xref: /linux/drivers/infiniband/hw/bnxt_re/qplib_fp.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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 
560 	/* Have a task to schedule CQ notifiers in post send case */
561 	nq->cqn_wq  = create_singlethread_workqueue("bnxt_qplib_nq");
562 	if (!nq->cqn_wq)
563 		return -ENOMEM;
564 
565 	rc = bnxt_qplib_map_nq_db(nq, bar_reg_offset);
566 	if (rc)
567 		goto fail;
568 
569 	rc = bnxt_qplib_nq_start_irq(nq, nq_idx, msix_vector, true);
570 	if (rc) {
571 		dev_err(&nq->pdev->dev,
572 			"Failed to request irq for nq-idx %d\n", nq_idx);
573 		goto fail;
574 	}
575 
576 	return 0;
577 fail:
578 	bnxt_qplib_disable_nq(nq);
579 	return rc;
580 }
581 
582 void bnxt_qplib_free_nq(struct bnxt_qplib_nq *nq)
583 {
584 	if (nq->hwq.max_elements) {
585 		bnxt_qplib_free_hwq(nq->res, &nq->hwq);
586 		nq->hwq.max_elements = 0;
587 	}
588 }
589 
590 int bnxt_qplib_alloc_nq(struct bnxt_qplib_res *res, struct bnxt_qplib_nq *nq)
591 {
592 	struct bnxt_qplib_hwq_attr hwq_attr = {};
593 	struct bnxt_qplib_sg_info sginfo = {};
594 
595 	nq->pdev = res->pdev;
596 	nq->res = res;
597 	if (!nq->hwq.max_elements ||
598 	    nq->hwq.max_elements > BNXT_QPLIB_NQE_MAX_CNT)
599 		nq->hwq.max_elements = BNXT_QPLIB_NQE_MAX_CNT;
600 
601 	sginfo.pgsize = PAGE_SIZE;
602 	sginfo.pgshft = PAGE_SHIFT;
603 	hwq_attr.res = res;
604 	hwq_attr.sginfo = &sginfo;
605 	hwq_attr.depth = nq->hwq.max_elements;
606 	hwq_attr.stride = sizeof(struct nq_base);
607 	hwq_attr.type = bnxt_qplib_get_hwq_type(nq->res);
608 	if (bnxt_qplib_alloc_init_hwq(&nq->hwq, &hwq_attr)) {
609 		dev_err(&nq->pdev->dev, "FP NQ allocation failed");
610 		return -ENOMEM;
611 	}
612 	nq->budget = 8;
613 	return 0;
614 }
615 
616 /* SRQ */
617 void bnxt_qplib_destroy_srq(struct bnxt_qplib_res *res,
618 			   struct bnxt_qplib_srq *srq)
619 {
620 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
621 	struct creq_destroy_srq_resp resp = {};
622 	struct bnxt_qplib_cmdqmsg msg = {};
623 	struct cmdq_destroy_srq req = {};
624 	int rc;
625 
626 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
627 				 CMDQ_BASE_OPCODE_DESTROY_SRQ,
628 				 sizeof(req));
629 
630 	/* Configure the request */
631 	req.srq_cid = cpu_to_le32(srq->id);
632 
633 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req), sizeof(resp), 0);
634 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
635 	kfree(srq->swq);
636 	if (rc)
637 		return;
638 	bnxt_qplib_free_hwq(res, &srq->hwq);
639 }
640 
641 int bnxt_qplib_create_srq(struct bnxt_qplib_res *res,
642 			  struct bnxt_qplib_srq *srq)
643 {
644 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
645 	struct bnxt_qplib_hwq_attr hwq_attr = {};
646 	struct creq_create_srq_resp resp = {};
647 	struct bnxt_qplib_cmdqmsg msg = {};
648 	struct cmdq_create_srq req = {};
649 	struct bnxt_qplib_pbl *pbl;
650 	u16 pg_sz_lvl;
651 	int rc, idx;
652 
653 	hwq_attr.res = res;
654 	hwq_attr.sginfo = &srq->sg_info;
655 	hwq_attr.depth = srq->max_wqe;
656 	hwq_attr.stride = srq->wqe_size;
657 	hwq_attr.type = HWQ_TYPE_QUEUE;
658 	rc = bnxt_qplib_alloc_init_hwq(&srq->hwq, &hwq_attr);
659 	if (rc)
660 		return rc;
661 
662 	srq->swq = kcalloc(srq->hwq.max_elements, sizeof(*srq->swq),
663 			   GFP_KERNEL);
664 	if (!srq->swq) {
665 		rc = -ENOMEM;
666 		goto fail;
667 	}
668 	srq->dbinfo.flags = 0;
669 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
670 				 CMDQ_BASE_OPCODE_CREATE_SRQ,
671 				 sizeof(req));
672 
673 	/* Configure the request */
674 	req.dpi = cpu_to_le32(srq->dpi->dpi);
675 	req.srq_handle = cpu_to_le64((uintptr_t)srq);
676 
677 	req.srq_size = cpu_to_le16((u16)srq->hwq.max_elements);
678 	pbl = &srq->hwq.pbl[PBL_LVL_0];
679 	pg_sz_lvl = ((u16)bnxt_qplib_base_pg_size(&srq->hwq) <<
680 		     CMDQ_CREATE_SRQ_PG_SIZE_SFT);
681 	pg_sz_lvl |= (srq->hwq.level & CMDQ_CREATE_SRQ_LVL_MASK) <<
682 		      CMDQ_CREATE_SRQ_LVL_SFT;
683 	req.pg_size_lvl = cpu_to_le16(pg_sz_lvl);
684 	req.pbl = cpu_to_le64(pbl->pg_map_arr[0]);
685 	req.pd_id = cpu_to_le32(srq->pd->id);
686 	req.eventq_id = cpu_to_le16(srq->eventq_hw_ring_id);
687 
688 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req), sizeof(resp), 0);
689 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
690 	if (rc)
691 		goto fail;
692 
693 	spin_lock_init(&srq->lock);
694 	srq->start_idx = 0;
695 	srq->last_idx = srq->hwq.max_elements - 1;
696 	for (idx = 0; idx < srq->hwq.max_elements; idx++)
697 		srq->swq[idx].next_idx = idx + 1;
698 	srq->swq[srq->last_idx].next_idx = -1;
699 
700 	srq->id = le32_to_cpu(resp.xid);
701 	srq->dbinfo.hwq = &srq->hwq;
702 	srq->dbinfo.xid = srq->id;
703 	srq->dbinfo.db = srq->dpi->dbr;
704 	srq->dbinfo.max_slot = 1;
705 	srq->dbinfo.priv_db = res->dpi_tbl.priv_db;
706 	if (srq->threshold)
707 		bnxt_qplib_armen_db(&srq->dbinfo, DBC_DBC_TYPE_SRQ_ARMENA);
708 	srq->arm_req = false;
709 
710 	return 0;
711 fail:
712 	bnxt_qplib_free_hwq(res, &srq->hwq);
713 	kfree(srq->swq);
714 
715 	return rc;
716 }
717 
718 int bnxt_qplib_modify_srq(struct bnxt_qplib_res *res,
719 			  struct bnxt_qplib_srq *srq)
720 {
721 	struct bnxt_qplib_hwq *srq_hwq = &srq->hwq;
722 	u32 count;
723 
724 	count = __bnxt_qplib_get_avail(srq_hwq);
725 	if (count > srq->threshold) {
726 		srq->arm_req = false;
727 		bnxt_qplib_srq_arm_db(&srq->dbinfo, srq->threshold);
728 	} else {
729 		/* Deferred arming */
730 		srq->arm_req = true;
731 	}
732 
733 	return 0;
734 }
735 
736 int bnxt_qplib_query_srq(struct bnxt_qplib_res *res,
737 			 struct bnxt_qplib_srq *srq)
738 {
739 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
740 	struct creq_query_srq_resp resp = {};
741 	struct bnxt_qplib_cmdqmsg msg = {};
742 	struct bnxt_qplib_rcfw_sbuf sbuf;
743 	struct creq_query_srq_resp_sb *sb;
744 	struct cmdq_query_srq req = {};
745 	int rc;
746 
747 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
748 				 CMDQ_BASE_OPCODE_QUERY_SRQ,
749 				 sizeof(req));
750 
751 	/* Configure the request */
752 	sbuf.size = ALIGN(sizeof(*sb), BNXT_QPLIB_CMDQE_UNITS);
753 	sbuf.sb = dma_alloc_coherent(&rcfw->pdev->dev, sbuf.size,
754 				     &sbuf.dma_addr, GFP_KERNEL);
755 	if (!sbuf.sb)
756 		return -ENOMEM;
757 	req.resp_size = sbuf.size / BNXT_QPLIB_CMDQE_UNITS;
758 	req.srq_cid = cpu_to_le32(srq->id);
759 	sb = sbuf.sb;
760 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, &sbuf, sizeof(req),
761 				sizeof(resp), 0);
762 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
763 	if (!rc)
764 		srq->threshold = le16_to_cpu(sb->srq_limit);
765 	dma_free_coherent(&rcfw->pdev->dev, sbuf.size,
766 			  sbuf.sb, sbuf.dma_addr);
767 
768 	return rc;
769 }
770 
771 int bnxt_qplib_post_srq_recv(struct bnxt_qplib_srq *srq,
772 			     struct bnxt_qplib_swqe *wqe)
773 {
774 	struct bnxt_qplib_hwq *srq_hwq = &srq->hwq;
775 	struct rq_wqe *srqe;
776 	struct sq_sge *hw_sge;
777 	u32 count = 0;
778 	int i, next;
779 
780 	spin_lock(&srq_hwq->lock);
781 	if (srq->start_idx == srq->last_idx) {
782 		dev_err(&srq_hwq->pdev->dev,
783 			"FP: SRQ (0x%x) is full!\n", srq->id);
784 		spin_unlock(&srq_hwq->lock);
785 		return -EINVAL;
786 	}
787 	next = srq->start_idx;
788 	srq->start_idx = srq->swq[next].next_idx;
789 	spin_unlock(&srq_hwq->lock);
790 
791 	srqe = bnxt_qplib_get_qe(srq_hwq, srq_hwq->prod, NULL);
792 	memset(srqe, 0, srq->wqe_size);
793 	/* Calculate wqe_size16 and data_len */
794 	for (i = 0, hw_sge = (struct sq_sge *)srqe->data;
795 	     i < wqe->num_sge; i++, hw_sge++) {
796 		hw_sge->va_or_pa = cpu_to_le64(wqe->sg_list[i].addr);
797 		hw_sge->l_key = cpu_to_le32(wqe->sg_list[i].lkey);
798 		hw_sge->size = cpu_to_le32(wqe->sg_list[i].size);
799 	}
800 	srqe->wqe_type = wqe->type;
801 	srqe->flags = wqe->flags;
802 	srqe->wqe_size = wqe->num_sge +
803 			((offsetof(typeof(*srqe), data) + 15) >> 4);
804 	srqe->wr_id[0] = cpu_to_le32((u32)next);
805 	srq->swq[next].wr_id = wqe->wr_id;
806 
807 	bnxt_qplib_hwq_incr_prod(&srq->dbinfo, srq_hwq, srq->dbinfo.max_slot);
808 
809 	spin_lock(&srq_hwq->lock);
810 	count = __bnxt_qplib_get_avail(srq_hwq);
811 	spin_unlock(&srq_hwq->lock);
812 	/* Ring DB */
813 	bnxt_qplib_ring_prod_db(&srq->dbinfo, DBC_DBC_TYPE_SRQ);
814 	if (srq->arm_req == true && count > srq->threshold) {
815 		srq->arm_req = false;
816 		bnxt_qplib_srq_arm_db(&srq->dbinfo, srq->threshold);
817 	}
818 
819 	return 0;
820 }
821 
822 /* QP */
823 
824 static int bnxt_qplib_alloc_init_swq(struct bnxt_qplib_q *que)
825 {
826 	int indx;
827 
828 	que->swq = kcalloc(que->max_sw_wqe, sizeof(*que->swq), GFP_KERNEL);
829 	if (!que->swq)
830 		return -ENOMEM;
831 
832 	que->swq_start = 0;
833 	que->swq_last = que->max_sw_wqe - 1;
834 	for (indx = 0; indx < que->max_sw_wqe; indx++)
835 		que->swq[indx].next_idx = indx + 1;
836 	que->swq[que->swq_last].next_idx = 0; /* Make it circular */
837 	que->swq_last = 0;
838 
839 	return 0;
840 }
841 
842 int bnxt_qplib_create_qp1(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
843 {
844 	struct bnxt_qplib_hwq_attr hwq_attr = {};
845 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
846 	struct creq_create_qp1_resp resp = {};
847 	struct bnxt_qplib_cmdqmsg msg = {};
848 	struct bnxt_qplib_q *sq = &qp->sq;
849 	struct bnxt_qplib_q *rq = &qp->rq;
850 	struct cmdq_create_qp1 req = {};
851 	struct bnxt_qplib_pbl *pbl;
852 	u32 qp_flags = 0;
853 	u8 pg_sz_lvl;
854 	u32 tbl_indx;
855 	int rc;
856 
857 	sq->dbinfo.flags = 0;
858 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
859 				 CMDQ_BASE_OPCODE_CREATE_QP1,
860 				 sizeof(req));
861 	/* General */
862 	req.type = qp->type;
863 	req.dpi = cpu_to_le32(qp->dpi->dpi);
864 	req.qp_handle = cpu_to_le64(qp->qp_handle);
865 
866 	/* SQ */
867 	hwq_attr.res = res;
868 	hwq_attr.sginfo = &sq->sg_info;
869 	hwq_attr.stride = sizeof(struct sq_sge);
870 	hwq_attr.depth = bnxt_qplib_get_depth(sq, qp->wqe_mode, false);
871 	hwq_attr.type = HWQ_TYPE_QUEUE;
872 	rc = bnxt_qplib_alloc_init_hwq(&sq->hwq, &hwq_attr);
873 	if (rc)
874 		return rc;
875 
876 	rc = bnxt_qplib_alloc_init_swq(sq);
877 	if (rc)
878 		goto fail_sq;
879 
880 	req.sq_size = cpu_to_le32(bnxt_qplib_set_sq_size(sq, qp->wqe_mode));
881 	pbl = &sq->hwq.pbl[PBL_LVL_0];
882 	req.sq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
883 	pg_sz_lvl = (bnxt_qplib_base_pg_size(&sq->hwq) <<
884 		     CMDQ_CREATE_QP1_SQ_PG_SIZE_SFT);
885 	pg_sz_lvl |= (sq->hwq.level & CMDQ_CREATE_QP1_SQ_LVL_MASK);
886 	req.sq_pg_size_sq_lvl = pg_sz_lvl;
887 	req.sq_fwo_sq_sge =
888 		cpu_to_le16((sq->max_sge & CMDQ_CREATE_QP1_SQ_SGE_MASK) <<
889 			     CMDQ_CREATE_QP1_SQ_SGE_SFT);
890 	req.scq_cid = cpu_to_le32(qp->scq->id);
891 
892 	/* RQ */
893 	if (rq->max_wqe) {
894 		rq->dbinfo.flags = 0;
895 		hwq_attr.res = res;
896 		hwq_attr.sginfo = &rq->sg_info;
897 		hwq_attr.stride = sizeof(struct sq_sge);
898 		hwq_attr.depth = bnxt_qplib_get_depth(rq, qp->wqe_mode, false);
899 		hwq_attr.type = HWQ_TYPE_QUEUE;
900 		rc = bnxt_qplib_alloc_init_hwq(&rq->hwq, &hwq_attr);
901 		if (rc)
902 			goto sq_swq;
903 		rc = bnxt_qplib_alloc_init_swq(rq);
904 		if (rc)
905 			goto fail_rq;
906 		req.rq_size = cpu_to_le32(rq->max_wqe);
907 		pbl = &rq->hwq.pbl[PBL_LVL_0];
908 		req.rq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
909 		pg_sz_lvl = (bnxt_qplib_base_pg_size(&rq->hwq) <<
910 			     CMDQ_CREATE_QP1_RQ_PG_SIZE_SFT);
911 		pg_sz_lvl |= (rq->hwq.level & CMDQ_CREATE_QP1_RQ_LVL_MASK);
912 		req.rq_pg_size_rq_lvl = pg_sz_lvl;
913 		req.rq_fwo_rq_sge =
914 			cpu_to_le16((rq->max_sge &
915 				     CMDQ_CREATE_QP1_RQ_SGE_MASK) <<
916 				    CMDQ_CREATE_QP1_RQ_SGE_SFT);
917 	}
918 	req.rcq_cid = cpu_to_le32(qp->rcq->id);
919 	/* Header buffer - allow hdr_buf pass in */
920 	rc = bnxt_qplib_alloc_qp_hdr_buf(res, qp);
921 	if (rc) {
922 		rc = -ENOMEM;
923 		goto rq_rwq;
924 	}
925 	qp_flags |= CMDQ_CREATE_QP1_QP_FLAGS_RESERVED_LKEY_ENABLE;
926 	req.qp_flags = cpu_to_le32(qp_flags);
927 	req.pd_id = cpu_to_le32(qp->pd->id);
928 
