xref: /linux/drivers/infiniband/hw/bnxt_re/qplib_fp.c (revision 68a052239fc4b351e961f698b824f7654a346091)
1 /*
2  * Broadcom NetXtreme-E RoCE driver.
3  *
4  * Copyright (c) 2016 - 2017, Broadcom. All rights reserved.  The term
5  * Broadcom refers to Broadcom Limited and/or its subsidiaries.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * BSD license below:
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  *
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in
21  *    the documentation and/or other materials provided with the
22  *    distribution.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS''
25  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
26  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS
28  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
32  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
33  * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
34  * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Description: Fast Path Operators
37  */
38 
39 #define dev_fmt(fmt) "QPLIB: " fmt
40 
41 #include <linux/interrupt.h>
42 #include <linux/spinlock.h>
43 #include <linux/sched.h>
44 #include <linux/slab.h>
45 #include <linux/pci.h>
46 #include <linux/delay.h>
47 #include <linux/prefetch.h>
48 #include <linux/if_ether.h>
49 #include <rdma/ib_mad.h>
50 
51 #include "roce_hsi.h"
52 
53 #include "qplib_res.h"
54 #include "qplib_rcfw.h"
55 #include "qplib_sp.h"
56 #include "qplib_fp.h"
57 #include <rdma/ib_addr.h>
58 #include "bnxt_ulp.h"
59 #include "bnxt_re.h"
60 #include "ib_verbs.h"
61 
62 static void __clean_cq(struct bnxt_qplib_cq *cq, u64 qp);
63 
64 static void bnxt_qplib_cancel_phantom_processing(struct bnxt_qplib_qp *qp)
65 {
66 	qp->sq.condition = false;
67 	qp->sq.send_phantom = false;
68 	qp->sq.single = false;
69 }
70 
71 /* Flush list */
72 static void __bnxt_qplib_add_flush_qp(struct bnxt_qplib_qp *qp)
73 {
74 	struct bnxt_qplib_cq *scq, *rcq;
75 
76 	scq = qp->scq;
77 	rcq = qp->rcq;
78 
79 	if (!qp->sq.flushed) {
80 		dev_dbg(&scq->hwq.pdev->dev,
81 			"FP: Adding to SQ Flush list = %p\n", qp);
82 		bnxt_qplib_cancel_phantom_processing(qp);
83 		list_add_tail(&qp->sq_flush, &scq->sqf_head);
84 		qp->sq.flushed = true;
85 	}
86 	if (!qp->srq) {
87 		if (!qp->rq.flushed) {
88 			dev_dbg(&rcq->hwq.pdev->dev,
89 				"FP: Adding to RQ Flush list = %p\n", qp);
90 			list_add_tail(&qp->rq_flush, &rcq->rqf_head);
91 			qp->rq.flushed = true;
92 		}
93 	}
94 }
95 
96 static void bnxt_qplib_acquire_cq_flush_locks(struct bnxt_qplib_qp *qp,
97 				       unsigned long *flags)
98 	__acquires(&qp->scq->flush_lock) __acquires(&qp->rcq->flush_lock)
99 {
100 	spin_lock_irqsave(&qp->scq->flush_lock, *flags);
101 	if (qp->scq == qp->rcq)
102 		__acquire(&qp->rcq->flush_lock);
103 	else
104 		spin_lock(&qp->rcq->flush_lock);
105 }
106 
107 static void bnxt_qplib_release_cq_flush_locks(struct bnxt_qplib_qp *qp,
108 				       unsigned long *flags)
109 	__releases(&qp->scq->flush_lock) __releases(&qp->rcq->flush_lock)
110 {
111 	if (qp->scq == qp->rcq)
112 		__release(&qp->rcq->flush_lock);
113 	else
114 		spin_unlock(&qp->rcq->flush_lock);
115 	spin_unlock_irqrestore(&qp->scq->flush_lock, *flags);
116 }
117 
118 void bnxt_qplib_add_flush_qp(struct bnxt_qplib_qp *qp)
119 {
120 	unsigned long flags;
121 
122 	bnxt_qplib_acquire_cq_flush_locks(qp, &flags);
123 	__bnxt_qplib_add_flush_qp(qp);
124 	bnxt_qplib_release_cq_flush_locks(qp, &flags);
125 }
126 
127 static void __bnxt_qplib_del_flush_qp(struct bnxt_qplib_qp *qp)
128 {
129 	if (qp->sq.flushed) {
130 		qp->sq.flushed = false;
131 		list_del(&qp->sq_flush);
132 	}
133 	if (!qp->srq) {
134 		if (qp->rq.flushed) {
135 			qp->rq.flushed = false;
136 			list_del(&qp->rq_flush);
137 		}
138 	}
139 }
140 
141 void bnxt_qplib_clean_qp(struct bnxt_qplib_qp *qp)
142 {
143 	unsigned long flags;
144 
145 	bnxt_qplib_acquire_cq_flush_locks(qp, &flags);
146 	__clean_cq(qp->scq, (u64)(unsigned long)qp);
147 	qp->sq.hwq.prod = 0;
148 	qp->sq.hwq.cons = 0;
149 	__clean_cq(qp->rcq, (u64)(unsigned long)qp);
150 	qp->rq.hwq.prod = 0;
151 	qp->rq.hwq.cons = 0;
152 
153 	__bnxt_qplib_del_flush_qp(qp);
154 	bnxt_qplib_release_cq_flush_locks(qp, &flags);
155 }
156 
157 static void bnxt_qpn_cqn_sched_task(struct work_struct *work)
158 {
159 	struct bnxt_qplib_nq_work *nq_work =
160 			container_of(work, struct bnxt_qplib_nq_work, work);
161 
162 	struct bnxt_qplib_cq *cq = nq_work->cq;
163 	struct bnxt_qplib_nq *nq = nq_work->nq;
164 
165 	if (cq && nq) {
166 		spin_lock_bh(&cq->compl_lock);
167 		if (atomic_read(&cq->arm_state) && nq->cqn_handler) {
168 			dev_dbg(&nq->pdev->dev,
169 				"%s:Trigger cq  = %p event nq = %p\n",
170 				__func__, cq, nq);
171 			nq->cqn_handler(nq, cq);
172 		}
173 		spin_unlock_bh(&cq->compl_lock);
174 	}
175 	kfree(nq_work);
176 }
177 
178 static void bnxt_qplib_free_qp_hdr_buf(struct bnxt_qplib_res *res,
179 				       struct bnxt_qplib_qp *qp)
180 {
181 	struct bnxt_qplib_q *rq = &qp->rq;
182 	struct bnxt_qplib_q *sq = &qp->sq;
183 
184 	if (qp->rq_hdr_buf)
185 		dma_free_coherent(&res->pdev->dev,
186 				  rq->max_wqe * qp->rq_hdr_buf_size,
187 				  qp->rq_hdr_buf, qp->rq_hdr_buf_map);
188 	if (qp->sq_hdr_buf)
189 		dma_free_coherent(&res->pdev->dev,
190 				  sq->max_wqe * qp->sq_hdr_buf_size,
191 				  qp->sq_hdr_buf, qp->sq_hdr_buf_map);
192 	qp->rq_hdr_buf = NULL;
193 	qp->sq_hdr_buf = NULL;
194 	qp->rq_hdr_buf_map = 0;
195 	qp->sq_hdr_buf_map = 0;
196 	qp->sq_hdr_buf_size = 0;
197 	qp->rq_hdr_buf_size = 0;
198 }
199 
200 static int bnxt_qplib_alloc_qp_hdr_buf(struct bnxt_qplib_res *res,
201 				       struct bnxt_qplib_qp *qp)
202 {
203 	struct bnxt_qplib_q *rq = &qp->rq;
204 	struct bnxt_qplib_q *sq = &qp->sq;
205 	int rc = 0;
206 
207 	if (qp->sq_hdr_buf_size && sq->max_wqe) {
208 		qp->sq_hdr_buf = dma_alloc_coherent(&res->pdev->dev,
209 					sq->max_wqe * qp->sq_hdr_buf_size,
210 					&qp->sq_hdr_buf_map, GFP_KERNEL);
211 		if (!qp->sq_hdr_buf) {
212 			rc = -ENOMEM;
213 			dev_err(&res->pdev->dev,
214 				"Failed to create sq_hdr_buf\n");
215 			goto fail;
216 		}
217 	}
218 
219 	if (qp->rq_hdr_buf_size && rq->max_wqe) {
220 		qp->rq_hdr_buf = dma_alloc_coherent(&res->pdev->dev,
221 						    rq->max_wqe *
222 						    qp->rq_hdr_buf_size,
223 						    &qp->rq_hdr_buf_map,
224 						    GFP_KERNEL);
225 		if (!qp->rq_hdr_buf) {
226 			rc = -ENOMEM;
227 			dev_err(&res->pdev->dev,
228 				"Failed to create rq_hdr_buf\n");
229 			goto fail;
230 		}
231 	}
232 	return 0;
233 
234 fail:
235 	bnxt_qplib_free_qp_hdr_buf(res, qp);
236 	return rc;
237 }
238 
239 static void clean_nq(struct bnxt_qplib_nq *nq, struct bnxt_qplib_cq *cq)
240 {
241 	struct bnxt_qplib_hwq *hwq = &nq->hwq;
242 	struct nq_base *nqe, **nq_ptr;
243 	int budget = nq->budget;
244 	uintptr_t q_handle;
245 	u16 type;
246 
247 	spin_lock_bh(&hwq->lock);
248 	/* Service the NQ until empty */
249 	while (budget--) {
250 		nq_ptr = (struct nq_base **)hwq->pbl_ptr;
251 		nqe = &nq_ptr[NQE_PG(hwq->cons)][NQE_IDX(hwq->cons)];
252 		if (!NQE_CMP_VALID(nqe, nq->nq_db.dbinfo.flags))
253 			break;
254 
255 		/*
256 		 * The valid test of the entry must be done first before
257 		 * reading any further.
258 		 */
259 		dma_rmb();
260 
261 		type = le16_to_cpu(nqe->info10_type) & NQ_BASE_TYPE_MASK;
262 		switch (type) {
263 		case NQ_BASE_TYPE_CQ_NOTIFICATION:
264 		{
265 			struct nq_cn *nqcne = (struct nq_cn *)nqe;
266 
267 			q_handle = le32_to_cpu(nqcne->cq_handle_low);
268 			q_handle |= (u64)le32_to_cpu(nqcne->cq_handle_high)
269 						     << 32;
270 			if ((unsigned long)cq == q_handle) {
271 				nqcne->cq_handle_low = 0;
272 				nqcne->cq_handle_high = 0;
273 				cq->cnq_events++;
274 			}
275 			break;
276 		}
277 		default:
278 			break;
279 		}
280 		bnxt_qplib_hwq_incr_cons(hwq->max_elements, &hwq->cons,
281 					 1, &nq->nq_db.dbinfo.flags);
282 	}
283 	spin_unlock_bh(&hwq->lock);
284 }
285 
286 /* Wait for receiving all NQEs for this CQ and clean the NQEs associated with
287  * this CQ.
288  */
289 static void __wait_for_all_nqes(struct bnxt_qplib_cq *cq, u16 cnq_events)
290 {
291 	u32 retry_cnt = 100;
292 
293 	while (retry_cnt--) {
294 		if (cnq_events == cq->cnq_events)
295 			return;
296 		usleep_range(50, 100);
297 		clean_nq(cq->nq, cq);
298 	}
299 }
300 
301 static void bnxt_qplib_service_nq(struct tasklet_struct *t)
302 {
303 	struct bnxt_qplib_nq *nq = from_tasklet(nq, t, nq_tasklet);
304 	struct bnxt_qplib_hwq *hwq = &nq->hwq;
305 	struct bnxt_qplib_cq *cq;
306 	int budget = nq->budget;
307 	struct nq_base *nqe;
308 	uintptr_t q_handle;
309 	u32 hw_polled = 0;
310 	u16 type;
311 
312 	spin_lock_bh(&hwq->lock);
313 	/* Service the NQ until empty */
314 	while (budget--) {
315 		nqe = bnxt_qplib_get_qe(hwq, hwq->cons, NULL);
316 		if (!NQE_CMP_VALID(nqe, nq->nq_db.dbinfo.flags))
317 			break;
318 
319 		/*
320 		 * The valid test of the entry must be done first before
321 		 * reading any further.
322 		 */
323 		dma_rmb();
324 
325 		type = le16_to_cpu(nqe->info10_type) & NQ_BASE_TYPE_MASK;
326 		switch (type) {
327 		case NQ_BASE_TYPE_CQ_NOTIFICATION:
328 		{
329 			struct nq_cn *nqcne = (struct nq_cn *)nqe;
330 			struct bnxt_re_cq *cq_p;
331 
332 			q_handle = le32_to_cpu(nqcne->cq_handle_low);
333 			q_handle |= (u64)le32_to_cpu(nqcne->cq_handle_high)
334 						     << 32;
335 			cq = (struct bnxt_qplib_cq *)(unsigned long)q_handle;
336 			if (!cq)
337 				break;
338 			cq->toggle = (le16_to_cpu(nqe->info10_type) &
339 					NQ_CN_TOGGLE_MASK) >> NQ_CN_TOGGLE_SFT;
340 			cq->dbinfo.toggle = cq->toggle;
341 			cq_p = container_of(cq, struct bnxt_re_cq, qplib_cq);
342 			if (cq_p->uctx_cq_page)
343 				*((u32 *)cq_p->uctx_cq_page) = cq->toggle;
344 
345 			bnxt_qplib_armen_db(&cq->dbinfo,
346 					    DBC_DBC_TYPE_CQ_ARMENA);
347 			spin_lock_bh(&cq->compl_lock);
348 			atomic_set(&cq->arm_state, 0);
349 			if (nq->cqn_handler(nq, (cq)))
350 				dev_warn(&nq->pdev->dev,
351 					 "cqn - type 0x%x not handled\n", type);
352 			cq->cnq_events++;
353 			spin_unlock_bh(&cq->compl_lock);
354 			break;
355 		}
356 		case NQ_BASE_TYPE_SRQ_EVENT:
357 		{
358 			struct bnxt_qplib_srq *srq;
359 			struct bnxt_re_srq *srq_p;
360 			struct nq_srq_event *nqsrqe =
361 						(struct nq_srq_event *)nqe;
362 
363 			q_handle = le32_to_cpu(nqsrqe->srq_handle_low);
364 			q_handle |= (u64)le32_to_cpu(nqsrqe->srq_handle_high)
365 				     << 32;
366 			srq = (struct bnxt_qplib_srq *)q_handle;
367 			srq->toggle = (le16_to_cpu(nqe->info10_type) & NQ_CN_TOGGLE_MASK)
368 				      >> NQ_CN_TOGGLE_SFT;
369 			srq->dbinfo.toggle = srq->toggle;
370 			srq_p = container_of(srq, struct bnxt_re_srq, qplib_srq);
371 			if (srq_p->uctx_srq_page)
372 				*((u32 *)srq_p->uctx_srq_page) = srq->toggle;
373 			bnxt_qplib_armen_db(&srq->dbinfo,
374 					    DBC_DBC_TYPE_SRQ_ARMENA);
375 			if (nq->srqn_handler(nq,
376 					     (struct bnxt_qplib_srq *)q_handle,
377 					     nqsrqe->event))
378 				dev_warn(&nq->pdev->dev,
379 					 "SRQ event 0x%x not handled\n",
380 					 nqsrqe->event);
381 			break;
382 		}
383 		case NQ_BASE_TYPE_DBQ_EVENT:
384 			break;
385 		default:
386 			dev_warn(&nq->pdev->dev,
387 				 "nqe with type = 0x%x not handled\n", type);
388 			break;
389 		}
390 		hw_polled++;
391 		bnxt_qplib_hwq_incr_cons(hwq->max_elements, &hwq->cons,
392 					 1, &nq->nq_db.dbinfo.flags);
393 	}
394 	if (hw_polled)
395 		bnxt_qplib_ring_nq_db(&nq->nq_db.dbinfo, nq->res->cctx, true);
396 	spin_unlock_bh(&hwq->lock);
397 }
398 
399 /* bnxt_re_synchronize_nq - self polling notification queue.
400  * @nq      -     notification queue pointer
401  *
402  * This function will start polling entries of a given notification queue
403  * for all pending  entries.
404  * This function is useful to synchronize notification entries while resources
405  * are going away.
