xref: /linux/drivers/infiniband/hw/bnxt_re/qplib_fp.c (revision 55f3538c4923e9dfca132e99ebec370e8094afda)
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 #include <linux/interrupt.h>
40 #include <linux/spinlock.h>
41 #include <linux/sched.h>
42 #include <linux/slab.h>
43 #include <linux/pci.h>
44 #include <linux/prefetch.h>
45 
46 #include "roce_hsi.h"
47 
48 #include "qplib_res.h"
49 #include "qplib_rcfw.h"
50 #include "qplib_sp.h"
51 #include "qplib_fp.h"
52 
53 static void bnxt_qplib_arm_cq_enable(struct bnxt_qplib_cq *cq);
54 static void __clean_cq(struct bnxt_qplib_cq *cq, u64 qp);
55 static void bnxt_qplib_arm_srq(struct bnxt_qplib_srq *srq, u32 arm_type);
56 
57 static void bnxt_qplib_cancel_phantom_processing(struct bnxt_qplib_qp *qp)
58 {
59 	qp->sq.condition = false;
60 	qp->sq.send_phantom = false;
61 	qp->sq.single = false;
62 }
63 
64 /* Flush list */
65 static void __bnxt_qplib_add_flush_qp(struct bnxt_qplib_qp *qp)
66 {
67 	struct bnxt_qplib_cq *scq, *rcq;
68 
69 	scq = qp->scq;
70 	rcq = qp->rcq;
71 
72 	if (!qp->sq.flushed) {
73 		dev_dbg(&scq->hwq.pdev->dev,
74 			"QPLIB: FP: Adding to SQ Flush list = %p",
75 			qp);
76 		bnxt_qplib_cancel_phantom_processing(qp);
77 		list_add_tail(&qp->sq_flush, &scq->sqf_head);
78 		qp->sq.flushed = true;
79 	}
80 	if (!qp->srq) {
81 		if (!qp->rq.flushed) {
82 			dev_dbg(&rcq->hwq.pdev->dev,
83 				"QPLIB: FP: Adding to RQ Flush list = %p",
84 				qp);
85 			list_add_tail(&qp->rq_flush, &rcq->rqf_head);
86 			qp->rq.flushed = true;
87 		}
88 	}
89 }
90 
91 void bnxt_qplib_acquire_cq_locks(struct bnxt_qplib_qp *qp,
92 				 unsigned long *flags)
93 	__acquires(&qp->scq->hwq.lock) __acquires(&qp->rcq->hwq.lock)
94 {
95 	spin_lock_irqsave(&qp->scq->hwq.lock, *flags);
96 	if (qp->scq == qp->rcq)
97 		__acquire(&qp->rcq->hwq.lock);
98 	else
99 		spin_lock(&qp->rcq->hwq.lock);
100 }
101 
102 void bnxt_qplib_release_cq_locks(struct bnxt_qplib_qp *qp,
103 				 unsigned long *flags)
104 	__releases(&qp->scq->hwq.lock) __releases(&qp->rcq->hwq.lock)
105 {
106 	if (qp->scq == qp->rcq)
107 		__release(&qp->rcq->hwq.lock);
108 	else
109 		spin_unlock(&qp->rcq->hwq.lock);
110 	spin_unlock_irqrestore(&qp->scq->hwq.lock, *flags);
111 }
112 
113 static struct bnxt_qplib_cq *bnxt_qplib_find_buddy_cq(struct bnxt_qplib_qp *qp,
114 						      struct bnxt_qplib_cq *cq)
115 {
116 	struct bnxt_qplib_cq *buddy_cq = NULL;
117 
118 	if (qp->scq == qp->rcq)
119 		buddy_cq = NULL;
120 	else if (qp->scq == cq)
121 		buddy_cq = qp->rcq;
122 	else
123 		buddy_cq = qp->scq;
124 	return buddy_cq;
125 }
126 
127 static void bnxt_qplib_lock_buddy_cq(struct bnxt_qplib_qp *qp,
128 				     struct bnxt_qplib_cq *cq)
129 	__acquires(&buddy_cq->hwq.lock)
130 {
131 	struct bnxt_qplib_cq *buddy_cq = NULL;
132 
133 	buddy_cq = bnxt_qplib_find_buddy_cq(qp, cq);
134 	if (!buddy_cq)
135 		__acquire(&cq->hwq.lock);
136 	else
137 		spin_lock(&buddy_cq->hwq.lock);
138 }
139 
140 static void bnxt_qplib_unlock_buddy_cq(struct bnxt_qplib_qp *qp,
141 				       struct bnxt_qplib_cq *cq)
142 	__releases(&buddy_cq->hwq.lock)
143 {
144 	struct bnxt_qplib_cq *buddy_cq = NULL;
145 
146 	buddy_cq = bnxt_qplib_find_buddy_cq(qp, cq);
147 	if (!buddy_cq)
148 		__release(&cq->hwq.lock);
149 	else
150 		spin_unlock(&buddy_cq->hwq.lock);
151 }
152 
153 void bnxt_qplib_add_flush_qp(struct bnxt_qplib_qp *qp)
154 {
155 	unsigned long flags;
156 
157 	bnxt_qplib_acquire_cq_locks(qp, &flags);
158 	__bnxt_qplib_add_flush_qp(qp);
159 	bnxt_qplib_release_cq_locks(qp, &flags);
160 }
161 
162 static void __bnxt_qplib_del_flush_qp(struct bnxt_qplib_qp *qp)
163 {
164 	if (qp->sq.flushed) {
165 		qp->sq.flushed = false;
166 		list_del(&qp->sq_flush);
167 	}
168 	if (!qp->srq) {
169 		if (qp->rq.flushed) {
170 			qp->rq.flushed = false;
171 			list_del(&qp->rq_flush);
172 		}
173 	}
174 }
175 
176 void bnxt_qplib_del_flush_qp(struct bnxt_qplib_qp *qp)
177 {
178 	unsigned long flags;
179 
180 	bnxt_qplib_acquire_cq_locks(qp, &flags);
181 	__clean_cq(qp->scq, (u64)(unsigned long)qp);
182 	qp->sq.hwq.prod = 0;
183 	qp->sq.hwq.cons = 0;
184 	__clean_cq(qp->rcq, (u64)(unsigned long)qp);
185 	qp->rq.hwq.prod = 0;
186 	qp->rq.hwq.cons = 0;
187 
188 	__bnxt_qplib_del_flush_qp(qp);
189 	bnxt_qplib_release_cq_locks(qp, &flags);
190 }
191 
192 static void bnxt_qpn_cqn_sched_task(struct work_struct *work)
193 {
194 	struct bnxt_qplib_nq_work *nq_work =
195 			container_of(work, struct bnxt_qplib_nq_work, work);
196 
197 	struct bnxt_qplib_cq *cq = nq_work->cq;
198 	struct bnxt_qplib_nq *nq = nq_work->nq;
199 
200 	if (cq && nq) {
201 		spin_lock_bh(&cq->compl_lock);
202 		if (atomic_read(&cq->arm_state) && nq->cqn_handler) {
203 			dev_dbg(&nq->pdev->dev,
204 				"%s:Trigger cq  = %p event nq = %p\n",
205 				__func__, cq, nq);
206 			nq->cqn_handler(nq, cq);
207 		}
208 		spin_unlock_bh(&cq->compl_lock);
209 	}
210 	kfree(nq_work);
211 }
212 
213 static void bnxt_qplib_free_qp_hdr_buf(struct bnxt_qplib_res *res,
214 				       struct bnxt_qplib_qp *qp)
215 {
216 	struct bnxt_qplib_q *rq = &qp->rq;
217 	struct bnxt_qplib_q *sq = &qp->sq;
218 
219 	if (qp->rq_hdr_buf)
220 		dma_free_coherent(&res->pdev->dev,
221 				  rq->hwq.max_elements * qp->rq_hdr_buf_size,
222 				  qp->rq_hdr_buf, qp->rq_hdr_buf_map);
223 	if (qp->sq_hdr_buf)
224 		dma_free_coherent(&res->pdev->dev,
225 				  sq->hwq.max_elements * qp->sq_hdr_buf_size,
226 				  qp->sq_hdr_buf, qp->sq_hdr_buf_map);
227 	qp->rq_hdr_buf = NULL;
228 	qp->sq_hdr_buf = NULL;
229 	qp->rq_hdr_buf_map = 0;
230 	qp->sq_hdr_buf_map = 0;
231 	qp->sq_hdr_buf_size = 0;
232 	qp->rq_hdr_buf_size = 0;
233 }
234 
235 static int bnxt_qplib_alloc_qp_hdr_buf(struct bnxt_qplib_res *res,
236 				       struct bnxt_qplib_qp *qp)
237 {
238 	struct bnxt_qplib_q *rq = &qp->rq;
239 	struct bnxt_qplib_q *sq = &qp->rq;
240 	int rc = 0;
241 
242 	if (qp->sq_hdr_buf_size && sq->hwq.max_elements) {
243 		qp->sq_hdr_buf = dma_alloc_coherent(&res->pdev->dev,
244 					sq->hwq.max_elements *
245 					qp->sq_hdr_buf_size,
246 					&qp->sq_hdr_buf_map, GFP_KERNEL);
247 		if (!qp->sq_hdr_buf) {
248 			rc = -ENOMEM;
249 			dev_err(&res->pdev->dev,
250 				"QPLIB: Failed to create sq_hdr_buf");
251 			goto fail;
252 		}
253 	}
254 
255 	if (qp->rq_hdr_buf_size && rq->hwq.max_elements) {
256 		qp->rq_hdr_buf = dma_alloc_coherent(&res->pdev->dev,
257 						    rq->hwq.max_elements *
258 						    qp->rq_hdr_buf_size,
259 						    &qp->rq_hdr_buf_map,
260 						    GFP_KERNEL);
261 		if (!qp->rq_hdr_buf) {
262 			rc = -ENOMEM;
263 			dev_err(&res->pdev->dev,
264 				"QPLIB: Failed to create rq_hdr_buf");
265 			goto fail;
266 		}
267 	}
268 	return 0;
269 
270 fail:
271 	bnxt_qplib_free_qp_hdr_buf(res, qp);
272 	return rc;
273 }
274 
275 static void bnxt_qplib_service_nq(unsigned long data)
276 {
277 	struct bnxt_qplib_nq *nq = (struct bnxt_qplib_nq *)data;
278 	struct bnxt_qplib_hwq *hwq = &nq->hwq;
279 	struct nq_base *nqe, **nq_ptr;
280 	struct bnxt_qplib_cq *cq;
281 	int num_cqne_processed = 0;
282 	int num_srqne_processed = 0;
283 	u32 sw_cons, raw_cons;
284 	u16 type;
285 	int budget = nq->budget;
286 	u64 q_handle;
287 
288 	/* Service the NQ until empty */
289 	raw_cons = hwq->cons;
290 	while (budget--) {
291 		sw_cons = HWQ_CMP(raw_cons, hwq);
292 		nq_ptr = (struct nq_base **)hwq->pbl_ptr;
293 		nqe = &nq_ptr[NQE_PG(sw_cons)][NQE_IDX(sw_cons)];
294 		if (!NQE_CMP_VALID(nqe, raw_cons, hwq->max_elements))
295 			break;
296 
297 		/*
298 		 * The valid test of the entry must be done first before
299 		 * reading any further.