929 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req), sizeof(resp), 0);
930 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
931 	if (rc)
932 		goto fail;
933 
934 	qp->id = le32_to_cpu(resp.xid);
935 	qp->cur_qp_state = CMDQ_MODIFY_QP_NEW_STATE_RESET;
936 	qp->cctx = res->cctx;
937 	sq->dbinfo.hwq = &sq->hwq;
938 	sq->dbinfo.xid = qp->id;
939 	sq->dbinfo.db = qp->dpi->dbr;
940 	sq->dbinfo.max_slot = bnxt_qplib_set_sq_max_slot(qp->wqe_mode);
941 	if (rq->max_wqe) {
942 		rq->dbinfo.hwq = &rq->hwq;
943 		rq->dbinfo.xid = qp->id;
944 		rq->dbinfo.db = qp->dpi->dbr;
945 		rq->dbinfo.max_slot = bnxt_qplib_set_rq_max_slot(rq->wqe_size);
946 	}
947 	tbl_indx = map_qp_id_to_tbl_indx(qp->id, rcfw);
948 	rcfw->qp_tbl[tbl_indx].qp_id = qp->id;
949 	rcfw->qp_tbl[tbl_indx].qp_handle = (void *)qp;
950 
951 	return 0;
952 
953 fail:
954 	bnxt_qplib_free_qp_hdr_buf(res, qp);
955 rq_rwq:
956 	kfree(rq->swq);
957 fail_rq:
958 	bnxt_qplib_free_hwq(res, &rq->hwq);
959 sq_swq:
960 	kfree(sq->swq);
961 fail_sq:
962 	bnxt_qplib_free_hwq(res, &sq->hwq);
963 	return rc;
964 }
965 
966 static void bnxt_qplib_init_psn_ptr(struct bnxt_qplib_qp *qp, int size)
967 {
968 	struct bnxt_qplib_hwq *hwq;
969 	struct bnxt_qplib_q *sq;
970 	u64 fpsne, psn_pg;
971 	u16 indx_pad = 0;
972 
973 	sq = &qp->sq;
974 	hwq = &sq->hwq;
975 	/* First psn entry */
976 	fpsne = (u64)bnxt_qplib_get_qe(hwq, hwq->depth, &psn_pg);
977 	if (!IS_ALIGNED(fpsne, PAGE_SIZE))
978 		indx_pad = (fpsne & ~PAGE_MASK) / size;
979 	hwq->pad_pgofft = indx_pad;
980 	hwq->pad_pg = (u64 *)psn_pg;
981 	hwq->pad_stride = size;
982 }
983 
984 int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
985 {
986 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
987 	struct bnxt_qplib_hwq_attr hwq_attr = {};
988 	struct bnxt_qplib_sg_info sginfo = {};
989 	struct creq_create_qp_resp resp = {};
990 	struct bnxt_qplib_cmdqmsg msg = {};
991 	struct bnxt_qplib_q *sq = &qp->sq;
992 	struct bnxt_qplib_q *rq = &qp->rq;
993 	struct cmdq_create_qp req = {};
994 	int rc, req_size, psn_sz = 0;
995 	struct bnxt_qplib_hwq *xrrq;
996 	struct bnxt_qplib_pbl *pbl;
997 	u32 qp_flags = 0;
998 	u8 pg_sz_lvl;
999 	u32 tbl_indx;
1000 	u16 nsge;
1001 
1002 	if (res->dattr)
1003 		qp->is_host_msn_tbl = _is_host_msn_table(res->dattr->dev_cap_flags2);
1004 
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 		hwq_attr.aux_depth = roundup_pow_of_two(bnxt_qplib_set_sq_size(sq, qp->wqe_mode));
1037 		qp->msn_tbl_sz = hwq_attr.aux_depth;
1038 		qp->msn = 0;
1039 	}
1040 
1041 	hwq_attr.type = HWQ_TYPE_QUEUE;
1042 	rc = bnxt_qplib_alloc_init_hwq(&sq->hwq, &hwq_attr);
1043 	if (rc)
1044 		return rc;
1045 
1046 	rc = bnxt_qplib_alloc_init_swq(sq);
1047 	if (rc)
1048 		goto fail_sq;
1049 
1050 	if (psn_sz)
1051 		bnxt_qplib_init_psn_ptr(qp, psn_sz);
1052 
1053 	req.sq_size = cpu_to_le32(bnxt_qplib_set_sq_size(sq, qp->wqe_mode));
1054 	pbl = &sq->hwq.pbl[PBL_LVL_0];
1055 	req.sq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
1056 	pg_sz_lvl = (bnxt_qplib_base_pg_size(&sq->hwq) <<
1057 		     CMDQ_CREATE_QP_SQ_PG_SIZE_SFT);
1058 	pg_sz_lvl |= (sq->hwq.level & CMDQ_CREATE_QP_SQ_LVL_MASK);
1059 	req.sq_pg_size_sq_lvl = pg_sz_lvl;
1060 	req.sq_fwo_sq_sge =
1061 		cpu_to_le16(((sq->max_sge & CMDQ_CREATE_QP_SQ_SGE_MASK) <<
1062 			     CMDQ_CREATE_QP_SQ_SGE_SFT) | 0);
1063 	req.scq_cid = cpu_to_le32(qp->scq->id);
1064 
1065 	/* RQ */
1066 	if (!qp->srq) {
1067 		rq->dbinfo.flags = 0;
1068 		hwq_attr.res = res;
1069 		hwq_attr.sginfo = &rq->sg_info;
1070 		hwq_attr.stride = sizeof(struct sq_sge);
1071 		hwq_attr.depth = bnxt_qplib_get_depth(rq, qp->wqe_mode, false);
1072 		hwq_attr.aux_stride = 0;
1073 		hwq_attr.aux_depth = 0;
1074 		hwq_attr.type = HWQ_TYPE_QUEUE;
1075 		rc = bnxt_qplib_alloc_init_hwq(&rq->hwq, &hwq_attr);
1076 		if (rc)
1077 			goto sq_swq;
1078 		rc = bnxt_qplib_alloc_init_swq(rq);
1079 		if (rc)
1080 			goto fail_rq;
1081 
1082 		req.rq_size = cpu_to_le32(rq->max_wqe);
1083 		pbl = &rq->hwq.pbl[PBL_LVL_0];
1084 		req.rq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
1085 		pg_sz_lvl = (bnxt_qplib_base_pg_size(&rq->hwq) <<
1086 			     CMDQ_CREATE_QP_RQ_PG_SIZE_SFT);
1087 		pg_sz_lvl |= (rq->hwq.level & CMDQ_CREATE_QP_RQ_LVL_MASK);
1088 		req.rq_pg_size_rq_lvl = pg_sz_lvl;
1089 		nsge = (qp->wqe_mode == BNXT_QPLIB_WQE_MODE_STATIC) ?
1090 			6 : rq->max_sge;
1091 		req.rq_fwo_rq_sge =
1092 			cpu_to_le16(((nsge &
1093 				      CMDQ_CREATE_QP_RQ_SGE_MASK) <<
1094 				     CMDQ_CREATE_QP_RQ_SGE_SFT) | 0);
1095 	} else {
1096 		/* SRQ */
1097 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_SRQ_USED;
1098 		req.srq_cid = cpu_to_le32(qp->srq->id);
1099 	}
1100 	req.rcq_cid = cpu_to_le32(qp->rcq->id);
1101 
1102 	qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_RESERVED_LKEY_ENABLE;
1103 	qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_FR_PMR_ENABLED;
1104 	if (qp->sig_type)
1105 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_FORCE_COMPLETION;
1106 	if (qp->wqe_mode == BNXT_QPLIB_WQE_MODE_VARIABLE)
1107 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_VARIABLE_SIZED_WQE_ENABLED;
1108 	if (_is_ext_stats_supported(res->dattr->dev_cap_flags) && !res->is_vf)
1109 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_EXT_STATS_ENABLED;
1110 
1111 	req.qp_flags = cpu_to_le32(qp_flags);
1112 
1113 	/* ORRQ and IRRQ */
1114 	if (psn_sz) {
1115 		xrrq = &qp->orrq;
1116 		xrrq->max_elements =
1117 			ORD_LIMIT_TO_ORRQ_SLOTS(qp->max_rd_atomic);
1118 		req_size = xrrq->max_elements *
1119 			   BNXT_QPLIB_MAX_ORRQE_ENTRY_SIZE + PAGE_SIZE - 1;
1120 		req_size &= ~(PAGE_SIZE - 1);
1121 		sginfo.pgsize = req_size;
1122 		sginfo.pgshft = PAGE_SHIFT;
1123 
1124 		hwq_attr.res = res;
1125 		hwq_attr.sginfo = &sginfo;
1126 		hwq_attr.depth = xrrq->max_elements;
1127 		hwq_attr.stride = BNXT_QPLIB_MAX_ORRQE_ENTRY_SIZE;
1128 		hwq_attr.aux_stride = 0;
1129 		hwq_attr.aux_depth = 0;
1130 		hwq_attr.type = HWQ_TYPE_CTX;
1131 		rc = bnxt_qplib_alloc_init_hwq(xrrq, &hwq_attr);
1132 		if (rc)
1133 			goto rq_swq;
1134 		pbl = &xrrq->pbl[PBL_LVL_0];
1135 		req.orrq_addr = cpu_to_le64(pbl->pg_map_arr[0]);
1136 
1137 		xrrq = &qp->irrq;
1138 		xrrq->max_elements = IRD_LIMIT_TO_IRRQ_SLOTS(
1139 						qp->max_dest_rd_atomic);
1140 		req_size = xrrq->max_elements *
1141 			   BNXT_QPLIB_MAX_IRRQE_ENTRY_SIZE + PAGE_SIZE - 1;
1142 		req_size &= ~(PAGE_SIZE - 1);
1143 		sginfo.pgsize = req_size;
1144 		hwq_attr.depth =  xrrq->max_elements;
1145 		hwq_attr.stride = BNXT_QPLIB_MAX_IRRQE_ENTRY_SIZE;
1146 		rc = bnxt_qplib_alloc_init_hwq(xrrq, &hwq_attr);
1147 		if (rc)
1148 			goto fail_orrq;
1149 
1150 		pbl = &xrrq->pbl[PBL_LVL_0];
1151 		req.irrq_addr = cpu_to_le64(pbl->pg_map_arr[0]);
1152 	}
1153 	req.pd_id = cpu_to_le32(qp->pd->id);
1154 
1155 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
1156 				sizeof(resp), 0);
1157 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
1158 	if (rc)
1159 		goto fail;
1160 
1161 	qp->id = le32_to_cpu(resp.xid);
1162 	qp->cur_qp_state = CMDQ_MODIFY_QP_NEW_STATE_RESET;
1163 	INIT_LIST_HEAD(&qp->sq_flush);
1164 	INIT_LIST_HEAD(&qp->rq_flush);
1165 	qp->cctx = res->cctx;
1166 	sq->dbinfo.hwq = &sq->hwq;
1167 	sq->dbinfo.xid = qp->id;
1168 	sq->dbinfo.db = qp->dpi->dbr;
1169 	sq->dbinfo.max_slot = bnxt_qplib_set_sq_max_slot(qp->wqe_mode);
1170 	if (rq->max_wqe) {
1171 		rq->dbinfo.hwq = &rq->hwq;
1172 		rq->dbinfo.xid = qp->id;
1173 		rq->dbinfo.db = qp->dpi->dbr;
1174 		rq->dbinfo.max_slot = bnxt_qplib_set_rq_max_slot(rq->wqe_size);
1175 	}
1176 	tbl_indx = map_qp_id_to_tbl_indx(qp->id, rcfw);
1177 	rcfw->qp_tbl[tbl_indx].qp_id = qp->id;
1178 	rcfw->qp_tbl[tbl_indx].qp_handle = (void *)qp;
1179 
1180 	return 0;
1181 fail:
1182 	bnxt_qplib_free_hwq(res, &qp->irrq);
1183 fail_orrq:
1184 	bnxt_qplib_free_hwq(res, &qp->orrq);
1185 rq_swq:
1186 	kfree(rq->swq);
1187 fail_rq:
1188 	bnxt_qplib_free_hwq(res, &rq->hwq);
1189 sq_swq:
1190 	kfree(sq->swq);
1191 fail_sq:
1192 	bnxt_qplib_free_hwq(res, &sq->hwq);
1193 	return rc;
1194 }
1195 
1196 static void __modify_flags_from_init_state(struct bnxt_qplib_qp *qp)
1197 {
1198 	switch (qp->state) {
1199 	case CMDQ_MODIFY_QP_NEW_STATE_RTR:
1200 		/* INIT->RTR, configure the path_mtu to the default
1201 		 * 2048 if not being requested
1202 		 */
1203 		if (!(qp->modify_flags &
1204 		    CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU)) {
1205 			qp->modify_flags |=
1206 				CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU;
1207 			qp->path_mtu =
1208 				CMDQ_MODIFY_QP_PATH_MTU_MTU_2048;
1209 		}
1210 		qp->modify_flags &=
1211 			~CMDQ_MODIFY_QP_MODIFY_MASK_VLAN_ID;
1212 		/* Bono FW require the max_dest_rd_atomic to be >= 1 */
1213 		if (qp->max_dest_rd_atomic < 1)
1214 			qp->max_dest_rd_atomic = 1;
1215 		qp->modify_flags &= ~CMDQ_MODIFY_QP_MODIFY_MASK_SRC_MAC;
1216 		/* Bono FW 20.6.5 requires SGID_INDEX configuration */
1217 		if (!(qp->modify_flags &
1218 		    CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX)) {
1219 			qp->modify_flags |=
1220 				CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX;
1221 			qp->ah.sgid_index = 0;
1222 		}
1223 		break;
1224 	default:
1225 		break;
1226 	}
1227 }
1228 
1229 static void __modify_flags_from_rtr_state(struct bnxt_qplib_qp *qp)
1230 {
1231 	switch (qp->state) {
1232 	case CMDQ_MODIFY_QP_NEW_STATE_RTS:
1233 		/* Bono FW requires the max_rd_atomic to be >= 1 */
1234 		if (qp->max_rd_atomic < 1)
1235 			qp->max_rd_atomic = 1;
1236 		/* Bono FW does not allow PKEY_INDEX,
1237 		 * DGID, FLOW_LABEL, SGID_INDEX, HOP_LIMIT,
1238 		 * TRAFFIC_CLASS, DEST_MAC, PATH_MTU, RQ_PSN,
1239 		 * MIN_RNR_TIMER, MAX_DEST_RD_ATOMIC, DEST_QP_ID
1240 		 * modification
1241 		 */
1242 		qp->modify_flags &=
1243 			~(CMDQ_MODIFY_QP_MODIFY_MASK_PKEY |
1244 			  CMDQ_MODIFY_QP_MODIFY_MASK_DGID |
1245 			  CMDQ_MODIFY_QP_MODIFY_MASK_FLOW_LABEL |
1246 			  CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX |
1247 			  CMDQ_MODIFY_QP_MODIFY_MASK_HOP_LIMIT |
1248 			  CMDQ_MODIFY_QP_MODIFY_MASK_TRAFFIC_CLASS |
1249 			  CMDQ_MODIFY_QP_MODIFY_MASK_DEST_MAC |
1250 			  CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU |
1251 			  CMDQ_MODIFY_QP_MODIFY_MASK_RQ_PSN |
1252 			  CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER |
1253 			  CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC |
1254 			  CMDQ_MODIFY_QP_MODIFY_MASK_DEST_QP_ID);
1255 		break;
1256 	default:
1257 		break;
1258 	}
1259 }
1260 
1261 static void __filter_modify_flags(struct bnxt_qplib_qp *qp)
1262 {
1263 	switch (qp->cur_qp_state) {
1264 	case CMDQ_MODIFY_QP_NEW_STATE_RESET:
1265 		break;
1266 	case CMDQ_MODIFY_QP_NEW_STATE_INIT:
1267 		__modify_flags_from_init_state(qp);
1268 		break;
1269 	case CMDQ_MODIFY_QP_NEW_STATE_RTR:
1270 		__modify_flags_from_rtr_state(qp);
1271 		break;
1272 	case CMDQ_MODIFY_QP_NEW_STATE_RTS:
1273 		break;
1274 	case CMDQ_MODIFY_QP_NEW_STATE_SQD:
1275 		break;
1276 	case CMDQ_MODIFY_QP_NEW_STATE_SQE:
1277 		break;
1278 	case CMDQ_MODIFY_QP_NEW_STATE_ERR:
1279 		break;
1280 	default:
1281 		break;
1282 	}
1283 }
1284 
1285 int bnxt_qplib_modify_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
1286 {
1287 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1288 	struct creq_modify_qp_resp resp = {};
1289 	struct bnxt_qplib_cmdqmsg msg = {};
1290 	struct cmdq_modify_qp req = {};
1291 	u32 temp32[4];
1292 	u32 bmask;
1293 	int rc;
1294 
1295 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
1296 				 CMDQ_BASE_OPCODE_MODIFY_QP,
1297 				 sizeof(req));
1298 
1299 	/* Filter out the qp_attr_mask based on the state->new transition */
1300 	__filter_modify_flags(qp);
1301 	bmask = qp->modify_flags;
1302 	req.modify_mask = cpu_to_le32(qp->modify_flags);
1303 	req.qp_cid = cpu_to_le32(qp->id);
1304 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_STATE) {
1305 		req.network_type_en_sqd_async_notify_new_state =
1306 				(qp->state & CMDQ_MODIFY_QP_NEW_STATE_MASK) |
1307 				(qp->en_sqd_async_notify ?