406  */
407 
408 void bnxt_re_synchronize_nq(struct bnxt_qplib_nq *nq)
409 {
410 	int budget = nq->budget;
411 
412 	nq->budget = nq->hwq.max_elements;
413 	bnxt_qplib_service_nq(&nq->nq_tasklet);
414 	nq->budget = budget;
415 }
416 
417 static irqreturn_t bnxt_qplib_nq_irq(int irq, void *dev_instance)
418 {
419 	struct bnxt_qplib_nq *nq = dev_instance;
420 	struct bnxt_qplib_hwq *hwq = &nq->hwq;
421 	u32 sw_cons;
422 
423 	/* Prefetch the NQ element */
424 	sw_cons = HWQ_CMP(hwq->cons, hwq);
425 	prefetch(bnxt_qplib_get_qe(hwq, sw_cons, NULL));
426 
427 	/* Fan out to CPU affinitized kthreads? */
428 	tasklet_schedule(&nq->nq_tasklet);
429 
430 	return IRQ_HANDLED;
431 }
432 
433 void bnxt_qplib_nq_stop_irq(struct bnxt_qplib_nq *nq, bool kill)
434 {
435 	if (!nq->requested)
436 		return;
437 
438 	nq->requested = false;
439 	/* Mask h/w interrupt */
440 	bnxt_qplib_ring_nq_db(&nq->nq_db.dbinfo, nq->res->cctx, false);
441 	/* Sync with last running IRQ handler */
442 	synchronize_irq(nq->msix_vec);
443 	irq_set_affinity_hint(nq->msix_vec, NULL);
444 	free_irq(nq->msix_vec, nq);
445 	kfree(nq->name);
446 	nq->name = NULL;
447 
448 	if (kill)
449 		tasklet_kill(&nq->nq_tasklet);
450 	tasklet_disable(&nq->nq_tasklet);
451 }
452 
453 void bnxt_qplib_disable_nq(struct bnxt_qplib_nq *nq)
454 {
455 	if (nq->cqn_wq) {
456 		destroy_workqueue(nq->cqn_wq);
457 		nq->cqn_wq = NULL;
458 	}
459 
460 	/* Make sure the HW is stopped! */
461 	bnxt_qplib_nq_stop_irq(nq, true);
462 
463 	if (nq->nq_db.reg.bar_reg) {
464 		iounmap(nq->nq_db.reg.bar_reg);
465 		nq->nq_db.reg.bar_reg = NULL;
466 	}
467 
468 	nq->cqn_handler = NULL;
469 	nq->srqn_handler = NULL;
470 	nq->msix_vec = 0;
471 }
472 
473 int bnxt_qplib_nq_start_irq(struct bnxt_qplib_nq *nq, int nq_indx,
474 			    int msix_vector, bool need_init)
475 {
476 	struct bnxt_qplib_res *res = nq->res;
477 	int rc;
478 
479 	if (nq->requested)
480 		return -EFAULT;
481 
482 	nq->msix_vec = msix_vector;
483 	if (need_init)
484 		tasklet_setup(&nq->nq_tasklet, bnxt_qplib_service_nq);
485 	else
486 		tasklet_enable(&nq->nq_tasklet);
487 
488 	nq->name = kasprintf(GFP_KERNEL, "bnxt_re-nq-%d@pci:%s",
489 			     nq_indx, pci_name(res->pdev));
490 	if (!nq->name)
491 		return -ENOMEM;
492 	rc = request_irq(nq->msix_vec, bnxt_qplib_nq_irq, 0, nq->name, nq);
493 	if (rc) {
494 		kfree(nq->name);
495 		nq->name = NULL;
496 		tasklet_disable(&nq->nq_tasklet);
497 		return rc;
498 	}
499 
500 	cpumask_clear(&nq->mask);
501 	cpumask_set_cpu(nq_indx, &nq->mask);
502 	rc = irq_set_affinity_hint(nq->msix_vec, &nq->mask);
503 	if (rc) {
504 		dev_warn(&nq->pdev->dev,
505 			 "set affinity failed; vector: %d nq_idx: %d\n",
506 			 nq->msix_vec, nq_indx);
507 	}
508 	nq->requested = true;
509 	bnxt_qplib_ring_nq_db(&nq->nq_db.dbinfo, res->cctx, true);
510 
511 	return rc;
512 }
513 
514 static int bnxt_qplib_map_nq_db(struct bnxt_qplib_nq *nq,  u32 reg_offt)
515 {
516 	resource_size_t reg_base;
517 	struct bnxt_qplib_nq_db *nq_db;
518 	struct pci_dev *pdev;
519 
520 	pdev = nq->pdev;
521 	nq_db = &nq->nq_db;
522 
523 	nq_db->dbinfo.flags = 0;
524 	nq_db->reg.bar_id = NQ_CONS_PCI_BAR_REGION;
525 	nq_db->reg.bar_base = pci_resource_start(pdev, nq_db->reg.bar_id);
526 	if (!nq_db->reg.bar_base) {
527 		dev_err(&pdev->dev, "QPLIB: NQ BAR region %d resc start is 0!",
528 			nq_db->reg.bar_id);
529 		return -ENOMEM;
530 	}
531 
532 	reg_base = nq_db->reg.bar_base + reg_offt;
533 	/* Unconditionally map 8 bytes to support 57500 series */
534 	nq_db->reg.len = 8;
535 	nq_db->reg.bar_reg = ioremap(reg_base, nq_db->reg.len);
536 	if (!nq_db->reg.bar_reg) {
537 		dev_err(&pdev->dev, "QPLIB: NQ BAR region %d mapping failed",
538 			nq_db->reg.bar_id);
539 		return -ENOMEM;
540 	}
541 
542 	nq_db->dbinfo.db = nq_db->reg.bar_reg;
543 	nq_db->dbinfo.hwq = &nq->hwq;
544 	nq_db->dbinfo.xid = nq->ring_id;
545 
546 	return 0;
547 }
548 
549 int bnxt_qplib_enable_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq,
550 			 int nq_idx, int msix_vector, int bar_reg_offset,
551 			 cqn_handler_t cqn_handler,
552 			 srqn_handler_t srqn_handler)
553 {
554 	int rc;
555 
556 	nq->pdev = pdev;
557 	nq->cqn_handler = cqn_handler;
558 	nq->srqn_handler = srqn_handler;
559 	nq->load = 0;
560 
561 	/* Have a task to schedule CQ notifiers in post send case */
562 	nq->cqn_wq  = create_singlethread_workqueue("bnxt_qplib_nq");
563 	if (!nq->cqn_wq)
564 		return -ENOMEM;
565 
566 	rc = bnxt_qplib_map_nq_db(nq, bar_reg_offset);
567 	if (rc)
568 		goto fail;
569 
570 	rc = bnxt_qplib_nq_start_irq(nq, nq_idx, msix_vector, true);
571 	if (rc) {
572 		dev_err(&nq->pdev->dev,
573 			"Failed to request irq for nq-idx %d\n", nq_idx);
574 		goto fail;
575 	}
576 
577 	return 0;
578 fail:
579 	bnxt_qplib_disable_nq(nq);
580 	return rc;
581 }
582 
583 void bnxt_qplib_free_nq(struct bnxt_qplib_nq *nq)
584 {
585 	if (nq->hwq.max_elements) {
586 		bnxt_qplib_free_hwq(nq->res, &nq->hwq);
587 		nq->hwq.max_elements = 0;
588 	}
589 }
590 
591 int bnxt_qplib_alloc_nq(struct bnxt_qplib_res *res, struct bnxt_qplib_nq *nq)
592 {
593 	struct bnxt_qplib_hwq_attr hwq_attr = {};
594 	struct bnxt_qplib_sg_info sginfo = {};
595 
596 	nq->pdev = res->pdev;
597 	nq->res = res;
598 	if (!nq->hwq.max_elements ||
599 	    nq->hwq.max_elements > BNXT_QPLIB_NQE_MAX_CNT)
600 		nq->hwq.max_elements = BNXT_QPLIB_NQE_MAX_CNT;
601 
602 	sginfo.pgsize = PAGE_SIZE;
603 	sginfo.pgshft = PAGE_SHIFT;
604 	hwq_attr.res = res;
605 	hwq_attr.sginfo = &sginfo;
606 	hwq_attr.depth = nq->hwq.max_elements;
607 	hwq_attr.stride = sizeof(struct nq_base);
608 	hwq_attr.type = bnxt_qplib_get_hwq_type(nq->res);
609 	if (bnxt_qplib_alloc_init_hwq(&nq->hwq, &hwq_attr)) {
610 		dev_err(&nq->pdev->dev, "FP NQ allocation failed");
611 		return -ENOMEM;
612 	}
613 	nq->budget = 8;
614 	return 0;
615 }
616 
617 /* SRQ */
618 void bnxt_qplib_destroy_srq(struct bnxt_qplib_res *res,
619 			   struct bnxt_qplib_srq *srq)
620 {
621 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
622 	struct creq_destroy_srq_resp resp = {};
623 	struct bnxt_qplib_cmdqmsg msg = {};
624 	struct cmdq_destroy_srq req = {};
625 	int rc;
626 
627 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
628 				 CMDQ_BASE_OPCODE_DESTROY_SRQ,
629 				 sizeof(req));
630 
631 	/* Configure the request */
632 	req.srq_cid = cpu_to_le32(srq->id);
633 
634 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req), sizeof(resp), 0);
635 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
636 	kfree(srq->swq);
637 	if (rc)
638 		return;
639 	bnxt_qplib_free_hwq(res, &srq->hwq);
640 }
641 
642 int bnxt_qplib_create_srq(struct bnxt_qplib_res *res,
643 			  struct bnxt_qplib_srq *srq)
644 {
645 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
646 	struct bnxt_qplib_hwq_attr hwq_attr = {};
647 	struct creq_create_srq_resp resp = {};
648 	struct bnxt_qplib_cmdqmsg msg = {};
649 	struct cmdq_create_srq req = {};
650 	struct bnxt_qplib_pbl *pbl;
651 	u16 pg_sz_lvl;
652 	int rc, idx;
653 
654 	hwq_attr.res = res;
655 	hwq_attr.sginfo = &srq->sg_info;
656 	hwq_attr.depth = srq->max_wqe;
657 	hwq_attr.stride = srq->wqe_size;
658 	hwq_attr.type = HWQ_TYPE_QUEUE;
659 	rc = bnxt_qplib_alloc_init_hwq(&srq->hwq, &hwq_attr);
660 	if (rc)
661 		return rc;
662 	srq->dbinfo.flags = 0;
663 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
664 				 CMDQ_BASE_OPCODE_CREATE_SRQ,
665 				 sizeof(req));
666 
667 	/* Configure the request */
668 	req.dpi = cpu_to_le32(srq->dpi->dpi);
669 	req.srq_handle = cpu_to_le64((uintptr_t)srq);
670 
671 	req.srq_size = cpu_to_le16((u16)srq->hwq.max_elements);
672 	pbl = &srq->hwq.pbl[PBL_LVL_0];
673 	pg_sz_lvl = ((u16)bnxt_qplib_base_pg_size(&srq->hwq) <<
674 		     CMDQ_CREATE_SRQ_PG_SIZE_SFT);
675 	pg_sz_lvl |= (srq->hwq.level & CMDQ_CREATE_SRQ_LVL_MASK) <<
676 		      CMDQ_CREATE_SRQ_LVL_SFT;
677 	req.pg_size_lvl = cpu_to_le16(pg_sz_lvl);
678 	req.pbl = cpu_to_le64(pbl->pg_map_arr[0]);
679 	req.pd_id = cpu_to_le32(srq->pd->id);
680 	req.eventq_id = cpu_to_le16(srq->eventq_hw_ring_id);
681 
682 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req), sizeof(resp), 0);
683 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
684 	if (rc)
685 		goto fail;
686 
687 	spin_lock_init(&srq->lock);
688 	srq->start_idx = 0;
689 	srq->last_idx = srq->hwq.max_elements - 1;
690 	if (!srq->hwq.is_user) {
691 		srq->swq = kcalloc(srq->hwq.max_elements, sizeof(*srq->swq),
692 				   GFP_KERNEL);
693 		if (!srq->swq) {
694 			rc = -ENOMEM;
695 			goto fail;
696 		}
697 		for (idx = 0; idx < srq->hwq.max_elements; idx++)
698 			srq->swq[idx].next_idx = idx + 1;
699 		srq->swq[srq->last_idx].next_idx = -1;
700 	}
701 
702 	srq->id = le32_to_cpu(resp.xid);
703 	srq->dbinfo.hwq = &srq->hwq;
704 	srq->dbinfo.xid = srq->id;
705 	srq->dbinfo.db = srq->dpi->dbr;
706 	srq->dbinfo.max_slot = 1;
707 	srq->dbinfo.priv_db = res->dpi_tbl.priv_db;
708 	bnxt_qplib_armen_db(&srq->dbinfo, DBC_DBC_TYPE_SRQ_ARMENA);
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_query_srq(struct bnxt_qplib_res *res,
719 			 struct bnxt_qplib_srq *srq)
720 {
721 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
722 	struct creq_query_srq_resp resp = {};
723 	struct bnxt_qplib_cmdqmsg msg = {};
724 	struct bnxt_qplib_rcfw_sbuf sbuf;
725 	struct creq_query_srq_resp_sb *sb;
726 	struct cmdq_query_srq req = {};
727 	int rc;
728 
729 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
730 				 CMDQ_BASE_OPCODE_QUERY_SRQ,
731 				 sizeof(req));
732 
733 	/* Configure the request */
734 	sbuf.size = ALIGN(sizeof(*sb), BNXT_QPLIB_CMDQE_UNITS);
735 	sbuf.sb = dma_alloc_coherent(&rcfw->pdev->dev, sbuf.size,
736 				     &sbuf.dma_addr, GFP_KERNEL);
737 	if (!sbuf.sb)
738 		return -ENOMEM;
739 	req.resp_size = sbuf.size / BNXT_QPLIB_CMDQE_UNITS;
740 	req.srq_cid = cpu_to_le32(srq->id);
741 	sb = sbuf.sb;
742 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, &sbuf, sizeof(req),
743 				sizeof(resp), 0);
744 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
745 	if (!rc)
746 		srq->threshold = le16_to_cpu(sb->srq_limit);
747 	dma_free_coherent(&rcfw->pdev->dev, sbuf.size,
748 			  sbuf.sb, sbuf.dma_addr);
749 
750 	return rc;
751 }
752 
753 int bnxt_qplib_post_srq_recv(struct bnxt_qplib_srq *srq,
754 			     struct bnxt_qplib_swqe *wqe)
755 {
756 	struct bnxt_qplib_hwq *srq_hwq = &srq->hwq;
757 	struct rq_wqe *srqe;
758 	struct sq_sge *hw_sge;
759 	int i, next;
760 
761 	spin_lock(&srq_hwq->lock);
762 	if (srq->start_idx == srq->last_idx) {
763 		dev_err(&srq_hwq->pdev->dev,
764 			"FP: SRQ (0x%x) is full!\n", srq->id);
765 		spin_unlock(&srq_hwq->lock);
766 		return -EINVAL;
767 	}
768 	next = srq->start_idx;
769 	srq->start_idx = srq->swq[next].next_idx;
770 	spin_unlock(&srq_hwq->lock);
771 
772 	srqe = bnxt_qplib_get_qe(srq_hwq, srq_hwq->prod, NULL);
773 	memset(srqe, 0, srq->wqe_size);
774 	/* Calculate wqe_size16 and data_len */
775 	for (i = 0, hw_sge = (struct sq_sge *)srqe->data;
776 	     i < wqe->num_sge; i++, hw_sge++) {
777 		hw_sge->va_or_pa = cpu_to_le64(wqe->sg_list[i].addr);
778 		hw_sge->l_key = cpu_to_le32(wqe->sg_list[i].lkey);
779 		hw_sge->size = cpu_to_le32(wqe->sg_list[i].size);
780 	}
781 	srqe->wqe_type = wqe->type;
782 	srqe->flags = wqe->flags;
783 	srqe->wqe_size = wqe->num_sge +
784 			((offsetof(typeof(*srqe), data) + 15) >> 4);
785 	srqe->wr_id[0] = cpu_to_le32((u32)next);
786 	srq->swq[next].wr_id = wqe->wr_id;
787 
788 	bnxt_qplib_hwq_incr_prod(&srq->dbinfo, srq_hwq, srq->dbinfo.max_slot);
789 
790 	/* Ring DB */
791 	bnxt_qplib_ring_prod_db(&srq->dbinfo, DBC_DBC_TYPE_SRQ);
792 
793 	return 0;
794 }
795 
796 /* QP */
797 
798 static int bnxt_qplib_alloc_init_swq(struct bnxt_qplib_q *que)
799 {
800 	int indx;
801 
802 	que->swq = kcalloc(que->max_sw_wqe, sizeof(*que->swq), GFP_KERNEL);
803 	if (!que->swq)
804 		return -ENOMEM;
805 
806 	que->swq_start = 0;
807 	que->swq_last = que->max_sw_wqe - 1;
808 	for (indx = 0; indx < que->max_sw_wqe; indx++)
809 		que->swq[indx].next_idx = indx + 1;
810 	que->swq[que->swq_last].next_idx = 0; /* Make it circular */
811 	que->swq_last = 0;
812 
813 	return 0;
814 }
815 
816 int bnxt_qplib_create_qp1(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
817 {
818 	struct bnxt_qplib_hwq_attr hwq_attr = {};
819 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
820 	struct creq_create_qp1_resp resp = {};
821 	struct bnxt_qplib_cmdqmsg msg = {};
822 	struct bnxt_qplib_q *sq = &qp->sq;
823 	struct bnxt_qplib_q *rq = &qp->rq;
824 	struct cmdq_create_qp1 req = {};
825 	struct bnxt_qplib_pbl *pbl;
826 	u32 qp_flags = 0;
827 	u8 pg_sz_lvl;
828 	u32 tbl_indx;
829 	int rc;
830 
831 	sq->dbinfo.flags = 0;
832 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
833 				 CMDQ_BASE_OPCODE_CREATE_QP1,
834 				 sizeof(req));
835 	/* General */
836 	req.type = qp->type;
837 	req.dpi = cpu_to_le32(qp->dpi->dpi);
838 	req.qp_handle = cpu_to_le64(qp->qp_handle);
839 
840 	/* SQ */
841 	hwq_attr.res = res;
842 	hwq_attr.sginfo = &sq->sg_info;
843 	hwq_attr.stride = sizeof(struct sq_sge);
844 	hwq_attr.depth = bnxt_qplib_get_depth(sq, qp->wqe_mode, false);
845 	hwq_attr.type = HWQ_TYPE_QUEUE;
846 	rc = bnxt_qplib_alloc_init_hwq(&sq->hwq, &hwq_attr);
847 	if (rc)
848 		return rc;
849 
850 	rc = bnxt_qplib_alloc_init_swq(sq);
851 	if (rc)
852 		goto fail_sq;
853 
854 	req.sq_size = cpu_to_le32(bnxt_qplib_set_sq_size(sq, qp->wqe_mode));
855 	pbl = &sq->hwq.pbl[PBL_LVL_0];
856 	req.sq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
857 	pg_sz_lvl = (bnxt_qplib_base_pg_size(&sq->hwq) <<
858 		     CMDQ_CREATE_QP1_SQ_PG_SIZE_SFT);
859 	pg_sz_lvl |= (sq->hwq.level & CMDQ_CREATE_QP1_SQ_LVL_MASK);
860 	req.sq_pg_size_sq_lvl = pg_sz_lvl;
861 	req.sq_fwo_sq_sge =
862 		cpu_to_le16((sq->max_sge & CMDQ_CREATE_QP1_SQ_SGE_MASK) <<
863 			     CMDQ_CREATE_QP1_SQ_SGE_SFT);
864 	req.scq_cid = cpu_to_le32(qp->scq->id);
865 
866 	/* RQ */
867 	if (rq->max_wqe) {
868 		rq->dbinfo.flags = 0;
869 		hwq_attr.res = res;
870 		hwq_attr.sginfo = &rq->sg_info;
871 		hwq_attr.stride = sizeof(struct sq_sge);
872 		hwq_attr.depth = bnxt_qplib_get_depth(rq, qp->wqe_mode, false);
873 		hwq_attr.type = HWQ_TYPE_QUEUE;
874 		rc = bnxt_qplib_alloc_init_hwq(&rq->hwq, &hwq_attr);
875 		if (rc)
876 			goto sq_swq;
877 		rc = bnxt_qplib_alloc_init_swq(rq);
878 		if (rc)
879 			goto fail_rq;
880 		req.rq_size = cpu_to_le32(rq->max_wqe);
881 		pbl = &rq->hwq.pbl[PBL_LVL_0];
882 		req.rq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
883 		pg_sz_lvl = (bnxt_qplib_base_pg_size(&rq->hwq) <<
884 			     CMDQ_CREATE_QP1_RQ_PG_SIZE_SFT);
885 		pg_sz_lvl |= (rq->hwq.level & CMDQ_CREATE_QP1_RQ_LVL_MASK);
886 		req.rq_pg_size_rq_lvl = pg_sz_lvl;
887 		req.rq_fwo_rq_sge =
888 			cpu_to_le16((rq->max_sge &
889 				     CMDQ_CREATE_QP1_RQ_SGE_MASK) <<
890 				    CMDQ_CREATE_QP1_RQ_SGE_SFT);
891 	}
892 	req.rcq_cid = cpu_to_le32(qp->rcq->id);
893 	/* Header buffer - allow hdr_buf pass in */
894 	rc = bnxt_qplib_alloc_qp_hdr_buf(res, qp);
895 	if (rc) {
896 		rc = -ENOMEM;
897 		goto rq_rwq;
898 	}
899 	qp_flags |= CMDQ_CREATE_QP1_QP_FLAGS_RESERVED_LKEY_ENABLE;
900 	req.qp_flags = cpu_to_le32(qp_flags);
901 	req.pd_id = cpu_to_le32(qp->pd->id);
902 
903 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req), sizeof(resp), 0);
904 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
905 	if (rc)
906 		goto fail;
907 
908 	qp->id = le32_to_cpu(resp.xid);
909 	qp->cur_qp_state = CMDQ_MODIFY_QP_NEW_STATE_RESET;
910 	qp->cctx = res->cctx;
911 	sq->dbinfo.hwq = &sq->hwq;
912 	sq->dbinfo.xid = qp->id;
913 	sq->dbinfo.db = qp->dpi->dbr;
914 	sq->dbinfo.max_slot = bnxt_qplib_set_sq_max_slot(qp->wqe_mode);
915 	if (rq->max_wqe) {
916 		rq->dbinfo.hwq = &rq->hwq;
917 		rq->dbinfo.xid = qp->id;
918 		rq->dbinfo.db = qp->dpi->dbr;
919 		rq->dbinfo.max_slot = bnxt_qplib_set_rq_max_slot(rq->wqe_size);
920 	}
921 	tbl_indx = map_qp_id_to_tbl_indx(qp->id, rcfw);
922 	rcfw->qp_tbl[tbl_indx].qp_id = qp->id;
923 	rcfw->qp_tbl[tbl_indx].qp_handle = (void *)qp;
924 
925 	return 0;
926 
927 fail:
928 	bnxt_qplib_free_qp_hdr_buf(res, qp);
929 rq_rwq:
930 	kfree(rq->swq);
931 fail_rq:
932 	bnxt_qplib_free_hwq(res, &rq->hwq);
933 sq_swq:
934 	kfree(sq->swq);
935 fail_sq:
936 	bnxt_qplib_free_hwq(res, &sq->hwq);
937 	return rc;
938 }
939 
940 static void bnxt_qplib_init_psn_ptr(struct bnxt_qplib_qp *qp, int size)
941 {
942 	struct bnxt_qplib_hwq *hwq;
943 	struct bnxt_qplib_q *sq;
944 	u64 fpsne, psn_pg;
945 	u16 indx_pad = 0;
946 
947 	sq = &qp->sq;
948 	hwq = &sq->hwq;
949 	/* First psn entry */
950 	fpsne = (u64)bnxt_qplib_get_qe(hwq, hwq->depth, &psn_pg);
951 	if (!IS_ALIGNED(fpsne, PAGE_SIZE))
952 		indx_pad = (fpsne & ~PAGE_MASK) / size;
953 	hwq->pad_pgofft = indx_pad;
954 	hwq->pad_pg = (u64 *)psn_pg;
955 	hwq->pad_stride = size;
956 }
957 
958 int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
959 {
960 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
961 	struct bnxt_qplib_hwq_attr hwq_attr = {};
962 	struct bnxt_qplib_sg_info sginfo = {};
963 	struct creq_create_qp_resp resp = {};
964 	struct bnxt_qplib_cmdqmsg msg = {};
965 	struct bnxt_qplib_q *sq = &qp->sq;
966 	struct bnxt_qplib_q *rq = &qp->rq;
967 	struct cmdq_create_qp req = {};
968 	int rc, req_size, psn_sz = 0;
969 	struct bnxt_qplib_hwq *xrrq;
970 	struct bnxt_qplib_pbl *pbl;
971 	u32 qp_flags = 0;
972 	u8 pg_sz_lvl;
973 	u32 tbl_indx;
974 	u16 nsge;
975 
976 	qp->is_host_msn_tbl = _is_host_msn_table(res->dattr->dev_cap_flags2);
977 	sq->dbinfo.flags = 0;
978 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
979 				 CMDQ_BASE_OPCODE_CREATE_QP,
980 				 sizeof(req));
981 
982 	/* General */
983 	req.type = qp->type;
984 	req.dpi = cpu_to_le32(qp->dpi->dpi);
985 	req.qp_handle = cpu_to_le64(qp->qp_handle);
986 
987 	/* SQ */
988 	if (qp->type == CMDQ_CREATE_QP_TYPE_RC) {
989 		psn_sz = bnxt_qplib_is_chip_gen_p5_p7(res->cctx) ?