300 		 */
301 		dma_rmb();
302 
303 		type = le16_to_cpu(nqe->info10_type) & NQ_BASE_TYPE_MASK;
304 		switch (type) {
305 		case NQ_BASE_TYPE_CQ_NOTIFICATION:
306 		{
307 			struct nq_cn *nqcne = (struct nq_cn *)nqe;
308 
309 			q_handle = le32_to_cpu(nqcne->cq_handle_low);
310 			q_handle |= (u64)le32_to_cpu(nqcne->cq_handle_high)
311 						     << 32;
312 			cq = (struct bnxt_qplib_cq *)(unsigned long)q_handle;
313 			bnxt_qplib_arm_cq_enable(cq);
314 			spin_lock_bh(&cq->compl_lock);
315 			atomic_set(&cq->arm_state, 0);
316 			if (!nq->cqn_handler(nq, (cq)))
317 				num_cqne_processed++;
318 			else
319 				dev_warn(&nq->pdev->dev,
320 					 "QPLIB: cqn - type 0x%x not handled",
321 					 type);
322 			spin_unlock_bh(&cq->compl_lock);
323 			break;
324 		}
325 		case NQ_BASE_TYPE_SRQ_EVENT:
326 		{
327 			struct nq_srq_event *nqsrqe =
328 						(struct nq_srq_event *)nqe;
329 
330 			q_handle = le32_to_cpu(nqsrqe->srq_handle_low);
331 			q_handle |= (u64)le32_to_cpu(nqsrqe->srq_handle_high)
332 				     << 32;
333 			bnxt_qplib_arm_srq((struct bnxt_qplib_srq *)q_handle,
334 					   DBR_DBR_TYPE_SRQ_ARMENA);
335 			if (!nq->srqn_handler(nq,
336 					      (struct bnxt_qplib_srq *)q_handle,
337 					      nqsrqe->event))
338 				num_srqne_processed++;
339 			else
340 				dev_warn(&nq->pdev->dev,
341 					 "QPLIB: SRQ event 0x%x not handled",
342 					 nqsrqe->event);
343 			break;
344 		}
345 		case NQ_BASE_TYPE_DBQ_EVENT:
346 			break;
347 		default:
348 			dev_warn(&nq->pdev->dev,
349 				 "QPLIB: nqe with type = 0x%x not handled",
350 				 type);
351 			break;
352 		}
353 		raw_cons++;
354 	}
355 	if (hwq->cons != raw_cons) {
356 		hwq->cons = raw_cons;
357 		NQ_DB_REARM(nq->bar_reg_iomem, hwq->cons, hwq->max_elements);
358 	}
359 }
360 
361 static irqreturn_t bnxt_qplib_nq_irq(int irq, void *dev_instance)
362 {
363 	struct bnxt_qplib_nq *nq = dev_instance;
364 	struct bnxt_qplib_hwq *hwq = &nq->hwq;
365 	struct nq_base **nq_ptr;
366 	u32 sw_cons;
367 
368 	/* Prefetch the NQ element */
369 	sw_cons = HWQ_CMP(hwq->cons, hwq);
370 	nq_ptr = (struct nq_base **)nq->hwq.pbl_ptr;
371 	prefetch(&nq_ptr[NQE_PG(sw_cons)][NQE_IDX(sw_cons)]);
372 
373 	/* Fan out to CPU affinitized kthreads? */
374 	tasklet_schedule(&nq->worker);
375 
376 	return IRQ_HANDLED;
377 }
378 
379 void bnxt_qplib_disable_nq(struct bnxt_qplib_nq *nq)
380 {
381 	if (nq->cqn_wq) {
382 		destroy_workqueue(nq->cqn_wq);
383 		nq->cqn_wq = NULL;
384 	}
385 	/* Make sure the HW is stopped! */
386 	synchronize_irq(nq->vector);
387 	tasklet_disable(&nq->worker);
388 	tasklet_kill(&nq->worker);
389 
390 	if (nq->requested) {
391 		irq_set_affinity_hint(nq->vector, NULL);
392 		free_irq(nq->vector, nq);
393 		nq->requested = false;
394 	}
395 	if (nq->bar_reg_iomem)
396 		iounmap(nq->bar_reg_iomem);
397 	nq->bar_reg_iomem = NULL;
398 
399 	nq->cqn_handler = NULL;
400 	nq->srqn_handler = NULL;
401 	nq->vector = 0;
402 }
403 
404 int bnxt_qplib_enable_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq,
405 			 int nq_idx, int msix_vector, int bar_reg_offset,
406 			 int (*cqn_handler)(struct bnxt_qplib_nq *nq,
407 					    struct bnxt_qplib_cq *),
408 			 int (*srqn_handler)(struct bnxt_qplib_nq *nq,
409 					     struct bnxt_qplib_srq *,
410 					     u8 event))
411 {
412 	resource_size_t nq_base;
413 	int rc = -1;
414 
415 	nq->pdev = pdev;
416 	nq->vector = msix_vector;
417 	if (cqn_handler)
418 		nq->cqn_handler = cqn_handler;
419 
420 	if (srqn_handler)
421 		nq->srqn_handler = srqn_handler;
422 
423 	tasklet_init(&nq->worker, bnxt_qplib_service_nq, (unsigned long)nq);
424 
425 	/* Have a task to schedule CQ notifiers in post send case */
426 	nq->cqn_wq  = create_singlethread_workqueue("bnxt_qplib_nq");
427 	if (!nq->cqn_wq)
428 		goto fail;
429 
430 	nq->requested = false;
431 	memset(nq->name, 0, 32);
432 	sprintf(nq->name, "bnxt_qplib_nq-%d", nq_idx);
433 	rc = request_irq(nq->vector, bnxt_qplib_nq_irq, 0, nq->name, nq);
434 	if (rc) {
435 		dev_err(&nq->pdev->dev,
436 			"Failed to request IRQ for NQ: %#x", rc);
437 		goto fail;
438 	}
439 
440 	cpumask_clear(&nq->mask);
441 	cpumask_set_cpu(nq_idx, &nq->mask);
442 	rc = irq_set_affinity_hint(nq->vector, &nq->mask);
443 	if (rc) {
444 		dev_warn(&nq->pdev->dev,
445 			 "QPLIB: set affinity failed; vector: %d nq_idx: %d\n",
446 			 nq->vector, nq_idx);
447 	}
448 
449 	nq->requested = true;
