751 		break;
752 	case OCRDMA_SRQCAT_ERROR:
753 		ib_evt.element.srq = &qp->srq->ibsrq;
754 		ib_evt.event = IB_EVENT_SRQ_ERR;
755 		srq_event = 1;
756 		qp_event = 0;
757 		break;
758 	case OCRDMA_SRQ_LIMIT_EVENT:
759 		ib_evt.element.srq = &qp->srq->ibsrq;
760 		ib_evt.event = IB_EVENT_SRQ_LIMIT_REACHED;
761 		srq_event = 1;
762 		qp_event = 0;
763 		break;
764 	case OCRDMA_QP_LAST_WQE_EVENT:
765 		ib_evt.element.qp = &qp->ibqp;
766 		ib_evt.event = IB_EVENT_QP_LAST_WQE_REACHED;
767 		break;
768 	default:
769 		cq_event = 0;
770 		qp_event = 0;
771 		srq_event = 0;
772 		dev_event = 0;
773 		pr_err("%s() unknown type=0x%x\n", __func__, type);
774 		break;
775 	}
776 
777 	if (type < OCRDMA_MAX_ASYNC_ERRORS)
778 		atomic_inc(&dev->async_err_stats[type]);
779 
780 	if (qp_event) {
781 		if (qp->ibqp.event_handler)
782 			qp->ibqp.event_handler(&ib_evt, qp->ibqp.qp_context);
783 	} else if (cq_event) {
784 		if (cq->ibcq.event_handler)
785 			cq->ibcq.event_handler(&ib_evt, cq->ibcq.cq_context);
786 	} else if (srq_event) {
787 		if (qp->srq->ibsrq.event_handler)
788 			qp->srq->ibsrq.event_handler(&ib_evt,
789 						     qp->srq->ibsrq.
790 						     srq_context);
791 	} else if (dev_event) {
792 		pr_err("%s: Fatal event received\n", dev->ibdev.name);
793 		ib_dispatch_event(&ib_evt);
794 	}
795 
796 }
797 
798 static void ocrdma_process_grp5_aync(struct ocrdma_dev *dev,
799 					struct ocrdma_ae_mcqe *cqe)
800 {
801 	struct ocrdma_ae_pvid_mcqe *evt;
802 	int type = (cqe->valid_ae_event & OCRDMA_AE_MCQE_EVENT_TYPE_MASK) >>
803 			OCRDMA_AE_MCQE_EVENT_TYPE_SHIFT;
804 
805 	switch (type) {
806 	case OCRDMA_ASYNC_EVENT_PVID_STATE:
807 		evt = (struct ocrdma_ae_pvid_mcqe *)cqe;
808 		if ((evt->tag_enabled & OCRDMA_AE_PVID_MCQE_ENABLED_MASK) >>
809 			OCRDMA_AE_PVID_MCQE_ENABLED_SHIFT)
810 			dev->pvid = ((evt->tag_enabled &
811 					OCRDMA_AE_PVID_MCQE_TAG_MASK) >>
812 					OCRDMA_AE_PVID_MCQE_TAG_SHIFT);
813 		break;
814 
815 	case OCRDMA_ASYNC_EVENT_COS_VALUE:
816 		atomic_set(&dev->update_sl, 1);
817 		break;
818 	default:
819 		/* Not interested evts. */
820 		break;
821 	}
822 }
823 
824 static void ocrdma_process_link_state(struct ocrdma_dev *dev,
825 				      struct ocrdma_ae_mcqe *cqe)
826 {
827 	struct ocrdma_ae_lnkst_mcqe *evt;
828 	u8 lstate;
829 
830 	evt = (struct ocrdma_ae_lnkst_mcqe *)cqe;
831 	lstate = ocrdma_get_ae_link_state(evt->speed_state_ptn);
832 
833 	if (!(lstate & OCRDMA_AE_LSC_LLINK_MASK))
834 		return;
835 
836 	if (dev->flags & OCRDMA_FLAGS_LINK_STATUS_INIT)
837 		ocrdma_update_link_state(dev, (lstate & OCRDMA_LINK_ST_MASK));
838 }
839 
840 static void ocrdma_process_acqe(struct ocrdma_dev *dev, void *ae_cqe)
841 {
842 	/* async CQE processing */
843 	struct ocrdma_ae_mcqe *cqe = ae_cqe;
844 	u32 evt_code = (cqe->valid_ae_event & OCRDMA_AE_MCQE_EVENT_CODE_MASK) >>
845 			OCRDMA_AE_MCQE_EVENT_CODE_SHIFT;
846 	switch (evt_code) {
847 	case OCRDMA_ASYNC_LINK_EVE_CODE:
848 		ocrdma_process_link_state(dev, cqe);
849 		break;
850 	case OCRDMA_ASYNC_RDMA_EVE_CODE:
851 		ocrdma_dispatch_ibevent(dev, cqe);
852 		break;
853 	case OCRDMA_ASYNC_GRP5_EVE_CODE:
854 		ocrdma_process_grp5_aync(dev, cqe);
855 		break;
856 	default:
857 		pr_err("%s(%d) invalid evt code=0x%x\n", __func__,
858 		       dev->id, evt_code);
859 	}
860 }
861 
862 static void ocrdma_process_mcqe(struct ocrdma_dev *dev, struct ocrdma_mcqe *cqe)
863 {
864 	if (dev->mqe_ctx.tag == cqe->tag_lo && dev->mqe_ctx.cmd_done == false) {
865 		dev->mqe_ctx.cqe_status = (cqe->status &
866 		     OCRDMA_MCQE_STATUS_MASK) >> OCRDMA_MCQE_STATUS_SHIFT;
867 		dev->mqe_ctx.ext_status =
868 		    (cqe->status & OCRDMA_MCQE_ESTATUS_MASK)
869 		    >> OCRDMA_MCQE_ESTATUS_SHIFT;
870 		dev->mqe_ctx.cmd_done = true;
871 		wake_up(&dev->mqe_ctx.cmd_wait);
872 	} else
873 		pr_err("%s() cqe for invalid tag0x%x.expected=0x%x\n",
874 		       __func__, cqe->tag_lo, dev->mqe_ctx.tag);
875 }
876 
877 static int ocrdma_mq_cq_handler(struct ocrdma_dev *dev, u16 cq_id)
878 {
879 	u16 cqe_popped = 0;
880 	struct ocrdma_mcqe *cqe;
881 
882 	while (1) {
883 		cqe = ocrdma_get_mcqe(dev);
884 		if (cqe == NULL)
885 			break;
886 		ocrdma_le32_to_cpu(cqe, sizeof(*cqe));
887 		cqe_popped += 1;
888 		if (cqe->valid_ae_cmpl_cons & OCRDMA_MCQE_AE_MASK)
889 			ocrdma_process_acqe(dev, cqe);
890 		else if (cqe->valid_ae_cmpl_cons & OCRDMA_MCQE_CMPL_MASK)
891 			ocrdma_process_mcqe(dev, cqe);
892 		memset(cqe, 0, sizeof(struct ocrdma_mcqe));
893 		ocrdma_mcq_inc_tail(dev);
894 	}
895 	ocrdma_ring_cq_db(dev, dev->mq.cq.id, true, false, cqe_popped);
896 	return 0;
897 }
898 
899 static struct ocrdma_cq *_ocrdma_qp_buddy_cq_handler(struct ocrdma_dev *dev,
900 				struct ocrdma_cq *cq, bool sq)
901 {
902 	struct ocrdma_qp *qp;
903 	struct list_head *cur;
904 	struct ocrdma_cq *bcq = NULL;
905 	struct list_head *head = sq?(&cq->sq_head):(&cq->rq_head);
906 
907 	list_for_each(cur, head) {
908 		if (sq)
909 			qp = list_entry(cur, struct ocrdma_qp, sq_entry);
910 		else
911 			qp = list_entry(cur, struct ocrdma_qp, rq_entry);
912 
913 		if (qp->srq)
914 			continue;
915 		/* if wq and rq share the same cq, than comp_handler
916 		 * is already invoked.
917 		 */
918 		if (qp->sq_cq == qp->rq_cq)
919 			continue;
920 		/* if completion came on sq, rq's cq is buddy cq.
921 		 * if completion came on rq, sq's cq is buddy cq.
922 		 */
923 		if (qp->sq_cq == cq)
924 			bcq = qp->rq_cq;
925 		else
926 			bcq = qp->sq_cq;
927 		return bcq;
928 	}
929 	return NULL;
930 }
931 
932 static void ocrdma_qp_buddy_cq_handler(struct ocrdma_dev *dev,
933 				       struct ocrdma_cq *cq)
934 {
935 	unsigned long flags;
936 	struct ocrdma_cq *bcq = NULL;
937 
938 	/* Go through list of QPs in error state which are using this CQ
939 	 * and invoke its callback handler to trigger CQE processing for
940 	 * error/flushed CQE. It is rare to find more than few entries in
941 	 * this list as most consumers stops after getting error CQE.
942 	 * List is traversed only once when a matching buddy cq found for a QP.
943 	 */
944 	spin_lock_irqsave(&dev->flush_q_lock, flags);
945 	/* Check if buddy CQ is present.
946 	 * true - Check for  SQ CQ
947 	 * false - Check for RQ CQ
948 	 */
949 	bcq = _ocrdma_qp_buddy_cq_handler(dev, cq, true);
950 	if (bcq == NULL)
951 		bcq = _ocrdma_qp_buddy_cq_handler(dev, cq, false);
952 	spin_unlock_irqrestore(&dev->flush_q_lock, flags);
953 
954 	/* if there is valid buddy cq, look for its completion handler */
955 	if (bcq && bcq->ibcq.comp_handler) {
956 		spin_lock_irqsave(&bcq->comp_handler_lock, flags);
957 		(*bcq->ibcq.comp_handler) (&bcq->ibcq, bcq->ibcq.cq_context);
958 		spin_unlock_irqrestore(&bcq->comp_handler_lock, flags);
959 	}
960 }
961 
962 static void ocrdma_qp_cq_handler(struct ocrdma_dev *dev, u16 cq_idx)
963 {
964 	unsigned long flags;
965 	struct ocrdma_cq *cq;
966 
967 	if (cq_idx >= OCRDMA_MAX_CQ)
968 		BUG();
969 
970 	cq = dev->cq_tbl[cq_idx];
971 	if (cq == NULL)
972 		return;
973 
974 	if (cq->ibcq.comp_handler) {
975 		spin_lock_irqsave(&cq->comp_handler_lock, flags);
976 		(*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context);
977 		spin_unlock_irqrestore(&cq->comp_handler_lock, flags);
978 	}
979 	ocrdma_qp_buddy_cq_handler(dev, cq);
980 }
981 
982 static void ocrdma_cq_handler(struct ocrdma_dev *dev, u16 cq_id)
983 {
984 	/* process the MQ-CQE. */
985 	if (cq_id == dev->mq.cq.id)
986 		ocrdma_mq_cq_handler(dev, cq_id);
987 	else
988 		ocrdma_qp_cq_handler(dev, cq_id);
989 }
990 
991 static irqreturn_t ocrdma_irq_handler(int irq, void *handle)
992 {
993 	struct ocrdma_eq *eq = handle;
994 	struct ocrdma_dev *dev = eq->dev;
995 	struct ocrdma_eqe eqe;
996 	struct ocrdma_eqe *ptr;
997 	u16 cq_id;
998 	u8 mcode;
999 	int budget = eq->cq_cnt;
1000 
1001 	do {
1002 		ptr = ocrdma_get_eqe(eq);
1003 		eqe = *ptr;
1004 		ocrdma_le32_to_cpu(&eqe, sizeof(eqe));
1005 		mcode = (eqe.id_valid & OCRDMA_EQE_MAJOR_CODE_MASK)
1006 				>> OCRDMA_EQE_MAJOR_CODE_SHIFT;
1007 		if (mcode == OCRDMA_MAJOR_CODE_SENTINAL)
1008 			pr_err("EQ full on eqid = 0x%x, eqe = 0x%x\n",
1009 			       eq->q.id, eqe.id_valid);
1010 		if ((eqe.id_valid & OCRDMA_EQE_VALID_MASK) == 0)
1011 			break;
1012 
1013 		ptr->id_valid = 0;
1014 		/* ring eq doorbell as soon as its consumed. */
1015 		ocrdma_ring_eq_db(dev, eq->q.id, false, true, 1);
1016 		/* check whether its CQE or not. */
1017 		if ((eqe.id_valid & OCRDMA_EQE_FOR_CQE_MASK) == 0) {
1018 			cq_id = eqe.id_valid >> OCRDMA_EQE_RESOURCE_ID_SHIFT;
1019 			ocrdma_cq_handler(dev, cq_id);
1020 		}
1021 		ocrdma_eq_inc_tail(eq);
1022 
1023 		/* There can be a stale EQE after the last bound CQ is
1024 		 * destroyed. EQE valid and budget == 0 implies this.
1025 		 */
1026 		if (budget)
1027 			budget--;
1028 
1029 	} while (budget);
1030 
1031 	eq->aic_obj.eq_intr_cnt++;
1032 	ocrdma_ring_eq_db(dev, eq->q.id, true, true, 0);
1033 	return IRQ_HANDLED;
1034 }
1035 
1036 static void ocrdma_post_mqe(struct ocrdma_dev *dev, struct ocrdma_mqe *cmd)
1037 {
1038 	struct ocrdma_mqe *mqe;
1039 
1040 	dev->mqe_ctx.tag = dev->mq.sq.head;
1041 	dev->mqe_ctx.cmd_done = false;
1042 	mqe = ocrdma_get_mqe(dev);
1043 	cmd->hdr.tag_lo = dev->mq.sq.head;
1044 	ocrdma_copy_cpu_to_le32(mqe, cmd, sizeof(*mqe));
1045 	/* make sure descriptor is written before ringing doorbell */
1046 	wmb();
1047 	ocrdma_mq_inc_head(dev);
1048 	ocrdma_ring_mq_db(dev);
1049 }
1050 
1051 static int ocrdma_wait_mqe_cmpl(struct ocrdma_dev *dev)
1052 {
1053 	long status;
1054 	/* 30 sec timeout */
1055 	status = wait_event_timeout(dev->mqe_ctx.cmd_wait,
1056 				    (dev->mqe_ctx.cmd_done != false),
1057 				    msecs_to_jiffies(30000));
1058 	if (status)
1059 		return 0;
1060 	else {
1061 		dev->mqe_ctx.fw_error_state = true;
1062 		pr_err("%s(%d) mailbox timeout: fw not responding\n",
1063 		       __func__, dev->id);
1064 		return -1;
1065 	}
1066 }
1067 
1068 /* issue a mailbox command on the MQ */
1069 static int ocrdma_mbx_cmd(struct ocrdma_dev *dev, struct ocrdma_mqe *mqe)
1070 {
1071 	int status = 0;
1072 	u16 cqe_status, ext_status;
1073 	struct ocrdma_mqe *rsp_mqe;
1074 	struct ocrdma_mbx_rsp *rsp = NULL;
1075 
1076 	mutex_lock(&dev->mqe_ctx.lock);
1077 	if (dev->mqe_ctx.fw_error_state)
1078 		goto mbx_err;
1079 	ocrdma_post_mqe(dev, mqe);
1080 	status = ocrdma_wait_mqe_cmpl(dev);
1081 	if (status)
1082 		goto mbx_err;
1083 	cqe_status = dev->mqe_ctx.cqe_status;
1084 	ext_status = dev->mqe_ctx.ext_status;
1085 	rsp_mqe = ocrdma_get_mqe_rsp(dev);
1086 	ocrdma_copy_le32_to_cpu(mqe, rsp_mqe, (sizeof(*mqe)));
1087 	if ((mqe->hdr.spcl_sge_cnt_emb & OCRDMA_MQE_HDR_EMB_MASK) >>
1088 				OCRDMA_MQE_HDR_EMB_SHIFT)
1089 		rsp = &mqe->u.rsp;
1090 
1091 	if (cqe_status || ext_status) {
1092 		pr_err("%s() cqe_status=0x%x, ext_status=0x%x,",
1093 		       __func__, cqe_status, ext_status);
1094 		if (rsp) {
1095 			/* This is for embedded cmds. */
1096 			pr_err("opcode=0x%x, subsystem=0x%x\n",
1097 			       (rsp->subsys_op & OCRDMA_MBX_RSP_OPCODE_MASK) >>
1098 				OCRDMA_MBX_RSP_OPCODE_SHIFT,
1099 				(rsp->subsys_op & OCRDMA_MBX_RSP_SUBSYS_MASK) >>
1100 				OCRDMA_MBX_RSP_SUBSYS_SHIFT);
1101 		}
1102 		status = ocrdma_get_mbx_cqe_errno(cqe_status);
1103 		goto mbx_err;
1104 	}
1105 	/* For non embedded, rsp errors are handled in ocrdma_nonemb_mbx_cmd */
1106 	if (rsp && (mqe->u.rsp.status & OCRDMA_MBX_RSP_STATUS_MASK))
1107 		status = ocrdma_get_mbx_errno(mqe->u.rsp.status);
1108 mbx_err:
1109 	mutex_unlock(&dev->mqe_ctx.lock);
1110 	return status;
1111 }
1112 
1113 static int ocrdma_nonemb_mbx_cmd(struct ocrdma_dev *dev, struct ocrdma_mqe *mqe,
1114 				 void *payload_va)
1115 {
1116 	int status;
1117 	struct ocrdma_mbx_rsp *rsp = payload_va;
1118 
1119 	if ((mqe->hdr.spcl_sge_cnt_emb & OCRDMA_MQE_HDR_EMB_MASK) >>
1120 				OCRDMA_MQE_HDR_EMB_SHIFT)
1121 		BUG();
1122 
1123 	status = ocrdma_mbx_cmd(dev, mqe);
1124 	if (!status)
1125 		/* For non embedded, only CQE failures are handled in
1126 		 * ocrdma_mbx_cmd. We need to check for RSP errors.
