xref: /linux/drivers/infiniband/hw/bnxt_re/qplib_fp.c (revision f9aa1fb9f8c0542f5f6e6e620de320995d5622ad)
1 /*
2  * Broadcom NetXtreme-E RoCE driver.
3  *
4  * Copyright (c) 2016 - 2017, Broadcom. All rights reserved.  The term
5  * Broadcom refers to Broadcom Limited and/or its subsidiaries.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * BSD license below:
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  *
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. 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
22  *    distribution.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS''
25  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
26  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS
28  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
32  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
33  * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
34  * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Description: Fast Path Operators
37  */
38 
39 #define dev_fmt(fmt) "QPLIB: " fmt
40 
41 #include <linux/interrupt.h>
42 #include <linux/spinlock.h>
43 #include <linux/sched.h>
44 #include <linux/slab.h>
45 #include <linux/pci.h>
46 #include <linux/delay.h>
47 #include <linux/prefetch.h>
48 #include <linux/if_ether.h>
49 #include <rdma/ib_mad.h>
50 
51 #include "roce_hsi.h"
52 
53 #include "qplib_res.h"
54 #include "qplib_rcfw.h"
55 #include "qplib_sp.h"
56 #include "qplib_fp.h"
57 #include <rdma/ib_addr.h>
58 #include "bnxt_ulp.h"
59 #include "bnxt_re.h"
60 #include "ib_verbs.h"
61 
62 static void __clean_cq(struct bnxt_qplib_cq *cq, u64 qp);
63 
64 static void bnxt_qplib_cancel_phantom_processing(struct bnxt_qplib_qp *qp)
65 {
66 	qp->sq.condition = false;
67 	qp->sq.send_phantom = false;
68 	qp->sq.single = false;
69 }
70 
71 /* Flush list */
72 static void __bnxt_qplib_add_flush_qp(struct bnxt_qplib_qp *qp)
73 {
74 	struct bnxt_qplib_cq *scq, *rcq;
75 
76 	scq = qp->scq;
77 	rcq = qp->rcq;
78 
79 	if (!qp->sq.flushed) {
80 		dev_dbg(&scq->hwq.pdev->dev,
81 			"FP: Adding to SQ Flush list = %p\n", qp);
82 		bnxt_qplib_cancel_phantom_processing(qp);
83 		list_add_tail(&qp->sq_flush, &scq->sqf_head);
84 		qp->sq.flushed = true;
85 	}
86 	if (!qp->srq) {
87 		if (!qp->rq.flushed) {
88 			dev_dbg(&rcq->hwq.pdev->dev,
89 				"FP: Adding to RQ Flush list = %p\n", qp);
90 			list_add_tail(&qp->rq_flush, &rcq->rqf_head);
91 			qp->rq.flushed = true;
92 		}
93 	}
94 }
95 
96 static void bnxt_qplib_acquire_cq_flush_locks(struct bnxt_qplib_qp *qp,
97 				       unsigned long *flags)
98 	__acquires(&qp->scq->flush_lock) __acquires(&qp->rcq->flush_lock)
99 {
100 	spin_lock_irqsave(&qp->scq->flush_lock, *flags);
101 	if (qp->scq == qp->rcq)
102 		__acquire(&qp->rcq->flush_lock);
103 	else
104 		spin_lock(&qp->rcq->flush_lock);
105 }
106 
107 static void bnxt_qplib_release_cq_flush_locks(struct bnxt_qplib_qp *qp,
108 				       unsigned long *flags)
109 	__releases(&qp->scq->flush_lock) __releases(&qp->rcq->flush_lock)
110 {
111 	if (qp->scq == qp->rcq)
112 		__release(&qp->rcq->flush_lock);
113 	else
114 		spin_unlock(&qp->rcq->flush_lock);
115 	spin_unlock_irqrestore(&qp->scq->flush_lock, *flags);
116 }
117 
118 void bnxt_qplib_add_flush_qp(struct bnxt_qplib_qp *qp)
119 {
120 	unsigned long flags;
121 
122 	bnxt_qplib_acquire_cq_flush_locks(qp, &flags);
123 	__bnxt_qplib_add_flush_qp(qp);
124 	bnxt_qplib_release_cq_flush_locks(qp, &flags);
125 }
126 
127 static void __bnxt_qplib_del_flush_qp(struct bnxt_qplib_qp *qp)
128 {
129 	if (qp->sq.flushed) {
130 		qp->sq.flushed = false;
131 		list_del(&qp->sq_flush);
132 	}
133 	if (!qp->srq) {
134 		if (qp->rq.flushed) {
135 			qp->rq.flushed = false;
136 			list_del(&qp->rq_flush);
137 		}
138 	}
139 }
140 
141 void bnxt_qplib_clean_qp(struct bnxt_qplib_qp *qp)
142 {
143 	unsigned long flags;
144 
145 	bnxt_qplib_acquire_cq_flush_locks(qp, &flags);
146 	__clean_cq(qp->scq, (u64)(unsigned long)qp);
147 	qp->sq.hwq.prod = 0;
148 	qp->sq.hwq.cons = 0;
149 	__clean_cq(qp->rcq, (u64)(unsigned long)qp);
150 	qp->rq.hwq.prod = 0;
151 	qp->rq.hwq.cons = 0;
152 
153 	__bnxt_qplib_del_flush_qp(qp);
154 	bnxt_qplib_release_cq_flush_locks(qp, &flags);
155 }
156 
157 static void bnxt_qpn_cqn_sched_task(struct work_struct *work)
158 {
159 	struct bnxt_qplib_nq_work *nq_work =
160 			container_of(work, struct bnxt_qplib_nq_work, work);
161 
162 	struct bnxt_qplib_cq *cq = nq_work->cq;
163 	struct bnxt_qplib_nq *nq = nq_work->nq;
164 
165 	if (cq && nq) {
166 		spin_lock_bh(&cq->compl_lock);
167 		if (atomic_read(&cq->arm_state) && nq->cqn_handler) {
168 			dev_dbg(&nq->pdev->dev,
169 				"%s:Trigger cq  = %p event nq = %p\n",
170 				__func__, cq, nq);
171 			nq->cqn_handler(nq, cq);
172 		}
173 		spin_unlock_bh(&cq->compl_lock);
174 	}
175 	kfree(nq_work);
176 }
177 
178 static void bnxt_qplib_free_qp_hdr_buf(struct bnxt_qplib_res *res,
179 				       struct bnxt_qplib_qp *qp)
180 {
181 	struct bnxt_qplib_q *rq = &qp->rq;
182 	struct bnxt_qplib_q *sq = &qp->sq;
183 
184 	if (qp->rq_hdr_buf)
185 		dma_free_coherent(&res->pdev->dev,
186 				  rq->max_wqe * qp->rq_hdr_buf_size,
187 				  qp->rq_hdr_buf, qp->rq_hdr_buf_map);
188 	if (qp->sq_hdr_buf)
189 		dma_free_coherent(&res->pdev->dev,
190 				  sq->max_wqe * qp->sq_hdr_buf_size,
191 				  qp->sq_hdr_buf, qp->sq_hdr_buf_map);
192 	qp->rq_hdr_buf = NULL;
193 	qp->sq_hdr_buf = NULL;
194 	qp->rq_hdr_buf_map = 0;
195 	qp->sq_hdr_buf_map = 0;
196 	qp->sq_hdr_buf_size = 0;
197 	qp->rq_hdr_buf_size = 0;
198 }
199 
200 static int bnxt_qplib_alloc_qp_hdr_buf(struct bnxt_qplib_res *res,
201 				       struct bnxt_qplib_qp *qp)
202 {
203 	struct bnxt_qplib_q *rq = &qp->rq;
204 	struct bnxt_qplib_q *sq = &qp->sq;
205 	int rc = 0;
206 
207 	if (qp->sq_hdr_buf_size && sq->max_wqe) {
208 		qp->sq_hdr_buf = dma_alloc_coherent(&res->pdev->dev,
209 					sq->max_wqe * qp->sq_hdr_buf_size,
210 					&qp->sq_hdr_buf_map, GFP_KERNEL);
211 		if (!qp->sq_hdr_buf) {
212 			rc = -ENOMEM;
213 			dev_err(&res->pdev->dev,
214 				"Failed to create sq_hdr_buf\n");
215 			goto fail;
216 		}
217 	}
218 
219 	if (qp->rq_hdr_buf_size && rq->max_wqe) {
220 		qp->rq_hdr_buf = dma_alloc_coherent(&res->pdev->dev,
221 						    rq->max_wqe *
222 						    qp->rq_hdr_buf_size,
223 						    &qp->rq_hdr_buf_map,
224 						    GFP_KERNEL);
225 		if (!qp->rq_hdr_buf) {
226 			rc = -ENOMEM;
227 			dev_err(&res->pdev->dev,
228 				"Failed to create rq_hdr_buf\n");
229 			goto fail;
230 		}
231 	}
232 	return 0;
233 
234 fail:
235 	bnxt_qplib_free_qp_hdr_buf(res, qp);
236 	return rc;
237 }
238 
239 static void clean_nq(struct bnxt_qplib_nq *nq, struct bnxt_qplib_cq *cq)
240 {
241 	struct bnxt_qplib_hwq *hwq = &nq->hwq;
242 	struct nq_base *nqe, **nq_ptr;
243 	int budget = nq->budget;
244 	uintptr_t q_handle;
245 	u16 type;
246 
247 	spin_lock_bh(&hwq->lock);
248 	/* Service the NQ until empty */
249 	while (budget--) {
250 		nq_ptr = (struct nq_base **)hwq->pbl_ptr;
251 		nqe = &nq_ptr[NQE_PG(hwq->cons)][NQE_IDX(hwq->cons)];
252 		if (!NQE_CMP_VALID(nqe, nq->nq_db.dbinfo.flags))
253 			break;
254 
255 		/*
256 		 * The valid test of the entry must be done first before
257 		 * reading any further.
258 		 */
259 		dma_rmb();
260 
261 		type = le16_to_cpu(nqe->info10_type) & NQ_BASE_TYPE_MASK;
262 		switch (type) {
263 		case NQ_BASE_TYPE_CQ_NOTIFICATION:
264 		{
265 			struct nq_cn *nqcne = (struct nq_cn *)nqe;
266 
267 			q_handle = le32_to_cpu(nqcne->cq_handle_low);
268 			q_handle |= (u64)le32_to_cpu(nqcne->cq_handle_high)
269 						     << 32;
270 			if ((unsigned long)cq == q_handle) {
271 				nqcne->cq_handle_low = 0;
272 				nqcne->cq_handle_high = 0;
273 				cq->cnq_events++;
274 			}
275 			break;
276 		}
277 		default:
278 			break;
279 		}
280 		bnxt_qplib_hwq_incr_cons(hwq->max_elements, &hwq->cons,
281 					 1, &nq->nq_db.dbinfo.flags);
282 	}
283 	spin_unlock_bh(&hwq->lock);
284 }
285 
286 /* Wait for receiving all NQEs for this CQ and clean the NQEs associated with
287  * this CQ.
288  */
289 static void __wait_for_all_nqes(struct bnxt_qplib_cq *cq, u16 cnq_events)
290 {
291 	u32 retry_cnt = 100;
292 
293 	while (retry_cnt--) {
294 		if (cnq_events == cq->cnq_events)
295 			return;
296 		usleep_range(50, 100);
297 		clean_nq(cq->nq, cq);
298 	}
299 }
300 
301 static void bnxt_qplib_service_nq(struct tasklet_struct *t)
302 {
303 	struct bnxt_qplib_nq *nq = from_tasklet(nq, t, nq_tasklet);
304 	struct bnxt_qplib_hwq *hwq = &nq->hwq;
305 	struct bnxt_qplib_cq *cq;
306 	int budget = nq->budget;
307 	struct nq_base *nqe;
308 	uintptr_t q_handle;
309 	u32 hw_polled = 0;
310 	u16 type;
311 
312 	spin_lock_bh(&hwq->lock);
313 	/* Service the NQ until empty */
314 	while (budget--) {
315 		nqe = bnxt_qplib_get_qe(hwq, hwq->cons, NULL);
316 		if (!NQE_CMP_VALID(nqe, nq->nq_db.dbinfo.flags))
317 			break;
318 
319 		/*
320 		 * The valid test of the entry must be done first before
321 		 * reading any further.
322 		 */
323 		dma_rmb();
324 
325 		type = le16_to_cpu(nqe->info10_type) & NQ_BASE_TYPE_MASK;
326 		switch (type) {
327 		case NQ_BASE_TYPE_CQ_NOTIFICATION:
328 		{
329 			struct nq_cn *nqcne = (struct nq_cn *)nqe;
330 			struct bnxt_re_cq *cq_p;
331 
332 			q_handle = le32_to_cpu(nqcne->cq_handle_low);
333 			q_handle |= (u64)le32_to_cpu(nqcne->cq_handle_high)
334 						     << 32;
335 			cq = (struct bnxt_qplib_cq *)(unsigned long)q_handle;
336 			if (!cq)
337 				break;
338 			cq->toggle = (le16_to_cpu(nqe->info10_type) &
339 					NQ_CN_TOGGLE_MASK) >> NQ_CN_TOGGLE_SFT;
340 			cq->dbinfo.toggle = cq->toggle;
341 			cq_p = container_of(cq, struct bnxt_re_cq, qplib_cq);
342 			if (cq_p->uctx_cq_page)
343 				*((u32 *)cq_p->uctx_cq_page) = cq->toggle;
344 
345 			bnxt_qplib_armen_db(&cq->dbinfo,
346 					    DBC_DBC_TYPE_CQ_ARMENA);
347 			spin_lock_bh(&cq->compl_lock);
348 			atomic_set(&cq->arm_state, 0);
349 			if (nq->cqn_handler(nq, (cq)))
350 				dev_warn(&nq->pdev->dev,
351 					 "cqn - type 0x%x not handled\n", type);
352 			cq->cnq_events++;
353 			spin_unlock_bh(&cq->compl_lock);
354 			break;
355 		}
356 		case NQ_BASE_TYPE_SRQ_EVENT:
357 		{
358 			struct bnxt_qplib_srq *srq;
359 			struct bnxt_re_srq *srq_p;
360 			struct nq_srq_event *nqsrqe =
361 						(struct nq_srq_event *)nqe;
362 
363 			q_handle = le32_to_cpu(nqsrqe->srq_handle_low);
364 			q_handle |= (u64)le32_to_cpu(nqsrqe->srq_handle_high)
365 				     << 32;
366 			srq = (struct bnxt_qplib_srq *)q_handle;
367 			srq->toggle = (le16_to_cpu(nqe->info10_type) & NQ_CN_TOGGLE_MASK)
368 				      >> NQ_CN_TOGGLE_SFT;
369 			srq->dbinfo.toggle = srq->toggle;
370 			srq_p = container_of(srq, struct bnxt_re_srq, qplib_srq);
371 			if (srq_p->uctx_srq_page)
372 				*((u32 *)srq_p->uctx_srq_page) = srq->toggle;
373 			bnxt_qplib_armen_db(&srq->dbinfo,
374 					    DBC_DBC_TYPE_SRQ_ARMENA);
375 			if (nq->srqn_handler(nq,
376 					     (struct bnxt_qplib_srq *)q_handle,
377 					     nqsrqe->event))
378 				dev_warn(&nq->pdev->dev,
379 					 "SRQ event 0x%x not handled\n",
380 					 nqsrqe->event);
381 			break;
382 		}
383 		case NQ_BASE_TYPE_DBQ_EVENT:
384 			break;
385 		default:
386 			dev_warn(&nq->pdev->dev,
387 				 "nqe with type = 0x%x not handled\n", type);
388 			break;
389 		}
390 		hw_polled++;
391 		bnxt_qplib_hwq_incr_cons(hwq->max_elements, &hwq->cons,
392 					 1, &nq->nq_db.dbinfo.flags);
393 	}
394 	if (hw_polled)
395 		bnxt_qplib_ring_nq_db(&nq->nq_db.dbinfo, nq->res->cctx, true);
396 	spin_unlock_bh(&hwq->lock);
397 }
398 
399 /* bnxt_re_synchronize_nq - self polling notification queue.
400  * @nq      -     notification queue pointer
401  *
402  * This function will start polling entries of a given notification queue
403  * for all pending  entries.
404  * This function is useful to synchronize notification entries while resources
405  * are going away.