450 	nq->bar_reg = NQ_CONS_PCI_BAR_REGION;
451 	nq->bar_reg_off = bar_reg_offset;
452 	nq_base = pci_resource_start(pdev, nq->bar_reg);
453 	if (!nq_base) {
454 		rc = -ENOMEM;
455 		goto fail;
456 	}
457 	nq->bar_reg_iomem = ioremap_nocache(nq_base + nq->bar_reg_off, 4);
458 	if (!nq->bar_reg_iomem) {
459 		rc = -ENOMEM;
460 		goto fail;
461 	}
462 	NQ_DB_REARM(nq->bar_reg_iomem, nq->hwq.cons, nq->hwq.max_elements);
463 
464 	return 0;
465 fail:
466 	bnxt_qplib_disable_nq(nq);
467 	return rc;
468 }
469 
470 void bnxt_qplib_free_nq(struct bnxt_qplib_nq *nq)
471 {
472 	if (nq->hwq.max_elements) {
473 		bnxt_qplib_free_hwq(nq->pdev, &nq->hwq);
474 		nq->hwq.max_elements = 0;
475 	}
476 }
477 
478 int bnxt_qplib_alloc_nq(struct pci_dev *pdev, struct bnxt_qplib_nq *nq)
479 {
480 	nq->pdev = pdev;
481 	if (!nq->hwq.max_elements ||
482 	    nq->hwq.max_elements > BNXT_QPLIB_NQE_MAX_CNT)
483 		nq->hwq.max_elements = BNXT_QPLIB_NQE_MAX_CNT;
484 
485 	if (bnxt_qplib_alloc_init_hwq(nq->pdev, &nq->hwq, NULL, 0,
486 				      &nq->hwq.max_elements,
487 				      BNXT_QPLIB_MAX_NQE_ENTRY_SIZE, 0,
488 				      PAGE_SIZE, HWQ_TYPE_L2_CMPL))
489 		return -ENOMEM;
490 
491 	nq->budget = 8;
492 	return 0;
493 }
494 
495 /* SRQ */
496 static void bnxt_qplib_arm_srq(struct bnxt_qplib_srq *srq, u32 arm_type)
497 {
498 	struct bnxt_qplib_hwq *srq_hwq = &srq->hwq;
499 	struct dbr_dbr db_msg = { 0 };
500 	void __iomem *db;
501 	u32 sw_prod = 0;
502 
503 	/* Ring DB */
504 	sw_prod = (arm_type == DBR_DBR_TYPE_SRQ_ARM) ? srq->threshold :
505 		   HWQ_CMP(srq_hwq->prod, srq_hwq);
506 	db_msg.index = cpu_to_le32((sw_prod << DBR_DBR_INDEX_SFT) &
507 				   DBR_DBR_INDEX_MASK);
508 	db_msg.type_xid = cpu_to_le32(((srq->id << DBR_DBR_XID_SFT) &
509 					DBR_DBR_XID_MASK) | arm_type);
510 	db = (arm_type == DBR_DBR_TYPE_SRQ_ARMENA) ?
511 		srq->dbr_base : srq->dpi->dbr;
512 	wmb(); /* barrier before db ring */
513 	__iowrite64_copy(db, &db_msg, sizeof(db_msg) / sizeof(u64));
514 }
515 
516 int bnxt_qplib_destroy_srq(struct bnxt_qplib_res *res,
517 			   struct bnxt_qplib_srq *srq)
518 {
519 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
520 	struct cmdq_destroy_srq req;
521 	struct creq_destroy_srq_resp resp;
522 	u16 cmd_flags = 0;
523 	int rc;
524 
525 	RCFW_CMD_PREP(req, DESTROY_SRQ, cmd_flags);
526 
527 	/* Configure the request */
528 	req.srq_cid = cpu_to_le32(srq->id);
529 
530 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
531 					  (void *)&resp, NULL, 0);
532 	if (rc)
533 		return rc;
534 
535 	bnxt_qplib_free_hwq(res->pdev, &srq->hwq);
536 	kfree(srq->swq);
537 	return 0;
538 }
539 
540 int bnxt_qplib_create_srq(struct bnxt_qplib_res *res,
541 			  struct bnxt_qplib_srq *srq)
542 {
543 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
544 	struct cmdq_create_srq req;
545 	struct creq_create_srq_resp resp;
546 	struct bnxt_qplib_pbl *pbl;
547 	u16 cmd_flags = 0;
548 	int rc, idx;
549 
550 	srq->hwq.max_elements = srq->max_wqe;
551 	rc = bnxt_qplib_alloc_init_hwq(res->pdev, &srq->hwq, srq->sglist,
552 				       srq->nmap, &srq->hwq.max_elements,
553 				       BNXT_QPLIB_MAX_RQE_ENTRY_SIZE, 0,
554 				       PAGE_SIZE, HWQ_TYPE_QUEUE);
555 	if (rc)
556 		goto exit;
557 
558 	srq->swq = kcalloc(srq->hwq.max_elements, sizeof(*srq->swq),
559 			   GFP_KERNEL);
560 	if (!srq->swq)
561 		goto fail;
562 
563 	RCFW_CMD_PREP(req, CREATE_SRQ, cmd_flags);
564 
565 	/* Configure the request */
566 	req.dpi = cpu_to_le32(srq->dpi->dpi);
567 	req.srq_handle = cpu_to_le64(srq);
568 
569 	req.srq_size = cpu_to_le16((u16)srq->hwq.max_elements);
570 	pbl = &srq->hwq.pbl[PBL_LVL_0];
571 	req.pg_size_lvl = cpu_to_le16((((u16)srq->hwq.level &
572 				      CMDQ_CREATE_SRQ_LVL_MASK) <<
573 				      CMDQ_CREATE_SRQ_LVL_SFT) |
574 				      (pbl->pg_size == ROCE_PG_SIZE_4K ?
575 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_4K :
576 				       pbl->pg_size == ROCE_PG_SIZE_8K ?
577 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_8K :
578 				       pbl->pg_size == ROCE_PG_SIZE_64K ?
579 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_64K :
580 				       pbl->pg_size == ROCE_PG_SIZE_2M ?
581 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_2M :
582 				       pbl->pg_size == ROCE_PG_SIZE_8M ?
583 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_8M :
584 				       pbl->pg_size == ROCE_PG_SIZE_1G ?