1308 					CMDQ_MODIFY_QP_EN_SQD_ASYNC_NOTIFY : 0);
1309 	}
1310 	req.network_type_en_sqd_async_notify_new_state |= qp->nw_type;
1311 
1312 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_ACCESS)
1313 		req.access = qp->access;
1314 
1315 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_PKEY)
1316 		req.pkey = cpu_to_le16(IB_DEFAULT_PKEY_FULL);
1317 
1318 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_QKEY)
1319 		req.qkey = cpu_to_le32(qp->qkey);
1320 
1321 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DGID) {
1322 		memcpy(temp32, qp->ah.dgid.data, sizeof(struct bnxt_qplib_gid));
1323 		req.dgid[0] = cpu_to_le32(temp32[0]);
1324 		req.dgid[1] = cpu_to_le32(temp32[1]);
1325 		req.dgid[2] = cpu_to_le32(temp32[2]);
1326 		req.dgid[3] = cpu_to_le32(temp32[3]);
1327 	}
1328 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_FLOW_LABEL)
1329 		req.flow_label = cpu_to_le32(qp->ah.flow_label);
1330 
1331 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX)
1332 		req.sgid_index = cpu_to_le16(res->sgid_tbl.hw_id
1333 					     [qp->ah.sgid_index]);
1334 
1335 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_HOP_LIMIT)
1336 		req.hop_limit = qp->ah.hop_limit;
1337 
1338 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_TRAFFIC_CLASS)
1339 		req.traffic_class = qp->ah.traffic_class;
1340 
1341 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DEST_MAC)
1342 		memcpy(req.dest_mac, qp->ah.dmac, 6);
1343 
1344 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU)
1345 		req.path_mtu_pingpong_push_enable |= qp->path_mtu;
1346 
1347 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_TIMEOUT)
1348 		req.timeout = qp->timeout;
1349 
1350 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RETRY_CNT)
1351 		req.retry_cnt = qp->retry_cnt;
1352 
1353 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RNR_RETRY)
1354 		req.rnr_retry = qp->rnr_retry;
1355 
1356 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER)
1357 		req.min_rnr_timer = qp->min_rnr_timer;
1358 
1359 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RQ_PSN)
1360 		req.rq_psn = cpu_to_le32(qp->rq.psn);
1361 
1362 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_SQ_PSN)
1363 		req.sq_psn = cpu_to_le32(qp->sq.psn);
1364 
1365 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MAX_RD_ATOMIC)
1366 		req.max_rd_atomic =
1367 			ORD_LIMIT_TO_ORRQ_SLOTS(qp->max_rd_atomic);
1368 
1369 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC)
1370 		req.max_dest_rd_atomic =
1371 			IRD_LIMIT_TO_IRRQ_SLOTS(qp->max_dest_rd_atomic);
1372 
1373 	req.sq_size = cpu_to_le32(qp->sq.hwq.max_elements);
1374 	req.rq_size = cpu_to_le32(qp->rq.hwq.max_elements);
1375 	req.sq_sge = cpu_to_le16(qp->sq.max_sge);
1376 	req.rq_sge = cpu_to_le16(qp->rq.max_sge);
1377 	req.max_inline_data = cpu_to_le32(qp->max_inline_data);
1378 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DEST_QP_ID)
1379 		req.dest_qp_id = cpu_to_le32(qp->dest_qpn);
1380 
1381 	req.vlan_pcp_vlan_dei_vlan_id = cpu_to_le16(qp->vlan_id);
1382 
1383 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),  sizeof(resp), 0);
1384 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
1385 	if (rc)
1386 		return rc;
1387 	qp->cur_qp_state = qp->state;
1388 	return 0;
1389 }
1390 
1391 int bnxt_qplib_query_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
1392 {
1393 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1394 	struct creq_query_qp_resp resp = {};
1395 	struct bnxt_qplib_cmdqmsg msg = {};
1396 	struct bnxt_qplib_rcfw_sbuf sbuf;
1397 	struct creq_query_qp_resp_sb *sb;
1398 	struct cmdq_query_qp req = {};
1399 	u32 temp32[4];
1400 	int i, rc;
1401 
1402 	sbuf.size = ALIGN(sizeof(*sb), BNXT_QPLIB_CMDQE_UNITS);
1403 	sbuf.sb = dma_alloc_coherent(&rcfw->pdev->dev, sbuf.size,
1404 				     &sbuf.dma_addr, GFP_KERNEL);
1405 	if (!sbuf.sb)
1406 		return -ENOMEM;
1407 	sb = sbuf.sb;
1408 
1409 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
1410 				 CMDQ_BASE_OPCODE_QUERY_QP,
1411 				 sizeof(req));
1412 
1413 	req.qp_cid = cpu_to_le32(qp->id);
1414 	req.resp_size = sbuf.size / BNXT_QPLIB_CMDQE_UNITS;
1415 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, &sbuf, sizeof(req),
1416 				sizeof(resp), 0);
1417 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
1418 	if (rc)
1419 		goto bail;
1420 	/* Extract the context from the side buffer */
1421 	qp->state = sb->en_sqd_async_notify_state &
1422 			CREQ_QUERY_QP_RESP_SB_STATE_MASK;
1423 	qp->en_sqd_async_notify = sb->en_sqd_async_notify_state &
1424 				  CREQ_QUERY_QP_RESP_SB_EN_SQD_ASYNC_NOTIFY;
1425 	qp->access = sb->access;
1426 	qp->pkey_index = le16_to_cpu(sb->pkey);
1427 	qp->qkey = le32_to_cpu(sb->qkey);
1428 
1429 	temp32[0] = le32_to_cpu(sb->dgid[0]);
1430 	temp32[1] = le32_to_cpu(sb->dgid[1]);
1431 	temp32[2] = le32_to_cpu(sb->dgid[2]);
1432 	temp32[3] = le32_to_cpu(sb->dgid[3]);
1433 	memcpy(qp->ah.dgid.data, temp32, sizeof(qp->ah.dgid.data));
1434 
1435 	qp->ah.flow_label = le32_to_cpu(sb->flow_label);
1436 
1437 	qp->ah.sgid_index = 0;
1438 	for (i = 0; i < res->sgid_tbl.max; i++) {
1439 		if (res->sgid_tbl.hw_id[i] == le16_to_cpu(sb->sgid_index)) {
1440 			qp->ah.sgid_index = i;
1441 			break;
1442 		}
1443 	}
1444 	if (i == res->sgid_tbl.max)
1445 		dev_warn(&res->pdev->dev, "SGID not found??\n");
1446 
1447 	qp->ah.hop_limit = sb->hop_limit;
1448 	qp->ah.traffic_class = sb->traffic_class;
1449 	memcpy(qp->ah.dmac, sb->dest_mac, 6);
1450 	qp->ah.vlan_id = (le16_to_cpu(sb->path_mtu_dest_vlan_id) &
1451 				CREQ_QUERY_QP_RESP_SB_VLAN_ID_MASK) >>
1452 				CREQ_QUERY_QP_RESP_SB_VLAN_ID_SFT;
1453 	qp->path_mtu = (le16_to_cpu(sb->path_mtu_dest_vlan_id) &
1454 				    CREQ_QUERY_QP_RESP_SB_PATH_MTU_MASK) >>
1455 				    CREQ_QUERY_QP_RESP_SB_PATH_MTU_SFT;
1456 	qp->timeout = sb->timeout;
1457 	qp->retry_cnt = sb->retry_cnt;
1458 	qp->rnr_retry = sb->rnr_retry;
1459 	qp->min_rnr_timer = sb->min_rnr_timer;
1460 	qp->rq.psn = le32_to_cpu(sb->rq_psn);
1461 	qp->max_rd_atomic = ORRQ_SLOTS_TO_ORD_LIMIT(sb->max_rd_atomic);
1462 	qp->sq.psn = le32_to_cpu(sb->sq_psn);
1463 	qp->max_dest_rd_atomic =
1464 			IRRQ_SLOTS_TO_IRD_LIMIT(sb->max_dest_rd_atomic);
1465 	qp->sq.max_wqe = qp->sq.hwq.max_elements;
1466 	qp->rq.max_wqe = qp->rq.hwq.max_elements;
1467 	qp->sq.max_sge = le16_to_cpu(sb->sq_sge);
1468 	qp->rq.max_sge = le16_to_cpu(sb->rq_sge);
1469 	qp->max_inline_data = le32_to_cpu(sb->max_inline_data);
1470 	qp->dest_qpn = le32_to_cpu(sb->dest_qp_id);
1471 	memcpy(qp->smac, sb->src_mac, 6);
1472 	qp->vlan_id = le16_to_cpu(sb->vlan_pcp_vlan_dei_vlan_id);
1473 bail:
1474 	dma_free_coherent(&rcfw->pdev->dev, sbuf.size,
1475 			  sbuf.sb, sbuf.dma_addr);
1476 	return rc;
1477 }
1478 
1479 static void __clean_cq(struct bnxt_qplib_cq *cq, u64 qp)
1480 {
1481 	struct bnxt_qplib_hwq *cq_hwq = &cq->hwq;
1482 	u32 peek_flags, peek_cons;
1483 	struct cq_base *hw_cqe;
1484 	int i;
1485 
1486 	peek_flags = cq->dbinfo.flags;
1487 	peek_cons = cq_hwq->cons;
1488 	for (i = 0; i < cq_hwq->max_elements; i++) {
1489 		hw_cqe = bnxt_qplib_get_qe(cq_hwq, peek_cons, NULL);
1490 		if (!CQE_CMP_VALID(hw_cqe, peek_flags))
1491 			continue;
1492 		/*
1493 		 * The valid test of the entry must be done first before
1494 		 * reading any further.