1127 		 */
1128 		if (rsp->status & OCRDMA_MBX_RSP_STATUS_MASK)
1129 			status = ocrdma_get_mbx_errno(rsp->status);
1130 
1131 	if (status)
1132 		pr_err("opcode=0x%x, subsystem=0x%x\n",
1133 		       (rsp->subsys_op & OCRDMA_MBX_RSP_OPCODE_MASK) >>
1134 			OCRDMA_MBX_RSP_OPCODE_SHIFT,
1135 			(rsp->subsys_op & OCRDMA_MBX_RSP_SUBSYS_MASK) >>
1136 			OCRDMA_MBX_RSP_SUBSYS_SHIFT);
1137 	return status;
1138 }
1139 
1140 static void ocrdma_get_attr(struct ocrdma_dev *dev,
1141 			      struct ocrdma_dev_attr *attr,
1142 			      struct ocrdma_mbx_query_config *rsp)
1143 {
1144 	attr->max_pd =
1145 	    (rsp->max_pd_ca_ack_delay & OCRDMA_MBX_QUERY_CFG_MAX_PD_MASK) >>
1146 	    OCRDMA_MBX_QUERY_CFG_MAX_PD_SHIFT;
1147 	attr->udp_encap = (rsp->max_pd_ca_ack_delay &
1148 			   OCRDMA_MBX_QUERY_CFG_L3_TYPE_MASK) >>
1149 			   OCRDMA_MBX_QUERY_CFG_L3_TYPE_SHIFT;
1150 	attr->max_dpp_pds =
1151 	   (rsp->max_dpp_pds_credits & OCRDMA_MBX_QUERY_CFG_MAX_DPP_PDS_MASK) >>
1152 	    OCRDMA_MBX_QUERY_CFG_MAX_DPP_PDS_OFFSET;
1153 	attr->max_qp =
1154 	    (rsp->qp_srq_cq_ird_ord & OCRDMA_MBX_QUERY_CFG_MAX_QP_MASK) >>
1155 	    OCRDMA_MBX_QUERY_CFG_MAX_QP_SHIFT;
1156 	attr->max_srq =
1157 		(rsp->max_srq_rpir_qps & OCRDMA_MBX_QUERY_CFG_MAX_SRQ_MASK) >>
1158 		OCRDMA_MBX_QUERY_CFG_MAX_SRQ_OFFSET;
1159 	attr->max_send_sge = ((rsp->max_recv_send_sge &
1160 			       OCRDMA_MBX_QUERY_CFG_MAX_SEND_SGE_MASK) >>
1161 			      OCRDMA_MBX_QUERY_CFG_MAX_SEND_SGE_SHIFT);
1162 	attr->max_recv_sge = (rsp->max_recv_send_sge &
1163 			      OCRDMA_MBX_QUERY_CFG_MAX_RECV_SGE_MASK) >>
1164 	    OCRDMA_MBX_QUERY_CFG_MAX_RECV_SGE_SHIFT;
1165 	attr->max_srq_sge = (rsp->max_srq_rqe_sge &
1166 			      OCRDMA_MBX_QUERY_CFG_MAX_SRQ_SGE_MASK) >>
1167 	    OCRDMA_MBX_QUERY_CFG_MAX_SRQ_SGE_OFFSET;
1168 	attr->max_rdma_sge = (rsp->max_wr_rd_sge &
1169 			      OCRDMA_MBX_QUERY_CFG_MAX_RD_SGE_MASK) >>
1170 	    OCRDMA_MBX_QUERY_CFG_MAX_RD_SGE_SHIFT;
1171 	attr->max_ord_per_qp = (rsp->max_ird_ord_per_qp &
1172 				OCRDMA_MBX_QUERY_CFG_MAX_ORD_PER_QP_MASK) >>
1173 	    OCRDMA_MBX_QUERY_CFG_MAX_ORD_PER_QP_SHIFT;
1174 	attr->max_ird_per_qp = (rsp->max_ird_ord_per_qp &
1175 				OCRDMA_MBX_QUERY_CFG_MAX_IRD_PER_QP_MASK) >>
1176 	    OCRDMA_MBX_QUERY_CFG_MAX_IRD_PER_QP_SHIFT;
1177 	attr->cq_overflow_detect = (rsp->qp_srq_cq_ird_ord &
1178 				    OCRDMA_MBX_QUERY_CFG_CQ_OVERFLOW_MASK) >>
1179 	    OCRDMA_MBX_QUERY_CFG_CQ_OVERFLOW_SHIFT;
1180 	attr->srq_supported = (rsp->qp_srq_cq_ird_ord &
1181 			       OCRDMA_MBX_QUERY_CFG_SRQ_SUPPORTED_MASK) >>
1182 	    OCRDMA_MBX_QUERY_CFG_SRQ_SUPPORTED_SHIFT;
1183 	attr->local_ca_ack_delay = (rsp->max_pd_ca_ack_delay &
1184 				    OCRDMA_MBX_QUERY_CFG_CA_ACK_DELAY_MASK) >>
1185 	    OCRDMA_MBX_QUERY_CFG_CA_ACK_DELAY_SHIFT;
1186 	attr->max_mw = rsp->max_mw;
1187 	attr->max_mr = rsp->max_mr;
1188 	attr->max_mr_size = ((u64)rsp->max_mr_size_hi << 32) |
1189 			      rsp->max_mr_size_lo;
1190 	attr->max_fmr = 0;
1191 	attr->max_pages_per_frmr = rsp->max_pages_per_frmr;
1192 	attr->max_num_mr_pbl = rsp->max_num_mr_pbl;
1193 	attr->max_cqe = rsp->max_cq_cqes_per_cq &
1194 			OCRDMA_MBX_QUERY_CFG_MAX_CQES_PER_CQ_MASK;
1195 	attr->max_cq = (rsp->max_cq_cqes_per_cq &
1196 			OCRDMA_MBX_QUERY_CFG_MAX_CQ_MASK) >>
1197 			OCRDMA_MBX_QUERY_CFG_MAX_CQ_OFFSET;
1198 	attr->wqe_size = ((rsp->wqe_rqe_stride_max_dpp_cqs &
1199 		OCRDMA_MBX_QUERY_CFG_MAX_WQE_SIZE_MASK) >>
1200 		OCRDMA_MBX_QUERY_CFG_MAX_WQE_SIZE_OFFSET) *
1201 		OCRDMA_WQE_STRIDE;
1202 	attr->rqe_size = ((rsp->wqe_rqe_stride_max_dpp_cqs &
1203 		OCRDMA_MBX_QUERY_CFG_MAX_RQE_SIZE_MASK) >>
1204 		OCRDMA_MBX_QUERY_CFG_MAX_RQE_SIZE_OFFSET) *
1205 		OCRDMA_WQE_STRIDE;
1206 	attr->max_inline_data =
1207 	    attr->wqe_size - (sizeof(struct ocrdma_hdr_wqe) +
1208 			      sizeof(struct ocrdma_sge));
1209 	if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) {
1210 		attr->ird = 1;
1211 		attr->ird_page_size = OCRDMA_MIN_Q_PAGE_SIZE;
1212 		attr->num_ird_pages = MAX_OCRDMA_IRD_PAGES;
1213 	}
1214 	dev->attr.max_wqe = rsp->max_wqes_rqes_per_q >>
1215 		 OCRDMA_MBX_QUERY_CFG_MAX_WQES_PER_WQ_OFFSET;
1216 	dev->attr.max_rqe = rsp->max_wqes_rqes_per_q &
1217 		OCRDMA_MBX_QUERY_CFG_MAX_RQES_PER_RQ_MASK;
1218 }
1219 
1220 static int ocrdma_check_fw_config(struct ocrdma_dev *dev,
1221 				   struct ocrdma_fw_conf_rsp *conf)
1222 {
1223 	u32 fn_mode;
1224 
1225 	fn_mode = conf->fn_mode & OCRDMA_FN_MODE_RDMA;
1226 	if (fn_mode != OCRDMA_FN_MODE_RDMA)
1227 		return -EINVAL;
1228 	dev->base_eqid = conf->base_eqid;
1229 	dev->max_eq = conf->max_eq;
1230 	return 0;
1231 }
1232 
1233 /* can be issued only during init time. */
1234 static int ocrdma_mbx_query_fw_ver(struct ocrdma_dev *dev)
1235 {
1236 	int status = -ENOMEM;
1237 	struct ocrdma_mqe *cmd;
1238 	struct ocrdma_fw_ver_rsp *rsp;
1239 
1240 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_GET_FW_VER, sizeof(*cmd));
1241 	if (!cmd)
1242 		return -ENOMEM;
1243 	ocrdma_init_mch((struct ocrdma_mbx_hdr *)&cmd->u.cmd[0],
1244 			OCRDMA_CMD_GET_FW_VER,
1245 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
1246 
1247 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1248 	if (status)
1249 		goto mbx_err;
1250 	rsp = (struct ocrdma_fw_ver_rsp *)cmd;
1251 	memset(&dev->attr.fw_ver[0], 0, sizeof(dev->attr.fw_ver));
1252 	memcpy(&dev->attr.fw_ver[0], &rsp->running_ver[0],
1253 	       sizeof(rsp->running_ver));
1254 	ocrdma_le32_to_cpu(dev->attr.fw_ver, sizeof(rsp->running_ver));
1255 mbx_err:
1256 	kfree(cmd);
1257 	return status;
1258 }
1259 
1260 /* can be issued only during init time. */
1261 static int ocrdma_mbx_query_fw_config(struct ocrdma_dev *dev)
1262 {
1263 	int status = -ENOMEM;
1264 	struct ocrdma_mqe *cmd;
1265 	struct ocrdma_fw_conf_rsp *rsp;
1266 
1267 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_GET_FW_CONFIG, sizeof(*cmd));
1268 	if (!cmd)
1269 		return -ENOMEM;
1270 	ocrdma_init_mch((struct ocrdma_mbx_hdr *)&cmd->u.cmd[0],
1271 			OCRDMA_CMD_GET_FW_CONFIG,
1272 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
1273 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1274 	if (status)
1275 		goto mbx_err;
1276 	rsp = (struct ocrdma_fw_conf_rsp *)cmd;
1277 	status = ocrdma_check_fw_config(dev, rsp);
1278 mbx_err:
1279 	kfree(cmd);
1280 	return status;
1281 }
1282 
1283 int ocrdma_mbx_rdma_stats(struct ocrdma_dev *dev, bool reset)
1284 {
1285 	struct ocrdma_rdma_stats_req *req = dev->stats_mem.va;
1286 	struct ocrdma_mqe *mqe = &dev->stats_mem.mqe;
1287 	struct ocrdma_rdma_stats_resp *old_stats;
1288 	int status;
1289 
1290 	old_stats = kmalloc(sizeof(*old_stats), GFP_KERNEL);
1291 	if (old_stats == NULL)
1292 		return -ENOMEM;
1293 
1294 	memset(mqe, 0, sizeof(*mqe));
1295 	mqe->hdr.pyld_len = dev->stats_mem.size;
1296 	mqe->hdr.spcl_sge_cnt_emb |=
1297 			(1 << OCRDMA_MQE_HDR_SGE_CNT_SHIFT) &
1298 				OCRDMA_MQE_HDR_SGE_CNT_MASK;
1299 	mqe->u.nonemb_req.sge[0].pa_lo = (u32) (dev->stats_mem.pa & 0xffffffff);
1300 	mqe->u.nonemb_req.sge[0].pa_hi = (u32) upper_32_bits(dev->stats_mem.pa);
1301 	mqe->u.nonemb_req.sge[0].len = dev->stats_mem.size;
1302 
1303 	/* Cache the old stats */
1304 	memcpy(old_stats, req, sizeof(struct ocrdma_rdma_stats_resp));
1305 	memset(req, 0, dev->stats_mem.size);
1306 
1307 	ocrdma_init_mch((struct ocrdma_mbx_hdr *)req,
1308 			OCRDMA_CMD_GET_RDMA_STATS,
1309 			OCRDMA_SUBSYS_ROCE,
1310 			dev->stats_mem.size);
1311 	if (reset)
1312 		req->reset_stats = reset;
1313 
1314 	status = ocrdma_nonemb_mbx_cmd(dev, mqe, dev->stats_mem.va);
1315 	if (status)
1316 		/* Copy from cache, if mbox fails */
1317 		memcpy(req, old_stats, sizeof(struct ocrdma_rdma_stats_resp));
1318 	else
1319 		ocrdma_le32_to_cpu(req, dev->stats_mem.size);
1320 
1321 	kfree(old_stats);
1322 	return status;
1323 }
1324 
1325 static int ocrdma_mbx_get_ctrl_attribs(struct ocrdma_dev *dev)
1326 {
1327 	int status = -ENOMEM;
1328 	struct ocrdma_dma_mem dma;
1329 	struct ocrdma_mqe *mqe;
1330 	struct ocrdma_get_ctrl_attribs_rsp *ctrl_attr_rsp;
1331 	struct mgmt_hba_attribs *hba_attribs;
1332 
1333 	mqe = kzalloc(sizeof(struct ocrdma_mqe), GFP_KERNEL);
1334 	if (!mqe)
1335 		return status;
1336 
1337 	dma.size = sizeof(struct ocrdma_get_ctrl_attribs_rsp);
1338 	dma.va	 = dma_alloc_coherent(&dev->nic_info.pdev->dev,
1339 					dma.size, &dma.pa, GFP_KERNEL);
1340 	if (!dma.va)
1341 		goto free_mqe;
1342 
1343 	mqe->hdr.pyld_len = dma.size;
1344 	mqe->hdr.spcl_sge_cnt_emb |=
1345 			(1 << OCRDMA_MQE_HDR_SGE_CNT_SHIFT) &
1346 			OCRDMA_MQE_HDR_SGE_CNT_MASK;
1347 	mqe->u.nonemb_req.sge[0].pa_lo = (u32) (dma.pa & 0xffffffff);
1348 	mqe->u.nonemb_req.sge[0].pa_hi = (u32) upper_32_bits(dma.pa);
1349 	mqe->u.nonemb_req.sge[0].len = dma.size;
1350 
1351 	memset(dma.va, 0, dma.size);
1352 	ocrdma_init_mch((struct ocrdma_mbx_hdr *)dma.va,
1353 			OCRDMA_CMD_GET_CTRL_ATTRIBUTES,
1354 			OCRDMA_SUBSYS_COMMON,
1355 			dma.size);
1356 
1357 	status = ocrdma_nonemb_mbx_cmd(dev, mqe, dma.va);
1358 	if (!status) {
1359 		ctrl_attr_rsp = (struct ocrdma_get_ctrl_attribs_rsp *)dma.va;
1360 		hba_attribs = &ctrl_attr_rsp->ctrl_attribs.hba_attribs;
1361 
1362 		dev->hba_port_num = (hba_attribs->ptpnum_maxdoms_hbast_cv &
1363 					OCRDMA_HBA_ATTRB_PTNUM_MASK)
1364 					>> OCRDMA_HBA_ATTRB_PTNUM_SHIFT;
1365 		strncpy(dev->model_number,
1366 			hba_attribs->controller_model_number, 31);
1367 	}
1368 	dma_free_coherent(&dev->nic_info.pdev->dev, dma.size, dma.va, dma.pa);
1369 free_mqe:
1370 	kfree(mqe);
1371 	return status;
1372 }
1373 
1374 static int ocrdma_mbx_query_dev(struct ocrdma_dev *dev)
1375 {
1376 	int status = -ENOMEM;
1377 	struct ocrdma_mbx_query_config *rsp;
1378 	struct ocrdma_mqe *cmd;
1379 
1380 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_CONFIG, sizeof(*cmd));
1381 	if (!cmd)
1382 		return status;
1383 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1384 	if (status)
1385 		goto mbx_err;
1386 	rsp = (struct ocrdma_mbx_query_config *)cmd;
1387 	ocrdma_get_attr(dev, &dev->attr, rsp);
1388 mbx_err:
1389 	kfree(cmd);
1390 	return status;
1391 }
1392 
1393 int ocrdma_mbx_get_link_speed(struct ocrdma_dev *dev, u8 *lnk_speed,
1394 			      u8 *lnk_state)
1395 {
1396 	int status = -ENOMEM;
1397 	struct ocrdma_get_link_speed_rsp *rsp;
1398 	struct ocrdma_mqe *cmd;
1399 
1400 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_NTWK_LINK_CONFIG_V1,
1401 				  sizeof(*cmd));
1402 	if (!cmd)
1403 		return status;
1404 	ocrdma_init_mch((struct ocrdma_mbx_hdr *)&cmd->u.cmd[0],
1405 			OCRDMA_CMD_QUERY_NTWK_LINK_CONFIG_V1,
1406 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
1407 
1408 	((struct ocrdma_mbx_hdr *)cmd->u.cmd)->rsvd_version = 0x1;
1409 
1410 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1411 	if (status)
1412 		goto mbx_err;
1413 
1414 	rsp = (struct ocrdma_get_link_speed_rsp *)cmd;
1415 	if (lnk_speed)
1416 		*lnk_speed = (rsp->pflt_pps_ld_pnum & OCRDMA_PHY_PS_MASK)
1417 			      >> OCRDMA_PHY_PS_SHIFT;