585 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_1G :
586 				       CMDQ_CREATE_SRQ_PG_SIZE_PG_4K));
587 	req.pbl = cpu_to_le64(pbl->pg_map_arr[0]);
588 	req.pd_id = cpu_to_le32(srq->pd->id);
589 	req.eventq_id = cpu_to_le16(srq->eventq_hw_ring_id);
590 
591 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
592 					  (void *)&resp, NULL, 0);
593 	if (rc)
594 		goto fail;
595 
596 	spin_lock_init(&srq->lock);
597 	srq->start_idx = 0;
598 	srq->last_idx = srq->hwq.max_elements - 1;
599 	for (idx = 0; idx < srq->hwq.max_elements; idx++)
600 		srq->swq[idx].next_idx = idx + 1;
601 	srq->swq[srq->last_idx].next_idx = -1;
602 
603 	srq->id = le32_to_cpu(resp.xid);
604 	srq->dbr_base = res->dpi_tbl.dbr_bar_reg_iomem;
605 	if (srq->threshold)
606 		bnxt_qplib_arm_srq(srq, DBR_DBR_TYPE_SRQ_ARMENA);
607 	srq->arm_req = false;
608 
609 	return 0;
610 fail:
611 	bnxt_qplib_free_hwq(res->pdev, &srq->hwq);
612 	kfree(srq->swq);
613 exit:
614 	return rc;
615 }
616 
617 int bnxt_qplib_modify_srq(struct bnxt_qplib_res *res,
618 			  struct bnxt_qplib_srq *srq)
619 {
620 	struct bnxt_qplib_hwq *srq_hwq = &srq->hwq;
621 	u32 sw_prod, sw_cons, count = 0;
622 
623 	sw_prod = HWQ_CMP(srq_hwq->prod, srq_hwq);
624 	sw_cons = HWQ_CMP(srq_hwq->cons, srq_hwq);
625 
626 	count = sw_prod > sw_cons ? sw_prod - sw_cons :
627 				    srq_hwq->max_elements - sw_cons + sw_prod;
628 	if (count > srq->threshold) {
629 		srq->arm_req = false;
630 		bnxt_qplib_arm_srq(srq, DBR_DBR_TYPE_SRQ_ARM);
631 	} else {
632 		/* Deferred arming */
633 		srq->arm_req = true;
634 	}
635 
636 	return 0;
637 }
638 
639 int bnxt_qplib_query_srq(struct bnxt_qplib_res *res,
640 			 struct bnxt_qplib_srq *srq)
641 {
642 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
643 	struct cmdq_query_srq req;
644 	struct creq_query_srq_resp resp;
645 	struct bnxt_qplib_rcfw_sbuf *sbuf;
646 	struct creq_query_srq_resp_sb *sb;
647 	u16 cmd_flags = 0;
648 	int rc = 0;
649 
650 	RCFW_CMD_PREP(req, QUERY_SRQ, cmd_flags);
651 	req.srq_cid = cpu_to_le32(srq->id);
652 
653 	/* Configure the request */
654 	sbuf = bnxt_qplib_rcfw_alloc_sbuf(rcfw, sizeof(*sb));
655 	if (!sbuf)
656 		return -ENOMEM;
657 	sb = sbuf->sb;
658 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req, (void *)&resp,
659 					  (void *)sbuf, 0);
660 	srq->threshold = le16_to_cpu(sb->srq_limit);
661 	bnxt_qplib_rcfw_free_sbuf(rcfw, sbuf);
662 
663 	return rc;
664 }
665 
666 int bnxt_qplib_post_srq_recv(struct bnxt_qplib_srq *srq,
667 			     struct bnxt_qplib_swqe *wqe)
668 {
669 	struct bnxt_qplib_hwq *srq_hwq = &srq->hwq;
670 	struct rq_wqe *srqe, **srqe_ptr;
671 	struct sq_sge *hw_sge;
672 	u32 sw_prod, sw_cons, count = 0;
673 	int i, rc = 0, next;
674 
675 	spin_lock(&srq_hwq->lock);
676 	if (srq->start_idx == srq->last_idx) {
677 		dev_err(&srq_hwq->pdev->dev, "QPLIB: FP: SRQ (0x%x) is full!",
678 			srq->id);
679 		rc = -EINVAL;
680 		spin_unlock(&srq_hwq->lock);
681 		goto done;
682 	}
683 	next = srq->start_idx;
684 	srq->start_idx = srq->swq[next].next_idx;
685 	spin_unlock(&srq_hwq->lock);
686 
687 	sw_prod = HWQ_CMP(srq_hwq->prod, srq_hwq);
688 	srqe_ptr = (struct rq_wqe **)srq_hwq->pbl_ptr;
689 	srqe = &srqe_ptr[RQE_PG(sw_prod)][RQE_IDX(sw_prod)];
690 	memset(srqe, 0, BNXT_QPLIB_MAX_RQE_ENTRY_SIZE);
691 	/* Calculate wqe_size16 and data_len */
692 	for (i = 0, hw_sge = (struct sq_sge *)srqe->data;
693 	     i < wqe->num_sge; i++, hw_sge++) {
694 		hw_sge->va_or_pa = cpu_to_le64(wqe->sg_list[i].addr);
695 		hw_sge->l_key = cpu_to_le32(wqe->sg_list[i].lkey);
696 		hw_sge->size = cpu_to_le32(wqe->sg_list[i].size);
697 	}
698 	srqe->wqe_type = wqe->type;
699 	srqe->flags = wqe->flags;
700 	srqe->wqe_size = wqe->num_sge +
701 			((offsetof(typeof(*srqe), data) + 15) >> 4);
702 	srqe->wr_id[0] = cpu_to_le32((u32)next);
703 	srq->swq[next].wr_id = wqe->wr_id;
704 
705 	srq_hwq->prod++;
706 
707 	spin_lock(&srq_hwq->lock);
708 	sw_prod = HWQ_CMP(srq_hwq->prod, srq_hwq);
709 	/* retaining srq_hwq->cons for this logic
710 	 * actually the lock is only required to
711 	 * read srq_hwq->cons.
712 	 */
713 	sw_cons = HWQ_CMP(srq_hwq->cons, srq_hwq);
714 	count = sw_prod > sw_cons ? sw_prod - sw_cons :
715 				    srq_hwq->max_elements - sw_cons + sw_prod;
716 	spin_unlock(&srq_hwq->lock);
717 	/* Ring DB */
718 	bnxt_qplib_arm_srq(srq, DBR_DBR_TYPE_SRQ);
719 	if (srq->arm_req == true && count > srq->threshold) {
720 		srq->arm_req = false;
721 		bnxt_qplib_arm_srq(srq, DBR_DBR_TYPE_SRQ_ARM);
722 	}
723 done:
724 	return rc;
725 }
726 
727 /* QP */
728 int bnxt_qplib_create_qp1(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
729 {
730 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
731 	struct cmdq_create_qp1 req;
732 	struct creq_create_qp1_resp resp;
733 	struct bnxt_qplib_pbl *pbl;
734 	struct bnxt_qplib_q *sq = &qp->sq;
735 	struct bnxt_qplib_q *rq = &qp->rq;
736 	int rc;
737 	u16 cmd_flags = 0;
738 	u32 qp_flags = 0;
739 
740 	RCFW_CMD_PREP(req, CREATE_QP1, cmd_flags);
741 
742 	/* General */
743 	req.type = qp->type;
744 	req.dpi = cpu_to_le32(qp->dpi->dpi);
745 	req.qp_handle = cpu_to_le64(qp->qp_handle);
746 
747 	/* SQ */
748 	sq->hwq.max_elements = sq->max_wqe;
749 	rc = bnxt_qplib_alloc_init_hwq(res->pdev, &sq->hwq, NULL, 0,
750 				       &sq->hwq.max_elements,
751 				       BNXT_QPLIB_MAX_SQE_ENTRY_SIZE, 0,
752 				       PAGE_SIZE, HWQ_TYPE_QUEUE);
753 	if (rc)
754 		goto exit;
755 
756 	sq->swq = kcalloc(sq->hwq.max_elements, sizeof(*sq->swq), GFP_KERNEL);
757 	if (!sq->swq) {
758 		rc = -ENOMEM;
759 		goto fail_sq;
760 	}
761 	pbl = &sq->hwq.pbl[PBL_LVL_0];
762 	req.sq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
763 	req.sq_pg_size_sq_lvl =
764 		((sq->hwq.level & CMDQ_CREATE_QP1_SQ_LVL_MASK)
765 				<<  CMDQ_CREATE_QP1_SQ_LVL_SFT) |
766 		(pbl->pg_size == ROCE_PG_SIZE_4K ?
767 				CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_4K :
768 		 pbl->pg_size == ROCE_PG_SIZE_8K ?
769 				CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_8K :
770 		 pbl->pg_size == ROCE_PG_SIZE_64K ?
771 				CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_64K :
772 		 pbl->pg_size == ROCE_PG_SIZE_2M ?
773 				CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_2M :
774 		 pbl->pg_size == ROCE_PG_SIZE_8M ?
775 				CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_8M :
776 		 pbl->pg_size == ROCE_PG_SIZE_1G ?
777 				CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_1G :
778 		 CMDQ_CREATE_QP1_SQ_PG_SIZE_PG_4K);
779 
780 	if (qp->scq)
781 		req.scq_cid = cpu_to_le32(qp->scq->id);
782 
783 	qp_flags |= CMDQ_CREATE_QP1_QP_FLAGS_RESERVED_LKEY_ENABLE;
784 
785 	/* RQ */
786 	if (rq->max_wqe) {
787 		rq->hwq.max_elements = qp->rq.max_wqe;
788 		rc = bnxt_qplib_alloc_init_hwq(res->pdev, &rq->hwq, NULL, 0,
789 					       &rq->hwq.max_elements,
790 					       BNXT_QPLIB_MAX_RQE_ENTRY_SIZE, 0,
791 					       PAGE_SIZE, HWQ_TYPE_QUEUE);
792 		if (rc)
793 			goto fail_sq;
794 
795 		rq->swq = kcalloc(rq->hwq.max_elements, sizeof(*rq->swq),
796 				  GFP_KERNEL);
797 		if (!rq->swq) {
798 			rc = -ENOMEM;
799 			goto fail_rq;
800 		}
801 		pbl = &rq->hwq.pbl[PBL_LVL_0];
802 		req.rq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
803 		req.rq_pg_size_rq_lvl =
804 			((rq->hwq.level & CMDQ_CREATE_QP1_RQ_LVL_MASK) <<
805 			 CMDQ_CREATE_QP1_RQ_LVL_SFT) |
806 				(pbl->pg_size == ROCE_PG_SIZE_4K ?
807 					CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_4K :
808 				 pbl->pg_size == ROCE_PG_SIZE_8K ?
809 					CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_8K :
810 				 pbl->pg_size == ROCE_PG_SIZE_64K ?
811 					CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_64K :
812 				 pbl->pg_size == ROCE_PG_SIZE_2M ?
813 					CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_2M :
814 				 pbl->pg_size == ROCE_PG_SIZE_8M ?
815 					CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_8M :
816 				 pbl->pg_size == ROCE_PG_SIZE_1G ?