990 			 sizeof(struct sq_psn_search_ext) :
991 			 sizeof(struct sq_psn_search);
992 
993 		if (qp->is_host_msn_tbl) {
994 			psn_sz = sizeof(struct sq_msn_search);
995 			qp->msn = 0;
996 		}
997 	}
998 
999 	hwq_attr.res = res;
1000 	hwq_attr.sginfo = &sq->sg_info;
1001 	hwq_attr.stride = sizeof(struct sq_sge);
1002 	hwq_attr.depth = bnxt_qplib_get_depth(sq, qp->wqe_mode, true);
1003 	hwq_attr.aux_stride = psn_sz;
1004 	hwq_attr.aux_depth = psn_sz ? bnxt_qplib_set_sq_size(sq, qp->wqe_mode)
1005 				    : 0;
1006 	/* Update msn tbl size */
1007 	if (qp->is_host_msn_tbl && psn_sz) {
1008 		if (qp->wqe_mode == BNXT_QPLIB_WQE_MODE_STATIC)
1009 			hwq_attr.aux_depth =
1010 				roundup_pow_of_two(bnxt_qplib_set_sq_size(sq, qp->wqe_mode));
1011 		else
1012 			hwq_attr.aux_depth =
1013 				roundup_pow_of_two(bnxt_qplib_set_sq_size(sq, qp->wqe_mode)) / 2;
1014 		qp->msn_tbl_sz = hwq_attr.aux_depth;
1015 		qp->msn = 0;
1016 	}
1017 
1018 	hwq_attr.type = HWQ_TYPE_QUEUE;
1019 	rc = bnxt_qplib_alloc_init_hwq(&sq->hwq, &hwq_attr);
1020 	if (rc)
1021 		return rc;
1022 
1023 	if (!sq->hwq.is_user) {
1024 		rc = bnxt_qplib_alloc_init_swq(sq);
1025 		if (rc)
1026 			goto fail_sq;
1027 
1028 		if (psn_sz)
1029 			bnxt_qplib_init_psn_ptr(qp, psn_sz);
1030 	}
1031 	req.sq_size = cpu_to_le32(bnxt_qplib_set_sq_size(sq, qp->wqe_mode));
1032 	pbl = &sq->hwq.pbl[PBL_LVL_0];
1033 	req.sq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
1034 	pg_sz_lvl = (bnxt_qplib_base_pg_size(&sq->hwq) <<
1035 		     CMDQ_CREATE_QP_SQ_PG_SIZE_SFT);
1036 	pg_sz_lvl |= (sq->hwq.level & CMDQ_CREATE_QP_SQ_LVL_MASK);
1037 	req.sq_pg_size_sq_lvl = pg_sz_lvl;
1038 	req.sq_fwo_sq_sge =
1039 		cpu_to_le16(((sq->max_sge & CMDQ_CREATE_QP_SQ_SGE_MASK) <<
1040 			     CMDQ_CREATE_QP_SQ_SGE_SFT) | 0);
1041 	req.scq_cid = cpu_to_le32(qp->scq->id);
1042 
1043 	/* RQ */
1044 	if (!qp->srq) {
1045 		rq->dbinfo.flags = 0;
1046 		hwq_attr.res = res;
1047 		hwq_attr.sginfo = &rq->sg_info;
1048 		hwq_attr.stride = sizeof(struct sq_sge);
1049 		hwq_attr.depth = bnxt_qplib_get_depth(rq, qp->wqe_mode, false);
1050 		hwq_attr.aux_stride = 0;
1051 		hwq_attr.aux_depth = 0;
1052 		hwq_attr.type = HWQ_TYPE_QUEUE;
1053 		rc = bnxt_qplib_alloc_init_hwq(&rq->hwq, &hwq_attr);
1054 		if (rc)
1055 			goto sq_swq;
1056 		if (!rq->hwq.is_user) {
1057 			rc = bnxt_qplib_alloc_init_swq(rq);
1058 			if (rc)
1059 				goto fail_rq;
1060 		}
1061 
1062 		req.rq_size = cpu_to_le32(rq->max_wqe);
1063 		pbl = &rq->hwq.pbl[PBL_LVL_0];
1064 		req.rq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
1065 		pg_sz_lvl = (bnxt_qplib_base_pg_size(&rq->hwq) <<
1066 			     CMDQ_CREATE_QP_RQ_PG_SIZE_SFT);
1067 		pg_sz_lvl |= (rq->hwq.level & CMDQ_CREATE_QP_RQ_LVL_MASK);
1068 		req.rq_pg_size_rq_lvl = pg_sz_lvl;
1069 		nsge = (qp->wqe_mode == BNXT_QPLIB_WQE_MODE_STATIC) ?
1070 			6 : rq->max_sge;
1071 		req.rq_fwo_rq_sge =
1072 			cpu_to_le16(((nsge &
1073 				      CMDQ_CREATE_QP_RQ_SGE_MASK) <<
1074 				     CMDQ_CREATE_QP_RQ_SGE_SFT) | 0);
1075 	} else {
1076 		/* SRQ */
1077 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_SRQ_USED;
1078 		req.srq_cid = cpu_to_le32(qp->srq->id);
1079 	}
1080 	req.rcq_cid = cpu_to_le32(qp->rcq->id);
1081 
1082 	qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_RESERVED_LKEY_ENABLE;
1083 	qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_FR_PMR_ENABLED;
1084 	if (qp->sig_type)
1085 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_FORCE_COMPLETION;
1086 	if (qp->wqe_mode == BNXT_QPLIB_WQE_MODE_VARIABLE)
1087 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_VARIABLE_SIZED_WQE_ENABLED;
1088 	if (bnxt_ext_stats_supported(res->cctx, res->dattr->dev_cap_flags, res->is_vf))
1089 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_EXT_STATS_ENABLED;
1090 
1091 	req.qp_flags = cpu_to_le32(qp_flags);
1092 
1093 	/* ORRQ and IRRQ */
1094 	if (psn_sz) {
1095 		xrrq = &qp->orrq;
1096 		xrrq->max_elements =
1097 			ORD_LIMIT_TO_ORRQ_SLOTS(qp->max_rd_atomic);
1098 		req_size = xrrq->max_elements *
1099 			   BNXT_QPLIB_MAX_ORRQE_ENTRY_SIZE + PAGE_SIZE - 1;
1100 		req_size &= ~(PAGE_SIZE - 1);
1101 		sginfo.pgsize = req_size;
1102 		sginfo.pgshft = PAGE_SHIFT;
1103 
1104 		hwq_attr.res = res;
1105 		hwq_attr.sginfo = &sginfo;
1106 		hwq_attr.depth = xrrq->max_elements;
1107 		hwq_attr.stride = BNXT_QPLIB_MAX_ORRQE_ENTRY_SIZE;
1108 		hwq_attr.aux_stride = 0;
1109 		hwq_attr.aux_depth = 0;
1110 		hwq_attr.type = HWQ_TYPE_CTX;
1111 		rc = bnxt_qplib_alloc_init_hwq(xrrq, &hwq_attr);
1112 		if (rc)
1113 			goto rq_swq;
1114 		pbl = &xrrq->pbl[PBL_LVL_0];
1115 		req.orrq_addr = cpu_to_le64(pbl->pg_map_arr[0]);
1116 
1117 		xrrq = &qp->irrq;
1118 		xrrq->max_elements = IRD_LIMIT_TO_IRRQ_SLOTS(
1119 						qp->max_dest_rd_atomic);
1120 		req_size = xrrq->max_elements *
1121 			   BNXT_QPLIB_MAX_IRRQE_ENTRY_SIZE + PAGE_SIZE - 1;
1122 		req_size &= ~(PAGE_SIZE - 1);
1123 		sginfo.pgsize = req_size;
1124 		hwq_attr.depth =  xrrq->max_elements;
1125 		hwq_attr.stride = BNXT_QPLIB_MAX_IRRQE_ENTRY_SIZE;
1126 		rc = bnxt_qplib_alloc_init_hwq(xrrq, &hwq_attr);
1127 		if (rc)
1128 			goto fail_orrq;
1129 
1130 		pbl = &xrrq->pbl[PBL_LVL_0];
1131 		req.irrq_addr = cpu_to_le64(pbl->pg_map_arr[0]);
1132 	}
1133 	req.pd_id = cpu_to_le32(qp->pd->id);
1134 
1135 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
1136 				sizeof(resp), 0);
1137 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
1138 	if (rc)
1139 		goto fail;
1140 
1141 	qp->id = le32_to_cpu(resp.xid);
1142 	qp->cur_qp_state = CMDQ_MODIFY_QP_NEW_STATE_RESET;
1143 	INIT_LIST_HEAD(&qp->sq_flush);
1144 	INIT_LIST_HEAD(&qp->rq_flush);
1145 	qp->cctx = res->cctx;
1146 	sq->dbinfo.hwq = &sq->hwq;
1147 	sq->dbinfo.xid = qp->id;
1148 	sq->dbinfo.db = qp->dpi->dbr;
1149 	sq->dbinfo.max_slot = bnxt_qplib_set_sq_max_slot(qp->wqe_mode);
1150 	if (rq->max_wqe) {
1151 		rq->dbinfo.hwq = &rq->hwq;
1152 		rq->dbinfo.xid = qp->id;
1153 		rq->dbinfo.db = qp->dpi->dbr;
1154 		rq->dbinfo.max_slot = bnxt_qplib_set_rq_max_slot(rq->wqe_size);
1155 	}
1156 	spin_lock_bh(&rcfw->tbl_lock);
1157 	tbl_indx = map_qp_id_to_tbl_indx(qp->id, rcfw);
1158 	rcfw->qp_tbl[tbl_indx].qp_id = qp->id;
1159 	rcfw->qp_tbl[tbl_indx].qp_handle = (void *)qp;
1160 	spin_unlock_bh(&rcfw->tbl_lock);
1161 
1162 	return 0;
1163 fail:
1164 	bnxt_qplib_free_hwq(res, &qp->irrq);
1165 fail_orrq:
1166 	bnxt_qplib_free_hwq(res, &qp->orrq);
1167 rq_swq:
1168 	kfree(rq->swq);
1169 fail_rq:
1170 	bnxt_qplib_free_hwq(res, &rq->hwq);
1171 sq_swq:
1172 	kfree(sq->swq);
1173 fail_sq:
1174 	bnxt_qplib_free_hwq(res, &sq->hwq);
1175 	return rc;
1176 }
1177 
1178 static void __modify_flags_from_init_state(struct bnxt_qplib_qp *qp)
1179 {
1180 	switch (qp->state) {
1181 	case CMDQ_MODIFY_QP_NEW_STATE_RTR:
1182 		/* INIT->RTR, configure the path_mtu to the default
1183 		 * 2048 if not being requested
1184 		 */
1185 		if (!(qp->modify_flags &
1186 		    CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU)) {
1187 			qp->modify_flags |=
1188 				CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU;
1189 			qp->path_mtu =
1190 				CMDQ_MODIFY_QP_PATH_MTU_MTU_2048;
1191 		}
1192 		/* Bono FW require the max_dest_rd_atomic to be >= 1 */
1193 		if (qp->max_dest_rd_atomic < 1)
1194 			qp->max_dest_rd_atomic = 1;
1195 		qp->modify_flags &= ~CMDQ_MODIFY_QP_MODIFY_MASK_SRC_MAC;
1196 		/* Bono FW 20.6.5 requires SGID_INDEX configuration */
1197 		if (!(qp->modify_flags &
1198 		    CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX)) {
1199 			qp->modify_flags |=
1200 				CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX;
1201 			qp->ah.sgid_index = 0;
1202 		}
1203 		break;
1204 	default:
1205 		break;
1206 	}
1207 }
1208 
1209 static void __modify_flags_from_rtr_state(struct bnxt_qplib_qp *qp)
1210 {
1211 	switch (qp->state) {
1212 	case CMDQ_MODIFY_QP_NEW_STATE_RTS:
1213 		/* Bono FW requires the max_rd_atomic to be >= 1 */
1214 		if (qp->max_rd_atomic < 1)
1215 			qp->max_rd_atomic = 1;
1216 		/* Bono FW does not allow PKEY_INDEX,
1217 		 * DGID, FLOW_LABEL, SGID_INDEX, HOP_LIMIT,
1218 		 * TRAFFIC_CLASS, DEST_MAC, PATH_MTU, RQ_PSN,
1219 		 * MIN_RNR_TIMER, MAX_DEST_RD_ATOMIC, DEST_QP_ID
1220 		 * modification
1221 		 */
1222 		qp->modify_flags &=
1223 			~(CMDQ_MODIFY_QP_MODIFY_MASK_PKEY |
1224 			  CMDQ_MODIFY_QP_MODIFY_MASK_DGID |
1225 			  CMDQ_MODIFY_QP_MODIFY_MASK_FLOW_LABEL |
1226 			  CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX |
1227 			  CMDQ_MODIFY_QP_MODIFY_MASK_HOP_LIMIT |
1228 			  CMDQ_MODIFY_QP_MODIFY_MASK_TRAFFIC_CLASS |
1229 			  CMDQ_MODIFY_QP_MODIFY_MASK_DEST_MAC |
1230 			  CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU |
1231 			  CMDQ_MODIFY_QP_MODIFY_MASK_RQ_PSN |
1232 			  CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER |
1233 			  CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC |
1234 			  CMDQ_MODIFY_QP_MODIFY_MASK_DEST_QP_ID);
1235 		break;
1236 	default:
1237 		break;
1238 	}
1239 }
1240 
1241 static void __filter_modify_flags(struct bnxt_qplib_qp *qp)
1242 {
1243 	switch (qp->cur_qp_state) {
1244 	case CMDQ_MODIFY_QP_NEW_STATE_RESET:
1245 		break;
1246 	case CMDQ_MODIFY_QP_NEW_STATE_INIT:
1247 		__modify_flags_from_init_state(qp);
1248 		break;
1249 	case CMDQ_MODIFY_QP_NEW_STATE_RTR:
1250 		__modify_flags_from_rtr_state(qp);
1251 		break;
1252 	case CMDQ_MODIFY_QP_NEW_STATE_RTS:
1253 		break;
1254 	case CMDQ_MODIFY_QP_NEW_STATE_SQD:
1255 		break;
1256 	case CMDQ_MODIFY_QP_NEW_STATE_SQE:
1257 		break;
1258 	case CMDQ_MODIFY_QP_NEW_STATE_ERR:
1259 		break;
1260 	default:
1261 		break;
1262 	}
1263 }
1264 
1265 static void bnxt_set_mandatory_attributes(struct bnxt_qplib_res *res,
1266 					  struct bnxt_qplib_qp *qp,
1267 					  struct cmdq_modify_qp *req)
1268 {
1269 	u32 mandatory_flags = 0;
1270 
1271 	if (qp->type == CMDQ_MODIFY_QP_QP_TYPE_RC)
1272 		mandatory_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_ACCESS;
1273 
1274 	if (qp->cur_qp_state == CMDQ_MODIFY_QP_NEW_STATE_INIT &&
1275 	    qp->state == CMDQ_MODIFY_QP_NEW_STATE_RTR) {
1276 		if (qp->type == CMDQ_MODIFY_QP_QP_TYPE_RC && qp->srq)
1277 			req->flags = cpu_to_le16(CMDQ_MODIFY_QP_FLAGS_SRQ_USED);
1278 		mandatory_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_PKEY;
1279 	}
1280 
1281 	if (_is_min_rnr_in_rtr_rts_mandatory(res->dattr->dev_cap_flags2) &&
1282 	    (qp->cur_qp_state == CMDQ_MODIFY_QP_NEW_STATE_RTR &&
1283 	     qp->state == CMDQ_MODIFY_QP_NEW_STATE_RTS)) {
1284 		if (qp->type == CMDQ_MODIFY_QP_QP_TYPE_RC)
1285 			mandatory_flags |=
1286 				CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER;
1287 	}
1288 
1289 	if (qp->type == CMDQ_MODIFY_QP_QP_TYPE_UD ||
1290 	    qp->type == CMDQ_MODIFY_QP_QP_TYPE_GSI)
1291 		mandatory_flags |= CMDQ_MODIFY_QP_MODIFY_MASK_QKEY;
1292 
1293 	qp->modify_flags |= mandatory_flags;
1294 	req->qp_type = qp->type;
1295 }
1296 
1297 static bool is_optimized_state_transition(struct bnxt_qplib_qp *qp)
1298 {
1299 	if ((qp->cur_qp_state == CMDQ_MODIFY_QP_NEW_STATE_INIT &&
1300 	     qp->state == CMDQ_MODIFY_QP_NEW_STATE_RTR) ||
1301 	    (qp->cur_qp_state == CMDQ_MODIFY_QP_NEW_STATE_RTR &&
1302 	     qp->state == CMDQ_MODIFY_QP_NEW_STATE_RTS))
1303 		return true;
1304 
1305 	return false;
1306 }
1307 
1308 int bnxt_qplib_modify_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
1309 {
1310 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &res->sgid_tbl;
1311 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1312 	struct creq_modify_qp_resp resp = {};
1313 	struct bnxt_qplib_cmdqmsg msg = {};
1314 	struct cmdq_modify_qp req = {};
1315 	u16 vlan_pcp_vlan_dei_vlan_id;
1316 	u32 temp32[4];
1317 	u32 bmask;
1318 	int rc;
1319 
1320 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
1321 				 CMDQ_BASE_OPCODE_MODIFY_QP,
1322 				 sizeof(req));
1323 
1324 	/* Filter out the qp_attr_mask based on the state->new transition */
1325 	__filter_modify_flags(qp);
1326 	if (qp->modify_flags & CMDQ_MODIFY_QP_MODIFY_MASK_STATE) {
1327 		/* Set mandatory attributes for INIT -> RTR and RTR -> RTS transition */
1328 		if (_is_optimize_modify_qp_supported(res->dattr->dev_cap_flags2) &&
1329 		    is_optimized_state_transition(qp))
1330 			bnxt_set_mandatory_attributes(res, qp, &req);
1331 	}
1332 	bmask = qp->modify_flags;
1333 	req.modify_mask = cpu_to_le32(qp->modify_flags);
1334 	req.qp_cid = cpu_to_le32(qp->id);
1335 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_STATE) {
1336 		req.network_type_en_sqd_async_notify_new_state =
1337 				(qp->state & CMDQ_MODIFY_QP_NEW_STATE_MASK) |
1338 				(qp->en_sqd_async_notify ?