1418 	if (lnk_state)
1419 		*lnk_state = (rsp->res_lnk_st & OCRDMA_LINK_ST_MASK);
1420 
1421 mbx_err:
1422 	kfree(cmd);
1423 	return status;
1424 }
1425 
1426 static int ocrdma_mbx_get_phy_info(struct ocrdma_dev *dev)
1427 {
1428 	int status = -ENOMEM;
1429 	struct ocrdma_mqe *cmd;
1430 	struct ocrdma_get_phy_info_rsp *rsp;
1431 
1432 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_PHY_DETAILS, sizeof(*cmd));
1433 	if (!cmd)
1434 		return status;
1435 
1436 	ocrdma_init_mch((struct ocrdma_mbx_hdr *)&cmd->u.cmd[0],
1437 			OCRDMA_CMD_PHY_DETAILS, OCRDMA_SUBSYS_COMMON,
1438 			sizeof(*cmd));
1439 
1440 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1441 	if (status)
1442 		goto mbx_err;
1443 
1444 	rsp = (struct ocrdma_get_phy_info_rsp *)cmd;
1445 	dev->phy.phy_type =
1446 			(rsp->ityp_ptyp & OCRDMA_PHY_TYPE_MASK);
1447 	dev->phy.interface_type =
1448 			(rsp->ityp_ptyp & OCRDMA_IF_TYPE_MASK)
1449 				>> OCRDMA_IF_TYPE_SHIFT;
1450 	dev->phy.auto_speeds_supported  =
1451 			(rsp->fspeed_aspeed & OCRDMA_ASPEED_SUPP_MASK);
1452 	dev->phy.fixed_speeds_supported =
1453 			(rsp->fspeed_aspeed & OCRDMA_FSPEED_SUPP_MASK)
1454 				>> OCRDMA_FSPEED_SUPP_SHIFT;
1455 mbx_err:
1456 	kfree(cmd);
1457 	return status;
1458 }
1459 
1460 int ocrdma_mbx_alloc_pd(struct ocrdma_dev *dev, struct ocrdma_pd *pd)
1461 {
1462 	int status = -ENOMEM;
1463 	struct ocrdma_alloc_pd *cmd;
1464 	struct ocrdma_alloc_pd_rsp *rsp;
1465 
1466 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_ALLOC_PD, sizeof(*cmd));
1467 	if (!cmd)
1468 		return status;
1469 	if (pd->dpp_enabled)
1470 		cmd->enable_dpp_rsvd |= OCRDMA_ALLOC_PD_ENABLE_DPP;
1471 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1472 	if (status)
1473 		goto mbx_err;
1474 	rsp = (struct ocrdma_alloc_pd_rsp *)cmd;
1475 	pd->id = rsp->dpp_page_pdid & OCRDMA_ALLOC_PD_RSP_PDID_MASK;
1476 	if (rsp->dpp_page_pdid & OCRDMA_ALLOC_PD_RSP_DPP) {
1477 		pd->dpp_enabled = true;
1478 		pd->dpp_page = rsp->dpp_page_pdid >>
1479 				OCRDMA_ALLOC_PD_RSP_DPP_PAGE_SHIFT;
1480 	} else {
1481 		pd->dpp_enabled = false;
1482 		pd->num_dpp_qp = 0;
1483 	}
1484 mbx_err:
1485 	kfree(cmd);
1486 	return status;
1487 }
1488 
1489 int ocrdma_mbx_dealloc_pd(struct ocrdma_dev *dev, struct ocrdma_pd *pd)
1490 {
1491 	int status = -ENOMEM;
1492 	struct ocrdma_dealloc_pd *cmd;
1493 
1494 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DEALLOC_PD, sizeof(*cmd));
1495 	if (!cmd)
1496 		return status;
1497 	cmd->id = pd->id;
1498 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1499 	kfree(cmd);
1500 	return status;
1501 }
1502 
1503 
1504 static int ocrdma_mbx_alloc_pd_range(struct ocrdma_dev *dev)
1505 {
1506 	int status = -ENOMEM;
1507 	size_t pd_bitmap_size;
1508 	struct ocrdma_alloc_pd_range *cmd;
1509 	struct ocrdma_alloc_pd_range_rsp *rsp;
1510 
1511 	/* Pre allocate the DPP PDs */
1512 	if (dev->attr.max_dpp_pds) {
1513 		cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_ALLOC_PD_RANGE,
1514 					  sizeof(*cmd));
1515 		if (!cmd)
1516 			return -ENOMEM;
1517 		cmd->pd_count = dev->attr.max_dpp_pds;
1518 		cmd->enable_dpp_rsvd |= OCRDMA_ALLOC_PD_ENABLE_DPP;
1519 		status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1520 		rsp = (struct ocrdma_alloc_pd_range_rsp *)cmd;
1521 
1522 		if (!status && (rsp->dpp_page_pdid & OCRDMA_ALLOC_PD_RSP_DPP) &&
1523 		    rsp->pd_count) {
1524 			dev->pd_mgr->dpp_page_index = rsp->dpp_page_pdid >>
1525 					OCRDMA_ALLOC_PD_RSP_DPP_PAGE_SHIFT;
1526 			dev->pd_mgr->pd_dpp_start = rsp->dpp_page_pdid &
1527 					OCRDMA_ALLOC_PD_RNG_RSP_START_PDID_MASK;
1528 			dev->pd_mgr->max_dpp_pd = rsp->pd_count;
1529 			pd_bitmap_size =
1530 				BITS_TO_LONGS(rsp->pd_count) * sizeof(long);
1531 			dev->pd_mgr->pd_dpp_bitmap = kzalloc(pd_bitmap_size,
1532 							     GFP_KERNEL);
1533 		}
1534 		kfree(cmd);
1535 	}
1536 
1537 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_ALLOC_PD_RANGE, sizeof(*cmd));
1538 	if (!cmd)
1539 		return -ENOMEM;
1540 
1541 	cmd->pd_count = dev->attr.max_pd - dev->attr.max_dpp_pds;
1542 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1543 	rsp = (struct ocrdma_alloc_pd_range_rsp *)cmd;
1544 	if (!status && rsp->pd_count) {
1545 		dev->pd_mgr->pd_norm_start = rsp->dpp_page_pdid &
1546 					OCRDMA_ALLOC_PD_RNG_RSP_START_PDID_MASK;
1547 		dev->pd_mgr->max_normal_pd = rsp->pd_count;
1548 		pd_bitmap_size = BITS_TO_LONGS(rsp->pd_count) * sizeof(long);
1549 		dev->pd_mgr->pd_norm_bitmap = kzalloc(pd_bitmap_size,
1550 						      GFP_KERNEL);
1551 	}
1552 	kfree(cmd);
1553 
1554 	if (dev->pd_mgr->pd_norm_bitmap || dev->pd_mgr->pd_dpp_bitmap) {
1555 		/* Enable PD resource manager */
1556 		dev->pd_mgr->pd_prealloc_valid = true;
1557 		return 0;
1558 	}
1559 	return status;
1560 }
1561 
1562 static void ocrdma_mbx_dealloc_pd_range(struct ocrdma_dev *dev)
1563 {
1564 	struct ocrdma_dealloc_pd_range *cmd;
1565 
1566 	/* return normal PDs to firmware */
1567 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DEALLOC_PD_RANGE, sizeof(*cmd));
1568 	if (!cmd)
1569 		goto mbx_err;
1570 
1571 	if (dev->pd_mgr->max_normal_pd) {
1572 		cmd->start_pd_id = dev->pd_mgr->pd_norm_start;
1573 		cmd->pd_count = dev->pd_mgr->max_normal_pd;
1574 		ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1575 	}
1576 
1577 	if (dev->pd_mgr->max_dpp_pd) {
1578 		kfree(cmd);
1579 		/* return DPP PDs to firmware */
1580 		cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DEALLOC_PD_RANGE,
1581 					  sizeof(*cmd));
1582 		if (!cmd)
1583 			goto mbx_err;
1584 
1585 		cmd->start_pd_id = dev->pd_mgr->pd_dpp_start;
1586 		cmd->pd_count = dev->pd_mgr->max_dpp_pd;
1587 		ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1588 	}
1589 mbx_err:
1590 	kfree(cmd);
1591 }
1592 
1593 void ocrdma_alloc_pd_pool(struct ocrdma_dev *dev)
1594 {
1595 	int status;
1596 
1597 	dev->pd_mgr = kzalloc(sizeof(struct ocrdma_pd_resource_mgr),
1598 			      GFP_KERNEL);
1599 	if (!dev->pd_mgr) {
1600 		pr_err("%s(%d)Memory allocation failure.\n", __func__, dev->id);
1601 		return;
1602 	}
1603 	status = ocrdma_mbx_alloc_pd_range(dev);
1604 	if (status) {
1605 		pr_err("%s(%d) Unable to initialize PD pool, using default.\n",
1606 			 __func__, dev->id);
1607 	}
1608 }
1609 
1610 static void ocrdma_free_pd_pool(struct ocrdma_dev *dev)
1611 {
1612 	ocrdma_mbx_dealloc_pd_range(dev);
1613 	kfree(dev->pd_mgr->pd_norm_bitmap);
1614 	kfree(dev->pd_mgr->pd_dpp_bitmap);
1615 	kfree(dev->pd_mgr);
1616 }
1617 
1618 static int ocrdma_build_q_conf(u32 *num_entries, int entry_size,
1619 			       int *num_pages, int *page_size)
1620 {
1621 	int i;
1622 	int mem_size;
1623 
1624 	*num_entries = roundup_pow_of_two(*num_entries);
1625 	mem_size = *num_entries * entry_size;
1626 	/* find the possible lowest possible multiplier */
1627 	for (i = 0; i < OCRDMA_MAX_Q_PAGE_SIZE_CNT; i++) {
1628 		if (mem_size <= (OCRDMA_Q_PAGE_BASE_SIZE << i))
1629 			break;
1630 	}
1631 	if (i >= OCRDMA_MAX_Q_PAGE_SIZE_CNT)
1632 		return -EINVAL;
1633 	mem_size = roundup(mem_size,
1634 		       ((OCRDMA_Q_PAGE_BASE_SIZE << i) / OCRDMA_MAX_Q_PAGES));
1635 	*num_pages =
1636 	    mem_size / ((OCRDMA_Q_PAGE_BASE_SIZE << i) / OCRDMA_MAX_Q_PAGES);
1637 	*page_size = ((OCRDMA_Q_PAGE_BASE_SIZE << i) / OCRDMA_MAX_Q_PAGES);
1638 	*num_entries = mem_size / entry_size;
1639 	return 0;
1640 }
1641 
1642 static int ocrdma_mbx_create_ah_tbl(struct ocrdma_dev *dev)
1643 {
1644 	int i;
1645 	int status = 0;
1646 	int max_ah;
1647 	struct ocrdma_create_ah_tbl *cmd;
1648 	struct ocrdma_create_ah_tbl_rsp *rsp;
1649 	struct pci_dev *pdev = dev->nic_info.pdev;
1650 	dma_addr_t pa;
1651 	struct ocrdma_pbe *pbes;
1652 
1653 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_CREATE_AH_TBL, sizeof(*cmd));
1654 	if (!cmd)
1655 		return status;
1656 
1657 	max_ah = OCRDMA_MAX_AH;
1658 	dev->av_tbl.size = sizeof(struct ocrdma_av) * max_ah;
1659 
1660 	/* number of PBEs in PBL */
1661 	cmd->ah_conf = (OCRDMA_AH_TBL_PAGES <<
1662 				OCRDMA_CREATE_AH_NUM_PAGES_SHIFT) &
1663 				OCRDMA_CREATE_AH_NUM_PAGES_MASK;
1664 
1665 	/* page size */
1666 	for (i = 0; i < OCRDMA_MAX_Q_PAGE_SIZE_CNT; i++) {
1667 		if (PAGE_SIZE == (OCRDMA_MIN_Q_PAGE_SIZE << i))
1668 			break;
1669 	}
1670 	cmd->ah_conf |= (i << OCRDMA_CREATE_AH_PAGE_SIZE_SHIFT) &
1671 				OCRDMA_CREATE_AH_PAGE_SIZE_MASK;
1672 
1673 	/* ah_entry size */
1674 	cmd->ah_conf |= (sizeof(struct ocrdma_av) <<
1675 				OCRDMA_CREATE_AH_ENTRY_SIZE_SHIFT) &
1676 				OCRDMA_CREATE_AH_ENTRY_SIZE_MASK;
1677 
1678 	dev->av_tbl.pbl.va = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
1679 						&dev->av_tbl.pbl.pa,
1680 						GFP_KERNEL);
1681 	if (dev->av_tbl.pbl.va == NULL)
1682 		goto mem_err;
1683 
1684 	dev->av_tbl.va = dma_alloc_coherent(&pdev->dev, dev->av_tbl.size,
1685 					    &pa, GFP_KERNEL);
1686 	if (dev->av_tbl.va == NULL)
1687 		goto mem_err_ah;
1688 	dev->av_tbl.pa = pa;
1689 	dev->av_tbl.num_ah = max_ah;
1690 	memset(dev->av_tbl.va, 0, dev->av_tbl.size);
1691 
1692 	pbes = (struct ocrdma_pbe *)dev->av_tbl.pbl.va;
1693 	for (i = 0; i < dev->av_tbl.size / OCRDMA_MIN_Q_PAGE_SIZE; i++) {
1694 		pbes[i].pa_lo = (u32)cpu_to_le32(pa & 0xffffffff);
1695 		pbes[i].pa_hi = (u32)cpu_to_le32(upper_32_bits(pa));
1696 		pa += PAGE_SIZE;
1697 	}
1698 	cmd->tbl_addr[0].lo = (u32)(dev->av_tbl.pbl.pa & 0xFFFFFFFF);
1699 	cmd->tbl_addr[0].hi = (u32)upper_32_bits(dev->av_tbl.pbl.pa);
1700 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1701 	if (status)
1702 		goto mbx_err;
1703 	rsp = (struct ocrdma_create_ah_tbl_rsp *)cmd;
1704 	dev->av_tbl.ahid = rsp->ahid & 0xFFFF;
1705 	kfree(cmd);
1706 	return 0;
1707 
1708 mbx_err:
1709 	dma_free_coherent(&pdev->dev, dev->av_tbl.size, dev->av_tbl.va,
1710 			  dev->av_tbl.pa);
1711 	dev->av_tbl.va = NULL;
1712 mem_err_ah:
1713 	dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->av_tbl.pbl.va,
1714 			  dev->av_tbl.pbl.pa);
1715 	dev->av_tbl.pbl.va = NULL;
1716 	dev->av_tbl.size = 0;
1717 mem_err:
1718 	kfree(cmd);
1719 	return status;
1720 }
1721 
1722 static void ocrdma_mbx_delete_ah_tbl(struct ocrdma_dev *dev)
1723 {
1724 	struct ocrdma_delete_ah_tbl *cmd;
1725 	struct pci_dev *pdev = dev->nic_info.pdev;
1726 
1727 	if (dev->av_tbl.va == NULL)
1728 		return;
1729 
1730 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DELETE_AH_TBL, sizeof(*cmd));
1731 	if (!cmd)
1732 		return;
1733 	cmd->ahid = dev->av_tbl.ahid;
1734 
1735 	ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1736 	dma_free_coherent(&pdev->dev, dev->av_tbl.size, dev->av_tbl.va,
1737 			  dev->av_tbl.pa);
1738 	dev->av_tbl.va = NULL;
1739 	dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->av_tbl.pbl.va,
1740 			  dev->av_tbl.pbl.pa);
1741 	kfree(cmd);
1742 }
1743 
1744 /* Multiple CQs uses the EQ. This routine returns least used
1745  * EQ to associate with CQ. This will distributes the interrupt
1746  * processing and CPU load to associated EQ, vector and so to that CPU.