406  */
407 
408 void bnxt_re_synchronize_nq(struct bnxt_qplib_nq *nq)
409 {
410 	int budget = nq->budget;
411 
412 	nq->budget = nq->hwq.max_elements;
413 	bnxt_qplib_service_nq(&nq->nq_tasklet);
414 	nq->budget = budget;
415 }
416 
417 static irqreturn_t bnxt_qplib_nq_irq(int irq, void *dev_instance)
418 {
419 	struct bnxt_qplib_nq *nq = dev_instance;
420 	struct bnxt_qplib_hwq *hwq = &nq->hwq;
421 	u32 sw_cons;
422 
423 	/* Prefetch the NQ element */
424 	sw_cons = HWQ_CMP(hwq->cons, hwq);
425 	prefetch(bnxt_qplib_get_qe(hwq, sw_cons, NULL));
426 
427 	/* Fan out to CPU affinitized kthreads? */
428 	tasklet_schedule(&nq->nq_tasklet);
429 
430 	return IRQ_HANDLED;
431 }
432 
433 void bnxt_qplib_nq_stop_irq(struct bnxt_qplib_nq *nq, bool kill)
434 {
435 	if (!nq->requested)
436 		return;
437 
438 	nq->requested = false;
439 	/* Mask h/w interrupt */
440 	bnxt_qplib_ring_nq_db(&nq->nq_db.dbinfo, nq->res->cctx, false);
441 	/* Sync with last running IRQ handler */
442 	synchronize_irq(nq->msix_vec);
443 	irq_set_affinity_hint(nq->msix_vec, NULL);
444 	free_irq(nq->msix_vec, nq);
445 	kfree(nq->name);
446 	nq->name = NULL;
447 
448 	if (kill)
449 		tasklet_kill(&nq->nq_tasklet);
450 	tasklet_disable(&nq->nq_tasklet);
451 }
452 
453 void bnxt_qplib_disable_nq(struct bnxt_qplib_nq *nq)
454 {
455 	if (nq->cqn_wq) {
456 		destroy_workqueue(nq->cqn_wq);
457 		nq->cqn_wq = NULL;
458 	}
459 
460 	/* Make sure the HW is stopped! */
461 	bnxt_qplib_nq_stop_irq(nq, true);
462 
463 	if (nq->nq_db.reg.bar_reg) {
464 		iounmap(nq->nq_db.reg.bar_reg);
465 		nq->nq_db.reg.bar_reg = NULL;
466 	}
467 
468 	nq->cqn_handler = NULL;
469 	nq->srqn_handler = NULL;
470 	nq->msix_vec = 0;
471 }
472 
473 int bnxt_qplib_nq_start_irq(struct bnxt_qplib_nq *nq, int nq_indx,
474 			    int msix_vector, bool need_init)
475 {
476 	struct bnxt_qplib_res *res = nq->res;
477 	int rc;
478 
479 	if (nq->requested)
480 		return -EFAULT;
481 
482 	nq->msix_vec = msix_vector;
483 	if (need_init)
484 		tasklet_setup(&nq->nq_tasklet, bnxt_qplib_service_nq);
485 	else
486 		tasklet_enable(&nq->nq_tasklet);
487 
488 	nq->name = kasprintf(GFP_KERNEL, "bnxt_re-nq-%d@pci:%s",
489 			     nq_indx, pci_name(res->pdev));
490 	if (!nq->name)
491 		return -ENOMEM;
492 	rc = request_irq(nq->msix_vec, bnxt_qplib_nq_irq, 0, nq->name, nq);
493 	if (rc) {
494 		kfree(nq->name);
495 		nq->name = NULL;
496 		tasklet_disable(&nq->nq_tasklet);
497 		return rc;
498 	}
499 
500 	cpumask_clear(&nq->mask);
501 	cpumask_set_cpu(nq_indx, &nq->mask);
502 	rc = irq_set_affinity_hint(nq->msix_vec, &nq->mask);
503 	if (rc) {
504 		dev_warn(&nq->pdev->dev,
505 			 "set affinity failed; vector: %d nq_idx: %d\n",
506 			 nq->msix_vec, nq_indx);
507 	}
508 	nq->requested = true;
509 	bnxt_qplib_ring_nq_db(&nq->nq_db.dbinfo, res->cctx, true);
510 
511 	return rc;
512 }
513 
514 static int bnxt_qplib_map_nq_db(struct bnxt_qplib_nq *nq,  u32 reg_offt)
515 {
516 	resource_size_t reg_base;
517 	struct bnxt_qplib_nq_db *nq_db;
518 	struct pci_dev *pdev;
519 
520 	pdev = nq->pdev;
521 	nq_db = &nq->nq_db;
522 
523 	nq_db->dbinfo.flags = 0;
524 	nq_db->reg.bar_id = NQ_CONS_PCI_BAR_REGION;
525 	nq_db->reg.bar_base = pci_resource_start(pdev, nq_db->reg.bar_id);
526 	if (!nq_db->reg.bar_base) {
527 		dev_err(&pdev->dev, "QPLIB: NQ BAR region %d resc start is 0!",
528 			nq_db->reg.bar_id);
529 		return -ENOMEM;
530 	}
531 
532 	reg_base = nq_db->reg.bar_base + reg_offt;
533 	/* Unconditionally map 8 bytes to support 57500 series */
534 	nq_db->reg.len = 8;
535 	nq_db->reg.bar_reg = ioremap(reg_base, nq_db->reg.len);
536 	if (!nq_db->reg.bar_reg) {
537 		dev_err(&pdev->dev, "QPLIB: NQ BAR region %d mapping failed",
538 			nq_db->reg.bar_id);
539 		return -ENOMEM;
540 	}
541 
542 	nq_db->dbinfo.db = nq_db->reg.bar_reg;
543 	nq_db->dbinfo.hwq = &nq->hwq;
544 	nq_db->dbinfo.xid = nq->ring_id;
545 
546 	return 0;
547 }
548 
549 int bnxt_qplib_enable_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq,
550 			 int nq_idx, int msix_vector, int bar_reg_offset,
551 			 cqn_handler_t cqn_handler,
552 			 srqn_handler_t srqn_handler)
553 {
554 	int rc;
555 
556 	nq->pdev = pdev;
557 	nq->cqn_handler = cqn_handler;
558 	nq->srqn_handler = srqn_handler;
559 	nq->load = 0;
560 
561 	/* Have a task to schedule CQ notifiers in post send case */
562 	nq->cqn_wq  = create_singlethread_workqueue("bnxt_qplib_nq");
563 	if (!nq->cqn_wq)
564 		return -ENOMEM;
565 
566 	rc = bnxt_qplib_map_nq_db(nq, bar_reg_offset);
567 	if (rc)
568 		goto fail;
569 
570 	rc = bnxt_qplib_nq_start_irq(nq, nq_idx, msix_vector, true);
571 	if (rc) {
572 		dev_err(&nq->pdev->dev,
573 			"Failed to request irq for nq-idx %d\n", nq_idx);
574 		goto fail;
575 	}
576 
577 	return 0;
578 fail:
579 	bnxt_qplib_disable_nq(nq);
580 	return rc;
581 }
582 
583 void bnxt_qplib_free_nq(struct bnxt_qplib_nq *nq)
584 {
585 	if (nq->hwq.max_elements) {
586 		bnxt_qplib_free_hwq(nq->res, &nq->hwq);
587 		nq->hwq.max_elements = 0;
588 	}
589 }
590 
591 int bnxt_qplib_alloc_nq(struct bnxt_qplib_res *res, struct bnxt_qplib_nq *nq)
592 {
593 	struct bnxt_qplib_hwq_attr hwq_attr = {};
594 	struct bnxt_qplib_sg_info sginfo = {};
595 
596 	nq->pdev = res->pdev;
597 	nq->res = res;
598 	if (!nq->hwq.max_elements ||
599 	    nq->hwq.max_elements > BNXT_QPLIB_NQE_MAX_CNT)
600 		nq->hwq.max_elements = BNXT_QPLIB_NQE_MAX_CNT;
601 
602 	sginfo.pgsize = PAGE_SIZE;
603 	sginfo.pgshft = PAGE_SHIFT;
604 	hwq_attr.res = res;
605 	hwq_attr.sginfo = &sginfo;
606 	hwq_attr.depth = nq->hwq.max_elements;
607 	hwq_attr.stride = sizeof(struct nq_base);
608 	hwq_attr.type = bnxt_qplib_get_hwq_type(nq->res);
609 	if (bnxt_qplib_alloc_init_hwq(&nq->hwq, &hwq_attr)) {
610 		dev_err(&nq->pdev->dev, "FP NQ allocation failed");
611 		return -ENOMEM;
612 	}
613 	nq->budget = 8;
614 	return 0;
615 }
616 
617 /* SRQ */
618 void bnxt_qplib_destroy_srq(struct bnxt_qplib_res *res,
619 			   struct bnxt_qplib_srq *srq)
620 {
621 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
622 	struct creq_destroy_srq_resp resp = {};
623 	struct bnxt_qplib_cmdqmsg msg = {};
624 	struct cmdq_destroy_srq req = {};
625 	int rc;
626 
627 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
628 				 CMDQ_BASE_OPCODE_DESTROY_SRQ,
629 				 sizeof(req));
630 
631 	/* Configure the request */
632 	req.srq_cid = cpu_to_le32(srq->id);
633 
634 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req), sizeof(resp), 0);
635 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
636 	kfree(srq->swq);
637 	if (rc)
638 		return;
639 	bnxt_qplib_free_hwq(res, &srq->hwq);
640 }
641 
642 int bnxt_qplib_create_srq(struct bnxt_qplib_res *res,
643 			  struct bnxt_qplib_srq *srq)
644 {
645 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
646 	struct bnxt_qplib_hwq_attr hwq_attr = {};
647 	struct creq_create_srq_resp resp = {};
648 	struct bnxt_qplib_cmdqmsg msg = {};
649 	struct cmdq_create_srq req = {};
650 	struct bnxt_qplib_pbl *pbl;
651 	u16 pg_sz_lvl;
652 	int rc, idx;
653 
654 	hwq_attr.res = res;
655 	hwq_attr.sginfo = &srq->sg_info;
656 	hwq_attr.depth = srq->max_wqe;
657 	hwq_attr.stride = srq->wqe_size;
658 	hwq_attr.type = HWQ_TYPE_QUEUE;
659 	rc = bnxt_qplib_alloc_init_hwq(&srq->hwq, &hwq_attr);
660 	if (rc)
661 		return rc;
662 	srq->dbinfo.flags = 0;
663 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
664 				 CMDQ_BASE_OPCODE_CREATE_SRQ,
665 				 sizeof(req));
666 
667 	/* Configure the request */
668 	req.dpi = cpu_to_le32(srq->dpi->dpi);
669 	req.srq_handle = cpu_to_le64((uintptr_t)srq);
670 
671 	req.srq_size = cpu_to_le16((u16)srq->hwq.max_elements);
672 	pbl = &srq->hwq.pbl[PBL_LVL_0];
673 	pg_sz_lvl = ((u16)bnxt_qplib_base_pg_size(&srq->hwq) <<
674 		     CMDQ_CREATE_SRQ_PG_SIZE_SFT);
675 	pg_sz_lvl |= (srq->hwq.level & CMDQ_CREATE_SRQ_LVL_MASK) <<
676 		      CMDQ_CREATE_SRQ_LVL_SFT;
677 	req.pg_size_lvl = cpu_to_le16(pg_sz_lvl);
678 	req.pbl = cpu_to_le64(pbl->pg_map_arr[0]);
679 	req.pd_id = cpu_to_le32(srq->pd->id);
680 	req.eventq_id = cpu_to_le16(srq->eventq_hw_ring_id);
681 
682 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req), sizeof(resp), 0);
683 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
684 	if (rc)
685 		goto fail;
686 
687 	spin_lock_init(&srq->lock);
688 	srq->start_idx = 0;
689 	srq->last_idx = srq->hwq.max_elements - 1;
690 	if (!srq->hwq.is_user) {
691 		srq->swq = kcalloc(srq->hwq.max_elements, sizeof(*srq->swq),
692 				   GFP_KERNEL);
693 		if (!srq->swq) {
694 			rc = -ENOMEM;
695 			goto fail;
696 		}
697 		for (idx = 0; idx < srq->hwq.max_elements; idx++)
698 			srq->swq[idx].next_idx = idx + 1;
699 		srq->swq[srq->last_idx].next_idx = -1;
700 	}
701 
702 	srq->id = le32_to_cpu(resp.xid);
703 	srq->dbinfo.hwq = &srq->hwq;
704 	srq->dbinfo.xid = srq->id;
705 	srq->dbinfo.db = srq->dpi->dbr;
706 	srq->dbinfo.max_slot = 1;
707 	srq->dbinfo.priv_db = res->dpi_tbl.priv_db;
708 	if (srq->threshold)
709 		bnxt_qplib_armen_db(&srq->dbinfo, DBC_DBC_TYPE_SRQ_ARMENA);
710 	srq->arm_req = false;
711 
712 	return 0;
713 fail:
714 	bnxt_qplib_free_hwq(res, &srq->hwq);
715 	kfree(srq->swq);
716 
717 	return rc;
718 }
719 
720 int bnxt_qplib_modify_srq(struct bnxt_qplib_res *res,
721 			  struct bnxt_qplib_srq *srq)
722 {
723 	struct bnxt_qplib_hwq *srq_hwq = &srq->hwq;
724 	u32 count;
725 
726 	count = __bnxt_qplib_get_avail(srq_hwq);
727 	if (count > srq->threshold) {
728 		srq->arm_req = false;
729 		bnxt_qplib_srq_arm_db(&srq->dbinfo, srq->threshold);
730 	} else {
731 		/* Deferred arming */
732 		srq->arm_req = true;
733 	}
734 
735 	return 0;
736 }
737 
738 int bnxt_qplib_query_srq(struct bnxt_qplib_res *res,
739 			 struct bnxt_qplib_srq *srq)
740 {
741 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
742 	struct creq_query_srq_resp resp = {};
743 	struct bnxt_qplib_cmdqmsg msg = {};
744 	struct bnxt_qplib_rcfw_sbuf sbuf;
745 	struct creq_query_srq_resp_sb *sb;
746 	struct cmdq_query_srq req = {};
747 	int rc;
748 
749 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
750 				 CMDQ_BASE_OPCODE_QUERY_SRQ,
751 				 sizeof(req));
752 
753 	/* Configure the request */
754 	sbuf.size = ALIGN(sizeof(*sb), BNXT_QPLIB_CMDQE_UNITS);
755 	sbuf.sb = dma_alloc_coherent(&rcfw->pdev->dev, sbuf.size,
756 				     &sbuf.dma_addr, GFP_KERNEL);
757 	if (!sbuf.sb)
758 		return -ENOMEM;
759 	req.resp_size = sbuf.size / BNXT_QPLIB_CMDQE_UNITS;
760 	req.srq_cid = cpu_to_le32(srq->id);
761 	sb = sbuf.sb;
762 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, &sbuf, sizeof(req),
763 				sizeof(resp), 0);
764 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
765 	if (!rc)
766 		srq->threshold = le16_to_cpu(sb->srq_limit);
767 	dma_free_coherent(&rcfw->pdev->dev, sbuf.size,
768 			  sbuf.sb, sbuf.dma_addr);
769 
770 	return rc;
771 }
772 
773 int bnxt_qplib_post_srq_recv(struct bnxt_qplib_srq *srq,
774 			     struct bnxt_qplib_swqe *wqe)
775 {
776 	struct bnxt_qplib_hwq *srq_hwq = &srq->hwq;
777 	struct rq_wqe *srqe;
778 	struct sq_sge *hw_sge;
779 	u32 count = 0;
780 	int i, next;
781 
782 	spin_lock(&srq_hwq->lock);
783 	if (srq->start_idx == srq->last_idx) {
784 		dev_err(&srq_hwq->pdev->dev,
785 			"FP: SRQ (0x%x) is full!\n", srq->id);
786 		spin_unlock(&srq_hwq->lock);
787 		return -EINVAL;
788 	}
789 	next = srq->start_idx;
790 	srq->start_idx = srq->swq[next].next_idx;
791 	spin_unlock(&srq_hwq->lock);
792 
793 	srqe = bnxt_qplib_get_qe(srq_hwq, srq_hwq->prod, NULL);
794 	memset(srqe, 0, srq->wqe_size);
795 	/* Calculate wqe_size16 and data_len */
796 	for (i = 0, hw_sge = (struct sq_sge *)srqe->data;
797 	     i < wqe->num_sge; i++, hw_sge++) {
798 		hw_sge->va_or_pa = cpu_to_le64(wqe->sg_list[i].addr);
799 		hw_sge->l_key = cpu_to_le32(wqe->sg_list[i].lkey);
800 		hw_sge->size = cpu_to_le32(wqe->sg_list[i].size);
801 	}
802 	srqe->wqe_type = wqe->type;
803 	srqe->flags = wqe->flags;
804 	srqe->wqe_size = wqe->num_sge +
805 			((offsetof(typeof(*srqe), data) + 15) >> 4);
806 	srqe->wr_id[0] = cpu_to_le32((u32)next);
807 	srq->swq[next].wr_id = wqe->wr_id;
808 
809 	bnxt_qplib_hwq_incr_prod(&srq->dbinfo, srq_hwq, srq->dbinfo.max_slot);
810 
811 	spin_lock(&srq_hwq->lock);
812 	count = __bnxt_qplib_get_avail(srq_hwq);
813 	spin_unlock(&srq_hwq->lock);
814 	/* Ring DB */
815 	bnxt_qplib_ring_prod_db(&srq->dbinfo, DBC_DBC_TYPE_SRQ);
816 	if (srq->arm_req == true && count > srq->threshold) {
817 		srq->arm_req = false;
818 		bnxt_qplib_srq_arm_db(&srq->dbinfo, srq->threshold);
819 	}
820 
821 	return 0;
822 }
823 
824 /* QP */
825 
826 static int bnxt_qplib_alloc_init_swq(struct bnxt_qplib_q *que)
827 {
828 	int indx;
829 
830 	que->swq = kcalloc(que->max_sw_wqe, sizeof(*que->swq), GFP_KERNEL);
831 	if (!que->swq)
832 		return -ENOMEM;
833 
834 	que->swq_start = 0;
835 	que->swq_last = que->max_sw_wqe - 1;
836 	for (indx = 0; indx < que->max_sw_wqe; indx++)
837 		que->swq[indx].next_idx = indx + 1;
838 	que->swq[que->swq_last].next_idx = 0; /* Make it circular */
839 	que->swq_last = 0;
840 
841 	return 0;
842 }
843 
844 int bnxt_qplib_create_qp1(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
845 {
846 	struct bnxt_qplib_hwq_attr hwq_attr = {};
847 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
848 	struct creq_create_qp1_resp resp = {};
849 	struct bnxt_qplib_cmdqmsg msg = {};
850 	struct bnxt_qplib_q *sq = &qp->sq;
851 	struct bnxt_qplib_q *rq = &qp->rq;
852 	struct cmdq_create_qp1 req = {};
853 	struct bnxt_qplib_pbl *pbl;
854 	u32 qp_flags = 0;
855 	u8 pg_sz_lvl;
856 	u32 tbl_indx;
857 	int rc;
858 
859 	sq->dbinfo.flags = 0;
860 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
861 				 CMDQ_BASE_OPCODE_CREATE_QP1,
862 				 sizeof(req));
863 	/* General */
864 	req.type = qp->type;
865 	req.dpi = cpu_to_le32(qp->dpi->dpi);
866 	req.qp_handle = cpu_to_le64(qp->qp_handle);
867 
868 	/* SQ */
869 	hwq_attr.res = res;
870 	hwq_attr.sginfo = &sq->sg_info;
871 	hwq_attr.stride = sizeof(struct sq_sge);
872 	hwq_attr.depth = bnxt_qplib_get_depth(sq, qp->wqe_mode, false);
873 	hwq_attr.type = HWQ_TYPE_QUEUE;
874 	rc = bnxt_qplib_alloc_init_hwq(&sq->hwq, &hwq_attr);
875 	if (rc)
876 		return rc;
877 
878 	rc = bnxt_qplib_alloc_init_swq(sq);
879 	if (rc)
880 		goto fail_sq;
881 
882 	req.sq_size = cpu_to_le32(bnxt_qplib_set_sq_size(sq, qp->wqe_mode));
883 	pbl = &sq->hwq.pbl[PBL_LVL_0];
884 	req.sq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
885 	pg_sz_lvl = (bnxt_qplib_base_pg_size(&sq->hwq) <<
886 		     CMDQ_CREATE_QP1_SQ_PG_SIZE_SFT);
887 	pg_sz_lvl |= (sq->hwq.level & CMDQ_CREATE_QP1_SQ_LVL_MASK);
888 	req.sq_pg_size_sq_lvl = pg_sz_lvl;
889 	req.sq_fwo_sq_sge =
890 		cpu_to_le16((sq->max_sge & CMDQ_CREATE_QP1_SQ_SGE_MASK) <<
891 			     CMDQ_CREATE_QP1_SQ_SGE_SFT);
892 	req.scq_cid = cpu_to_le32(qp->scq->id);
893 
894 	/* RQ */
895 	if (rq->max_wqe) {
896 		rq->dbinfo.flags = 0;
897 		hwq_attr.res = res;
898 		hwq_attr.sginfo = &rq->sg_info;
899 		hwq_attr.stride = sizeof(struct sq_sge);
900 		hwq_attr.depth = bnxt_qplib_get_depth(rq, qp->wqe_mode, false);
901 		hwq_attr.type = HWQ_TYPE_QUEUE;
902 		rc = bnxt_qplib_alloc_init_hwq(&rq->hwq, &hwq_attr);
903 		if (rc)
904 			goto sq_swq;
905 		rc = bnxt_qplib_alloc_init_swq(rq);
906 		if (rc)
907 			goto fail_rq;
908 		req.rq_size = cpu_to_le32(rq->max_wqe);
909 		pbl = &rq->hwq.pbl[PBL_LVL_0];
910 		req.rq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
911 		pg_sz_lvl = (bnxt_qplib_base_pg_size(&rq->hwq) <<
912 			     CMDQ_CREATE_QP1_RQ_PG_SIZE_SFT);
913 		pg_sz_lvl |= (rq->hwq.level & CMDQ_CREATE_QP1_RQ_LVL_MASK);
914 		req.rq_pg_size_rq_lvl = pg_sz_lvl;
915 		req.rq_fwo_rq_sge =
916 			cpu_to_le16((rq->max_sge &
917 				     CMDQ_CREATE_QP1_RQ_SGE_MASK) <<