1495 		 */
1496 		dma_rmb();
1497 		switch (hw_cqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK) {
1498 		case CQ_BASE_CQE_TYPE_REQ:
1499 		case CQ_BASE_CQE_TYPE_TERMINAL:
1500 		{
1501 			struct cq_req *cqe = (struct cq_req *)hw_cqe;
1502 
1503 			if (qp == le64_to_cpu(cqe->qp_handle))
1504 				cqe->qp_handle = 0;
1505 			break;
1506 		}
1507 		case CQ_BASE_CQE_TYPE_RES_RC:
1508 		case CQ_BASE_CQE_TYPE_RES_UD:
1509 		case CQ_BASE_CQE_TYPE_RES_RAWETH_QP1:
1510 		{
1511 			struct cq_res_rc *cqe = (struct cq_res_rc *)hw_cqe;
1512 
1513 			if (qp == le64_to_cpu(cqe->qp_handle))
1514 				cqe->qp_handle = 0;
1515 			break;
1516 		}
1517 		default:
1518 			break;
1519 		}
1520 		bnxt_qplib_hwq_incr_cons(cq_hwq->max_elements, &peek_cons,
1521 					 1, &peek_flags);
1522 	}
1523 }
1524 
1525 int bnxt_qplib_destroy_qp(struct bnxt_qplib_res *res,
1526 			  struct bnxt_qplib_qp *qp)
1527 {
1528 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1529 	struct creq_destroy_qp_resp resp = {};
1530 	struct bnxt_qplib_cmdqmsg msg = {};
1531 	struct cmdq_destroy_qp req = {};
1532 	u32 tbl_indx;
1533 	int rc;
1534 
1535 	spin_lock_bh(&rcfw->tbl_lock);
1536 	tbl_indx = map_qp_id_to_tbl_indx(qp->id, rcfw);
1537 	rcfw->qp_tbl[tbl_indx].qp_id = BNXT_QPLIB_QP_ID_INVALID;
1538 	rcfw->qp_tbl[tbl_indx].qp_handle = NULL;
1539 	spin_unlock_bh(&rcfw->tbl_lock);
1540 
1541 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
1542 				 CMDQ_BASE_OPCODE_DESTROY_QP,
1543 				 sizeof(req));
1544 
1545 	req.qp_cid = cpu_to_le32(qp->id);
1546 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
1547 				sizeof(resp), 0);
1548 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
1549 	if (rc) {
1550 		spin_lock_bh(&rcfw->tbl_lock);
1551 		rcfw->qp_tbl[tbl_indx].qp_id = qp->id;
1552 		rcfw->qp_tbl[tbl_indx].qp_handle = qp;
1553 		spin_unlock_bh(&rcfw->tbl_lock);
1554 		return rc;
1555 	}
1556 
1557 	return 0;
1558 }
1559 
1560 void bnxt_qplib_free_qp_res(struct bnxt_qplib_res *res,
1561 			    struct bnxt_qplib_qp *qp)
1562 {
1563 	bnxt_qplib_free_qp_hdr_buf(res, qp);
1564 	bnxt_qplib_free_hwq(res, &qp->sq.hwq);
1565 	kfree(qp->sq.swq);
1566 
1567 	bnxt_qplib_free_hwq(res, &qp->rq.hwq);
1568 	kfree(qp->rq.swq);
1569 
1570 	if (qp->irrq.max_elements)
1571 		bnxt_qplib_free_hwq(res, &qp->irrq);
1572 	if (qp->orrq.max_elements)
1573 		bnxt_qplib_free_hwq(res, &qp->orrq);
1574 
1575 }
1576 
1577 void *bnxt_qplib_get_qp1_sq_buf(struct bnxt_qplib_qp *qp,
1578 				struct bnxt_qplib_sge *sge)
1579 {
1580 	struct bnxt_qplib_q *sq = &qp->sq;
1581 	u32 sw_prod;
1582 
1583 	memset(sge, 0, sizeof(*sge));
1584 
1585 	if (qp->sq_hdr_buf) {
1586 		sw_prod = sq->swq_start;
1587 		sge->addr = (dma_addr_t)(qp->sq_hdr_buf_map +
1588 					 sw_prod * qp->sq_hdr_buf_size);
1589 		sge->lkey = 0xFFFFFFFF;
1590 		sge->size = qp->sq_hdr_buf_size;
1591 		return qp->sq_hdr_buf + sw_prod * sge->size;
1592 	}
1593 	return NULL;
1594 }
1595 
1596 u32 bnxt_qplib_get_rq_prod_index(struct bnxt_qplib_qp *qp)
1597 {
1598 	struct bnxt_qplib_q *rq = &qp->rq;
1599 
1600 	return rq->swq_start;
1601 }
1602 
1603 dma_addr_t bnxt_qplib_get_qp_buf_from_index(struct bnxt_qplib_qp *qp, u32 index)
1604 {
1605 	return (qp->rq_hdr_buf_map + index * qp->rq_hdr_buf_size);
1606 }
1607 
1608 void *bnxt_qplib_get_qp1_rq_buf(struct bnxt_qplib_qp *qp,
1609 				struct bnxt_qplib_sge *sge)
1610 {
1611 	struct bnxt_qplib_q *rq = &qp->rq;
1612 	u32 sw_prod;
1613 
1614 	memset(sge, 0, sizeof(*sge));
1615 
1616 	if (qp->rq_hdr_buf) {
1617 		sw_prod = rq->swq_start;
1618 		sge->addr = (dma_addr_t)(qp->rq_hdr_buf_map +
1619 					 sw_prod * qp->rq_hdr_buf_size);
1620 		sge->lkey = 0xFFFFFFFF;
1621 		sge->size = qp->rq_hdr_buf_size;
1622 		return qp->rq_hdr_buf + sw_prod * sge->size;
1623 	}
1624 	return NULL;
1625 }
1626 
1627 /* Fil the MSN table into the next psn row */
1628 static void bnxt_qplib_fill_msn_search(struct bnxt_qplib_qp *qp,
1629 				       struct bnxt_qplib_swqe *wqe,
1630 				       struct bnxt_qplib_swq *swq)
1631 {
1632 	struct sq_msn_search *msns;
1633 	u32 start_psn, next_psn;
1634 	u16 start_idx;
1635 
1636 	msns = (struct sq_msn_search *)swq->psn_search;
1637 	msns->start_idx_next_psn_start_psn = 0;
1638 
1639 	start_psn = swq->start_psn;
1640 	next_psn = swq->next_psn;
1641 	start_idx = swq->slot_idx;
1642 	msns->start_idx_next_psn_start_psn |=
1643 		bnxt_re_update_msn_tbl(start_idx, next_psn, start_psn);
1644 	qp->msn++;
1645 	qp->msn %= qp->msn_tbl_sz;
1646 }
1647 
1648 static void bnxt_qplib_fill_psn_search(struct bnxt_qplib_qp *qp,
1649 				       struct bnxt_qplib_swqe *wqe,
1650 				       struct bnxt_qplib_swq *swq)
1651 {
1652 	struct sq_psn_search_ext *psns_ext;
1653 	struct sq_psn_search *psns;
1654 	u32 flg_npsn;
1655 	u32 op_spsn;
1656 
1657 	if (!swq->psn_search)
1658 		return;
1659 	/* Handle MSN differently on cap flags  */
1660 	if (qp->is_host_msn_tbl) {
1661 		bnxt_qplib_fill_msn_search(qp, wqe, swq);
1662 		return;
1663 	}
1664 	psns = (struct sq_psn_search *)swq->psn_search;
1665 	psns = swq->psn_search;
1666 	psns_ext = swq->psn_ext;
1667 
1668 	op_spsn = ((swq->start_psn << SQ_PSN_SEARCH_START_PSN_SFT) &
1669 		    SQ_PSN_SEARCH_START_PSN_MASK);
1670 	op_spsn |= ((wqe->type << SQ_PSN_SEARCH_OPCODE_SFT) &
1671 		     SQ_PSN_SEARCH_OPCODE_MASK);
1672 	flg_npsn = ((swq->next_psn << SQ_PSN_SEARCH_NEXT_PSN_SFT) &
1673 		     SQ_PSN_SEARCH_NEXT_PSN_MASK);
1674 
1675 	if (bnxt_qplib_is_chip_gen_p5_p7(qp->cctx)) {
1676 		psns_ext->opcode_start_psn = cpu_to_le32(op_spsn);
1677 		psns_ext->flags_next_psn = cpu_to_le32(flg_npsn);
1678 		psns_ext->start_slot_idx = cpu_to_le16(swq->slot_idx);
1679 	} else {
1680 		psns->opcode_start_psn = cpu_to_le32(op_spsn);
1681 		psns->flags_next_psn = cpu_to_le32(flg_npsn);
1682 	}
1683 }
1684 
1685 static int bnxt_qplib_put_inline(struct bnxt_qplib_qp *qp,
1686 				 struct bnxt_qplib_swqe *wqe,
1687 				 u16 *idx)
1688 {
1689 	struct bnxt_qplib_hwq *hwq;
1690 	int len, t_len, offt;
1691 	bool pull_dst = true;
1692 	void *il_dst = NULL;
1693 	void *il_src = NULL;
1694 	int t_cplen, cplen;
1695 	int indx;
1696 
1697 	hwq = &qp->sq.hwq;
1698 	t_len = 0;
1699 	for (indx = 0; indx < wqe->num_sge; indx++) {
1700 		len = wqe->sg_list[indx].size;
1701 		il_src = (void *)wqe->sg_list[indx].addr;
1702 		t_len += len;
1703 		if (t_len > qp->max_inline_data)
1704 			return -ENOMEM;
1705 		while (len) {
1706 			if (pull_dst) {
1707 				pull_dst = false;
1708 				il_dst = bnxt_qplib_get_prod_qe(hwq, *idx);
1709 				(*idx)++;
1710 				t_cplen = 0;
1711 				offt = 0;
1712 			}
1713 			cplen = min_t(int, len, sizeof(struct sq_sge));
1714 			cplen = min_t(int, cplen,
1715 					(sizeof(struct sq_sge) - offt));
1716 			memcpy(il_dst, il_src, cplen);
1717 			t_cplen += cplen;
1718 			il_src += cplen;
1719 			il_dst += cplen;
1720 			offt += cplen;
1721 			len -= cplen;
1722 			if (t_cplen == sizeof(struct sq_sge))
1723 				pull_dst = true;
1724 		}
1725 	}
1726 
1727 	return t_len;
1728 }
1729 
1730 static u32 bnxt_qplib_put_sges(struct bnxt_qplib_hwq *hwq,
1731 			       struct bnxt_qplib_sge *ssge,
1732 			       u16 nsge, u16 *idx)
1733 {
1734 	struct sq_sge *dsge;
1735 	int indx, len = 0;
1736 
1737 	for (indx = 0; indx < nsge; indx++, (*idx)++) {
1738 		dsge = bnxt_qplib_get_prod_qe(hwq, *idx);
1739 		dsge->va_or_pa = cpu_to_le64(ssge[indx].addr);
1740 		dsge->l_key = cpu_to_le32(ssge[indx].lkey);
1741 		dsge->size = cpu_to_le32(ssge[indx].size);
1742 		len += ssge[indx].size;
1743 	}
1744 
1745 	return len;
1746 }
1747 
1748 static u16 bnxt_qplib_required_slots(struct bnxt_qplib_qp *qp,
1749 				     struct bnxt_qplib_swqe *wqe,
1750 				     u16 *wqe_sz, u16 *qdf, u8 mode)
1751 {
1752 	u32 ilsize, bytes;
1753 	u16 nsge;
1754 	u16 slot;
1755 
1756 	nsge = wqe->num_sge;
1757 	/* Adding sq_send_hdr is a misnomer, for rq also hdr size is same. */
1758 	bytes = sizeof(struct sq_send_hdr) + nsge * sizeof(struct sq_sge);
1759 	if (wqe->flags & BNXT_QPLIB_SWQE_FLAGS_INLINE) {
1760 		ilsize = bnxt_qplib_calc_ilsize(wqe, qp->max_inline_data);
1761 		bytes = ALIGN(ilsize, sizeof(struct sq_sge));
1762 		bytes += sizeof(struct sq_send_hdr);
1763 	}
1764 
1765 	*qdf =  __xlate_qfd(qp->sq.q_full_delta, bytes);
1766 	slot = bytes >> 4;
1767 	*wqe_sz = slot;
1768 	if (mode == BNXT_QPLIB_WQE_MODE_STATIC)
1769 		slot = 8;
1770 	return slot;
1771 }
1772 
1773 static void bnxt_qplib_pull_psn_buff(struct bnxt_qplib_qp *qp, struct bnxt_qplib_q *sq,
1774 				     struct bnxt_qplib_swq *swq, bool hw_retx)
1775 {
1776 	struct bnxt_qplib_hwq *hwq;
1777 	u32 pg_num, pg_indx;
1778 	void *buff;
1779 	u32 tail;
1780 
1781 	hwq = &sq->hwq;
1782 	if (!hwq->pad_pg)
1783 		return;
1784 	tail = swq->slot_idx / sq->dbinfo.max_slot;
1785 	if (hw_retx) {
1786 		/* For HW retx use qp msn index */
1787 		tail = qp->msn;
1788 		tail %= qp->msn_tbl_sz;
1789 	}
1790 	pg_num = (tail + hwq->pad_pgofft) / (PAGE_SIZE / hwq->pad_stride);
1791 	pg_indx = (tail + hwq->pad_pgofft) % (PAGE_SIZE / hwq->pad_stride);
1792 	buff = (void *)(hwq->pad_pg[pg_num] + pg_indx * hwq->pad_stride);
1793 	swq->psn_ext = buff;
1794 	swq->psn_search = buff;
1795 }
1796 
1797 void bnxt_qplib_post_send_db(struct bnxt_qplib_qp *qp)
1798 {
1799 	struct bnxt_qplib_q *sq = &qp->sq;
1800 
1801 	bnxt_qplib_ring_prod_db(&sq->dbinfo, DBC_DBC_TYPE_SQ);
1802 }
1803 
1804 int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp,
1805 			 struct bnxt_qplib_swqe *wqe)
1806 {
1807 	struct bnxt_qplib_nq_work *nq_work = NULL;
1808 	int i, rc = 0, data_len = 0, pkt_num = 0;
1809 	struct bnxt_qplib_q *sq = &qp->sq;
1810 	struct bnxt_qplib_hwq *hwq;
1811 	struct bnxt_qplib_swq *swq;
1812 	bool sch_handler = false;
1813 	u16 wqe_sz, qdf = 0;
1814 	bool msn_update;
1815 	void *base_hdr;
1816 	void *ext_hdr;
1817 	__le32 temp32;
1818 	u32 wqe_idx;
1819 	u32 slots;
1820 	u16 idx;
1821 
1822 	hwq = &sq->hwq;
1823 	if (qp->state != CMDQ_MODIFY_QP_NEW_STATE_RTS &&
1824 	    qp->state != CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1825 		dev_err(&hwq->pdev->dev,
1826 			"QPLIB: FP: QP (0x%x) is in the 0x%x state",
1827 			qp->id, qp->state);
1828 		rc = -EINVAL;
1829 		goto done;
1830 	}
1831 
1832 	slots = bnxt_qplib_required_slots(qp, wqe, &wqe_sz, &qdf, qp->wqe_mode);
1833 	if (bnxt_qplib_queue_full(sq, slots + qdf)) {
1834 		dev_err(&hwq->pdev->dev,