1747  */
1748 static u16 ocrdma_bind_eq(struct ocrdma_dev *dev)
1749 {
1750 	int i, selected_eq = 0, cq_cnt = 0;
1751 	u16 eq_id;
1752 
1753 	mutex_lock(&dev->dev_lock);
1754 	cq_cnt = dev->eq_tbl[0].cq_cnt;
1755 	eq_id = dev->eq_tbl[0].q.id;
1756 	/* find the EQ which is has the least number of
1757 	 * CQs associated with it.
1758 	 */
1759 	for (i = 0; i < dev->eq_cnt; i++) {
1760 		if (dev->eq_tbl[i].cq_cnt < cq_cnt) {
1761 			cq_cnt = dev->eq_tbl[i].cq_cnt;
1762 			eq_id = dev->eq_tbl[i].q.id;
1763 			selected_eq = i;
1764 		}
1765 	}
1766 	dev->eq_tbl[selected_eq].cq_cnt += 1;
1767 	mutex_unlock(&dev->dev_lock);
1768 	return eq_id;
1769 }
1770 
1771 static void ocrdma_unbind_eq(struct ocrdma_dev *dev, u16 eq_id)
1772 {
1773 	int i;
1774 
1775 	mutex_lock(&dev->dev_lock);
1776 	i = ocrdma_get_eq_table_index(dev, eq_id);
1777 	if (i == -EINVAL)
1778 		BUG();
1779 	dev->eq_tbl[i].cq_cnt -= 1;
1780 	mutex_unlock(&dev->dev_lock);
1781 }
1782 
1783 int ocrdma_mbx_create_cq(struct ocrdma_dev *dev, struct ocrdma_cq *cq,
1784 			 int entries, int dpp_cq, u16 pd_id)
1785 {
1786 	int status = -ENOMEM; int max_hw_cqe;
1787 	struct pci_dev *pdev = dev->nic_info.pdev;
1788 	struct ocrdma_create_cq *cmd;
1789 	struct ocrdma_create_cq_rsp *rsp;
1790 	u32 hw_pages, cqe_size, page_size, cqe_count;
1791 
1792 	if (entries > dev->attr.max_cqe) {
1793 		pr_err("%s(%d) max_cqe=0x%x, requester_cqe=0x%x\n",
1794 		       __func__, dev->id, dev->attr.max_cqe, entries);
1795 		return -EINVAL;
1796 	}
1797 	if (dpp_cq && (ocrdma_get_asic_type(dev) != OCRDMA_ASIC_GEN_SKH_R))
1798 		return -EINVAL;
1799 
1800 	if (dpp_cq) {
1801 		cq->max_hw_cqe = 1;
1802 		max_hw_cqe = 1;
1803 		cqe_size = OCRDMA_DPP_CQE_SIZE;
1804 		hw_pages = 1;
1805 	} else {
1806 		cq->max_hw_cqe = dev->attr.max_cqe;
1807 		max_hw_cqe = dev->attr.max_cqe;
1808 		cqe_size = sizeof(struct ocrdma_cqe);
1809 		hw_pages = OCRDMA_CREATE_CQ_MAX_PAGES;
1810 	}
1811 
1812 	cq->len = roundup(max_hw_cqe * cqe_size, OCRDMA_MIN_Q_PAGE_SIZE);
1813 
1814 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_CREATE_CQ, sizeof(*cmd));
1815 	if (!cmd)
1816 		return -ENOMEM;
1817 	ocrdma_init_mch(&cmd->cmd.req, OCRDMA_CMD_CREATE_CQ,
1818 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
1819 	cq->va = dma_alloc_coherent(&pdev->dev, cq->len, &cq->pa, GFP_KERNEL);
1820 	if (!cq->va) {
1821 		status = -ENOMEM;
1822 		goto mem_err;
1823 	}
1824 	memset(cq->va, 0, cq->len);
1825 	page_size = cq->len / hw_pages;
1826 	cmd->cmd.pgsz_pgcnt = (page_size / OCRDMA_MIN_Q_PAGE_SIZE) <<
1827 					OCRDMA_CREATE_CQ_PAGE_SIZE_SHIFT;
1828 	cmd->cmd.pgsz_pgcnt |= hw_pages;
1829 	cmd->cmd.ev_cnt_flags = OCRDMA_CREATE_CQ_DEF_FLAGS;
1830 
1831 	cq->eqn = ocrdma_bind_eq(dev);
1832 	cmd->cmd.req.rsvd_version = OCRDMA_CREATE_CQ_VER3;
1833 	cqe_count = cq->len / cqe_size;
1834 	cq->cqe_cnt = cqe_count;
1835 	if (cqe_count > 1024) {
1836 		/* Set cnt to 3 to indicate more than 1024 cq entries */
1837 		cmd->cmd.ev_cnt_flags |= (0x3 << OCRDMA_CREATE_CQ_CNT_SHIFT);
1838 	} else {
1839 		u8 count = 0;
1840 		switch (cqe_count) {
1841 		case 256:
1842 			count = 0;
1843 			break;
1844 		case 512:
1845 			count = 1;
1846 			break;
1847 		case 1024:
1848 			count = 2;
1849 			break;
1850 		default:
1851 			goto mbx_err;
1852 		}
1853 		cmd->cmd.ev_cnt_flags |= (count << OCRDMA_CREATE_CQ_CNT_SHIFT);
1854 	}
1855 	/* shared eq between all the consumer cqs. */
1856 	cmd->cmd.eqn = cq->eqn;
1857 	if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) {
1858 		if (dpp_cq)
1859 			cmd->cmd.pgsz_pgcnt |= OCRDMA_CREATE_CQ_DPP <<
1860 				OCRDMA_CREATE_CQ_TYPE_SHIFT;
1861 		cq->phase_change = false;
1862 		cmd->cmd.pdid_cqecnt = (cq->len / cqe_size);
1863 	} else {
1864 		cmd->cmd.pdid_cqecnt = (cq->len / cqe_size) - 1;
1865 		cmd->cmd.ev_cnt_flags |= OCRDMA_CREATE_CQ_FLAGS_AUTO_VALID;
1866 		cq->phase_change = true;
1867 	}
1868 
1869 	/* pd_id valid only for v3 */
1870 	cmd->cmd.pdid_cqecnt |= (pd_id <<
1871 		OCRDMA_CREATE_CQ_CMD_PDID_SHIFT);
1872 	ocrdma_build_q_pages(&cmd->cmd.pa[0], hw_pages, cq->pa, page_size);
1873 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1874 	if (status)
1875 		goto mbx_err;
1876 
1877 	rsp = (struct ocrdma_create_cq_rsp *)cmd;
1878 	cq->id = (u16) (rsp->rsp.cq_id & OCRDMA_CREATE_CQ_RSP_CQ_ID_MASK);
1879 	kfree(cmd);
1880 	return 0;
1881 mbx_err:
1882 	ocrdma_unbind_eq(dev, cq->eqn);
1883 	dma_free_coherent(&pdev->dev, cq->len, cq->va, cq->pa);
1884 mem_err:
1885 	kfree(cmd);
1886 	return status;
1887 }
1888 
1889 int ocrdma_mbx_destroy_cq(struct ocrdma_dev *dev, struct ocrdma_cq *cq)
1890 {
1891 	int status = -ENOMEM;
1892 	struct ocrdma_destroy_cq *cmd;
1893 
1894 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DELETE_CQ, sizeof(*cmd));
1895 	if (!cmd)
1896 		return status;
1897 	ocrdma_init_mch(&cmd->req, OCRDMA_CMD_DELETE_CQ,
1898 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
1899 
1900 	cmd->bypass_flush_qid |=
1901 	    (cq->id << OCRDMA_DESTROY_CQ_QID_SHIFT) &
1902 	    OCRDMA_DESTROY_CQ_QID_MASK;
1903 
1904 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1905 	ocrdma_unbind_eq(dev, cq->eqn);
1906 	dma_free_coherent(&dev->nic_info.pdev->dev, cq->len, cq->va, cq->pa);
1907 	kfree(cmd);
1908 	return status;
1909 }
1910 
1911 int ocrdma_mbx_alloc_lkey(struct ocrdma_dev *dev, struct ocrdma_hw_mr *hwmr,
1912 			  u32 pdid, int addr_check)
1913 {
1914 	int status = -ENOMEM;
1915 	struct ocrdma_alloc_lkey *cmd;
1916 	struct ocrdma_alloc_lkey_rsp *rsp;
1917 
1918 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_ALLOC_LKEY, sizeof(*cmd));
1919 	if (!cmd)
1920 		return status;
1921 	cmd->pdid = pdid;
1922 	cmd->pbl_sz_flags |= addr_check;
1923 	cmd->pbl_sz_flags |= (hwmr->fr_mr << OCRDMA_ALLOC_LKEY_FMR_SHIFT);
1924 	cmd->pbl_sz_flags |=
1925 	    (hwmr->remote_wr << OCRDMA_ALLOC_LKEY_REMOTE_WR_SHIFT);
1926 	cmd->pbl_sz_flags |=
1927 	    (hwmr->remote_rd << OCRDMA_ALLOC_LKEY_REMOTE_RD_SHIFT);
1928 	cmd->pbl_sz_flags |=
1929 	    (hwmr->local_wr << OCRDMA_ALLOC_LKEY_LOCAL_WR_SHIFT);
1930 	cmd->pbl_sz_flags |=
1931 	    (hwmr->remote_atomic << OCRDMA_ALLOC_LKEY_REMOTE_ATOMIC_SHIFT);
1932 	cmd->pbl_sz_flags |=
1933 	    (hwmr->num_pbls << OCRDMA_ALLOC_LKEY_PBL_SIZE_SHIFT);
1934 
1935 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1936 	if (status)
1937 		goto mbx_err;
1938 	rsp = (struct ocrdma_alloc_lkey_rsp *)cmd;
1939 	hwmr->lkey = rsp->lrkey;
1940 mbx_err:
1941 	kfree(cmd);
1942 	return status;
1943 }
1944 
1945 int ocrdma_mbx_dealloc_lkey(struct ocrdma_dev *dev, int fr_mr, u32 lkey)
1946 {
1947 	int status = -ENOMEM;
1948 	struct ocrdma_dealloc_lkey *cmd;
1949 
1950 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DEALLOC_LKEY, sizeof(*cmd));
1951 	if (!cmd)
1952 		return -ENOMEM;
1953 	cmd->lkey = lkey;
1954 	cmd->rsvd_frmr = fr_mr ? 1 : 0;
1955 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
1956 	if (status)
1957 		goto mbx_err;
1958 mbx_err:
1959 	kfree(cmd);
1960 	return status;
1961 }
1962 
1963 static int ocrdma_mbx_reg_mr(struct ocrdma_dev *dev, struct ocrdma_hw_mr *hwmr,
1964 			     u32 pdid, u32 pbl_cnt, u32 pbe_size, u32 last)
1965 {
1966 	int status = -ENOMEM;
1967 	int i;
1968 	struct ocrdma_reg_nsmr *cmd;
1969 	struct ocrdma_reg_nsmr_rsp *rsp;
1970 
1971 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_REGISTER_NSMR, sizeof(*cmd));
1972 	if (!cmd)
1973 		return -ENOMEM;
1974 	cmd->num_pbl_pdid =
1975 	    pdid | (hwmr->num_pbls << OCRDMA_REG_NSMR_NUM_PBL_SHIFT);
1976 	cmd->fr_mr = hwmr->fr_mr;
1977 
1978 	cmd->flags_hpage_pbe_sz |= (hwmr->remote_wr <<
1979 				    OCRDMA_REG_NSMR_REMOTE_WR_SHIFT);
1980 	cmd->flags_hpage_pbe_sz |= (hwmr->remote_rd <<
1981 				    OCRDMA_REG_NSMR_REMOTE_RD_SHIFT);
1982 	cmd->flags_hpage_pbe_sz |= (hwmr->local_wr <<
1983 				    OCRDMA_REG_NSMR_LOCAL_WR_SHIFT);
1984 	cmd->flags_hpage_pbe_sz |= (hwmr->remote_atomic <<
1985 				    OCRDMA_REG_NSMR_REMOTE_ATOMIC_SHIFT);
1986 	cmd->flags_hpage_pbe_sz |= (hwmr->mw_bind <<
1987 				    OCRDMA_REG_NSMR_BIND_MEMWIN_SHIFT);
1988 	cmd->flags_hpage_pbe_sz |= (last << OCRDMA_REG_NSMR_LAST_SHIFT);
1989 
1990 	cmd->flags_hpage_pbe_sz |= (hwmr->pbe_size / OCRDMA_MIN_HPAGE_SIZE);
1991 	cmd->flags_hpage_pbe_sz |= (hwmr->pbl_size / OCRDMA_MIN_HPAGE_SIZE) <<
1992 					OCRDMA_REG_NSMR_HPAGE_SIZE_SHIFT;
1993 	cmd->totlen_low = hwmr->len;
1994 	cmd->totlen_high = upper_32_bits(hwmr->len);
1995 	cmd->fbo_low = (u32) (hwmr->fbo & 0xffffffff);
1996 	cmd->fbo_high = (u32) upper_32_bits(hwmr->fbo);
1997 	cmd->va_loaddr = (u32) hwmr->va;
1998 	cmd->va_hiaddr = (u32) upper_32_bits(hwmr->va);
1999 
2000 	for (i = 0; i < pbl_cnt; i++) {
2001 		cmd->pbl[i].lo = (u32) (hwmr->pbl_table[i].pa & 0xffffffff);
2002 		cmd->pbl[i].hi = upper_32_bits(hwmr->pbl_table[i].pa);
2003 	}
2004 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2005 	if (status)
2006 		goto mbx_err;
2007 	rsp = (struct ocrdma_reg_nsmr_rsp *)cmd;
2008 	hwmr->lkey = rsp->lrkey;
2009 mbx_err:
2010 	kfree(cmd);
2011 	return status;
2012 }
2013 
2014 static int ocrdma_mbx_reg_mr_cont(struct ocrdma_dev *dev,
2015 				  struct ocrdma_hw_mr *hwmr, u32 pbl_cnt,
2016 				  u32 pbl_offset, u32 last)
2017 {
2018 	int status = -ENOMEM;
2019 	int i;
2020 	struct ocrdma_reg_nsmr_cont *cmd;
2021 
2022 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_REGISTER_NSMR_CONT, sizeof(*cmd));
2023 	if (!cmd)
2024 		return -ENOMEM;
2025 	cmd->lrkey = hwmr->lkey;
2026 	cmd->num_pbl_offset = (pbl_cnt << OCRDMA_REG_NSMR_CONT_NUM_PBL_SHIFT) |
2027 	    (pbl_offset & OCRDMA_REG_NSMR_CONT_PBL_SHIFT_MASK);
2028 	cmd->last = last << OCRDMA_REG_NSMR_CONT_LAST_SHIFT;
2029 
2030 	for (i = 0; i < pbl_cnt; i++) {
2031 		cmd->pbl[i].lo =
2032 		    (u32) (hwmr->pbl_table[i + pbl_offset].pa & 0xffffffff);
2033 		cmd->pbl[i].hi =
2034 		    upper_32_bits(hwmr->pbl_table[i + pbl_offset].pa);
2035 	}
2036 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2037 	if (status)
2038 		goto mbx_err;
2039 mbx_err:
2040 	kfree(cmd);
2041 	return status;
2042 }
2043 
2044 int ocrdma_reg_mr(struct ocrdma_dev *dev,
2045 		  struct ocrdma_hw_mr *hwmr, u32 pdid, int acc)
2046 {
2047 	int status;
2048 	u32 last = 0;
2049 	u32 cur_pbl_cnt, pbl_offset;
2050 	u32 pending_pbl_cnt = hwmr->num_pbls;
2051 
2052 	pbl_offset = 0;
2053 	cur_pbl_cnt = min(pending_pbl_cnt, MAX_OCRDMA_NSMR_PBL);
2054 	if (cur_pbl_cnt == pending_pbl_cnt)
2055 		last = 1;
2056 
2057 	status = ocrdma_mbx_reg_mr(dev, hwmr, pdid,
2058 				   cur_pbl_cnt, hwmr->pbe_size, last);
2059 	if (status) {
2060 		pr_err("%s() status=%d\n", __func__, status);
2061 		return status;
2062 	}
2063 	/* if there is no more pbls to register then exit. */
2064 	if (last)
2065 		return 0;
2066 
2067 	while (!last) {
2068 		pbl_offset += cur_pbl_cnt;
2069 		pending_pbl_cnt -= cur_pbl_cnt;
2070 		cur_pbl_cnt = min(pending_pbl_cnt, MAX_OCRDMA_NSMR_PBL);
2071 		/* if we reach the end of the pbls, then need to set the last
2072 		 * bit, indicating no more pbls to register for this memory key.