817 					CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_1G :
818 				 CMDQ_CREATE_QP1_RQ_PG_SIZE_PG_4K);
819 		if (qp->rcq)
820 			req.rcq_cid = cpu_to_le32(qp->rcq->id);
821 	}
822 
823 	/* Header buffer - allow hdr_buf pass in */
824 	rc = bnxt_qplib_alloc_qp_hdr_buf(res, qp);
825 	if (rc) {
826 		rc = -ENOMEM;
827 		goto fail;
828 	}
829 	req.qp_flags = cpu_to_le32(qp_flags);
830 	req.sq_size = cpu_to_le32(sq->hwq.max_elements);
831 	req.rq_size = cpu_to_le32(rq->hwq.max_elements);
832 
833 	req.sq_fwo_sq_sge =
834 		cpu_to_le16((sq->max_sge & CMDQ_CREATE_QP1_SQ_SGE_MASK) <<
835 			    CMDQ_CREATE_QP1_SQ_SGE_SFT);
836 	req.rq_fwo_rq_sge =
837 		cpu_to_le16((rq->max_sge & CMDQ_CREATE_QP1_RQ_SGE_MASK) <<
838 			    CMDQ_CREATE_QP1_RQ_SGE_SFT);
839 
840 	req.pd_id = cpu_to_le32(qp->pd->id);
841 
842 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
843 					  (void *)&resp, NULL, 0);
844 	if (rc)
845 		goto fail;
846 
847 	qp->id = le32_to_cpu(resp.xid);
848 	qp->cur_qp_state = CMDQ_MODIFY_QP_NEW_STATE_RESET;
849 	rcfw->qp_tbl[qp->id].qp_id = qp->id;
850 	rcfw->qp_tbl[qp->id].qp_handle = (void *)qp;
851 
852 	return 0;
853 
854 fail:
855 	bnxt_qplib_free_qp_hdr_buf(res, qp);
856 fail_rq:
857 	bnxt_qplib_free_hwq(res->pdev, &rq->hwq);
858 	kfree(rq->swq);
859 fail_sq:
860 	bnxt_qplib_free_hwq(res->pdev, &sq->hwq);
861 	kfree(sq->swq);
862 exit:
863 	return rc;
864 }
865 
866 int bnxt_qplib_create_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
867 {
868 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
869 	struct sq_send *hw_sq_send_hdr, **hw_sq_send_ptr;
870 	struct cmdq_create_qp req;
871 	struct creq_create_qp_resp resp;
872 	struct bnxt_qplib_pbl *pbl;
873 	struct sq_psn_search **psn_search_ptr;
874 	unsigned long int psn_search, poff = 0;
875 	struct bnxt_qplib_q *sq = &qp->sq;
876 	struct bnxt_qplib_q *rq = &qp->rq;
877 	struct bnxt_qplib_hwq *xrrq;
878 	int i, rc, req_size, psn_sz;
879 	u16 cmd_flags = 0, max_ssge;
880 	u32 sw_prod, qp_flags = 0;
881 
882 	RCFW_CMD_PREP(req, CREATE_QP, cmd_flags);
883 
884 	/* General */
885 	req.type = qp->type;
886 	req.dpi = cpu_to_le32(qp->dpi->dpi);
887 	req.qp_handle = cpu_to_le64(qp->qp_handle);
888 
889 	/* SQ */
890 	psn_sz = (qp->type == CMDQ_CREATE_QP_TYPE_RC) ?
891 		 sizeof(struct sq_psn_search) : 0;
892 	sq->hwq.max_elements = sq->max_wqe;
893 	rc = bnxt_qplib_alloc_init_hwq(res->pdev, &sq->hwq, sq->sglist,
894 				       sq->nmap, &sq->hwq.max_elements,
895 				       BNXT_QPLIB_MAX_SQE_ENTRY_SIZE,
896 				       psn_sz,
897 				       PAGE_SIZE, HWQ_TYPE_QUEUE);
898 	if (rc)
899 		goto exit;
900 
901 	sq->swq = kcalloc(sq->hwq.max_elements, sizeof(*sq->swq), GFP_KERNEL);
902 	if (!sq->swq) {
903 		rc = -ENOMEM;
904 		goto fail_sq;
905 	}
906 	hw_sq_send_ptr = (struct sq_send **)sq->hwq.pbl_ptr;
907 	if (psn_sz) {
908 		psn_search_ptr = (struct sq_psn_search **)
909 				  &hw_sq_send_ptr[get_sqe_pg
910 					(sq->hwq.max_elements)];
911 		psn_search = (unsigned long int)
912 			      &hw_sq_send_ptr[get_sqe_pg(sq->hwq.max_elements)]
913 			      [get_sqe_idx(sq->hwq.max_elements)];
914 		if (psn_search & ~PAGE_MASK) {
915 			/* If the psn_search does not start on a page boundary,
916 			 * then calculate the offset
917 			 */
918 			poff = (psn_search & ~PAGE_MASK) /
919 				BNXT_QPLIB_MAX_PSNE_ENTRY_SIZE;
920 		}
921 		for (i = 0; i < sq->hwq.max_elements; i++)
922 			sq->swq[i].psn_search =
923 				&psn_search_ptr[get_psne_pg(i + poff)]
924 					       [get_psne_idx(i + poff)];
925 	}
926 	pbl = &sq->hwq.pbl[PBL_LVL_0];
927 	req.sq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
928 	req.sq_pg_size_sq_lvl =
929 		((sq->hwq.level & CMDQ_CREATE_QP_SQ_LVL_MASK)
930 				 <<  CMDQ_CREATE_QP_SQ_LVL_SFT) |
931 		(pbl->pg_size == ROCE_PG_SIZE_4K ?
932 				CMDQ_CREATE_QP_SQ_PG_SIZE_PG_4K :
933 		 pbl->pg_size == ROCE_PG_SIZE_8K ?
934 				CMDQ_CREATE_QP_SQ_PG_SIZE_PG_8K :
935 		 pbl->pg_size == ROCE_PG_SIZE_64K ?
936 				CMDQ_CREATE_QP_SQ_PG_SIZE_PG_64K :
937 		 pbl->pg_size == ROCE_PG_SIZE_2M ?
938 				CMDQ_CREATE_QP_SQ_PG_SIZE_PG_2M :
939 		 pbl->pg_size == ROCE_PG_SIZE_8M ?
940 				CMDQ_CREATE_QP_SQ_PG_SIZE_PG_8M :
941 		 pbl->pg_size == ROCE_PG_SIZE_1G ?
942 				CMDQ_CREATE_QP_SQ_PG_SIZE_PG_1G :
943 		 CMDQ_CREATE_QP_SQ_PG_SIZE_PG_4K);
944 
945 	/* initialize all SQ WQEs to LOCAL_INVALID (sq prep for hw fetch) */
946 	hw_sq_send_ptr = (struct sq_send **)sq->hwq.pbl_ptr;
947 	for (sw_prod = 0; sw_prod < sq->hwq.max_elements; sw_prod++) {
948 		hw_sq_send_hdr = &hw_sq_send_ptr[get_sqe_pg(sw_prod)]
949 						[get_sqe_idx(sw_prod)];
950 		hw_sq_send_hdr->wqe_type = SQ_BASE_WQE_TYPE_LOCAL_INVALID;
951 	}
952 
953 	if (qp->scq)
954 		req.scq_cid = cpu_to_le32(qp->scq->id);
955 
956 	qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_RESERVED_LKEY_ENABLE;
957 	qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_FR_PMR_ENABLED;
958 	if (qp->sig_type)
959 		qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_FORCE_COMPLETION;
960 
961 	/* RQ */
962 	if (rq->max_wqe) {
963 		rq->hwq.max_elements = rq->max_wqe;
964 		rc = bnxt_qplib_alloc_init_hwq(res->pdev, &rq->hwq, rq->sglist,
965 					       rq->nmap, &rq->hwq.max_elements,
966 					       BNXT_QPLIB_MAX_RQE_ENTRY_SIZE, 0,
967 					       PAGE_SIZE, HWQ_TYPE_QUEUE);
968 		if (rc)
969 			goto fail_sq;
970 
971 		rq->swq = kcalloc(rq->hwq.max_elements, sizeof(*rq->swq),
972 				  GFP_KERNEL);
973 		if (!rq->swq) {
974 			rc = -ENOMEM;
975 			goto fail_rq;
976 		}
977 		pbl = &rq->hwq.pbl[PBL_LVL_0];
978 		req.rq_pbl = cpu_to_le64(pbl->pg_map_arr[0]);
979 		req.rq_pg_size_rq_lvl =
980 			((rq->hwq.level & CMDQ_CREATE_QP_RQ_LVL_MASK) <<
981 			 CMDQ_CREATE_QP_RQ_LVL_SFT) |
982 				(pbl->pg_size == ROCE_PG_SIZE_4K ?
983 					CMDQ_CREATE_QP_RQ_PG_SIZE_PG_4K :
984 				 pbl->pg_size == ROCE_PG_SIZE_8K ?
