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/prefetch.h>
47 #include <linux/if_ether.h>
48 
49 #include "roce_hsi.h"
50 
51 #include "qplib_res.h"
52 #include "qplib_rcfw.h"
53 #include "qplib_sp.h"
54 #include "qplib_fp.h"
55 
56 static void bnxt_qplib_arm_cq_enable(struct bnxt_qplib_cq *cq);
57 static void __clean_cq(struct bnxt_qplib_cq *cq, u64 qp);
58 static void bnxt_qplib_arm_srq(struct bnxt_qplib_srq *srq, u32 arm_type);
59 
60 static void bnxt_qplib_cancel_phantom_processing(struct bnxt_qplib_qp *qp)
61 {
62 	qp->sq.condition = false;
63 	qp->sq.send_phantom = false;
64 	qp->sq.single = false;
65 }
66 
67 /* Flush list */
68 static void __bnxt_qplib_add_flush_qp(struct bnxt_qplib_qp *qp)
69 {
70 	struct bnxt_qplib_cq *scq, *rcq;
71 
72 	scq = qp->scq;
73 	rcq = qp->rcq;
74 
75 	if (!qp->sq.flushed) {
76 		dev_dbg(&scq->hwq.pdev->dev,
77 			"FP: Adding to SQ Flush list = %p\n", qp);
78 		bnxt_qplib_cancel_phantom_processing(qp);
79 		list_add_tail(&qp->sq_flush, &scq->sqf_head);
80 		qp->sq.flushed = true;
81 	}
82 	if (!qp->srq) {
83 		if (!qp->rq.flushed) {
84 			dev_dbg(&rcq->hwq.pdev->dev,
85 				"FP: Adding to RQ Flush list = %p\n", qp);
86 			list_add_tail(&qp->rq_flush, &rcq->rqf_head);
87 			qp->rq.flushed = true;
88 		}
89 	}
90 }
91 
92 static void bnxt_qplib_acquire_cq_flush_locks(struct bnxt_qplib_qp *qp,
93 				       unsigned long *flags)
94 	__acquires(&qp->scq->flush_lock) __acquires(&qp->rcq->flush_lock)
95 {
96 	spin_lock_irqsave(&qp->scq->flush_lock, *flags);
97 	if (qp->scq == qp->rcq)
98 		__acquire(&qp->rcq->flush_lock);
99 	else
100 		spin_lock(&qp->rcq->flush_lock);
101 }
102 
103 static void bnxt_qplib_release_cq_flush_locks(struct bnxt_qplib_qp *qp,
104 				       unsigned long *flags)
105 	__releases(&qp->scq->flush_lock) __releases(&qp->rcq->flush_lock)
106 {
107 	if (qp->scq == qp->rcq)
108 		__release(&qp->rcq->flush_lock);
109 	else
110 		spin_unlock(&qp->rcq->flush_lock);
111 	spin_unlock_irqrestore(&qp->scq->flush_lock, *flags);
112 }
113 
114 void bnxt_qplib_add_flush_qp(struct bnxt_qplib_qp *qp)
115 {
116 	unsigned long flags;
117 
118 	bnxt_qplib_acquire_cq_flush_locks(qp, &flags);
119 	__bnxt_qplib_add_flush_qp(qp);
120 	bnxt_qplib_release_cq_flush_locks(qp, &flags);
121 }
122 
123 static void __bnxt_qplib_del_flush_qp(struct bnxt_qplib_qp *qp)
124 {
125 	if (qp->sq.flushed) {
126 		qp->sq.flushed = false;
127 		list_del(&qp->sq_flush);
128 	}
129 	if (!qp->srq) {
130 		if (qp->rq.flushed) {
131 			qp->rq.flushed = false;
132 			list_del(&qp->rq_flush);
133 		}
134 	}
135 }
136 
137 void bnxt_qplib_clean_qp(struct bnxt_qplib_qp *qp)
138 {
139 	unsigned long flags;
140 
141 	bnxt_qplib_acquire_cq_flush_locks(qp, &flags);
142 	__clean_cq(qp->scq, (u64)(unsigned long)qp);
143 	qp->sq.hwq.prod = 0;
144 	qp->sq.hwq.cons = 0;
145 	__clean_cq(qp->rcq, (u64)(unsigned long)qp);
146 	qp->rq.hwq.prod = 0;
147 	qp->rq.hwq.cons = 0;
148 
149 	__bnxt_qplib_del_flush_qp(qp);
150 	bnxt_qplib_release_cq_flush_locks(qp, &flags);
151 }
152 
153 static void bnxt_qpn_cqn_sched_task(struct work_struct *work)
154 {
155 	struct bnxt_qplib_nq_work *nq_work =
156 			container_of(work, struct bnxt_qplib_nq_work, work);
157 
158 	struct bnxt_qplib_cq *cq = nq_work->cq;
159 	struct bnxt_qplib_nq *nq = nq_work->nq;
160 
161 	if (cq && nq) {
162 		spin_lock_bh(&cq->compl_lock);
163 		if (atomic_read(&cq->arm_state) && nq->cqn_handler) {
164 			dev_dbg(&nq->pdev->dev,
165 				"%s:Trigger cq  = %p event nq = %p\n",
166 				__func__, cq, nq);
167 			nq->cqn_handler(nq, cq);
168 		}
169 		spin_unlock_bh(&cq->compl_lock);
170 	}
171 	kfree(nq_work);
172 }
173 
174 static void bnxt_qplib_free_qp_hdr_buf(struct bnxt_qplib_res *res,
175 				       struct bnxt_qplib_qp *qp)
176 {
177 	struct bnxt_qplib_q *rq = &qp->rq;
178 	struct bnxt_qplib_q *sq = &qp->sq;
179 
180 	if (qp->rq_hdr_buf)
181 		dma_free_coherent(&res->pdev->dev,
182 				  rq->hwq.max_elements * qp->rq_hdr_buf_size,
183 				  qp->rq_hdr_buf, qp->rq_hdr_buf_map);
184 	if (qp->sq_hdr_buf)
185 		dma_free_coherent(&res->pdev->dev,
186 				  sq->hwq.max_elements * qp->sq_hdr_buf_size,
187 				  qp->sq_hdr_buf, qp->sq_hdr_buf_map);
188 	qp->rq_hdr_buf = NULL;
189 	qp->sq_hdr_buf = NULL;
190 	qp->rq_hdr_buf_map = 0;
191 	qp->sq_hdr_buf_map = 0;
192 	qp->sq_hdr_buf_size = 0;
193 	qp->rq_hdr_buf_size = 0;
194 }
195 
196 static int bnxt_qplib_alloc_qp_hdr_buf(struct bnxt_qplib_res *res,
197 				       struct bnxt_qplib_qp *qp)
198 {
199 	struct bnxt_qplib_q *rq = &qp->rq;
200 	struct bnxt_qplib_q *sq = &qp->sq;
201 	int rc = 0;
202 
203 	if (qp->sq_hdr_buf_size && sq->hwq.max_elements) {
204 		qp->sq_hdr_buf = dma_alloc_coherent(&res->pdev->dev,
205 					sq->hwq.max_elements *
206 					qp->sq_hdr_buf_size,
207 					&qp->sq_hdr_buf_map, GFP_KERNEL);
208 		if (!qp->sq_hdr_buf) {
209 			rc = -ENOMEM;
210 			dev_err(&res->pdev->dev,
211 				"Failed to create sq_hdr_buf\n");
212 			goto fail;
213 		}
214 	}
215 
216 	if (qp->rq_hdr_buf_size && rq->hwq.max_elements) {
217 		qp->rq_hdr_buf = dma_alloc_coherent(&res->pdev->dev,
218 						    rq->hwq.max_elements *
219 						    qp->rq_hdr_buf_size,
220 						    &qp->rq_hdr_buf_map,
221 						    GFP_KERNEL);
222 		if (!qp->rq_hdr_buf) {
223 			rc = -ENOMEM;
224 			dev_err(&res->pdev->dev,
225 				"Failed to create rq_hdr_buf\n");
226 			goto fail;
227 		}
228 	}
229 	return 0;
230 
231 fail:
232 	bnxt_qplib_free_qp_hdr_buf(res, qp);
233 	return rc;
234 }
235 
236 static void bnxt_qplib_service_nq(unsigned long data)
237 {
238 	struct bnxt_qplib_nq *nq = (struct bnxt_qplib_nq *)data;
239 	struct bnxt_qplib_hwq *hwq = &nq->hwq;
240 	struct nq_base *nqe, **nq_ptr;
241 	struct bnxt_qplib_cq *cq;
242 	int num_cqne_processed = 0;
243 	int num_srqne_processed = 0;
244 	u32 sw_cons, raw_cons;
245 	u16 type;
246 	int budget = nq->budget;
247 	uintptr_t q_handle;
248 	bool gen_p5 = bnxt_qplib_is_chip_gen_p5(nq->res->cctx);
249 
250 	/* Service the NQ until empty */
251 	raw_cons = hwq->cons;
252 	while (budget--) {
253 		sw_cons = HWQ_CMP(raw_cons, hwq);
254 		nq_ptr = (struct nq_base **)hwq->pbl_ptr;
255 		nqe = &nq_ptr[NQE_PG(sw_cons)][NQE_IDX(sw_cons)];
256 		if (!NQE_CMP_VALID(nqe, raw_cons, hwq->max_elements))
257 			break;
258 
259 		/*
260 		 * The valid test of the entry must be done first before
261 		 * reading any further.
262 		 */
263 		dma_rmb();
264 
265 		type = le16_to_cpu(nqe->info10_type) & NQ_BASE_TYPE_MASK;
266 		switch (type) {
267 		case NQ_BASE_TYPE_CQ_NOTIFICATION:
268 		{
269 			struct nq_cn *nqcne = (struct nq_cn *)nqe;
270 
271 			q_handle = le32_to_cpu(nqcne->cq_handle_low);
272 			q_handle |= (u64)le32_to_cpu(nqcne->cq_handle_high)
273 						     << 32;
274 			cq = (struct bnxt_qplib_cq *)(unsigned long)q_handle;
275 			bnxt_qplib_arm_cq_enable(cq);
276 			spin_lock_bh(&cq->compl_lock);
277 			atomic_set(&cq->arm_state, 0);
278 			if (!nq->cqn_handler(nq, (cq)))
279 				num_cqne_processed++;
280 			else
281 				dev_warn(&nq->pdev->dev,
282 					 "cqn - type 0x%x not handled\n", type);
283 			spin_unlock_bh(&cq->compl_lock);
284 			break;
285 		}
286 		case NQ_BASE_TYPE_SRQ_EVENT:
287 		{
288 			struct nq_srq_event *nqsrqe =
289 						(struct nq_srq_event *)nqe;
290 
291 			q_handle = le32_to_cpu(nqsrqe->srq_handle_low);
292 			q_handle |= (u64)le32_to_cpu(nqsrqe->srq_handle_high)
293 				     << 32;
294 			bnxt_qplib_arm_srq((struct bnxt_qplib_srq *)q_handle,
295 					   DBC_DBC_TYPE_SRQ_ARMENA);
296 			if (!nq->srqn_handler(nq,
297 					      (struct bnxt_qplib_srq *)q_handle,
298 					      nqsrqe->event))
299 				num_srqne_processed++;
300 			else
301 				dev_warn(&nq->pdev->dev,
302 					 "SRQ event 0x%x not handled\n",
303 					 nqsrqe->event);
304 			break;
305 		}
306 		case NQ_BASE_TYPE_DBQ_EVENT:
307 			break;
308 		default:
309 			dev_warn(&nq->pdev->dev,
310 				 "nqe with type = 0x%x not handled\n", type);
311 			break;
312 		}
313 		raw_cons++;
314 	}
315 	if (hwq->cons != raw_cons) {
316 		hwq->cons = raw_cons;
317 		bnxt_qplib_ring_nq_db_rearm(nq->bar_reg_iomem, hwq->cons,
318 					    hwq->max_elements, nq->ring_id,
319 					    gen_p5);
320 	}
321 }
322 
323 static irqreturn_t bnxt_qplib_nq_irq(int irq, void *dev_instance)
324 {
325 	struct bnxt_qplib_nq *nq = dev_instance;
326 	struct bnxt_qplib_hwq *hwq = &nq->hwq;
327 	struct nq_base **nq_ptr;
328 	u32 sw_cons;
329 
330 	/* Prefetch the NQ element */
331 	sw_cons = HWQ_CMP(hwq->cons, hwq);
332 	nq_ptr = (struct nq_base **)nq->hwq.pbl_ptr;
333 	prefetch(&nq_ptr[NQE_PG(sw_cons)][NQE_IDX(sw_cons)]);
334 
335 	/* Fan out to CPU affinitized kthreads? */
336 	tasklet_schedule(&nq->worker);
337 
338 	return IRQ_HANDLED;
339 }
340 
341 void bnxt_qplib_nq_stop_irq(struct bnxt_qplib_nq *nq, bool kill)
342 {
343 	bool gen_p5 = bnxt_qplib_is_chip_gen_p5(nq->res->cctx);
344 	tasklet_disable(&nq->worker);
345 	/* Mask h/w interrupt */
346 	bnxt_qplib_ring_nq_db(nq->bar_reg_iomem, nq->hwq.cons,
347 			      nq->hwq.max_elements, nq->ring_id, gen_p5);
348 	/* Sync with last running IRQ handler */
349 	synchronize_irq(nq->vector);
350 	if (kill)
351 		tasklet_kill(&nq->worker);
352 	if (nq->requested) {
353 		irq_set_affinity_hint(nq->vector, NULL);
354 		free_irq(nq->vector, nq);
355 		nq->requested = false;
356 	}
357 }
358 
359 void bnxt_qplib_disable_nq(struct bnxt_qplib_nq *nq)
360 {
361 	if (nq->cqn_wq) {
362 		destroy_workqueue(nq->cqn_wq);
363 		nq->cqn_wq = NULL;
364 	}
365 
366 	/* Make sure the HW is stopped! */
367 	if (nq->requested)
368 		bnxt_qplib_nq_stop_irq(nq, true);
369 
370 	if (nq->bar_reg_iomem)
371 		iounmap(nq->bar_reg_iomem);
372 	nq->bar_reg_iomem = NULL;
373 
374 	nq->cqn_handler = NULL;
375 	nq->srqn_handler = NULL;
376 	nq->vector = 0;
377 }
378 
379 int bnxt_qplib_nq_start_irq(struct bnxt_qplib_nq *nq, int nq_indx,
380 			    int msix_vector, bool need_init)
381 {
382 	bool gen_p5 = bnxt_qplib_is_chip_gen_p5(nq->res->cctx);
383 	int rc;
384 
385 	if (nq->requested)
386 		return -EFAULT;
387 
388 	nq->vector = msix_vector;
389 	if (need_init)
390 		tasklet_init(&nq->worker, bnxt_qplib_service_nq,
391 			     (unsigned long)nq);
392 	else
393 		tasklet_enable(&nq->worker);
394 
395 	snprintf(nq->name, sizeof(nq->name), "bnxt_qplib_nq-%d", nq_indx);
396 	rc = request_irq(nq->vector, bnxt_qplib_nq_irq, 0, nq->name, nq);
397 	if (rc)
398 		return rc;
399 
400 	cpumask_clear(&nq->mask);
401 	cpumask_set_cpu(nq_indx, &nq->mask);
402 	rc = irq_set_affinity_hint(nq->vector, &nq->mask);
403 	if (rc) {
404 		dev_warn(&nq->pdev->dev,
405 			 "set affinity failed; vector: %d nq_idx: %d\n",
406 			 nq->vector, nq_indx);
407 	}
408 	nq->requested = true;
409 	bnxt_qplib_ring_nq_db_rearm(nq->bar_reg_iomem, nq->hwq.cons,
410 				    nq->hwq.max_elements, nq->ring_id, gen_p5);
411 
412 	return rc;
413 }
414 
415 int bnxt_qplib_enable_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq,
416 			 int nq_idx, int msix_vector, int bar_reg_offset,
417 			 int (*cqn_handler)(struct bnxt_qplib_nq *nq,
418 					    struct bnxt_qplib_cq *),
419 			 int (*srqn_handler)(struct bnxt_qplib_nq *nq,
420 					     struct bnxt_qplib_srq *,
421 					     u8 event))
422 {
423 	resource_size_t nq_base;
424 	int rc = -1;
425 
426 	if (cqn_handler)
427 		nq->cqn_handler = cqn_handler;
428 
429 	if (srqn_handler)
430 		nq->srqn_handler = srqn_handler;
431 
432 	/* Have a task to schedule CQ notifiers in post send case */
433 	nq->cqn_wq  = create_singlethread_workqueue("bnxt_qplib_nq");
434 	if (!nq->cqn_wq)
435 		return -ENOMEM;
436 
437 	nq->bar_reg = NQ_CONS_PCI_BAR_REGION;
438 	nq->bar_reg_off = bar_reg_offset;
439 	nq_base = pci_resource_start(pdev, nq->bar_reg);
440 	if (!nq_base) {
441 		rc = -ENOMEM;
442 		goto fail;
443 	}
444 	/* Unconditionally map 8 bytes to support 57500 series */
445 	nq->bar_reg_iomem = ioremap(nq_base + nq->bar_reg_off, 8);
446 	if (!nq->bar_reg_iomem) {
447 		rc = -ENOMEM;
448 		goto fail;
449 	}
450 
451 	rc = bnxt_qplib_nq_start_irq(nq, nq_idx, msix_vector, true);
452 	if (rc) {
453 		dev_err(&nq->pdev->dev,
454 			"Failed to request irq for nq-idx %d\n", nq_idx);
455 		goto fail;
456 	}
457 
458 	return 0;
459 fail:
460 	bnxt_qplib_disable_nq(nq);
461 	return rc;
462 }
463 
464 void bnxt_qplib_free_nq(struct bnxt_qplib_nq *nq)
465 {
466 	if (nq->hwq.max_elements) {
467 		bnxt_qplib_free_hwq(nq->pdev, &nq->hwq);
468 		nq->hwq.max_elements = 0;
469 	}
470 }
471 
472 int bnxt_qplib_alloc_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq)
473 {
474 	u8 hwq_type;
475 
476 	nq->pdev = pdev;
477 	if (!nq->hwq.max_elements ||
478 	    nq->hwq.max_elements > BNXT_QPLIB_NQE_MAX_CNT)
479 		nq->hwq.max_elements = BNXT_QPLIB_NQE_MAX_CNT;
480 	hwq_type = bnxt_qplib_get_hwq_type(nq->res);
481 	if (bnxt_qplib_alloc_init_hwq(nq->pdev, &nq->hwq, NULL,
482 				      &nq->hwq.max_elements,
483 				      BNXT_QPLIB_MAX_NQE_ENTRY_SIZE, 0,
484 				      PAGE_SIZE, hwq_type))
485 		return -ENOMEM;
486 
487 	nq->budget = 8;
488 	return 0;
489 }
490 
491 /* SRQ */
492 static void bnxt_qplib_arm_srq(struct bnxt_qplib_srq *srq, u32 arm_type)
493 {
494 	struct bnxt_qplib_hwq *srq_hwq = &srq->hwq;
495 	void __iomem *db;
496 	u32 sw_prod;
497 	u64 val = 0;
498 
499 	/* Ring DB */
500 	sw_prod = (arm_type == DBC_DBC_TYPE_SRQ_ARM) ?