1835 			"prod = %#x cons = %#x qdepth = %#x delta = %#x\n",
1836 			hwq->prod, hwq->cons, hwq->depth, sq->q_full_delta);
1837 		rc = -ENOMEM;
1838 		goto done;
1839 	}
1840 
1841 	swq = bnxt_qplib_get_swqe(sq, &wqe_idx);
1842 	bnxt_qplib_pull_psn_buff(qp, sq, swq, qp->is_host_msn_tbl);
1843 
1844 	idx = 0;
1845 	swq->slot_idx = hwq->prod;
1846 	swq->slots = slots;
1847 	swq->wr_id = wqe->wr_id;
1848 	swq->type = wqe->type;
1849 	swq->flags = wqe->flags;
1850 	swq->start_psn = sq->psn & BTH_PSN_MASK;
1851 	if (qp->sig_type)
1852 		swq->flags |= SQ_SEND_FLAGS_SIGNAL_COMP;
1853 
1854 	if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1855 		sch_handler = true;
1856 		dev_dbg(&hwq->pdev->dev,
1857 			"%s Error QP. Scheduling for poll_cq\n", __func__);
1858 		goto queue_err;
1859 	}
1860 
1861 	base_hdr = bnxt_qplib_get_prod_qe(hwq, idx++);
1862 	ext_hdr = bnxt_qplib_get_prod_qe(hwq, idx++);
1863 	memset(base_hdr, 0, sizeof(struct sq_sge));
1864 	memset(ext_hdr, 0, sizeof(struct sq_sge));
1865 
1866 	if (wqe->flags & BNXT_QPLIB_SWQE_FLAGS_INLINE)
1867 		/* Copy the inline data */
1868 		data_len = bnxt_qplib_put_inline(qp, wqe, &idx);
1869 	else
1870 		data_len = bnxt_qplib_put_sges(hwq, wqe->sg_list, wqe->num_sge,
1871 					       &idx);
1872 	if (data_len < 0)
1873 		goto queue_err;
1874 	/* Make sure we update MSN table only for wired wqes */
1875 	msn_update = true;
1876 	/* Specifics */
1877 	switch (wqe->type) {
1878 	case BNXT_QPLIB_SWQE_TYPE_SEND:
1879 		if (qp->type == CMDQ_CREATE_QP1_TYPE_GSI) {
1880 			struct sq_send_raweth_qp1_hdr *sqe = base_hdr;
1881 			struct sq_raw_ext_hdr *ext_sqe = ext_hdr;
1882 			/* Assemble info for Raw Ethertype QPs */
1883 
1884 			sqe->wqe_type = wqe->type;
1885 			sqe->flags = wqe->flags;
1886 			sqe->wqe_size = wqe_sz;
1887 			sqe->cfa_action = cpu_to_le16(wqe->rawqp1.cfa_action);
1888 			sqe->lflags = cpu_to_le16(wqe->rawqp1.lflags);
1889 			sqe->length = cpu_to_le32(data_len);
1890 			ext_sqe->cfa_meta = cpu_to_le32((wqe->rawqp1.cfa_meta &
1891 				SQ_SEND_RAWETH_QP1_CFA_META_VLAN_VID_MASK) <<
1892 				SQ_SEND_RAWETH_QP1_CFA_META_VLAN_VID_SFT);
1893 
1894 			break;
1895 		}
1896 		fallthrough;
1897 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_IMM:
1898 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_INV:
1899 	{
1900 		struct sq_ud_ext_hdr *ext_sqe = ext_hdr;
1901 		struct sq_send_hdr *sqe = base_hdr;
1902 
1903 		sqe->wqe_type = wqe->type;
1904 		sqe->flags = wqe->flags;
1905 		sqe->wqe_size = wqe_sz;
1906 		sqe->inv_key_or_imm_data = cpu_to_le32(wqe->send.inv_key);
1907 		if (qp->type == CMDQ_CREATE_QP_TYPE_UD ||
1908 		    qp->type == CMDQ_CREATE_QP_TYPE_GSI) {
1909 			sqe->q_key = cpu_to_le32(wqe->send.q_key);
1910 			sqe->length = cpu_to_le32(data_len);
1911 			sq->psn = (sq->psn + 1) & BTH_PSN_MASK;
1912 			ext_sqe->dst_qp = cpu_to_le32(wqe->send.dst_qp &
1913 						      SQ_SEND_DST_QP_MASK);
1914 			ext_sqe->avid = cpu_to_le32(wqe->send.avid &
1915 						    SQ_SEND_AVID_MASK);
1916 			msn_update = false;
1917 		} else {
1918 			sqe->length = cpu_to_le32(data_len);
1919 			if (qp->mtu)
1920 				pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
1921 			if (!pkt_num)
1922 				pkt_num = 1;
1923 			sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
1924 		}
1925 		break;
1926 	}
1927 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE:
1928 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE_WITH_IMM:
1929 	case BNXT_QPLIB_SWQE_TYPE_RDMA_READ:
1930 	{
1931 		struct sq_rdma_ext_hdr *ext_sqe = ext_hdr;
1932 		struct sq_rdma_hdr *sqe = base_hdr;
1933 
1934 		sqe->wqe_type = wqe->type;
1935 		sqe->flags = wqe->flags;
1936 		sqe->wqe_size = wqe_sz;
1937 		sqe->imm_data = cpu_to_le32(wqe->rdma.inv_key);
1938 		sqe->length = cpu_to_le32((u32)data_len);
1939 		ext_sqe->remote_va = cpu_to_le64(wqe->rdma.remote_va);
1940 		ext_sqe->remote_key = cpu_to_le32(wqe->rdma.r_key);
1941 		if (qp->mtu)
1942 			pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
1943 		if (!pkt_num)
1944 			pkt_num = 1;
1945 		sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
1946 		break;
1947 	}
1948 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_CMP_AND_SWP:
1949 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_FETCH_AND_ADD:
1950 	{
1951 		struct sq_atomic_ext_hdr *ext_sqe = ext_hdr;
1952 		struct sq_atomic_hdr *sqe = base_hdr;
1953 
1954 		sqe->wqe_type = wqe->type;
1955 		sqe->flags = wqe->flags;
1956 		sqe->remote_key = cpu_to_le32(wqe->atomic.r_key);
1957 		sqe->remote_va = cpu_to_le64(wqe->atomic.remote_va);
1958 		ext_sqe->swap_data = cpu_to_le64(wqe->atomic.swap_data);
1959 		ext_sqe->cmp_data = cpu_to_le64(wqe->atomic.cmp_data);
1960 		if (qp->mtu)
1961 			pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
1962 		if (!pkt_num)
1963 			pkt_num = 1;
1964 		sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
1965 		break;
1966 	}
1967 	case BNXT_QPLIB_SWQE_TYPE_LOCAL_INV:
1968 	{
1969 		struct sq_localinvalidate *sqe = base_hdr;
1970 
1971 		sqe->wqe_type = wqe->type;
1972 		sqe->flags = wqe->flags;
1973 		sqe->inv_l_key = cpu_to_le32(wqe->local_inv.inv_l_key);
1974 		msn_update = false;
1975 		break;
1976 	}
1977 	case BNXT_QPLIB_SWQE_TYPE_FAST_REG_MR:
1978 	{
1979 		struct sq_fr_pmr_ext_hdr *ext_sqe = ext_hdr;
1980 		struct sq_fr_pmr_hdr *sqe = base_hdr;
1981 
1982 		sqe->wqe_type = wqe->type;
1983 		sqe->flags = wqe->flags;
1984 		sqe->access_cntl = wqe->frmr.access_cntl |
1985 				   SQ_FR_PMR_ACCESS_CNTL_LOCAL_WRITE;
1986 		sqe->zero_based_page_size_log =
1987 			(wqe->frmr.pg_sz_log & SQ_FR_PMR_PAGE_SIZE_LOG_MASK) <<
1988 			SQ_FR_PMR_PAGE_SIZE_LOG_SFT |
1989 			(wqe->frmr.zero_based ? SQ_FR_PMR_ZERO_BASED : 0);
1990 		sqe->l_key = cpu_to_le32(wqe->frmr.l_key);
1991 		temp32 = cpu_to_le32(wqe->frmr.length);
1992 		memcpy(sqe->length, &temp32, sizeof(wqe->frmr.length));
1993 		sqe->numlevels_pbl_page_size_log =
1994 			((wqe->frmr.pbl_pg_sz_log <<
1995 					SQ_FR_PMR_PBL_PAGE_SIZE_LOG_SFT) &
1996 					SQ_FR_PMR_PBL_PAGE_SIZE_LOG_MASK) |
1997 			((wqe->frmr.levels << SQ_FR_PMR_NUMLEVELS_SFT) &
1998 					SQ_FR_PMR_NUMLEVELS_MASK);
1999 
2000 		for (i = 0; i < wqe->frmr.page_list_len; i++)
2001 			wqe->frmr.pbl_ptr[i] = cpu_to_le64(
2002 						wqe->frmr.page_list[i] |
2003 						PTU_PTE_VALID);
2004 		ext_sqe->pblptr = cpu_to_le64(wqe->frmr.pbl_dma_ptr);
2005 		ext_sqe->va = cpu_to_le64(wqe->frmr.va);
2006 		msn_update = false;
2007 
2008 		break;
2009 	}
2010 	case BNXT_QPLIB_SWQE_TYPE_BIND_MW:
2011 	{
2012 		struct sq_bind_ext_hdr *ext_sqe = ext_hdr;
2013 		struct sq_bind_hdr *sqe = base_hdr;
2014 
2015 		sqe->wqe_type = wqe->type;
2016 		sqe->flags = wqe->flags;
2017 		sqe->access_cntl = wqe->bind.access_cntl;
2018 		sqe->mw_type_zero_based = wqe->bind.mw_type |
2019 			(wqe->bind.zero_based ? SQ_BIND_ZERO_BASED : 0);
2020 		sqe->parent_l_key = cpu_to_le32(wqe->bind.parent_l_key);
2021 		sqe->l_key = cpu_to_le32(wqe->bind.r_key);
2022 		ext_sqe->va = cpu_to_le64(wqe->bind.va);
2023 		ext_sqe->length_lo = cpu_to_le32(wqe->bind.length);
2024 		msn_update = false;
2025 		break;
2026 	}
2027 	default:
2028 		/* Bad wqe, return error */
2029 		rc = -EINVAL;
2030 		goto done;
2031 	}
2032 	if (!qp->is_host_msn_tbl || msn_update) {
2033 		swq->next_psn = sq->psn & BTH_PSN_MASK;
2034 		bnxt_qplib_fill_psn_search(qp, wqe, swq);
2035 	}
2036 queue_err:
2037 	bnxt_qplib_swq_mod_start(sq, wqe_idx);
2038 	bnxt_qplib_hwq_incr_prod(&sq->dbinfo, hwq, swq->slots);
2039 	qp->wqe_cnt++;
2040 done:
2041 	if (sch_handler) {
2042 		nq_work = kzalloc(sizeof(*nq_work), GFP_ATOMIC);
2043 		if (nq_work) {
2044 			nq_work->cq = qp->scq;
2045 			nq_work->nq = qp->scq->nq;
2046 			INIT_WORK(&nq_work->work, bnxt_qpn_cqn_sched_task);
2047 			queue_work(qp->scq->nq->cqn_wq, &nq_work->work);
2048 		} else {
2049 			dev_err(&hwq->pdev->dev,
2050 				"FP: Failed to allocate SQ nq_work!\n");
2051 			rc = -ENOMEM;
2052 		}
2053 	}
2054 	return rc;
2055 }
2056 
2057 void bnxt_qplib_post_recv_db(struct bnxt_qplib_qp *qp)
2058 {
2059 	struct bnxt_qplib_q *rq = &qp->rq;
2060 
2061 	bnxt_qplib_ring_prod_db(&rq->dbinfo, DBC_DBC_TYPE_RQ);
2062 }
2063 
2064 int bnxt_qplib_post_recv(struct bnxt_qplib_qp *qp,
2065 			 struct bnxt_qplib_swqe *wqe)
2066 {
2067 	struct bnxt_qplib_nq_work *nq_work = NULL;
2068 	struct bnxt_qplib_q *rq = &qp->rq;
2069 	struct rq_wqe_hdr *base_hdr;
2070 	struct rq_ext_hdr *ext_hdr;
2071 	struct bnxt_qplib_hwq *hwq;
2072 	struct bnxt_qplib_swq *swq;
2073 	bool sch_handler = false;
2074 	u16 wqe_sz, idx;
2075 	u32 wqe_idx;
2076 	int rc = 0;
2077 
2078 	hwq = &rq->hwq;
2079 	if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_RESET) {
2080 		dev_err(&hwq->pdev->dev,
2081 			"QPLIB: FP: QP (0x%x) is in the 0x%x state",
2082 			qp->id, qp->state);
2083 		rc = -EINVAL;
2084 		goto done;
2085 	}
2086 
2087 	if (bnxt_qplib_queue_full(rq, rq->dbinfo.max_slot)) {
2088 		dev_err(&hwq->pdev->dev,
2089 			"FP: QP (0x%x) RQ is full!\n", qp->id);
2090 		rc = -EINVAL;
2091 		goto done;
2092 	}
2093 
2094 	swq = bnxt_qplib_get_swqe(rq, &wqe_idx);
2095 	swq->wr_id = wqe->wr_id;
2096 	swq->slots = rq->dbinfo.max_slot;
2097 
2098 	if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
2099 		sch_handler = true;
2100 		dev_dbg(&hwq->pdev->dev,
2101 			"%s: Error QP. Scheduling for poll_cq\n", __func__);
2102 		goto queue_err;
2103 	}
2104 
2105 	idx = 0;
2106 	base_hdr = bnxt_qplib_get_prod_qe(hwq, idx++);
2107 	ext_hdr = bnxt_qplib_get_prod_qe(hwq, idx++);
2108 	memset(base_hdr, 0, sizeof(struct sq_sge));
2109 	memset(ext_hdr, 0, sizeof(struct sq_sge));
2110 	wqe_sz = (sizeof(struct rq_wqe_hdr) +
2111 	wqe->num_sge * sizeof(struct sq_sge)) >> 4;
2112 	bnxt_qplib_put_sges(hwq, wqe->sg_list, wqe->num_sge, &idx);
2113 	if (!wqe->num_sge) {
2114 		struct sq_sge *sge;
2115 
2116 		sge = bnxt_qplib_get_prod_qe(hwq, idx++);
2117 		sge->size = 0;
2118 		wqe_sz++;
2119 	}
2120 	base_hdr->wqe_type = wqe->type;
2121 	base_hdr->flags = wqe->flags;
2122 	base_hdr->wqe_size = wqe_sz;
2123 	base_hdr->wr_id[0] = cpu_to_le32(wqe_idx);
2124 queue_err:
2125 	bnxt_qplib_swq_mod_start(rq, wqe_idx);
2126 	bnxt_qplib_hwq_incr_prod(&rq->dbinfo, hwq, swq->slots);
2127 done:
2128 	if (sch_handler) {
2129 		nq_work = kzalloc(sizeof(*nq_work), GFP_ATOMIC);
2130 		if (nq_work) {