1339 					CMDQ_MODIFY_QP_EN_SQD_ASYNC_NOTIFY : 0);
1340 	}
1341 	req.network_type_en_sqd_async_notify_new_state |= qp->nw_type;
1342 
1343 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_ACCESS)
1344 		req.access = qp->access;
1345 
1346 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_PKEY)
1347 		req.pkey = cpu_to_le16(IB_DEFAULT_PKEY_FULL);
1348 
1349 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_QKEY)
1350 		req.qkey = cpu_to_le32(qp->qkey);
1351 
1352 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DGID) {
1353 		memcpy(temp32, qp->ah.dgid.data, sizeof(struct bnxt_qplib_gid));
1354 		req.dgid[0] = cpu_to_le32(temp32[0]);
1355 		req.dgid[1] = cpu_to_le32(temp32[1]);
1356 		req.dgid[2] = cpu_to_le32(temp32[2]);
1357 		req.dgid[3] = cpu_to_le32(temp32[3]);
1358 	}
1359 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_FLOW_LABEL)
1360 		req.flow_label = cpu_to_le32(qp->ah.flow_label);
1361 
1362 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX) {
1363 		if (qp->type == CMDQ_CREATE_QP_TYPE_RAW_ETHERTYPE)
1364 			req.sgid_index =
1365 				cpu_to_le16(sgid_tbl->hw_id[qp->ugid_index]);
1366 		else
1367 			req.sgid_index =
1368 				cpu_to_le16(sgid_tbl->hw_id[qp->ah.sgid_index]);
1369 	}
1370 
1371 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_HOP_LIMIT)
1372 		req.hop_limit = qp->ah.hop_limit;
1373 
1374 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_TRAFFIC_CLASS)
1375 		req.traffic_class = qp->ah.traffic_class;
1376 
1377 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DEST_MAC)
1378 		memcpy(req.dest_mac, qp->ah.dmac, 6);
1379 
1380 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU)
1381 		req.path_mtu_pingpong_push_enable |= qp->path_mtu;
1382 
1383 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_TIMEOUT)
1384 		req.timeout = qp->timeout;
1385 
1386 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RETRY_CNT)
1387 		req.retry_cnt = qp->retry_cnt;
1388 
1389 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RNR_RETRY)
1390 		req.rnr_retry = qp->rnr_retry;
1391 
1392 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER)
1393 		req.min_rnr_timer = qp->min_rnr_timer;
1394 
1395 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RQ_PSN)
1396 		req.rq_psn = cpu_to_le32(qp->rq.psn);
1397 
1398 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_SQ_PSN)
1399 		req.sq_psn = cpu_to_le32(qp->sq.psn);
1400 
1401 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MAX_RD_ATOMIC)
1402 		req.max_rd_atomic =
1403 			ORD_LIMIT_TO_ORRQ_SLOTS(qp->max_rd_atomic);
1404 
1405 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC)
1406 		req.max_dest_rd_atomic =
1407 			IRD_LIMIT_TO_IRRQ_SLOTS(qp->max_dest_rd_atomic);
1408 
1409 	req.sq_size = cpu_to_le32(qp->sq.hwq.max_elements);
1410 	req.rq_size = cpu_to_le32(qp->rq.hwq.max_elements);
1411 	req.sq_sge = cpu_to_le16(qp->sq.max_sge);
1412 	req.rq_sge = cpu_to_le16(qp->rq.max_sge);
1413 	req.max_inline_data = cpu_to_le32(qp->max_inline_data);
1414 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DEST_QP_ID)
1415 		req.dest_qp_id = cpu_to_le32(qp->dest_qpn);
1416 
1417 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_VLAN_ID) {
1418 		vlan_pcp_vlan_dei_vlan_id =
1419 			((res->sgid_tbl.tbl[qp->ah.sgid_index].vlan_id <<
1420 			  CMDQ_MODIFY_QP_VLAN_ID_SFT) &
1421 			 CMDQ_MODIFY_QP_VLAN_ID_MASK);
1422 		vlan_pcp_vlan_dei_vlan_id |=
1423 			((qp->ah.sl << CMDQ_MODIFY_QP_VLAN_PCP_SFT) &
1424 			 CMDQ_MODIFY_QP_VLAN_PCP_MASK);
1425 		req.vlan_pcp_vlan_dei_vlan_id = cpu_to_le16(vlan_pcp_vlan_dei_vlan_id);
1426 	}
1427 
1428 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),  sizeof(resp), 0);
1429 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
1430 	if (rc)
1431 		return rc;
1432 	qp->cur_qp_state = qp->state;
1433 	return 0;
1434 }
1435 
1436 int bnxt_qplib_query_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
1437 {
1438 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1439 	struct creq_query_qp_resp resp = {};
1440 	struct bnxt_qplib_cmdqmsg msg = {};
1441 	struct bnxt_qplib_rcfw_sbuf sbuf;
1442 	struct creq_query_qp_resp_sb *sb;
1443 	struct cmdq_query_qp req = {};
1444 	u32 temp32[4];
1445 	int i, rc;
1446 
1447 	sbuf.size = ALIGN(sizeof(*sb), BNXT_QPLIB_CMDQE_UNITS);
1448 	sbuf.sb = dma_alloc_coherent(&rcfw->pdev->dev, sbuf.size,
1449 				     &sbuf.dma_addr, GFP_KERNEL);
1450 	if (!sbuf.sb)
1451 		return -ENOMEM;
1452 	sb = sbuf.sb;
1453 
1454 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
1455 				 CMDQ_BASE_OPCODE_QUERY_QP,
1456 				 sizeof(req));
1457 
1458 	req.qp_cid = cpu_to_le32(qp->id);
1459 	req.resp_size = sbuf.size / BNXT_QPLIB_CMDQE_UNITS;
1460 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, &sbuf, sizeof(req),
1461 				sizeof(resp), 0);
1462 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
1463 	if (rc)
1464 		goto bail;
1465 	/* Extract the context from the side buffer */
1466 	qp->state = sb->en_sqd_async_notify_state &
1467 			CREQ_QUERY_QP_RESP_SB_STATE_MASK;
1468 	qp->en_sqd_async_notify = sb->en_sqd_async_notify_state &
1469 				  CREQ_QUERY_QP_RESP_SB_EN_SQD_ASYNC_NOTIFY;
1470 	qp->access = sb->access;
1471 	qp->pkey_index = le16_to_cpu(sb->pkey);
1472 	qp->qkey = le32_to_cpu(sb->qkey);
1473 	qp->udp_sport = le16_to_cpu(sb->udp_src_port);
1474 
1475 	temp32[0] = le32_to_cpu(sb->dgid[0]);
1476 	temp32[1] = le32_to_cpu(sb->dgid[1]);
1477 	temp32[2] = le32_to_cpu(sb->dgid[2]);
1478 	temp32[3] = le32_to_cpu(sb->dgid[3]);
1479 	memcpy(qp->ah.dgid.data, temp32, sizeof(qp->ah.dgid.data));
1480 
1481 	qp->ah.flow_label = le32_to_cpu(sb->flow_label);
1482 
1483 	qp->ah.sgid_index = 0;
1484 	for (i = 0; i < res->sgid_tbl.max; i++) {
1485 		if (res->sgid_tbl.hw_id[i] == le16_to_cpu(sb->sgid_index)) {
1486 			qp->ah.sgid_index = i;
1487 			break;
1488 		}
1489 	}
1490 	if (i == res->sgid_tbl.max)
1491 		dev_warn(&res->pdev->dev, "SGID not found??\n");
1492 
1493 	qp->ah.hop_limit = sb->hop_limit;
1494 	qp->ah.traffic_class = sb->traffic_class;
1495 	memcpy(qp->ah.dmac, sb->dest_mac, 6);
1496 	qp->ah.vlan_id = (le16_to_cpu(sb->path_mtu_dest_vlan_id) &
1497 				CREQ_QUERY_QP_RESP_SB_VLAN_ID_MASK) >>
1498 				CREQ_QUERY_QP_RESP_SB_VLAN_ID_SFT;
1499 	qp->path_mtu = (le16_to_cpu(sb->path_mtu_dest_vlan_id) &
1500 				    CREQ_QUERY_QP_RESP_SB_PATH_MTU_MASK) >>
1501 				    CREQ_QUERY_QP_RESP_SB_PATH_MTU_SFT;
1502 	qp->timeout = sb->timeout;
1503 	qp->retry_cnt = sb->retry_cnt;
1504 	qp->rnr_retry = sb->rnr_retry;
1505 	qp->min_rnr_timer = sb->min_rnr_timer;
1506 	qp->rq.psn = le32_to_cpu(sb->rq_psn);
1507 	qp->max_rd_atomic = ORRQ_SLOTS_TO_ORD_LIMIT(sb->max_rd_atomic);
1508 	qp->sq.psn = le32_to_cpu(sb->sq_psn);
1509 	qp->max_dest_rd_atomic =
1510 			IRRQ_SLOTS_TO_IRD_LIMIT(sb->max_dest_rd_atomic);
1511 	qp->sq.max_wqe = qp->sq.hwq.max_elements;
1512 	qp->rq.max_wqe = qp->rq.hwq.max_elements;
1513 	qp->sq.max_sge = le16_to_cpu(sb->sq_sge);
1514 	qp->rq.max_sge = le16_to_cpu(sb->rq_sge);
1515 	qp->max_inline_data = le32_to_cpu(sb->max_inline_data);
1516 	qp->dest_qpn = le32_to_cpu(sb->dest_qp_id);
1517 	memcpy(qp->smac, sb->src_mac, 6);
1518 	qp->vlan_id = le16_to_cpu(sb->vlan_pcp_vlan_dei_vlan_id);
1519 	qp->port_id = le16_to_cpu(sb->port_id);
1520 bail:
1521 	dma_free_coherent(&rcfw->pdev->dev, sbuf.size,
1522 			  sbuf.sb, sbuf.dma_addr);
1523 	return rc;
1524 }
1525 
1526 static void __clean_cq(struct bnxt_qplib_cq *cq, u64 qp)
1527 {
1528 	struct bnxt_qplib_hwq *cq_hwq = &cq->hwq;
1529 	u32 peek_flags, peek_cons;
1530 	struct cq_base *hw_cqe;
1531 	int i;
1532 
1533 	peek_flags = cq->dbinfo.flags;
1534 	peek_cons = cq_hwq->cons;
1535 	for (i = 0; i < cq_hwq->max_elements; i++) {
1536 		hw_cqe = bnxt_qplib_get_qe(cq_hwq, peek_cons, NULL);
1537 		if (!CQE_CMP_VALID(hw_cqe, peek_flags))
1538 			continue;
1539 		/*
1540 		 * The valid test of the entry must be done first before
1541 		 * reading any further.