501 		   srq->threshold : HWQ_CMP(srq_hwq->prod, srq_hwq);
502 	db = (arm_type == DBC_DBC_TYPE_SRQ_ARMENA) ? srq->dbr_base :
503 						     srq->dpi->dbr;
504 	val = ((srq->id << DBC_DBC_XID_SFT) & DBC_DBC_XID_MASK) | arm_type;
505 	val <<= 32;
506 	val |= (sw_prod << DBC_DBC_INDEX_SFT) & DBC_DBC_INDEX_MASK;
507 	writeq(val, db);
508 }
509 
510 void bnxt_qplib_destroy_srq(struct bnxt_qplib_res *res,
511 			   struct bnxt_qplib_srq *srq)
512 {
513 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
514 	struct cmdq_destroy_srq req;
515 	struct creq_destroy_srq_resp resp;
516 	u16 cmd_flags = 0;
517 	int rc;
518 
519 	RCFW_CMD_PREP(req, DESTROY_SRQ, cmd_flags);
520 
521 	/* Configure the request */
522 	req.srq_cid = cpu_to_le32(srq->id);
523 
524 	rc = bnxt_qplib_rcfw_send_message(rcfw, (struct cmdq_base *)&req,
525 					  (struct creq_base *)&resp, NULL, 0);
526 	kfree(srq->swq);
527 	if (rc)
528 		return;
529 	bnxt_qplib_free_hwq(res->pdev, &srq->hwq);
530 }
531 
532 int bnxt_qplib_create_srq(struct bnxt_qplib_res *res,
533 			  struct bnxt_qplib_srq *srq)
534 {
535 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
536 	struct cmdq_create_srq req;
537 	struct creq_create_srq_resp resp;
538 	struct bnxt_qplib_pbl *pbl;
539 	u16 cmd_flags = 0;
540 	int rc, idx;
541 
542 	srq->hwq.max_elements = srq->max_wqe;
543 	rc = bnxt_qplib_alloc_init_hwq(res->pdev, &srq->hwq, &srq->sg_info,
544 				       &srq->hwq.max_elements,
545 				       BNXT_QPLIB_MAX_RQE_ENTRY_SIZE, 0,
546 				       PAGE_SIZE, HWQ_TYPE_QUEUE);
547 	if (rc)
548 		goto exit;
549 
550 	srq->swq = kcalloc(srq->hwq.max_elements, sizeof(*srq->swq),
551 			   GFP_KERNEL);
552 	if (!srq->swq) {
553 		rc = -ENOMEM;
554 		goto fail;
555 	}
556 
557 	RCFW_CMD_PREP(req, CREATE_SRQ, cmd_flags);
558 
559 	/* Configure the request */
560 	req.dpi = cpu_to_le32(srq->dpi->dpi);
561 	req.srq_handle = cpu_to_le64((uintptr_t)srq);
562 
563 	req.srq_size = cpu_to_le16((u16)srq->hwq.max_elements);
564 	pbl = &srq->hwq.pbl[PBL_LVL_0];
565 	req.pg_size_lvl = cpu_to_le16((((u16)srq->hwq.level &
566 				      CMDQ_CREATE_SRQ_LVL_MASK) <<
567 				      CMDQ_CREATE_SRQ_LVL_SFT) |
568 				      (pbl->pg_size == ROCE_PG_SIZE_4K ?
569 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_4K :
570 				       pbl->pg_size == ROCE_PG_SIZE_8K ?
571 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_8K :
572 				       pbl->pg_size == ROCE_PG_SIZE_64K ?
573 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_64K :
574 				       pbl->pg_size == ROCE_PG_SIZE_2M ?
575 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_2M :
576 				       pbl->pg_size == ROCE_PG_SIZE_8M ?
577 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_8M :
578 				       pbl->pg_size == ROCE_PG_SIZE_1G ?
579 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_1G :
580 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_4K));
581 	req.pbl = cpu_to_le64(pbl->pg_map_arr[0]);
582 	req.pd_id = cpu_to_le32(srq->pd->id);
583 	req.eventq_id = cpu_to_le16(srq->eventq_hw_ring_id);
584 
585 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
586 					  (void *)&resp, NULL, 0);
587 	if (rc)
588 		goto fail;
589 
590 	spin_lock_init(&srq->lock);
591 	srq->start_idx = 0;
592 	srq->last_idx = srq->hwq.max_elements - 1;
593 	for (idx = 0; idx < srq->hwq.max_elements; idx++)
594 		srq->swq[idx].next_idx = idx + 1;
595 	srq->swq[srq->last_idx].next_idx = -1;
596 
597 	srq->id = le32_to_cpu(resp.xid);
598 	srq->dbr_base = res->dpi_tbl.dbr_bar_reg_iomem;
599 	if (srq->threshold)
600 		bnxt_qplib_arm_srq(srq, DBC_DBC_TYPE_SRQ_ARMENA);
601 	srq->arm_req = false;
602 
603 	return 0;
604 fail:
605 	bnxt_qplib_free_hwq(res->pdev, &srq->hwq);
606 	kfree(srq->swq);
607 exit:
608 	return rc;
609 }
610 
611 int bnxt_qplib_modify_srq(struct bnxt_qplib_res *res,
612 			  struct bnxt_qplib_srq *srq)
613 {
614 	struct bnxt_qplib_hwq *srq_hwq = &srq->hwq;
615 	u32 sw_prod, sw_cons, count = 0;
616 
617 	sw_prod = HWQ_CMP(srq_hwq->prod, srq_hwq);
618 	sw_cons = HWQ_CMP(srq_hwq->cons, srq_hwq);
619 
620 	count = sw_prod > sw_cons ? sw_prod - sw_cons :
621 				    srq_hwq->max_elements - sw_cons + sw_prod;
622 	if (count > srq->threshold) {
623 		srq->arm_req = false;
624 		bnxt_qplib_arm_srq(srq, DBC_DBC_TYPE_SRQ_ARM);
625 	} else {
626 		/* Deferred arming */
627 		srq->arm_req = true;
628 	}
629 
630 	return 0;
631 }
632 
633 int bnxt_qplib_query_srq(struct bnxt_qplib_res *res,
634 			 struct bnxt_qplib_srq *srq)
635 {
636 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
637 	struct cmdq_query_srq req;
638 	struct creq_query_srq_resp resp;
639 	struct bnxt_qplib_rcfw_sbuf *sbuf;
640 	struct creq_query_srq_resp_sb *sb;
641 	u16 cmd_flags = 0;
642 	int rc = 0;
643 
644 	RCFW_CMD_PREP(req, QUERY_SRQ, cmd_flags);
645 	req.srq_cid = cpu_to_le32(srq->id);
646 
647 	/* Configure the request */
648 	sbuf = bnxt_qplib_rcfw_alloc_sbuf(rcfw, sizeof(*sb));
649 	if (!sbuf)
650 		return -ENOMEM;
651 	sb = sbuf->sb;
652 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req, (void *)&resp,
653 					  (void *)sbuf, 0);
654 	srq->threshold = le16_to_cpu(sb->srq_limit);
655 	bnxt_qplib_rcfw_free_sbuf(rcfw, sbuf);
656 
657 	return rc;
658 }
659 
660 int bnxt_qplib_post_srq_recv(struct bnxt_qplib_srq *srq,
661 			     struct bnxt_qplib_swqe *wqe)
662 {
663 	struct bnxt_qplib_hwq *srq_hwq = &srq->hwq;
664 	struct rq_wqe *srqe, **srqe_ptr;
665 	struct sq_sge *hw_sge;
666 	u32 sw_prod, sw_cons, count = 0;
667 	int i, rc = 0, next;
668 
669 	spin_lock(&srq_hwq->lock);
670 	if (srq->start_idx == srq->last_idx) {
671 		dev_err(&srq_hwq->pdev->dev,
672 			"FP: SRQ (0x%x) is full!\n", srq->id);
673 		rc = -EINVAL;
674 		spin_unlock(&srq_hwq->lock);
675 		goto done;
676 	}
677 	next = srq->start_idx;
678 	srq->start_idx = srq->swq[next].next_idx;
679 	spin_unlock(&srq_hwq->lock);
680 
681 	sw_prod = HWQ_CMP(srq_hwq->prod, srq_hwq);
682 	srqe_ptr = (struct rq_wqe **)srq_hwq->pbl_ptr;
683 	srqe = &srqe_ptr[RQE_PG(sw_prod)][RQE_IDX(sw_prod)];
684 	memset(srqe, 0, BNXT_QPLIB_MAX_RQE_ENTRY_SIZE);
685 	/* Calculate wqe_size16 and data_len */
686 	for (i = 0, hw_sge = (struct sq_sge *)srqe->data;
687 	     i < wqe->num_sge; i++, hw_sge++) {
688 		hw_sge->va_or_pa = cpu_to_le64(wqe->sg_list[i].addr);
689 		hw_sge->l_key = cpu_to_le32(wqe->sg_list[i].lkey);
690 		hw_sge->size = cpu_to_le32(wqe->sg_list[i].size);
691 	}
692 	srqe->wqe_type = wqe->type;
693 	srqe->flags = wqe->flags;
694 	srqe->wqe_size = wqe->num_sge +
695 			((offsetof(typeof(*srqe), data) + 15) >> 4);
696 	srqe->wr_id[0] = cpu_to_le32((u32)next);
697 	srq->swq[next].wr_id = wqe->wr_id;
698 
699 	srq_hwq->prod++;
700 
701 	spin_lock(&srq_hwq->lock);
702 	sw_prod = HWQ_CMP(srq_hwq->prod, srq_hwq);
703 	/* retaining srq_hwq->cons for this logic
704 	 * actually the lock is only required to
705 	 * read srq_hwq->cons.
706 	 */
707 	sw_cons = HWQ_CMP(srq_hwq->cons, srq_hwq);
708 	count = sw_prod > sw_cons ? sw_prod - sw_cons :
709 				    srq_hwq->max_elements - sw_cons + sw_prod;
710 	spin_unlock(&srq_hwq->lock);
711 	/* Ring DB */
712 	bnxt_qplib_arm_srq(srq, DBC_DBC_TYPE_SRQ);
713 	if (srq->arm_req == true && count > srq->threshold) {
714 		srq->arm_req = false;
715 		bnxt_qplib_arm_srq(srq, DBC_DBC_TYPE_SRQ_ARM);
716 	}
717 done:
718 	return rc;
719 }
720 
721 /* QP */
722 int bnxt_qplib_create_qp1(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
723 {
724 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
725 	struct cmdq_create_qp1 req;
726 	struct creq_create_qp1_resp resp;
727 	struct bnxt_qplib_pbl *pbl;
728 	struct bnxt_qplib_q *sq = &qp->sq;
729 	struct bnxt_qplib_q *rq = &qp->rq;
730 	int rc;
731 	u16 cmd_flags = 0;
732 	u32 qp_flags = 0;
733 
734 	RCFW_CMD_PREP(req, CREATE_QP1, cmd_flags);
735 
736 	/* General */
737 	req.type = qp->type;
738 	req.dpi = cpu_to_le32(qp->dpi->dpi);
739 	req.qp_handle = cpu_to_le64(qp->qp_handle);
740 
741 	/* SQ */
742 	sq->hwq.max_elements = sq->max_wqe;
743 	rc = bnxt_qplib_alloc_init_hwq(res->pdev, &sq->hwq, NULL,
744 				       &sq->hwq.max_elements,
745 				       BNXT_QPLIB_MAX_SQE_ENTRY_SIZE, 0,
746 				       PAGE_SIZE, HWQ_TYPE_QUEUE);
747 	if (rc)
748 		goto exit;
749 
750 	sq->swq = kcalloc(sq->hwq.max_elements, sizeof(*sq->swq), GFP_KERNEL);
751 	if (!sq->swq) {
752 		rc = -ENOMEM;
753 		goto fail_sq;
754 	}
755 	pbl = &sq->hwq.pbl[PBL_LVL_0];
756 	req.sq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
757 	req.sq_pg_size_sq_lvl =
758 		((sq->hwq.level & CMDQ_CREATE_QP1_SQ_LVL_MASK)
759 				<<  CMDQ_CREATE_QP1_SQ_LVL_SFT) |
760 		(pbl->pg_size == ROCE_PG_SIZE_4K ?
761 				CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_4K :
762 		 pbl->pg_size == ROCE_PG_SIZE_8K ?
763 				CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_8K :
764 		 pbl->pg_size == ROCE_PG_SIZE_64K ?
765 				CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_64K :
766 		 pbl->pg_size == ROCE_PG_SIZE_2M ?
767 				CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_2M :
768 		 pbl->pg_size == ROCE_PG_SIZE_8M ?
769 				CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_8M :
770 		 pbl->pg_size == ROCE_PG_SIZE_1G ?
771 				CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_1G :
772 		 CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_4K);
773 
774 	if (qp->scq)
775 		req.scq_cid = cpu_to_le32(qp->scq->id);
776 
777 	qp_flags |= CMDQ_CREATE_QP1_QP_FLAGS_RESERVED_LKEY_ENABLE;
778 
779 	/* RQ */
780 	if (rq->max_wqe) {
781 		rq->hwq.max_elements = qp->rq.max_wqe;
782 		rc = bnxt_qplib_alloc_init_hwq(res->pdev, &rq->hwq, NULL,
783 					       &rq->hwq.max_elements,
784 					       BNXT_QPLIB_MAX_RQE_ENTRY_SIZE, 0,
785 					       PAGE_SIZE, HWQ_TYPE_QUEUE);
786 		if (rc)
787 			goto fail_sq;
788 
789 		rq->swq = kcalloc(rq->hwq.max_elements, sizeof(*rq->swq),
790 				  GFP_KERNEL);
791 		if (!rq->swq) {
792 			rc = -ENOMEM;
793 			goto fail_rq;
794 		}
795 		pbl = &rq->hwq.pbl[PBL_LVL_0];
796 		req.rq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
797 		req.rq_pg_size_rq_lvl =
798 			((rq->hwq.level & CMDQ_CREATE_QP1_RQ_LVL_MASK) <<
799 			 CMDQ_CREATE_QP1_RQ_LVL_SFT) |
800 				(pbl->pg_size == ROCE_PG_SIZE_4K ?