918 				    CMDQ_CREATE_QP1_RQ_SGE_SFT);
919 	}
920 	req.rcq_cid = cpu_to_le32(qp->rcq->id);
921 	/* Header buffer - allow hdr_buf pass in */
922 	rc = bnxt_qplib_alloc_qp_hdr_buf(res, qp);
923 	if (rc) {
924 		rc = -ENOMEM;
925 		goto rq_rwq;
926 	}
927 	qp_flags |= CMDQ_CREATE_QP1_QP_FLAGS_RESERVED_LKEY_ENABLE;
928 	req.qp_flags = cpu_to_le32(qp_flags);
929 	req.pd_id = cpu_to_le32(qp->pd->id);
930 
931 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req), sizeof(resp), 0);
932 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
933 	if (rc)
934 		goto fail;
935 
936 	qp->id = le32_to_cpu(resp.xid);
937 	qp->cur_qp_state = CMDQ_MODIFY_QP_NEW_STATE_RESET;
938 	qp->cctx = res->cctx;
939 	sq->dbinfo.hwq = &sq->hwq;
940 	sq->dbinfo.xid = qp->id;
941 	sq->dbinfo.db = qp->dpi->dbr;
942 	sq->dbinfo.max_slot = bnxt_qplib_set_sq_max_slot(qp->wqe_mode);
943 	if (rq->max_wqe) {
944 		rq->dbinfo.hwq = &rq->hwq;
945 		rq->dbinfo.xid = qp->id;
946 		rq->dbinfo.db = qp->dpi->dbr;
947 		rq->dbinfo.max_slot = bnxt_qplib_set_rq_max_slot(rq->wqe_size);
948 	}
949 	tbl_indx = map_qp_id_to_tbl_indx(qp->id, rcfw);
950 	rcfw->qp_tbl[tbl_indx].qp_id = qp->id;
951 	rcfw->qp_tbl[tbl_indx].qp_handle = (void *)qp;
952 
953 	return 0;
954 
955 fail:
956 	bnxt_qplib_free_qp_hdr_buf(res, qp);
957 rq_rwq:
958 	kfree(rq->swq);
959 fail_rq:
960 	bnxt_qplib_free_hwq(res, &rq->hwq);
961 sq_swq:
962 	kfree(sq->swq);
963 fail_sq:
964 	bnxt_qplib_free_hwq(res, &sq->hwq);
965 	return rc;
966 }
967 
968 static void bnxt_qplib_init_psn_ptr(struct bnxt_qplib_qp *qp, int size)
969 {
970 	struct bnxt_qplib_hwq *hwq;
971 	struct bnxt_qplib_q *sq;
972 	u64 fpsne, psn_pg;
973 	u16 indx_pad = 0;
974 
975 	sq = &qp->sq;
976 	hwq = &sq->hwq;
977 	/* First psn entry */
978 	fpsne = (u64)bnxt_qplib_get_qe(hwq, hwq->depth, &psn_pg);
979 	if (!IS_ALIGNED(fpsne, PAGE_SIZE))
980 		indx_pad = (fpsne & ~PAGE_MASK) / size;
981 	hwq->pad_pgofft = indx_pad;
982 	hwq->pad_pg = (u64 *)psn_pg;
983 	hwq->pad_stride = size;
984 }
985 
986 int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
987 {
988 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
989 	struct bnxt_qplib_hwq_attr hwq_attr = {};
990 	struct bnxt_qplib_sg_info sginfo = {};
991 	struct creq_create_qp_resp resp = {};
992 	struct bnxt_qplib_cmdqmsg msg = {};
993 	struct bnxt_qplib_q *sq = &qp->sq;
994 	struct bnxt_qplib_q *rq = &qp->rq;
995 	struct cmdq_create_qp req = {};
996 	int rc, req_size, psn_sz = 0;
997 	struct bnxt_qplib_hwq *xrrq;
998 	struct bnxt_qplib_pbl *pbl;
999 	u32 qp_flags = 0;
1000 	u8 pg_sz_lvl;
1001 	u32 tbl_indx;
1002 	u16 nsge;
1003 
1004 	qp->is_host_msn_tbl = _is_host_msn_table(res->dattr->dev_cap_flags2);
1005 	sq->dbinfo.flags = 0;
1006 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
1007 				 CMDQ_BASE_OPCODE_CREATE_QP,
1008 				 sizeof(req));
1009 
1010 	/* General */
1011 	req.type = qp->type;
1012 	req.dpi = cpu_to_le32(qp->dpi->dpi);
1013 	req.qp_handle = cpu_to_le64(qp->qp_handle);
1014 
1015 	/* SQ */
1016 	if (qp->type == CMDQ_CREATE_QP_TYPE_RC) {
1017 		psn_sz = bnxt_qplib_is_chip_gen_p5_p7(res->cctx) ?
1018 			 sizeof(struct sq_psn_search_ext) :
1019 			 sizeof(struct sq_psn_search);
1020 
1021 		if (qp->is_host_msn_tbl) {
1022 			psn_sz = sizeof(struct sq_msn_search);
1023 			qp->msn = 0;
1024 		}
1025 	}
1026 
1027 	hwq_attr.res = res;
1028 	hwq_attr.sginfo = &sq->sg_info;
1029 	hwq_attr.stride = sizeof(struct sq_sge);
1030 	hwq_attr.depth = bnxt_qplib_get_depth(sq, qp->wqe_mode, true);
1031 	hwq_attr.aux_stride = psn_sz;
1032 	hwq_attr.aux_depth = psn_sz ? bnxt_qplib_set_sq_size(sq, qp->wqe_mode)
1033 				    : 0;
1034 	/* Update msn tbl size */
1035 	if (qp->is_host_msn_tbl && psn_sz) {
1036 		if (qp->wqe_mode == BNXT_QPLIB_WQE_MODE_STATIC)
1037 			hwq_attr.aux_depth =
1038 				roundup_pow_of_two(bnxt_qplib_set_sq_size(sq, qp->wqe_mode));
1039 		else
1040 			hwq_attr.aux_depth =
1041 				roundup_pow_of_two(bnxt_qplib_set_sq_size(sq, qp->wqe_mode)) / 2;
1042 		qp->msn_tbl_sz = hwq_attr.aux_depth;
1043 		qp->msn = 0;
1044 	}
1045 
1046 	hwq_attr.type = HWQ_TYPE_QUEUE;
1047 	rc = bnxt_qplib_alloc_init_hwq(&sq->hwq, &hwq_attr);
1048 	if (rc)
1049 		return rc;
1050 
1051 	if (!sq->hwq.is_user) {
1052 		rc = bnxt_qplib_alloc_init_swq(sq);
1053 		if (rc)
1054 			goto fail_sq;
1055 
1056 		if (psn_sz)
1057 			bnxt_qplib_init_psn_ptr(qp, psn_sz);
1058 	}
1059 	req.sq_size = cpu_to_le32(bnxt_qplib_set_sq_size(sq, qp->wqe_mode));
1060 	pbl = &sq->hwq.pbl[PBL_LVL_0];
1061 	req.sq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
1062 	pg_sz_lvl = (bnxt_qplib_base_pg_size(&sq->hwq) <<
1063 		     CMDQ_CREATE_QP_SQ_PG_SIZE_SFT);
1064 	pg_sz_lvl |= (sq->hwq.level & CMDQ_CREATE_QP_SQ_LVL_MASK);
1065 	req.sq_pg_size_sq_lvl = pg_sz_lvl;
1066 	req.sq_fwo_sq_sge =
1067 		cpu_to_le16(((sq->max_sge & CMDQ_CREATE_QP_SQ_SGE_MASK) <<
1068 			     CMDQ_CREATE_QP_SQ_SGE_SFT) | 0);
1069 	req.scq_cid = cpu_to_le32(qp->scq->id);
1070 
1071 	/* RQ */
1072 	if (!qp->srq) {
1073 		rq->dbinfo.flags = 0;
1074 		hwq_attr.res = res;
1075 		hwq_attr.sginfo = &rq->sg_info;
1076 		hwq_attr.stride = sizeof(struct sq_sge);
1077 		hwq_attr.depth = bnxt_qplib_get_depth(rq, qp->wqe_mode, false);
1078 		hwq_attr.aux_stride = 0;
1079 		hwq_attr.aux_depth = 0;
1080 		hwq_attr.type = HWQ_TYPE_QUEUE;
1081 		rc = bnxt_qplib_alloc_init_hwq(&rq->hwq, &hwq_attr);
1082 		if (rc)
1083 			goto sq_swq;
1084 		if (!rq->hwq.is_user) {
1085 			rc = bnxt_qplib_alloc_init_swq(rq);
1086 			if (rc)
1087 				goto fail_rq;
1088 		}
1089 
1090 		req.rq_size = cpu_to_le32(rq->max_wqe);
1091 		pbl = &rq->hwq.pbl[PBL_LVL_0];
1092 		req.rq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
1093 		pg_sz_lvl = (bnxt_qplib_base_pg_size(&rq->hwq) <<
1094 			     CMDQ_CREATE_QP_RQ_PG_SIZE_SFT);
1095 		pg_sz_lvl |= (rq->hwq.level & CMDQ_CREATE_QP_RQ_LVL_MASK);
1096 		req.rq_pg_size_rq_lvl = pg_sz_lvl;
1097 		nsge = (qp->wqe_mode == BNXT_QPLIB_WQE_MODE_STATIC) ?
1098 			6 : rq->max_sge;
1099 		req.rq_fwo_rq_sge =
1100 			cpu_to_le16(((nsge &
1101 				      CMDQ_CREATE_QP_RQ_SGE_MASK) <<
1102 				     CMDQ_CREATE_QP_RQ_SGE_SFT) | 0);
1103 	} else {
1104 		/* SRQ */
1105 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_SRQ_USED;
1106 		req.srq_cid = cpu_to_le32(qp->srq->id);
1107 	}
1108 	req.rcq_cid = cpu_to_le32(qp->rcq->id);
1109 
1110 	qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_RESERVED_LKEY_ENABLE;
1111 	qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_FR_PMR_ENABLED;
1112 	if (qp->sig_type)
1113 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_FORCE_COMPLETION;
1114 	if (qp->wqe_mode == BNXT_QPLIB_WQE_MODE_VARIABLE)
1115 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_VARIABLE_SIZED_WQE_ENABLED;
1116 	if (_is_ext_stats_supported(res->dattr->dev_cap_flags) && !res->is_vf)
1117 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_EXT_STATS_ENABLED;
1118 
1119 	req.qp_flags = cpu_to_le32(qp_flags);
1120 
1121 	/* ORRQ and IRRQ */
1122 	if (psn_sz) {
1123 		xrrq = &qp->orrq;
1124 		xrrq->max_elements =
1125 			ORD_LIMIT_TO_ORRQ_SLOTS(qp->max_rd_atomic);
1126 		req_size = xrrq->max_elements *
1127 			   BNXT_QPLIB_MAX_ORRQE_ENTRY_SIZE + PAGE_SIZE - 1;
1128 		req_size &= ~(PAGE_SIZE - 1);
1129 		sginfo.pgsize = req_size;
1130 		sginfo.pgshft = PAGE_SHIFT;
1131 
1132 		hwq_attr.res = res;
1133 		hwq_attr.sginfo = &sginfo;
1134 		hwq_attr.depth = xrrq->max_elements;
1135 		hwq_attr.stride = BNXT_QPLIB_MAX_ORRQE_ENTRY_SIZE;
1136 		hwq_attr.aux_stride = 0;
1137 		hwq_attr.aux_depth = 0;
1138 		hwq_attr.type = HWQ_TYPE_CTX;
1139 		rc = bnxt_qplib_alloc_init_hwq(xrrq, &hwq_attr);
1140 		if (rc)
1141 			goto rq_swq;
1142 		pbl = &xrrq->pbl[PBL_LVL_0];
1143 		req.orrq_addr = cpu_to_le64(pbl->pg_map_arr[0]);
1144 
1145 		xrrq = &qp->irrq;
1146 		xrrq->max_elements = IRD_LIMIT_TO_IRRQ_SLOTS(
1147 						qp->max_dest_rd_atomic);
1148 		req_size = xrrq->max_elements *
1149 			   BNXT_QPLIB_MAX_IRRQE_ENTRY_SIZE + PAGE_SIZE - 1;
1150 		req_size &= ~(PAGE_SIZE - 1);
1151 		sginfo.pgsize = req_size;
1152 		hwq_attr.depth =  xrrq->max_elements;
1153 		hwq_attr.stride = BNXT_QPLIB_MAX_IRRQE_ENTRY_SIZE;
1154 		rc = bnxt_qplib_alloc_init_hwq(xrrq, &hwq_attr);
1155 		if (rc)
1156 			goto fail_orrq;
1157 
1158 		pbl = &xrrq->pbl[PBL_LVL_0];
1159 		req.irrq_addr = cpu_to_le64(pbl->pg_map_arr[0]);
1160 	}
1161 	req.pd_id = cpu_to_le32(qp->pd->id);
1162 
1163 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
1164 				sizeof(resp), 0);
1165 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
1166 	if (rc)
1167 		goto fail;
1168 
1169 	qp->id = le32_to_cpu(resp.xid);
1170 	qp->cur_qp_state = CMDQ_MODIFY_QP_NEW_STATE_RESET;
1171 	INIT_LIST_HEAD(&qp->sq_flush);
1172 	INIT_LIST_HEAD(&qp->rq_flush);
1173 	qp->cctx = res->cctx;
1174 	sq->dbinfo.hwq = &sq->hwq;
1175 	sq->dbinfo.xid = qp->id;
1176 	sq->dbinfo.db = qp->dpi->dbr;
1177 	sq->dbinfo.max_slot = bnxt_qplib_set_sq_max_slot(qp->wqe_mode);
1178 	if (rq->max_wqe) {
1179 		rq->dbinfo.hwq = &rq->hwq;
1180 		rq->dbinfo.xid = qp->id;
1181 		rq->dbinfo.db = qp->dpi->dbr;
1182 		rq->dbinfo.max_slot = bnxt_qplib_set_rq_max_slot(rq->wqe_size);
1183 	}
1184 	spin_lock_bh(&rcfw->tbl_lock);
1185 	tbl_indx = map_qp_id_to_tbl_indx(qp->id, rcfw);
1186 	rcfw->qp_tbl[tbl_indx].qp_id = qp->id;
1187 	rcfw->qp_tbl[tbl_indx].qp_handle = (void *)qp;
1188 	spin_unlock_bh(&rcfw->tbl_lock);
1189 
1190 	return 0;
1191 fail:
1192 	bnxt_qplib_free_hwq(res, &qp->irrq);
1193 fail_orrq:
1194 	bnxt_qplib_free_hwq(res, &qp->orrq);
1195 rq_swq:
1196 	kfree(rq->swq);
1197 fail_rq:
1198 	bnxt_qplib_free_hwq(res, &rq->hwq);
1199 sq_swq:
1200 	kfree(sq->swq);
1201 fail_sq:
1202 	bnxt_qplib_free_hwq(res, &sq->hwq);
1203 	return rc;
1204 }
1205 
1206 static void __modify_flags_from_init_state(struct bnxt_qplib_qp *qp)
1207 {
1208 	switch (qp->state) {
1209 	case CMDQ_MODIFY_QP_NEW_STATE_RTR:
1210 		/* INIT->RTR, configure the path_mtu to the default
1211 		 * 2048 if not being requested
1212 		 */
1213 		if (!(qp->modify_flags &
1214 		    CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU)) {
1215 			qp->modify_flags |=
1216 				CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU;
1217 			qp->path_mtu =
1218 				CMDQ_MODIFY_QP_PATH_MTU_MTU_2048;
1219 		}
1220 		qp->modify_flags &=
1221 			~CMDQ_MODIFY_QP_MODIFY_MASK_VLAN_ID;
1222 		/* Bono FW require the max_dest_rd_atomic to be >= 1 */
1223 		if (qp->max_dest_rd_atomic < 1)
1224 			qp->max_dest_rd_atomic = 1;
1225 		qp->modify_flags &= ~CMDQ_MODIFY_QP_MODIFY_MASK_SRC_MAC;
1226 		/* Bono FW 20.6.5 requires SGID_INDEX configuration */
1227 		if (!(qp->modify_flags &
1228 		    CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX)) {
1229 			qp->modify_flags |=
1230 				CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX;
1231 			qp->ah.sgid_index = 0;
1232 		}
1233 		break;
1234 	default:
1235 		break;
1236 	}
1237 }
1238 
1239 static void __modify_flags_from_rtr_state(struct bnxt_qplib_qp *qp)
1240 {
1241 	switch (qp->state) {
1242 	case CMDQ_MODIFY_QP_NEW_STATE_RTS:
1243 		/* Bono FW requires the max_rd_atomic to be >= 1 */
1244 		if (qp->max_rd_atomic < 1)
1245 			qp->max_rd_atomic = 1;
1246 		/* Bono FW does not allow PKEY_INDEX,
1247 		 * DGID, FLOW_LABEL, SGID_INDEX, HOP_LIMIT,
1248 		 * TRAFFIC_CLASS, DEST_MAC, PATH_MTU, RQ_PSN,
1249 		 * MIN_RNR_TIMER, MAX_DEST_RD_ATOMIC, DEST_QP_ID
1250 		 * modification
1251 		 */
1252 		qp->modify_flags &=
1253 			~(CMDQ_MODIFY_QP_MODIFY_MASK_PKEY |
1254 			  CMDQ_MODIFY_QP_MODIFY_MASK_DGID |
1255 			  CMDQ_MODIFY_QP_MODIFY_MASK_FLOW_LABEL |
1256 			  CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX |
1257 			  CMDQ_MODIFY_QP_MODIFY_MASK_HOP_LIMIT |
1258 			  CMDQ_MODIFY_QP_MODIFY_MASK_TRAFFIC_CLASS |
1259 			  CMDQ_MODIFY_QP_MODIFY_MASK_DEST_MAC |
1260 			  CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU |
1261 			  CMDQ_MODIFY_QP_MODIFY_MASK_RQ_PSN |
1262 			  CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER |
1263 			  CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC |
1264 			  CMDQ_MODIFY_QP_MODIFY_MASK_DEST_QP_ID);
1265 		break;
1266 	default:
1267 		break;
1268 	}
1269 }
1270 
1271 static void __filter_modify_flags(struct bnxt_qplib_qp *qp)
1272 {
1273 	switch (qp->cur_qp_state) {
1274 	case CMDQ_MODIFY_QP_NEW_STATE_RESET:
1275 		break;
1276 	case CMDQ_MODIFY_QP_NEW_STATE_INIT:
1277 		__modify_flags_from_init_state(qp);
1278 		break;
1279 	case CMDQ_MODIFY_QP_NEW_STATE_RTR:
1280 		__modify_flags_from_rtr_state(qp);
1281 		break;
1282 	case CMDQ_MODIFY_QP_NEW_STATE_RTS:
1283 		break;
1284 	case CMDQ_MODIFY_QP_NEW_STATE_SQD:
1285 		break;
1286 	case CMDQ_MODIFY_QP_NEW_STATE_SQE:
1287 		break;
1288 	case CMDQ_MODIFY_QP_NEW_STATE_ERR:
1289 		break;
1290 	default:
1291 		break;
1292 	}
1293 }
1294 
1295 static void bnxt_set_mandatory_attributes(struct bnxt_qplib_res *res,
1296 					  struct bnxt_qplib_qp *qp,
1297 					  struct cmdq_modify_qp *req)
1298 {
1299 	u32 mandatory_flags = 0;
1300 
1301 	if (qp->type == CMDQ_MODIFY_QP_QP_TYPE_RC)
1302 		mandatory_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_ACCESS;
1303 
1304 	if (qp->cur_qp_state == CMDQ_MODIFY_QP_NEW_STATE_INIT &&
1305 	    qp->state == CMDQ_MODIFY_QP_NEW_STATE_RTR) {
1306 		if (qp->type == CMDQ_MODIFY_QP_QP_TYPE_RC && qp->srq)
1307 			req->flags = cpu_to_le16(CMDQ_MODIFY_QP_FLAGS_SRQ_USED);
1308 		mandatory_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_PKEY;
1309 	}
1310 
1311 	if (_is_min_rnr_in_rtr_rts_mandatory(res->dattr->dev_cap_flags2) &&
1312 	    (qp->cur_qp_state == CMDQ_MODIFY_QP_NEW_STATE_RTR &&
1313 	     qp->state == CMDQ_MODIFY_QP_NEW_STATE_RTS)) {
1314 		if (qp->type == CMDQ_MODIFY_QP_QP_TYPE_RC)
1315 			mandatory_flags |=
1316 				CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER;
1317 	}
1318 
1319 	if (qp->type == CMDQ_MODIFY_QP_QP_TYPE_UD ||
1320 	    qp->type == CMDQ_MODIFY_QP_QP_TYPE_GSI)
1321 		mandatory_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_QKEY;
1322 
1323 	qp->modify_flags |= mandatory_flags;
1324 	req->qp_type = qp->type;
1325 }
1326 
1327 static bool is_optimized_state_transition(struct bnxt_qplib_qp *qp)
1328 {
1329 	if ((qp->cur_qp_state == CMDQ_MODIFY_QP_NEW_STATE_INIT &&
1330 	     qp->state == CMDQ_MODIFY_QP_NEW_STATE_RTR) ||
1331 	    (qp->cur_qp_state == CMDQ_MODIFY_QP_NEW_STATE_RTR &&
1332 	     qp->state == CMDQ_MODIFY_QP_NEW_STATE_RTS))
1333 		return true;
1334 
1335 	return false;
1336 }
1337 
1338 int bnxt_qplib_modify_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
1339 {
1340 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1341 	struct creq_modify_qp_resp resp = {};
1342 	struct bnxt_qplib_cmdqmsg msg = {};
1343 	struct cmdq_modify_qp req = {};
1344 	u16 vlan_pcp_vlan_dei_vlan_id;
1345 	u32 temp32[4];
1346 	u32 bmask;
1347 	int rc;
1348 
1349 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
1350 				 CMDQ_BASE_OPCODE_MODIFY_QP,
1351 				 sizeof(req));
1352 
1353 	/* Filter out the qp_attr_mask based on the state->new transition */
1354 	__filter_modify_flags(qp);
1355 	if (qp->modify_flags & CMDQ_MODIFY_QP_MODIFY_MASK_STATE) {
1356 		/* Set mandatory attributes for INIT -> RTR and RTR -> RTS transition */
1357 		if (_is_optimize_modify_qp_supported(res->dattr->dev_cap_flags2) &&
1358 		    is_optimized_state_transition(qp))
1359 			bnxt_set_mandatory_attributes(res, qp, &req);
1360 	}
1361 	bmask = qp->modify_flags;
1362 	req.modify_mask = cpu_to_le32(qp->modify_flags);
1363 	req.qp_cid = cpu_to_le32(qp->id);
1364 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_STATE) {
1365 		req.network_type_en_sqd_async_notify_new_state =
1366 				(qp->state & CMDQ_MODIFY_QP_NEW_STATE_MASK) |
1367 				(qp->en_sqd_async_notify ?