1542 		 */
1543 		dma_rmb();
1544 		switch (hw_cqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK) {
1545 		case CQ_BASE_CQE_TYPE_REQ:
1546 		case CQ_BASE_CQE_TYPE_TERMINAL:
1547 		{
1548 			struct cq_req *cqe = (struct cq_req *)hw_cqe;
1549 
1550 			if (qp == le64_to_cpu(cqe->qp_handle))
1551 				cqe->qp_handle = 0;
1552 			break;
1553 		}
1554 		case CQ_BASE_CQE_TYPE_RES_RC:
1555 		case CQ_BASE_CQE_TYPE_RES_UD:
1556 		case CQ_BASE_CQE_TYPE_RES_RAWETH_QP1:
1557 		{
1558 			struct cq_res_rc *cqe = (struct cq_res_rc *)hw_cqe;
1559 
1560 			if (qp == le64_to_cpu(cqe->qp_handle))
1561 				cqe->qp_handle = 0;
1562 			break;
1563 		}
1564 		default:
1565 			break;
1566 		}
1567 		bnxt_qplib_hwq_incr_cons(cq_hwq->max_elements, &peek_cons,
1568 					 1, &peek_flags);
1569 	}
1570 }
1571 
1572 int bnxt_qplib_destroy_qp(struct bnxt_qplib_res *res,
1573 			  struct bnxt_qplib_qp *qp)
1574 {
1575 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1576 	struct creq_destroy_qp_resp resp = {};
1577 	struct bnxt_qplib_cmdqmsg msg = {};
1578 	struct cmdq_destroy_qp req = {};
1579 	u32 tbl_indx;
1580 	int rc;
1581 
1582 	spin_lock_bh(&rcfw->tbl_lock);
1583 	tbl_indx = map_qp_id_to_tbl_indx(qp->id, rcfw);
1584 	rcfw->qp_tbl[tbl_indx].qp_id = BNXT_QPLIB_QP_ID_INVALID;
1585 	rcfw->qp_tbl[tbl_indx].qp_handle = NULL;
1586 	spin_unlock_bh(&rcfw->tbl_lock);
1587 
1588 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
1589 				 CMDQ_BASE_OPCODE_DESTROY_QP,
1590 				 sizeof(req));
1591 
1592 	req.qp_cid = cpu_to_le32(qp->id);
1593 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
1594 				sizeof(resp), 0);
1595 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
1596 	if (rc) {
1597 		spin_lock_bh(&rcfw->tbl_lock);
1598 		rcfw->qp_tbl[tbl_indx].qp_id = qp->id;
1599 		rcfw->qp_tbl[tbl_indx].qp_handle = qp;
1600 		spin_unlock_bh(&rcfw->tbl_lock);
1601 		return rc;
1602 	}
1603 
1604 	return 0;
1605 }
1606 
1607 void bnxt_qplib_free_qp_res(struct bnxt_qplib_res *res,
1608 			    struct bnxt_qplib_qp *qp)
1609 {
1610 	bnxt_qplib_free_qp_hdr_buf(res, qp);
1611 	bnxt_qplib_free_hwq(res, &qp->sq.hwq);
1612 	kfree(qp->sq.swq);
1613 
1614 	bnxt_qplib_free_hwq(res, &qp->rq.hwq);
1615 	kfree(qp->rq.swq);
1616 
1617 	if (qp->irrq.max_elements)
1618 		bnxt_qplib_free_hwq(res, &qp->irrq);
1619 	if (qp->orrq.max_elements)
1620 		bnxt_qplib_free_hwq(res, &qp->orrq);
1621 
1622 }
1623 
1624 void *bnxt_qplib_get_qp1_sq_buf(struct bnxt_qplib_qp *qp,
1625 				struct bnxt_qplib_sge *sge)
1626 {
1627 	struct bnxt_qplib_q *sq = &qp->sq;
1628 	u32 sw_prod;
1629 
1630 	memset(sge, 0, sizeof(*sge));
1631 
1632 	if (qp->sq_hdr_buf) {
1633 		sw_prod = sq->swq_start;
1634 		sge->addr = (dma_addr_t)(qp->sq_hdr_buf_map +
1635 					 sw_prod * qp->sq_hdr_buf_size);
1636 		sge->lkey = 0xFFFFFFFF;
1637 		sge->size = qp->sq_hdr_buf_size;
1638 		return qp->sq_hdr_buf + sw_prod * sge->size;
1639 	}
1640 	return NULL;
1641 }
1642 
1643 u32 bnxt_qplib_get_rq_prod_index(struct bnxt_qplib_qp *qp)
1644 {
1645 	struct bnxt_qplib_q *rq = &qp->rq;
1646 
1647 	return rq->swq_start;
1648 }
1649 
1650 dma_addr_t bnxt_qplib_get_qp_buf_from_index(struct bnxt_qplib_qp *qp, u32 index)
1651 {
1652 	return (qp->rq_hdr_buf_map + index * qp->rq_hdr_buf_size);
1653 }
1654 
1655 void *bnxt_qplib_get_qp1_rq_buf(struct bnxt_qplib_qp *qp,
1656 				struct bnxt_qplib_sge *sge)
1657 {
1658 	struct bnxt_qplib_q *rq = &qp->rq;
1659 	u32 sw_prod;
1660 
1661 	memset(sge, 0, sizeof(*sge));
1662 
1663 	if (qp->rq_hdr_buf) {
1664 		sw_prod = rq->swq_start;
1665 		sge->addr = (dma_addr_t)(qp->rq_hdr_buf_map +
1666 					 sw_prod * qp->rq_hdr_buf_size);
1667 		sge->lkey = 0xFFFFFFFF;
1668 		sge->size = qp->rq_hdr_buf_size;
1669 		return qp->rq_hdr_buf + sw_prod * sge->size;
1670 	}
1671 	return NULL;
1672 }
1673 
1674 /* Fil the MSN table into the next psn row */
1675 static void bnxt_qplib_fill_msn_search(struct bnxt_qplib_qp *qp,
1676 				       struct bnxt_qplib_swqe *wqe,
1677 				       struct bnxt_qplib_swq *swq)
1678 {
1679 	struct sq_msn_search *msns;
1680 	u32 start_psn, next_psn;
1681 	u16 start_idx;
1682 
1683 	msns = (struct sq_msn_search *)swq->psn_search;
1684 	msns->start_idx_next_psn_start_psn = 0;
1685 
1686 	start_psn = swq->start_psn;
1687 	next_psn = swq->next_psn;
1688 	start_idx = swq->slot_idx;
1689 	msns->start_idx_next_psn_start_psn |=
1690 		bnxt_re_update_msn_tbl(start_idx, next_psn, start_psn);
1691 	qp->msn++;
1692 	qp->msn %= qp->msn_tbl_sz;
1693 }
1694 
1695 static void bnxt_qplib_fill_psn_search(struct bnxt_qplib_qp *qp,
1696 				       struct bnxt_qplib_swqe *wqe,
1697 				       struct bnxt_qplib_swq *swq)
1698 {
1699 	struct sq_psn_search_ext *psns_ext;
1700 	struct sq_psn_search *psns;
1701 	u32 flg_npsn;
1702 	u32 op_spsn;
1703 
1704 	if (!swq->psn_search)
1705 		return;
1706 	/* Handle MSN differently on cap flags  */
1707 	if (qp->is_host_msn_tbl) {
1708 		bnxt_qplib_fill_msn_search(qp, wqe, swq);
1709 		return;
1710 	}
1711 	psns = (struct sq_psn_search *)swq->psn_search;
1712 	psns = swq->psn_search;
1713 	psns_ext = swq->psn_ext;
1714 
1715 	op_spsn = ((swq->start_psn << SQ_PSN_SEARCH_START_PSN_SFT) &
1716 		    SQ_PSN_SEARCH_START_PSN_MASK);
1717 	op_spsn |= ((wqe->type << SQ_PSN_SEARCH_OPCODE_SFT) &
1718 		     SQ_PSN_SEARCH_OPCODE_MASK);
1719 	flg_npsn = ((swq->next_psn << SQ_PSN_SEARCH_NEXT_PSN_SFT) &
1720 		     SQ_PSN_SEARCH_NEXT_PSN_MASK);
1721 
1722 	if (bnxt_qplib_is_chip_gen_p5_p7(qp->cctx)) {
1723 		psns_ext->opcode_start_psn = cpu_to_le32(op_spsn);
1724 		psns_ext->flags_next_psn = cpu_to_le32(flg_npsn);
1725 		psns_ext->start_slot_idx = cpu_to_le16(swq->slot_idx);
1726 	} else {
1727 		psns->opcode_start_psn = cpu_to_le32(op_spsn);
1728 		psns->flags_next_psn = cpu_to_le32(flg_npsn);
1729 	}
1730 }
1731 
1732 static unsigned int bnxt_qplib_put_inline(struct bnxt_qplib_qp *qp,
1733 					  struct bnxt_qplib_swqe *wqe,
1734 					  u32 *idx)
1735 {
1736 	struct bnxt_qplib_hwq *hwq;
1737 	int len, t_len, offt;
1738 	bool pull_dst = true;
1739 	void *il_dst = NULL;
1740 	void *il_src = NULL;
1741 	int t_cplen, cplen;
1742 	int indx;
1743 
1744 	hwq = &qp->sq.hwq;
1745 	t_len = 0;
1746 	for (indx = 0; indx < wqe->num_sge; indx++) {
1747 		len = wqe->sg_list[indx].size;
1748 		il_src = (void *)wqe->sg_list[indx].addr;
1749 		t_len += len;
1750 		if (t_len > qp->max_inline_data)
1751 			return BNXT_RE_INVAL_MSG_SIZE;
1752 		while (len) {
1753 			if (pull_dst) {
1754 				pull_dst = false;
1755 				il_dst = bnxt_qplib_get_prod_qe(hwq, *idx);
1756 				(*idx)++;
1757 				t_cplen = 0;
1758 				offt = 0;
1759 			}
1760 			cplen = min_t(int, len, sizeof(struct sq_sge));
1761 			cplen = min_t(int, cplen,
1762 					(sizeof(struct sq_sge) - offt));
1763 			memcpy(il_dst, il_src, cplen);
1764 			t_cplen += cplen;
1765 			il_src += cplen;
1766 			il_dst += cplen;
1767 			offt += cplen;
1768 			len -= cplen;
1769 			if (t_cplen == sizeof(struct sq_sge))
1770 				pull_dst = true;
1771 		}
1772 	}
1773 
1774 	return t_len;
1775 }
1776 
1777 static unsigned int bnxt_qplib_put_sges(struct bnxt_qplib_hwq *hwq,
1778 					struct bnxt_qplib_sge *ssge,
1779 					u32 nsge, u32 *idx)
1780 {
1781 	struct sq_sge *dsge;
1782 	int indx, len = 0;
1783 
1784 	for (indx = 0; indx < nsge; indx++, (*idx)++) {
1785 		dsge = bnxt_qplib_get_prod_qe(hwq, *idx);
1786 		dsge->va_or_pa = cpu_to_le64(ssge[indx].addr);
1787 		dsge->l_key = cpu_to_le32(ssge[indx].lkey);
1788 		dsge->size = cpu_to_le32(ssge[indx].size);
1789 		len += ssge[indx].size;
1790 	}
1791 
1792 	return len;
1793 }
1794 
1795 static u16 bnxt_qplib_required_slots(struct bnxt_qplib_qp *qp,
1796 				     struct bnxt_qplib_swqe *wqe,
1797 				     u16 *wqe_sz, u16 *qdf, u8 mode)
1798 {
1799 	u32 ilsize, bytes;
1800 	u16 nsge;
1801 	u16 slot;
1802 
1803 	nsge = wqe->num_sge;
1804 	/* Adding sq_send_hdr is a misnomer, for rq also hdr size is same. */
1805 	bytes = sizeof(struct sq_send_hdr) + nsge * sizeof(struct sq_sge);
1806 	if (wqe->flags & BNXT_QPLIB_SWQE_FLAGS_INLINE) {
1807 		ilsize = bnxt_qplib_calc_ilsize(wqe, qp->max_inline_data);
1808 		bytes = ALIGN(ilsize, sizeof(struct sq_sge));
1809 		bytes += sizeof(struct sq_send_hdr);
1810 	}
1811 
1812 	*qdf =  __xlate_qfd(qp->sq.q_full_delta, bytes);
1813 	slot = bytes >> 4;
1814 	*wqe_sz = slot;
1815 	if (mode == BNXT_QPLIB_WQE_MODE_STATIC)
1816 		slot = 8;
1817 	return slot;
1818 }
1819 
1820 static void bnxt_qplib_pull_psn_buff(struct bnxt_qplib_qp *qp, struct bnxt_qplib_q *sq,
1821 				     struct bnxt_qplib_swq *swq, bool hw_retx)
1822 {
1823 	struct bnxt_qplib_hwq *hwq;
1824 	u32 pg_num, pg_indx;
1825 	void *buff;
1826 	u32 tail;
1827 
1828 	hwq = &sq->hwq;
1829 	if (!hwq->pad_pg)
1830 		return;
1831 	tail = swq->slot_idx / sq->dbinfo.max_slot;
1832 	if (hw_retx) {
1833 		/* For HW retx use qp msn index */
1834 		tail = qp->msn;
1835 		tail %= qp->msn_tbl_sz;
1836 	}
1837 	pg_num = (tail + hwq->pad_pgofft) / (PAGE_SIZE / hwq->pad_stride);
1838 	pg_indx = (tail + hwq->pad_pgofft) % (PAGE_SIZE / hwq->pad_stride);
1839 	buff = (void *)(hwq->pad_pg[pg_num] + pg_indx * hwq->pad_stride);
1840 	swq->psn_ext = buff;
1841 	swq->psn_search = buff;
1842 }
1843 
1844 void bnxt_qplib_post_send_db(struct bnxt_qplib_qp *qp)
1845 {
1846 	struct bnxt_qplib_q *sq = &qp->sq;
1847 
1848 	bnxt_qplib_ring_prod_db(&sq->dbinfo, DBC_DBC_TYPE_SQ);
1849 }
1850 
1851 int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp,
1852 			 struct bnxt_qplib_swqe *wqe)
1853 {
1854 	struct bnxt_qplib_nq_work *nq_work = NULL;
1855 	int i, rc = 0, data_len = 0, pkt_num = 0;
1856 	struct bnxt_qplib_q *sq = &qp->sq;
1857 	struct bnxt_qplib_hwq *hwq;
1858 	struct bnxt_qplib_swq *swq;
1859 	bool sch_handler = false;
1860 	u32 wqe_idx, slots, idx;
1861 	u16 wqe_sz, qdf = 0;
1862 	bool msn_update;
1863 	void *base_hdr;
1864 	void *ext_hdr;
1865 	__le32 temp32;
1866 
1867 	hwq = &sq->hwq;
1868 	if (qp->state != CMDQ_MODIFY_QP_NEW_STATE_RTS &&
1869 	    qp->state != CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1870 		dev_err(&hwq->pdev->dev,
1871 			"QPLIB: FP: QP (0x%x) is in the 0x%x state",
1872 			qp->id, qp->state);
1873 		rc = -EINVAL;
1874 		goto done;
1875 	}
1876 
1877 	slots = bnxt_qplib_required_slots(qp, wqe, &wqe_sz, &qdf, qp->wqe_mode);
1878 	if (bnxt_qplib_queue_full(sq, slots + qdf)) {
1879 		dev_err(&hwq->pdev->dev,
1880 			"prod = %#x cons = %#x qdepth = %#x delta = %#x\n",
1881 			hwq->prod, hwq->cons, hwq->depth, sq->q_full_delta);
1882 		rc = -ENOMEM;
1883 		goto done;
1884 	}
1885 
1886 	swq = bnxt_qplib_get_swqe(sq, &wqe_idx);
1887 	bnxt_qplib_pull_psn_buff(qp, sq, swq, qp->is_host_msn_tbl);
1888 
1889 	idx = 0;
1890 	swq->slot_idx = hwq->prod;
1891 	swq->slots = slots;
1892 	swq->wr_id = wqe->wr_id;
1893 	swq->type = wqe->type;
1894 	swq->flags = wqe->flags;
1895 	swq->start_psn = sq->psn & BTH_PSN_MASK;
1896 	if (qp->sig_type)
1897 		swq->flags |= SQ_SEND_FLAGS_SIGNAL_COMP;
1898 
1899 	if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1900 		sch_handler = true;
1901 		dev_dbg(&hwq->pdev->dev,
1902 			"%s Error QP. Scheduling for poll_cq\n", __func__);
1903 		goto queue_err;
1904 	}
1905 
1906 	base_hdr = bnxt_qplib_get_prod_qe(hwq, idx++);
1907 	ext_hdr = bnxt_qplib_get_prod_qe(hwq, idx++);
1908 	memset(base_hdr, 0, sizeof(struct sq_sge));
1909 	memset(ext_hdr, 0, sizeof(struct sq_sge));
1910 
1911 	if (wqe->flags & BNXT_QPLIB_SWQE_FLAGS_INLINE)
1912 		/* Copy the inline data */
1913 		data_len = bnxt_qplib_put_inline(qp, wqe, &idx);
1914 	else
1915 		data_len = bnxt_qplib_put_sges(hwq, wqe->sg_list, wqe->num_sge,
1916 					       &idx);
1917 	if (data_len > BNXT_RE_MAX_MSG_SIZE) {
1918 		rc = -EINVAL;
1919 		goto done;
1920 	}
1921 	/* Make sure we update MSN table only for wired wqes */
1922 	msn_update = true;
1923 	/* Specifics */
1924 	switch (wqe->type) {
1925 	case BNXT_QPLIB_SWQE_TYPE_SEND:
1926 		if (qp->type == CMDQ_CREATE_QP1_TYPE_GSI) {
1927 			struct sq_send_raweth_qp1_hdr *sqe = base_hdr;
1928 			struct sq_raw_ext_hdr *ext_sqe = ext_hdr;
1929 			/* Assemble info for Raw Ethertype QPs */
1930 
1931 			sqe->wqe_type = wqe->type;
1932 			sqe->flags = wqe->flags;
1933 			sqe->wqe_size = wqe_sz;
1934 			sqe->cfa_action = cpu_to_le16(wqe->rawqp1.cfa_action);
1935 			sqe->lflags = cpu_to_le16(wqe->rawqp1.lflags);
1936 			sqe->length = cpu_to_le32(data_len);
1937 			ext_sqe->cfa_meta = cpu_to_le32((wqe->rawqp1.cfa_meta &
1938 				SQ_SEND_RAWETH_QP1_CFA_META_VLAN_VID_MASK) <<
1939 				SQ_SEND_RAWETH_QP1_CFA_META_VLAN_VID_SFT);
1940 
1941 			break;
1942 		}
1943 		fallthrough;
1944 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_IMM:
1945 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_INV:
1946 	{
1947 		struct sq_ud_ext_hdr *ext_sqe = ext_hdr;
1948 		struct sq_send_hdr *sqe = base_hdr;
1949 
1950 		sqe->wqe_type = wqe->type;
1951 		sqe->flags = wqe->flags;
1952 		sqe->wqe_size = wqe_sz;
1953 		sqe->inv_key_or_imm_data = cpu_to_le32(wqe->send.inv_key);
1954 		if (qp->type == CMDQ_CREATE_QP_TYPE_UD ||
1955 		    qp->type == CMDQ_CREATE_QP_TYPE_GSI) {
1956 			sqe->q_key = cpu_to_le32(wqe->send.q_key);
1957 			sqe->length = cpu_to_le32(data_len);
1958 			sq->psn = (sq->psn + 1) & BTH_PSN_MASK;
1959 			ext_sqe->dst_qp = cpu_to_le32(wqe->send.dst_qp &
1960 						      SQ_SEND_DST_QP_MASK);
1961 			ext_sqe->avid = cpu_to_le32(wqe->send.avid &
1962 						    SQ_SEND_AVID_MASK);
1963 			msn_update = false;
1964 		} else {
1965 			sqe->length = cpu_to_le32(data_len);
1966 			if (qp->mtu)
1967 				pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
1968 			if (!pkt_num)
1969 				pkt_num = 1;
1970 			sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
1971 		}
1972 		break;
1973 	}
1974 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE:
1975 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE_WITH_IMM:
1976 	case BNXT_QPLIB_SWQE_TYPE_RDMA_READ:
1977 	{
1978 		struct sq_rdma_ext_hdr *ext_sqe = ext_hdr;
1979 		struct sq_rdma_hdr *sqe = base_hdr;
1980 
1981 		sqe->wqe_type = wqe->type;
1982 		sqe->flags = wqe->flags;
1983 		sqe->wqe_size = wqe_sz;
1984 		sqe->imm_data = cpu_to_le32(wqe->rdma.inv_key);
1985 		sqe->length = cpu_to_le32((u32)data_len);
1986 		ext_sqe->remote_va = cpu_to_le64(wqe->rdma.remote_va);
1987 		ext_sqe->remote_key = cpu_to_le32(wqe->rdma.r_key);
1988 		if (qp->mtu)
1989 			pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
1990 		if (!pkt_num)
1991 			pkt_num = 1;
1992 		sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
1993 		break;
1994 	}
1995 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_CMP_AND_SWP:
1996 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_FETCH_AND_ADD:
1997 	{
1998 		struct sq_atomic_ext_hdr *ext_sqe = ext_hdr;
1999 		struct sq_atomic_hdr *sqe = base_hdr;
2000 
2001 		sqe->wqe_type = wqe->type;
2002 		sqe->flags = wqe->flags;
2003 		sqe->remote_key = cpu_to_le32(wqe->atomic.r_key);
2004 		sqe->remote_va = cpu_to_le64(wqe->atomic.remote_va);
2005 		ext_sqe->swap_data = cpu_to_le64(wqe->atomic.swap_data);
2006 		ext_sqe->cmp_data = cpu_to_le64(wqe->atomic.cmp_data);
2007 		if (qp->mtu)
2008 			pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
2009 		if (!pkt_num)
2010 			pkt_num = 1;
2011 		sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
2012 		break;
2013 	}
2014 	case BNXT_QPLIB_SWQE_TYPE_LOCAL_INV:
2015 	{
2016 		struct sq_localinvalidate *sqe = base_hdr;
2017 
2018 		sqe->wqe_type = wqe->type;
2019 		sqe->flags = wqe->flags;
2020 		sqe->inv_l_key = cpu_to_le32(wqe->local_inv.inv_l_key);
2021 		msn_update = false;
2022 		break;
2023 	}
2024 	case BNXT_QPLIB_SWQE_TYPE_FAST_REG_MR:
2025 	{
2026 		struct sq_fr_pmr_ext_hdr *ext_sqe = ext_hdr;
2027 		struct sq_fr_pmr_hdr *sqe = base_hdr;
2028 
2029 		sqe->wqe_type = wqe->type;
2030 		sqe->flags = wqe->flags;
2031 		sqe->access_cntl = wqe->frmr.access_cntl |
2032 				   SQ_FR_PMR_ACCESS_CNTL_LOCAL_WRITE;
2033 		sqe->zero_based_page_size_log =