801 					CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_4K :
802 				 pbl->pg_size == ROCE_PG_SIZE_8K ?
803 					CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_8K :
804 				 pbl->pg_size == ROCE_PG_SIZE_64K ?
805 					CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_64K :
806 				 pbl->pg_size == ROCE_PG_SIZE_2M ?
807 					CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_2M :
808 				 pbl->pg_size == ROCE_PG_SIZE_8M ?
809 					CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_8M :
810 				 pbl->pg_size == ROCE_PG_SIZE_1G ?
811 					CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_1G :
812 				 CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_4K);
813 		if (qp->rcq)
814 			req.rcq_cid = cpu_to_le32(qp->rcq->id);
815 	}
816 
817 	/* Header buffer - allow hdr_buf pass in */
818 	rc = bnxt_qplib_alloc_qp_hdr_buf(res, qp);
819 	if (rc) {
820 		rc = -ENOMEM;
821 		goto fail;
822 	}
823 	req.qp_flags = cpu_to_le32(qp_flags);
824 	req.sq_size = cpu_to_le32(sq->hwq.max_elements);
825 	req.rq_size = cpu_to_le32(rq->hwq.max_elements);
826 
827 	req.sq_fwo_sq_sge =
828 		cpu_to_le16((sq->max_sge & CMDQ_CREATE_QP1_SQ_SGE_MASK) <<
829 			    CMDQ_CREATE_QP1_SQ_SGE_SFT);
830 	req.rq_fwo_rq_sge =
831 		cpu_to_le16((rq->max_sge & CMDQ_CREATE_QP1_RQ_SGE_MASK) <<
832 			    CMDQ_CREATE_QP1_RQ_SGE_SFT);
833 
834 	req.pd_id = cpu_to_le32(qp->pd->id);
835 
836 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
837 					  (void *)&resp, NULL, 0);
838 	if (rc)
839 		goto fail;
840 
841 	qp->id = le32_to_cpu(resp.xid);
842 	qp->cur_qp_state = CMDQ_MODIFY_QP_NEW_STATE_RESET;
843 	rcfw->qp_tbl[qp->id].qp_id = qp->id;
844 	rcfw->qp_tbl[qp->id].qp_handle = (void *)qp;
845 
846 	return 0;
847 
848 fail:
849 	bnxt_qplib_free_qp_hdr_buf(res, qp);
850 fail_rq:
851 	bnxt_qplib_free_hwq(res->pdev, &rq->hwq);
852 	kfree(rq->swq);
853 fail_sq:
854 	bnxt_qplib_free_hwq(res->pdev, &sq->hwq);
855 	kfree(sq->swq);
856 exit:
857 	return rc;
858 }
859 
860 int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
861 {
862 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
863 	unsigned long int psn_search, poff = 0;
864 	struct sq_psn_search **psn_search_ptr;
865 	struct bnxt_qplib_q *sq = &qp->sq;
866 	struct bnxt_qplib_q *rq = &qp->rq;
867 	int i, rc, req_size, psn_sz = 0;
868 	struct sq_send **hw_sq_send_ptr;
869 	struct creq_create_qp_resp resp;
870 	struct bnxt_qplib_hwq *xrrq;
871 	u16 cmd_flags = 0, max_ssge;
872 	struct cmdq_create_qp req;
873 	struct bnxt_qplib_pbl *pbl;
874 	u32 qp_flags = 0;
875 	u16 max_rsge;
876 
877 	RCFW_CMD_PREP(req, CREATE_QP, cmd_flags);
878 
879 	/* General */
880 	req.type = qp->type;
881 	req.dpi = cpu_to_le32(qp->dpi->dpi);
882 	req.qp_handle = cpu_to_le64(qp->qp_handle);
883 
884 	/* SQ */
885 	if (qp->type == CMDQ_CREATE_QP_TYPE_RC) {
886 		psn_sz = bnxt_qplib_is_chip_gen_p5(res->cctx) ?
887 			 sizeof(struct sq_psn_search_ext) :
888 			 sizeof(struct sq_psn_search);
889 	}
890 	sq->hwq.max_elements = sq->max_wqe;
891 	rc = bnxt_qplib_alloc_init_hwq(res->pdev, &sq->hwq, &sq->sg_info,
892 				       &sq->hwq.max_elements,
893 				       BNXT_QPLIB_MAX_SQE_ENTRY_SIZE,
894 				       psn_sz,
895 				       PAGE_SIZE, HWQ_TYPE_QUEUE);
896 	if (rc)
897 		goto exit;
898 
899 	sq->swq = kcalloc(sq->hwq.max_elements, sizeof(*sq->swq), GFP_KERNEL);
900 	if (!sq->swq) {
901 		rc = -ENOMEM;
902 		goto fail_sq;
903 	}
904 	hw_sq_send_ptr = (struct sq_send **)sq->hwq.pbl_ptr;
905 	if (psn_sz) {
906 		psn_search_ptr = (struct sq_psn_search **)
907 				  &hw_sq_send_ptr[get_sqe_pg
908 					(sq->hwq.max_elements)];
909 		psn_search = (unsigned long int)
910 			      &hw_sq_send_ptr[get_sqe_pg(sq->hwq.max_elements)]
911 			      [get_sqe_idx(sq->hwq.max_elements)];
912 		if (psn_search & ~PAGE_MASK) {
913 			/* If the psn_search does not start on a page boundary,
914 			 * then calculate the offset
915 			 */
916 			poff = (psn_search & ~PAGE_MASK) /
917 				BNXT_QPLIB_MAX_PSNE_ENTRY_SIZE;
918 		}
919 		for (i = 0; i < sq->hwq.max_elements; i++) {
920 			sq->swq[i].psn_search =
921 				&psn_search_ptr[get_psne_pg(i + poff)]
922 					       [get_psne_idx(i + poff)];
923 			/*psns_ext will be used only for P5 chips. */
924 			sq->swq[i].psn_ext =
925 				(struct sq_psn_search_ext *)
926 				&psn_search_ptr[get_psne_pg(i + poff)]
927 					       [get_psne_idx(i + poff)];
928 		}
929 	}
930 	pbl = &sq->hwq.pbl[PBL_LVL_0];
931 	req.sq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
932 	req.sq_pg_size_sq_lvl =
933 		((sq->hwq.level & CMDQ_CREATE_QP_SQ_LVL_MASK)
934 				 <<  CMDQ_CREATE_QP_SQ_LVL_SFT) |
935 		(pbl->pg_size == ROCE_PG_SIZE_4K ?
936 				CMDQ_CREATE_QP_SQ_PG_SIZE_PG_4K :
937 		 pbl->pg_size == ROCE_PG_SIZE_8K ?
938 				CMDQ_CREATE_QP_SQ_PG_SIZE_PG_8K :
939 		 pbl->pg_size == ROCE_PG_SIZE_64K ?
940 				CMDQ_CREATE_QP_SQ_PG_SIZE_PG_64K :
941 		 pbl->pg_size == ROCE_PG_SIZE_2M ?
942 				CMDQ_CREATE_QP_SQ_PG_SIZE_PG_2M :
943 		 pbl->pg_size == ROCE_PG_SIZE_8M ?
944 				CMDQ_CREATE_QP_SQ_PG_SIZE_PG_8M :
945 		 pbl->pg_size == ROCE_PG_SIZE_1G ?
946 				CMDQ_CREATE_QP_SQ_PG_SIZE_PG_1G :
947 		 CMDQ_CREATE_QP_SQ_PG_SIZE_PG_4K);
948 
949 	if (qp->scq)
950 		req.scq_cid = cpu_to_le32(qp->scq->id);
951 
952 	qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_RESERVED_LKEY_ENABLE;
953 	qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_FR_PMR_ENABLED;
954 	if (qp->sig_type)
955 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_FORCE_COMPLETION;
956 
957 	/* RQ */
958 	if (rq->max_wqe) {
959 		rq->hwq.max_elements = rq->max_wqe;
960 		rc = bnxt_qplib_alloc_init_hwq(res->pdev, &rq->hwq,
961 					       &rq->sg_info,
962 					       &rq->hwq.max_elements,
963 					       BNXT_QPLIB_MAX_RQE_ENTRY_SIZE, 0,
964 					       PAGE_SIZE, HWQ_TYPE_QUEUE);
965 		if (rc)
966 			goto fail_sq;
967 
968 		rq->swq = kcalloc(rq->hwq.max_elements, sizeof(*rq->swq),
969 				  GFP_KERNEL);
970 		if (!rq->swq) {
971 			rc = -ENOMEM;
972 			goto fail_rq;
973 		}
974 		pbl = &rq->hwq.pbl[PBL_LVL_0];
975 		req.rq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
976 		req.rq_pg_size_rq_lvl =
977 			((rq->hwq.level & CMDQ_CREATE_QP_RQ_LVL_MASK) <<
978 			 CMDQ_CREATE_QP_RQ_LVL_SFT) |
979 				(pbl->pg_size == ROCE_PG_SIZE_4K ?
980 					CMDQ_CREATE_QP_RQ_PG_SIZE_PG_4K :
981 				 pbl->pg_size == ROCE_PG_SIZE_8K ?
982 					CMDQ_CREATE_QP_RQ_PG_SIZE_PG_8K :
983 				 pbl->pg_size == ROCE_PG_SIZE_64K ?
984 					CMDQ_CREATE_QP_RQ_PG_SIZE_PG_64K :
985 				 pbl->pg_size == ROCE_PG_SIZE_2M ?
986 					CMDQ_CREATE_QP_RQ_PG_SIZE_PG_2M :
987 				 pbl->pg_size == ROCE_PG_SIZE_8M ?
988 					CMDQ_CREATE_QP_RQ_PG_SIZE_PG_8M :
989 				 pbl->pg_size == ROCE_PG_SIZE_1G ?
990 					CMDQ_CREATE_QP_RQ_PG_SIZE_PG_1G :
991 				 CMDQ_CREATE_QP_RQ_PG_SIZE_PG_4K);
992 	} else {
993 		/* SRQ */
994 		if (qp->srq) {
995 			qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_SRQ_USED;
996 			req.srq_cid = cpu_to_le32(qp->srq->id);
997 		}
998 	}
999 
1000 	if (qp->rcq)
1001 		req.rcq_cid = cpu_to_le32(qp->rcq->id);
1002 	req.qp_flags = cpu_to_le32(qp_flags);
1003 	req.sq_size = cpu_to_le32(sq->hwq.max_elements);
1004 	req.rq_size = cpu_to_le32(rq->hwq.max_elements);
1005 	qp->sq_hdr_buf = NULL;
1006 	qp->rq_hdr_buf = NULL;
1007 
1008 	rc = bnxt_qplib_alloc_qp_hdr_buf(res, qp);
1009 	if (rc)
1010 		goto fail_rq;
1011 
1012 	/* CTRL-22434: Irrespective of the requested SGE count on the SQ
1013 	 * always create the QP with max send sges possible if the requested
1014 	 * inline size is greater than 0.
1015 	 */
1016 	max_ssge = qp->max_inline_data ? 6 : sq->max_sge;
1017 	req.sq_fwo_sq_sge = cpu_to_le16(
1018 				((max_ssge & CMDQ_CREATE_QP_SQ_SGE_MASK)
1019 				 << CMDQ_CREATE_QP_SQ_SGE_SFT) | 0);
1020 	max_rsge = bnxt_qplib_is_chip_gen_p5(res->cctx) ? 6 : rq->max_sge;
1021 	req.rq_fwo_rq_sge = cpu_to_le16(
1022 				((max_rsge & CMDQ_CREATE_QP_RQ_SGE_MASK)
1023 				 << CMDQ_CREATE_QP_RQ_SGE_SFT) | 0);
1024 	/* ORRQ and IRRQ */
1025 	if (psn_sz) {
1026 		xrrq = &qp->orrq;
1027 		xrrq->max_elements =
1028 			ORD_LIMIT_TO_ORRQ_SLOTS(qp->max_rd_atomic);
1029 		req_size = xrrq->max_elements *
1030 			   BNXT_QPLIB_MAX_ORRQE_ENTRY_SIZE + PAGE_SIZE - 1;
1031 		req_size &= ~(PAGE_SIZE - 1);
1032 		rc = bnxt_qplib_alloc_init_hwq(res->pdev, xrrq, NULL,
1033 					       &xrrq->max_elements,
1034 					       BNXT_QPLIB_MAX_ORRQE_ENTRY_SIZE,
1035 					       0, req_size, HWQ_TYPE_CTX);
1036 		if (rc)
1037 			goto fail_buf_free;
1038 		pbl = &xrrq->pbl[PBL_LVL_0];
1039 		req.orrq_addr = cpu_to_le64(pbl->pg_map_arr[0]);
1040 
1041 		xrrq = &qp->irrq;
1042 		xrrq->max_elements = IRD_LIMIT_TO_IRRQ_SLOTS(
1043 						qp->max_dest_rd_atomic);
1044 		req_size = xrrq->max_elements *
1045 			   BNXT_QPLIB_MAX_IRRQE_ENTRY_SIZE + PAGE_SIZE - 1;
1046 		req_size &= ~(PAGE_SIZE - 1);
1047 
1048 		rc = bnxt_qplib_alloc_init_hwq(res->pdev, xrrq, NULL,
1049 					       &xrrq->max_elements,
1050 					       BNXT_QPLIB_MAX_IRRQE_ENTRY_SIZE,
1051 					       0, req_size, HWQ_TYPE_CTX);
1052 		if (rc)
1053 			goto fail_orrq;
1054 
1055 		pbl = &xrrq->pbl[PBL_LVL_0];
1056 		req.irrq_addr = cpu_to_le64(pbl->pg_map_arr[0]);
1057 	}
1058 	req.pd_id = cpu_to_le32(qp->pd->id);
1059 
1060 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
1061 					  (void *)&resp, NULL, 0);
1062 	if (rc)
1063 		goto fail;
1064 
1065 	qp->id = le32_to_cpu(resp.xid);
1066 	qp->cur_qp_state = CMDQ_MODIFY_QP_NEW_STATE_RESET;
1067 	qp->cctx = res->cctx;
1068 	INIT_LIST_HEAD(&qp->sq_flush);
1069 	INIT_LIST_HEAD(&qp->rq_flush);
1070 	rcfw->qp_tbl[qp->id].qp_id = qp->id;
1071 	rcfw->qp_tbl[qp->id].qp_handle = (void *)qp;
1072 
1073 	return 0;
1074 
1075 fail:
1076 	if (qp->irrq.max_elements)
1077 		bnxt_qplib_free_hwq(res->pdev, &qp->irrq);
1078 fail_orrq:
1079 	if (qp->orrq.max_elements)
1080 		bnxt_qplib_free_hwq(res->pdev, &qp->orrq);
1081 fail_buf_free:
1082 	bnxt_qplib_free_qp_hdr_buf(res, qp);
1083 fail_rq:
1084 	bnxt_qplib_free_hwq(res->pdev, &rq->hwq);
1085 	kfree(rq->swq);
1086 fail_sq:
1087 	bnxt_qplib_free_hwq(res->pdev, &sq->hwq);
1088 	kfree(sq->swq);
1089 exit:
1090 	return rc;
1091 }
1092 
1093 static void __modify_flags_from_init_state(struct bnxt_qplib_qp *qp)
1094 {
1095 	switch (qp->state) {
1096 	case CMDQ_MODIFY_QP_NEW_STATE_RTR:
1097 		/* INIT->RTR, configure the path_mtu to the default
1098 		 * 2048 if not being requested
1099 		 */
1100 		if (!(qp->modify_flags &
1101 		    CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU)) {
1102 			qp->modify_flags |=
1103 				CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU;
1104 			qp->path_mtu =
1105 				CMDQ_MODIFY_QP_PATH_MTU_MTU_2048;
1106 		}
1107 		qp->modify_flags &=
1108 			~CMDQ_MODIFY_QP_MODIFY_MASK_VLAN_ID;
1109 		/* Bono FW require the max_dest_rd_atomic to be >= 1 */
1110 		if (qp->max_dest_rd_atomic < 1)
1111 			qp->max_dest_rd_atomic = 1;
1112 		qp->modify_flags &= ~CMDQ_MODIFY_QP_MODIFY_MASK_SRC_MAC;
1113 		/* Bono FW 20.6.5 requires SGID_INDEX configuration */
1114 		if (!(qp->modify_flags &
1115 		    CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX)) {
1116 			qp->modify_flags |=
1117 				CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX;
1118 			qp->ah.sgid_index = 0;
1119 		}
1120 		break;
1121 	default:
1122 		break;
1123 	}
1124 }
1125 
1126 static void __modify_flags_from_rtr_state(struct bnxt_qplib_qp *qp)
1127 {
1128 	switch (qp->state) {
1129 	case CMDQ_MODIFY_QP_NEW_STATE_RTS:
1130 		/* Bono FW requires the max_rd_atomic to be >= 1 */
1131 		if (qp->max_rd_atomic < 1)
1132 			qp->max_rd_atomic = 1;
1133 		/* Bono FW does not allow PKEY_INDEX,
1134 		 * DGID, FLOW_LABEL, SGID_INDEX, HOP_LIMIT,
1135 		 * TRAFFIC_CLASS, DEST_MAC, PATH_MTU, RQ_PSN,
1136 		 * MIN_RNR_TIMER, MAX_DEST_RD_ATOMIC, DEST_QP_ID
1137 		 * modification
1138 		 */
1139 		qp->modify_flags &=
1140 			~(CMDQ_MODIFY_QP_MODIFY_MASK_PKEY |
1141 			  CMDQ_MODIFY_QP_MODIFY_MASK_DGID |
1142 			  CMDQ_MODIFY_QP_MODIFY_MASK_FLOW_LABEL |
1143 			  CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX |
1144 			  CMDQ_MODIFY_QP_MODIFY_MASK_HOP_LIMIT |
1145 			  CMDQ_MODIFY_QP_MODIFY_MASK_TRAFFIC_CLASS |
1146 			  CMDQ_MODIFY_QP_MODIFY_MASK_DEST_MAC |
1147 			  CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU |
1148 			  CMDQ_MODIFY_QP_MODIFY_MASK_RQ_PSN |
1149 			  CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER |
1150 			  CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC |
1151 			  CMDQ_MODIFY_QP_MODIFY_MASK_DEST_QP_ID);
1152 		break;
1153 	default:
1154 		break;
1155 	}
1156 }
1157 
1158 static void __filter_modify_flags(struct bnxt_qplib_qp *qp)
1159 {
1160 	switch (qp->cur_qp_state) {
1161 	case CMDQ_MODIFY_QP_NEW_STATE_RESET:
1162 		break;
1163 	case CMDQ_MODIFY_QP_NEW_STATE_INIT:
1164 		__modify_flags_from_init_state(qp);
1165 		break;
1166 	case CMDQ_MODIFY_QP_NEW_STATE_RTR:
1167 		__modify_flags_from_rtr_state(qp);
1168 		break;
1169 	case CMDQ_MODIFY_QP_NEW_STATE_RTS:
1170 		break;
1171 	case CMDQ_MODIFY_QP_NEW_STATE_SQD:
1172 		break;
1173 	case CMDQ_MODIFY_QP_NEW_STATE_SQE:
1174 		break;
1175 	case CMDQ_MODIFY_QP_NEW_STATE_ERR:
1176 		break;
1177 	default:
1178 		break;
1179 	}
1180 }
1181 
1182 int bnxt_qplib_modify_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
1183 {
1184 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1185 	struct cmdq_modify_qp req;
1186 	struct creq_modify_qp_resp resp;
1187 	u16 cmd_flags = 0, pkey;
1188 	u32 temp32[4];
1189 	u32 bmask;
1190 	int rc;
1191 
1192 	RCFW_CMD_PREP(req, MODIFY_QP, cmd_flags);
1193 
1194 	/* Filter out the qp_attr_mask based on the state->new transition */
1195 	__filter_modify_flags(qp);
1196 	bmask = qp->modify_flags;
1197 	req.modify_mask = cpu_to_le32(qp->modify_flags);
1198 	req.qp_cid = cpu_to_le32(qp->id);
1199 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_STATE) {
1200 		req.network_type_en_sqd_async_notify_new_state =
1201 				(qp->state & CMDQ_MODIFY_QP_NEW_STATE_MASK) |
1202 				(qp->en_sqd_async_notify ?