985 					CMDQ_CREATE_QP_RQ_PG_SIZE_PG_8K :
986 				 pbl->pg_size == ROCE_PG_SIZE_64K ?
987 					CMDQ_CREATE_QP_RQ_PG_SIZE_PG_64K :
988 				 pbl->pg_size == ROCE_PG_SIZE_2M ?
989 					CMDQ_CREATE_QP_RQ_PG_SIZE_PG_2M :
990 				 pbl->pg_size == ROCE_PG_SIZE_8M ?
991 					CMDQ_CREATE_QP_RQ_PG_SIZE_PG_8M :
992 				 pbl->pg_size == ROCE_PG_SIZE_1G ?
993 					CMDQ_CREATE_QP_RQ_PG_SIZE_PG_1G :
994 				 CMDQ_CREATE_QP_RQ_PG_SIZE_PG_4K);
995 	} else {
996 		/* SRQ */
997 		if (qp->srq) {
998 			qp_flags |= CMDQ_CREATE_QP_QP_FLAGS_SRQ_USED;
999 			req.srq_cid = cpu_to_le32(qp->srq->id);
1000 		}
1001 	}
1002 
1003 	if (qp->rcq)
1004 		req.rcq_cid = cpu_to_le32(qp->rcq->id);
1005 	req.qp_flags = cpu_to_le32(qp_flags);
1006 	req.sq_size = cpu_to_le32(sq->hwq.max_elements);
1007 	req.rq_size = cpu_to_le32(rq->hwq.max_elements);
1008 	qp->sq_hdr_buf = NULL;
1009 	qp->rq_hdr_buf = NULL;
1010 
1011 	rc = bnxt_qplib_alloc_qp_hdr_buf(res, qp);
1012 	if (rc)
1013 		goto fail_rq;
1014 
1015 	/* CTRL-22434: Irrespective of the requested SGE count on the SQ
1016 	 * always create the QP with max send sges possible if the requested
1017 	 * inline size is greater than 0.
1018 	 */
1019 	max_ssge = qp->max_inline_data ? 6 : sq->max_sge;
1020 	req.sq_fwo_sq_sge = cpu_to_le16(
1021 				((max_ssge & CMDQ_CREATE_QP_SQ_SGE_MASK)
1022 				 << CMDQ_CREATE_QP_SQ_SGE_SFT) | 0);
1023 	req.rq_fwo_rq_sge = cpu_to_le16(
1024 				((rq->max_sge & CMDQ_CREATE_QP_RQ_SGE_MASK)
1025 				 << CMDQ_CREATE_QP_RQ_SGE_SFT) | 0);
1026 	/* ORRQ and IRRQ */
1027 	if (psn_sz) {
1028 		xrrq = &qp->orrq;
1029 		xrrq->max_elements =
1030 			ORD_LIMIT_TO_ORRQ_SLOTS(qp->max_rd_atomic);
1031 		req_size = xrrq->max_elements *
1032 			   BNXT_QPLIB_MAX_ORRQE_ENTRY_SIZE + PAGE_SIZE - 1;
1033 		req_size &= ~(PAGE_SIZE - 1);
1034 		rc = bnxt_qplib_alloc_init_hwq(res->pdev, xrrq, NULL, 0,
1035 					       &xrrq->max_elements,
1036 					       BNXT_QPLIB_MAX_ORRQE_ENTRY_SIZE,
1037 					       0, req_size, HWQ_TYPE_CTX);
1038 		if (rc)
1039 			goto fail_buf_free;
1040 		pbl = &xrrq->pbl[PBL_LVL_0];
1041 		req.orrq_addr = cpu_to_le64(pbl->pg_map_arr[0]);
1042 
1043 		xrrq = &qp->irrq;
1044 		xrrq->max_elements = IRD_LIMIT_TO_IRRQ_SLOTS(
1045 						qp->max_dest_rd_atomic);
1046 		req_size = xrrq->max_elements *
1047 			   BNXT_QPLIB_MAX_IRRQE_ENTRY_SIZE + PAGE_SIZE - 1;
1048 		req_size &= ~(PAGE_SIZE - 1);
1049 
1050 		rc = bnxt_qplib_alloc_init_hwq(res->pdev, xrrq, NULL, 0,
1051 					       &xrrq->max_elements,
1052 					       BNXT_QPLIB_MAX_IRRQE_ENTRY_SIZE,
1053 					       0, req_size, HWQ_TYPE_CTX);
1054 		if (rc)
1055 			goto fail_orrq;
1056 
1057 		pbl = &xrrq->pbl[PBL_LVL_0];
1058 		req.irrq_addr = cpu_to_le64(pbl->pg_map_arr[0]);
1059 	}
1060 	req.pd_id = cpu_to_le32(qp->pd->id);
1061 
1062 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
1063 					  (void *)&resp, NULL, 0);
1064 	if (rc)
1065 		goto fail;
1066 
1067 	qp->id = le32_to_cpu(resp.xid);
1068 	qp->cur_qp_state = CMDQ_MODIFY_QP_NEW_STATE_RESET;
1069 	INIT_LIST_HEAD(&qp->sq_flush);
1070 	INIT_LIST_HEAD(&qp->rq_flush);
1071 	rcfw->qp_tbl[qp->id].qp_id = qp->id;
1072 	rcfw->qp_tbl[qp->id].qp_handle = (void *)qp;
1073 
1074 	return 0;
1075 
1076 fail:
1077 	if (qp->irrq.max_elements)
1078 		bnxt_qplib_free_hwq(res->pdev, &qp->irrq);
1079 fail_orrq:
1080 	if (qp->orrq.max_elements)
1081 		bnxt_qplib_free_hwq(res->pdev, &qp->orrq);
1082 fail_buf_free:
1083 	bnxt_qplib_free_qp_hdr_buf(res, qp);
1084 fail_rq:
1085 	bnxt_qplib_free_hwq(res->pdev, &rq->hwq);
1086 	kfree(rq->swq);
1087 fail_sq:
1088 	bnxt_qplib_free_hwq(res->pdev, &sq->hwq);
1089 	kfree(sq->swq);
1090 exit:
1091 	return rc;
1092 }
1093 
1094 static void __modify_flags_from_init_state(struct bnxt_qplib_qp *qp)
1095 {
1096 	switch (qp->state) {
1097 	case CMDQ_MODIFY_QP_NEW_STATE_RTR:
1098 		/* INIT->RTR, configure the path_mtu to the default
1099 		 * 2048 if not being requested
1100 		 */
1101 		if (!(qp->modify_flags &
1102 		    CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU)) {
1103 			qp->modify_flags |=
1104 				CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU;
1105 			qp->path_mtu =
1106 				CMDQ_MODIFY_QP_PATH_MTU_MTU_2048;
1107 		}
1108 		qp->modify_flags &=
1109 			~CMDQ_MODIFY_QP_MODIFY_MASK_VLAN_ID;
1110 		/* Bono FW require the max_dest_rd_atomic to be >= 1 */
1111 		if (qp->max_dest_rd_atomic < 1)
1112 			qp->max_dest_rd_atomic = 1;
1113 		qp->modify_flags &= ~CMDQ_MODIFY_QP_MODIFY_MASK_SRC_MAC;
1114 		/* Bono FW 20.6.5 requires SGID_INDEX configuration */
1115 		if (!(qp->modify_flags &
1116 		    CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX)) {
1117 			qp->modify_flags |=
1118 				CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX;
1119 			qp->ah.sgid_index = 0;
1120 		}
1121 		break;
1122 	default:
1123 		break;
1124 	}
1125 }
1126 
1127 static void __modify_flags_from_rtr_state(struct bnxt_qplib_qp *qp)
1128 {
1129 	switch (qp->state) {
1130 	case CMDQ_MODIFY_QP_NEW_STATE_RTS:
1131 		/* Bono FW requires the max_rd_atomic to be >= 1 */
1132 		if (qp->max_rd_atomic < 1)
1133 			qp->max_rd_atomic = 1;
1134 		/* Bono FW does not allow PKEY_INDEX,
1135 		 * DGID, FLOW_LABEL, SGID_INDEX, HOP_LIMIT,
1136 		 * TRAFFIC_CLASS, DEST_MAC, PATH_MTU, RQ_PSN,
1137 		 * MIN_RNR_TIMER, MAX_DEST_RD_ATOMIC, DEST_QP_ID
1138 		 * modification
1139 		 */
1140 		qp->modify_flags &=
1141 			~(CMDQ_MODIFY_QP_MODIFY_MASK_PKEY |
1142 			  CMDQ_MODIFY_QP_MODIFY_MASK_DGID |
1143 			  CMDQ_MODIFY_QP_MODIFY_MASK_FLOW_LABEL |
1144 			  CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX |
1145 			  CMDQ_MODIFY_QP_MODIFY_MASK_HOP_LIMIT |
1146 			  CMDQ_MODIFY_QP_MODIFY_MASK_TRAFFIC_CLASS |
1147 			  CMDQ_MODIFY_QP_MODIFY_MASK_DEST_MAC |
1148 			  CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU |
1149 			  CMDQ_MODIFY_QP_MODIFY_MASK_RQ_PSN |
1150 			  CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER |
1151 			  CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC |
1152 			  CMDQ_MODIFY_QP_MODIFY_MASK_DEST_QP_ID);
1153 		break;
1154 	default:
1155 		break;
1156 	}
1157 }
1158 
1159 static void __filter_modify_flags(struct bnxt_qplib_qp *qp)
1160 {
1161 	switch (qp->cur_qp_state) {
1162 	case CMDQ_MODIFY_QP_NEW_STATE_RESET:
1163 		break;
1164 	case CMDQ_MODIFY_QP_NEW_STATE_INIT:
1165 		__modify_flags_from_init_state(qp);
1166 		break;
1167 	case CMDQ_MODIFY_QP_NEW_STATE_RTR:
1168 		__modify_flags_from_rtr_state(qp);
1169 		break;
1170 	case CMDQ_MODIFY_QP_NEW_STATE_RTS:
1171 		break;
1172 	case CMDQ_MODIFY_QP_NEW_STATE_SQD:
1173 		break;
1174 	case CMDQ_MODIFY_QP_NEW_STATE_SQE:
1175 		break;
1176 	case CMDQ_MODIFY_QP_NEW_STATE_ERR:
1177 		break;
1178 	default:
1179 		break;
1180 	}
1181 }
1182 
1183 int bnxt_qplib_modify_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
1184 {
1185 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1186 	struct cmdq_modify_qp req;
1187 	struct creq_modify_qp_resp resp;
1188 	u16 cmd_flags = 0, pkey;
1189 	u32 temp32[4];
1190 	u32 bmask;
1191 	int rc;
1192 
1193 	RCFW_CMD_PREP(req, MODIFY_QP, cmd_flags);
1194 
1195 	/* Filter out the qp_attr_mask based on the state->new transition */
1196 	__filter_modify_flags(qp);
1197 	bmask = qp->modify_flags;
1198 	req.modify_mask = cpu_to_le32(qp->modify_flags);
1199 	req.qp_cid = cpu_to_le32(qp->id);
1200 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_STATE) {
1201 		req.network_type_en_sqd_async_notify_new_state =
1202 				(qp->state & CMDQ_MODIFY_QP_NEW_STATE_MASK) |
1203 				(qp->en_sqd_async_notify ?