2034 			(wqe->frmr.pg_sz_log & SQ_FR_PMR_PAGE_SIZE_LOG_MASK) <<
2035 			SQ_FR_PMR_PAGE_SIZE_LOG_SFT |
2036 			(wqe->frmr.zero_based ? SQ_FR_PMR_ZERO_BASED : 0);
2037 		sqe->l_key = cpu_to_le32(wqe->frmr.l_key);
2038 		temp32 = cpu_to_le32(wqe->frmr.length);
2039 		memcpy(sqe->length, &temp32, sizeof(wqe->frmr.length));
2040 		sqe->numlevels_pbl_page_size_log =
2041 			((wqe->frmr.pbl_pg_sz_log <<
2042 					SQ_FR_PMR_PBL_PAGE_SIZE_LOG_SFT) &
2043 					SQ_FR_PMR_PBL_PAGE_SIZE_LOG_MASK) |
2044 			((wqe->frmr.levels << SQ_FR_PMR_NUMLEVELS_SFT) &
2045 					SQ_FR_PMR_NUMLEVELS_MASK);
2046 
2047 		for (i = 0; i < wqe->frmr.page_list_len; i++)
2048 			wqe->frmr.pbl_ptr[i] = cpu_to_le64(
2049 						wqe->frmr.page_list[i] |
2050 						PTU_PTE_VALID);
2051 		ext_sqe->pblptr = cpu_to_le64(wqe->frmr.pbl_dma_ptr);
2052 		ext_sqe->va = cpu_to_le64(wqe->frmr.va);
2053 		msn_update = false;
2054 
2055 		break;
2056 	}
2057 	case BNXT_QPLIB_SWQE_TYPE_BIND_MW:
2058 	{
2059 		struct sq_bind_ext_hdr *ext_sqe = ext_hdr;
2060 		struct sq_bind_hdr *sqe = base_hdr;
2061 
2062 		sqe->wqe_type = wqe->type;
2063 		sqe->flags = wqe->flags;
2064 		sqe->access_cntl = wqe->bind.access_cntl;
2065 		sqe->mw_type_zero_based = wqe->bind.mw_type |
2066 			(wqe->bind.zero_based ? SQ_BIND_ZERO_BASED : 0);
2067 		sqe->parent_l_key = cpu_to_le32(wqe->bind.parent_l_key);
2068 		sqe->l_key = cpu_to_le32(wqe->bind.r_key);
2069 		ext_sqe->va = cpu_to_le64(wqe->bind.va);
2070 		ext_sqe->length_lo = cpu_to_le32(wqe->bind.length);
2071 		msn_update = false;
2072 		break;
2073 	}
2074 	default:
2075 		/* Bad wqe, return error */
2076 		rc = -EINVAL;
2077 		goto done;
2078 	}
2079 	if (!qp->is_host_msn_tbl || msn_update) {
2080 		swq->next_psn = sq->psn & BTH_PSN_MASK;
2081 		bnxt_qplib_fill_psn_search(qp, wqe, swq);
2082 	}
2083 queue_err:
2084 	bnxt_qplib_swq_mod_start(sq, wqe_idx);
2085 	bnxt_qplib_hwq_incr_prod(&sq->dbinfo, hwq, swq->slots);
2086 	qp->wqe_cnt++;
2087 done:
2088 	if (sch_handler) {
2089 		nq_work = kzalloc(sizeof(*nq_work), GFP_ATOMIC);
2090 		if (nq_work) {
2091 			nq_work->cq = qp->scq;
2092 			nq_work->nq = qp->scq->nq;
2093 			INIT_WORK(&nq_work->work, bnxt_qpn_cqn_sched_task);
2094 			queue_work(qp->scq->nq->cqn_wq, &nq_work->work);
2095 		} else {
2096 			dev_err(&hwq->pdev->dev,
2097 				"FP: Failed to allocate SQ nq_work!\n");
2098 			rc = -ENOMEM;
2099 		}
2100 	}
2101 	return rc;
2102 }
2103 
2104 void bnxt_qplib_post_recv_db(struct bnxt_qplib_qp *qp)
2105 {
2106 	struct bnxt_qplib_q *rq = &qp->rq;
2107 
2108 	bnxt_qplib_ring_prod_db(&rq->dbinfo, DBC_DBC_TYPE_RQ);
2109 }
2110 
2111 int bnxt_qplib_post_recv(struct bnxt_qplib_qp *qp,
2112 			 struct bnxt_qplib_swqe *wqe)
2113 {
2114 	struct bnxt_qplib_nq_work *nq_work = NULL;
2115 	struct bnxt_qplib_q *rq = &qp->rq;
2116 	struct rq_wqe_hdr *base_hdr;
2117 	struct rq_ext_hdr *ext_hdr;
2118 	struct bnxt_qplib_hwq *hwq;
2119 	struct bnxt_qplib_swq *swq;
2120 	bool sch_handler = false;
2121 	u32 wqe_idx, idx;
2122 	u16 wqe_sz;
2123 	int rc = 0;
2124 
2125 	hwq = &rq->hwq;
2126 	if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_RESET) {
2127 		dev_err(&hwq->pdev->dev,
2128 			"QPLIB: FP: QP (0x%x) is in the 0x%x state",
2129 			qp->id, qp->state);
2130 		rc = -EINVAL;
2131 		goto done;
2132 	}
2133 
2134 	if (bnxt_qplib_queue_full(rq, rq->dbinfo.max_slot)) {
2135 		dev_err(&hwq->pdev->dev,
2136 			"FP: QP (0x%x) RQ is full!\n", qp->id);
2137 		rc = -EINVAL;
2138 		goto done;
2139 	}
2140 
2141 	swq = bnxt_qplib_get_swqe(rq, &wqe_idx);
2142 	swq->wr_id = wqe->wr_id;
2143 	swq->slots = rq->dbinfo.max_slot;
2144 
2145 	if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
2146 		sch_handler = true;
2147 		dev_dbg(&hwq->pdev->dev,
2148 			"%s: Error QP. Scheduling for poll_cq\n", __func__);
2149 		goto queue_err;
2150 	}
2151 
2152 	idx = 0;
2153 	base_hdr = bnxt_qplib_get_prod_qe(hwq, idx++);
2154 	ext_hdr = bnxt_qplib_get_prod_qe(hwq, idx++);
2155 	memset(base_hdr, 0, sizeof(struct sq_sge));
2156 	memset(ext_hdr, 0, sizeof(struct sq_sge));
2157 	wqe_sz = (sizeof(struct rq_wqe_hdr) +
2158 	wqe->num_sge * sizeof(struct sq_sge)) >> 4;
2159 	bnxt_qplib_put_sges(hwq, wqe->sg_list, wqe->num_sge, &idx);
2160 	if (!wqe->num_sge) {
2161 		struct sq_sge *sge;
2162 
2163 		sge = bnxt_qplib_get_prod_qe(hwq, idx++);
2164 		sge->size = 0;
2165 		wqe_sz++;
2166 	}
2167 	base_hdr->wqe_type = wqe->type;
2168 	base_hdr->flags = wqe->flags;
2169 	base_hdr->wqe_size = wqe_sz;
2170 	base_hdr->wr_id[0] = cpu_to_le32(wqe_idx);
2171 queue_err:
2172 	bnxt_qplib_swq_mod_start(rq, wqe_idx);
2173 	bnxt_qplib_hwq_incr_prod(&rq->dbinfo, hwq, swq->slots);
2174 done:
2175 	if (sch_handler) {
2176 		nq_work = kzalloc(sizeof(*nq_work), GFP_ATOMIC);
2177 		if (nq_work) {
2178 			nq_work->cq = qp->rcq;
2179 			nq_work->nq = qp->rcq->nq;
2180 			INIT_WORK(&nq_work->work, bnxt_qpn_cqn_sched_task);
2181 			queue_work(qp->rcq->nq->cqn_wq, &nq_work->work);
2182 		} else {
2183 			dev_err(&hwq->pdev->dev,
2184 				"FP: Failed to allocate RQ nq_work!\n");
2185 			rc = -ENOMEM;
2186 		}
2187 	}
2188 
2189 	return rc;
2190 }
2191 
2192 /* CQ */
2193 int bnxt_qplib_create_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq)
2194 {
2195 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
2196 	struct bnxt_qplib_hwq_attr hwq_attr = {};
2197 	struct creq_create_cq_resp resp = {};
2198 	struct bnxt_qplib_cmdqmsg msg = {};
2199 	struct cmdq_create_cq req = {};
2200 	struct bnxt_qplib_pbl *pbl;
2201 	u32 coalescing = 0;
2202 	u32 pg_sz_lvl;
2203 	int rc;
2204 
2205 	if (!cq->dpi) {
2206 		dev_err(&rcfw->pdev->dev,
2207 			"FP: CREATE_CQ failed due to NULL DPI\n");
2208 		return -EINVAL;
2209 	}
2210 
2211 	cq->dbinfo.flags = 0;
2212 	hwq_attr.res = res;
2213 	hwq_attr.depth = cq->max_wqe;
2214 	hwq_attr.stride = sizeof(struct cq_base);
2215 	hwq_attr.type = HWQ_TYPE_QUEUE;
2216 	hwq_attr.sginfo = &cq->sg_info;
2217 	rc = bnxt_qplib_alloc_init_hwq(&cq->hwq, &hwq_attr);
2218 	if (rc)
2219 		return rc;
2220 
2221 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
2222 				 CMDQ_BASE_OPCODE_CREATE_CQ,
2223 				 sizeof(req));
2224 
2225 	req.dpi = cpu_to_le32(cq->dpi->dpi);
2226 	req.cq_handle = cpu_to_le64(cq->cq_handle);
2227 	req.cq_size = cpu_to_le32(cq->max_wqe);
2228 
2229 	if (_is_cq_coalescing_supported(res->dattr->dev_cap_flags2)) {
2230 		req.flags |= cpu_to_le16(CMDQ_CREATE_CQ_FLAGS_COALESCING_VALID);
2231 		coalescing |= ((cq->coalescing->buf_maxtime <<
2232 				CMDQ_CREATE_CQ_BUF_MAXTIME_SFT) &
2233 			       CMDQ_CREATE_CQ_BUF_MAXTIME_MASK);
2234 		coalescing |= ((cq->coalescing->normal_maxbuf <<
2235 				CMDQ_CREATE_CQ_NORMAL_MAXBUF_SFT) &
2236 			       CMDQ_CREATE_CQ_NORMAL_MAXBUF_MASK);
2237 		coalescing |= ((cq->coalescing->during_maxbuf <<
2238 				CMDQ_CREATE_CQ_DURING_MAXBUF_SFT) &
2239 			       CMDQ_CREATE_CQ_DURING_MAXBUF_MASK);
2240 		if (cq->coalescing->en_ring_idle_mode)
2241 			coalescing |= CMDQ_CREATE_CQ_ENABLE_RING_IDLE_MODE;
2242 		else
2243 			coalescing &= ~CMDQ_CREATE_CQ_ENABLE_RING_IDLE_MODE;
2244 		req.coalescing = cpu_to_le32(coalescing);
2245 	}
2246 
2247 	pbl = &cq->hwq.pbl[PBL_LVL_0];
2248 	pg_sz_lvl = (bnxt_qplib_base_pg_size(&cq->hwq) <<
2249 		     CMDQ_CREATE_CQ_PG_SIZE_SFT);
2250 	pg_sz_lvl |= (cq->hwq.level & CMDQ_CREATE_CQ_LVL_MASK);
2251 	req.pg_size_lvl = cpu_to_le32(pg_sz_lvl);
2252 	req.pbl = cpu_to_le64(pbl->pg_map_arr[0]);
2253 	req.cq_fco_cnq_id = cpu_to_le32(
2254 			(cq->cnq_hw_ring_id & CMDQ_CREATE_CQ_CNQ_ID_MASK) <<
2255 			 CMDQ_CREATE_CQ_CNQ_ID_SFT);
2256 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
2257 				sizeof(resp), 0);
2258 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
2259 	if (rc)
2260 		goto fail;
2261 
2262 	cq->id = le32_to_cpu(resp.xid);
2263 	cq->period = BNXT_QPLIB_QUEUE_START_PERIOD;
2264 	init_waitqueue_head(&cq->waitq);
2265 	INIT_LIST_HEAD(&cq->sqf_head);
2266 	INIT_LIST_HEAD(&cq->rqf_head);
2267 	spin_lock_init(&cq->compl_lock);
2268 	spin_lock_init(&cq->flush_lock);
2269 
2270 	cq->dbinfo.hwq = &cq->hwq;
2271 	cq->dbinfo.xid = cq->id;
2272 	cq->dbinfo.db = cq->dpi->dbr;
2273 	cq->dbinfo.priv_db = res->dpi_tbl.priv_db;
2274 	cq->dbinfo.flags = 0;
2275 	cq->dbinfo.toggle = 0;
2276 
2277 	bnxt_qplib_armen_db(&cq->dbinfo, DBC_DBC_TYPE_CQ_ARMENA);
2278 
2279 	return 0;
2280 
2281 fail:
2282 	bnxt_qplib_free_hwq(res, &cq->hwq);
2283 	return rc;
2284 }
2285 
2286 void bnxt_qplib_resize_cq_complete(struct bnxt_qplib_res *res,
2287 				   struct bnxt_qplib_cq *cq)
2288 {
2289 	bnxt_qplib_free_hwq(res, &cq->hwq);
2290 	memcpy(&cq->hwq, &cq->resize_hwq, sizeof(cq->hwq));
2291        /* Reset only the cons bit in the flags */
2292 	cq->dbinfo.flags &= ~(1UL << BNXT_QPLIB_FLAG_EPOCH_CONS_SHIFT);
2293 }
2294 
2295 int bnxt_qplib_resize_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq,
2296 			 int new_cqes)
2297 {
2298 	struct bnxt_qplib_hwq_attr hwq_attr = {};
2299 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
2300 	struct creq_resize_cq_resp resp = {};
2301 	struct bnxt_qplib_cmdqmsg msg = {};
2302 	struct cmdq_resize_cq req = {};
2303 	struct bnxt_qplib_pbl *pbl;
2304 	u32 pg_sz, lvl, new_sz;
2305 	int rc;
2306 
2307 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
2308 				 CMDQ_BASE_OPCODE_RESIZE_CQ,
2309 				 sizeof(req));
2310 	hwq_attr.sginfo = &cq->sg_info;
2311 	hwq_attr.res = res;
2312 	hwq_attr.depth = new_cqes;
2313 	hwq_attr.stride = sizeof(struct cq_base);
2314 	hwq_attr.type = HWQ_TYPE_QUEUE;
2315 	rc = bnxt_qplib_alloc_init_hwq(&cq->resize_hwq, &hwq_attr);
2316 	if (rc)
2317 		return rc;
2318 
2319 	req.cq_cid = cpu_to_le32(cq->id);
2320 	pbl = &cq->resize_hwq.pbl[PBL_LVL_0];
2321 	pg_sz = bnxt_qplib_base_pg_size(&cq->resize_hwq);
2322 	lvl = (cq->resize_hwq.level << CMDQ_RESIZE_CQ_LVL_SFT) &
2323 				       CMDQ_RESIZE_CQ_LVL_MASK;
2324 	new_sz = (new_cqes << CMDQ_RESIZE_CQ_NEW_CQ_SIZE_SFT) &
2325 		  CMDQ_RESIZE_CQ_NEW_CQ_SIZE_MASK;
2326 	req.new_cq_size_pg_size_lvl = cpu_to_le32(new_sz | pg_sz | lvl);
2327 	req.new_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
2328 
2329 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
2330 				sizeof(resp), 0);
2331 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
2332 	return rc;
2333 }
2334 
2335 int bnxt_qplib_destroy_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq)
2336 {
2337 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
2338 	struct creq_destroy_cq_resp resp = {};
2339 	struct bnxt_qplib_cmdqmsg msg = {};
2340 	struct cmdq_destroy_cq req = {};
2341 	u16 total_cnq_events;
2342 	int rc;
2343 
2344 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
2345 				 CMDQ_BASE_OPCODE_DESTROY_CQ,
2346 				 sizeof(req));
2347 
2348 	req.cq_cid = cpu_to_le32(cq->id);
2349 	bnxt_qplib_fill_cmdqmsg(&msg, &req, &resp, NULL, sizeof(req),
2350 				sizeof(resp), 0);
2351 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
2352 	if (rc)
2353 		return rc;
2354 	total_cnq_events = le16_to_cpu(resp.total_cnq_events);
2355 	__wait_for_all_nqes(cq, total_cnq_events);
2356 	bnxt_qplib_free_hwq(res, &cq->hwq);
2357 	return 0;
2358 }
2359 
2360 static int __flush_sq(struct bnxt_qplib_q *sq, struct bnxt_qplib_qp *qp,
2361 		      struct bnxt_qplib_cqe **pcqe, int *budget)
2362 {
2363 	struct bnxt_qplib_cqe *cqe;
2364 	u32 start, last;
2365 	int rc = 0;
2366 
2367 	/* Now complete all outstanding SQEs with FLUSHED_ERR */
2368 	start = sq->swq_start;
2369 	cqe = *pcqe;
2370 	while (*budget) {
2371 		last = sq->swq_last;
2372 		if (start == last)
2373 			break;
2374 		/* Skip the FENCE WQE completions */
2375 		if (sq->swq[last].wr_id == BNXT_QPLIB_FENCE_WRID) {
2376 			bnxt_qplib_cancel_phantom_processing(qp);
2377 			goto skip_compl;
2378 		}
2379 		memset(cqe, 0, sizeof(*cqe));
2380 		cqe->status = CQ_REQ_STATUS_WORK_REQUEST_FLUSHED_ERR;
2381 		cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
2382 		cqe->qp_handle = (u64)(unsigned long)qp;
2383 		cqe->wr_id = sq->swq[last].wr_id;
2384 		cqe->src_qp = qp->id;
2385 		cqe->type = sq->swq[last].type;
2386 		cqe++;
2387 		(*budget)--;
2388 skip_compl:
2389 		bnxt_qplib_hwq_incr_cons(sq->hwq.max_elements, &sq->hwq.cons,
2390 					 sq->swq[last].slots, &sq->dbinfo.flags);
2391 		sq->swq_last = sq->swq[last].next_idx;
2392 	}
2393 	*pcqe = cqe;
2394 	if (!(*budget) && sq->swq_last != start)
2395 		/* Out of budget */
2396 		rc = -EAGAIN;
2397 
2398 	return rc;
2399 }
2400 
2401 static int __flush_rq(struct bnxt_qplib_q *rq, struct bnxt_qplib_qp *qp,
2402 		      struct bnxt_qplib_cqe **pcqe, int *budget)
2403 {
2404 	struct bnxt_qplib_cqe *cqe;
2405 	u32 start, last;
2406 	int opcode = 0;
2407 	int rc = 0;
2408 
2409 	switch (qp->type) {
2410 	case CMDQ_CREATE_QP1_TYPE_GSI:
2411 		opcode = CQ_BASE_CQE_TYPE_RES_RAWETH_QP1;
2412 		break;
2413 	case CMDQ_CREATE_QP_TYPE_RC:
2414 		opcode = CQ_BASE_CQE_TYPE_RES_RC;
2415 		break;
2416 	case CMDQ_CREATE_QP_TYPE_UD:
2417 	case CMDQ_CREATE_QP_TYPE_GSI:
2418 		opcode = CQ_BASE_CQE_TYPE_RES_UD;
2419 		break;
2420 	}
2421 
2422 	/* Flush the rest of the RQ */
2423 	start = rq->swq_start;
2424 	cqe = *pcqe;
2425 	while (*budget) {
2426 		last = rq->swq_last;
2427 		if (last == start)
2428 			break;
2429 		memset(cqe, 0, sizeof(*cqe));
2430 		cqe->status =
2431 		    CQ_RES_RC_STATUS_WORK_REQUEST_FLUSHED_ERR;
2432 		cqe->opcode = opcode;
2433 		cqe->qp_handle = (unsigned long)qp;
2434 		cqe->wr_id = rq->swq[last].wr_id;
2435 		cqe++;
2436 		(*budget)--;
2437 		bnxt_qplib_hwq_incr_cons(rq->hwq.max_elements, &rq->hwq.cons,
2438 					 rq->swq[last].slots, &rq->dbinfo.flags);
2439 		rq->swq_last = rq->swq[last].next_idx;
2440 	}
2441 	*pcqe = cqe;
2442 	if (!*budget && rq->swq_last != start)
2443 		/* Out of budget */
2444 		rc = -EAGAIN;
2445 
2446 	return rc;
2447 }
2448 
2449 void bnxt_qplib_mark_qp_error(void *qp_handle)
2450 {
2451 	struct bnxt_qplib_qp *qp = qp_handle;
2452 
2453 	if (!qp)
2454 		return;
2455 
2456 	/* Must block new posting of SQ and RQ */
2457 	qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2458 	bnxt_qplib_cancel_phantom_processing(qp);
2459 }
2460 
2461 /* Note: SQE is valid from sw_sq_cons up to cqe_sq_cons (exclusive)
2462  *       CQE is track from sw_cq_cons to max_element but valid only if VALID=1
2463  */
2464 static int do_wa9060(struct bnxt_qplib_qp *qp, struct bnxt_qplib_cq *cq,
2465 		     u32 cq_cons, u32 swq_last, u32 cqe_sq_cons)
2466 {
2467 	u32 peek_sw_cq_cons, peek_sq_cons_idx, peek_flags;
2468 	struct bnxt_qplib_q *sq = &qp->sq;
2469 	struct cq_req *peek_req_hwcqe;
2470 	struct bnxt_qplib_qp *peek_qp;
2471 	struct bnxt_qplib_q *peek_sq;
2472 	struct bnxt_qplib_swq *swq;
2473 	struct cq_base *peek_hwcqe;
2474 	int i, rc = 0;
2475 
2476 	/* Normal mode */
2477 	/* Check for the psn_search marking before completing */
2478 	swq = &sq->swq[swq_last];
2479 	if (swq->psn_search &&
2480 	    le32_to_cpu(swq->psn_search->flags_next_psn) & 0x80000000) {
2481 		/* Unmark */
2482 		swq->psn_search->flags_next_psn = cpu_to_le32
2483 			(le32_to_cpu(swq->psn_search->flags_next_psn)
2484 				     & ~0x80000000);
2485 		dev_dbg(&cq->hwq.pdev->dev,
2486 			"FP: Process Req cq_cons=0x%x qp=0x%x sq cons sw=0x%x cqe=0x%x marked!\n",
2487 			cq_cons, qp->id, swq_last, cqe_sq_cons);
2488 		sq->condition = true;
2489 		sq->send_phantom = true;
2490 
2491 		/* TODO: Only ARM if the previous SQE is ARMALL */
2492 		bnxt_qplib_ring_db(&cq->dbinfo, DBC_DBC_TYPE_CQ_ARMALL);
2493 		rc = -EAGAIN;
2494 		goto out;
2495 	}
2496 	if (sq->condition) {
2497 		/* Peek at the completions */
2498 		peek_flags = cq->dbinfo.flags;
2499 		peek_sw_cq_cons = cq_cons;
2500 		i = cq->hwq.max_elements;
2501 		while (i--) {
2502 			peek_hwcqe = bnxt_qplib_get_qe(&cq->hwq,
2503 						       peek_sw_cq_cons, NULL);
2504 			/* If the next hwcqe is VALID */
2505 			if (CQE_CMP_VALID(peek_hwcqe, peek_flags)) {
2506 			/*
2507 			 * The valid test of the entry must be done first before
2508 			 * reading any further.