1203 					CMDQ_MODIFY_QP_EN_SQD_ASYNC_NOTIFY : 0);
1204 	}
1205 	req.network_type_en_sqd_async_notify_new_state |= qp->nw_type;
1206 
1207 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_ACCESS)
1208 		req.access = qp->access;
1209 
1210 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_PKEY) {
1211 		if (!bnxt_qplib_get_pkey(res, &res->pkey_tbl,
1212 					 qp->pkey_index, &pkey))
1213 			req.pkey = cpu_to_le16(pkey);
1214 	}
1215 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_QKEY)
1216 		req.qkey = cpu_to_le32(qp->qkey);
1217 
1218 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DGID) {
1219 		memcpy(temp32, qp->ah.dgid.data, sizeof(struct bnxt_qplib_gid));
1220 		req.dgid[0] = cpu_to_le32(temp32[0]);
1221 		req.dgid[1] = cpu_to_le32(temp32[1]);
1222 		req.dgid[2] = cpu_to_le32(temp32[2]);
1223 		req.dgid[3] = cpu_to_le32(temp32[3]);
1224 	}
1225 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_FLOW_LABEL)
1226 		req.flow_label = cpu_to_le32(qp->ah.flow_label);
1227 
1228 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX)
1229 		req.sgid_index = cpu_to_le16(res->sgid_tbl.hw_id
1230 					     [qp->ah.sgid_index]);
1231 
1232 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_HOP_LIMIT)
1233 		req.hop_limit = qp->ah.hop_limit;
1234 
1235 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_TRAFFIC_CLASS)
1236 		req.traffic_class = qp->ah.traffic_class;
1237 
1238 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DEST_MAC)
1239 		memcpy(req.dest_mac, qp->ah.dmac, 6);
1240 
1241 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU)
1242 		req.path_mtu = qp->path_mtu;
1243 
1244 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_TIMEOUT)
1245 		req.timeout = qp->timeout;
1246 
1247 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RETRY_CNT)
1248 		req.retry_cnt = qp->retry_cnt;
1249 
1250 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RNR_RETRY)
1251 		req.rnr_retry = qp->rnr_retry;
1252 
1253 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER)
1254 		req.min_rnr_timer = qp->min_rnr_timer;
1255 
1256 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RQ_PSN)
1257 		req.rq_psn = cpu_to_le32(qp->rq.psn);
1258 
1259 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_SQ_PSN)
1260 		req.sq_psn = cpu_to_le32(qp->sq.psn);
1261 
1262 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MAX_RD_ATOMIC)
1263 		req.max_rd_atomic =
1264 			ORD_LIMIT_TO_ORRQ_SLOTS(qp->max_rd_atomic);
1265 
1266 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC)
1267 		req.max_dest_rd_atomic =
1268 			IRD_LIMIT_TO_IRRQ_SLOTS(qp->max_dest_rd_atomic);
1269 
1270 	req.sq_size = cpu_to_le32(qp->sq.hwq.max_elements);
1271 	req.rq_size = cpu_to_le32(qp->rq.hwq.max_elements);
1272 	req.sq_sge = cpu_to_le16(qp->sq.max_sge);
1273 	req.rq_sge = cpu_to_le16(qp->rq.max_sge);
1274 	req.max_inline_data = cpu_to_le32(qp->max_inline_data);
1275 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DEST_QP_ID)
1276 		req.dest_qp_id = cpu_to_le32(qp->dest_qpn);
1277 
1278 	req.vlan_pcp_vlan_dei_vlan_id = cpu_to_le16(qp->vlan_id);
1279 
1280 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
1281 					  (void *)&resp, NULL, 0);
1282 	if (rc)
1283 		return rc;
1284 	qp->cur_qp_state = qp->state;
1285 	return 0;
1286 }
1287 
1288 int bnxt_qplib_query_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
1289 {
1290 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1291 	struct cmdq_query_qp req;
1292 	struct creq_query_qp_resp resp;
1293 	struct bnxt_qplib_rcfw_sbuf *sbuf;
1294 	struct creq_query_qp_resp_sb *sb;
1295 	u16 cmd_flags = 0;
1296 	u32 temp32[4];
1297 	int i, rc = 0;
1298 
1299 	RCFW_CMD_PREP(req, QUERY_QP, cmd_flags);
1300 
1301 	sbuf = bnxt_qplib_rcfw_alloc_sbuf(rcfw, sizeof(*sb));
1302 	if (!sbuf)
1303 		return -ENOMEM;
1304 	sb = sbuf->sb;
1305 
1306 	req.qp_cid = cpu_to_le32(qp->id);
1307 	req.resp_size = sizeof(*sb) / BNXT_QPLIB_CMDQE_UNITS;
1308 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req, (void *)&resp,
1309 					  (void *)sbuf, 0);
1310 	if (rc)
1311 		goto bail;
1312 	/* Extract the context from the side buffer */
1313 	qp->state = sb->en_sqd_async_notify_state &
1314 			CREQ_QUERY_QP_RESP_SB_STATE_MASK;
1315 	qp->en_sqd_async_notify = sb->en_sqd_async_notify_state &
1316 				  CREQ_QUERY_QP_RESP_SB_EN_SQD_ASYNC_NOTIFY ?
1317 				  true : false;
1318 	qp->access = sb->access;
1319 	qp->pkey_index = le16_to_cpu(sb->pkey);
1320 	qp->qkey = le32_to_cpu(sb->qkey);
1321 
1322 	temp32[0] = le32_to_cpu(sb->dgid[0]);
1323 	temp32[1] = le32_to_cpu(sb->dgid[1]);
1324 	temp32[2] = le32_to_cpu(sb->dgid[2]);
1325 	temp32[3] = le32_to_cpu(sb->dgid[3]);
1326 	memcpy(qp->ah.dgid.data, temp32, sizeof(qp->ah.dgid.data));
1327 
1328 	qp->ah.flow_label = le32_to_cpu(sb->flow_label);
1329 
1330 	qp->ah.sgid_index = 0;
1331 	for (i = 0; i < res->sgid_tbl.max; i++) {
1332 		if (res->sgid_tbl.hw_id[i] == le16_to_cpu(sb->sgid_index)) {
1333 			qp->ah.sgid_index = i;
1334 			break;
1335 		}
1336 	}
1337 	if (i == res->sgid_tbl.max)
1338 		dev_warn(&res->pdev->dev, "SGID not found??\n");
1339 
1340 	qp->ah.hop_limit = sb->hop_limit;
1341 	qp->ah.traffic_class = sb->traffic_class;
1342 	memcpy(qp->ah.dmac, sb->dest_mac, 6);
1343 	qp->ah.vlan_id = (le16_to_cpu(sb->path_mtu_dest_vlan_id) &
1344 				CREQ_QUERY_QP_RESP_SB_VLAN_ID_MASK) >>
1345 				CREQ_QUERY_QP_RESP_SB_VLAN_ID_SFT;
1346 	qp->path_mtu = (le16_to_cpu(sb->path_mtu_dest_vlan_id) &
1347 				    CREQ_QUERY_QP_RESP_SB_PATH_MTU_MASK) >>
1348 				    CREQ_QUERY_QP_RESP_SB_PATH_MTU_SFT;
1349 	qp->timeout = sb->timeout;
1350 	qp->retry_cnt = sb->retry_cnt;
1351 	qp->rnr_retry = sb->rnr_retry;
1352 	qp->min_rnr_timer = sb->min_rnr_timer;
1353 	qp->rq.psn = le32_to_cpu(sb->rq_psn);
1354 	qp->max_rd_atomic = ORRQ_SLOTS_TO_ORD_LIMIT(sb->max_rd_atomic);
1355 	qp->sq.psn = le32_to_cpu(sb->sq_psn);
1356 	qp->max_dest_rd_atomic =
1357 			IRRQ_SLOTS_TO_IRD_LIMIT(sb->max_dest_rd_atomic);
1358 	qp->sq.max_wqe = qp->sq.hwq.max_elements;
1359 	qp->rq.max_wqe = qp->rq.hwq.max_elements;
1360 	qp->sq.max_sge = le16_to_cpu(sb->sq_sge);
1361 	qp->rq.max_sge = le16_to_cpu(sb->rq_sge);
1362 	qp->max_inline_data = le32_to_cpu(sb->max_inline_data);
1363 	qp->dest_qpn = le32_to_cpu(sb->dest_qp_id);
1364 	memcpy(qp->smac, sb->src_mac, 6);
1365 	qp->vlan_id = le16_to_cpu(sb->vlan_pcp_vlan_dei_vlan_id);
1366 bail:
1367 	bnxt_qplib_rcfw_free_sbuf(rcfw, sbuf);
1368 	return rc;
1369 }
1370 
1371 static void __clean_cq(struct bnxt_qplib_cq *cq, u64 qp)
1372 {
1373 	struct bnxt_qplib_hwq *cq_hwq = &cq->hwq;
1374 	struct cq_base *hw_cqe, **hw_cqe_ptr;
1375 	int i;
1376 
1377 	for (i = 0; i < cq_hwq->max_elements; i++) {
1378 		hw_cqe_ptr = (struct cq_base **)cq_hwq->pbl_ptr;
1379 		hw_cqe = &hw_cqe_ptr[CQE_PG(i)][CQE_IDX(i)];
1380 		if (!CQE_CMP_VALID(hw_cqe, i, cq_hwq->max_elements))
1381 			continue;
1382 		/*
1383 		 * The valid test of the entry must be done first before
1384 		 * reading any further.