2131 			nq_work->cq = qp->rcq;
2132 			nq_work->nq = qp->rcq->nq;
2133 			INIT_WORK(&nq_work->work, bnxt_qpn_cqn_sched_task);
2134 			queue_work(qp->rcq->nq->cqn_wq, &nq_work->work);
2135 		} else {
2136 			dev_err(&hwq->pdev->dev,
2137 				"FP: Failed to allocate RQ nq_work!\n");
2138 			rc = -ENOMEM;
2139 		}
2140 	}
2141 
2142 	return rc;
2143 }
2144 
2145 /* CQ */
2146 int bnxt_qplib_create_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq)
2147 {
2148 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
2149 	struct bnxt_qplib_hwq_attr hwq_attr = {};
2150 	struct creq_create_cq_resp resp = {};
2151 	struct bnxt_qplib_cmdqmsg msg = {};
2152 	struct cmdq_create_cq req = {};
2153 	struct bnxt_qplib_pbl *pbl;
2154 	u32 pg_sz_lvl;
2155 	int rc;
2156 
2157 	if (!cq->dpi) {
2158 		dev_err(&rcfw->pdev->dev,
2159 			"FP: CREATE_CQ failed due to NULL DPI\n");
2160 		return -EINVAL;
2161 	}
2162 
2163 	cq->dbinfo.flags = 0;
2164 	hwq_attr.res = res;
2165 	hwq_attr.depth = cq->max_wqe;
2166 	hwq_attr.stride = sizeof(struct cq_base);
2167 	hwq_attr.type = HWQ_TYPE_QUEUE;
2168 	hwq_attr.sginfo = &cq->sg_info;
2169 	rc = bnxt_qplib_alloc_init_hwq(&cq->hwq, &hwq_attr);
2170 	if (rc)
2171 		return rc;
2172 
2173 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
2174 				 CMDQ_BASE_OPCODE_CREATE_CQ,
2175 				 sizeof(req));
2176 
2177 	req.dpi = cpu_to_le32(cq->dpi->dpi);
2178 	req.cq_handle = cpu_to_le64(cq->cq_handle);
2179 	req.cq_size = cpu_to_le32(cq->max_wqe);
2180 	pbl = &cq->hwq.pbl[PBL_LVL_0];
2181 	pg_sz_lvl = (bnxt_qplib_base_pg_size(&cq->hwq) <<
2182 		     CMDQ_CREATE_CQ_PG_SIZE_SFT);
2183 	pg_sz_lvl |= (cq->hwq.level & CMDQ_CREATE_CQ_LVL_MASK);
2184 	req.pg_size_lvl = cpu_to_le32(pg_sz_lvl);
2185 	req.pbl = cpu_to_le64(pbl->pg_map_arr[0]);
2186 	req.cq_fco_cnq_id = cpu_to_le32(
2187 			(cq->cnq_hw_ring_id & CMDQ_CREATE_CQ_CNQ_ID_MASK) <<
2188 			 CMDQ_CREATE_CQ_CNQ_ID_SFT);
2189 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
2190 				sizeof(resp), 0);
2191 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
2192 	if (rc)
2193 		goto fail;
2194 
2195 	cq->id = le32_to_cpu(resp.xid);
2196 	cq->period = BNXT_QPLIB_QUEUE_START_PERIOD;
2197 	init_waitqueue_head(&cq->waitq);
2198 	INIT_LIST_HEAD(&cq->sqf_head);
2199 	INIT_LIST_HEAD(&cq->rqf_head);
2200 	spin_lock_init(&cq->compl_lock);
2201 	spin_lock_init(&cq->flush_lock);
2202 
2203 	cq->dbinfo.hwq = &cq->hwq;
2204 	cq->dbinfo.xid = cq->id;
2205 	cq->dbinfo.db = cq->dpi->dbr;
2206 	cq->dbinfo.priv_db = res->dpi_tbl.priv_db;
2207 	cq->dbinfo.flags = 0;
2208 	cq->dbinfo.toggle = 0;
2209 
2210 	bnxt_qplib_armen_db(&cq->dbinfo, DBC_DBC_TYPE_CQ_ARMENA);
2211 
2212 	return 0;
2213 
2214 fail:
2215 	bnxt_qplib_free_hwq(res, &cq->hwq);
2216 	return rc;
2217 }
2218 
2219 void bnxt_qplib_resize_cq_complete(struct bnxt_qplib_res *res,
2220 				   struct bnxt_qplib_cq *cq)
2221 {
2222 	bnxt_qplib_free_hwq(res, &cq->hwq);
2223 	memcpy(&cq->hwq, &cq->resize_hwq, sizeof(cq->hwq));
2224        /* Reset only the cons bit in the flags */
2225 	cq->dbinfo.flags &= ~(1UL << BNXT_QPLIB_FLAG_EPOCH_CONS_SHIFT);
2226 }
2227 
2228 int bnxt_qplib_resize_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq,
2229 			 int new_cqes)
2230 {
2231 	struct bnxt_qplib_hwq_attr hwq_attr = {};
2232 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
2233 	struct creq_resize_cq_resp resp = {};
2234 	struct bnxt_qplib_cmdqmsg msg = {};
2235 	struct cmdq_resize_cq req = {};
2236 	struct bnxt_qplib_pbl *pbl;
2237 	u32 pg_sz, lvl, new_sz;
2238 	int rc;
2239 
2240 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
2241 				 CMDQ_BASE_OPCODE_RESIZE_CQ,
2242 				 sizeof(req));
2243 	hwq_attr.sginfo = &cq->sg_info;
2244 	hwq_attr.res = res;
2245 	hwq_attr.depth = new_cqes;
2246 	hwq_attr.stride = sizeof(struct cq_base);
2247 	hwq_attr.type = HWQ_TYPE_QUEUE;
2248 	rc = bnxt_qplib_alloc_init_hwq(&cq->resize_hwq, &hwq_attr);
2249 	if (rc)
2250 		return rc;
2251 
2252 	req.cq_cid = cpu_to_le32(cq->id);
2253 	pbl = &cq->resize_hwq.pbl[PBL_LVL_0];
2254 	pg_sz = bnxt_qplib_base_pg_size(&cq->resize_hwq);
2255 	lvl = (cq->resize_hwq.level << CMDQ_RESIZE_CQ_LVL_SFT) &
2256 				       CMDQ_RESIZE_CQ_LVL_MASK;
2257 	new_sz = (new_cqes << CMDQ_RESIZE_CQ_NEW_CQ_SIZE_SFT) &
2258 		  CMDQ_RESIZE_CQ_NEW_CQ_SIZE_MASK;
2259 	req.new_cq_size_pg_size_lvl = cpu_to_le32(new_sz | pg_sz | lvl);
2260 	req.new_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
2261 
2262 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
2263 				sizeof(resp), 0);
2264 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
2265 	return rc;
2266 }
2267 
2268 int bnxt_qplib_destroy_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq)
2269 {
2270 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
2271 	struct creq_destroy_cq_resp resp = {};
2272 	struct bnxt_qplib_cmdqmsg msg = {};
2273 	struct cmdq_destroy_cq req = {};
2274 	u16 total_cnq_events;
2275 	int rc;
2276 
2277 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
2278 				 CMDQ_BASE_OPCODE_DESTROY_CQ,
2279 				 sizeof(req));
2280 
2281 	req.cq_cid = cpu_to_le32(cq->id);
2282 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
2283 				sizeof(resp), 0);
2284 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
2285 	if (rc)
2286 		return rc;
2287 	total_cnq_events = le16_to_cpu(resp.total_cnq_events);
2288 	__wait_for_all_nqes(cq, total_cnq_events);
2289 	bnxt_qplib_free_hwq(res, &cq->hwq);
2290 	return 0;
2291 }
2292 
2293 static int __flush_sq(struct bnxt_qplib_q *sq, struct bnxt_qplib_qp *qp,
2294 		      struct bnxt_qplib_cqe **pcqe, int *budget)
2295 {
2296 	struct bnxt_qplib_cqe *cqe;
2297 	u32 start, last;
2298 	int rc = 0;
2299 
2300 	/* Now complete all outstanding SQEs with FLUSHED_ERR */
2301 	start = sq->swq_start;
2302 	cqe = *pcqe;
2303 	while (*budget) {
2304 		last = sq->swq_last;
2305 		if (start == last)
2306 			break;
2307 		/* Skip the FENCE WQE completions */
2308 		if (sq->swq[last].wr_id == BNXT_QPLIB_FENCE_WRID) {
2309 			bnxt_qplib_cancel_phantom_processing(qp);
2310 			goto skip_compl;
2311 		}
2312 		memset(cqe, 0, sizeof(*cqe));
2313 		cqe->status = CQ_REQ_STATUS_WORK_REQUEST_FLUSHED_ERR;
2314 		cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
2315 		cqe->qp_handle = (u64)(unsigned long)qp;
2316 		cqe->wr_id = sq->swq[last].wr_id;
2317 		cqe->src_qp = qp->id;
2318 		cqe->type = sq->swq[last].type;
2319 		cqe++;
2320 		(*budget)--;
2321 skip_compl:
2322 		bnxt_qplib_hwq_incr_cons(sq->hwq.max_elements, &sq->hwq.cons,
2323 					 sq->swq[last].slots, &sq->dbinfo.flags);
2324 		sq->swq_last = sq->swq[last].next_idx;
2325 	}
2326 	*pcqe = cqe;
2327 	if (!(*budget) && sq->swq_last != start)
2328 		/* Out of budget */
2329 		rc = -EAGAIN;
2330 
2331 	return rc;
2332 }
2333 
2334 static int __flush_rq(struct bnxt_qplib_q *rq, struct bnxt_qplib_qp *qp,
2335 		      struct bnxt_qplib_cqe **pcqe, int *budget)
2336 {
2337 	struct bnxt_qplib_cqe *cqe;
2338 	u32 start, last;
2339 	int opcode = 0;
2340 	int rc = 0;
2341 
2342 	switch (qp->type) {
2343 	case CMDQ_CREATE_QP1_TYPE_GSI:
2344 		opcode = CQ_BASE_CQE_TYPE_RES_RAWETH_QP1;
2345 		break;
2346 	case CMDQ_CREATE_QP_TYPE_RC:
2347 		opcode = CQ_BASE_CQE_TYPE_RES_RC;
2348 		break;
2349 	case CMDQ_CREATE_QP_TYPE_UD:
2350 	case CMDQ_CREATE_QP_TYPE_GSI:
2351 		opcode = CQ_BASE_CQE_TYPE_RES_UD;
2352 		break;
2353 	}
2354 
2355 	/* Flush the rest of the RQ */
2356 	start = rq->swq_start;
2357 	cqe = *pcqe;
2358 	while (*budget) {
2359 		last = rq->swq_last;
2360 		if (last == start)
2361 			break;
2362 		memset(cqe, 0, sizeof(*cqe));
2363 		cqe->status =
2364 		    CQ_RES_RC_STATUS_WORK_REQUEST_FLUSHED_ERR;
2365 		cqe->opcode = opcode;
2366 		cqe->qp_handle = (unsigned long)qp;
2367 		cqe->wr_id = rq->swq[last].wr_id;
2368 		cqe++;
2369 		(*budget)--;
2370 		bnxt_qplib_hwq_incr_cons(rq->hwq.max_elements, &rq->hwq.cons,
2371 					 rq->swq[last].slots, &rq->dbinfo.flags);
2372 		rq->swq_last = rq->swq[last].next_idx;
2373 	}
2374 	*pcqe = cqe;
2375 	if (!*budget && rq->swq_last != start)
2376 		/* Out of budget */
2377 		rc = -EAGAIN;
2378 
2379 	return rc;
2380 }
2381 
2382 void bnxt_qplib_mark_qp_error(void *qp_handle)
2383 {
2384 	struct bnxt_qplib_qp *qp = qp_handle;
2385 
2386 	if (!qp)
2387 		return;
2388 
2389 	/* Must block new posting of SQ and RQ */
2390 	qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2391 	bnxt_qplib_cancel_phantom_processing(qp);
2392 }
2393 
2394 /* Note: SQE is valid from sw_sq_cons up to cqe_sq_cons (exclusive)
2395  *       CQE is track from sw_cq_cons to max_element but valid only if VALID=1
2396  */
2397 static int do_wa9060(struct bnxt_qplib_qp *qp, struct bnxt_qplib_cq *cq,
2398 		     u32 cq_cons, u32 swq_last, u32 cqe_sq_cons)
2399 {
2400 	u32 peek_sw_cq_cons, peek_sq_cons_idx, peek_flags;
2401 	struct bnxt_qplib_q *sq = &qp->sq;
2402 	struct cq_req *peek_req_hwcqe;
2403 	struct bnxt_qplib_qp *peek_qp;
2404 	struct bnxt_qplib_q *peek_sq;
2405 	struct bnxt_qplib_swq *swq;
2406 	struct cq_base *peek_hwcqe;
2407 	int i, rc = 0;
2408 
2409 	/* Normal mode */
2410 	/* Check for the psn_search marking before completing */
2411 	swq = &sq->swq[swq_last];
2412 	if (swq->psn_search &&
2413 	    le32_to_cpu(swq->psn_search->flags_next_psn) & 0x80000000) {
2414 		/* Unmark */
2415 		swq->psn_search->flags_next_psn = cpu_to_le32
2416 			(le32_to_cpu(swq->psn_search->flags_next_psn)
2417 				     & ~0x80000000);
2418 		dev_dbg(&cq->hwq.pdev->dev,
2419 			"FP: Process Req cq_cons=0x%x qp=0x%x sq cons sw=0x%x cqe=0x%x marked!\n",
2420 			cq_cons, qp->id, swq_last, cqe_sq_cons);
2421 		sq->condition = true;
2422 		sq->send_phantom = true;
2423 
2424 		/* TODO: Only ARM if the previous SQE is ARMALL */
2425 		bnxt_qplib_ring_db(&cq->dbinfo, DBC_DBC_TYPE_CQ_ARMALL);
2426 		rc = -EAGAIN;
2427 		goto out;
2428 	}
2429 	if (sq->condition) {
2430 		/* Peek at the completions */
2431 		peek_flags = cq->dbinfo.flags;
2432 		peek_sw_cq_cons = cq_cons;
2433 		i = cq->hwq.max_elements;
2434 		while (i--) {
2435 			peek_hwcqe = bnxt_qplib_get_qe(&cq->hwq,
2436 						       peek_sw_cq_cons, NULL);
2437 			/* If the next hwcqe is VALID */
2438 			if (CQE_CMP_VALID(peek_hwcqe, peek_flags)) {
2439 			/*
2440 			 * The valid test of the entry must be done first before
2441 			 * reading any further.