1204 					CMDQ_MODIFY_QP_EN_SQD_ASYNC_NOTIFY : 0);
1205 	}
1206 	req.network_type_en_sqd_async_notify_new_state |= qp->nw_type;
1207 
1208 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_ACCESS)
1209 		req.access = qp->access;
1210 
1211 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_PKEY) {
1212 		if (!bnxt_qplib_get_pkey(res, &res->pkey_tbl,
1213 					 qp->pkey_index, &pkey))
1214 			req.pkey = cpu_to_le16(pkey);
1215 	}
1216 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_QKEY)
1217 		req.qkey = cpu_to_le32(qp->qkey);
1218 
1219 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DGID) {
1220 		memcpy(temp32, qp->ah.dgid.data, sizeof(struct bnxt_qplib_gid));
1221 		req.dgid[0] = cpu_to_le32(temp32[0]);
1222 		req.dgid[1] = cpu_to_le32(temp32[1]);
1223 		req.dgid[2] = cpu_to_le32(temp32[2]);
1224 		req.dgid[3] = cpu_to_le32(temp32[3]);
1225 	}
1226 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_FLOW_LABEL)
1227 		req.flow_label = cpu_to_le32(qp->ah.flow_label);
1228 
1229 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_SGID_INDEX)
1230 		req.sgid_index = cpu_to_le16(res->sgid_tbl.hw_id
1231 					     [qp->ah.sgid_index]);
1232 
1233 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_HOP_LIMIT)
1234 		req.hop_limit = qp->ah.hop_limit;
1235 
1236 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_TRAFFIC_CLASS)
1237 		req.traffic_class = qp->ah.traffic_class;
1238 
1239 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DEST_MAC)
1240 		memcpy(req.dest_mac, qp->ah.dmac, 6);
1241 
1242 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_PATH_MTU)
1243 		req.path_mtu = qp->path_mtu;
1244 
1245 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_TIMEOUT)
1246 		req.timeout = qp->timeout;
1247 
1248 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RETRY_CNT)
1249 		req.retry_cnt = qp->retry_cnt;
1250 
1251 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RNR_RETRY)
1252 		req.rnr_retry = qp->rnr_retry;
1253 
1254 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MIN_RNR_TIMER)
1255 		req.min_rnr_timer = qp->min_rnr_timer;
1256 
1257 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_RQ_PSN)
1258 		req.rq_psn = cpu_to_le32(qp->rq.psn);
1259 
1260 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_SQ_PSN)
1261 		req.sq_psn = cpu_to_le32(qp->sq.psn);
1262 
1263 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MAX_RD_ATOMIC)
1264 		req.max_rd_atomic =
1265 			ORD_LIMIT_TO_ORRQ_SLOTS(qp->max_rd_atomic);
1266 
1267 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_MAX_DEST_RD_ATOMIC)
1268 		req.max_dest_rd_atomic =
1269 			IRD_LIMIT_TO_IRRQ_SLOTS(qp->max_dest_rd_atomic);
1270 
1271 	req.sq_size = cpu_to_le32(qp->sq.hwq.max_elements);
1272 	req.rq_size = cpu_to_le32(qp->rq.hwq.max_elements);
1273 	req.sq_sge = cpu_to_le16(qp->sq.max_sge);
1274 	req.rq_sge = cpu_to_le16(qp->rq.max_sge);
1275 	req.max_inline_data = cpu_to_le32(qp->max_inline_data);
1276 	if (bmask & CMDQ_MODIFY_QP_MODIFY_MASK_DEST_QP_ID)
1277 		req.dest_qp_id = cpu_to_le32(qp->dest_qpn);
1278 
1279 	req.vlan_pcp_vlan_dei_vlan_id = cpu_to_le16(qp->vlan_id);
1280 
1281 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
1282 					  (void *)&resp, NULL, 0);
1283 	if (rc)
1284 		return rc;
1285 	qp->cur_qp_state = qp->state;
1286 	return 0;
1287 }
1288 
1289 int bnxt_qplib_query_qp(struct bnxt_qplib_res *res, struct bnxt_qplib_qp *qp)
1290 {
1291 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1292 	struct cmdq_query_qp req;
1293 	struct creq_query_qp_resp resp;
1294 	struct bnxt_qplib_rcfw_sbuf *sbuf;
1295 	struct creq_query_qp_resp_sb *sb;
1296 	u16 cmd_flags = 0;
1297 	u32 temp32[4];
1298 	int i, rc = 0;
1299 
1300 	RCFW_CMD_PREP(req, QUERY_QP, cmd_flags);
1301 
1302 	sbuf = bnxt_qplib_rcfw_alloc_sbuf(rcfw, sizeof(*sb));
1303 	if (!sbuf)
1304 		return -ENOMEM;
1305 	sb = sbuf->sb;
1306 
1307 	req.qp_cid = cpu_to_le32(qp->id);
1308 	req.resp_size = sizeof(*sb) / BNXT_QPLIB_CMDQE_UNITS;
1309 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req, (void *)&resp,
1310 					  (void *)sbuf, 0);
1311 	if (rc)
1312 		goto bail;