1385 		 */
1386 		dma_rmb();
1387 		switch (hw_cqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK) {
1388 		case CQ_BASE_CQE_TYPE_REQ:
1389 		case CQ_BASE_CQE_TYPE_TERMINAL:
1390 		{
1391 			struct cq_req *cqe = (struct cq_req *)hw_cqe;
1392 
1393 			if (qp == le64_to_cpu(cqe->qp_handle))
1394 				cqe->qp_handle = 0;
1395 			break;
1396 		}
1397 		case CQ_BASE_CQE_TYPE_RES_RC:
1398 		case CQ_BASE_CQE_TYPE_RES_UD:
1399 		case CQ_BASE_CQE_TYPE_RES_RAWETH_QP1:
1400 		{
1401 			struct cq_res_rc *cqe = (struct cq_res_rc *)hw_cqe;
1402 
1403 			if (qp == le64_to_cpu(cqe->qp_handle))
1404 				cqe->qp_handle = 0;
1405 			break;
1406 		}
1407 		default:
1408 			break;
1409 		}
1410 	}
1411 }
1412 
1413 int bnxt_qplib_destroy_qp(struct bnxt_qplib_res *res,
1414 			  struct bnxt_qplib_qp *qp)
1415 {
1416 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1417 	struct cmdq_destroy_qp req;
1418 	struct creq_destroy_qp_resp resp;
1419 	u16 cmd_flags = 0;
1420 	int rc;
1421 
1422 	rcfw->qp_tbl[qp->id].qp_id = BNXT_QPLIB_QP_ID_INVALID;
1423 	rcfw->qp_tbl[qp->id].qp_handle = NULL;
1424 
1425 	RCFW_CMD_PREP(req, DESTROY_QP, cmd_flags);
1426 
1427 	req.qp_cid = cpu_to_le32(qp->id);
1428 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
1429 					  (void *)&resp, NULL, 0);
1430 	if (rc) {
1431 		rcfw->qp_tbl[qp->id].qp_id = qp->id;
1432 		rcfw->qp_tbl[qp->id].qp_handle = qp;
1433 		return rc;
1434 	}
1435 
1436 	return 0;
1437 }
1438 
1439 void bnxt_qplib_free_qp_res(struct bnxt_qplib_res *res,
1440 			    struct bnxt_qplib_qp *qp)
1441 {
1442 	bnxt_qplib_free_qp_hdr_buf(res, qp);
1443 	bnxt_qplib_free_hwq(res->pdev, &qp->sq.hwq);
1444 	kfree(qp->sq.swq);
1445 
1446 	bnxt_qplib_free_hwq(res->pdev, &qp->rq.hwq);
1447 	kfree(qp->rq.swq);
1448 
1449 	if (qp->irrq.max_elements)
1450 		bnxt_qplib_free_hwq(res->pdev, &qp->irrq);
1451 	if (qp->orrq.max_elements)
1452 		bnxt_qplib_free_hwq(res->pdev, &qp->orrq);
1453 
1454 }
1455 
1456 void *bnxt_qplib_get_qp1_sq_buf(struct bnxt_qplib_qp *qp,
1457 				struct bnxt_qplib_sge *sge)
1458 {
1459 	struct bnxt_qplib_q *sq = &qp->sq;
1460 	u32 sw_prod;
1461 
1462 	memset(sge, 0, sizeof(*sge));
1463 
1464 	if (qp->sq_hdr_buf) {
1465 		sw_prod = HWQ_CMP(sq->hwq.prod, &sq->hwq);
1466 		sge->addr = (dma_addr_t)(qp->sq_hdr_buf_map +
1467 					 sw_prod * qp->sq_hdr_buf_size);
1468 		sge->lkey = 0xFFFFFFFF;
1469 		sge->size = qp->sq_hdr_buf_size;
1470 		return qp->sq_hdr_buf + sw_prod * sge->size;
1471 	}
1472 	return NULL;
1473 }
1474 
1475 u32 bnxt_qplib_get_rq_prod_index(struct bnxt_qplib_qp *qp)
1476 {
1477 	struct bnxt_qplib_q *rq = &qp->rq;
1478 
1479 	return HWQ_CMP(rq->hwq.prod, &rq->hwq);
1480 }
1481 
1482 dma_addr_t bnxt_qplib_get_qp_buf_from_index(struct bnxt_qplib_qp *qp, u32 index)
1483 {
1484 	return (qp->rq_hdr_buf_map + index * qp->rq_hdr_buf_size);
1485 }
1486 
1487 void *bnxt_qplib_get_qp1_rq_buf(struct bnxt_qplib_qp *qp,
1488 				struct bnxt_qplib_sge *sge)
1489 {
1490 	struct bnxt_qplib_q *rq = &qp->rq;
1491 	u32 sw_prod;
1492 
1493 	memset(sge, 0, sizeof(*sge));
1494 
1495 	if (qp->rq_hdr_buf) {
1496 		sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq);
1497 		sge->addr = (dma_addr_t)(qp->rq_hdr_buf_map +
1498 					 sw_prod * qp->rq_hdr_buf_size);
1499 		sge->lkey = 0xFFFFFFFF;
1500 		sge->size = qp->rq_hdr_buf_size;
1501 		return qp->rq_hdr_buf + sw_prod * sge->size;
1502 	}
1503 	return NULL;
1504 }
1505 
1506 void bnxt_qplib_post_send_db(struct bnxt_qplib_qp *qp)
1507 {
1508 	struct bnxt_qplib_q *sq = &qp->sq;
1509 	u32 sw_prod;
1510 	u64 val = 0;
1511 
1512 	val = (((qp->id << DBC_DBC_XID_SFT) & DBC_DBC_XID_MASK) |
1513 	       DBC_DBC_TYPE_SQ);
1514 	val <<= 32;
1515 	sw_prod = HWQ_CMP(sq->hwq.prod, &sq->hwq);
1516 	val |= (sw_prod << DBC_DBC_INDEX_SFT) & DBC_DBC_INDEX_MASK;
1517 	/* Flush all the WQE writes to HW */
1518 	writeq(val, qp->dpi->dbr);
1519 }
1520 
1521 int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp,
1522 			 struct bnxt_qplib_swqe *wqe)
1523 {
1524 	struct bnxt_qplib_q *sq = &qp->sq;
1525 	struct bnxt_qplib_swq *swq;
1526 	struct sq_send *hw_sq_send_hdr, **hw_sq_send_ptr;
1527 	struct sq_sge *hw_sge;
1528 	struct bnxt_qplib_nq_work *nq_work = NULL;
1529 	bool sch_handler = false;
1530 	u32 sw_prod;
1531 	u8 wqe_size16;
1532 	int i, rc = 0, data_len = 0, pkt_num = 0;
1533 	__le32 temp32;
1534 
1535 	if (qp->state != CMDQ_MODIFY_QP_NEW_STATE_RTS) {
1536 		if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1537 			sch_handler = true;
1538 			dev_dbg(&sq->hwq.pdev->dev,
1539 				"%s Error QP. Scheduling for poll_cq\n",
1540 				__func__);
1541 			goto queue_err;
1542 		}
1543 	}
1544 
1545 	if (bnxt_qplib_queue_full(sq)) {
1546 		dev_err(&sq->hwq.pdev->dev,
1547 			"prod = %#x cons = %#x qdepth = %#x delta = %#x\n",
1548 			sq->hwq.prod, sq->hwq.cons, sq->hwq.max_elements,
1549 			sq->q_full_delta);
1550 		rc = -ENOMEM;
1551 		goto done;
1552 	}
1553 	sw_prod = HWQ_CMP(sq->hwq.prod, &sq->hwq);
1554 	swq = &sq->swq[sw_prod];
1555 	swq->wr_id = wqe->wr_id;
1556 	swq->type = wqe->type;
1557 	swq->flags = wqe->flags;
1558 	if (qp->sig_type)
1559 		swq->flags |= SQ_SEND_FLAGS_SIGNAL_COMP;
1560 	swq->start_psn = sq->psn & BTH_PSN_MASK;
1561 
1562 	hw_sq_send_ptr = (struct sq_send **)sq->hwq.pbl_ptr;
1563 	hw_sq_send_hdr = &hw_sq_send_ptr[get_sqe_pg(sw_prod)]
1564 					[get_sqe_idx(sw_prod)];
1565 
1566 	memset(hw_sq_send_hdr, 0, BNXT_QPLIB_MAX_SQE_ENTRY_SIZE);
1567 
1568 	if (wqe->flags & BNXT_QPLIB_SWQE_FLAGS_INLINE) {
1569 		/* Copy the inline data */
1570 		if (wqe->inline_len > BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH) {
1571 			dev_warn(&sq->hwq.pdev->dev,
1572 				 "Inline data length > 96 detected\n");
1573 			data_len = BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH;
1574 		} else {
1575 			data_len = wqe->inline_len;
1576 		}
1577 		memcpy(hw_sq_send_hdr->data, wqe->inline_data, data_len);
1578 		wqe_size16 = (data_len + 15) >> 4;
1579 	} else {
1580 		for (i = 0, hw_sge = (struct sq_sge *)hw_sq_send_hdr->data;
1581 		     i < wqe->num_sge; i++, hw_sge++) {
1582 			hw_sge->va_or_pa = cpu_to_le64(wqe->sg_list[i].addr);
1583 			hw_sge->l_key = cpu_to_le32(wqe->sg_list[i].lkey);
1584 			hw_sge->size = cpu_to_le32(wqe->sg_list[i].size);
1585 			data_len += wqe->sg_list[i].size;
1586 		}
1587 		/* Each SGE entry = 1 WQE size16 */
1588 		wqe_size16 = wqe->num_sge;
1589 		/* HW requires wqe size has room for atleast one SGE even if
1590 		 * none was supplied by ULP
1591 		 */
1592 		if (!wqe->num_sge)
1593 			wqe_size16++;
1594 	}
1595 
1596 	/* Specifics */
1597 	switch (wqe->type) {
1598 	case BNXT_QPLIB_SWQE_TYPE_SEND:
1599 		if (qp->type == CMDQ_CREATE_QP1_TYPE_GSI) {
1600 			/* Assemble info for Raw Ethertype QPs */
1601 			struct sq_send_raweth_qp1 *sqe =
1602 				(struct sq_send_raweth_qp1 *)hw_sq_send_hdr;
1603 
1604 			sqe->wqe_type = wqe->type;
1605 			sqe->flags = wqe->flags;
1606 			sqe->wqe_size = wqe_size16 +
1607 				((offsetof(typeof(*sqe), data) + 15) >> 4);
1608 			sqe->cfa_action = cpu_to_le16(wqe->rawqp1.cfa_action);
1609 			sqe->lflags = cpu_to_le16(wqe->rawqp1.lflags);
1610 			sqe->length = cpu_to_le32(data_len);
1611 			sqe->cfa_meta = cpu_to_le32((wqe->rawqp1.cfa_meta &
1612 				SQ_SEND_RAWETH_QP1_CFA_META_VLAN_VID_MASK) <<
1613 				SQ_SEND_RAWETH_QP1_CFA_META_VLAN_VID_SFT);
1614 
1615 			break;
1616 		}
1617 		/* fall thru */
1618 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_IMM:
1619 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_INV:
1620 	{
1621 		struct sq_send *sqe = (struct sq_send *)hw_sq_send_hdr;
1622 
1623 		sqe->wqe_type = wqe->type;
1624 		sqe->flags = wqe->flags;
1625 		sqe->wqe_size = wqe_size16 +
1626 				((offsetof(typeof(*sqe), data) + 15) >> 4);
1627 		sqe->inv_key_or_imm_data = cpu_to_le32(
1628 						wqe->send.inv_key);
1629 		if (qp->type == CMDQ_CREATE_QP_TYPE_UD ||
1630 		    qp->type == CMDQ_CREATE_QP_TYPE_GSI) {
1631 			sqe->q_key = cpu_to_le32(wqe->send.q_key);
1632 			sqe->dst_qp = cpu_to_le32(
1633 					wqe->send.dst_qp & SQ_SEND_DST_QP_MASK);
1634 			sqe->length = cpu_to_le32(data_len);
1635 			sqe->avid = cpu_to_le32(wqe->send.avid &
1636 						SQ_SEND_AVID_MASK);
1637 			sq->psn = (sq->psn + 1) & BTH_PSN_MASK;
1638 		} else {
1639 			sqe->length = cpu_to_le32(data_len);
1640 			sqe->dst_qp = 0;
1641 			sqe->avid = 0;
1642 			if (qp->mtu)
1643 				pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
1644 			if (!pkt_num)
1645 				pkt_num = 1;
1646 			sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
1647 		}
1648 		break;
1649 	}
1650 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE:
1651 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE_WITH_IMM:
1652 	case BNXT_QPLIB_SWQE_TYPE_RDMA_READ:
1653 	{
1654 		struct sq_rdma *sqe = (struct sq_rdma *)hw_sq_send_hdr;
1655 
1656 		sqe->wqe_type = wqe->type;
1657 		sqe->flags = wqe->flags;
1658 		sqe->wqe_size = wqe_size16 +
1659 				((offsetof(typeof(*sqe), data) + 15) >> 4);
1660 		sqe->imm_data = cpu_to_le32(wqe->rdma.inv_key);
1661 		sqe->length = cpu_to_le32((u32)data_len);
1662 		sqe->remote_va = cpu_to_le64(wqe->rdma.remote_va);
1663 		sqe->remote_key = cpu_to_le32(wqe->rdma.r_key);
1664 		if (qp->mtu)
1665 			pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
1666 		if (!pkt_num)
1667 			pkt_num = 1;
1668 		sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
1669 		break;
1670 	}
1671 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_CMP_AND_SWP:
1672 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_FETCH_AND_ADD:
1673 	{
1674 		struct sq_atomic *sqe = (struct sq_atomic *)hw_sq_send_hdr;
1675 
1676 		sqe->wqe_type = wqe->type;
1677 		sqe->flags = wqe->flags;
1678 		sqe->remote_key = cpu_to_le32(wqe->atomic.r_key);
1679 		sqe->remote_va = cpu_to_le64(wqe->atomic.remote_va);
1680 		sqe->swap_data = cpu_to_le64(wqe->atomic.swap_data);
1681 		sqe->cmp_data = cpu_to_le64(wqe->atomic.cmp_data);
1682 		if (qp->mtu)
1683 			pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
1684 		if (!pkt_num)
1685 			pkt_num = 1;
1686 		sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
1687 		break;
1688 	}
1689 	case BNXT_QPLIB_SWQE_TYPE_LOCAL_INV:
1690 	{
1691 		struct sq_localinvalidate *sqe =
1692 				(struct sq_localinvalidate *)hw_sq_send_hdr;
1693 
1694 		sqe->wqe_type = wqe->type;
1695 		sqe->flags = wqe->flags;
1696 		sqe->inv_l_key = cpu_to_le32(wqe->local_inv.inv_l_key);
1697 
1698 		break;
1699 	}
1700 	case BNXT_QPLIB_SWQE_TYPE_FAST_REG_MR:
1701 	{
1702 		struct sq_fr_pmr *sqe = (struct sq_fr_pmr *)hw_sq_send_hdr;
1703 
1704 		sqe->wqe_type = wqe->type;
1705 		sqe->flags = wqe->flags;
1706 		sqe->access_cntl = wqe->frmr.access_cntl |
1707 				   SQ_FR_PMR_ACCESS_CNTL_LOCAL_WRITE;
1708 		sqe->zero_based_page_size_log =
1709 			(wqe->frmr.pg_sz_log & SQ_FR_PMR_PAGE_SIZE_LOG_MASK) <<
1710 			SQ_FR_PMR_PAGE_SIZE_LOG_SFT |
1711 			(wqe->frmr.zero_based ? SQ_FR_PMR_ZERO_BASED : 0);
1712 		sqe->l_key = cpu_to_le32(wqe->frmr.l_key);
1713 		temp32 = cpu_to_le32(wqe->frmr.length);
1714 		memcpy(sqe->length, &temp32, sizeof(wqe->frmr.length));
1715 		sqe->numlevels_pbl_page_size_log =
1716 			((wqe->frmr.pbl_pg_sz_log <<
1717 					SQ_FR_PMR_PBL_PAGE_SIZE_LOG_SFT) &
1718 					SQ_FR_PMR_PBL_PAGE_SIZE_LOG_MASK) |
1719 			((wqe->frmr.levels << SQ_FR_PMR_NUMLEVELS_SFT) &
1720 					SQ_FR_PMR_NUMLEVELS_MASK);
1721 
1722 		for (i = 0; i < wqe->frmr.page_list_len; i++)
1723 			wqe->frmr.pbl_ptr[i] = cpu_to_le64(
1724 						wqe->frmr.page_list[i] |
1725 						PTU_PTE_VALID);
1726 		sqe->pblptr = cpu_to_le64(wqe->frmr.pbl_dma_ptr);
1727 		sqe->va = cpu_to_le64(wqe->frmr.va);
1728 
1729 		break;
1730 	}
1731 	case BNXT_QPLIB_SWQE_TYPE_BIND_MW:
1732 	{
1733 		struct sq_bind *sqe = (struct sq_bind *)hw_sq_send_hdr;
1734 
1735 		sqe->wqe_type = wqe->type;
1736 		sqe->flags = wqe->flags;
1737 		sqe->access_cntl = wqe->bind.access_cntl;
1738 		sqe->mw_type_zero_based = wqe->bind.mw_type |
1739 			(wqe->bind.zero_based ? SQ_BIND_ZERO_BASED : 0);
1740 		sqe->parent_l_key = cpu_to_le32(wqe->bind.parent_l_key);
1741 		sqe->l_key = cpu_to_le32(wqe->bind.r_key);
1742 		sqe->va = cpu_to_le64(wqe->bind.va);
1743 		temp32 = cpu_to_le32(wqe->bind.length);
1744 		memcpy(&sqe->length, &temp32, sizeof(wqe->bind.length));
1745 		break;
1746 	}
1747 	default:
1748 		/* Bad wqe, return error */
1749 		rc = -EINVAL;
1750 		goto done;
1751 	}
1752 	swq->next_psn = sq->psn & BTH_PSN_MASK;
1753 	if (swq->psn_search) {
1754 		u32 opcd_spsn;
1755 		u32 flg_npsn;
1756 
1757 		opcd_spsn = ((swq->start_psn << SQ_PSN_SEARCH_START_PSN_SFT) &
1758 			      SQ_PSN_SEARCH_START_PSN_MASK);
1759 		opcd_spsn |= ((wqe->type << SQ_PSN_SEARCH_OPCODE_SFT) &
1760 			       SQ_PSN_SEARCH_OPCODE_MASK);
1761 		flg_npsn = ((swq->next_psn << SQ_PSN_SEARCH_NEXT_PSN_SFT) &
1762 			     SQ_PSN_SEARCH_NEXT_PSN_MASK);
1763 		if (bnxt_qplib_is_chip_gen_p5(qp->cctx)) {
1764 			swq->psn_ext->opcode_start_psn =
1765 						cpu_to_le32(opcd_spsn);