2073 		 */
2074 		if (cur_pbl_cnt == pending_pbl_cnt)
2075 			last = 1;
2076 
2077 		status = ocrdma_mbx_reg_mr_cont(dev, hwmr, cur_pbl_cnt,
2078 						pbl_offset, last);
2079 		if (status)
2080 			break;
2081 	}
2082 	if (status)
2083 		pr_err("%s() err. status=%d\n", __func__, status);
2084 
2085 	return status;
2086 }
2087 
2088 bool ocrdma_is_qp_in_sq_flushlist(struct ocrdma_cq *cq, struct ocrdma_qp *qp)
2089 {
2090 	struct ocrdma_qp *tmp;
2091 	bool found = false;
2092 	list_for_each_entry(tmp, &cq->sq_head, sq_entry) {
2093 		if (qp == tmp) {
2094 			found = true;
2095 			break;
2096 		}
2097 	}
2098 	return found;
2099 }
2100 
2101 bool ocrdma_is_qp_in_rq_flushlist(struct ocrdma_cq *cq, struct ocrdma_qp *qp)
2102 {
2103 	struct ocrdma_qp *tmp;
2104 	bool found = false;
2105 	list_for_each_entry(tmp, &cq->rq_head, rq_entry) {
2106 		if (qp == tmp) {
2107 			found = true;
2108 			break;
2109 		}
2110 	}
2111 	return found;
2112 }
2113 
2114 void ocrdma_flush_qp(struct ocrdma_qp *qp)
2115 {
2116 	bool found;
2117 	unsigned long flags;
2118 	struct ocrdma_dev *dev = get_ocrdma_dev(qp->ibqp.device);
2119 
2120 	spin_lock_irqsave(&dev->flush_q_lock, flags);
2121 	found = ocrdma_is_qp_in_sq_flushlist(qp->sq_cq, qp);
2122 	if (!found)
2123 		list_add_tail(&qp->sq_entry, &qp->sq_cq->sq_head);
2124 	if (!qp->srq) {
2125 		found = ocrdma_is_qp_in_rq_flushlist(qp->rq_cq, qp);
2126 		if (!found)
2127 			list_add_tail(&qp->rq_entry, &qp->rq_cq->rq_head);
2128 	}
2129 	spin_unlock_irqrestore(&dev->flush_q_lock, flags);
2130 }
2131 
2132 static void ocrdma_init_hwq_ptr(struct ocrdma_qp *qp)
2133 {
2134 	qp->sq.head = 0;
2135 	qp->sq.tail = 0;
2136 	qp->rq.head = 0;
2137 	qp->rq.tail = 0;
2138 }
2139 
2140 int ocrdma_qp_state_change(struct ocrdma_qp *qp, enum ib_qp_state new_ib_state,
2141 			   enum ib_qp_state *old_ib_state)
2142 {
2143 	unsigned long flags;
2144 	enum ocrdma_qp_state new_state;
2145 	new_state = get_ocrdma_qp_state(new_ib_state);
2146 
2147 	/* sync with wqe and rqe posting */
2148 	spin_lock_irqsave(&qp->q_lock, flags);
2149 
2150 	if (old_ib_state)
2151 		*old_ib_state = get_ibqp_state(qp->state);
2152 	if (new_state == qp->state) {
2153 		spin_unlock_irqrestore(&qp->q_lock, flags);
2154 		return 1;
2155 	}
2156 
2157 
2158 	if (new_state == OCRDMA_QPS_INIT) {
2159 		ocrdma_init_hwq_ptr(qp);
2160 		ocrdma_del_flush_qp(qp);
2161 	} else if (new_state == OCRDMA_QPS_ERR) {
2162 		ocrdma_flush_qp(qp);
2163 	}
2164 
2165 	qp->state = new_state;
2166 
2167 	spin_unlock_irqrestore(&qp->q_lock, flags);
2168 	return 0;
2169 }
2170 
2171 static u32 ocrdma_set_create_qp_mbx_access_flags(struct ocrdma_qp *qp)
2172 {
2173 	u32 flags = 0;
2174 	if (qp->cap_flags & OCRDMA_QP_INB_RD)
2175 		flags |= OCRDMA_CREATE_QP_REQ_INB_RDEN_MASK;
2176 	if (qp->cap_flags & OCRDMA_QP_INB_WR)
2177 		flags |= OCRDMA_CREATE_QP_REQ_INB_WREN_MASK;
2178 	if (qp->cap_flags & OCRDMA_QP_MW_BIND)
2179 		flags |= OCRDMA_CREATE_QP_REQ_BIND_MEMWIN_MASK;
2180 	if (qp->cap_flags & OCRDMA_QP_LKEY0)
2181 		flags |= OCRDMA_CREATE_QP_REQ_ZERO_LKEYEN_MASK;
2182 	if (qp->cap_flags & OCRDMA_QP_FAST_REG)
2183 		flags |= OCRDMA_CREATE_QP_REQ_FMR_EN_MASK;
2184 	return flags;
2185 }
2186 
2187 static int ocrdma_set_create_qp_sq_cmd(struct ocrdma_create_qp_req *cmd,
2188 					struct ib_qp_init_attr *attrs,
2189 					struct ocrdma_qp *qp)
2190 {
2191 	int status;
2192 	u32 len, hw_pages, hw_page_size;
2193 	dma_addr_t pa;
2194 	struct ocrdma_pd *pd = qp->pd;
2195 	struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device);
2196 	struct pci_dev *pdev = dev->nic_info.pdev;
2197 	u32 max_wqe_allocated;
2198 	u32 max_sges = attrs->cap.max_send_sge;
2199 
2200 	/* QP1 may exceed 127 */
2201 	max_wqe_allocated = min_t(u32, attrs->cap.max_send_wr + 1,
2202 				dev->attr.max_wqe);
2203 
2204 	status = ocrdma_build_q_conf(&max_wqe_allocated,
2205 		dev->attr.wqe_size, &hw_pages, &hw_page_size);
2206 	if (status) {
2207 		pr_err("%s() req. max_send_wr=0x%x\n", __func__,
2208 		       max_wqe_allocated);
2209 		return -EINVAL;
2210 	}
2211 	qp->sq.max_cnt = max_wqe_allocated;
2212 	len = (hw_pages * hw_page_size);
2213 
2214 	qp->sq.va = dma_alloc_coherent(&pdev->dev, len, &pa, GFP_KERNEL);
2215 	if (!qp->sq.va)
2216 		return -EINVAL;
2217 	memset(qp->sq.va, 0, len);
2218 	qp->sq.len = len;
2219 	qp->sq.pa = pa;
2220 	qp->sq.entry_size = dev->attr.wqe_size;
2221 	ocrdma_build_q_pages(&cmd->wq_addr[0], hw_pages, pa, hw_page_size);
2222 
2223 	cmd->type_pgsz_pdn |= (ilog2(hw_page_size / OCRDMA_MIN_Q_PAGE_SIZE)
2224 				<< OCRDMA_CREATE_QP_REQ_SQ_PAGE_SIZE_SHIFT);
2225 	cmd->num_wq_rq_pages |= (hw_pages <<
2226 				 OCRDMA_CREATE_QP_REQ_NUM_WQ_PAGES_SHIFT) &
2227 	    OCRDMA_CREATE_QP_REQ_NUM_WQ_PAGES_MASK;
2228 	cmd->max_sge_send_write |= (max_sges <<
2229 				    OCRDMA_CREATE_QP_REQ_MAX_SGE_SEND_SHIFT) &
2230 	    OCRDMA_CREATE_QP_REQ_MAX_SGE_SEND_MASK;
2231 	cmd->max_sge_send_write |= (max_sges <<
2232 				    OCRDMA_CREATE_QP_REQ_MAX_SGE_WRITE_SHIFT) &
2233 					OCRDMA_CREATE_QP_REQ_MAX_SGE_WRITE_MASK;
2234 	cmd->max_wqe_rqe |= (ilog2(qp->sq.max_cnt) <<
2235 			     OCRDMA_CREATE_QP_REQ_MAX_WQE_SHIFT) &
2236 				OCRDMA_CREATE_QP_REQ_MAX_WQE_MASK;
2237 	cmd->wqe_rqe_size |= (dev->attr.wqe_size <<
2238 			      OCRDMA_CREATE_QP_REQ_WQE_SIZE_SHIFT) &
2239 				OCRDMA_CREATE_QP_REQ_WQE_SIZE_MASK;
2240 	return 0;
2241 }
2242 
2243 static int ocrdma_set_create_qp_rq_cmd(struct ocrdma_create_qp_req *cmd,
2244 					struct ib_qp_init_attr *attrs,
2245 					struct ocrdma_qp *qp)
2246 {
2247 	int status;
2248 	u32 len, hw_pages, hw_page_size;
2249 	dma_addr_t pa = 0;
2250 	struct ocrdma_pd *pd = qp->pd;
2251 	struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device);
2252 	struct pci_dev *pdev = dev->nic_info.pdev;
2253 	u32 max_rqe_allocated = attrs->cap.max_recv_wr + 1;
2254 
2255 	status = ocrdma_build_q_conf(&max_rqe_allocated, dev->attr.rqe_size,
2256 				     &hw_pages, &hw_page_size);
2257 	if (status) {
2258 		pr_err("%s() req. max_recv_wr=0x%x\n", __func__,
2259 		       attrs->cap.max_recv_wr + 1);
2260 		return status;
2261 	}
2262 	qp->rq.max_cnt = max_rqe_allocated;
2263 	len = (hw_pages * hw_page_size);
2264 
2265 	qp->rq.va = dma_alloc_coherent(&pdev->dev, len, &pa, GFP_KERNEL);
2266 	if (!qp->rq.va)
2267 		return -ENOMEM;
2268 	memset(qp->rq.va, 0, len);
2269 	qp->rq.pa = pa;
2270 	qp->rq.len = len;
2271 	qp->rq.entry_size = dev->attr.rqe_size;
2272 
2273 	ocrdma_build_q_pages(&cmd->rq_addr[0], hw_pages, pa, hw_page_size);
2274 	cmd->type_pgsz_pdn |= (ilog2(hw_page_size / OCRDMA_MIN_Q_PAGE_SIZE) <<
2275 		OCRDMA_CREATE_QP_REQ_RQ_PAGE_SIZE_SHIFT);
2276 	cmd->num_wq_rq_pages |=
2277 	    (hw_pages << OCRDMA_CREATE_QP_REQ_NUM_RQ_PAGES_SHIFT) &
2278 	    OCRDMA_CREATE_QP_REQ_NUM_RQ_PAGES_MASK;
2279 	cmd->max_sge_recv_flags |= (attrs->cap.max_recv_sge <<
2280 				OCRDMA_CREATE_QP_REQ_MAX_SGE_RECV_SHIFT) &
2281 				OCRDMA_CREATE_QP_REQ_MAX_SGE_RECV_MASK;
2282 	cmd->max_wqe_rqe |= (ilog2(qp->rq.max_cnt) <<
2283 				OCRDMA_CREATE_QP_REQ_MAX_RQE_SHIFT) &
2284 				OCRDMA_CREATE_QP_REQ_MAX_RQE_MASK;
2285 	cmd->wqe_rqe_size |= (dev->attr.rqe_size <<
2286 			OCRDMA_CREATE_QP_REQ_RQE_SIZE_SHIFT) &
2287 			OCRDMA_CREATE_QP_REQ_RQE_SIZE_MASK;
2288 	return 0;
2289 }
2290 
2291 static void ocrdma_set_create_qp_dpp_cmd(struct ocrdma_create_qp_req *cmd,
2292 					 struct ocrdma_pd *pd,
2293 					 struct ocrdma_qp *qp,
2294 					 u8 enable_dpp_cq, u16 dpp_cq_id)
2295 {
2296 	pd->num_dpp_qp--;
2297 	qp->dpp_enabled = true;
2298 	cmd->max_sge_recv_flags |= OCRDMA_CREATE_QP_REQ_ENABLE_DPP_MASK;
2299 	if (!enable_dpp_cq)
2300 		return;
2301 	cmd->max_sge_recv_flags |= OCRDMA_CREATE_QP_REQ_ENABLE_DPP_MASK;
2302 	cmd->dpp_credits_cqid = dpp_cq_id;
2303 	cmd->dpp_credits_cqid |= OCRDMA_CREATE_QP_REQ_DPP_CREDIT_LIMIT <<
2304 					OCRDMA_CREATE_QP_REQ_DPP_CREDIT_SHIFT;
2305 }
2306 
2307 static int ocrdma_set_create_qp_ird_cmd(struct ocrdma_create_qp_req *cmd,
2308 					struct ocrdma_qp *qp)
2309 {
2310 	struct ocrdma_pd *pd = qp->pd;
2311 	struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device);
2312 	struct pci_dev *pdev = dev->nic_info.pdev;
2313 	dma_addr_t pa = 0;
2314 	int ird_page_size = dev->attr.ird_page_size;
2315 	int ird_q_len = dev->attr.num_ird_pages * ird_page_size;
2316 	struct ocrdma_hdr_wqe *rqe;
2317 	int i  = 0;
2318 
2319 	if (dev->attr.ird == 0)
2320 		return 0;
2321 
2322 	qp->ird_q_va = dma_alloc_coherent(&pdev->dev, ird_q_len,
2323 					&pa, GFP_KERNEL);
2324 	if (!qp->ird_q_va)
2325 		return -ENOMEM;
2326 	memset(qp->ird_q_va, 0, ird_q_len);
2327 	ocrdma_build_q_pages(&cmd->ird_addr[0], dev->attr.num_ird_pages,
2328 			     pa, ird_page_size);
2329 	for (; i < ird_q_len / dev->attr.rqe_size; i++) {
2330 		rqe = (struct ocrdma_hdr_wqe *)(qp->ird_q_va +
2331 			(i * dev->attr.rqe_size));
2332 		rqe->cw = 0;
2333 		rqe->cw |= 2;
2334 		rqe->cw |= (OCRDMA_TYPE_LKEY << OCRDMA_WQE_TYPE_SHIFT);
2335 		rqe->cw |= (8 << OCRDMA_WQE_SIZE_SHIFT);
2336 		rqe->cw |= (8 << OCRDMA_WQE_NXT_WQE_SIZE_SHIFT);
2337 	}
2338 	return 0;
2339 }
2340 
2341 static void ocrdma_get_create_qp_rsp(struct ocrdma_create_qp_rsp *rsp,
2342 				     struct ocrdma_qp *qp,
2343 				     struct ib_qp_init_attr *attrs,
2344 				     u16 *dpp_offset, u16 *dpp_credit_lmt)
2345 {
2346 	u32 max_wqe_allocated, max_rqe_allocated;
2347 	qp->id = rsp->qp_id & OCRDMA_CREATE_QP_RSP_QP_ID_MASK;
2348 	qp->rq.dbid = rsp->sq_rq_id & OCRDMA_CREATE_QP_RSP_RQ_ID_MASK;
2349 	qp->sq.dbid = rsp->sq_rq_id >> OCRDMA_CREATE_QP_RSP_SQ_ID_SHIFT;
2350 	qp->max_ird = rsp->max_ord_ird & OCRDMA_CREATE_QP_RSP_MAX_IRD_MASK;
2351 	qp->max_ord = (rsp->max_ord_ird >> OCRDMA_CREATE_QP_RSP_MAX_ORD_SHIFT);
2352 	qp->dpp_enabled = false;
2353 	if (rsp->dpp_response & OCRDMA_CREATE_QP_RSP_DPP_ENABLED_MASK) {
2354 		qp->dpp_enabled = true;
2355 		*dpp_credit_lmt = (rsp->dpp_response &
2356 				OCRDMA_CREATE_QP_RSP_DPP_CREDITS_MASK) >>
2357 				OCRDMA_CREATE_QP_RSP_DPP_CREDITS_SHIFT;
2358 		*dpp_offset = (rsp->dpp_response &
2359 				OCRDMA_CREATE_QP_RSP_DPP_PAGE_OFFSET_MASK) >>
2360 				OCRDMA_CREATE_QP_RSP_DPP_PAGE_OFFSET_SHIFT;
2361 	}
2362 	max_wqe_allocated =
2363 		rsp->max_wqe_rqe >> OCRDMA_CREATE_QP_RSP_MAX_WQE_SHIFT;
2364 	max_wqe_allocated = 1 << max_wqe_allocated;
2365 	max_rqe_allocated = 1 << ((u16)rsp->max_wqe_rqe);
2366 
2367 	qp->sq.max_cnt = max_wqe_allocated;
2368 	qp->sq.max_wqe_idx = max_wqe_allocated - 1;
2369 
2370 	if (!attrs->srq) {
2371 		qp->rq.max_cnt = max_rqe_allocated;
2372 		qp->rq.max_wqe_idx = max_rqe_allocated - 1;