1368 					CMDQ_MODIFY_QP_EN_SQD_ASYNC_NOTIFY : 0);
1369 	}
1370 	req.network_type_en_sqd_async_notify_new_state |= qp->nw_type;
1371 
1372 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_ACCESS)
1373 		req.access = qp->access;
1374 
1375 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_PKEY)
1376 		req.pkey = cpu_to_le16(IB_DEFAULT_PKEY_FULL);
1377 
1378 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_QKEY)
1379 		req.qkey = cpu_to_le32(qp->qkey);
1380 
1381 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DGID) {
1382 		memcpy(temp32, qp->ah.dgid.data, sizeof(struct bnxt_qplib_gid));
1383 		req.dgid[0] = cpu_to_le32(temp32[0]);
1384 		req.dgid[1] = cpu_to_le32(temp32[1]);
1385 		req.dgid[2] = cpu_to_le32(temp32[2]);
1386 		req.dgid[3] = cpu_to_le32(temp32[3]);
1387 	}
1388 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_FLOW_LABEL)
1389 		req.flow_label = cpu_to_le32(qp->ah.flow_label);
1390 
1391 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX)
1392 		req.sgid_index = cpu_to_le16(res->sgid_tbl.hw_id
1393 					     [qp->ah.sgid_index]);
1394 
1395 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_HOP_LIMIT)
1396 		req.hop_limit = qp->ah.hop_limit;
1397 
1398 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_TRAFFIC_CLASS)
1399 		req.traffic_class = qp->ah.traffic_class;
1400 
1401 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DEST_MAC)
1402 		memcpy(req.dest_mac, qp->ah.dmac, 6);
1403 
1404 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU)
1405 		req.path_mtu_pingpong_push_enable |= qp->path_mtu;
1406 
1407 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_TIMEOUT)
1408 		req.timeout = qp->timeout;
1409 
1410 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RETRY_CNT)
1411 		req.retry_cnt = qp->retry_cnt;
1412 
1413 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RNR_RETRY)
1414 		req.rnr_retry = qp->rnr_retry;
1415 
1416 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER)
1417 		req.min_rnr_timer = qp->min_rnr_timer;
1418 
1419 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RQ_PSN)
1420 		req.rq_psn = cpu_to_le32(qp->rq.psn);
1421 
1422 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_SQ_PSN)
1423 		req.sq_psn = cpu_to_le32(qp->sq.psn);
1424 
1425 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MAX_RD_ATOMIC)
1426 		req.max_rd_atomic =
1427 			ORD_LIMIT_TO_ORRQ_SLOTS(qp->max_rd_atomic);
1428 
1429 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC)
1430 		req.max_dest_rd_atomic =
1431 			IRD_LIMIT_TO_IRRQ_SLOTS(qp->max_dest_rd_atomic);
1432 
1433 	req.sq_size = cpu_to_le32(qp->sq.hwq.max_elements);
1434 	req.rq_size = cpu_to_le32(qp->rq.hwq.max_elements);
1435 	req.sq_sge = cpu_to_le16(qp->sq.max_sge);
1436 	req.rq_sge = cpu_to_le16(qp->rq.max_sge);
1437 	req.max_inline_data = cpu_to_le32(qp->max_inline_data);
1438 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DEST_QP_ID)
1439 		req.dest_qp_id = cpu_to_le32(qp->dest_qpn);
1440 
1441 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_VLAN_ID) {
1442 		vlan_pcp_vlan_dei_vlan_id =
1443 			((res->sgid_tbl.tbl[qp->ah.sgid_index].vlan_id <<
1444 			  CMDQ_MODIFY_QP_VLAN_ID_SFT) &
1445 			 CMDQ_MODIFY_QP_VLAN_ID_MASK);
1446 		vlan_pcp_vlan_dei_vlan_id |=
1447 			((qp->ah.sl << CMDQ_MODIFY_QP_VLAN_PCP_SFT) &
1448 			 CMDQ_MODIFY_QP_VLAN_PCP_MASK);
1449 		req.vlan_pcp_vlan_dei_vlan_id = cpu_to_le16(vlan_pcp_vlan_dei_vlan_id);
1450 	}
1451 
1452 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),  sizeof(resp), 0);
1453 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
1454 	if (rc)
1455 		return rc;
1456 	qp->cur_qp_state = qp->state;
1457 	return 0;
1458 }
1459 
1460 int bnxt_qplib_query_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
1461 {
1462 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1463 	struct creq_query_qp_resp resp = {};
1464 	struct bnxt_qplib_cmdqmsg msg = {};
1465 	struct bnxt_qplib_rcfw_sbuf sbuf;
1466 	struct creq_query_qp_resp_sb *sb;
1467 	struct cmdq_query_qp req = {};
1468 	u32 temp32[4];
1469 	int i, rc;
1470 
1471 	sbuf.size = ALIGN(sizeof(*sb), BNXT_QPLIB_CMDQE_UNITS);
1472 	sbuf.sb = dma_alloc_coherent(&rcfw->pdev->dev, sbuf.size,
1473 				     &sbuf.dma_addr, GFP_KERNEL);
1474 	if (!sbuf.sb)
1475 		return -ENOMEM;
1476 	sb = sbuf.sb;
1477 
1478 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
1479 				 CMDQ_BASE_OPCODE_QUERY_QP,
1480 				 sizeof(req));
1481 
1482 	req.qp_cid = cpu_to_le32(qp->id);
1483 	req.resp_size = sbuf.size / BNXT_QPLIB_CMDQE_UNITS;
1484 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, &sbuf, sizeof(req),
1485 				sizeof(resp), 0);
1486 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
1487 	if (rc)
1488 		goto bail;
1489 	/* Extract the context from the side buffer */
1490 	qp->state = sb->en_sqd_async_notify_state &
1491 			CREQ_QUERY_QP_RESP_SB_STATE_MASK;
1492 	qp->en_sqd_async_notify = sb->en_sqd_async_notify_state &
1493 				  CREQ_QUERY_QP_RESP_SB_EN_SQD_ASYNC_NOTIFY;
1494 	qp->access = sb->access;
1495 	qp->pkey_index = le16_to_cpu(sb->pkey);
1496 	qp->qkey = le32_to_cpu(sb->qkey);
1497 
1498 	temp32[0] = le32_to_cpu(sb->dgid[0]);
1499 	temp32[1] = le32_to_cpu(sb->dgid[1]);
1500 	temp32[2] = le32_to_cpu(sb->dgid[2]);
1501 	temp32[3] = le32_to_cpu(sb->dgid[3]);
1502 	memcpy(qp->ah.dgid.data, temp32, sizeof(qp->ah.dgid.data));
1503 
1504 	qp->ah.flow_label = le32_to_cpu(sb->flow_label);
1505 
1506 	qp->ah.sgid_index = 0;
1507 	for (i = 0; i < res->sgid_tbl.max; i++) {
1508 		if (res->sgid_tbl.hw_id[i] == le16_to_cpu(sb->sgid_index)) {
1509 			qp->ah.sgid_index = i;
1510 			break;
1511 		}
1512 	}
1513 	if (i == res->sgid_tbl.max)
1514 		dev_warn(&res->pdev->dev, "SGID not found??\n");
1515 
1516 	qp->ah.hop_limit = sb->hop_limit;
1517 	qp->ah.traffic_class = sb->traffic_class;
1518 	memcpy(qp->ah.dmac, sb->dest_mac, 6);
1519 	qp->ah.vlan_id = (le16_to_cpu(sb->path_mtu_dest_vlan_id) &
1520 				CREQ_QUERY_QP_RESP_SB_VLAN_ID_MASK) >>
1521 				CREQ_QUERY_QP_RESP_SB_VLAN_ID_SFT;
1522 	qp->path_mtu = (le16_to_cpu(sb->path_mtu_dest_vlan_id) &
1523 				    CREQ_QUERY_QP_RESP_SB_PATH_MTU_MASK) >>
1524 				    CREQ_QUERY_QP_RESP_SB_PATH_MTU_SFT;
1525 	qp->timeout = sb->timeout;
1526 	qp->retry_cnt = sb->retry_cnt;
1527 	qp->rnr_retry = sb->rnr_retry;
1528 	qp->min_rnr_timer = sb->min_rnr_timer;
1529 	qp->rq.psn = le32_to_cpu(sb->rq_psn);
1530 	qp->max_rd_atomic = ORRQ_SLOTS_TO_ORD_LIMIT(sb->max_rd_atomic);
1531 	qp->sq.psn = le32_to_cpu(sb->sq_psn);
1532 	qp->max_dest_rd_atomic =
1533 			IRRQ_SLOTS_TO_IRD_LIMIT(sb->max_dest_rd_atomic);
1534 	qp->sq.max_wqe = qp->sq.hwq.max_elements;
1535 	qp->rq.max_wqe = qp->rq.hwq.max_elements;
1536 	qp->sq.max_sge = le16_to_cpu(sb->sq_sge);
1537 	qp->rq.max_sge = le16_to_cpu(sb->rq_sge);
1538 	qp->max_inline_data = le32_to_cpu(sb->max_inline_data);
1539 	qp->dest_qpn = le32_to_cpu(sb->dest_qp_id);
1540 	memcpy(qp->smac, sb->src_mac, 6);
1541 	qp->vlan_id = le16_to_cpu(sb->vlan_pcp_vlan_dei_vlan_id);
1542 	qp->port_id = le16_to_cpu(sb->port_id);
1543 bail:
1544 	dma_free_coherent(&rcfw->pdev->dev, sbuf.size,
1545 			  sbuf.sb, sbuf.dma_addr);
1546 	return rc;
1547 }
1548 
1549 static void __clean_cq(struct bnxt_qplib_cq *cq, u64 qp)
1550 {
1551 	struct bnxt_qplib_hwq *cq_hwq = &cq->hwq;
1552 	u32 peek_flags, peek_cons;
1553 	struct cq_base *hw_cqe;
1554 	int i;
1555 
1556 	peek_flags = cq->dbinfo.flags;
1557 	peek_cons = cq_hwq->cons;
1558 	for (i = 0; i < cq_hwq->max_elements; i++) {
1559 		hw_cqe = bnxt_qplib_get_qe(cq_hwq, peek_cons, NULL);
1560 		if (!CQE_CMP_VALID(hw_cqe, peek_flags))
1561 			continue;
1562 		/*
1563 		 * The valid test of the entry must be done first before
1564 		 * reading any further.