1766 			swq->psn_ext->flags_next_psn =
1767 						cpu_to_le32(flg_npsn);
1768 		} else {
1769 			swq->psn_search->opcode_start_psn =
1770 						cpu_to_le32(opcd_spsn);
1771 			swq->psn_search->flags_next_psn =
1772 						cpu_to_le32(flg_npsn);
1773 		}
1774 	}
1775 queue_err:
1776 	if (sch_handler) {
1777 		/* Store the ULP info in the software structures */
1778 		sw_prod = HWQ_CMP(sq->hwq.prod, &sq->hwq);
1779 		swq = &sq->swq[sw_prod];
1780 		swq->wr_id = wqe->wr_id;
1781 		swq->type = wqe->type;
1782 		swq->flags = wqe->flags;
1783 		if (qp->sig_type)
1784 			swq->flags |= SQ_SEND_FLAGS_SIGNAL_COMP;
1785 		swq->start_psn = sq->psn & BTH_PSN_MASK;
1786 	}
1787 	sq->hwq.prod++;
1788 	qp->wqe_cnt++;
1789 
1790 done:
1791 	if (sch_handler) {
1792 		nq_work = kzalloc(sizeof(*nq_work), GFP_ATOMIC);
1793 		if (nq_work) {
1794 			nq_work->cq = qp->scq;
1795 			nq_work->nq = qp->scq->nq;
1796 			INIT_WORK(&nq_work->work, bnxt_qpn_cqn_sched_task);
1797 			queue_work(qp->scq->nq->cqn_wq, &nq_work->work);
1798 		} else {
1799 			dev_err(&sq->hwq.pdev->dev,
1800 				"FP: Failed to allocate SQ nq_work!\n");
1801 			rc = -ENOMEM;
1802 		}
1803 	}
1804 	return rc;
1805 }
1806 
1807 void bnxt_qplib_post_recv_db(struct bnxt_qplib_qp *qp)
1808 {
1809 	struct bnxt_qplib_q *rq = &qp->rq;
1810 	u32 sw_prod;
1811 	u64 val = 0;
1812 
1813 	val = (((qp->id << DBC_DBC_XID_SFT) & DBC_DBC_XID_MASK) |
1814 	       DBC_DBC_TYPE_RQ);
1815 	val <<= 32;
1816 	sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq);
1817 	val |= (sw_prod << DBC_DBC_INDEX_SFT) & DBC_DBC_INDEX_MASK;
1818 	/* Flush the writes to HW Rx WQE before the ringing Rx DB */
1819 	writeq(val, qp->dpi->dbr);
1820 }
1821 
1822 int bnxt_qplib_post_recv(struct bnxt_qplib_qp *qp,
1823 			 struct bnxt_qplib_swqe *wqe)
1824 {
1825 	struct bnxt_qplib_q *rq = &qp->rq;
1826 	struct rq_wqe *rqe, **rqe_ptr;
1827 	struct sq_sge *hw_sge;
1828 	struct bnxt_qplib_nq_work *nq_work = NULL;
1829 	bool sch_handler = false;
1830 	u32 sw_prod;
1831 	int i, rc = 0;
1832 
1833 	if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1834 		sch_handler = true;
1835 		dev_dbg(&rq->hwq.pdev->dev,
1836 			"%s: Error QP. Scheduling for poll_cq\n", __func__);
1837 		goto queue_err;
1838 	}
1839 	if (bnxt_qplib_queue_full(rq)) {
1840 		dev_err(&rq->hwq.pdev->dev,
1841 			"FP: QP (0x%x) RQ is full!\n", qp->id);
1842 		rc = -EINVAL;
1843 		goto done;
1844 	}
1845 	sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq);
1846 	rq->swq[sw_prod].wr_id = wqe->wr_id;
1847 
1848 	rqe_ptr = (struct rq_wqe **)rq->hwq.pbl_ptr;
1849 	rqe = &rqe_ptr[RQE_PG(sw_prod)][RQE_IDX(sw_prod)];
1850 
1851 	memset(rqe, 0, BNXT_QPLIB_MAX_RQE_ENTRY_SIZE);
1852 
1853 	/* Calculate wqe_size16 and data_len */
1854 	for (i = 0, hw_sge = (struct sq_sge *)rqe->data;
1855 	     i < wqe->num_sge; i++, hw_sge++) {
1856 		hw_sge->va_or_pa = cpu_to_le64(wqe->sg_list[i].addr);
1857 		hw_sge->l_key = cpu_to_le32(wqe->sg_list[i].lkey);
1858 		hw_sge->size = cpu_to_le32(wqe->sg_list[i].size);
1859 	}
1860 	rqe->wqe_type = wqe->type;
1861 	rqe->flags = wqe->flags;
1862 	rqe->wqe_size = wqe->num_sge +
1863 			((offsetof(typeof(*rqe), data) + 15) >> 4);
1864 	/* HW requires wqe size has room for atleast one SGE even if none
1865 	 * was supplied by ULP
1866 	 */
1867 	if (!wqe->num_sge)
1868 		rqe->wqe_size++;
1869 
1870 	/* Supply the rqe->wr_id index to the wr_id_tbl for now */
1871 	rqe->wr_id[0] = cpu_to_le32(sw_prod);
1872 
1873 queue_err:
1874 	if (sch_handler) {
1875 		/* Store the ULP info in the software structures */
1876 		sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq);
1877 		rq->swq[sw_prod].wr_id = wqe->wr_id;
1878 	}
1879 
1880 	rq->hwq.prod++;
1881 	if (sch_handler) {
1882 		nq_work = kzalloc(sizeof(*nq_work), GFP_ATOMIC);
1883 		if (nq_work) {
1884 			nq_work->cq = qp->rcq;
1885 			nq_work->nq = qp->rcq->nq;
1886 			INIT_WORK(&nq_work->work, bnxt_qpn_cqn_sched_task);
1887 			queue_work(qp->rcq->nq->cqn_wq, &nq_work->work);
1888 		} else {
1889 			dev_err(&rq->hwq.pdev->dev,
1890 				"FP: Failed to allocate RQ nq_work!\n");
1891 			rc = -ENOMEM;
1892 		}
1893 	}
1894 done:
1895 	return rc;
1896 }
1897 
1898 /* CQ */
1899 
1900 /* Spinlock must be held */
1901 static void bnxt_qplib_arm_cq_enable(struct bnxt_qplib_cq *cq)
1902 {
1903 	u64 val = 0;
1904 
1905 	val = ((cq->id << DBC_DBC_XID_SFT) & DBC_DBC_XID_MASK) |
1906 	       DBC_DBC_TYPE_CQ_ARMENA;
1907 	val <<= 32;
1908 	/* Flush memory writes before enabling the CQ */
1909 	writeq(val, cq->dbr_base);
1910 }
1911 
1912 static void bnxt_qplib_arm_cq(struct bnxt_qplib_cq *cq, u32 arm_type)
1913 {
1914 	struct bnxt_qplib_hwq *cq_hwq = &cq->hwq;
1915 	u32 sw_cons;
1916 	u64 val = 0;
1917 
1918 	/* Ring DB */
1919 	val = ((cq->id << DBC_DBC_XID_SFT) & DBC_DBC_XID_MASK) | arm_type;
1920 	val <<= 32;
1921 	sw_cons = HWQ_CMP(cq_hwq->cons, cq_hwq);
1922 	val |= (sw_cons << DBC_DBC_INDEX_SFT) & DBC_DBC_INDEX_MASK;
1923 	/* flush memory writes before arming the CQ */
1924 	writeq(val, cq->dpi->dbr);
1925 }
1926 
1927 int bnxt_qplib_create_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq)
1928 {
1929 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1930 	struct cmdq_create_cq req;
1931 	struct creq_create_cq_resp resp;
1932 	struct bnxt_qplib_pbl *pbl;
1933 	u16 cmd_flags = 0;
1934 	int rc;
1935 
1936 	cq->hwq.max_elements = cq->max_wqe;
1937 	rc = bnxt_qplib_alloc_init_hwq(res->pdev, &cq->hwq, &cq->sg_info,
1938 				       &cq->hwq.max_elements,
1939 				       BNXT_QPLIB_MAX_CQE_ENTRY_SIZE, 0,
1940 				       PAGE_SIZE, HWQ_TYPE_QUEUE);
1941 	if (rc)
1942 		goto exit;
1943 
1944 	RCFW_CMD_PREP(req, CREATE_CQ, cmd_flags);
1945 
1946 	if (!cq->dpi) {
1947 		dev_err(&rcfw->pdev->dev,
1948 			"FP: CREATE_CQ failed due to NULL DPI\n");
1949 		return -EINVAL;
1950 	}
1951 	req.dpi = cpu_to_le32(cq->dpi->dpi);
1952 	req.cq_handle = cpu_to_le64(cq->cq_handle);
1953 
1954 	req.cq_size = cpu_to_le32(cq->hwq.max_elements);
1955 	pbl = &cq->hwq.pbl[PBL_LVL_0];
1956 	req.pg_size_lvl = cpu_to_le32(
1957 	    ((cq->hwq.level & CMDQ_CREATE_CQ_LVL_MASK) <<
1958 						CMDQ_CREATE_CQ_LVL_SFT) |
1959 	    (pbl->pg_size == ROCE_PG_SIZE_4K ? CMDQ_CREATE_CQ_PG_SIZE_PG_4K :
1960 	     pbl->pg_size == ROCE_PG_SIZE_8K ? CMDQ_CREATE_CQ_PG_SIZE_PG_8K :
1961 	     pbl->pg_size == ROCE_PG_SIZE_64K ? CMDQ_CREATE_CQ_PG_SIZE_PG_64K :
1962 	     pbl->pg_size == ROCE_PG_SIZE_2M ? CMDQ_CREATE_CQ_PG_SIZE_PG_2M :
1963 	     pbl->pg_size == ROCE_PG_SIZE_8M ? CMDQ_CREATE_CQ_PG_SIZE_PG_8M :
1964 	     pbl->pg_size == ROCE_PG_SIZE_1G ? CMDQ_CREATE_CQ_PG_SIZE_PG_1G :
1965 	     CMDQ_CREATE_CQ_PG_SIZE_PG_4K));
1966 
1967 	req.pbl = cpu_to_le64(pbl->pg_map_arr[0]);
1968 
1969 	req.cq_fco_cnq_id = cpu_to_le32(
1970 			(cq->cnq_hw_ring_id & CMDQ_CREATE_CQ_CNQ_ID_MASK) <<
1971 			 CMDQ_CREATE_CQ_CNQ_ID_SFT);
1972 
1973 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
1974 					  (void *)&resp, NULL, 0);
1975 	if (rc)
1976 		goto fail;
1977 
1978 	cq->id = le32_to_cpu(resp.xid);
1979 	cq->dbr_base = res->dpi_tbl.dbr_bar_reg_iomem;
1980 	cq->period = BNXT_QPLIB_QUEUE_START_PERIOD;
1981 	init_waitqueue_head(&cq->waitq);
1982 	INIT_LIST_HEAD(&cq->sqf_head);
1983 	INIT_LIST_HEAD(&cq->rqf_head);
1984 	spin_lock_init(&cq->compl_lock);
1985 	spin_lock_init(&cq->flush_lock);
1986 
1987 	bnxt_qplib_arm_cq_enable(cq);
1988 	return 0;
1989 
1990 fail:
1991 	bnxt_qplib_free_hwq(res->pdev, &cq->hwq);
1992 exit:
1993 	return rc;
1994 }
1995 
1996 int bnxt_qplib_destroy_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq)
1997 {
1998 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1999 	struct cmdq_destroy_cq req;
2000 	struct creq_destroy_cq_resp resp;
2001 	u16 cmd_flags = 0;
2002 	int rc;
2003 
2004 	RCFW_CMD_PREP(req, DESTROY_CQ, cmd_flags);
2005 
2006 	req.cq_cid = cpu_to_le32(cq->id);
2007 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
2008 					  (void *)&resp, NULL, 0);
2009 	if (rc)
2010 		return rc;
2011 	bnxt_qplib_free_hwq(res->pdev, &cq->hwq);
2012 	return 0;
2013 }
2014 
2015 static int __flush_sq(struct bnxt_qplib_q *sq, struct bnxt_qplib_qp *qp,
2016 		      struct bnxt_qplib_cqe **pcqe, int *budget)
2017 {
2018 	u32 sw_prod, sw_cons;
2019 	struct bnxt_qplib_cqe *cqe;
2020 	int rc = 0;
2021 
2022 	/* Now complete all outstanding SQEs with FLUSHED_ERR */
2023 	sw_prod = HWQ_CMP(sq->hwq.prod, &sq->hwq);
2024 	cqe = *pcqe;
2025 	while (*budget) {
2026 		sw_cons = HWQ_CMP(sq->hwq.cons, &sq->hwq);
2027 		if (sw_cons == sw_prod) {
2028 			break;
2029 		}
2030 		/* Skip the FENCE WQE completions */
2031 		if (sq->swq[sw_cons].wr_id == BNXT_QPLIB_FENCE_WRID) {
2032 			bnxt_qplib_cancel_phantom_processing(qp);
2033 			goto skip_compl;
2034 		}
2035 		memset(cqe, 0, sizeof(*cqe));
2036 		cqe->status = CQ_REQ_STATUS_WORK_REQUEST_FLUSHED_ERR;
2037 		cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
2038 		cqe->qp_handle = (u64)(unsigned long)qp;
2039 		cqe->wr_id = sq->swq[sw_cons].wr_id;
2040 		cqe->src_qp = qp->id;
2041 		cqe->type = sq->swq[sw_cons].type;
2042 		cqe++;
2043 		(*budget)--;
2044 skip_compl:
2045 		sq->hwq.cons++;
2046 	}
2047 	*pcqe = cqe;
2048 	if (!(*budget) && HWQ_CMP(sq->hwq.cons, &sq->hwq) != sw_prod)
2049 		/* Out of budget */
2050 		rc = -EAGAIN;
2051 
2052 	return rc;
2053 }
2054 
2055 static int __flush_rq(struct bnxt_qplib_q *rq, struct bnxt_qplib_qp *qp,
2056 		      struct bnxt_qplib_cqe **pcqe, int *budget)
2057 {
2058 	struct bnxt_qplib_cqe *cqe;
2059 	u32 sw_prod, sw_cons;
2060 	int rc = 0;
2061 	int opcode = 0;
2062 
2063 	switch (qp->type) {
2064 	case CMDQ_CREATE_QP1_TYPE_GSI:
2065 		opcode = CQ_BASE_CQE_TYPE_RES_RAWETH_QP1;
2066 		break;
2067 	case CMDQ_CREATE_QP_TYPE_RC:
2068 		opcode = CQ_BASE_CQE_TYPE_RES_RC;
2069 		break;
2070 	case CMDQ_CREATE_QP_TYPE_UD:
2071 	case CMDQ_CREATE_QP_TYPE_GSI:
2072 		opcode = CQ_BASE_CQE_TYPE_RES_UD;
2073 		break;
2074 	}
2075 
2076 	/* Flush the rest of the RQ */
2077 	sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq);
2078 	cqe = *pcqe;
2079 	while (*budget) {
2080 		sw_cons = HWQ_CMP(rq->hwq.cons, &rq->hwq);
2081 		if (sw_cons == sw_prod)
2082 			break;
2083 		memset(cqe, 0, sizeof(*cqe));
2084 		cqe->status =
2085 		    CQ_RES_RC_STATUS_WORK_REQUEST_FLUSHED_ERR;
2086 		cqe->opcode = opcode;
2087 		cqe->qp_handle = (unsigned long)qp;
2088 		cqe->wr_id = rq->swq[sw_cons].wr_id;
2089 		cqe++;
2090 		(*budget)--;
2091 		rq->hwq.cons++;
2092 	}
2093 	*pcqe = cqe;
2094 	if (!*budget && HWQ_CMP(rq->hwq.cons, &rq->hwq) != sw_prod)
2095 		/* Out of budget */
2096 		rc = -EAGAIN;
2097 
2098 	return rc;
2099 }
2100 
2101 void bnxt_qplib_mark_qp_error(void *qp_handle)
2102 {
2103 	struct bnxt_qplib_qp *qp = qp_handle;
2104 
2105 	if (!qp)
2106 		return;
2107 
2108 	/* Must block new posting of SQ and RQ */
2109 	qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2110 	bnxt_qplib_cancel_phantom_processing(qp);
2111 }
2112 
2113 /* Note: SQE is valid from sw_sq_cons up to cqe_sq_cons (exclusive)
2114  *       CQE is track from sw_cq_cons to max_element but valid only if VALID=1
2115  */
2116 static int do_wa9060(struct bnxt_qplib_qp *qp, struct bnxt_qplib_cq *cq,
2117 		     u32 cq_cons, u32 sw_sq_cons, u32 cqe_sq_cons)
2118 {
2119 	struct bnxt_qplib_q *sq = &qp->sq;
2120 	struct bnxt_qplib_swq *swq;
2121 	u32 peek_sw_cq_cons, peek_raw_cq_cons, peek_sq_cons_idx;
2122 	struct cq_base *peek_hwcqe, **peek_hw_cqe_ptr;
2123 	struct cq_req *peek_req_hwcqe;
2124 	struct bnxt_qplib_qp *peek_qp;
2125 	struct bnxt_qplib_q *peek_sq;
2126 	int i, rc = 0;
2127 
2128 	/* Normal mode */
2129 	/* Check for the psn_search marking before completing */
2130 	swq = &sq->swq[sw_sq_cons];
2131 	if (swq->psn_search &&
2132 	    le32_to_cpu(swq->psn_search->flags_next_psn) & 0x80000000) {
2133 		/* Unmark */
2134 		swq->psn_search->flags_next_psn = cpu_to_le32
2135 			(le32_to_cpu(swq->psn_search->flags_next_psn)
2136 				     & ~0x80000000);
2137 		dev_dbg(&cq->hwq.pdev->dev,
2138 			"FP: Process Req cq_cons=0x%x qp=0x%x sq cons sw=0x%x cqe=0x%x marked!\n",
2139 			cq_cons, qp->id, sw_sq_cons, cqe_sq_cons);
2140 		sq->condition = true;
2141 		sq->send_phantom = true;
2142 
2143 		/* TODO: Only ARM if the previous SQE is ARMALL */
2144 		bnxt_qplib_arm_cq(cq, DBC_DBC_TYPE_CQ_ARMALL);
2145 
2146 		rc = -EAGAIN;
2147 		goto out;
2148 	}
2149 	if (sq->condition) {
2150 		/* Peek at the completions */
2151 		peek_raw_cq_cons = cq->hwq.cons;
2152 		peek_sw_cq_cons = cq_cons;
2153 		i = cq->hwq.max_elements;
2154 		while (i--) {
2155 			peek_sw_cq_cons = HWQ_CMP((peek_sw_cq_cons), &cq->hwq);
2156 			peek_hw_cqe_ptr = (struct cq_base **)cq->hwq.pbl_ptr;
2157 			peek_hwcqe = &peek_hw_cqe_ptr[CQE_PG(peek_sw_cq_cons)]
2158 						     [CQE_IDX(peek_sw_cq_cons)];
2159 			/* If the next hwcqe is VALID */
2160 			if (CQE_CMP_VALID(peek_hwcqe, peek_raw_cq_cons,
2161 					  cq->hwq.max_elements)) {
2162 			/*
2163 			 * The valid test of the entry must be done first before
2164 			 * reading any further.