2442 			 */
2443 				dma_rmb();
2444 				/* If the next hwcqe is a REQ */
2445 				if ((peek_hwcqe->cqe_type_toggle &
2446 				    CQ_BASE_CQE_TYPE_MASK) ==
2447 				    CQ_BASE_CQE_TYPE_REQ) {
2448 					peek_req_hwcqe = (struct cq_req *)
2449 							 peek_hwcqe;
2450 					peek_qp = (struct bnxt_qplib_qp *)
2451 						((unsigned long)
2452 						 le64_to_cpu
2453 						 (peek_req_hwcqe->qp_handle));
2454 					peek_sq = &peek_qp->sq;
2455 					peek_sq_cons_idx =
2456 						((le16_to_cpu(
2457 						  peek_req_hwcqe->sq_cons_idx)
2458 						  - 1) % sq->max_wqe);
2459 					/* If the hwcqe's sq's wr_id matches */
2460 					if (peek_sq == sq &&
2461 					    sq->swq[peek_sq_cons_idx].wr_id ==
2462 					    BNXT_QPLIB_FENCE_WRID) {
2463 						/*
2464 						 *  Unbreak only if the phantom
2465 						 *  comes back
2466 						 */
2467 						dev_dbg(&cq->hwq.pdev->dev,
2468 							"FP: Got Phantom CQE\n");
2469 						sq->condition = false;
2470 						sq->single = true;
2471 						rc = 0;
2472 						goto out;
2473 					}
2474 				}
2475 				/* Valid but not the phantom, so keep looping */
2476 			} else {
2477 				/* Not valid yet, just exit and wait */
2478 				rc = -EINVAL;
2479 				goto out;
2480 			}
2481 			bnxt_qplib_hwq_incr_cons(cq->hwq.max_elements,
2482 						 &peek_sw_cq_cons,
2483 						 1, &peek_flags);
2484 		}
2485 		dev_err(&cq->hwq.pdev->dev,
2486 			"Should not have come here! cq_cons=0x%x qp=0x%x sq cons sw=0x%x hw=0x%x\n",
2487 			cq_cons, qp->id, swq_last, cqe_sq_cons);
2488 		rc = -EINVAL;
2489 	}
2490 out:
2491 	return rc;
2492 }
2493 
2494 static int bnxt_qplib_get_cqe_sq_cons(struct bnxt_qplib_q *sq, u32 cqe_slot)
2495 {
2496 	struct bnxt_qplib_hwq *sq_hwq;
2497 	struct bnxt_qplib_swq *swq;
2498 	int cqe_sq_cons = -1;
2499 	u32 start, last;
2500 
2501 	sq_hwq = &sq->hwq;
2502 
2503 	start = sq->swq_start;
2504 	last = sq->swq_last;
2505 
2506 	while (last != start) {
2507 		swq = &sq->swq[last];
2508 		if (swq->slot_idx  == cqe_slot) {
2509 			cqe_sq_cons = swq->next_idx;
2510 			dev_err(&sq_hwq->pdev->dev, "%s: Found cons wqe = %d slot = %d\n",
2511 				__func__, cqe_sq_cons, cqe_slot);
2512 			break;
2513 		}
2514 
2515 		last = swq->next_idx;
2516 	}
2517 	return cqe_sq_cons;
2518 }
2519 
2520 static int bnxt_qplib_cq_process_req(struct bnxt_qplib_cq *cq,
2521 				     struct cq_req *hwcqe,
2522 				     struct bnxt_qplib_cqe **pcqe, int *budget,
2523 				     u32 cq_cons, struct bnxt_qplib_qp **lib_qp)
2524 {
2525 	struct bnxt_qplib_swq *swq;
2526 	struct bnxt_qplib_cqe *cqe;
2527 	u32 cqe_sq_cons, slot_num;
2528 	struct bnxt_qplib_qp *qp;
2529 	struct bnxt_qplib_q *sq;
2530 	int cqe_cons;
2531 	int rc = 0;
2532 
2533 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2534 				      le64_to_cpu(hwcqe->qp_handle));
2535 	if (!qp) {
2536 		dev_err(&cq->hwq.pdev->dev,
2537 			"FP: Process Req qp is NULL\n");
2538 		return -EINVAL;
2539 	}
2540 	sq = &qp->sq;
2541 
2542 	cqe_sq_cons = le16_to_cpu(hwcqe->sq_cons_idx) % sq->max_sw_wqe;
2543 	if (qp->sq.flushed) {
2544 		dev_dbg(&cq->hwq.pdev->dev,
2545 			"%s: QP in Flush QP = %p\n", __func__, qp);
2546 		goto done;
2547 	}
2548 
2549 	if (__is_err_cqe_for_var_wqe(qp, hwcqe->status)) {
2550 		slot_num = le16_to_cpu(hwcqe->sq_cons_idx);
2551 		cqe_cons = bnxt_qplib_get_cqe_sq_cons(sq, slot_num);
2552 		if (cqe_cons < 0) {
2553 			dev_err(&cq->hwq.pdev->dev, "%s: Wrong SQ cons cqe_slot_indx = %d\n",
2554 				__func__, slot_num);
2555 			goto done;
2556 		}
2557 		cqe_sq_cons = cqe_cons;
2558 		dev_err(&cq->hwq.pdev->dev, "%s: cqe_sq_cons = %d swq_last = %d swq_start = %d\n",
2559 			__func__, cqe_sq_cons, sq->swq_last, sq->swq_start);
2560 	}
2561 
2562 	/* Require to walk the sq's swq to fabricate CQEs for all previously
2563 	 * signaled SWQEs due to CQE aggregation from the current sq cons
2564 	 * to the cqe_sq_cons
2565 	 */
2566 	cqe = *pcqe;
2567 	while (*budget) {
2568 		if (sq->swq_last == cqe_sq_cons)
2569 			/* Done */
2570 			break;
2571 
2572 		swq = &sq->swq[sq->swq_last];
2573 		memset(cqe, 0, sizeof(*cqe));
2574 		cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
2575 		cqe->qp_handle = (u64)(unsigned long)qp;
2576 		cqe->src_qp = qp->id;
2577 		cqe->wr_id = swq->wr_id;
2578 		if (cqe->wr_id == BNXT_QPLIB_FENCE_WRID)
2579 			goto skip;
2580 		cqe->type = swq->type;
2581 
2582 		/* For the last CQE, check for status.  For errors, regardless
2583 		 * of the request being signaled or not, it must complete with
2584 		 * the hwcqe error status
2585 		 */
2586 		if (swq->next_idx == cqe_sq_cons &&
2587 		    hwcqe->status != CQ_REQ_STATUS_OK) {
2588 			cqe->status = hwcqe->status;
2589 			dev_err(&cq->hwq.pdev->dev,
2590 				"FP: CQ Processed Req wr_id[%d] = 0x%llx with status 0x%x\n",
2591 				sq->swq_last, cqe->wr_id, cqe->status);
2592 			cqe++;
2593 			(*budget)--;
2594 			bnxt_qplib_mark_qp_error(qp);
2595 			/* Add qp to flush list of the CQ */
2596 			bnxt_qplib_add_flush_qp(qp);
2597 		} else {
2598 			/* Before we complete, do WA 9060 */
2599 			if (do_wa9060(qp, cq, cq_cons, sq->swq_last,
2600 				      cqe_sq_cons)) {
2601 				*lib_qp = qp;
2602 				goto out;
2603 			}
2604 			if (swq->flags & SQ_SEND_FLAGS_SIGNAL_COMP) {
2605 				cqe->status = CQ_REQ_STATUS_OK;
2606 				cqe++;
2607 				(*budget)--;
2608 			}
2609 		}
2610 skip:
2611 		bnxt_qplib_hwq_incr_cons(sq->hwq.max_elements, &sq->hwq.cons,
2612 					 swq->slots, &sq->dbinfo.flags);
2613 		sq->swq_last = swq->next_idx;
2614 		if (sq->single)
2615 			break;
2616 	}
2617 out:
2618 	*pcqe = cqe;
2619 	if (sq->swq_last != cqe_sq_cons) {
2620 		/* Out of budget */
2621 		rc = -EAGAIN;
2622 		goto done;
2623 	}
2624 	/*
2625 	 * Back to normal completion mode only after it has completed all of
2626 	 * the WC for this CQE
2627 	 */
2628 	sq->single = false;
2629 done:
2630 	return rc;
2631 }
2632 
2633 static void bnxt_qplib_release_srqe(struct bnxt_qplib_srq *srq, u32 tag)
2634 {
2635 	spin_lock(&srq->hwq.lock);
2636 	srq->swq[srq->last_idx].next_idx = (int)tag;
2637 	srq->last_idx = (int)tag;
2638 	srq->swq[srq->last_idx].next_idx = -1;
2639 	bnxt_qplib_hwq_incr_cons(srq->hwq.max_elements, &srq->hwq.cons,
2640 				 srq->dbinfo.max_slot, &srq->dbinfo.flags);
2641 	spin_unlock(&srq->hwq.lock);
2642 }
2643 
2644 static int bnxt_qplib_cq_process_res_rc(struct bnxt_qplib_cq *cq,
2645 					struct cq_res_rc *hwcqe,
2646 					struct bnxt_qplib_cqe **pcqe,
2647 					int *budget)
2648 {
2649 	struct bnxt_qplib_srq *srq;
2650 	struct bnxt_qplib_cqe *cqe;
2651 	struct bnxt_qplib_qp *qp;
2652 	struct bnxt_qplib_q *rq;
2653 	u32 wr_id_idx;
2654 
2655 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2656 				      le64_to_cpu(hwcqe->qp_handle));
2657 	if (!qp) {
2658 		dev_err(&cq->hwq.pdev->dev, "process_cq RC qp is NULL\n");
2659 		return -EINVAL;
2660 	}
2661 	if (qp->rq.flushed) {
2662 		dev_dbg(&cq->hwq.pdev->dev,
2663 			"%s: QP in Flush QP = %p\n", __func__, qp);
2664 		return 0;
2665 	}
2666 
2667 	cqe = *pcqe;
2668 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2669 	cqe->length = le32_to_cpu(hwcqe->length);
2670 	cqe->invrkey = le32_to_cpu(hwcqe->imm_data_or_inv_r_key);
2671 	cqe->mr_handle = le64_to_cpu(hwcqe->mr_handle);
2672 	cqe->flags = le16_to_cpu(hwcqe->flags);
2673 	cqe->status = hwcqe->status;
2674 	cqe->qp_handle = (u64)(unsigned long)qp;
2675 
2676 	wr_id_idx = le32_to_cpu(hwcqe->srq_or_rq_wr_id) &
2677 				CQ_RES_RC_SRQ_OR_RQ_WR_ID_MASK;
2678 	if (cqe->flags & CQ_RES_RC_FLAGS_SRQ_SRQ) {
2679 		srq = qp->srq;
2680 		if (!srq)
2681 			return -EINVAL;
2682 		if (wr_id_idx >= srq->hwq.max_elements) {
2683 			dev_err(&cq->hwq.pdev->dev,
2684 				"FP: CQ Process RC wr_id idx 0x%x exceeded SRQ max 0x%x\n",
2685 				wr_id_idx, srq->hwq.max_elements);
2686 			return -EINVAL;
2687 		}
2688 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2689 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2690 		cqe++;
2691 		(*budget)--;
2692 		*pcqe = cqe;
2693 	} else {
2694 		struct bnxt_qplib_swq *swq;
2695 
2696 		rq = &qp->rq;
2697 		if (wr_id_idx > (rq->max_wqe - 1)) {
2698 			dev_err(&cq->hwq.pdev->dev,
2699 				"FP: CQ Process RC wr_id idx 0x%x exceeded RQ max 0x%x\n",
2700 				wr_id_idx, rq->max_wqe);
2701 			return -EINVAL;
2702 		}
2703 		if (wr_id_idx != rq->swq_last)
2704 			return -EINVAL;
2705 		swq = &rq->swq[rq->swq_last];
2706 		cqe->wr_id = swq->wr_id;
2707 		cqe++;
2708 		(*budget)--;
2709 		bnxt_qplib_hwq_incr_cons(rq->hwq.max_elements, &rq->hwq.cons,
2710 					 swq->slots, &rq->dbinfo.flags);
2711 		rq->swq_last = swq->next_idx;
2712 		*pcqe = cqe;
2713 
2714 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2715 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2716 			/* Add qp to flush list of the CQ */
2717 			bnxt_qplib_add_flush_qp(qp);
2718 		}
2719 	}
2720 
2721 	return 0;
2722 }
2723 
2724 static int bnxt_qplib_cq_process_res_ud(struct bnxt_qplib_cq *cq,
2725 					struct cq_res_ud *hwcqe,
2726 					struct bnxt_qplib_cqe **pcqe,
2727 					int *budget)
2728 {
2729 	struct bnxt_qplib_srq *srq;
2730 	struct bnxt_qplib_cqe *cqe;
2731 	struct bnxt_qplib_qp *qp;
2732 	struct bnxt_qplib_q *rq;
2733 	u32 wr_id_idx;
2734 
2735 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2736 				      le64_to_cpu(hwcqe->qp_handle));
2737 	if (!qp) {
2738 		dev_err(&cq->hwq.pdev->dev, "process_cq UD qp is NULL\n");
2739 		return -EINVAL;
2740 	}
2741 	if (qp->rq.flushed) {
2742 		dev_dbg(&cq->hwq.pdev->dev,
2743 			"%s: QP in Flush QP = %p\n", __func__, qp);
2744 		return 0;
2745 	}
2746 	cqe = *pcqe;
2747 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2748 	cqe->length = le16_to_cpu(hwcqe->length) & CQ_RES_UD_LENGTH_MASK;
2749 	cqe->cfa_meta = le16_to_cpu(hwcqe->cfa_metadata);
2750 	cqe->invrkey = le32_to_cpu(hwcqe->imm_data);
2751 	cqe->flags = le16_to_cpu(hwcqe->flags);
2752 	cqe->status = hwcqe->status;
2753 	cqe->qp_handle = (u64)(unsigned long)qp;
2754 	/*FIXME: Endianness fix needed for smace */
2755 	memcpy(cqe->smac, hwcqe->src_mac, ETH_ALEN);
2756 	wr_id_idx = le32_to_cpu(hwcqe->src_qp_high_srq_or_rq_wr_id)
2757 				& CQ_RES_UD_SRQ_OR_RQ_WR_ID_MASK;
2758 	cqe->src_qp = le16_to_cpu(hwcqe->src_qp_low) |
2759 				  ((le32_to_cpu(
2760 				  hwcqe->src_qp_high_srq_or_rq_wr_id) &
2761 				 CQ_RES_UD_SRC_QP_HIGH_MASK) >> 8);
2762 
2763 	if (cqe->flags & CQ_RES_RC_FLAGS_SRQ_SRQ) {
2764 		srq = qp->srq;
2765 		if (!srq)
2766 			return -EINVAL;
2767 
2768 		if (wr_id_idx >= srq->hwq.max_elements) {
2769 			dev_err(&cq->hwq.pdev->dev,
2770 				"FP: CQ Process UD wr_id idx 0x%x exceeded SRQ max 0x%x\n",
2771 				wr_id_idx, srq->hwq.max_elements);
2772 			return -EINVAL;
2773 		}
2774 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2775 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2776 		cqe++;
2777 		(*budget)--;
2778 		*pcqe = cqe;
2779 	} else {
2780 		struct bnxt_qplib_swq *swq;
2781 
2782 		rq = &qp->rq;
2783 		if (wr_id_idx > (rq->max_wqe - 1)) {
2784 			dev_err(&cq->hwq.pdev->dev,
2785 				"FP: CQ Process UD wr_id idx 0x%x exceeded RQ max 0x%x\n",
2786 				wr_id_idx, rq->max_wqe);
2787 			return -EINVAL;
2788 		}
2789 
2790 		if (rq->swq_last != wr_id_idx)
2791 			return -EINVAL;
2792 		swq = &rq->swq[rq->swq_last];
2793 		cqe->wr_id = swq->wr_id;
2794 		cqe++;
2795 		(*budget)--;