2509 			 */
2510 				dma_rmb();
2511 				/* If the next hwcqe is a REQ */
2512 				if ((peek_hwcqe->cqe_type_toggle &
2513 				    CQ_BASE_CQE_TYPE_MASK) ==
2514 				    CQ_BASE_CQE_TYPE_REQ) {
2515 					peek_req_hwcqe = (struct cq_req *)
2516 							 peek_hwcqe;
2517 					peek_qp = (struct bnxt_qplib_qp *)
2518 						((unsigned long)
2519 						 le64_to_cpu
2520 						 (peek_req_hwcqe->qp_handle));
2521 					peek_sq = &peek_qp->sq;
2522 					peek_sq_cons_idx =
2523 						((le16_to_cpu(
2524 						  peek_req_hwcqe->sq_cons_idx)
2525 						  - 1) % sq->max_wqe);
2526 					/* If the hwcqe's sq's wr_id matches */
2527 					if (peek_sq == sq &&
2528 					    sq->swq[peek_sq_cons_idx].wr_id ==
2529 					    BNXT_QPLIB_FENCE_WRID) {
2530 						/*
2531 						 *  Unbreak only if the phantom
2532 						 *  comes back
2533 						 */
2534 						dev_dbg(&cq->hwq.pdev->dev,
2535 							"FP: Got Phantom CQE\n");
2536 						sq->condition = false;
2537 						sq->single = true;
2538 						rc = 0;
2539 						goto out;
2540 					}
2541 				}
2542 				/* Valid but not the phantom, so keep looping */
2543 			} else {
2544 				/* Not valid yet, just exit and wait */
2545 				rc = -EINVAL;
2546 				goto out;
2547 			}
2548 			bnxt_qplib_hwq_incr_cons(cq->hwq.max_elements,
2549 						 &peek_sw_cq_cons,
2550 						 1, &peek_flags);
2551 		}
2552 		dev_err(&cq->hwq.pdev->dev,
2553 			"Should not have come here! cq_cons=0x%x qp=0x%x sq cons sw=0x%x hw=0x%x\n",
2554 			cq_cons, qp->id, swq_last, cqe_sq_cons);
2555 		rc = -EINVAL;
2556 	}
2557 out:
2558 	return rc;
2559 }
2560 
2561 static int bnxt_qplib_get_cqe_sq_cons(struct bnxt_qplib_q *sq, u32 cqe_slot)
2562 {
2563 	struct bnxt_qplib_hwq *sq_hwq;
2564 	struct bnxt_qplib_swq *swq;
2565 	int cqe_sq_cons = -1;
2566 	u32 start, last;
2567 
2568 	sq_hwq = &sq->hwq;
2569 
2570 	start = sq->swq_start;
2571 	last = sq->swq_last;
2572 
2573 	while (last != start) {
2574 		swq = &sq->swq[last];
2575 		if (swq->slot_idx  == cqe_slot) {
2576 			cqe_sq_cons = swq->next_idx;
2577 			dev_err(&sq_hwq->pdev->dev, "%s: Found cons wqe = %d slot = %d\n",
2578 				__func__, cqe_sq_cons, cqe_slot);
2579 			break;
2580 		}
2581 
2582 		last = swq->next_idx;
2583 	}
2584 	return cqe_sq_cons;
2585 }
2586 
2587 static int bnxt_qplib_cq_process_req(struct bnxt_qplib_cq *cq,
2588 				     struct cq_req *hwcqe,
2589 				     struct bnxt_qplib_cqe **pcqe, int *budget,
2590 				     u32 cq_cons, struct bnxt_qplib_qp **lib_qp)
2591 {
2592 	struct bnxt_qplib_swq *swq;
2593 	struct bnxt_qplib_cqe *cqe;
2594 	u32 cqe_sq_cons, slot_num;
2595 	struct bnxt_qplib_qp *qp;
2596 	struct bnxt_qplib_q *sq;
2597 	int cqe_cons;
2598 	int rc = 0;
2599 
2600 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2601 				      le64_to_cpu(hwcqe->qp_handle));
2602 	if (!qp) {
2603 		dev_err(&cq->hwq.pdev->dev,
2604 			"FP: Process Req qp is NULL\n");
2605 		return -EINVAL;
2606 	}
2607 	sq = &qp->sq;
2608 
2609 	cqe_sq_cons = le16_to_cpu(hwcqe->sq_cons_idx) % sq->max_sw_wqe;
2610 	if (qp->sq.flushed) {
2611 		dev_dbg(&cq->hwq.pdev->dev,
2612 			"%s: QP in Flush QP = %p\n", __func__, qp);
2613 		goto done;
2614 	}
2615 
2616 	if (__is_err_cqe_for_var_wqe(qp, hwcqe->status)) {
2617 		slot_num = le16_to_cpu(hwcqe->sq_cons_idx);
2618 		cqe_cons = bnxt_qplib_get_cqe_sq_cons(sq, slot_num);
2619 		if (cqe_cons < 0) {
2620 			dev_err(&cq->hwq.pdev->dev, "%s: Wrong SQ cons cqe_slot_indx = %d\n",
2621 				__func__, slot_num);
2622 			goto done;
2623 		}
2624 		cqe_sq_cons = cqe_cons;
2625 		dev_err(&cq->hwq.pdev->dev, "%s: cqe_sq_cons = %d swq_last = %d swq_start = %d\n",
2626 			__func__, cqe_sq_cons, sq->swq_last, sq->swq_start);
2627 	}
2628 
2629 	/* Require to walk the sq's swq to fabricate CQEs for all previously
2630 	 * signaled SWQEs due to CQE aggregation from the current sq cons
2631 	 * to the cqe_sq_cons
2632 	 */
2633 	cqe = *pcqe;
2634 	while (*budget) {
2635 		if (sq->swq_last == cqe_sq_cons)
2636 			/* Done */
2637 			break;
2638 
2639 		swq = &sq->swq[sq->swq_last];
2640 		memset(cqe, 0, sizeof(*cqe));
2641 		cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
2642 		cqe->qp_handle = (u64)(unsigned long)qp;
2643 		cqe->src_qp = qp->id;
2644 		cqe->wr_id = swq->wr_id;
2645 		if (cqe->wr_id == BNXT_QPLIB_FENCE_WRID)
2646 			goto skip;
2647 		cqe->type = swq->type;
2648 
2649 		/* For the last CQE, check for status.  For errors, regardless
2650 		 * of the request being signaled or not, it must complete with
2651 		 * the hwcqe error status
2652 		 */
2653 		if (swq->next_idx == cqe_sq_cons &&
2654 		    hwcqe->status != CQ_REQ_STATUS_OK) {
2655 			cqe->status = hwcqe->status;
2656 			dev_err(&cq->hwq.pdev->dev,
2657 				"FP: CQ Processed Req wr_id[%d] = 0x%llx with status 0x%x\n",
2658 				sq->swq_last, cqe->wr_id, cqe->status);
2659 			cqe++;
2660 			(*budget)--;
2661 			bnxt_qplib_mark_qp_error(qp);
2662 			/* Add qp to flush list of the CQ */
2663 			bnxt_qplib_add_flush_qp(qp);
2664 		} else {
2665 			/* Before we complete, do WA 9060 */
2666 			if (!bnxt_qplib_is_chip_gen_p5_p7(qp->cctx)) {
2667 				if (do_wa9060(qp, cq, cq_cons, sq->swq_last,
2668 					      cqe_sq_cons)) {
2669 					*lib_qp = qp;
2670 					goto out;
2671 				}
2672 			}
2673 			if (swq->flags & SQ_SEND_FLAGS_SIGNAL_COMP) {
2674 				cqe->status = CQ_REQ_STATUS_OK;
2675 				cqe++;
2676 				(*budget)--;
2677 			}
2678 		}
2679 skip:
2680 		bnxt_qplib_hwq_incr_cons(sq->hwq.max_elements, &sq->hwq.cons,
2681 					 swq->slots, &sq->dbinfo.flags);
2682 		sq->swq_last = swq->next_idx;
2683 		if (sq->single)
2684 			break;
2685 	}
2686 out:
2687 	*pcqe = cqe;
2688 	if (sq->swq_last != cqe_sq_cons) {
2689 		/* Out of budget */
2690 		rc = -EAGAIN;
2691 		goto done;
2692 	}
2693 	/*
2694 	 * Back to normal completion mode only after it has completed all of
2695 	 * the WC for this CQE
2696 	 */
2697 	sq->single = false;
2698 done:
2699 	return rc;
2700 }
2701 
2702 static void bnxt_qplib_release_srqe(struct bnxt_qplib_srq *srq, u32 tag)
2703 {
2704 	spin_lock(&srq->hwq.lock);
2705 	srq->swq[srq->last_idx].next_idx = (int)tag;
2706 	srq->last_idx = (int)tag;
2707 	srq->swq[srq->last_idx].next_idx = -1;
2708 	bnxt_qplib_hwq_incr_cons(srq->hwq.max_elements, &srq->hwq.cons,
2709 				 srq->dbinfo.max_slot, &srq->dbinfo.flags);
2710 	spin_unlock(&srq->hwq.lock);
2711 }
2712 
2713 static int bnxt_qplib_cq_process_res_rc(struct bnxt_qplib_cq *cq,
2714 					struct cq_res_rc *hwcqe,
2715 					struct bnxt_qplib_cqe **pcqe,
2716 					int *budget)
2717 {
2718 	struct bnxt_qplib_srq *srq;
2719 	struct bnxt_qplib_cqe *cqe;
2720 	struct bnxt_qplib_qp *qp;
2721 	struct bnxt_qplib_q *rq;
2722 	u32 wr_id_idx;
2723 
2724 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2725 				      le64_to_cpu(hwcqe->qp_handle));
2726 	if (!qp) {
2727 		dev_err(&cq->hwq.pdev->dev, "process_cq RC qp is NULL\n");
2728 		return -EINVAL;
2729 	}
2730 	if (qp->rq.flushed) {
2731 		dev_dbg(&cq->hwq.pdev->dev,
2732 			"%s: QP in Flush QP = %p\n", __func__, qp);
2733 		return 0;
2734 	}
2735 
2736 	cqe = *pcqe;
2737 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2738 	cqe->length = le32_to_cpu(hwcqe->length);
2739 	cqe->invrkey = le32_to_cpu(hwcqe->imm_data_or_inv_r_key);
2740 	cqe->mr_handle = le64_to_cpu(hwcqe->mr_handle);
2741 	cqe->flags = le16_to_cpu(hwcqe->flags);
2742 	cqe->status = hwcqe->status;
2743 	cqe->qp_handle = (u64)(unsigned long)qp;
2744 
2745 	wr_id_idx = le32_to_cpu(hwcqe->srq_or_rq_wr_id) &
2746 				CQ_RES_RC_SRQ_OR_RQ_WR_ID_MASK;
2747 	if (cqe->flags & CQ_RES_RC_FLAGS_SRQ_SRQ) {
2748 		srq = qp->srq;
2749 		if (!srq)
2750 			return -EINVAL;
2751 		if (wr_id_idx >= srq->hwq.max_elements) {
2752 			dev_err(&cq->hwq.pdev->dev,
2753 				"FP: CQ Process RC wr_id idx 0x%x exceeded SRQ max 0x%x\n",
2754 				wr_id_idx, srq->hwq.max_elements);
2755 			return -EINVAL;
2756 		}
2757 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2758 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2759 		cqe++;
2760 		(*budget)--;
2761 		*pcqe = cqe;
2762 	} else {
2763 		struct bnxt_qplib_swq *swq;
2764 
2765 		rq = &qp->rq;
2766 		if (wr_id_idx > (rq->max_wqe - 1)) {
2767 			dev_err(&cq->hwq.pdev->dev,
2768 				"FP: CQ Process RC wr_id idx 0x%x exceeded RQ max 0x%x\n",
2769 				wr_id_idx, rq->max_wqe);
2770 			return -EINVAL;
2771 		}
2772 		if (wr_id_idx != rq->swq_last)
2773 			return -EINVAL;
2774 		swq = &rq->swq[rq->swq_last];
2775 		cqe->wr_id = swq->wr_id;
2776 		cqe++;
2777 		(*budget)--;
2778 		bnxt_qplib_hwq_incr_cons(rq->hwq.max_elements, &rq->hwq.cons,
2779 					 swq->slots, &rq->dbinfo.flags);
2780 		rq->swq_last = swq->next_idx;
2781 		*pcqe = cqe;
2782 
2783 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2784 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2785 			/* Add qp to flush list of the CQ */
2786 			bnxt_qplib_add_flush_qp(qp);
2787 		}
2788 	}
2789 
2790 	return 0;
2791 }
2792 
2793 static int bnxt_qplib_cq_process_res_ud(struct bnxt_qplib_cq *cq,
2794 					struct cq_res_ud *hwcqe,
2795 					struct bnxt_qplib_cqe **pcqe,
2796 					int *budget)
2797 {
2798 	struct bnxt_qplib_srq *srq;
2799 	struct bnxt_qplib_cqe *cqe;
2800 	struct bnxt_qplib_qp *qp;
2801 	struct bnxt_qplib_q *rq;
2802 	u32 wr_id_idx;
2803 
2804 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2805 				      le64_to_cpu(hwcqe->qp_handle));
2806 	if (!qp) {
2807 		dev_err(&cq->hwq.pdev->dev, "process_cq UD qp is NULL\n");
2808 		return -EINVAL;
2809 	}
2810 	if (qp->rq.flushed) {
2811 		dev_dbg(&cq->hwq.pdev->dev,
2812 			"%s: QP in Flush QP = %p\n", __func__, qp);
2813 		return 0;
2814 	}
2815 	cqe = *pcqe;
2816 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2817 	cqe->length = le16_to_cpu(hwcqe->length) & CQ_RES_UD_LENGTH_MASK;
2818 	cqe->cfa_meta = le16_to_cpu(hwcqe->cfa_metadata);
2819 	cqe->invrkey = le32_to_cpu(hwcqe->imm_data);
2820 	cqe->flags = le16_to_cpu(hwcqe->flags);
2821 	cqe->status = hwcqe->status;
2822 	cqe->qp_handle = (u64)(unsigned long)qp;
2823 	/*FIXME: Endianness fix needed for smace */
2824 	memcpy(cqe->smac, hwcqe->src_mac, ETH_ALEN);
2825 	wr_id_idx = le32_to_cpu(hwcqe->src_qp_high_srq_or_rq_wr_id)
2826 				& CQ_RES_UD_SRQ_OR_RQ_WR_ID_MASK;
2827 	cqe->src_qp = le16_to_cpu(hwcqe->src_qp_low) |
2828 				  ((le32_to_cpu(
2829 				  hwcqe->src_qp_high_srq_or_rq_wr_id) &
2830 				 CQ_RES_UD_SRC_QP_HIGH_MASK) >> 8);
2831 
2832 	if (cqe->flags & CQ_RES_RC_FLAGS_SRQ_SRQ) {
2833 		srq = qp->srq;
2834 		if (!srq)
2835 			return -EINVAL;
2836 
2837 		if (wr_id_idx >= srq->hwq.max_elements) {
2838 			dev_err(&cq->hwq.pdev->dev,
2839 				"FP: CQ Process UD wr_id idx 0x%x exceeded SRQ max 0x%x\n",
2840 				wr_id_idx, srq->hwq.max_elements);
2841 			return -EINVAL;
2842 		}
2843 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2844 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2845 		cqe++;
2846 		(*budget)--;
2847 		*pcqe = cqe;
2848 	} else {
2849 		struct bnxt_qplib_swq *swq;
2850 
2851 		rq = &qp->rq;
2852 		if (wr_id_idx > (rq->max_wqe - 1)) {
2853 			dev_err(&cq->hwq.pdev->dev,
2854 				"FP: CQ Process UD wr_id idx 0x%x exceeded RQ max 0x%x\n",
2855 				wr_id_idx, rq->max_wqe);
2856 			return -EINVAL;
2857 		}
2858 
2859 		if (rq->swq_last != wr_id_idx)
2860 			return -EINVAL;
2861 		swq = &rq->swq[rq->swq_last];
2862 		cqe->wr_id = swq->wr_id;
2863 		cqe++;