2373 	}
2374 }
2375 
2376 int ocrdma_mbx_create_qp(struct ocrdma_qp *qp, struct ib_qp_init_attr *attrs,
2377 			 u8 enable_dpp_cq, u16 dpp_cq_id, u16 *dpp_offset,
2378 			 u16 *dpp_credit_lmt)
2379 {
2380 	int status = -ENOMEM;
2381 	u32 flags = 0;
2382 	struct ocrdma_pd *pd = qp->pd;
2383 	struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device);
2384 	struct pci_dev *pdev = dev->nic_info.pdev;
2385 	struct ocrdma_cq *cq;
2386 	struct ocrdma_create_qp_req *cmd;
2387 	struct ocrdma_create_qp_rsp *rsp;
2388 	int qptype;
2389 
2390 	switch (attrs->qp_type) {
2391 	case IB_QPT_GSI:
2392 		qptype = OCRDMA_QPT_GSI;
2393 		break;
2394 	case IB_QPT_RC:
2395 		qptype = OCRDMA_QPT_RC;
2396 		break;
2397 	case IB_QPT_UD:
2398 		qptype = OCRDMA_QPT_UD;
2399 		break;
2400 	default:
2401 		return -EINVAL;
2402 	}
2403 
2404 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_CREATE_QP, sizeof(*cmd));
2405 	if (!cmd)
2406 		return status;
2407 	cmd->type_pgsz_pdn |= (qptype << OCRDMA_CREATE_QP_REQ_QPT_SHIFT) &
2408 						OCRDMA_CREATE_QP_REQ_QPT_MASK;
2409 	status = ocrdma_set_create_qp_sq_cmd(cmd, attrs, qp);
2410 	if (status)
2411 		goto sq_err;
2412 
2413 	if (attrs->srq) {
2414 		struct ocrdma_srq *srq = get_ocrdma_srq(attrs->srq);
2415 		cmd->max_sge_recv_flags |= OCRDMA_CREATE_QP_REQ_USE_SRQ_MASK;
2416 		cmd->rq_addr[0].lo = srq->id;
2417 		qp->srq = srq;
2418 	} else {
2419 		status = ocrdma_set_create_qp_rq_cmd(cmd, attrs, qp);
2420 		if (status)
2421 			goto rq_err;
2422 	}
2423 
2424 	status = ocrdma_set_create_qp_ird_cmd(cmd, qp);
2425 	if (status)
2426 		goto mbx_err;
2427 
2428 	cmd->type_pgsz_pdn |= (pd->id << OCRDMA_CREATE_QP_REQ_PD_ID_SHIFT) &
2429 				OCRDMA_CREATE_QP_REQ_PD_ID_MASK;
2430 
2431 	flags = ocrdma_set_create_qp_mbx_access_flags(qp);
2432 
2433 	cmd->max_sge_recv_flags |= flags;
2434 	cmd->max_ord_ird |= (dev->attr.max_ord_per_qp <<
2435 			     OCRDMA_CREATE_QP_REQ_MAX_ORD_SHIFT) &
2436 				OCRDMA_CREATE_QP_REQ_MAX_ORD_MASK;
2437 	cmd->max_ord_ird |= (dev->attr.max_ird_per_qp <<
2438 			     OCRDMA_CREATE_QP_REQ_MAX_IRD_SHIFT) &
2439 				OCRDMA_CREATE_QP_REQ_MAX_IRD_MASK;
2440 	cq = get_ocrdma_cq(attrs->send_cq);
2441 	cmd->wq_rq_cqid |= (cq->id << OCRDMA_CREATE_QP_REQ_WQ_CQID_SHIFT) &
2442 				OCRDMA_CREATE_QP_REQ_WQ_CQID_MASK;
2443 	qp->sq_cq = cq;
2444 	cq = get_ocrdma_cq(attrs->recv_cq);
2445 	cmd->wq_rq_cqid |= (cq->id << OCRDMA_CREATE_QP_REQ_RQ_CQID_SHIFT) &
2446 				OCRDMA_CREATE_QP_REQ_RQ_CQID_MASK;
2447 	qp->rq_cq = cq;
2448 
2449 	if (pd->dpp_enabled && attrs->cap.max_inline_data && pd->num_dpp_qp &&
2450 	    (attrs->cap.max_inline_data <= dev->attr.max_inline_data)) {
2451 		ocrdma_set_create_qp_dpp_cmd(cmd, pd, qp, enable_dpp_cq,
2452 					     dpp_cq_id);
2453 	}
2454 
2455 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2456 	if (status)
2457 		goto mbx_err;
2458 	rsp = (struct ocrdma_create_qp_rsp *)cmd;
2459 	ocrdma_get_create_qp_rsp(rsp, qp, attrs, dpp_offset, dpp_credit_lmt);
2460 	qp->state = OCRDMA_QPS_RST;
2461 	kfree(cmd);
2462 	return 0;
2463 mbx_err:
2464 	if (qp->rq.va)
2465 		dma_free_coherent(&pdev->dev, qp->rq.len, qp->rq.va, qp->rq.pa);
2466 rq_err:
2467 	pr_err("%s(%d) rq_err\n", __func__, dev->id);
2468 	dma_free_coherent(&pdev->dev, qp->sq.len, qp->sq.va, qp->sq.pa);
2469 sq_err:
2470 	pr_err("%s(%d) sq_err\n", __func__, dev->id);
2471 	kfree(cmd);
2472 	return status;
2473 }
2474 
2475 int ocrdma_mbx_query_qp(struct ocrdma_dev *dev, struct ocrdma_qp *qp,
2476 			struct ocrdma_qp_params *param)
2477 {
2478 	int status = -ENOMEM;
2479 	struct ocrdma_query_qp *cmd;
2480 	struct ocrdma_query_qp_rsp *rsp;
2481 
2482 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_QP, sizeof(*rsp));
2483 	if (!cmd)
2484 		return status;
2485 	cmd->qp_id = qp->id;
2486 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2487 	if (status)
2488 		goto mbx_err;
2489 	rsp = (struct ocrdma_query_qp_rsp *)cmd;
2490 	memcpy(param, &rsp->params, sizeof(struct ocrdma_qp_params));
2491 mbx_err:
2492 	kfree(cmd);
2493 	return status;
2494 }
2495 
2496 static int ocrdma_set_av_params(struct ocrdma_qp *qp,
2497 				struct ocrdma_modify_qp *cmd,
2498 				struct ib_qp_attr *attrs,
2499 				int attr_mask)
2500 {
2501 	int status;
2502 	struct ib_ah_attr *ah_attr = &attrs->ah_attr;
2503 	union ib_gid sgid, zgid;
2504 	struct ib_gid_attr sgid_attr;
2505 	u32 vlan_id = 0xFFFF;
2506 	u8 mac_addr[6], hdr_type;
2507 	union {
2508 		struct sockaddr     _sockaddr;
2509 		struct sockaddr_in  _sockaddr_in;
2510 		struct sockaddr_in6 _sockaddr_in6;
2511 	} sgid_addr, dgid_addr;
2512 	struct ocrdma_dev *dev = get_ocrdma_dev(qp->ibqp.device);
2513 
2514 	if ((ah_attr->ah_flags & IB_AH_GRH) == 0)
2515 		return -EINVAL;
2516 	if (atomic_cmpxchg(&dev->update_sl, 1, 0))
2517 		ocrdma_init_service_level(dev);
2518 	cmd->params.tclass_sq_psn |=
2519 	    (ah_attr->grh.traffic_class << OCRDMA_QP_PARAMS_TCLASS_SHIFT);
2520 	cmd->params.rnt_rc_sl_fl |=
2521 	    (ah_attr->grh.flow_label & OCRDMA_QP_PARAMS_FLOW_LABEL_MASK);
2522 	cmd->params.rnt_rc_sl_fl |= (ah_attr->sl << OCRDMA_QP_PARAMS_SL_SHIFT);
2523 	cmd->params.hop_lmt_rq_psn |=
2524 	    (ah_attr->grh.hop_limit << OCRDMA_QP_PARAMS_HOP_LMT_SHIFT);
2525 	cmd->flags |= OCRDMA_QP_PARA_FLOW_LBL_VALID;
2526 
2527 	/* GIDs */
2528 	memcpy(&cmd->params.dgid[0], &ah_attr->grh.dgid.raw[0],
2529 	       sizeof(cmd->params.dgid));
2530 
2531 	status = ib_get_cached_gid(&dev->ibdev, 1, ah_attr->grh.sgid_index,
2532 				   &sgid, &sgid_attr);
2533 	if (!status && sgid_attr.ndev) {
2534 		vlan_id = rdma_vlan_dev_vlan_id(sgid_attr.ndev);
2535 		memcpy(mac_addr, sgid_attr.ndev->dev_addr, ETH_ALEN);
2536 		dev_put(sgid_attr.ndev);
2537 	}
2538 
2539 	memset(&zgid, 0, sizeof(zgid));
2540 	if (!memcmp(&sgid, &zgid, sizeof(zgid)))
2541 		return -EINVAL;
2542 
2543 	qp->sgid_idx = ah_attr->grh.sgid_index;
2544 	memcpy(&cmd->params.sgid[0], &sgid.raw[0], sizeof(cmd->params.sgid));
2545 	status = ocrdma_resolve_dmac(dev, ah_attr, &mac_addr[0]);
2546 	if (status)
2547 		return status;
2548 	cmd->params.dmac_b0_to_b3 = mac_addr[0] | (mac_addr[1] << 8) |
2549 				(mac_addr[2] << 16) | (mac_addr[3] << 24);
2550 
2551 	hdr_type = ib_gid_to_network_type(sgid_attr.gid_type, &sgid);
2552 	if (hdr_type == RDMA_NETWORK_IPV4) {
2553 		rdma_gid2ip(&sgid_addr._sockaddr, &sgid);
2554 		rdma_gid2ip(&dgid_addr._sockaddr, &ah_attr->grh.dgid);
2555 		memcpy(&cmd->params.dgid[0],
2556 		       &dgid_addr._sockaddr_in.sin_addr.s_addr, 4);
2557 		memcpy(&cmd->params.sgid[0],
2558 		       &sgid_addr._sockaddr_in.sin_addr.s_addr, 4);
2559 	}
2560 	/* convert them to LE format. */
2561 	ocrdma_cpu_to_le32(&cmd->params.dgid[0], sizeof(cmd->params.dgid));
2562 	ocrdma_cpu_to_le32(&cmd->params.sgid[0], sizeof(cmd->params.sgid));
2563 	cmd->params.vlan_dmac_b4_to_b5 = mac_addr[4] | (mac_addr[5] << 8);
2564 
2565 	if (vlan_id == 0xFFFF)
2566 		vlan_id = 0;
2567 	if (vlan_id || dev->pfc_state) {
2568 		if (!vlan_id) {
2569 			pr_err("ocrdma%d:Using VLAN with PFC is recommended\n",
2570 			       dev->id);
2571 			pr_err("ocrdma%d:Using VLAN 0 for this connection\n",
2572 			       dev->id);
2573 		}
2574 		cmd->params.vlan_dmac_b4_to_b5 |=
2575 		    vlan_id << OCRDMA_QP_PARAMS_VLAN_SHIFT;
2576 		cmd->flags |= OCRDMA_QP_PARA_VLAN_EN_VALID;
2577 		cmd->params.rnt_rc_sl_fl |=
2578 			(dev->sl & 0x07) << OCRDMA_QP_PARAMS_SL_SHIFT;
2579 	}
2580 	cmd->params.max_sge_recv_flags |= ((hdr_type <<
2581 					OCRDMA_QP_PARAMS_FLAGS_L3_TYPE_SHIFT) &
2582 					OCRDMA_QP_PARAMS_FLAGS_L3_TYPE_MASK);
2583 	return 0;
2584 }
2585 
2586 static int ocrdma_set_qp_params(struct ocrdma_qp *qp,
2587 				struct ocrdma_modify_qp *cmd,
2588 				struct ib_qp_attr *attrs, int attr_mask)
2589 {
2590 	int status = 0;
2591 	struct ocrdma_dev *dev = get_ocrdma_dev(qp->ibqp.device);
2592 
2593 	if (attr_mask & IB_QP_PKEY_INDEX) {
2594 		cmd->params.path_mtu_pkey_indx |= (attrs->pkey_index &
2595 					    OCRDMA_QP_PARAMS_PKEY_INDEX_MASK);
2596 		cmd->flags |= OCRDMA_QP_PARA_PKEY_VALID;
2597 	}
2598 	if (attr_mask & IB_QP_QKEY) {
2599 		qp->qkey = attrs->qkey;
2600 		cmd->params.qkey = attrs->qkey;
2601 		cmd->flags |= OCRDMA_QP_PARA_QKEY_VALID;
2602 	}
2603 	if (attr_mask & IB_QP_AV) {
2604 		status = ocrdma_set_av_params(qp, cmd, attrs, attr_mask);
2605 		if (status)
2606 			return status;
2607 	} else if (qp->qp_type == IB_QPT_GSI || qp->qp_type == IB_QPT_UD) {
2608 		/* set the default mac address for UD, GSI QPs */
2609 		cmd->params.dmac_b0_to_b3 = dev->nic_info.mac_addr[0] |
2610 			(dev->nic_info.mac_addr[1] << 8) |
2611 			(dev->nic_info.mac_addr[2] << 16) |
2612 			(dev->nic_info.mac_addr[3] << 24);
2613 		cmd->params.vlan_dmac_b4_to_b5 = dev->nic_info.mac_addr[4] |
2614 					(dev->nic_info.mac_addr[5] << 8);
2615 	}
2616 	if ((attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY) &&
2617 	    attrs->en_sqd_async_notify) {
2618 		cmd->params.max_sge_recv_flags |=
2619 			OCRDMA_QP_PARAMS_FLAGS_SQD_ASYNC;
2620 		cmd->flags |= OCRDMA_QP_PARA_DST_QPN_VALID;
2621 	}
2622 	if (attr_mask & IB_QP_DEST_QPN) {
2623 		cmd->params.ack_to_rnr_rtc_dest_qpn |= (attrs->dest_qp_num &
2624 				OCRDMA_QP_PARAMS_DEST_QPN_MASK);
2625 		cmd->flags |= OCRDMA_QP_PARA_DST_QPN_VALID;
2626 	}
2627 	if (attr_mask & IB_QP_PATH_MTU) {
2628 		if (attrs->path_mtu < IB_MTU_512 ||
2629 		    attrs->path_mtu > IB_MTU_4096) {
2630 			pr_err("ocrdma%d: IB MTU %d is not supported\n",
2631 			       dev->id, ib_mtu_enum_to_int(attrs->path_mtu));
2632 			status = -EINVAL;
2633 			goto pmtu_err;
2634 		}
2635 		cmd->params.path_mtu_pkey_indx |=
2636 		    (ib_mtu_enum_to_int(attrs->path_mtu) <<
2637 		     OCRDMA_QP_PARAMS_PATH_MTU_SHIFT) &
2638 		    OCRDMA_QP_PARAMS_PATH_MTU_MASK;
2639 		cmd->flags |= OCRDMA_QP_PARA_PMTU_VALID;
2640 	}
2641 	if (attr_mask & IB_QP_TIMEOUT) {
2642 		cmd->params.ack_to_rnr_rtc_dest_qpn |= attrs->timeout <<
2643 		    OCRDMA_QP_PARAMS_ACK_TIMEOUT_SHIFT;
2644 		cmd->flags |= OCRDMA_QP_PARA_ACK_TO_VALID;
2645 	}
2646 	if (attr_mask & IB_QP_RETRY_CNT) {
2647 		cmd->params.rnt_rc_sl_fl |= (attrs->retry_cnt <<
2648 				      OCRDMA_QP_PARAMS_RETRY_CNT_SHIFT) &
2649 		    OCRDMA_QP_PARAMS_RETRY_CNT_MASK;
2650 		cmd->flags |= OCRDMA_QP_PARA_RETRY_CNT_VALID;
2651 	}
2652 	if (attr_mask & IB_QP_MIN_RNR_TIMER) {
2653 		cmd->params.rnt_rc_sl_fl |= (attrs->min_rnr_timer <<
2654 				      OCRDMA_QP_PARAMS_RNR_NAK_TIMER_SHIFT) &
2655 		    OCRDMA_QP_PARAMS_RNR_NAK_TIMER_MASK;
2656 		cmd->flags |= OCRDMA_QP_PARA_RNT_VALID;
2657 	}
2658 	if (attr_mask & IB_QP_RNR_RETRY) {