1565 		 */
1566 		dma_rmb();
1567 		switch (hw_cqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK) {
1568 		case CQ_BASE_CQE_TYPE_REQ:
1569 		case CQ_BASE_CQE_TYPE_TERMINAL:
1570 		{
1571 			struct cq_req *cqe = (struct cq_req *)hw_cqe;
1572 
1573 			if (qp == le64_to_cpu(cqe->qp_handle))
1574 				cqe->qp_handle = 0;
1575 			break;
1576 		}
1577 		case CQ_BASE_CQE_TYPE_RES_RC:
1578 		case CQ_BASE_CQE_TYPE_RES_UD:
1579 		case CQ_BASE_CQE_TYPE_RES_RAWETH_QP1:
1580 		{
1581 			struct cq_res_rc *cqe = (struct cq_res_rc *)hw_cqe;
1582 
1583 			if (qp == le64_to_cpu(cqe->qp_handle))
1584 				cqe->qp_handle = 0;
1585 			break;
1586 		}
1587 		default:
1588 			break;
1589 		}
1590 		bnxt_qplib_hwq_incr_cons(cq_hwq->max_elements, &peek_cons,
1591 					 1, &peek_flags);
1592 	}
1593 }
1594 
1595 int bnxt_qplib_destroy_qp(struct bnxt_qplib_res *res,
1596 			  struct bnxt_qplib_qp *qp)
1597 {
1598 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1599 	struct creq_destroy_qp_resp resp = {};
1600 	struct bnxt_qplib_cmdqmsg msg = {};
1601 	struct cmdq_destroy_qp req = {};
1602 	u32 tbl_indx;
1603 	int rc;
1604 
1605 	spin_lock_bh(&rcfw->tbl_lock);
1606 	tbl_indx = map_qp_id_to_tbl_indx(qp->id, rcfw);
1607 	rcfw->qp_tbl[tbl_indx].qp_id = BNXT_QPLIB_QP_ID_INVALID;
1608 	rcfw->qp_tbl[tbl_indx].qp_handle = NULL;
1609 	spin_unlock_bh(&rcfw->tbl_lock);
1610 
1611 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
1612 				 CMDQ_BASE_OPCODE_DESTROY_QP,
1613 				 sizeof(req));
1614 
1615 	req.qp_cid = cpu_to_le32(qp->id);
1616 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
1617 				sizeof(resp), 0);
1618 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
1619 	if (rc) {
1620 		spin_lock_bh(&rcfw->tbl_lock);
1621 		rcfw->qp_tbl[tbl_indx].qp_id = qp->id;
1622 		rcfw->qp_tbl[tbl_indx].qp_handle = qp;
1623 		spin_unlock_bh(&rcfw->tbl_lock);
1624 		return rc;
1625 	}
1626 
1627 	return 0;
1628 }
1629 
1630 void bnxt_qplib_free_qp_res(struct bnxt_qplib_res *res,
1631 			    struct bnxt_qplib_qp *qp)
1632 {
1633 	bnxt_qplib_free_qp_hdr_buf(res, qp);
1634 	bnxt_qplib_free_hwq(res, &qp->sq.hwq);
1635 	kfree(qp->sq.swq);
1636 
1637 	bnxt_qplib_free_hwq(res, &qp->rq.hwq);
1638 	kfree(qp->rq.swq);
1639 
1640 	if (qp->irrq.max_elements)
1641 		bnxt_qplib_free_hwq(res, &qp->irrq);
1642 	if (qp->orrq.max_elements)
1643 		bnxt_qplib_free_hwq(res, &qp->orrq);
1644 
1645 }
1646 
1647 void *bnxt_qplib_get_qp1_sq_buf(struct bnxt_qplib_qp *qp,
1648 				struct bnxt_qplib_sge *sge)
1649 {
1650 	struct bnxt_qplib_q *sq = &qp->sq;
1651 	u32 sw_prod;
1652 
1653 	memset(sge, 0, sizeof(*sge));
1654 
1655 	if (qp->sq_hdr_buf) {
1656 		sw_prod = sq->swq_start;
1657 		sge->addr = (dma_addr_t)(qp->sq_hdr_buf_map +
1658 					 sw_prod * qp->sq_hdr_buf_size);
1659 		sge->lkey = 0xFFFFFFFF;
1660 		sge->size = qp->sq_hdr_buf_size;
1661 		return qp->sq_hdr_buf + sw_prod * sge->size;
1662 	}
1663 	return NULL;
1664 }
1665 
1666 u32 bnxt_qplib_get_rq_prod_index(struct bnxt_qplib_qp *qp)
1667 {
1668 	struct bnxt_qplib_q *rq = &qp->rq;
1669 
1670 	return rq->swq_start;
1671 }
1672 
1673 dma_addr_t bnxt_qplib_get_qp_buf_from_index(struct bnxt_qplib_qp *qp, u32 index)
1674 {
1675 	return (qp->rq_hdr_buf_map + index * qp->rq_hdr_buf_size);
1676 }
1677 
1678 void *bnxt_qplib_get_qp1_rq_buf(struct bnxt_qplib_qp *qp,
1679 				struct bnxt_qplib_sge *sge)
1680 {
1681 	struct bnxt_qplib_q *rq = &qp->rq;
1682 	u32 sw_prod;
1683 
1684 	memset(sge, 0, sizeof(*sge));
1685 
1686 	if (qp->rq_hdr_buf) {
1687 		sw_prod = rq->swq_start;
1688 		sge->addr = (dma_addr_t)(qp->rq_hdr_buf_map +
1689 					 sw_prod * qp->rq_hdr_buf_size);
1690 		sge->lkey = 0xFFFFFFFF;
1691 		sge->size = qp->rq_hdr_buf_size;
1692 		return qp->rq_hdr_buf + sw_prod * sge->size;
1693 	}
1694 	return NULL;
1695 }
1696 
1697 /* Fil the MSN table into the next psn row */
1698 static void bnxt_qplib_fill_msn_search(struct bnxt_qplib_qp *qp,
1699 				       struct bnxt_qplib_swqe *wqe,
1700 				       struct bnxt_qplib_swq *swq)
1701 {
1702 	struct sq_msn_search *msns;
1703 	u32 start_psn, next_psn;
1704 	u16 start_idx;
1705 
1706 	msns = (struct sq_msn_search *)swq->psn_search;
1707 	msns->start_idx_next_psn_start_psn = 0;
1708 
1709 	start_psn = swq->start_psn;
1710 	next_psn = swq->next_psn;
1711 	start_idx = swq->slot_idx;
1712 	msns->start_idx_next_psn_start_psn |=
1713 		bnxt_re_update_msn_tbl(start_idx, next_psn, start_psn);
1714 	qp->msn++;
1715 	qp->msn %= qp->msn_tbl_sz;
1716 }
1717 
1718 static void bnxt_qplib_fill_psn_search(struct bnxt_qplib_qp *qp,
1719 				       struct bnxt_qplib_swqe *wqe,
1720 				       struct bnxt_qplib_swq *swq)
1721 {
1722 	struct sq_psn_search_ext *psns_ext;
1723 	struct sq_psn_search *psns;
1724 	u32 flg_npsn;
1725 	u32 op_spsn;
1726 
1727 	if (!swq->psn_search)
1728 		return;
1729 	/* Handle MSN differently on cap flags  */
1730 	if (qp->is_host_msn_tbl) {
1731 		bnxt_qplib_fill_msn_search(qp, wqe, swq);
1732 		return;
1733 	}
1734 	psns = (struct sq_psn_search *)swq->psn_search;
1735 	psns = swq->psn_search;
1736 	psns_ext = swq->psn_ext;
1737 
1738 	op_spsn = ((swq->start_psn << SQ_PSN_SEARCH_START_PSN_SFT) &
1739 		    SQ_PSN_SEARCH_START_PSN_MASK);
1740 	op_spsn |= ((wqe->type << SQ_PSN_SEARCH_OPCODE_SFT) &
1741 		     SQ_PSN_SEARCH_OPCODE_MASK);
1742 	flg_npsn = ((swq->next_psn << SQ_PSN_SEARCH_NEXT_PSN_SFT) &
1743 		     SQ_PSN_SEARCH_NEXT_PSN_MASK);
1744 
1745 	if (bnxt_qplib_is_chip_gen_p5_p7(qp->cctx)) {
1746 		psns_ext->opcode_start_psn = cpu_to_le32(op_spsn);
1747 		psns_ext->flags_next_psn = cpu_to_le32(flg_npsn);
1748 		psns_ext->start_slot_idx = cpu_to_le16(swq->slot_idx);
1749 	} else {
1750 		psns->opcode_start_psn = cpu_to_le32(op_spsn);
1751 		psns->flags_next_psn = cpu_to_le32(flg_npsn);
1752 	}
1753 }
1754 
1755 static int bnxt_qplib_put_inline(struct bnxt_qplib_qp *qp,
1756 				 struct bnxt_qplib_swqe *wqe,
1757 				 u16 *idx)
1758 {
1759 	struct bnxt_qplib_hwq *hwq;
1760 	int len, t_len, offt;
1761 	bool pull_dst = true;
1762 	void *il_dst = NULL;
1763 	void *il_src = NULL;
1764 	int t_cplen, cplen;
1765 	int indx;
1766 
1767 	hwq = &qp->sq.hwq;
1768 	t_len = 0;
1769 	for (indx = 0; indx < wqe->num_sge; indx++) {
1770 		len = wqe->sg_list[indx].size;
1771 		il_src = (void *)wqe->sg_list[indx].addr;
1772 		t_len += len;
1773 		if (t_len > qp->max_inline_data)
1774 			return -ENOMEM;
1775 		while (len) {
1776 			if (pull_dst) {
1777 				pull_dst = false;
1778 				il_dst = bnxt_qplib_get_prod_qe(hwq, *idx);
1779 				(*idx)++;
1780 				t_cplen = 0;
1781 				offt = 0;
1782 			}
1783 			cplen = min_t(int, len, sizeof(struct sq_sge));
1784 			cplen = min_t(int, cplen,
1785 					(sizeof(struct sq_sge) - offt));
1786 			memcpy(il_dst, il_src, cplen);
1787 			t_cplen += cplen;
1788 			il_src += cplen;
1789 			il_dst += cplen;
1790 			offt += cplen;
1791 			len -= cplen;
1792 			if (t_cplen == sizeof(struct sq_sge))
1793 				pull_dst = true;
1794 		}
1795 	}
1796 
1797 	return t_len;
1798 }
1799 
1800 static u32 bnxt_qplib_put_sges(struct bnxt_qplib_hwq *hwq,
1801 			       struct bnxt_qplib_sge *ssge,
1802 			       u16 nsge, u16 *idx)
1803 {
1804 	struct sq_sge *dsge;
1805 	int indx, len = 0;
1806 
1807 	for (indx = 0; indx < nsge; indx++, (*idx)++) {
1808 		dsge = bnxt_qplib_get_prod_qe(hwq, *idx);
1809 		dsge->va_or_pa = cpu_to_le64(ssge[indx].addr);
1810 		dsge->l_key = cpu_to_le32(ssge[indx].lkey);
1811 		dsge->size = cpu_to_le32(ssge[indx].size);
1812 		len += ssge[indx].size;
1813 	}
1814 
1815 	return len;
1816 }
1817 
1818 static u16 bnxt_qplib_required_slots(struct bnxt_qplib_qp *qp,
1819 				     struct bnxt_qplib_swqe *wqe,
1820 				     u16 *wqe_sz, u16 *qdf, u8 mode)
1821 {
1822 	u32 ilsize, bytes;
1823 	u16 nsge;
1824 	u16 slot;
1825 
1826 	nsge = wqe->num_sge;
1827 	/* Adding sq_send_hdr is a misnomer, for rq also hdr size is same. */
1828 	bytes = sizeof(struct sq_send_hdr) + nsge * sizeof(struct sq_sge);
1829 	if (wqe->flags & BNXT_QPLIB_SWQE_FLAGS_INLINE) {
1830 		ilsize = bnxt_qplib_calc_ilsize(wqe, qp->max_inline_data);
1831 		bytes = ALIGN(ilsize, sizeof(struct sq_sge));
1832 		bytes += sizeof(struct sq_send_hdr);
1833 	}
1834 
1835 	*qdf =  __xlate_qfd(qp->sq.q_full_delta, bytes);
1836 	slot = bytes >> 4;
1837 	*wqe_sz = slot;
1838 	if (mode == BNXT_QPLIB_WQE_MODE_STATIC)
1839 		slot = 8;
1840 	return slot;
1841 }
1842 
1843 static void bnxt_qplib_pull_psn_buff(struct bnxt_qplib_qp *qp, struct bnxt_qplib_q *sq,
1844 				     struct bnxt_qplib_swq *swq, bool hw_retx)
1845 {
1846 	struct bnxt_qplib_hwq *hwq;
1847 	u32 pg_num, pg_indx;
1848 	void *buff;
1849 	u32 tail;
1850 
1851 	hwq = &sq->hwq;
1852 	if (!hwq->pad_pg)
1853 		return;
1854 	tail = swq->slot_idx / sq->dbinfo.max_slot;
1855 	if (hw_retx) {
1856 		/* For HW retx use qp msn index */
1857 		tail = qp->msn;
1858 		tail %= qp->msn_tbl_sz;
1859 	}
1860 	pg_num = (tail + hwq->pad_pgofft) / (PAGE_SIZE / hwq->pad_stride);
1861 	pg_indx = (tail + hwq->pad_pgofft) % (PAGE_SIZE / hwq->pad_stride);
1862 	buff = (void *)(hwq->pad_pg[pg_num] + pg_indx * hwq->pad_stride);
1863 	swq->psn_ext = buff;
1864 	swq->psn_search = buff;
1865 }
1866 
1867 void bnxt_qplib_post_send_db(struct bnxt_qplib_qp *qp)
1868 {
1869 	struct bnxt_qplib_q *sq = &qp->sq;
1870 
1871 	bnxt_qplib_ring_prod_db(&sq->dbinfo, DBC_DBC_TYPE_SQ);
1872 }
1873 
1874 int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp,
1875 			 struct bnxt_qplib_swqe *wqe)
1876 {
1877 	struct bnxt_qplib_nq_work *nq_work = NULL;
1878 	int i, rc = 0, data_len = 0, pkt_num = 0;
1879 	struct bnxt_qplib_q *sq = &qp->sq;
1880 	struct bnxt_qplib_hwq *hwq;
1881 	struct bnxt_qplib_swq *swq;
1882 	bool sch_handler = false;
1883 	u16 wqe_sz, qdf = 0;
1884 	bool msn_update;
1885 	void *base_hdr;
1886 	void *ext_hdr;
1887 	__le32 temp32;
1888 	u32 wqe_idx;
1889 	u32 slots;
1890 	u16 idx;
1891 
1892 	hwq = &sq->hwq;
1893 	if (qp->state != CMDQ_MODIFY_QP_NEW_STATE_RTS &&
1894 	    qp->state != CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1895 		dev_err(&hwq->pdev->dev,
1896 			"QPLIB: FP: QP (0x%x) is in the 0x%x state",
1897 			qp->id, qp->state);
1898 		rc = -EINVAL;
1899 		goto done;
1900 	}
1901 
1902 	slots = bnxt_qplib_required_slots(qp, wqe, &wqe_sz, &qdf, qp->wqe_mode);
1903 	if (bnxt_qplib_queue_full(sq, slots + qdf)) {
1904 		dev_err(&hwq->pdev->dev,
1905 			"prod = %#x cons = %#x qdepth = %#x delta = %#x\n",
1906 			hwq->prod, hwq->cons, hwq->depth, sq->q_full_delta);
1907 		rc = -ENOMEM;
1908 		goto done;
1909 	}
1910 
1911 	swq = bnxt_qplib_get_swqe(sq, &wqe_idx);
1912 	bnxt_qplib_pull_psn_buff(qp, sq, swq, qp->is_host_msn_tbl);
1913 
1914 	idx = 0;
1915 	swq->slot_idx = hwq->prod;
1916 	swq->slots = slots;
1917 	swq->wr_id = wqe->wr_id;
1918 	swq->type = wqe->type;
1919 	swq->flags = wqe->flags;
1920 	swq->start_psn = sq->psn & BTH_PSN_MASK;
1921 	if (qp->sig_type)
1922 		swq->flags |= SQ_SEND_FLAGS_SIGNAL_COMP;
1923 
1924 	if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1925 		sch_handler = true;
1926 		dev_dbg(&hwq->pdev->dev,
1927 			"%s Error QP. Scheduling for poll_cq\n", __func__);
1928 		goto queue_err;
1929 	}
1930 
1931 	base_hdr = bnxt_qplib_get_prod_qe(hwq, idx++);
1932 	ext_hdr = bnxt_qplib_get_prod_qe(hwq, idx++);
1933 	memset(base_hdr, 0, sizeof(struct sq_sge));
1934 	memset(ext_hdr, 0, sizeof(struct sq_sge));
1935 
1936 	if (wqe->flags & BNXT_QPLIB_SWQE_FLAGS_INLINE)
1937 		/* Copy the inline data */
1938 		data_len = bnxt_qplib_put_inline(qp, wqe, &idx);
1939 	else
1940 		data_len = bnxt_qplib_put_sges(hwq, wqe->sg_list, wqe->num_sge,
1941 					       &idx);
1942 	if (data_len < 0)
1943 		goto queue_err;
1944 	/* Make sure we update MSN table only for wired wqes */
1945 	msn_update = true;
1946 	/* Specifics */
1947 	switch (wqe->type) {
1948 	case BNXT_QPLIB_SWQE_TYPE_SEND:
1949 		if (qp->type == CMDQ_CREATE_QP1_TYPE_GSI) {
1950 			struct sq_send_raweth_qp1_hdr *sqe = base_hdr;
1951 			struct sq_raw_ext_hdr *ext_sqe = ext_hdr;
1952 			/* Assemble info for Raw Ethertype QPs */
1953 
1954 			sqe->wqe_type = wqe->type;
1955 			sqe->flags = wqe->flags;
1956 			sqe->wqe_size = wqe_sz;
1957 			sqe->cfa_action = cpu_to_le16(wqe->rawqp1.cfa_action);
1958 			sqe->lflags = cpu_to_le16(wqe->rawqp1.lflags);
1959 			sqe->length = cpu_to_le32(data_len);
1960 			ext_sqe->cfa_meta = cpu_to_le32((wqe->rawqp1.cfa_meta &
1961 				SQ_SEND_RAWETH_QP1_CFA_META_VLAN_VID_MASK) <<
1962 				SQ_SEND_RAWETH_QP1_CFA_META_VLAN_VID_SFT);
1963 
1964 			break;
1965 		}
1966 		fallthrough;
1967 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_IMM:
1968 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_INV:
1969 	{
1970 		struct sq_ud_ext_hdr *ext_sqe = ext_hdr;
1971 		struct sq_send_hdr *sqe = base_hdr;
1972 
1973 		sqe->wqe_type = wqe->type;
1974 		sqe->flags = wqe->flags;
1975 		sqe->wqe_size = wqe_sz;
1976 		sqe->inv_key_or_imm_data = cpu_to_le32(wqe->send.inv_key);
1977 		if (qp->type == CMDQ_CREATE_QP_TYPE_UD ||
1978 		    qp->type == CMDQ_CREATE_QP_TYPE_GSI) {
1979 			sqe->q_key = cpu_to_le32(wqe->send.q_key);
1980 			sqe->length = cpu_to_le32(data_len);
1981 			sq->psn = (sq->psn + 1) & BTH_PSN_MASK;
1982 			ext_sqe->dst_qp = cpu_to_le32(wqe->send.dst_qp &
1983 						      SQ_SEND_DST_QP_MASK);
1984 			ext_sqe->avid = cpu_to_le32(wqe->send.avid &
1985 						    SQ_SEND_AVID_MASK);
1986 			msn_update = false;
1987 		} else {
1988 			sqe->length = cpu_to_le32(data_len);
1989 			if (qp->mtu)
1990 				pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
1991 			if (!pkt_num)
1992 				pkt_num = 1;
1993 			sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
1994 		}
1995 		break;
1996 	}
1997 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE:
1998 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE_WITH_IMM:
1999 	case BNXT_QPLIB_SWQE_TYPE_RDMA_READ:
2000 	{
2001 		struct sq_rdma_ext_hdr *ext_sqe = ext_hdr;
2002 		struct sq_rdma_hdr *sqe = base_hdr;
2003 
2004 		sqe->wqe_type = wqe->type;
2005 		sqe->flags = wqe->flags;
2006 		sqe->wqe_size = wqe_sz;
2007 		sqe->imm_data = cpu_to_le32(wqe->rdma.inv_key);
2008 		sqe->length = cpu_to_le32((u32)data_len);
2009 		ext_sqe->remote_va = cpu_to_le64(wqe->rdma.remote_va);
2010 		ext_sqe->remote_key = cpu_to_le32(wqe->rdma.r_key);
2011 		if (qp->mtu)
2012 			pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
2013 		if (!pkt_num)
2014 			pkt_num = 1;
2015 		sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
2016 		break;
2017 	}
2018 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_CMP_AND_SWP:
2019 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_FETCH_AND_ADD:
2020 	{
2021 		struct sq_atomic_ext_hdr *ext_sqe = ext_hdr;
2022 		struct sq_atomic_hdr *sqe = base_hdr;
2023 
2024 		sqe->wqe_type = wqe->type;
2025 		sqe->flags = wqe->flags;
2026 		sqe->remote_key = cpu_to_le32(wqe->atomic.r_key);
2027 		sqe->remote_va = cpu_to_le64(wqe->atomic.remote_va);
2028 		ext_sqe->swap_data = cpu_to_le64(wqe->atomic.swap_data);
2029 		ext_sqe->cmp_data = cpu_to_le64(wqe->atomic.cmp_data);
2030 		if (qp->mtu)
2031 			pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
2032 		if (!pkt_num)
2033 			pkt_num = 1;
2034 		sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
2035 		break;
2036 	}
2037 	case BNXT_QPLIB_SWQE_TYPE_LOCAL_INV:
2038 	{
2039 		struct sq_localinvalidate *sqe = base_hdr;
2040 
2041 		sqe->wqe_type = wqe->type;
2042 		sqe->flags = wqe->flags;
2043 		sqe->inv_l_key = cpu_to_le32(wqe->local_inv.inv_l_key);
2044 		msn_update = false;
2045 		break;
2046 	}
2047 	case BNXT_QPLIB_SWQE_TYPE_FAST_REG_MR:
2048 	{
2049 		struct sq_fr_pmr_ext_hdr *ext_sqe = ext_hdr;
2050 		struct sq_fr_pmr_hdr *sqe = base_hdr;
2051 
2052 		sqe->wqe_type = wqe->type;
2053 		sqe->flags = wqe->flags;
2054 		sqe->access_cntl = wqe->frmr.access_cntl |
2055 				   SQ_FR_PMR_ACCESS_CNTL_LOCAL_WRITE;
2056 		sqe->zero_based_page_size_log =
2057 			(wqe->frmr.pg_sz_log & SQ_FR_PMR_PAGE_SIZE_LOG_MASK) <<