2165 			 */
2166 				dma_rmb();
2167 				/* If the next hwcqe is a REQ */
2168 				if ((peek_hwcqe->cqe_type_toggle &
2169 				    CQ_BASE_CQE_TYPE_MASK) ==
2170 				    CQ_BASE_CQE_TYPE_REQ) {
2171 					peek_req_hwcqe = (struct cq_req *)
2172 							 peek_hwcqe;
2173 					peek_qp = (struct bnxt_qplib_qp *)
2174 						((unsigned long)
2175 						 le64_to_cpu
2176 						 (peek_req_hwcqe->qp_handle));
2177 					peek_sq = &peek_qp->sq;
2178 					peek_sq_cons_idx = HWQ_CMP(le16_to_cpu(
2179 						peek_req_hwcqe->sq_cons_idx) - 1
2180 						, &sq->hwq);
2181 					/* If the hwcqe's sq's wr_id matches */
2182 					if (peek_sq == sq &&
2183 					    sq->swq[peek_sq_cons_idx].wr_id ==
2184 					    BNXT_QPLIB_FENCE_WRID) {
2185 						/*
2186 						 *  Unbreak only if the phantom
2187 						 *  comes back
2188 						 */
2189 						dev_dbg(&cq->hwq.pdev->dev,
2190 							"FP: Got Phantom CQE\n");
2191 						sq->condition = false;
2192 						sq->single = true;
2193 						rc = 0;
2194 						goto out;
2195 					}
2196 				}
2197 				/* Valid but not the phantom, so keep looping */
2198 			} else {
2199 				/* Not valid yet, just exit and wait */
2200 				rc = -EINVAL;
2201 				goto out;
2202 			}
2203 			peek_sw_cq_cons++;
2204 			peek_raw_cq_cons++;
2205 		}
2206 		dev_err(&cq->hwq.pdev->dev,
2207 			"Should not have come here! cq_cons=0x%x qp=0x%x sq cons sw=0x%x hw=0x%x\n",
2208 			cq_cons, qp->id, sw_sq_cons, cqe_sq_cons);
2209 		rc = -EINVAL;
2210 	}
2211 out:
2212 	return rc;
2213 }
2214 
2215 static int bnxt_qplib_cq_process_req(struct bnxt_qplib_cq *cq,
2216 				     struct cq_req *hwcqe,
2217 				     struct bnxt_qplib_cqe **pcqe, int *budget,
2218 				     u32 cq_cons, struct bnxt_qplib_qp **lib_qp)
2219 {
2220 	struct bnxt_qplib_qp *qp;
2221 	struct bnxt_qplib_q *sq;
2222 	struct bnxt_qplib_cqe *cqe;
2223 	u32 sw_sq_cons, cqe_sq_cons;
2224 	struct bnxt_qplib_swq *swq;
2225 	int rc = 0;
2226 
2227 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2228 				      le64_to_cpu(hwcqe->qp_handle));
2229 	if (!qp) {
2230 		dev_err(&cq->hwq.pdev->dev,
2231 			"FP: Process Req qp is NULL\n");
2232 		return -EINVAL;
2233 	}
2234 	sq = &qp->sq;
2235 
2236 	cqe_sq_cons = HWQ_CMP(le16_to_cpu(hwcqe->sq_cons_idx), &sq->hwq);
2237 	if (cqe_sq_cons > sq->hwq.max_elements) {
2238 		dev_err(&cq->hwq.pdev->dev,
2239 			"FP: CQ Process req reported sq_cons_idx 0x%x which exceeded max 0x%x\n",
2240 			cqe_sq_cons, sq->hwq.max_elements);
2241 		return -EINVAL;
2242 	}
2243 
2244 	if (qp->sq.flushed) {
2245 		dev_dbg(&cq->hwq.pdev->dev,
2246 			"%s: QP in Flush QP = %p\n", __func__, qp);
2247 		goto done;
2248 	}
2249 	/* Require to walk the sq's swq to fabricate CQEs for all previously
2250 	 * signaled SWQEs due to CQE aggregation from the current sq cons
2251 	 * to the cqe_sq_cons
2252 	 */
2253 	cqe = *pcqe;
2254 	while (*budget) {
2255 		sw_sq_cons = HWQ_CMP(sq->hwq.cons, &sq->hwq);
2256 		if (sw_sq_cons == cqe_sq_cons)
2257 			/* Done */
2258 			break;
2259 
2260 		swq = &sq->swq[sw_sq_cons];
2261 		memset(cqe, 0, sizeof(*cqe));
2262 		cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
2263 		cqe->qp_handle = (u64)(unsigned long)qp;
2264 		cqe->src_qp = qp->id;
2265 		cqe->wr_id = swq->wr_id;
2266 		if (cqe->wr_id == BNXT_QPLIB_FENCE_WRID)
2267 			goto skip;
2268 		cqe->type = swq->type;
2269 
2270 		/* For the last CQE, check for status.  For errors, regardless
2271 		 * of the request being signaled or not, it must complete with
2272 		 * the hwcqe error status
2273 		 */
2274 		if (HWQ_CMP((sw_sq_cons + 1), &sq->hwq) == cqe_sq_cons &&
2275 		    hwcqe->status != CQ_REQ_STATUS_OK) {
2276 			cqe->status = hwcqe->status;
2277 			dev_err(&cq->hwq.pdev->dev,
2278 				"FP: CQ Processed Req wr_id[%d] = 0x%llx with status 0x%x\n",
2279 				sw_sq_cons, cqe->wr_id, cqe->status);
2280 			cqe++;
2281 			(*budget)--;
2282 			bnxt_qplib_mark_qp_error(qp);
2283 			/* Add qp to flush list of the CQ */
2284 			bnxt_qplib_add_flush_qp(qp);
2285 		} else {
2286 			/* Before we complete, do WA 9060 */
2287 			if (do_wa9060(qp, cq, cq_cons, sw_sq_cons,
2288 				      cqe_sq_cons)) {
2289 				*lib_qp = qp;
2290 				goto out;
2291 			}
2292 			if (swq->flags & SQ_SEND_FLAGS_SIGNAL_COMP) {
2293 				cqe->status = CQ_REQ_STATUS_OK;
2294 				cqe++;
2295 				(*budget)--;
2296 			}
2297 		}
2298 skip:
2299 		sq->hwq.cons++;
2300 		if (sq->single)
2301 			break;
2302 	}
2303 out:
2304 	*pcqe = cqe;
2305 	if (HWQ_CMP(sq->hwq.cons, &sq->hwq) != cqe_sq_cons) {
2306 		/* Out of budget */
2307 		rc = -EAGAIN;
2308 		goto done;
2309 	}
2310 	/*
2311 	 * Back to normal completion mode only after it has completed all of
2312 	 * the WC for this CQE
2313 	 */
2314 	sq->single = false;
2315 done:
2316 	return rc;
2317 }
2318 
2319 static void bnxt_qplib_release_srqe(struct bnxt_qplib_srq *srq, u32 tag)
2320 {
2321 	spin_lock(&srq->hwq.lock);
2322 	srq->swq[srq->last_idx].next_idx = (int)tag;
2323 	srq->last_idx = (int)tag;
2324 	srq->swq[srq->last_idx].next_idx = -1;
2325 	srq->hwq.cons++; /* Support for SRQE counter */
2326 	spin_unlock(&srq->hwq.lock);
2327 }
2328 
2329 static int bnxt_qplib_cq_process_res_rc(struct bnxt_qplib_cq *cq,
2330 					struct cq_res_rc *hwcqe,
2331 					struct bnxt_qplib_cqe **pcqe,
2332 					int *budget)
2333 {
2334 	struct bnxt_qplib_qp *qp;
2335 	struct bnxt_qplib_q *rq;
2336 	struct bnxt_qplib_srq *srq;
2337 	struct bnxt_qplib_cqe *cqe;
2338 	u32 wr_id_idx;
2339 	int rc = 0;
2340 
2341 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2342 				      le64_to_cpu(hwcqe->qp_handle));
2343 	if (!qp) {
2344 		dev_err(&cq->hwq.pdev->dev, "process_cq RC qp is NULL\n");
2345 		return -EINVAL;
2346 	}
2347 	if (qp->rq.flushed) {
2348 		dev_dbg(&cq->hwq.pdev->dev,
2349 			"%s: QP in Flush QP = %p\n", __func__, qp);
2350 		goto done;
2351 	}
2352 
2353 	cqe = *pcqe;
2354 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2355 	cqe->length = le32_to_cpu(hwcqe->length);
2356 	cqe->invrkey = le32_to_cpu(hwcqe->imm_data_or_inv_r_key);
2357 	cqe->mr_handle = le64_to_cpu(hwcqe->mr_handle);
2358 	cqe->flags = le16_to_cpu(hwcqe->flags);
2359 	cqe->status = hwcqe->status;
2360 	cqe->qp_handle = (u64)(unsigned long)qp;
2361 
2362 	wr_id_idx = le32_to_cpu(hwcqe->srq_or_rq_wr_id) &
2363 				CQ_RES_RC_SRQ_OR_RQ_WR_ID_MASK;
2364 	if (cqe->flags & CQ_RES_RC_FLAGS_SRQ_SRQ) {
2365 		srq = qp->srq;
2366 		if (!srq)
2367 			return -EINVAL;
2368 		if (wr_id_idx >= srq->hwq.max_elements) {
2369 			dev_err(&cq->hwq.pdev->dev,
2370 				"FP: CQ Process RC wr_id idx 0x%x exceeded SRQ max 0x%x\n",
2371 				wr_id_idx, srq->hwq.max_elements);
2372 			return -EINVAL;
2373 		}
2374 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2375 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2376 		cqe++;
2377 		(*budget)--;
2378 		*pcqe = cqe;
2379 	} else {
2380 		rq = &qp->rq;
2381 		if (wr_id_idx >= rq->hwq.max_elements) {
2382 			dev_err(&cq->hwq.pdev->dev,
2383 				"FP: CQ Process RC wr_id idx 0x%x exceeded RQ max 0x%x\n",
2384 				wr_id_idx, rq->hwq.max_elements);
2385 			return -EINVAL;
2386 		}
2387 		cqe->wr_id = rq->swq[wr_id_idx].wr_id;
2388 		cqe++;
2389 		(*budget)--;
2390 		rq->hwq.cons++;
2391 		*pcqe = cqe;
2392 
2393 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2394 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2395 			/* Add qp to flush list of the CQ */
2396 			bnxt_qplib_add_flush_qp(qp);
2397 		}
2398 	}
2399 
2400 done:
2401 	return rc;
2402 }
2403 
2404 static int bnxt_qplib_cq_process_res_ud(struct bnxt_qplib_cq *cq,
2405 					struct cq_res_ud *hwcqe,
2406 					struct bnxt_qplib_cqe **pcqe,
2407 					int *budget)
2408 {
2409 	struct bnxt_qplib_qp *qp;
2410 	struct bnxt_qplib_q *rq;
2411 	struct bnxt_qplib_srq *srq;
2412 	struct bnxt_qplib_cqe *cqe;
2413 	u32 wr_id_idx;
2414 	int rc = 0;
2415 
2416 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2417 				      le64_to_cpu(hwcqe->qp_handle));
2418 	if (!qp) {
2419 		dev_err(&cq->hwq.pdev->dev, "process_cq UD qp is NULL\n");
2420 		return -EINVAL;
2421 	}
2422 	if (qp->rq.flushed) {
2423 		dev_dbg(&cq->hwq.pdev->dev,
2424 			"%s: QP in Flush QP = %p\n", __func__, qp);
2425 		goto done;
2426 	}
2427 	cqe = *pcqe;
2428 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2429 	cqe->length = (u32)le16_to_cpu(hwcqe->length);
2430 	cqe->cfa_meta = le16_to_cpu(hwcqe->cfa_metadata);
2431 	cqe->invrkey = le32_to_cpu(hwcqe->imm_data);
2432 	cqe->flags = le16_to_cpu(hwcqe->flags);
2433 	cqe->status = hwcqe->status;
2434 	cqe->qp_handle = (u64)(unsigned long)qp;
2435 	/*FIXME: Endianness fix needed for smace */
2436 	memcpy(cqe->smac, hwcqe->src_mac, ETH_ALEN);
2437 	wr_id_idx = le32_to_cpu(hwcqe->src_qp_high_srq_or_rq_wr_id)
2438 				& CQ_RES_UD_SRQ_OR_RQ_WR_ID_MASK;
2439 	cqe->src_qp = le16_to_cpu(hwcqe->src_qp_low) |
2440 				  ((le32_to_cpu(
2441 				  hwcqe->src_qp_high_srq_or_rq_wr_id) &
2442 				 CQ_RES_UD_SRC_QP_HIGH_MASK) >> 8);
2443 
2444 	if (cqe->flags & CQ_RES_RC_FLAGS_SRQ_SRQ) {
2445 		srq = qp->srq;
2446 		if (!srq)
2447 			return -EINVAL;
2448 
2449 		if (wr_id_idx >= srq->hwq.max_elements) {
2450 			dev_err(&cq->hwq.pdev->dev,
2451 				"FP: CQ Process UD wr_id idx 0x%x exceeded SRQ max 0x%x\n",
2452 				wr_id_idx, srq->hwq.max_elements);
2453 			return -EINVAL;
2454 		}
2455 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2456 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2457 		cqe++;
2458 		(*budget)--;
2459 		*pcqe = cqe;
2460 	} else {
2461 		rq = &qp->rq;
2462 		if (wr_id_idx >= rq->hwq.max_elements) {
2463 			dev_err(&cq->hwq.pdev->dev,
2464 				"FP: CQ Process UD wr_id idx 0x%x exceeded RQ max 0x%x\n",
2465 				wr_id_idx, rq->hwq.max_elements);
2466 			return -EINVAL;
2467 		}
2468 
2469 		cqe->wr_id = rq->swq[wr_id_idx].wr_id;
2470 		cqe++;
2471 		(*budget)--;
2472 		rq->hwq.cons++;
2473 		*pcqe = cqe;
2474 
2475 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2476 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2477 			/* Add qp to flush list of the CQ */
2478 			bnxt_qplib_add_flush_qp(qp);
2479 		}
2480 	}
2481 done:
2482 	return rc;
2483 }
2484 
2485 bool bnxt_qplib_is_cq_empty(struct bnxt_qplib_cq *cq)
2486 {
2487 	struct cq_base *hw_cqe, **hw_cqe_ptr;
2488 	u32 sw_cons, raw_cons;
2489 	bool rc = true;
2490 
2491 	raw_cons = cq->hwq.cons;
2492 	sw_cons = HWQ_CMP(raw_cons, &cq->hwq);
2493 	hw_cqe_ptr = (struct cq_base **)cq->hwq.pbl_ptr;
2494 	hw_cqe = &hw_cqe_ptr[CQE_PG(sw_cons)][CQE_IDX(sw_cons)];
2495 
2496 	 /* Check for Valid bit. If the CQE is valid, return false */