2864 		(*budget)--;
2865 		bnxt_qplib_hwq_incr_cons(rq->hwq.max_elements, &rq->hwq.cons,
2866 					 swq->slots, &rq->dbinfo.flags);
2867 		rq->swq_last = swq->next_idx;
2868 		*pcqe = cqe;
2869 
2870 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2871 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2872 			/* Add qp to flush list of the CQ */
2873 			bnxt_qplib_add_flush_qp(qp);
2874 		}
2875 	}
2876 
2877 	return 0;
2878 }
2879 
2880 bool bnxt_qplib_is_cq_empty(struct bnxt_qplib_cq *cq)
2881 {
2882 	struct cq_base *hw_cqe;
2883 	bool rc = true;
2884 
2885 	hw_cqe = bnxt_qplib_get_qe(&cq->hwq, cq->hwq.cons, NULL);
2886 	 /* Check for Valid bit. If the CQE is valid, return false */
2887 	rc = !CQE_CMP_VALID(hw_cqe, cq->dbinfo.flags);
2888 	return rc;
2889 }
2890 
2891 static int bnxt_qplib_cq_process_res_raweth_qp1(struct bnxt_qplib_cq *cq,
2892 						struct cq_res_raweth_qp1 *hwcqe,
2893 						struct bnxt_qplib_cqe **pcqe,
2894 						int *budget)
2895 {
2896 	struct bnxt_qplib_qp *qp;
2897 	struct bnxt_qplib_q *rq;
2898 	struct bnxt_qplib_srq *srq;
2899 	struct bnxt_qplib_cqe *cqe;
2900 	u32 wr_id_idx;
2901 
2902 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2903 				      le64_to_cpu(hwcqe->qp_handle));
2904 	if (!qp) {
2905 		dev_err(&cq->hwq.pdev->dev, "process_cq Raw/QP1 qp is NULL\n");
2906 		return -EINVAL;
2907 	}
2908 	if (qp->rq.flushed) {
2909 		dev_dbg(&cq->hwq.pdev->dev,
2910 			"%s: QP in Flush QP = %p\n", __func__, qp);
2911 		return 0;
2912 	}
2913 	cqe = *pcqe;
2914 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2915 	cqe->flags = le16_to_cpu(hwcqe->flags);
2916 	cqe->qp_handle = (u64)(unsigned long)qp;
2917 
2918 	wr_id_idx =
2919 		le32_to_cpu(hwcqe->raweth_qp1_payload_offset_srq_or_rq_wr_id)
2920 				& CQ_RES_RAWETH_QP1_SRQ_OR_RQ_WR_ID_MASK;
2921 	cqe->src_qp = qp->id;
2922 	if (qp->id == 1 && !cqe->length) {
2923 		/* Add workaround for the length misdetection */
2924 		cqe->length = 296;
2925 	} else {
2926 		cqe->length = le16_to_cpu(hwcqe->length);
2927 	}
2928 	cqe->pkey_index = qp->pkey_index;
2929 	memcpy(cqe->smac, qp->smac, 6);
2930 
2931 	cqe->raweth_qp1_flags = le16_to_cpu(hwcqe->raweth_qp1_flags);
2932 	cqe->raweth_qp1_flags2 = le32_to_cpu(hwcqe->raweth_qp1_flags2);
2933 	cqe->raweth_qp1_metadata = le32_to_cpu(hwcqe->raweth_qp1_metadata);
2934 
2935 	if (cqe->flags & CQ_RES_RAWETH_QP1_FLAGS_SRQ_SRQ) {
2936 		srq = qp->srq;
2937 		if (!srq) {
2938 			dev_err(&cq->hwq.pdev->dev,
2939 				"FP: SRQ used but not defined??\n");
2940 			return -EINVAL;
2941 		}
2942 		if (wr_id_idx >= srq->hwq.max_elements) {
2943 			dev_err(&cq->hwq.pdev->dev,
2944 				"FP: CQ Process Raw/QP1 wr_id idx 0x%x exceeded SRQ max 0x%x\n",
2945 				wr_id_idx, srq->hwq.max_elements);
2946 			return -EINVAL;
2947 		}
2948 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2949 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2950 		cqe++;
2951 		(*budget)--;
2952 		*pcqe = cqe;
2953 	} else {
2954 		struct bnxt_qplib_swq *swq;
2955 
2956 		rq = &qp->rq;
2957 		if (wr_id_idx > (rq->max_wqe - 1)) {
2958 			dev_err(&cq->hwq.pdev->dev,
2959 				"FP: CQ Process Raw/QP1 RQ wr_id idx 0x%x exceeded RQ max 0x%x\n",
2960 				wr_id_idx, rq->max_wqe);
2961 			return -EINVAL;
2962 		}
2963 		if (rq->swq_last != wr_id_idx)
2964 			return -EINVAL;
2965 		swq = &rq->swq[rq->swq_last];
2966 		cqe->wr_id = swq->wr_id;
2967 		cqe++;
2968 		(*budget)--;
2969 		bnxt_qplib_hwq_incr_cons(rq->hwq.max_elements, &rq->hwq.cons,
2970 					 swq->slots, &rq->dbinfo.flags);
2971 		rq->swq_last = swq->next_idx;
2972 		*pcqe = cqe;
2973 
2974 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2975 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2976 			/* Add qp to flush list of the CQ */
2977 			bnxt_qplib_add_flush_qp(qp);
2978 		}
2979 	}
2980 
2981 	return 0;
2982 }
2983 
2984 static int bnxt_qplib_cq_process_terminal(struct bnxt_qplib_cq *cq,
2985 					  struct cq_terminal *hwcqe,
2986 					  struct bnxt_qplib_cqe **pcqe,
2987 					  int *budget)
2988 {
2989 	struct bnxt_qplib_qp *qp;
2990 	struct bnxt_qplib_q *sq, *rq;
2991 	struct bnxt_qplib_cqe *cqe;
2992 	u32 swq_last = 0, cqe_cons;
2993 	int rc = 0;
2994 
2995 	/* Check the Status */
2996 	if (hwcqe->status != CQ_TERMINAL_STATUS_OK)
2997 		dev_warn(&cq->hwq.pdev->dev,
2998 			 "FP: CQ Process Terminal Error status = 0x%x\n",
2999 			 hwcqe->status);
3000 
3001 	qp = (struct bnxt_qplib_qp *)((unsigned long)
3002 				      le64_to_cpu(hwcqe->qp_handle));
3003 	if (!qp)
3004 		return -EINVAL;
3005 
3006 	/* Must block new posting of SQ and RQ */
3007 	qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
3008 
3009 	sq = &qp->sq;
3010 	rq = &qp->rq;
3011 
3012 	cqe_cons = le16_to_cpu(hwcqe->sq_cons_idx);
3013 	if (cqe_cons == 0xFFFF)
3014 		goto do_rq;
3015 	cqe_cons %= sq->max_sw_wqe;
3016 
3017 	if (qp->sq.flushed) {
3018 		dev_dbg(&cq->hwq.pdev->dev,
3019 			"%s: QP in Flush QP = %p\n", __func__, qp);
3020 		goto sq_done;
3021 	}
3022 
3023 	/* Terminal CQE can also include aggregated successful CQEs prior.
3024 	 * So we must complete all CQEs from the current sq's cons to the
3025 	 * cq_cons with status OK
3026 	 */
3027 	cqe = *pcqe;
3028 	while (*budget) {
3029 		swq_last = sq->swq_last;
3030 		if (swq_last == cqe_cons)
3031 			break;
3032 		if (sq->swq[swq_last].flags & SQ_SEND_FLAGS_SIGNAL_COMP) {
3033 			memset(cqe, 0, sizeof(*cqe));
3034 			cqe->status = CQ_REQ_STATUS_OK;
3035 			cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
3036 			cqe->qp_handle = (u64)(unsigned long)qp;
3037 			cqe->src_qp = qp->id;
3038 			cqe->wr_id = sq->swq[swq_last].wr_id;
3039 			cqe->type = sq->swq[swq_last].type;
3040 			cqe++;
3041 			(*budget)--;
3042 		}
3043 		bnxt_qplib_hwq_incr_cons(sq->hwq.max_elements, &sq->hwq.cons,
3044 					 sq->swq[swq_last].slots, &sq->dbinfo.flags);
3045 		sq->swq_last = sq->swq[swq_last].next_idx;
3046 	}
3047 	*pcqe = cqe;
3048 	if (!(*budget) && swq_last != cqe_cons) {
3049 		/* Out of budget */
3050 		rc = -EAGAIN;
3051 		goto sq_done;
3052 	}
3053 sq_done:
3054 	if (rc)
3055 		return rc;
3056 do_rq:
3057 	cqe_cons = le16_to_cpu(hwcqe->rq_cons_idx);
3058 	if (cqe_cons == 0xFFFF) {
3059 		goto done;
3060 	} else if (cqe_cons > rq->max_wqe - 1) {
3061 		dev_err(&cq->hwq.pdev->dev,
3062 			"FP: CQ Processed terminal reported rq_cons_idx 0x%x exceeds max 0x%x\n",
3063 			cqe_cons, rq->max_wqe);
3064 		rc = -EINVAL;
3065 		goto done;
3066 	}
3067 
3068 	if (qp->rq.flushed) {
3069 		dev_dbg(&cq->hwq.pdev->dev,
3070 			"%s: QP in Flush QP = %p\n", __func__, qp);
3071 		rc = 0;
3072 		goto done;
3073 	}
3074 
3075 	/* Terminal CQE requires all posted RQEs to complete with FLUSHED_ERR
3076 	 * from the current rq->cons to the rq->prod regardless what the
3077 	 * rq->cons the terminal CQE indicates
3078 	 */
3079 
3080 	/* Add qp to flush list of the CQ */
3081 	bnxt_qplib_add_flush_qp(qp);
3082 done:
3083 	return rc;
3084 }
3085 
3086 static int bnxt_qplib_cq_process_cutoff(struct bnxt_qplib_cq *cq,
3087 					struct cq_cutoff *hwcqe)
3088 {
3089 	/* Check the Status */
3090 	if (hwcqe->status != CQ_CUTOFF_STATUS_OK) {
3091 		dev_err(&cq->hwq.pdev->dev,
3092 			"FP: CQ Process Cutoff Error status = 0x%x\n",
3093 			hwcqe->status);
3094 		return -EINVAL;
3095 	}
3096 	clear_bit(CQ_FLAGS_RESIZE_IN_PROG, &cq->flags);
3097 	wake_up_interruptible(&cq->waitq);
3098 
3099 	return 0;
3100 }
3101 
3102 int bnxt_qplib_process_flush_list(struct bnxt_qplib_cq *cq,
3103 				  struct bnxt_qplib_cqe *cqe,
3104 				  int num_cqes)
3105 {
3106 	struct bnxt_qplib_qp *qp = NULL;
3107 	u32 budget = num_cqes;
3108 	unsigned long flags;
3109 
3110 	spin_lock_irqsave(&cq->flush_lock, flags);
3111 	list_for_each_entry(qp, &cq->sqf_head, sq_flush) {
3112 		dev_dbg(&cq->hwq.pdev->dev, "FP: Flushing SQ QP= %p\n", qp);
3113 		__flush_sq(&qp->sq, qp, &cqe, &budget);
3114 	}
3115 
3116 	list_for_each_entry(qp, &cq->rqf_head, rq_flush) {
3117 		dev_dbg(&cq->hwq.pdev->dev, "FP: Flushing RQ QP= %p\n", qp);
3118 		__flush_rq(&qp->rq, qp, &cqe, &budget);
3119 	}
3120 	spin_unlock_irqrestore(&cq->flush_lock, flags);
3121 
3122 	return num_cqes - budget;
3123 }
3124 
3125 int bnxt_qplib_poll_cq(struct bnxt_qplib_cq *cq, struct bnxt_qplib_cqe *cqe,
3126 		       int num_cqes, struct bnxt_qplib_qp **lib_qp)
3127 {
3128 	struct cq_base *hw_cqe;
3129 	int budget, rc = 0;
3130 	u32 hw_polled = 0;
3131 	u8 type;
3132 
3133 	budget = num_cqes;
3134 
3135 	while (budget) {
3136 		hw_cqe = bnxt_qplib_get_qe(&cq->hwq, cq->hwq.cons, NULL);
3137 
3138 		/* Check for Valid bit */
3139 		if (!CQE_CMP_VALID(hw_cqe, cq->dbinfo.flags))
3140 			break;
3141 
3142 		/*
3143 		 * The valid test of the entry must be done first before
3144 		 * reading any further.
3145 		 */
3146 		dma_rmb();
3147 		/* From the device's respective CQE format to qplib_wc*/
3148 		type = hw_cqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
3149 		switch (type) {
3150 		case CQ_BASE_CQE_TYPE_REQ:
3151 			rc = bnxt_qplib_cq_process_req(cq,
3152 						       (struct cq_req *)hw_cqe,
3153 						       &cqe, &budget,
3154 						       cq->hwq.cons, lib_qp);
3155 			break;
3156 		case CQ_BASE_CQE_TYPE_RES_RC:
3157 			rc = bnxt_qplib_cq_process_res_rc(cq,
3158 							  (struct cq_res_rc *)
3159 							  hw_cqe, &cqe,
3160 							  &budget);
3161 			break;
3162 		case CQ_BASE_CQE_TYPE_RES_UD:
3163 			rc = bnxt_qplib_cq_process_res_ud
3164 					(cq, (struct cq_res_ud *)hw_cqe, &cqe,
3165 					 &budget);
3166 			break;
3167 		case CQ_BASE_CQE_TYPE_RES_RAWETH_QP1:
3168 			rc = bnxt_qplib_cq_process_res_raweth_qp1
3169 					(cq, (struct cq_res_raweth_qp1 *)
3170 					 hw_cqe, &cqe, &budget);
3171 			break;
3172 		case CQ_BASE_CQE_TYPE_TERMINAL:
3173 			rc = bnxt_qplib_cq_process_terminal
3174 					(cq, (struct cq_terminal *)hw_cqe,
3175 					 &cqe, &budget);
3176 			break;
3177 		case CQ_BASE_CQE_TYPE_CUT_OFF:
3178 			bnxt_qplib_cq_process_cutoff
3179 					(cq, (struct cq_cutoff *)hw_cqe);
3180 			/* Done processing this CQ */
3181 			goto exit;
3182 		default:
3183 			dev_err(&cq->hwq.pdev->dev,
3184 				"process_cq unknown type 0x%lx\n",
3185 				hw_cqe->cqe_type_toggle &
3186 				CQ_BASE_CQE_TYPE_MASK);
3187 			rc = -EINVAL;
3188 			break;
3189 		}
3190 		if (rc < 0) {
3191 			if (rc == -EAGAIN)
3192 				break;
3193 			/* Error while processing the CQE, just skip to the
3194 			 * next one
3195 			 */
3196 			if (type != CQ_BASE_CQE_TYPE_TERMINAL)
3197 				dev_err(&cq->hwq.pdev->dev,
3198 					"process_cqe error rc = 0x%x\n", rc);
3199 		}
3200 		hw_polled++;
3201 		bnxt_qplib_hwq_incr_cons(cq->hwq.max_elements, &cq->hwq.cons,
3202 					 1, &cq->dbinfo.flags);
3203 
3204 	}
3205 	if (hw_polled)
3206 		bnxt_qplib_ring_db(&cq->dbinfo, DBC_DBC_TYPE_CQ);
3207 exit:
3208 	return num_cqes - budget;
3209 }
3210 
3211 void bnxt_qplib_req_notify_cq(struct bnxt_qplib_cq *cq, u32 arm_type)
3212 {
3213 	cq->dbinfo.toggle = cq->toggle;
3214 	if (arm_type)
3215 		bnxt_qplib_ring_db(&cq->dbinfo, arm_type);
3216 	/* Using cq->arm_state variable to track whether to issue cq handler */
3217 	atomic_set(&cq->arm_state, 1);
3218 }
3219 
3220 void bnxt_qplib_flush_cqn_wq(struct bnxt_qplib_qp *qp)
3221 {
3222 	flush_workqueue(qp->scq->nq->cqn_wq);
3223 	if (qp->scq != qp->rcq)
3224 		flush_workqueue(qp->rcq->nq->cqn_wq);
3225 }
3226