2659 		cmd->params.ack_to_rnr_rtc_dest_qpn |= (attrs->rnr_retry <<
2660 			OCRDMA_QP_PARAMS_RNR_RETRY_CNT_SHIFT)
2661 			& OCRDMA_QP_PARAMS_RNR_RETRY_CNT_MASK;
2662 		cmd->flags |= OCRDMA_QP_PARA_RRC_VALID;
2663 	}
2664 	if (attr_mask & IB_QP_SQ_PSN) {
2665 		cmd->params.tclass_sq_psn |= (attrs->sq_psn & 0x00ffffff);
2666 		cmd->flags |= OCRDMA_QP_PARA_SQPSN_VALID;
2667 	}
2668 	if (attr_mask & IB_QP_RQ_PSN) {
2669 		cmd->params.hop_lmt_rq_psn |= (attrs->rq_psn & 0x00ffffff);
2670 		cmd->flags |= OCRDMA_QP_PARA_RQPSN_VALID;
2671 	}
2672 	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
2673 		if (attrs->max_rd_atomic > dev->attr.max_ord_per_qp) {
2674 			status = -EINVAL;
2675 			goto pmtu_err;
2676 		}
2677 		qp->max_ord = attrs->max_rd_atomic;
2678 		cmd->flags |= OCRDMA_QP_PARA_MAX_ORD_VALID;
2679 	}
2680 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
2681 		if (attrs->max_dest_rd_atomic > dev->attr.max_ird_per_qp) {
2682 			status = -EINVAL;
2683 			goto pmtu_err;
2684 		}
2685 		qp->max_ird = attrs->max_dest_rd_atomic;
2686 		cmd->flags |= OCRDMA_QP_PARA_MAX_IRD_VALID;
2687 	}
2688 	cmd->params.max_ord_ird = (qp->max_ord <<
2689 				OCRDMA_QP_PARAMS_MAX_ORD_SHIFT) |
2690 				(qp->max_ird & OCRDMA_QP_PARAMS_MAX_IRD_MASK);
2691 pmtu_err:
2692 	return status;
2693 }
2694 
2695 int ocrdma_mbx_modify_qp(struct ocrdma_dev *dev, struct ocrdma_qp *qp,
2696 			 struct ib_qp_attr *attrs, int attr_mask)
2697 {
2698 	int status = -ENOMEM;
2699 	struct ocrdma_modify_qp *cmd;
2700 
2701 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_MODIFY_QP, sizeof(*cmd));
2702 	if (!cmd)
2703 		return status;
2704 
2705 	cmd->params.id = qp->id;
2706 	cmd->flags = 0;
2707 	if (attr_mask & IB_QP_STATE) {
2708 		cmd->params.max_sge_recv_flags |=
2709 		    (get_ocrdma_qp_state(attrs->qp_state) <<
2710 		     OCRDMA_QP_PARAMS_STATE_SHIFT) &
2711 		    OCRDMA_QP_PARAMS_STATE_MASK;
2712 		cmd->flags |= OCRDMA_QP_PARA_QPS_VALID;
2713 	} else {
2714 		cmd->params.max_sge_recv_flags |=
2715 		    (qp->state << OCRDMA_QP_PARAMS_STATE_SHIFT) &
2716 		    OCRDMA_QP_PARAMS_STATE_MASK;
2717 	}
2718 
2719 	status = ocrdma_set_qp_params(qp, cmd, attrs, attr_mask);
2720 	if (status)
2721 		goto mbx_err;
2722 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2723 	if (status)
2724 		goto mbx_err;
2725 
2726 mbx_err:
2727 	kfree(cmd);
2728 	return status;
2729 }
2730 
2731 int ocrdma_mbx_destroy_qp(struct ocrdma_dev *dev, struct ocrdma_qp *qp)
2732 {
2733 	int status = -ENOMEM;
2734 	struct ocrdma_destroy_qp *cmd;
2735 	struct pci_dev *pdev = dev->nic_info.pdev;
2736 
2737 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DELETE_QP, sizeof(*cmd));
2738 	if (!cmd)
2739 		return status;
2740 	cmd->qp_id = qp->id;
2741 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2742 	if (status)
2743 		goto mbx_err;
2744 
2745 mbx_err:
2746 	kfree(cmd);
2747 	if (qp->sq.va)
2748 		dma_free_coherent(&pdev->dev, qp->sq.len, qp->sq.va, qp->sq.pa);
2749 	if (!qp->srq && qp->rq.va)
2750 		dma_free_coherent(&pdev->dev, qp->rq.len, qp->rq.va, qp->rq.pa);
2751 	if (qp->dpp_enabled)
2752 		qp->pd->num_dpp_qp++;
2753 	return status;
2754 }
2755 
2756 int ocrdma_mbx_create_srq(struct ocrdma_dev *dev, struct ocrdma_srq *srq,
2757 			  struct ib_srq_init_attr *srq_attr,
2758 			  struct ocrdma_pd *pd)
2759 {
2760 	int status = -ENOMEM;
2761 	int hw_pages, hw_page_size;
2762 	int len;
2763 	struct ocrdma_create_srq_rsp *rsp;
2764 	struct ocrdma_create_srq *cmd;
2765 	dma_addr_t pa;
2766 	struct pci_dev *pdev = dev->nic_info.pdev;
2767 	u32 max_rqe_allocated;
2768 
2769 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_CREATE_SRQ, sizeof(*cmd));
2770 	if (!cmd)
2771 		return status;
2772 
2773 	cmd->pgsz_pdid = pd->id & OCRDMA_CREATE_SRQ_PD_ID_MASK;
2774 	max_rqe_allocated = srq_attr->attr.max_wr + 1;
2775 	status = ocrdma_build_q_conf(&max_rqe_allocated,
2776 				dev->attr.rqe_size,
2777 				&hw_pages, &hw_page_size);
2778 	if (status) {
2779 		pr_err("%s() req. max_wr=0x%x\n", __func__,
2780 		       srq_attr->attr.max_wr);
2781 		status = -EINVAL;
2782 		goto ret;
2783 	}
2784 	len = hw_pages * hw_page_size;
2785 	srq->rq.va = dma_alloc_coherent(&pdev->dev, len, &pa, GFP_KERNEL);
2786 	if (!srq->rq.va) {
2787 		status = -ENOMEM;
2788 		goto ret;
2789 	}
2790 	ocrdma_build_q_pages(&cmd->rq_addr[0], hw_pages, pa, hw_page_size);
2791 
2792 	srq->rq.entry_size = dev->attr.rqe_size;
2793 	srq->rq.pa = pa;
2794 	srq->rq.len = len;
2795 	srq->rq.max_cnt = max_rqe_allocated;
2796 
2797 	cmd->max_sge_rqe = ilog2(max_rqe_allocated);
2798 	cmd->max_sge_rqe |= srq_attr->attr.max_sge <<
2799 				OCRDMA_CREATE_SRQ_MAX_SGE_RECV_SHIFT;
2800 
2801 	cmd->pgsz_pdid |= (ilog2(hw_page_size / OCRDMA_MIN_Q_PAGE_SIZE)
2802 		<< OCRDMA_CREATE_SRQ_PG_SZ_SHIFT);
2803 	cmd->pages_rqe_sz |= (dev->attr.rqe_size
2804 		<< OCRDMA_CREATE_SRQ_RQE_SIZE_SHIFT)
2805 		& OCRDMA_CREATE_SRQ_RQE_SIZE_MASK;
2806 	cmd->pages_rqe_sz |= hw_pages << OCRDMA_CREATE_SRQ_NUM_RQ_PAGES_SHIFT;
2807 
2808 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2809 	if (status)
2810 		goto mbx_err;
2811 	rsp = (struct ocrdma_create_srq_rsp *)cmd;
2812 	srq->id = rsp->id;
2813 	srq->rq.dbid = rsp->id;
2814 	max_rqe_allocated = ((rsp->max_sge_rqe_allocated &
2815 		OCRDMA_CREATE_SRQ_RSP_MAX_RQE_ALLOCATED_MASK) >>
2816 		OCRDMA_CREATE_SRQ_RSP_MAX_RQE_ALLOCATED_SHIFT);
2817 	max_rqe_allocated = (1 << max_rqe_allocated);
2818 	srq->rq.max_cnt = max_rqe_allocated;
2819 	srq->rq.max_wqe_idx = max_rqe_allocated - 1;
2820 	srq->rq.max_sges = (rsp->max_sge_rqe_allocated &
2821 		OCRDMA_CREATE_SRQ_RSP_MAX_SGE_RECV_ALLOCATED_MASK) >>
2822 		OCRDMA_CREATE_SRQ_RSP_MAX_SGE_RECV_ALLOCATED_SHIFT;
2823 	goto ret;
2824 mbx_err:
2825 	dma_free_coherent(&pdev->dev, srq->rq.len, srq->rq.va, pa);
2826 ret:
2827 	kfree(cmd);
2828 	return status;
2829 }
2830 
2831 int ocrdma_mbx_modify_srq(struct ocrdma_srq *srq, struct ib_srq_attr *srq_attr)
2832 {
2833 	int status = -ENOMEM;
2834 	struct ocrdma_modify_srq *cmd;
2835 	struct ocrdma_pd *pd = srq->pd;
2836 	struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device);
2837 
2838 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_MODIFY_SRQ, sizeof(*cmd));
2839 	if (!cmd)
2840 		return status;
2841 	cmd->id = srq->id;
2842 	cmd->limit_max_rqe |= srq_attr->srq_limit <<
2843 	    OCRDMA_MODIFY_SRQ_LIMIT_SHIFT;
2844 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2845 	kfree(cmd);
2846 	return status;
2847 }
2848 
2849 int ocrdma_mbx_query_srq(struct ocrdma_srq *srq, struct ib_srq_attr *srq_attr)
2850 {
2851 	int status = -ENOMEM;
2852 	struct ocrdma_query_srq *cmd;
2853 	struct ocrdma_dev *dev = get_ocrdma_dev(srq->ibsrq.device);
2854 
2855 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_SRQ, sizeof(*cmd));
2856 	if (!cmd)
2857 		return status;
2858 	cmd->id = srq->rq.dbid;
2859 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2860 	if (status == 0) {
2861 		struct ocrdma_query_srq_rsp *rsp =
2862 		    (struct ocrdma_query_srq_rsp *)cmd;
2863 		srq_attr->max_sge =
2864 		    rsp->srq_lmt_max_sge &
2865 		    OCRDMA_QUERY_SRQ_RSP_MAX_SGE_RECV_MASK;
2866 		srq_attr->max_wr =
2867 		    rsp->max_rqe_pdid >> OCRDMA_QUERY_SRQ_RSP_MAX_RQE_SHIFT;
2868 		srq_attr->srq_limit = rsp->srq_lmt_max_sge >>
2869 		    OCRDMA_QUERY_SRQ_RSP_SRQ_LIMIT_SHIFT;
2870 	}
2871 	kfree(cmd);
2872 	return status;
2873 }
2874 
2875 int ocrdma_mbx_destroy_srq(struct ocrdma_dev *dev, struct ocrdma_srq *srq)
2876 {
2877 	int status = -ENOMEM;
2878 	struct ocrdma_destroy_srq *cmd;
2879 	struct pci_dev *pdev = dev->nic_info.pdev;
2880 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DELETE_SRQ, sizeof(*cmd));
2881 	if (!cmd)
2882 		return status;
2883 	cmd->id = srq->id;
2884 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
2885 	if (srq->rq.va)
2886 		dma_free_coherent(&pdev->dev, srq->rq.len,
2887 				  srq->rq.va, srq->rq.pa);
2888 	kfree(cmd);
2889 	return status;
2890 }
2891 
2892 static int ocrdma_mbx_get_dcbx_config(struct ocrdma_dev *dev, u32 ptype,
2893 				      struct ocrdma_dcbx_cfg *dcbxcfg)
2894 {
2895 	int status;
2896 	dma_addr_t pa;
2897 	struct ocrdma_mqe cmd;
2898 
2899 	struct ocrdma_get_dcbx_cfg_req *req = NULL;
2900 	struct ocrdma_get_dcbx_cfg_rsp *rsp = NULL;
2901 	struct pci_dev *pdev = dev->nic_info.pdev;
2902 	struct ocrdma_mqe_sge *mqe_sge = cmd.u.nonemb_req.sge;
2903 
2904 	memset(&cmd, 0, sizeof(struct ocrdma_mqe));
2905 	cmd.hdr.pyld_len = max_t (u32, sizeof(struct ocrdma_get_dcbx_cfg_rsp),
2906 					sizeof(struct ocrdma_get_dcbx_cfg_req));
2907 	req = dma_alloc_coherent(&pdev->dev, cmd.hdr.pyld_len, &pa, GFP_KERNEL);
2908 	if (!req) {
2909 		status = -ENOMEM;
2910 		goto mem_err;
2911 	}
2912 
2913 	cmd.hdr.spcl_sge_cnt_emb |= (1 << OCRDMA_MQE_HDR_SGE_CNT_SHIFT) &
2914 					OCRDMA_MQE_HDR_SGE_CNT_MASK;
2915 	mqe_sge->pa_lo = (u32) (pa & 0xFFFFFFFFUL);
2916 	mqe_sge->pa_hi = (u32) upper_32_bits(pa);
2917 	mqe_sge->len = cmd.hdr.pyld_len;
2918 
2919 	memset(req, 0, sizeof(struct ocrdma_get_dcbx_cfg_req));
2920 	ocrdma_init_mch(&req->hdr, OCRDMA_CMD_GET_DCBX_CONFIG,
2921 			OCRDMA_SUBSYS_DCBX, cmd.hdr.pyld_len);
2922 	req->param_type = ptype;
2923 
2924 	status = ocrdma_mbx_cmd(dev, &cmd);
2925 	if (status)
2926 		goto mbx_err;
2927 
2928 	rsp = (struct ocrdma_get_dcbx_cfg_rsp *)req;
2929 	ocrdma_le32_to_cpu(rsp, sizeof(struct ocrdma_get_dcbx_cfg_rsp));
2930 	memcpy(dcbxcfg, &rsp->cfg, sizeof(struct ocrdma_dcbx_cfg));
2931 
2932 mbx_err:
2933 	dma_free_coherent(&pdev->dev, cmd.hdr.pyld_len, req, pa);
2934 mem_err:
2935 	return status;
2936 }
2937 
2938 #define OCRDMA_MAX_SERVICE_LEVEL_INDEX	0x08
2939 #define OCRDMA_DEFAULT_SERVICE_LEVEL	0x05
2940 
2941 static int ocrdma_parse_dcbxcfg_rsp(struct ocrdma_dev *dev, int ptype,
2942 				    struct ocrdma_dcbx_cfg *dcbxcfg,
2943 				    u8 *srvc_lvl)
2944 {
2945 	int status = -EINVAL, indx, slindx;
2946 	int ventry_cnt;
2947 	struct ocrdma_app_parameter *app_param;
2948 	u8 valid, proto_sel;
2949 	u8 app_prio, pfc_prio;
2950 	u16 proto;
2951 
2952 	if (!(dcbxcfg->tcv_aev_opv_st & OCRDMA_DCBX_STATE_MASK)) {
2953 		pr_info("%s ocrdma%d DCBX is disabled\n",
2954 			dev_name(&dev->nic_info.pdev->dev), dev->id);
2955 		goto out;
2956 	}
2957 
2958 	if (!ocrdma_is_enabled_and_synced(dcbxcfg->pfc_state)) {
2959 		pr_info("%s ocrdma%d priority flow control(%s) is %s%s\n",
2960 			dev_name(&dev->nic_info.pdev->dev), dev->id,
2961 			(ptype > 0 ? "operational" : "admin"),
2962 			(dcbxcfg->pfc_state & OCRDMA_STATE_FLAG_ENABLED) ?