2058 			SQ_FR_PMR_PAGE_SIZE_LOG_SFT |
2059 			(wqe->frmr.zero_based ? SQ_FR_PMR_ZERO_BASED : 0);
2060 		sqe->l_key = cpu_to_le32(wqe->frmr.l_key);
2061 		temp32 = cpu_to_le32(wqe->frmr.length);
2062 		memcpy(sqe->length, &temp32, sizeof(wqe->frmr.length));
2063 		sqe->numlevels_pbl_page_size_log =
2064 			((wqe->frmr.pbl_pg_sz_log <<
2065 					SQ_FR_PMR_PBL_PAGE_SIZE_LOG_SFT) &
2066 					SQ_FR_PMR_PBL_PAGE_SIZE_LOG_MASK) |
2067 			((wqe->frmr.levels << SQ_FR_PMR_NUMLEVELS_SFT) &
2068 					SQ_FR_PMR_NUMLEVELS_MASK);
2069 
2070 		for (i = 0; i < wqe->frmr.page_list_len; i++)
2071 			wqe->frmr.pbl_ptr[i] = cpu_to_le64(
2072 						wqe->frmr.page_list[i] |
2073 						PTU_PTE_VALID);
2074 		ext_sqe->pblptr = cpu_to_le64(wqe->frmr.pbl_dma_ptr);
2075 		ext_sqe->va = cpu_to_le64(wqe->frmr.va);
2076 		msn_update = false;
2077 
2078 		break;
2079 	}
2080 	case BNXT_QPLIB_SWQE_TYPE_BIND_MW:
2081 	{
2082 		struct sq_bind_ext_hdr *ext_sqe = ext_hdr;
2083 		struct sq_bind_hdr *sqe = base_hdr;
2084 
2085 		sqe->wqe_type = wqe->type;
2086 		sqe->flags = wqe->flags;
2087 		sqe->access_cntl = wqe->bind.access_cntl;
2088 		sqe->mw_type_zero_based = wqe->bind.mw_type |
2089 			(wqe->bind.zero_based ? SQ_BIND_ZERO_BASED : 0);
2090 		sqe->parent_l_key = cpu_to_le32(wqe->bind.parent_l_key);
2091 		sqe->l_key = cpu_to_le32(wqe->bind.r_key);
2092 		ext_sqe->va = cpu_to_le64(wqe->bind.va);
2093 		ext_sqe->length_lo = cpu_to_le32(wqe->bind.length);
2094 		msn_update = false;
2095 		break;
2096 	}
2097 	default:
2098 		/* Bad wqe, return error */
2099 		rc = -EINVAL;
2100 		goto done;
2101 	}
2102 	if (!qp->is_host_msn_tbl || msn_update) {
2103 		swq->next_psn = sq->psn & BTH_PSN_MASK;
2104 		bnxt_qplib_fill_psn_search(qp, wqe, swq);
2105 	}
2106 queue_err:
2107 	bnxt_qplib_swq_mod_start(sq, wqe_idx);
2108 	bnxt_qplib_hwq_incr_prod(&sq->dbinfo, hwq, swq->slots);
2109 	qp->wqe_cnt++;
2110 done:
2111 	if (sch_handler) {
2112 		nq_work = kzalloc(sizeof(*nq_work), GFP_ATOMIC);
2113 		if (nq_work) {
2114 			nq_work->cq = qp->scq;
2115 			nq_work->nq = qp->scq->nq;
2116 			INIT_WORK(&nq_work->work, bnxt_qpn_cqn_sched_task);
2117 			queue_work(qp->scq->nq->cqn_wq, &nq_work->work);
2118 		} else {
2119 			dev_err(&hwq->pdev->dev,
2120 				"FP: Failed to allocate SQ nq_work!\n");
2121 			rc = -ENOMEM;
2122 		}
2123 	}
2124 	return rc;
2125 }
2126 
2127 void bnxt_qplib_post_recv_db(struct bnxt_qplib_qp *qp)
2128 {
2129 	struct bnxt_qplib_q *rq = &qp->rq;
2130 
2131 	bnxt_qplib_ring_prod_db(&rq->dbinfo, DBC_DBC_TYPE_RQ);
2132 }
2133 
2134 int bnxt_qplib_post_recv(struct bnxt_qplib_qp *qp,
2135 			 struct bnxt_qplib_swqe *wqe)
2136 {
2137 	struct bnxt_qplib_nq_work *nq_work = NULL;
2138 	struct bnxt_qplib_q *rq = &qp->rq;
2139 	struct rq_wqe_hdr *base_hdr;
2140 	struct rq_ext_hdr *ext_hdr;
2141 	struct bnxt_qplib_hwq *hwq;
2142 	struct bnxt_qplib_swq *swq;
2143 	bool sch_handler = false;
2144 	u16 wqe_sz, idx;
2145 	u32 wqe_idx;
2146 	int rc = 0;
2147 
2148 	hwq = &rq->hwq;
2149 	if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_RESET) {
2150 		dev_err(&hwq->pdev->dev,
2151 			"QPLIB: FP: QP (0x%x) is in the 0x%x state",
2152 			qp->id, qp->state);
2153 		rc = -EINVAL;
2154 		goto done;
2155 	}
2156 
2157 	if (bnxt_qplib_queue_full(rq, rq->dbinfo.max_slot)) {
2158 		dev_err(&hwq->pdev->dev,
2159 			"FP: QP (0x%x) RQ is full!\n", qp->id);
2160 		rc = -EINVAL;
2161 		goto done;
2162 	}
2163 
2164 	swq = bnxt_qplib_get_swqe(rq, &wqe_idx);
2165 	swq->wr_id = wqe->wr_id;
2166 	swq->slots = rq->dbinfo.max_slot;
2167 
2168 	if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
2169 		sch_handler = true;
2170 		dev_dbg(&hwq->pdev->dev,
2171 			"%s: Error QP. Scheduling for poll_cq\n", __func__);
2172 		goto queue_err;
2173 	}
2174 
2175 	idx = 0;
2176 	base_hdr = bnxt_qplib_get_prod_qe(hwq, idx++);
2177 	ext_hdr = bnxt_qplib_get_prod_qe(hwq, idx++);
2178 	memset(base_hdr, 0, sizeof(struct sq_sge));
2179 	memset(ext_hdr, 0, sizeof(struct sq_sge));
2180 	wqe_sz = (sizeof(struct rq_wqe_hdr) +
2181 	wqe->num_sge * sizeof(struct sq_sge)) >> 4;
2182 	bnxt_qplib_put_sges(hwq, wqe->sg_list, wqe->num_sge, &idx);
2183 	if (!wqe->num_sge) {
2184 		struct sq_sge *sge;
2185 
2186 		sge = bnxt_qplib_get_prod_qe(hwq, idx++);
2187 		sge->size = 0;
2188 		wqe_sz++;
2189 	}
2190 	base_hdr->wqe_type = wqe->type;
2191 	base_hdr->flags = wqe->flags;
2192 	base_hdr->wqe_size = wqe_sz;
2193 	base_hdr->wr_id[0] = cpu_to_le32(wqe_idx);
2194 queue_err:
2195 	bnxt_qplib_swq_mod_start(rq, wqe_idx);
2196 	bnxt_qplib_hwq_incr_prod(&rq->dbinfo, hwq, swq->slots);
2197 done:
2198 	if (sch_handler) {
2199 		nq_work = kzalloc(sizeof(*nq_work), GFP_ATOMIC);
2200 		if (nq_work) {
2201 			nq_work->cq = qp->rcq;
2202 			nq_work->nq = qp->rcq->nq;
2203 			INIT_WORK(&nq_work->work, bnxt_qpn_cqn_sched_task);
2204 			queue_work(qp->rcq->nq->cqn_wq, &nq_work->work);
2205 		} else {
2206 			dev_err(&hwq->pdev->dev,
2207 				"FP: Failed to allocate RQ nq_work!\n");
2208 			rc = -ENOMEM;
2209 		}
2210 	}
2211 
2212 	return rc;
2213 }
2214 
2215 /* CQ */
2216 int bnxt_qplib_create_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq)
2217 {
2218 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
2219 	struct bnxt_qplib_hwq_attr hwq_attr = {};
2220 	struct creq_create_cq_resp resp = {};
2221 	struct bnxt_qplib_cmdqmsg msg = {};
2222 	struct cmdq_create_cq req = {};
2223 	struct bnxt_qplib_pbl *pbl;
2224 	u32 coalescing = 0;
2225 	u32 pg_sz_lvl;
2226 	int rc;
2227 
2228 	if (!cq->dpi) {
2229 		dev_err(&rcfw->pdev->dev,
2230 			"FP: CREATE_CQ failed due to NULL DPI\n");
2231 		return -EINVAL;
2232 	}
2233 
2234 	cq->dbinfo.flags = 0;
2235 	hwq_attr.res = res;
2236 	hwq_attr.depth = cq->max_wqe;
2237 	hwq_attr.stride = sizeof(struct cq_base);
2238 	hwq_attr.type = HWQ_TYPE_QUEUE;
2239 	hwq_attr.sginfo = &cq->sg_info;
2240 	rc = bnxt_qplib_alloc_init_hwq(&cq->hwq, &hwq_attr);
2241 	if (rc)
2242 		return rc;
2243 
2244 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
2245 				 CMDQ_BASE_OPCODE_CREATE_CQ,
2246 				 sizeof(req));
2247 
2248 	req.dpi = cpu_to_le32(cq->dpi->dpi);
2249 	req.cq_handle = cpu_to_le64(cq->cq_handle);
2250 	req.cq_size = cpu_to_le32(cq->max_wqe);
2251 
2252 	if (_is_cq_coalescing_supported(res->dattr->dev_cap_flags2)) {
2253 		req.flags |= cpu_to_le16(CMDQ_CREATE_CQ_FLAGS_COALESCING_VALID);
2254 		coalescing |= ((cq->coalescing->buf_maxtime <<
2255 				CMDQ_CREATE_CQ_BUF_MAXTIME_SFT) &
2256 			       CMDQ_CREATE_CQ_BUF_MAXTIME_MASK);
2257 		coalescing |= ((cq->coalescing->normal_maxbuf <<
2258 				CMDQ_CREATE_CQ_NORMAL_MAXBUF_SFT) &
2259 			       CMDQ_CREATE_CQ_NORMAL_MAXBUF_MASK);
2260 		coalescing |= ((cq->coalescing->during_maxbuf <<
2261 				CMDQ_CREATE_CQ_DURING_MAXBUF_SFT) &
2262 			       CMDQ_CREATE_CQ_DURING_MAXBUF_MASK);
2263 		if (cq->coalescing->en_ring_idle_mode)
2264 			coalescing |= CMDQ_CREATE_CQ_ENABLE_RING_IDLE_MODE;
2265 		else
2266 			coalescing &= ~CMDQ_CREATE_CQ_ENABLE_RING_IDLE_MODE;
2267 		req.coalescing = cpu_to_le32(coalescing);
2268 	}
2269 
2270 	pbl = &cq->hwq.pbl[PBL_LVL_0];
2271 	pg_sz_lvl = (bnxt_qplib_base_pg_size(&cq->hwq) <<
2272 		     CMDQ_CREATE_CQ_PG_SIZE_SFT);
2273 	pg_sz_lvl |= (cq->hwq.level & CMDQ_CREATE_CQ_LVL_MASK);
2274 	req.pg_size_lvl = cpu_to_le32(pg_sz_lvl);
2275 	req.pbl = cpu_to_le64(pbl->pg_map_arr[0]);
2276 	req.cq_fco_cnq_id = cpu_to_le32(
2277 			(cq->cnq_hw_ring_id & CMDQ_CREATE_CQ_CNQ_ID_MASK) <<
2278 			 CMDQ_CREATE_CQ_CNQ_ID_SFT);
2279 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
2280 				sizeof(resp), 0);
2281 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
2282 	if (rc)
2283 		goto fail;
2284 
2285 	cq->id = le32_to_cpu(resp.xid);
2286 	cq->period = BNXT_QPLIB_QUEUE_START_PERIOD;
2287 	init_waitqueue_head(&cq->waitq);
2288 	INIT_LIST_HEAD(&cq->sqf_head);
2289 	INIT_LIST_HEAD(&cq->rqf_head);
2290 	spin_lock_init(&cq->compl_lock);
2291 	spin_lock_init(&cq->flush_lock);
2292 
2293 	cq->dbinfo.hwq = &cq->hwq;
2294 	cq->dbinfo.xid = cq->id;
2295 	cq->dbinfo.db = cq->dpi->dbr;
2296 	cq->dbinfo.priv_db = res->dpi_tbl.priv_db;
2297 	cq->dbinfo.flags = 0;
2298 	cq->dbinfo.toggle = 0;
2299 
2300 	bnxt_qplib_armen_db(&cq->dbinfo, DBC_DBC_TYPE_CQ_ARMENA);
2301 
2302 	return 0;
2303 
2304 fail:
2305 	bnxt_qplib_free_hwq(res, &cq->hwq);
2306 	return rc;
2307 }
2308 
2309 void bnxt_qplib_resize_cq_complete(struct bnxt_qplib_res *res,
2310 				   struct bnxt_qplib_cq *cq)
2311 {
2312 	bnxt_qplib_free_hwq(res, &cq->hwq);
2313 	memcpy(&cq->hwq, &cq->resize_hwq, sizeof(cq->hwq));
2314        /* Reset only the cons bit in the flags */
2315 	cq->dbinfo.flags &= ~(1UL << BNXT_QPLIB_FLAG_EPOCH_CONS_SHIFT);
2316 }
2317 
2318 int bnxt_qplib_resize_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq,
2319 			 int new_cqes)
2320 {
2321 	struct bnxt_qplib_hwq_attr hwq_attr = {};
2322 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
2323 	struct creq_resize_cq_resp resp = {};
2324 	struct bnxt_qplib_cmdqmsg msg = {};
2325 	struct cmdq_resize_cq req = {};
2326 	struct bnxt_qplib_pbl *pbl;
2327 	u32 pg_sz, lvl, new_sz;
2328 	int rc;
2329 
2330 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
2331 				 CMDQ_BASE_OPCODE_RESIZE_CQ,
2332 				 sizeof(req));
2333 	hwq_attr.sginfo = &cq->sg_info;
2334 	hwq_attr.res = res;
2335 	hwq_attr.depth = new_cqes;
2336 	hwq_attr.stride = sizeof(struct cq_base);
2337 	hwq_attr.type = HWQ_TYPE_QUEUE;
2338 	rc = bnxt_qplib_alloc_init_hwq(&cq->resize_hwq, &hwq_attr);
2339 	if (rc)
2340 		return rc;
2341 
2342 	req.cq_cid = cpu_to_le32(cq->id);
2343 	pbl = &cq->resize_hwq.pbl[PBL_LVL_0];
2344 	pg_sz = bnxt_qplib_base_pg_size(&cq->resize_hwq);
2345 	lvl = (cq->resize_hwq.level << CMDQ_RESIZE_CQ_LVL_SFT) &
2346 				       CMDQ_RESIZE_CQ_LVL_MASK;
2347 	new_sz = (new_cqes << CMDQ_RESIZE_CQ_NEW_CQ_SIZE_SFT) &
2348 		  CMDQ_RESIZE_CQ_NEW_CQ_SIZE_MASK;
2349 	req.new_cq_size_pg_size_lvl = cpu_to_le32(new_sz | pg_sz | lvl);
2350 	req.new_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
2351 
2352 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
2353 				sizeof(resp), 0);
2354 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
2355 	return rc;
2356 }
2357 
2358 int bnxt_qplib_destroy_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq)
2359 {
2360 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
2361 	struct creq_destroy_cq_resp resp = {};
2362 	struct bnxt_qplib_cmdqmsg msg = {};
2363 	struct cmdq_destroy_cq req = {};
2364 	u16 total_cnq_events;
2365 	int rc;
2366 
2367 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
2368 				 CMDQ_BASE_OPCODE_DESTROY_CQ,
2369 				 sizeof(req));
2370 
2371 	req.cq_cid = cpu_to_le32(cq->id);
2372 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
2373 				sizeof(resp), 0);
2374 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
2375 	if (rc)
2376 		return rc;
2377 	total_cnq_events = le16_to_cpu(resp.total_cnq_events);
2378 	__wait_for_all_nqes(cq, total_cnq_events);
2379 	bnxt_qplib_free_hwq(res, &cq->hwq);
2380 	return 0;
2381 }
2382 
2383 static int __flush_sq(struct bnxt_qplib_q *sq, struct bnxt_qplib_qp *qp,
2384 		      struct bnxt_qplib_cqe **pcqe, int *budget)
2385 {
2386 	struct bnxt_qplib_cqe *cqe;
2387 	u32 start, last;
2388 	int rc = 0;
2389 
2390 	/* Now complete all outstanding SQEs with FLUSHED_ERR */
2391 	start = sq->swq_start;
2392 	cqe = *pcqe;
2393 	while (*budget) {
2394 		last = sq->swq_last;
2395 		if (start == last)
2396 			break;
2397 		/* Skip the FENCE WQE completions */
2398 		if (sq->swq[last].wr_id == BNXT_QPLIB_FENCE_WRID) {
2399 			bnxt_qplib_cancel_phantom_processing(qp);
2400 			goto skip_compl;
2401 		}
2402 		memset(cqe, 0, sizeof(*cqe));
2403 		cqe->status = CQ_REQ_STATUS_WORK_REQUEST_FLUSHED_ERR;
2404 		cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
2405 		cqe->qp_handle = (u64)(unsigned long)qp;
2406 		cqe->wr_id = sq->swq[last].wr_id;
2407 		cqe->src_qp = qp->id;
2408 		cqe->type = sq->swq[last].type;
2409 		cqe++;
2410 		(*budget)--;
2411 skip_compl:
2412 		bnxt_qplib_hwq_incr_cons(sq->hwq.max_elements, &sq->hwq.cons,
2413 					 sq->swq[last].slots, &sq->dbinfo.flags);
2414 		sq->swq_last = sq->swq[last].next_idx;
2415 	}
2416 	*pcqe = cqe;
2417 	if (!(*budget) && sq->swq_last != start)
2418 		/* Out of budget */
2419 		rc = -EAGAIN;
2420 
2421 	return rc;
2422 }
2423 
2424 static int __flush_rq(struct bnxt_qplib_q *rq, struct bnxt_qplib_qp *qp,
2425 		      struct bnxt_qplib_cqe **pcqe, int *budget)
2426 {
2427 	struct bnxt_qplib_cqe *cqe;
2428 	u32 start, last;
2429 	int opcode = 0;
2430 	int rc = 0;
2431 
2432 	switch (qp->type) {
2433 	case CMDQ_CREATE_QP1_TYPE_GSI:
2434 		opcode = CQ_BASE_CQE_TYPE_RES_RAWETH_QP1;
2435 		break;
2436 	case CMDQ_CREATE_QP_TYPE_RC:
2437 		opcode = CQ_BASE_CQE_TYPE_RES_RC;
2438 		break;
2439 	case CMDQ_CREATE_QP_TYPE_UD:
2440 	case CMDQ_CREATE_QP_TYPE_GSI:
2441 		opcode = CQ_BASE_CQE_TYPE_RES_UD;
2442 		break;
2443 	}
2444 
2445 	/* Flush the rest of the RQ */
2446 	start = rq->swq_start;
2447 	cqe = *pcqe;
2448 	while (*budget) {
2449 		last = rq->swq_last;
2450 		if (last == start)
2451 			break;
2452 		memset(cqe, 0, sizeof(*cqe));
2453 		cqe->status =
2454 		    CQ_RES_RC_STATUS_WORK_REQUEST_FLUSHED_ERR;
2455 		cqe->opcode = opcode;
2456 		cqe->qp_handle = (unsigned long)qp;
2457 		cqe->wr_id = rq->swq[last].wr_id;
2458 		cqe++;
2459 		(*budget)--;
2460 		bnxt_qplib_hwq_incr_cons(rq->hwq.max_elements, &rq->hwq.cons,
2461 					 rq->swq[last].slots, &rq->dbinfo.flags);
2462 		rq->swq_last = rq->swq[last].next_idx;
2463 	}
2464 	*pcqe = cqe;
2465 	if (!*budget && rq->swq_last != start)
2466 		/* Out of budget */
2467 		rc = -EAGAIN;
2468 
2469 	return rc;
2470 }
2471 
2472 void bnxt_qplib_mark_qp_error(void *qp_handle)
2473 {
2474 	struct bnxt_qplib_qp *qp = qp_handle;
2475 
2476 	if (!qp)
2477 		return;
2478 
2479 	/* Must block new posting of SQ and RQ */
2480 	qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2481 	bnxt_qplib_cancel_phantom_processing(qp);
2482 }
2483 
2484 /* Note: SQE is valid from sw_sq_cons up to cqe_sq_cons (exclusive)
2485  *       CQE is track from sw_cq_cons to max_element but valid only if VALID=1
2486  */
2487 static int do_wa9060(struct bnxt_qplib_qp *qp, struct bnxt_qplib_cq *cq,
2488 		     u32 cq_cons, u32 swq_last, u32 cqe_sq_cons)
2489 {
2490 	u32 peek_sw_cq_cons, peek_sq_cons_idx, peek_flags;
2491 	struct bnxt_qplib_q *sq = &qp->sq;
2492 	struct cq_req *peek_req_hwcqe;
2493 	struct bnxt_qplib_qp *peek_qp;
2494 	struct bnxt_qplib_q *peek_sq;
2495 	struct bnxt_qplib_swq *swq;
2496 	struct cq_base *peek_hwcqe;
2497 	int i, rc = 0;
2498 
2499 	/* Normal mode */
2500 	/* Check for the psn_search marking before completing */
2501 	swq = &sq->swq[swq_last];
2502 	if (swq->psn_search &&
2503 	    le32_to_cpu(swq->psn_search->flags_next_psn) & 0x80000000) {
2504 		/* Unmark */
2505 		swq->psn_search->flags_next_psn = cpu_to_le32
2506 			(le32_to_cpu(swq->psn_search->flags_next_psn)
2507 				     & ~0x80000000);
2508 		dev_dbg(&cq->hwq.pdev->dev,
2509 			"FP: Process Req cq_cons=0x%x qp=0x%x sq cons sw=0x%x cqe=0x%x marked!\n",
2510 			cq_cons, qp->id, swq_last, cqe_sq_cons);
2511 		sq->condition = true;
2512 		sq->send_phantom = true;
2513 
2514 		/* TODO: Only ARM if the previous SQE is ARMALL */
2515 		bnxt_qplib_ring_db(&cq->dbinfo, DBC_DBC_TYPE_CQ_ARMALL);
2516 		rc = -EAGAIN;
2517 		goto out;
2518 	}
2519 	if (sq->condition) {
2520 		/* Peek at the completions */
2521 		peek_flags = cq->dbinfo.flags;
2522 		peek_sw_cq_cons = cq_cons;
2523 		i = cq->hwq.max_elements;
2524 		while (i--) {
2525 			peek_hwcqe = bnxt_qplib_get_qe(&cq->hwq,
2526 						       peek_sw_cq_cons, NULL);
2527 			/* If the next hwcqe is VALID */
2528 			if (CQE_CMP_VALID(peek_hwcqe, peek_flags)) {
2529 			/*
2530 			 * The valid test of the entry must be done first before
2531 			 * reading any further.