2497 	rc = !CQE_CMP_VALID(hw_cqe, raw_cons, cq->hwq.max_elements);
2498 	return rc;
2499 }
2500 
2501 static int bnxt_qplib_cq_process_res_raweth_qp1(struct bnxt_qplib_cq *cq,
2502 						struct cq_res_raweth_qp1 *hwcqe,
2503 						struct bnxt_qplib_cqe **pcqe,
2504 						int *budget)
2505 {
2506 	struct bnxt_qplib_qp *qp;
2507 	struct bnxt_qplib_q *rq;
2508 	struct bnxt_qplib_srq *srq;
2509 	struct bnxt_qplib_cqe *cqe;
2510 	u32 wr_id_idx;
2511 	int rc = 0;
2512 
2513 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2514 				      le64_to_cpu(hwcqe->qp_handle));
2515 	if (!qp) {
2516 		dev_err(&cq->hwq.pdev->dev, "process_cq Raw/QP1 qp is NULL\n");
2517 		return -EINVAL;
2518 	}
2519 	if (qp->rq.flushed) {
2520 		dev_dbg(&cq->hwq.pdev->dev,
2521 			"%s: QP in Flush QP = %p\n", __func__, qp);
2522 		goto done;
2523 	}
2524 	cqe = *pcqe;
2525 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2526 	cqe->flags = le16_to_cpu(hwcqe->flags);
2527 	cqe->qp_handle = (u64)(unsigned long)qp;
2528 
2529 	wr_id_idx =
2530 		le32_to_cpu(hwcqe->raweth_qp1_payload_offset_srq_or_rq_wr_id)
2531 				& CQ_RES_RAWETH_QP1_SRQ_OR_RQ_WR_ID_MASK;
2532 	cqe->src_qp = qp->id;
2533 	if (qp->id == 1 && !cqe->length) {
2534 		/* Add workaround for the length misdetection */
2535 		cqe->length = 296;
2536 	} else {
2537 		cqe->length = le16_to_cpu(hwcqe->length);
2538 	}
2539 	cqe->pkey_index = qp->pkey_index;
2540 	memcpy(cqe->smac, qp->smac, 6);
2541 
2542 	cqe->raweth_qp1_flags = le16_to_cpu(hwcqe->raweth_qp1_flags);
2543 	cqe->raweth_qp1_flags2 = le32_to_cpu(hwcqe->raweth_qp1_flags2);
2544 	cqe->raweth_qp1_metadata = le32_to_cpu(hwcqe->raweth_qp1_metadata);
2545 
2546 	if (cqe->flags & CQ_RES_RAWETH_QP1_FLAGS_SRQ_SRQ) {
2547 		srq = qp->srq;
2548 		if (!srq) {
2549 			dev_err(&cq->hwq.pdev->dev,
2550 				"FP: SRQ used but not defined??\n");
2551 			return -EINVAL;
2552 		}
2553 		if (wr_id_idx >= srq->hwq.max_elements) {
2554 			dev_err(&cq->hwq.pdev->dev,
2555 				"FP: CQ Process Raw/QP1 wr_id idx 0x%x exceeded SRQ max 0x%x\n",
2556 				wr_id_idx, srq->hwq.max_elements);
2557 			return -EINVAL;
2558 		}
2559 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2560 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2561 		cqe++;
2562 		(*budget)--;
2563 		*pcqe = cqe;
2564 	} else {
2565 		rq = &qp->rq;
2566 		if (wr_id_idx >= rq->hwq.max_elements) {
2567 			dev_err(&cq->hwq.pdev->dev,
2568 				"FP: CQ Process Raw/QP1 RQ wr_id idx 0x%x exceeded RQ max 0x%x\n",
2569 				wr_id_idx, rq->hwq.max_elements);
2570 			return -EINVAL;
2571 		}
2572 		cqe->wr_id = rq->swq[wr_id_idx].wr_id;
2573 		cqe++;
2574 		(*budget)--;
2575 		rq->hwq.cons++;
2576 		*pcqe = cqe;
2577 
2578 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2579 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2580 			/* Add qp to flush list of the CQ */
2581 			bnxt_qplib_add_flush_qp(qp);
2582 		}
2583 	}
2584 
2585 done:
2586 	return rc;
2587 }
2588 
2589 static int bnxt_qplib_cq_process_terminal(struct bnxt_qplib_cq *cq,
2590 					  struct cq_terminal *hwcqe,
2591 					  struct bnxt_qplib_cqe **pcqe,
2592 					  int *budget)
2593 {
2594 	struct bnxt_qplib_qp *qp;
2595 	struct bnxt_qplib_q *sq, *rq;
2596 	struct bnxt_qplib_cqe *cqe;
2597 	u32 sw_cons = 0, cqe_cons;
2598 	int rc = 0;
2599 
2600 	/* Check the Status */
2601 	if (hwcqe->status != CQ_TERMINAL_STATUS_OK)
2602 		dev_warn(&cq->hwq.pdev->dev,
2603 			 "FP: CQ Process Terminal Error status = 0x%x\n",
2604 			 hwcqe->status);
2605 
2606 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2607 				      le64_to_cpu(hwcqe->qp_handle));
2608 	if (!qp) {
2609 		dev_err(&cq->hwq.pdev->dev,
2610 			"FP: CQ Process terminal qp is NULL\n");
2611 		return -EINVAL;
2612 	}
2613 
2614 	/* Must block new posting of SQ and RQ */
2615 	qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2616 
2617 	sq = &qp->sq;
2618 	rq = &qp->rq;
2619 
2620 	cqe_cons = le16_to_cpu(hwcqe->sq_cons_idx);
2621 	if (cqe_cons == 0xFFFF)
2622 		goto do_rq;
2623 
2624 	if (cqe_cons > sq->hwq.max_elements) {
2625 		dev_err(&cq->hwq.pdev->dev,
2626 			"FP: CQ Process terminal reported sq_cons_idx 0x%x which exceeded max 0x%x\n",
2627 			cqe_cons, sq->hwq.max_elements);
2628 		goto do_rq;
2629 	}
2630 
2631 	if (qp->sq.flushed) {
2632 		dev_dbg(&cq->hwq.pdev->dev,
2633 			"%s: QP in Flush QP = %p\n", __func__, qp);
2634 		goto sq_done;
2635 	}
2636 
2637 	/* Terminal CQE can also include aggregated successful CQEs prior.
2638 	 * So we must complete all CQEs from the current sq's cons to the
2639 	 * cq_cons with status OK
2640 	 */
2641 	cqe = *pcqe;
2642 	while (*budget) {
2643 		sw_cons = HWQ_CMP(sq->hwq.cons, &sq->hwq);
2644 		if (sw_cons == cqe_cons)
2645 			break;
2646 		if (sq->swq[sw_cons].flags & SQ_SEND_FLAGS_SIGNAL_COMP) {
2647 			memset(cqe, 0, sizeof(*cqe));
2648 			cqe->status = CQ_REQ_STATUS_OK;
2649 			cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
2650 			cqe->qp_handle = (u64)(unsigned long)qp;
2651 			cqe->src_qp = qp->id;
2652 			cqe->wr_id = sq->swq[sw_cons].wr_id;
2653 			cqe->type = sq->swq[sw_cons].type;
2654 			cqe++;
2655 			(*budget)--;
2656 		}
2657 		sq->hwq.cons++;
2658 	}
2659 	*pcqe = cqe;
2660 	if (!(*budget) && sw_cons != cqe_cons) {
2661 		/* Out of budget */
2662 		rc = -EAGAIN;
2663 		goto sq_done;
2664 	}
2665 sq_done:
2666 	if (rc)
2667 		return rc;
2668 do_rq:
2669 	cqe_cons = le16_to_cpu(hwcqe->rq_cons_idx);
2670 	if (cqe_cons == 0xFFFF) {
2671 		goto done;
2672 	} else if (cqe_cons > rq->hwq.max_elements) {
2673 		dev_err(&cq->hwq.pdev->dev,
2674 			"FP: CQ Processed terminal reported rq_cons_idx 0x%x exceeds max 0x%x\n",
2675 			cqe_cons, rq->hwq.max_elements);
2676 		goto done;
2677 	}
2678 
2679 	if (qp->rq.flushed) {
2680 		dev_dbg(&cq->hwq.pdev->dev,
2681 			"%s: QP in Flush QP = %p\n", __func__, qp);
2682 		rc = 0;
2683 		goto done;
2684 	}
2685 
2686 	/* Terminal CQE requires all posted RQEs to complete with FLUSHED_ERR
2687 	 * from the current rq->cons to the rq->prod regardless what the
2688 	 * rq->cons the terminal CQE indicates
2689 	 */
2690 
2691 	/* Add qp to flush list of the CQ */
2692 	bnxt_qplib_add_flush_qp(qp);
2693 done:
2694 	return rc;
2695 }
2696 
2697 static int bnxt_qplib_cq_process_cutoff(struct bnxt_qplib_cq *cq,
2698 					struct cq_cutoff *hwcqe)
2699 {
2700 	/* Check the Status */
2701 	if (hwcqe->status != CQ_CUTOFF_STATUS_OK) {
2702 		dev_err(&cq->hwq.pdev->dev,
2703 			"FP: CQ Process Cutoff Error status = 0x%x\n",
2704 			hwcqe->status);
2705 		return -EINVAL;
2706 	}
2707 	clear_bit(CQ_FLAGS_RESIZE_IN_PROG, &cq->flags);
2708 	wake_up_interruptible(&cq->waitq);
2709 
2710 	return 0;
2711 }
2712 
2713 int bnxt_qplib_process_flush_list(struct bnxt_qplib_cq *cq,
2714 				  struct bnxt_qplib_cqe *cqe,
2715 				  int num_cqes)
2716 {
2717 	struct bnxt_qplib_qp *qp = NULL;
2718 	u32 budget = num_cqes;
2719 	unsigned long flags;
2720 
2721 	spin_lock_irqsave(&cq->flush_lock, flags);
2722 	list_for_each_entry(qp, &cq->sqf_head, sq_flush) {
2723 		dev_dbg(&cq->hwq.pdev->dev, "FP: Flushing SQ QP= %p\n", qp);
2724 		__flush_sq(&qp->sq, qp, &cqe, &budget);
2725 	}
2726 
2727 	list_for_each_entry(qp, &cq->rqf_head, rq_flush) {
2728 		dev_dbg(&cq->hwq.pdev->dev, "FP: Flushing RQ QP= %p\n", qp);
2729 		__flush_rq(&qp->rq, qp, &cqe, &budget);
2730 	}
2731 	spin_unlock_irqrestore(&cq->flush_lock, flags);
2732 
2733 	return num_cqes - budget;
2734 }
2735 
2736 int bnxt_qplib_poll_cq(struct bnxt_qplib_cq *cq, struct bnxt_qplib_cqe *cqe,
2737 		       int num_cqes, struct bnxt_qplib_qp **lib_qp)
2738 {
2739 	struct cq_base *hw_cqe, **hw_cqe_ptr;
2740 	u32 sw_cons, raw_cons;
2741 	int budget, rc = 0;
2742 
2743 	raw_cons = cq->hwq.cons;
2744 	budget = num_cqes;
2745 
2746 	while (budget) {
2747 		sw_cons = HWQ_CMP(raw_cons, &cq->hwq);
2748 		hw_cqe_ptr = (struct cq_base **)cq->hwq.pbl_ptr;
2749 		hw_cqe = &hw_cqe_ptr[CQE_PG(sw_cons)][CQE_IDX(sw_cons)];
2750 
2751 		/* Check for Valid bit */
2752 		if (!CQE_CMP_VALID(hw_cqe, raw_cons, cq->hwq.max_elements))
2753 			break;
2754 
2755 		/*
2756 		 * The valid test of the entry must be done first before
2757 		 * reading any further.
2758 		 */
2759 		dma_rmb();
2760 		/* From the device's respective CQE format to qplib_wc*/
2761 		switch (hw_cqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK) {
2762 		case CQ_BASE_CQE_TYPE_REQ:
2763 			rc = bnxt_qplib_cq_process_req(cq,
2764 						       (struct cq_req *)hw_cqe,
2765 						       &cqe, &budget,
2766 						       sw_cons, lib_qp);
2767 			break;
2768 		case CQ_BASE_CQE_TYPE_RES_RC:
2769 			rc = bnxt_qplib_cq_process_res_rc(cq,
2770 							  (struct cq_res_rc *)
2771 							  hw_cqe, &cqe,
2772 							  &budget);
2773 			break;
2774 		case CQ_BASE_CQE_TYPE_RES_UD:
2775 			rc = bnxt_qplib_cq_process_res_ud
2776 					(cq, (struct cq_res_ud *)hw_cqe, &cqe,
2777 					 &budget);
2778 			break;
2779 		case CQ_BASE_CQE_TYPE_RES_RAWETH_QP1:
2780 			rc = bnxt_qplib_cq_process_res_raweth_qp1
2781 					(cq, (struct cq_res_raweth_qp1 *)
2782 					 hw_cqe, &cqe, &budget);
2783 			break;
2784 		case CQ_BASE_CQE_TYPE_TERMINAL:
2785 			rc = bnxt_qplib_cq_process_terminal
2786 					(cq, (struct cq_terminal *)hw_cqe,
2787 					 &cqe, &budget);
2788 			break;
2789 		case CQ_BASE_CQE_TYPE_CUT_OFF:
2790 			bnxt_qplib_cq_process_cutoff
2791 					(cq, (struct cq_cutoff *)hw_cqe);
2792 			/* Done processing this CQ */
2793 			goto exit;
2794 		default:
2795 			dev_err(&cq->hwq.pdev->dev,
2796 				"process_cq unknown type 0x%lx\n",
2797 				hw_cqe->cqe_type_toggle &
2798 				CQ_BASE_CQE_TYPE_MASK);
2799 			rc = -EINVAL;
2800 			break;
2801 		}
2802 		if (rc < 0) {
2803 			if (rc == -EAGAIN)
2804 				break;
2805 			/* Error while processing the CQE, just skip to the
2806 			 * next one
2807 			 */
2808 			dev_err(&cq->hwq.pdev->dev,
2809 				"process_cqe error rc = 0x%x\n", rc);
2810 		}
2811 		raw_cons++;
2812 	}
2813 	if (cq->hwq.cons != raw_cons) {
2814 		cq->hwq.cons = raw_cons;
2815 		bnxt_qplib_arm_cq(cq, DBC_DBC_TYPE_CQ);
2816 	}
2817 exit:
2818 	return num_cqes - budget;
2819 }
2820 
2821 void bnxt_qplib_req_notify_cq(struct bnxt_qplib_cq *cq, u32 arm_type)
2822 {
2823 	if (arm_type)
2824 		bnxt_qplib_arm_cq(cq, arm_type);
2825 	/* Using cq->arm_state variable to track whether to issue cq handler */
2826 	atomic_set(&cq->arm_state, 1);
2827 }
2828 
2829 void bnxt_qplib_flush_cqn_wq(struct bnxt_qplib_qp *qp)
2830 {
2831 	flush_workqueue(qp->scq->nq->cqn_wq);
2832 	if (qp->scq != qp->rcq)
2833 		flush_workqueue(qp->rcq->nq->cqn_wq);
2834 }
2835