2963 			"enabled" : "disabled",
2964 			(dcbxcfg->pfc_state & OCRDMA_STATE_FLAG_SYNC) ?
2965 			"" : ", not sync'ed");
2966 		goto out;
2967 	} else {
2968 		pr_info("%s ocrdma%d priority flow control is enabled and sync'ed\n",
2969 			dev_name(&dev->nic_info.pdev->dev), dev->id);
2970 	}
2971 
2972 	ventry_cnt = (dcbxcfg->tcv_aev_opv_st >>
2973 				OCRDMA_DCBX_APP_ENTRY_SHIFT)
2974 				& OCRDMA_DCBX_STATE_MASK;
2975 
2976 	for (indx = 0; indx < ventry_cnt; indx++) {
2977 		app_param = &dcbxcfg->app_param[indx];
2978 		valid = (app_param->valid_proto_app >>
2979 				OCRDMA_APP_PARAM_VALID_SHIFT)
2980 				& OCRDMA_APP_PARAM_VALID_MASK;
2981 		proto_sel = (app_param->valid_proto_app
2982 				>>  OCRDMA_APP_PARAM_PROTO_SEL_SHIFT)
2983 				& OCRDMA_APP_PARAM_PROTO_SEL_MASK;
2984 		proto = app_param->valid_proto_app &
2985 				OCRDMA_APP_PARAM_APP_PROTO_MASK;
2986 
2987 		if (
2988 			valid && proto == OCRDMA_APP_PROTO_ROCE &&
2989 			proto_sel == OCRDMA_PROTO_SELECT_L2) {
2990 			for (slindx = 0; slindx <
2991 				OCRDMA_MAX_SERVICE_LEVEL_INDEX; slindx++) {
2992 				app_prio = ocrdma_get_app_prio(
2993 						(u8 *)app_param->app_prio,
2994 						slindx);
2995 				pfc_prio = ocrdma_get_pfc_prio(
2996 						(u8 *)dcbxcfg->pfc_prio,
2997 						slindx);
2998 
2999 				if (app_prio && pfc_prio) {
3000 					*srvc_lvl = slindx;
3001 					status = 0;
3002 					goto out;
3003 				}
3004 			}
3005 			if (slindx == OCRDMA_MAX_SERVICE_LEVEL_INDEX) {
3006 				pr_info("%s ocrdma%d application priority not set for 0x%x protocol\n",
3007 					dev_name(&dev->nic_info.pdev->dev),
3008 					dev->id, proto);
3009 			}
3010 		}
3011 	}
3012 
3013 out:
3014 	return status;
3015 }
3016 
3017 void ocrdma_init_service_level(struct ocrdma_dev *dev)
3018 {
3019 	int status = 0, indx;
3020 	struct ocrdma_dcbx_cfg dcbxcfg;
3021 	u8 srvc_lvl = OCRDMA_DEFAULT_SERVICE_LEVEL;
3022 	int ptype = OCRDMA_PARAMETER_TYPE_OPER;
3023 
3024 	for (indx = 0; indx < 2; indx++) {
3025 		status = ocrdma_mbx_get_dcbx_config(dev, ptype, &dcbxcfg);
3026 		if (status) {
3027 			pr_err("%s(): status=%d\n", __func__, status);
3028 			ptype = OCRDMA_PARAMETER_TYPE_ADMIN;
3029 			continue;
3030 		}
3031 
3032 		status = ocrdma_parse_dcbxcfg_rsp(dev, ptype,
3033 						  &dcbxcfg, &srvc_lvl);
3034 		if (status) {
3035 			ptype = OCRDMA_PARAMETER_TYPE_ADMIN;
3036 			continue;
3037 		}
3038 
3039 		break;
3040 	}
3041 
3042 	if (status)
3043 		pr_info("%s ocrdma%d service level default\n",
3044 			dev_name(&dev->nic_info.pdev->dev), dev->id);
3045 	else
3046 		pr_info("%s ocrdma%d service level %d\n",
3047 			dev_name(&dev->nic_info.pdev->dev), dev->id,
3048 			srvc_lvl);
3049 
3050 	dev->pfc_state = ocrdma_is_enabled_and_synced(dcbxcfg.pfc_state);
3051 	dev->sl = srvc_lvl;
3052 }
3053 
3054 int ocrdma_alloc_av(struct ocrdma_dev *dev, struct ocrdma_ah *ah)
3055 {
3056 	int i;
3057 	int status = -EINVAL;
3058 	struct ocrdma_av *av;
3059 	unsigned long flags;
3060 
3061 	av = dev->av_tbl.va;
3062 	spin_lock_irqsave(&dev->av_tbl.lock, flags);
3063 	for (i = 0; i < dev->av_tbl.num_ah; i++) {
3064 		if (av->valid == 0) {
3065 			av->valid = OCRDMA_AV_VALID;
3066 			ah->av = av;
3067 			ah->id = i;
3068 			status = 0;
3069 			break;
3070 		}
3071 		av++;
3072 	}
3073 	if (i == dev->av_tbl.num_ah)
3074 		status = -EAGAIN;
3075 	spin_unlock_irqrestore(&dev->av_tbl.lock, flags);
3076 	return status;
3077 }
3078 
3079 int ocrdma_free_av(struct ocrdma_dev *dev, struct ocrdma_ah *ah)
3080 {
3081 	unsigned long flags;
3082 	spin_lock_irqsave(&dev->av_tbl.lock, flags);
3083 	ah->av->valid = 0;
3084 	spin_unlock_irqrestore(&dev->av_tbl.lock, flags);
3085 	return 0;
3086 }
3087 
3088 static int ocrdma_create_eqs(struct ocrdma_dev *dev)
3089 {
3090 	int num_eq, i, status = 0;
3091 	int irq;
3092 	unsigned long flags = 0;
3093 
3094 	num_eq = dev->nic_info.msix.num_vectors -
3095 			dev->nic_info.msix.start_vector;
3096 	if (dev->nic_info.intr_mode == BE_INTERRUPT_MODE_INTX) {
3097 		num_eq = 1;
3098 		flags = IRQF_SHARED;
3099 	} else {
3100 		num_eq = min_t(u32, num_eq, num_online_cpus());
3101 	}
3102 
3103 	if (!num_eq)
3104 		return -EINVAL;
3105 
3106 	dev->eq_tbl = kzalloc(sizeof(struct ocrdma_eq) * num_eq, GFP_KERNEL);
3107 	if (!dev->eq_tbl)
3108 		return -ENOMEM;
3109 
3110 	for (i = 0; i < num_eq; i++) {
3111 		status = ocrdma_create_eq(dev, &dev->eq_tbl[i],
3112 					OCRDMA_EQ_LEN);
3113 		if (status) {
3114 			status = -EINVAL;
3115 			break;
3116 		}
3117 		sprintf(dev->eq_tbl[i].irq_name, "ocrdma%d-%d",
3118 			dev->id, i);
3119 		irq = ocrdma_get_irq(dev, &dev->eq_tbl[i]);
3120 		status = request_irq(irq, ocrdma_irq_handler, flags,
3121 				     dev->eq_tbl[i].irq_name,
3122 				     &dev->eq_tbl[i]);
3123 		if (status)
3124 			goto done;
3125 		dev->eq_cnt += 1;
3126 	}
3127 	/* one eq is sufficient for data path to work */
3128 	return 0;
3129 done:
3130 	ocrdma_destroy_eqs(dev);
3131 	return status;
3132 }
3133 
3134 static int ocrdma_mbx_modify_eqd(struct ocrdma_dev *dev, struct ocrdma_eq *eq,
3135 				 int num)
3136 {
3137 	int i, status = -ENOMEM;
3138 	struct ocrdma_modify_eqd_req *cmd;
3139 
3140 	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_MODIFY_EQ_DELAY, sizeof(*cmd));
3141 	if (!cmd)
3142 		return status;
3143 
3144 	ocrdma_init_mch(&cmd->cmd.req, OCRDMA_CMD_MODIFY_EQ_DELAY,
3145 			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
3146 
3147 	cmd->cmd.num_eq = num;
3148 	for (i = 0; i < num; i++) {
3149 		cmd->cmd.set_eqd[i].eq_id = eq[i].q.id;
3150 		cmd->cmd.set_eqd[i].phase = 0;
3151 		cmd->cmd.set_eqd[i].delay_multiplier =
3152 				(eq[i].aic_obj.prev_eqd * 65)/100;
3153 	}
3154 	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
3155 	if (status)
3156 		goto mbx_err;
3157 mbx_err:
3158 	kfree(cmd);
3159 	return status;
3160 }
3161 
3162 static int ocrdma_modify_eqd(struct ocrdma_dev *dev, struct ocrdma_eq *eq,
3163 			     int num)
3164 {
3165 	int num_eqs, i = 0;
3166 	if (num > 8) {
3167 		while (num) {
3168 			num_eqs = min(num, 8);
3169 			ocrdma_mbx_modify_eqd(dev, &eq[i], num_eqs);
3170 			i += num_eqs;
3171 			num -= num_eqs;
3172 		}
3173 	} else {
3174 		ocrdma_mbx_modify_eqd(dev, eq, num);
3175 	}
3176 	return 0;
3177 }
3178 
3179 void ocrdma_eqd_set_task(struct work_struct *work)
3180 {
3181 	struct ocrdma_dev *dev =
3182 		container_of(work, struct ocrdma_dev, eqd_work.work);
3183 	struct ocrdma_eq *eq = 0;
3184 	int i, num = 0, status = -EINVAL;
3185 	u64 eq_intr;
3186 
3187 	for (i = 0; i < dev->eq_cnt; i++) {
3188 		eq = &dev->eq_tbl[i];
3189 		if (eq->aic_obj.eq_intr_cnt > eq->aic_obj.prev_eq_intr_cnt) {
3190 			eq_intr = eq->aic_obj.eq_intr_cnt -
3191 				  eq->aic_obj.prev_eq_intr_cnt;
3192 			if ((eq_intr > EQ_INTR_PER_SEC_THRSH_HI) &&
3193 			    (eq->aic_obj.prev_eqd == EQ_AIC_MIN_EQD)) {
3194 				eq->aic_obj.prev_eqd = EQ_AIC_MAX_EQD;
3195 				num++;
3196 			} else if ((eq_intr < EQ_INTR_PER_SEC_THRSH_LOW) &&
3197 				   (eq->aic_obj.prev_eqd == EQ_AIC_MAX_EQD)) {
3198 				eq->aic_obj.prev_eqd = EQ_AIC_MIN_EQD;
3199 				num++;
3200 			}
3201 		}
3202 		eq->aic_obj.prev_eq_intr_cnt = eq->aic_obj.eq_intr_cnt;
3203 	}
3204 
3205 	if (num)
3206 		status = ocrdma_modify_eqd(dev, &dev->eq_tbl[0], num);
3207 	schedule_delayed_work(&dev->eqd_work, msecs_to_jiffies(1000));
3208 }
3209 
3210 int ocrdma_init_hw(struct ocrdma_dev *dev)
3211 {
3212 	int status;
3213 
3214 	/* create the eqs  */
3215 	status = ocrdma_create_eqs(dev);
3216 	if (status)
3217 		goto qpeq_err;
3218 	status = ocrdma_create_mq(dev);
3219 	if (status)
3220 		goto mq_err;
3221 	status = ocrdma_mbx_query_fw_config(dev);
3222 	if (status)
3223 		goto conf_err;
3224 	status = ocrdma_mbx_query_dev(dev);
3225 	if (status)
3226 		goto conf_err;
3227 	status = ocrdma_mbx_query_fw_ver(dev);
3228 	if (status)
3229 		goto conf_err;
3230 	status = ocrdma_mbx_create_ah_tbl(dev);
3231 	if (status)
3232 		goto conf_err;
3233 	status = ocrdma_mbx_get_phy_info(dev);
3234 	if (status)
3235 		goto info_attrb_err;
3236 	status = ocrdma_mbx_get_ctrl_attribs(dev);
3237 	if (status)
3238 		goto info_attrb_err;
3239 
3240 	return 0;
3241 
3242 info_attrb_err:
3243 	ocrdma_mbx_delete_ah_tbl(dev);
3244 conf_err:
3245 	ocrdma_destroy_mq(dev);
3246 mq_err:
3247 	ocrdma_destroy_eqs(dev);
3248 qpeq_err:
3249 	pr_err("%s() status=%d\n", __func__, status);
3250 	return status;
3251 }
3252 
3253 void ocrdma_cleanup_hw(struct ocrdma_dev *dev)
3254 {
3255 	ocrdma_free_pd_pool(dev);
3256 	ocrdma_mbx_delete_ah_tbl(dev);
3257 
3258 	/* cleanup the control path */
3259 	ocrdma_destroy_mq(dev);
3260 
3261 	/* cleanup the eqs */
3262 	ocrdma_destroy_eqs(dev);
3263 }
3264