2532 			 */
2533 				dma_rmb();
2534 				/* If the next hwcqe is a REQ */
2535 				if ((peek_hwcqe->cqe_type_toggle &
2536 				    CQ_BASE_CQE_TYPE_MASK) ==
2537 				    CQ_BASE_CQE_TYPE_REQ) {
2538 					peek_req_hwcqe = (struct cq_req *)
2539 							 peek_hwcqe;
2540 					peek_qp = (struct bnxt_qplib_qp *)
2541 						((unsigned long)
2542 						 le64_to_cpu
2543 						 (peek_req_hwcqe->qp_handle));
2544 					peek_sq = &peek_qp->sq;
2545 					peek_sq_cons_idx =
2546 						((le16_to_cpu(
2547 						  peek_req_hwcqe->sq_cons_idx)
2548 						  - 1) % sq->max_wqe);
2549 					/* If the hwcqe's sq's wr_id matches */
2550 					if (peek_sq == sq &&
2551 					    sq->swq[peek_sq_cons_idx].wr_id ==
2552 					    BNXT_QPLIB_FENCE_WRID) {
2553 						/*
2554 						 *  Unbreak only if the phantom
2555 						 *  comes back
2556 						 */
2557 						dev_dbg(&cq->hwq.pdev->dev,
2558 							"FP: Got Phantom CQE\n");
2559 						sq->condition = false;
2560 						sq->single = true;
2561 						rc = 0;
2562 						goto out;
2563 					}
2564 				}
2565 				/* Valid but not the phantom, so keep looping */
2566 			} else {
2567 				/* Not valid yet, just exit and wait */
2568 				rc = -EINVAL;
2569 				goto out;
2570 			}
2571 			bnxt_qplib_hwq_incr_cons(cq->hwq.max_elements,
2572 						 &peek_sw_cq_cons,
2573 						 1, &peek_flags);
2574 		}
2575 		dev_err(&cq->hwq.pdev->dev,
2576 			"Should not have come here! cq_cons=0x%x qp=0x%x sq cons sw=0x%x hw=0x%x\n",
2577 			cq_cons, qp->id, swq_last, cqe_sq_cons);
2578 		rc = -EINVAL;
2579 	}
2580 out:
2581 	return rc;
2582 }
2583 
2584 static int bnxt_qplib_get_cqe_sq_cons(struct bnxt_qplib_q *sq, u32 cqe_slot)
2585 {
2586 	struct bnxt_qplib_hwq *sq_hwq;
2587 	struct bnxt_qplib_swq *swq;
2588 	int cqe_sq_cons = -1;
2589 	u32 start, last;
2590 
2591 	sq_hwq = &sq->hwq;
2592 
2593 	start = sq->swq_start;
2594 	last = sq->swq_last;
2595 
2596 	while (last != start) {
2597 		swq = &sq->swq[last];
2598 		if (swq->slot_idx  == cqe_slot) {
2599 			cqe_sq_cons = swq->next_idx;
2600 			dev_err(&sq_hwq->pdev->dev, "%s: Found cons wqe = %d slot = %d\n",
2601 				__func__, cqe_sq_cons, cqe_slot);
2602 			break;
2603 		}
2604 
2605 		last = swq->next_idx;
2606 	}
2607 	return cqe_sq_cons;
2608 }
2609 
2610 static int bnxt_qplib_cq_process_req(struct bnxt_qplib_cq *cq,
2611 				     struct cq_req *hwcqe,
2612 				     struct bnxt_qplib_cqe **pcqe, int *budget,
2613 				     u32 cq_cons, struct bnxt_qplib_qp **lib_qp)
2614 {
2615 	struct bnxt_qplib_swq *swq;
2616 	struct bnxt_qplib_cqe *cqe;
2617 	u32 cqe_sq_cons, slot_num;
2618 	struct bnxt_qplib_qp *qp;
2619 	struct bnxt_qplib_q *sq;
2620 	int cqe_cons;
2621 	int rc = 0;
2622 
2623 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2624 				      le64_to_cpu(hwcqe->qp_handle));
2625 	if (!qp) {
2626 		dev_err(&cq->hwq.pdev->dev,
2627 			"FP: Process Req qp is NULL\n");
2628 		return -EINVAL;
2629 	}
2630 	sq = &qp->sq;
2631 
2632 	cqe_sq_cons = le16_to_cpu(hwcqe->sq_cons_idx) % sq->max_sw_wqe;
2633 	if (qp->sq.flushed) {
2634 		dev_dbg(&cq->hwq.pdev->dev,
2635 			"%s: QP in Flush QP = %p\n", __func__, qp);
2636 		goto done;
2637 	}
2638 
2639 	if (__is_err_cqe_for_var_wqe(qp, hwcqe->status)) {
2640 		slot_num = le16_to_cpu(hwcqe->sq_cons_idx);
2641 		cqe_cons = bnxt_qplib_get_cqe_sq_cons(sq, slot_num);
2642 		if (cqe_cons < 0) {
2643 			dev_err(&cq->hwq.pdev->dev, "%s: Wrong SQ cons cqe_slot_indx = %d\n",
2644 				__func__, slot_num);
2645 			goto done;
2646 		}
2647 		cqe_sq_cons = cqe_cons;
2648 		dev_err(&cq->hwq.pdev->dev, "%s: cqe_sq_cons = %d swq_last = %d swq_start = %d\n",
2649 			__func__, cqe_sq_cons, sq->swq_last, sq->swq_start);
2650 	}
2651 
2652 	/* Require to walk the sq's swq to fabricate CQEs for all previously
2653 	 * signaled SWQEs due to CQE aggregation from the current sq cons
2654 	 * to the cqe_sq_cons
2655 	 */
2656 	cqe = *pcqe;
2657 	while (*budget) {
2658 		if (sq->swq_last == cqe_sq_cons)
2659 			/* Done */
2660 			break;
2661 
2662 		swq = &sq->swq[sq->swq_last];
2663 		memset(cqe, 0, sizeof(*cqe));
2664 		cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
2665 		cqe->qp_handle = (u64)(unsigned long)qp;
2666 		cqe->src_qp = qp->id;
2667 		cqe->wr_id = swq->wr_id;
2668 		if (cqe->wr_id == BNXT_QPLIB_FENCE_WRID)
2669 			goto skip;
2670 		cqe->type = swq->type;
2671 
2672 		/* For the last CQE, check for status.  For errors, regardless
2673 		 * of the request being signaled or not, it must complete with
2674 		 * the hwcqe error status
2675 		 */
2676 		if (swq->next_idx == cqe_sq_cons &&
2677 		    hwcqe->status != CQ_REQ_STATUS_OK) {
2678 			cqe->status = hwcqe->status;
2679 			dev_err(&cq->hwq.pdev->dev,
2680 				"FP: CQ Processed Req wr_id[%d] = 0x%llx with status 0x%x\n",
2681 				sq->swq_last, cqe->wr_id, cqe->status);
2682 			cqe++;
2683 			(*budget)--;
2684 			bnxt_qplib_mark_qp_error(qp);
2685 			/* Add qp to flush list of the CQ */
2686 			bnxt_qplib_add_flush_qp(qp);
2687 		} else {
2688 			/* Before we complete, do WA 9060 */
2689 			if (!bnxt_qplib_is_chip_gen_p5_p7(qp->cctx)) {
2690 				if (do_wa9060(qp, cq, cq_cons, sq->swq_last,
2691 					      cqe_sq_cons)) {
2692 					*lib_qp = qp;
2693 					goto out;
2694 				}
2695 			}
2696 			if (swq->flags & SQ_SEND_FLAGS_SIGNAL_COMP) {
2697 				cqe->status = CQ_REQ_STATUS_OK;
2698 				cqe++;
2699 				(*budget)--;
2700 			}
2701 		}
2702 skip:
2703 		bnxt_qplib_hwq_incr_cons(sq->hwq.max_elements, &sq->hwq.cons,
2704 					 swq->slots, &sq->dbinfo.flags);
2705 		sq->swq_last = swq->next_idx;
2706 		if (sq->single)
2707 			break;
2708 	}
2709 out:
2710 	*pcqe = cqe;
2711 	if (sq->swq_last != cqe_sq_cons) {
2712 		/* Out of budget */
2713 		rc = -EAGAIN;
2714 		goto done;
2715 	}
2716 	/*
2717 	 * Back to normal completion mode only after it has completed all of
2718 	 * the WC for this CQE
2719 	 */
2720 	sq->single = false;
2721 done:
2722 	return rc;
2723 }
2724 
2725 static void bnxt_qplib_release_srqe(struct bnxt_qplib_srq *srq, u32 tag)
2726 {
2727 	spin_lock(&srq->hwq.lock);
2728 	srq->swq[srq->last_idx].next_idx = (int)tag;
2729 	srq->last_idx = (int)tag;
2730 	srq->swq[srq->last_idx].next_idx = -1;
2731 	bnxt_qplib_hwq_incr_cons(srq->hwq.max_elements, &srq->hwq.cons,
2732 				 srq->dbinfo.max_slot, &srq->dbinfo.flags);
2733 	spin_unlock(&srq->hwq.lock);
2734 }
2735 
2736 static int bnxt_qplib_cq_process_res_rc(struct bnxt_qplib_cq *cq,
2737 					struct cq_res_rc *hwcqe,
2738 					struct bnxt_qplib_cqe **pcqe,
2739 					int *budget)
2740 {
2741 	struct bnxt_qplib_srq *srq;
2742 	struct bnxt_qplib_cqe *cqe;
2743 	struct bnxt_qplib_qp *qp;
2744 	struct bnxt_qplib_q *rq;
2745 	u32 wr_id_idx;
2746 
2747 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2748 				      le64_to_cpu(hwcqe->qp_handle));
2749 	if (!qp) {
2750 		dev_err(&cq->hwq.pdev->dev, "process_cq RC qp is NULL\n");
2751 		return -EINVAL;
2752 	}
2753 	if (qp->rq.flushed) {
2754 		dev_dbg(&cq->hwq.pdev->dev,
2755 			"%s: QP in Flush QP = %p\n", __func__, qp);
2756 		return 0;
2757 	}
2758 
2759 	cqe = *pcqe;
2760 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2761 	cqe->length = le32_to_cpu(hwcqe->length);
2762 	cqe->invrkey = le32_to_cpu(hwcqe->imm_data_or_inv_r_key);
2763 	cqe->mr_handle = le64_to_cpu(hwcqe->mr_handle);
2764 	cqe->flags = le16_to_cpu(hwcqe->flags);
2765 	cqe->status = hwcqe->status;
2766 	cqe->qp_handle = (u64)(unsigned long)qp;
2767 
2768 	wr_id_idx = le32_to_cpu(hwcqe->srq_or_rq_wr_id) &
2769 				CQ_RES_RC_SRQ_OR_RQ_WR_ID_MASK;
2770 	if (cqe->flags & CQ_RES_RC_FLAGS_SRQ_SRQ) {
2771 		srq = qp->srq;
2772 		if (!srq)
2773 			return -EINVAL;
2774 		if (wr_id_idx >= srq->hwq.max_elements) {
2775 			dev_err(&cq->hwq.pdev->dev,
2776 				"FP: CQ Process RC wr_id idx 0x%x exceeded SRQ max 0x%x\n",
2777 				wr_id_idx, srq->hwq.max_elements);
2778 			return -EINVAL;
2779 		}
2780 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2781 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2782 		cqe++;
2783 		(*budget)--;
2784 		*pcqe = cqe;
2785 	} else {
2786 		struct bnxt_qplib_swq *swq;
2787 
2788 		rq = &qp->rq;
2789 		if (wr_id_idx > (rq->max_wqe - 1)) {
2790 			dev_err(&cq->hwq.pdev->dev,
2791 				"FP: CQ Process RC wr_id idx 0x%x exceeded RQ max 0x%x\n",
2792 				wr_id_idx, rq->max_wqe);
2793 			return -EINVAL;
2794 		}
2795 		if (wr_id_idx != rq->swq_last)
2796 			return -EINVAL;
2797 		swq = &rq->swq[rq->swq_last];
2798 		cqe->wr_id = swq->wr_id;
2799 		cqe++;
2800 		(*budget)--;
2801 		bnxt_qplib_hwq_incr_cons(rq->hwq.max_elements, &rq->hwq.cons,
2802 					 swq->slots, &rq->dbinfo.flags);
2803 		rq->swq_last = swq->next_idx;
2804 		*pcqe = cqe;
2805 
2806 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2807 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2808 			/* Add qp to flush list of the CQ */
2809 			bnxt_qplib_add_flush_qp(qp);
2810 		}
2811 	}
2812 
2813 	return 0;
2814 }
2815 
2816 static int bnxt_qplib_cq_process_res_ud(struct bnxt_qplib_cq *cq,
2817 					struct cq_res_ud *hwcqe,
2818 					struct bnxt_qplib_cqe **pcqe,
2819 					int *budget)
2820 {
2821 	struct bnxt_qplib_srq *srq;
2822 	struct bnxt_qplib_cqe *cqe;
2823 	struct bnxt_qplib_qp *qp;
2824 	struct bnxt_qplib_q *rq;
2825 	u32 wr_id_idx;
2826 
2827 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2828 				      le64_to_cpu(hwcqe->qp_handle));
2829 	if (!qp) {
2830 		dev_err(&cq->hwq.pdev->dev, "process_cq UD qp is NULL\n");
2831 		return -EINVAL;
2832 	}
2833 	if (qp->rq.flushed) {
2834 		dev_dbg(&cq->hwq.pdev->dev,
2835 			"%s: QP in Flush QP = %p\n", __func__, qp);
2836 		return 0;
2837 	}
2838 	cqe = *pcqe;
2839 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2840 	cqe->length = le16_to_cpu(hwcqe->length) & CQ_RES_UD_LENGTH_MASK;
2841 	cqe->cfa_meta = le16_to_cpu(hwcqe->cfa_metadata);
2842 	cqe->invrkey = le32_to_cpu(hwcqe->imm_data);
2843 	cqe->flags = le16_to_cpu(hwcqe->flags);
2844 	cqe->status = hwcqe->status;
2845 	cqe->qp_handle = (u64)(unsigned long)qp;
2846 	/*FIXME: Endianness fix needed for smace */
2847 	memcpy(cqe->smac, hwcqe->src_mac, ETH_ALEN);
2848 	wr_id_idx = le32_to_cpu(hwcqe->src_qp_high_srq_or_rq_wr_id)
2849 				& CQ_RES_UD_SRQ_OR_RQ_WR_ID_MASK;
2850 	cqe->src_qp = le16_to_cpu(hwcqe->src_qp_low) |
2851 				  ((le32_to_cpu(
2852 				  hwcqe->src_qp_high_srq_or_rq_wr_id) &
2853 				 CQ_RES_UD_SRC_QP_HIGH_MASK) >> 8);
2854 
2855 	if (cqe->flags & CQ_RES_RC_FLAGS_SRQ_SRQ) {
2856 		srq = qp->srq;
2857 		if (!srq)
2858 			return -EINVAL;
2859 
2860 		if (wr_id_idx >= srq->hwq.max_elements) {
2861 			dev_err(&cq->hwq.pdev->dev,
2862 				"FP: CQ Process UD wr_id idx 0x%x exceeded SRQ max 0x%x\n",
2863 				wr_id_idx, srq->hwq.max_elements);
2864 			return -EINVAL;
2865 		}
2866 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2867 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2868 		cqe++;
2869 		(*budget)--;
2870 		*pcqe = cqe;
2871 	} else {
2872 		struct bnxt_qplib_swq *swq;
2873 
2874 		rq = &qp->rq;
2875 		if (wr_id_idx > (rq->max_wqe - 1)) {
2876 			dev_err(&cq->hwq.pdev->dev,
2877 				"FP: CQ Process UD wr_id idx 0x%x exceeded RQ max 0x%x\n",
2878 				wr_id_idx, rq->max_wqe);
2879 			return -EINVAL;
2880 		}
2881 
2882 		if (rq->swq_last != wr_id_idx)
2883 			return -EINVAL;
2884 		swq = &rq->swq[rq->swq_last];
2885 		cqe->wr_id = swq->wr_id;
2886 		cqe++;