2796 		bnxt_qplib_hwq_incr_cons(rq->hwq.max_elements, &rq->hwq.cons,
2797 					 swq->slots, &rq->dbinfo.flags);
2798 		rq->swq_last = swq->next_idx;
2799 		*pcqe = cqe;
2800 
2801 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2802 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2803 			/* Add qp to flush list of the CQ */
2804 			bnxt_qplib_add_flush_qp(qp);
2805 		}
2806 	}
2807 
2808 	return 0;
2809 }
2810 
2811 bool bnxt_qplib_is_cq_empty(struct bnxt_qplib_cq *cq)
2812 {
2813 	struct cq_base *hw_cqe;
2814 	bool rc = true;
2815 
2816 	hw_cqe = bnxt_qplib_get_qe(&cq->hwq, cq->hwq.cons, NULL);
2817 	 /* Check for Valid bit. If the CQE is valid, return false */
2818 	rc = !CQE_CMP_VALID(hw_cqe, cq->dbinfo.flags);
2819 	return rc;
2820 }
2821 
2822 static int bnxt_qplib_cq_process_res_raweth_qp1(struct bnxt_qplib_cq *cq,
2823 						struct cq_res_raweth_qp1 *hwcqe,
2824 						struct bnxt_qplib_cqe **pcqe,
2825 						int *budget)
2826 {
2827 	struct bnxt_qplib_qp *qp;
2828 	struct bnxt_qplib_q *rq;
2829 	struct bnxt_qplib_srq *srq;
2830 	struct bnxt_qplib_cqe *cqe;
2831 	u32 wr_id_idx;
2832 
2833 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2834 				      le64_to_cpu(hwcqe->qp_handle));
2835 	if (!qp) {
2836 		dev_err(&cq->hwq.pdev->dev, "process_cq Raw/QP1 qp is NULL\n");
2837 		return -EINVAL;
2838 	}
2839 	if (qp->rq.flushed) {
2840 		dev_dbg(&cq->hwq.pdev->dev,
2841 			"%s: QP in Flush QP = %p\n", __func__, qp);
2842 		return 0;
2843 	}
2844 	cqe = *pcqe;
2845 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2846 	cqe->flags = le16_to_cpu(hwcqe->flags);
2847 	cqe->qp_handle = (u64)(unsigned long)qp;
2848 
2849 	wr_id_idx =
2850 		le32_to_cpu(hwcqe->raweth_qp1_payload_offset_srq_or_rq_wr_id)
2851 				& CQ_RES_RAWETH_QP1_SRQ_OR_RQ_WR_ID_MASK;
2852 	cqe->src_qp = qp->id;
2853 	if (qp->id == 1 && !cqe->length) {
2854 		/* Add workaround for the length misdetection */
2855 		cqe->length = 296;
2856 	} else {
2857 		cqe->length = le16_to_cpu(hwcqe->length);
2858 	}
2859 	cqe->pkey_index = qp->pkey_index;
2860 	memcpy(cqe->smac, qp->smac, 6);
2861 
2862 	cqe->raweth_qp1_flags = le16_to_cpu(hwcqe->raweth_qp1_flags);
2863 	cqe->raweth_qp1_flags2 = le32_to_cpu(hwcqe->raweth_qp1_flags2);
2864 	cqe->raweth_qp1_metadata = le32_to_cpu(hwcqe->raweth_qp1_metadata);
2865 
2866 	if (cqe->flags & CQ_RES_RAWETH_QP1_FLAGS_SRQ_SRQ) {
2867 		srq = qp->srq;
2868 		if (!srq) {
2869 			dev_err(&cq->hwq.pdev->dev,
2870 				"FP: SRQ used but not defined??\n");
2871 			return -EINVAL;
2872 		}
2873 		if (wr_id_idx >= srq->hwq.max_elements) {
2874 			dev_err(&cq->hwq.pdev->dev,
2875 				"FP: CQ Process Raw/QP1 wr_id idx 0x%x exceeded SRQ max 0x%x\n",
2876 				wr_id_idx, srq->hwq.max_elements);
2877 			return -EINVAL;
2878 		}
2879 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2880 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2881 		cqe++;
2882 		(*budget)--;
2883 		*pcqe = cqe;
2884 	} else {
2885 		struct bnxt_qplib_swq *swq;
2886 
2887 		rq = &qp->rq;
2888 		if (wr_id_idx > (rq->max_wqe - 1)) {
2889 			dev_err(&cq->hwq.pdev->dev,
2890 				"FP: CQ Process Raw/QP1 RQ wr_id idx 0x%x exceeded RQ max 0x%x\n",
2891 				wr_id_idx, rq->max_wqe);
2892 			return -EINVAL;
2893 		}
2894 		if (rq->swq_last != wr_id_idx)
2895 			return -EINVAL;
2896 		swq = &rq->swq[rq->swq_last];
2897 		cqe->wr_id = swq->wr_id;
2898 		cqe++;
2899 		(*budget)--;
2900 		bnxt_qplib_hwq_incr_cons(rq->hwq.max_elements, &rq->hwq.cons,
2901 					 swq->slots, &rq->dbinfo.flags);
2902 		rq->swq_last = swq->next_idx;
2903 		*pcqe = cqe;
2904 
2905 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2906 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2907 			/* Add qp to flush list of the CQ */
2908 			bnxt_qplib_add_flush_qp(qp);
2909 		}
2910 	}
2911 
2912 	return 0;
2913 }
2914 
2915 static int bnxt_qplib_cq_process_terminal(struct bnxt_qplib_cq *cq,
2916 					  struct cq_terminal *hwcqe,
2917 					  struct bnxt_qplib_cqe **pcqe,
2918 					  int *budget)
2919 {
2920 	struct bnxt_qplib_qp *qp;
2921 	struct bnxt_qplib_q *sq, *rq;
2922 	struct bnxt_qplib_cqe *cqe;
2923 	u32 swq_last = 0, cqe_cons;
2924 	int rc = 0;
2925 
2926 	/* Check the Status */
2927 	if (hwcqe->status != CQ_TERMINAL_STATUS_OK)
2928 		dev_warn(&cq->hwq.pdev->dev,
2929 			 "FP: CQ Process Terminal Error status = 0x%x\n",
2930 			 hwcqe->status);
2931 
2932 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2933 				      le64_to_cpu(hwcqe->qp_handle));
2934 	if (!qp)
2935 		return -EINVAL;
2936 
2937 	/* Must block new posting of SQ and RQ */
2938 	qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2939 
2940 	sq = &qp->sq;
2941 	rq = &qp->rq;
2942 
2943 	cqe_cons = le16_to_cpu(hwcqe->sq_cons_idx);
2944 	if (cqe_cons == 0xFFFF)
2945 		goto do_rq;
2946 	cqe_cons %= sq->max_sw_wqe;
2947 
2948 	if (qp->sq.flushed) {
2949 		dev_dbg(&cq->hwq.pdev->dev,
2950 			"%s: QP in Flush QP = %p\n", __func__, qp);
2951 		goto sq_done;
2952 	}
2953 
2954 	/* Terminal CQE can also include aggregated successful CQEs prior.
2955 	 * So we must complete all CQEs from the current sq's cons to the
2956 	 * cq_cons with status OK
2957 	 */
2958 	cqe = *pcqe;
2959 	while (*budget) {
2960 		swq_last = sq->swq_last;
2961 		if (swq_last == cqe_cons)
2962 			break;
2963 		if (sq->swq[swq_last].flags & SQ_SEND_FLAGS_SIGNAL_COMP) {
2964 			memset(cqe, 0, sizeof(*cqe));
2965 			cqe->status = CQ_REQ_STATUS_OK;
2966 			cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
2967 			cqe->qp_handle = (u64)(unsigned long)qp;
2968 			cqe->src_qp = qp->id;
2969 			cqe->wr_id = sq->swq[swq_last].wr_id;
2970 			cqe->type = sq->swq[swq_last].type;
2971 			cqe++;
2972 			(*budget)--;
2973 		}
2974 		bnxt_qplib_hwq_incr_cons(sq->hwq.max_elements, &sq->hwq.cons,
2975 					 sq->swq[swq_last].slots, &sq->dbinfo.flags);
2976 		sq->swq_last = sq->swq[swq_last].next_idx;
2977 	}
2978 	*pcqe = cqe;
2979 	if (!(*budget) && swq_last != cqe_cons) {
2980 		/* Out of budget */
2981 		rc = -EAGAIN;
2982 		goto sq_done;
2983 	}
2984 sq_done:
2985 	if (rc)
2986 		return rc;
2987 do_rq:
2988 	cqe_cons = le16_to_cpu(hwcqe->rq_cons_idx);
2989 	if (cqe_cons == 0xFFFF) {
2990 		goto done;
2991 	} else if (cqe_cons > rq->max_wqe - 1) {
2992 		dev_err(&cq->hwq.pdev->dev,
2993 			"FP: CQ Processed terminal reported rq_cons_idx 0x%x exceeds max 0x%x\n",
2994 			cqe_cons, rq->max_wqe);
2995 		rc = -EINVAL;
2996 		goto done;
2997 	}
2998 
2999 	if (qp->rq.flushed) {
3000 		dev_dbg(&cq->hwq.pdev->dev,
3001 			"%s: QP in Flush QP = %p\n", __func__, qp);
3002 		rc = 0;
3003 		goto done;
3004 	}
3005 
3006 	/* Terminal CQE requires all posted RQEs to complete with FLUSHED_ERR
3007 	 * from the current rq->cons to the rq->prod regardless what the
3008 	 * rq->cons the terminal CQE indicates
3009 	 */
3010 
3011 	/* Add qp to flush list of the CQ */
3012 	bnxt_qplib_add_flush_qp(qp);
3013 done:
3014 	return rc;
3015 }
3016 
3017 static int bnxt_qplib_cq_process_cutoff(struct bnxt_qplib_cq *cq,
3018 					struct cq_cutoff *hwcqe)
3019 {
3020 	/* Check the Status */
3021 	if (hwcqe->status != CQ_CUTOFF_STATUS_OK) {
3022 		dev_err(&cq->hwq.pdev->dev,
3023 			"FP: CQ Process Cutoff Error status = 0x%x\n",
3024 			hwcqe->status);
3025 		return -EINVAL;
3026 	}
3027 	clear_bit(CQ_FLAGS_RESIZE_IN_PROG, &cq->flags);
3028 	wake_up_interruptible(&cq->waitq);
3029 
3030 	return 0;
3031 }
3032 
3033 int bnxt_qplib_process_flush_list(struct bnxt_qplib_cq *cq,
3034 				  struct bnxt_qplib_cqe *cqe,
3035 				  int num_cqes)
3036 {
3037 	struct bnxt_qplib_qp *qp = NULL;
3038 	u32 budget = num_cqes;
3039 	unsigned long flags;
3040 
3041 	spin_lock_irqsave(&cq->flush_lock, flags);
3042 	list_for_each_entry(qp, &cq->sqf_head, sq_flush) {
3043 		dev_dbg(&cq->hwq.pdev->dev, "FP: Flushing SQ QP= %p\n", qp);
3044 		__flush_sq(&qp->sq, qp, &cqe, &budget);
3045 	}
3046 
3047 	list_for_each_entry(qp, &cq->rqf_head, rq_flush) {
3048 		dev_dbg(&cq->hwq.pdev->dev, "FP: Flushing RQ QP= %p\n", qp);
3049 		__flush_rq(&qp->rq, qp, &cqe, &budget);
3050 	}
3051 	spin_unlock_irqrestore(&cq->flush_lock, flags);
3052 
3053 	return num_cqes - budget;
3054 }
3055 
3056 int bnxt_qplib_poll_cq(struct bnxt_qplib_cq *cq, struct bnxt_qplib_cqe *cqe,
3057 		       int num_cqes, struct bnxt_qplib_qp **lib_qp)
3058 {
3059 	struct cq_base *hw_cqe;
3060 	int budget, rc = 0;
3061 	u32 hw_polled = 0;
3062 	u8 type;
3063 
3064 	budget = num_cqes;
3065 
3066 	while (budget) {
3067 		hw_cqe = bnxt_qplib_get_qe(&cq->hwq, cq->hwq.cons, NULL);
3068 
3069 		/* Check for Valid bit */
3070 		if (!CQE_CMP_VALID(hw_cqe, cq->dbinfo.flags))
3071 			break;
3072 
3073 		/*
3074 		 * The valid test of the entry must be done first before
3075 		 * reading any further.
3076 		 */
3077 		dma_rmb();
3078 		/* From the device's respective CQE format to qplib_wc*/
3079 		type = hw_cqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
3080 		switch (type) {
3081 		case CQ_BASE_CQE_TYPE_REQ:
3082 			rc = bnxt_qplib_cq_process_req(cq,
3083 						       (struct cq_req *)hw_cqe,
3084 						       &cqe, &budget,
3085 						       cq->hwq.cons, lib_qp);
3086 			break;
3087 		case CQ_BASE_CQE_TYPE_RES_RC:
3088 			rc = bnxt_qplib_cq_process_res_rc(cq,
3089 							  (struct cq_res_rc *)
3090 							  hw_cqe, &cqe,
3091 							  &budget);
3092 			break;
3093 		case CQ_BASE_CQE_TYPE_RES_UD:
3094 			rc = bnxt_qplib_cq_process_res_ud
3095 					(cq, (struct cq_res_ud *)hw_cqe, &cqe,
3096 					 &budget);
3097 			break;
3098 		case CQ_BASE_CQE_TYPE_RES_RAWETH_QP1:
3099 			rc = bnxt_qplib_cq_process_res_raweth_qp1
3100 					(cq, (struct cq_res_raweth_qp1 *)
3101 					 hw_cqe, &cqe, &budget);
3102 			break;
3103 		case CQ_BASE_CQE_TYPE_TERMINAL:
3104 			rc = bnxt_qplib_cq_process_terminal
3105 					(cq, (struct cq_terminal *)hw_cqe,
3106 					 &cqe, &budget);
3107 			break;
3108 		case CQ_BASE_CQE_TYPE_CUT_OFF:
3109 			bnxt_qplib_cq_process_cutoff
3110 					(cq, (struct cq_cutoff *)hw_cqe);
3111 			/* Done processing this CQ */
3112 			goto exit;
3113 		default:
3114 			dev_err(&cq->hwq.pdev->dev,
3115 				"process_cq unknown type 0x%lx\n",
3116 				hw_cqe->cqe_type_toggle &
3117 				CQ_BASE_CQE_TYPE_MASK);
3118 			rc = -EINVAL;
3119 			break;
3120 		}
3121 		if (rc < 0) {
3122 			if (rc == -EAGAIN)
3123 				break;
3124 			/* Error while processing the CQE, just skip to the
3125 			 * next one
3126 			 */
3127 			if (type != CQ_BASE_CQE_TYPE_TERMINAL)
3128 				dev_err(&cq->hwq.pdev->dev,
3129 					"process_cqe error rc = 0x%x\n", rc);
3130 		}
3131 		hw_polled++;
3132 		bnxt_qplib_hwq_incr_cons(cq->hwq.max_elements, &cq->hwq.cons,
3133 					 1, &cq->dbinfo.flags);
3134 
3135 	}
3136 	if (hw_polled)
3137 		bnxt_qplib_ring_db(&cq->dbinfo, DBC_DBC_TYPE_CQ);
3138 exit:
3139 	return num_cqes - budget;
3140 }
3141 
3142 void bnxt_qplib_req_notify_cq(struct bnxt_qplib_cq *cq, u32 arm_type)
3143 {
3144 	cq->dbinfo.toggle = cq->toggle;
3145 	if (arm_type)
3146 		bnxt_qplib_ring_db(&cq->dbinfo, arm_type);
3147 	/* Using cq->arm_state variable to track whether to issue cq handler */
3148 	atomic_set(&cq->arm_state, 1);
3149 }
3150 
3151 void bnxt_qplib_flush_cqn_wq(struct bnxt_qplib_qp *qp)
3152 {
3153 	flush_workqueue(qp->scq->nq->cqn_wq);
3154 	if (qp->scq != qp->rcq)
3155 		flush_workqueue(qp->rcq->nq->cqn_wq);
3156 }
3157