2887 		(*budget)--;
2888 		bnxt_qplib_hwq_incr_cons(rq->hwq.max_elements, &rq->hwq.cons,
2889 					 swq->slots, &rq->dbinfo.flags);
2890 		rq->swq_last = swq->next_idx;
2891 		*pcqe = cqe;
2892 
2893 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2894 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2895 			/* Add qp to flush list of the CQ */
2896 			bnxt_qplib_add_flush_qp(qp);
2897 		}
2898 	}
2899 
2900 	return 0;
2901 }
2902 
2903 bool bnxt_qplib_is_cq_empty(struct bnxt_qplib_cq *cq)
2904 {
2905 	struct cq_base *hw_cqe;
2906 	bool rc = true;
2907 
2908 	hw_cqe = bnxt_qplib_get_qe(&cq->hwq, cq->hwq.cons, NULL);
2909 	 /* Check for Valid bit. If the CQE is valid, return false */
2910 	rc = !CQE_CMP_VALID(hw_cqe, cq->dbinfo.flags);
2911 	return rc;
2912 }
2913 
2914 static int bnxt_qplib_cq_process_res_raweth_qp1(struct bnxt_qplib_cq *cq,
2915 						struct cq_res_raweth_qp1 *hwcqe,
2916 						struct bnxt_qplib_cqe **pcqe,
2917 						int *budget)
2918 {
2919 	struct bnxt_qplib_qp *qp;
2920 	struct bnxt_qplib_q *rq;
2921 	struct bnxt_qplib_srq *srq;
2922 	struct bnxt_qplib_cqe *cqe;
2923 	u32 wr_id_idx;
2924 
2925 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2926 				      le64_to_cpu(hwcqe->qp_handle));
2927 	if (!qp) {
2928 		dev_err(&cq->hwq.pdev->dev, "process_cq Raw/QP1 qp is NULL\n");
2929 		return -EINVAL;
2930 	}
2931 	if (qp->rq.flushed) {
2932 		dev_dbg(&cq->hwq.pdev->dev,
2933 			"%s: QP in Flush QP = %p\n", __func__, qp);
2934 		return 0;
2935 	}
2936 	cqe = *pcqe;
2937 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2938 	cqe->flags = le16_to_cpu(hwcqe->flags);
2939 	cqe->qp_handle = (u64)(unsigned long)qp;
2940 
2941 	wr_id_idx =
2942 		le32_to_cpu(hwcqe->raweth_qp1_payload_offset_srq_or_rq_wr_id)
2943 				& CQ_RES_RAWETH_QP1_SRQ_OR_RQ_WR_ID_MASK;
2944 	cqe->src_qp = qp->id;
2945 	if (qp->id == 1 && !cqe->length) {
2946 		/* Add workaround for the length misdetection */
2947 		cqe->length = 296;
2948 	} else {
2949 		cqe->length = le16_to_cpu(hwcqe->length);
2950 	}
2951 	cqe->pkey_index = qp->pkey_index;
2952 	memcpy(cqe->smac, qp->smac, 6);
2953 
2954 	cqe->raweth_qp1_flags = le16_to_cpu(hwcqe->raweth_qp1_flags);
2955 	cqe->raweth_qp1_flags2 = le32_to_cpu(hwcqe->raweth_qp1_flags2);
2956 	cqe->raweth_qp1_metadata = le32_to_cpu(hwcqe->raweth_qp1_metadata);
2957 
2958 	if (cqe->flags & CQ_RES_RAWETH_QP1_FLAGS_SRQ_SRQ) {
2959 		srq = qp->srq;
2960 		if (!srq) {
2961 			dev_err(&cq->hwq.pdev->dev,
2962 				"FP: SRQ used but not defined??\n");
2963 			return -EINVAL;
2964 		}
2965 		if (wr_id_idx >= srq->hwq.max_elements) {
2966 			dev_err(&cq->hwq.pdev->dev,
2967 				"FP: CQ Process Raw/QP1 wr_id idx 0x%x exceeded SRQ max 0x%x\n",
2968 				wr_id_idx, srq->hwq.max_elements);
2969 			return -EINVAL;
2970 		}
2971 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2972 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2973 		cqe++;
2974 		(*budget)--;
2975 		*pcqe = cqe;
2976 	} else {
2977 		struct bnxt_qplib_swq *swq;
2978 
2979 		rq = &qp->rq;
2980 		if (wr_id_idx > (rq->max_wqe - 1)) {
2981 			dev_err(&cq->hwq.pdev->dev,
2982 				"FP: CQ Process Raw/QP1 RQ wr_id idx 0x%x exceeded RQ max 0x%x\n",
2983 				wr_id_idx, rq->max_wqe);
2984 			return -EINVAL;
2985 		}
2986 		if (rq->swq_last != wr_id_idx)
2987 			return -EINVAL;
2988 		swq = &rq->swq[rq->swq_last];
2989 		cqe->wr_id = swq->wr_id;
2990 		cqe++;
2991 		(*budget)--;
2992 		bnxt_qplib_hwq_incr_cons(rq->hwq.max_elements, &rq->hwq.cons,
2993 					 swq->slots, &rq->dbinfo.flags);
2994 		rq->swq_last = swq->next_idx;
2995 		*pcqe = cqe;
2996 
2997 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2998 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2999 			/* Add qp to flush list of the CQ */
3000 			bnxt_qplib_add_flush_qp(qp);
3001 		}
3002 	}
3003 
3004 	return 0;
3005 }
3006 
3007 static int bnxt_qplib_cq_process_terminal(struct bnxt_qplib_cq *cq,
3008 					  struct cq_terminal *hwcqe,
3009 					  struct bnxt_qplib_cqe **pcqe,
3010 					  int *budget)
3011 {
3012 	struct bnxt_qplib_qp *qp;
3013 	struct bnxt_qplib_q *sq, *rq;
3014 	struct bnxt_qplib_cqe *cqe;
3015 	u32 swq_last = 0, cqe_cons;
3016 	int rc = 0;
3017 
3018 	/* Check the Status */
3019 	if (hwcqe->status != CQ_TERMINAL_STATUS_OK)
3020 		dev_warn(&cq->hwq.pdev->dev,
3021 			 "FP: CQ Process Terminal Error status = 0x%x\n",
3022 			 hwcqe->status);
3023 
3024 	qp = (struct bnxt_qplib_qp *)((unsigned long)
3025 				      le64_to_cpu(hwcqe->qp_handle));
3026 	if (!qp)
3027 		return -EINVAL;
3028 
3029 	/* Must block new posting of SQ and RQ */
3030 	qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
3031 
3032 	sq = &qp->sq;
3033 	rq = &qp->rq;
3034 
3035 	cqe_cons = le16_to_cpu(hwcqe->sq_cons_idx);
3036 	if (cqe_cons == 0xFFFF)
3037 		goto do_rq;
3038 	cqe_cons %= sq->max_sw_wqe;
3039 
3040 	if (qp->sq.flushed) {
3041 		dev_dbg(&cq->hwq.pdev->dev,
3042 			"%s: QP in Flush QP = %p\n", __func__, qp);
3043 		goto sq_done;
3044 	}
3045 
3046 	/* Terminal CQE can also include aggregated successful CQEs prior.
3047 	 * So we must complete all CQEs from the current sq's cons to the
3048 	 * cq_cons with status OK
3049 	 */
3050 	cqe = *pcqe;
3051 	while (*budget) {
3052 		swq_last = sq->swq_last;
3053 		if (swq_last == cqe_cons)
3054 			break;
3055 		if (sq->swq[swq_last].flags & SQ_SEND_FLAGS_SIGNAL_COMP) {
3056 			memset(cqe, 0, sizeof(*cqe));
3057 			cqe->status = CQ_REQ_STATUS_OK;
3058 			cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
3059 			cqe->qp_handle = (u64)(unsigned long)qp;
3060 			cqe->src_qp = qp->id;
3061 			cqe->wr_id = sq->swq[swq_last].wr_id;
3062 			cqe->type = sq->swq[swq_last].type;
3063 			cqe++;
3064 			(*budget)--;
3065 		}
3066 		bnxt_qplib_hwq_incr_cons(sq->hwq.max_elements, &sq->hwq.cons,
3067 					 sq->swq[swq_last].slots, &sq->dbinfo.flags);
3068 		sq->swq_last = sq->swq[swq_last].next_idx;
3069 	}
3070 	*pcqe = cqe;
3071 	if (!(*budget) && swq_last != cqe_cons) {
3072 		/* Out of budget */
3073 		rc = -EAGAIN;
3074 		goto sq_done;
3075 	}
3076 sq_done:
3077 	if (rc)
3078 		return rc;
3079 do_rq:
3080 	cqe_cons = le16_to_cpu(hwcqe->rq_cons_idx);
3081 	if (cqe_cons == 0xFFFF) {
3082 		goto done;
3083 	} else if (cqe_cons > rq->max_wqe - 1) {
3084 		dev_err(&cq->hwq.pdev->dev,
3085 			"FP: CQ Processed terminal reported rq_cons_idx 0x%x exceeds max 0x%x\n",
3086 			cqe_cons, rq->max_wqe);
3087 		rc = -EINVAL;
3088 		goto done;
3089 	}
3090 
3091 	if (qp->rq.flushed) {
3092 		dev_dbg(&cq->hwq.pdev->dev,
3093 			"%s: QP in Flush QP = %p\n", __func__, qp);
3094 		rc = 0;
3095 		goto done;
3096 	}
3097 
3098 	/* Terminal CQE requires all posted RQEs to complete with FLUSHED_ERR
3099 	 * from the current rq->cons to the rq->prod regardless what the
3100 	 * rq->cons the terminal CQE indicates
3101 	 */
3102 
3103 	/* Add qp to flush list of the CQ */
3104 	bnxt_qplib_add_flush_qp(qp);
3105 done:
3106 	return rc;
3107 }
3108 
3109 static int bnxt_qplib_cq_process_cutoff(struct bnxt_qplib_cq *cq,
3110 					struct cq_cutoff *hwcqe)
3111 {
3112 	/* Check the Status */
3113 	if (hwcqe->status != CQ_CUTOFF_STATUS_OK) {
3114 		dev_err(&cq->hwq.pdev->dev,
3115 			"FP: CQ Process Cutoff Error status = 0x%x\n",
3116 			hwcqe->status);
3117 		return -EINVAL;
3118 	}
3119 	clear_bit(CQ_FLAGS_RESIZE_IN_PROG, &cq->flags);
3120 	wake_up_interruptible(&cq->waitq);
3121 
3122 	return 0;
3123 }
3124 
3125 int bnxt_qplib_process_flush_list(struct bnxt_qplib_cq *cq,
3126 				  struct bnxt_qplib_cqe *cqe,
3127 				  int num_cqes)
3128 {
3129 	struct bnxt_qplib_qp *qp = NULL;
3130 	u32 budget = num_cqes;
3131 	unsigned long flags;
3132 
3133 	spin_lock_irqsave(&cq->flush_lock, flags);
3134 	list_for_each_entry(qp, &cq->sqf_head, sq_flush) {
3135 		dev_dbg(&cq->hwq.pdev->dev, "FP: Flushing SQ QP= %p\n", qp);
3136 		__flush_sq(&qp->sq, qp, &cqe, &budget);
3137 	}
3138 
3139 	list_for_each_entry(qp, &cq->rqf_head, rq_flush) {
3140 		dev_dbg(&cq->hwq.pdev->dev, "FP: Flushing RQ QP= %p\n", qp);
3141 		__flush_rq(&qp->rq, qp, &cqe, &budget);
3142 	}
3143 	spin_unlock_irqrestore(&cq->flush_lock, flags);
3144 
3145 	return num_cqes - budget;
3146 }
3147 
3148 int bnxt_qplib_poll_cq(struct bnxt_qplib_cq *cq, struct bnxt_qplib_cqe *cqe,
3149 		       int num_cqes, struct bnxt_qplib_qp **lib_qp)
3150 {
3151 	struct cq_base *hw_cqe;
3152 	int budget, rc = 0;
3153 	u32 hw_polled = 0;
3154 	u8 type;
3155 
3156 	budget = num_cqes;
3157 
3158 	while (budget) {
3159 		hw_cqe = bnxt_qplib_get_qe(&cq->hwq, cq->hwq.cons, NULL);
3160 
3161 		/* Check for Valid bit */
3162 		if (!CQE_CMP_VALID(hw_cqe, cq->dbinfo.flags))
3163 			break;
3164 
3165 		/*
3166 		 * The valid test of the entry must be done first before
3167 		 * reading any further.
3168 		 */
3169 		dma_rmb();
3170 		/* From the device's respective CQE format to qplib_wc*/
3171 		type = hw_cqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
3172 		switch (type) {
3173 		case CQ_BASE_CQE_TYPE_REQ:
3174 			rc = bnxt_qplib_cq_process_req(cq,
3175 						       (struct cq_req *)hw_cqe,
3176 						       &cqe, &budget,
3177 						       cq->hwq.cons, lib_qp);
3178 			break;
3179 		case CQ_BASE_CQE_TYPE_RES_RC:
3180 			rc = bnxt_qplib_cq_process_res_rc(cq,
3181 							  (struct cq_res_rc *)
3182 							  hw_cqe, &cqe,
3183 							  &budget);
3184 			break;
3185 		case CQ_BASE_CQE_TYPE_RES_UD:
3186 			rc = bnxt_qplib_cq_process_res_ud
3187 					(cq, (struct cq_res_ud *)hw_cqe, &cqe,
3188 					 &budget);
3189 			break;
3190 		case CQ_BASE_CQE_TYPE_RES_RAWETH_QP1:
3191 			rc = bnxt_qplib_cq_process_res_raweth_qp1
3192 					(cq, (struct cq_res_raweth_qp1 *)
3193 					 hw_cqe, &cqe, &budget);
3194 			break;
3195 		case CQ_BASE_CQE_TYPE_TERMINAL:
3196 			rc = bnxt_qplib_cq_process_terminal
3197 					(cq, (struct cq_terminal *)hw_cqe,
3198 					 &cqe, &budget);
3199 			break;
3200 		case CQ_BASE_CQE_TYPE_CUT_OFF:
3201 			bnxt_qplib_cq_process_cutoff
3202 					(cq, (struct cq_cutoff *)hw_cqe);
3203 			/* Done processing this CQ */
3204 			goto exit;
3205 		default:
3206 			dev_err(&cq->hwq.pdev->dev,
3207 				"process_cq unknown type 0x%lx\n",
3208 				hw_cqe->cqe_type_toggle &
3209 				CQ_BASE_CQE_TYPE_MASK);
3210 			rc = -EINVAL;
3211 			break;
3212 		}
3213 		if (rc < 0) {
3214 			if (rc == -EAGAIN)
3215 				break;
3216 			/* Error while processing the CQE, just skip to the
3217 			 * next one
3218 			 */
3219 			if (type != CQ_BASE_CQE_TYPE_TERMINAL)
3220 				dev_err(&cq->hwq.pdev->dev,
3221 					"process_cqe error rc = 0x%x\n", rc);
3222 		}
3223 		hw_polled++;
3224 		bnxt_qplib_hwq_incr_cons(cq->hwq.max_elements, &cq->hwq.cons,
3225 					 1, &cq->dbinfo.flags);
3226 
3227 	}
3228 	if (hw_polled)
3229 		bnxt_qplib_ring_db(&cq->dbinfo, DBC_DBC_TYPE_CQ);
3230 exit:
3231 	return num_cqes - budget;
3232 }
3233 
3234 void bnxt_qplib_req_notify_cq(struct bnxt_qplib_cq *cq, u32 arm_type)
3235 {
3236 	cq->dbinfo.toggle = cq->toggle;
3237 	if (arm_type)
3238 		bnxt_qplib_ring_db(&cq->dbinfo, arm_type);
3239 	/* Using cq->arm_state variable to track whether to issue cq handler */
3240 	atomic_set(&cq->arm_state, 1);
3241 }
3242 
3243 void bnxt_qplib_flush_cqn_wq(struct bnxt_qplib_qp *qp)
3244 {
3245 	flush_workqueue(qp->scq->nq->cqn_wq);
3246 	if (qp->scq != qp->rcq)
3247 		flush_workqueue(qp->rcq->nq->cqn_wq);
3248 }
3249