1313 	/* Extract the context from the side buffer */
1314 	qp->state = sb->en_sqd_async_notify_state &
1315 			CREQ_QUERY_QP_RESP_SB_STATE_MASK;
1316 	qp->en_sqd_async_notify = sb->en_sqd_async_notify_state &
1317 				  CREQ_QUERY_QP_RESP_SB_EN_SQD_ASYNC_NOTIFY ?
1318 				  true : false;
1319 	qp->access = sb->access;
1320 	qp->pkey_index = le16_to_cpu(sb->pkey);
1321 	qp->qkey = le32_to_cpu(sb->qkey);
1322 
1323 	temp32[0] = le32_to_cpu(sb->dgid[0]);
1324 	temp32[1] = le32_to_cpu(sb->dgid[1]);
1325 	temp32[2] = le32_to_cpu(sb->dgid[2]);
1326 	temp32[3] = le32_to_cpu(sb->dgid[3]);
1327 	memcpy(qp->ah.dgid.data, temp32, sizeof(qp->ah.dgid.data));
1328 
1329 	qp->ah.flow_label = le32_to_cpu(sb->flow_label);
1330 
1331 	qp->ah.sgid_index = 0;
1332 	for (i = 0; i < res->sgid_tbl.max; i++) {
1333 		if (res->sgid_tbl.hw_id[i] == le16_to_cpu(sb->sgid_index)) {
1334 			qp->ah.sgid_index = i;
1335 			break;
1336 		}
1337 	}
1338 	if (i == res->sgid_tbl.max)
1339 		dev_warn(&res->pdev->dev, "QPLIB: SGID not found??");
1340 
1341 	qp->ah.hop_limit = sb->hop_limit;
1342 	qp->ah.traffic_class = sb->traffic_class;
1343 	memcpy(qp->ah.dmac, sb->dest_mac, 6);
1344 	qp->ah.vlan_id = (le16_to_cpu(sb->path_mtu_dest_vlan_id) &
1345 				CREQ_QUERY_QP_RESP_SB_VLAN_ID_MASK) >>
1346 				CREQ_QUERY_QP_RESP_SB_VLAN_ID_SFT;
1347 	qp->path_mtu = (le16_to_cpu(sb->path_mtu_dest_vlan_id) &
1348 				    CREQ_QUERY_QP_RESP_SB_PATH_MTU_MASK) >>
1349 				    CREQ_QUERY_QP_RESP_SB_PATH_MTU_SFT;
1350 	qp->timeout = sb->timeout;
1351 	qp->retry_cnt = sb->retry_cnt;
1352 	qp->rnr_retry = sb->rnr_retry;
1353 	qp->min_rnr_timer = sb->min_rnr_timer;
1354 	qp->rq.psn = le32_to_cpu(sb->rq_psn);
1355 	qp->max_rd_atomic = ORRQ_SLOTS_TO_ORD_LIMIT(sb->max_rd_atomic);
1356 	qp->sq.psn = le32_to_cpu(sb->sq_psn);
1357 	qp->max_dest_rd_atomic =
1358 			IRRQ_SLOTS_TO_IRD_LIMIT(sb->max_dest_rd_atomic);
1359 	qp->sq.max_wqe = qp->sq.hwq.max_elements;
1360 	qp->rq.max_wqe = qp->rq.hwq.max_elements;
1361 	qp->sq.max_sge = le16_to_cpu(sb->sq_sge);
1362 	qp->rq.max_sge = le16_to_cpu(sb->rq_sge);
1363 	qp->max_inline_data = le32_to_cpu(sb->max_inline_data);
1364 	qp->dest_qpn = le32_to_cpu(sb->dest_qp_id);
1365 	memcpy(qp->smac, sb->src_mac, 6);
1366 	qp->vlan_id = le16_to_cpu(sb->vlan_pcp_vlan_dei_vlan_id);
1367 bail:
1368 	bnxt_qplib_rcfw_free_sbuf(rcfw, sbuf);
1369 	return rc;
1370 }
1371 
1372 static void __clean_cq(struct bnxt_qplib_cq *cq, u64 qp)
1373 {
1374 	struct bnxt_qplib_hwq *cq_hwq = &cq->hwq;
1375 	struct cq_base *hw_cqe, **hw_cqe_ptr;
1376 	int i;
1377 
1378 	for (i = 0; i < cq_hwq->max_elements; i++) {
1379 		hw_cqe_ptr = (struct cq_base **)cq_hwq->pbl_ptr;
1380 		hw_cqe = &hw_cqe_ptr[CQE_PG(i)][CQE_IDX(i)];
1381 		if (!CQE_CMP_VALID(hw_cqe, i, cq_hwq->max_elements))
1382 			continue;
1383 		/*
1384 		 * The valid test of the entry must be done first before
1385 		 * reading any further.
1386 		 */
1387 		dma_rmb();
1388 		switch (hw_cqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK) {
1389 		case CQ_BASE_CQE_TYPE_REQ:
1390 		case CQ_BASE_CQE_TYPE_TERMINAL:
1391 		{
1392 			struct cq_req *cqe = (struct cq_req *)hw_cqe;
1393 
1394 			if (qp == le64_to_cpu(cqe->qp_handle))
1395 				cqe->qp_handle = 0;
1396 			break;
1397 		}
1398 		case CQ_BASE_CQE_TYPE_RES_RC:
1399 		case CQ_BASE_CQE_TYPE_RES_UD:
1400 		case CQ_BASE_CQE_TYPE_RES_RAWETH_QP1:
1401 		{
1402 			struct cq_res_rc *cqe = (struct cq_res_rc *)hw_cqe;
1403 
1404 			if (qp == le64_to_cpu(cqe->qp_handle))
1405 				cqe->qp_handle = 0;
1406 			break;
1407 		}
1408 		default:
1409 			break;
1410 		}
1411 	}
1412 }
1413 
1414 int bnxt_qplib_destroy_qp(struct bnxt_qplib_res *res,
1415 			  struct bnxt_qplib_qp *qp)
1416 {
1417 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1418 	struct cmdq_destroy_qp req;
1419 	struct creq_destroy_qp_resp resp;
1420 	unsigned long flags;
1421 	u16 cmd_flags = 0;
1422 	int rc;
1423 
1424 	rcfw->qp_tbl[qp->id].qp_id = BNXT_QPLIB_QP_ID_INVALID;
1425 	rcfw->qp_tbl[qp->id].qp_handle = NULL;
1426 
1427 	RCFW_CMD_PREP(req, DESTROY_QP, cmd_flags);
1428 
1429 	req.qp_cid = cpu_to_le32(qp->id);
1430 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
1431 					  (void *)&resp, NULL, 0);
1432 	if (rc) {
1433 		rcfw->qp_tbl[qp->id].qp_id = qp->id;
1434 		rcfw->qp_tbl[qp->id].qp_handle = qp;
1435 		return rc;
1436 	}
1437 
1438 	/* Must walk the associated CQs to nullified the QP ptr */
1439 	spin_lock_irqsave(&qp->scq->hwq.lock, flags);
1440 
1441 	__clean_cq(qp->scq, (u64)(unsigned long)qp);
1442 
1443 	if (qp->rcq && qp->rcq != qp->scq) {
1444 		spin_lock(&qp->rcq->hwq.lock);
1445 		__clean_cq(qp->rcq, (u64)(unsigned long)qp);
1446 		spin_unlock(&qp->rcq->hwq.lock);
1447 	}
1448 
1449 	spin_unlock_irqrestore(&qp->scq->hwq.lock, flags);
1450 
1451 	bnxt_qplib_free_qp_hdr_buf(res, qp);
1452 	bnxt_qplib_free_hwq(res->pdev, &qp->sq.hwq);
1453 	kfree(qp->sq.swq);
1454 
1455 	bnxt_qplib_free_hwq(res->pdev, &qp->rq.hwq);
1456 	kfree(qp->rq.swq);
1457 
1458 	if (qp->irrq.max_elements)
1459 		bnxt_qplib_free_hwq(res->pdev, &qp->irrq);
1460 	if (qp->orrq.max_elements)
1461 		bnxt_qplib_free_hwq(res->pdev, &qp->orrq);
1462 
1463 	return 0;
1464 }
1465 
1466 void *bnxt_qplib_get_qp1_sq_buf(struct bnxt_qplib_qp *qp,
1467 				struct bnxt_qplib_sge *sge)
1468 {
1469 	struct bnxt_qplib_q *sq = &qp->sq;
1470 	u32 sw_prod;
1471 
1472 	memset(sge, 0, sizeof(*sge));
1473 
1474 	if (qp->sq_hdr_buf) {
1475 		sw_prod = HWQ_CMP(sq->hwq.prod, &sq->hwq);
1476 		sge->addr = (dma_addr_t)(qp->sq_hdr_buf_map +
1477 					 sw_prod * qp->sq_hdr_buf_size);
1478 		sge->lkey = 0xFFFFFFFF;
1479 		sge->size = qp->sq_hdr_buf_size;
1480 		return qp->sq_hdr_buf + sw_prod * sge->size;
1481 	}
1482 	return NULL;
1483 }
1484 
1485 u32 bnxt_qplib_get_rq_prod_index(struct bnxt_qplib_qp *qp)
1486 {
1487 	struct bnxt_qplib_q *rq = &qp->rq;
1488 
1489 	return HWQ_CMP(rq->hwq.prod, &rq->hwq);
1490 }
1491 
1492 dma_addr_t bnxt_qplib_get_qp_buf_from_index(struct bnxt_qplib_qp *qp, u32 index)
1493 {
1494 	return (qp->rq_hdr_buf_map + index * qp->rq_hdr_buf_size);
1495 }
1496 
1497 void *bnxt_qplib_get_qp1_rq_buf(struct bnxt_qplib_qp *qp,
1498 				struct bnxt_qplib_sge *sge)
1499 {
1500 	struct bnxt_qplib_q *rq = &qp->rq;
1501 	u32 sw_prod;
1502 
1503 	memset(sge, 0, sizeof(*sge));
1504 
1505 	if (qp->rq_hdr_buf) {
1506 		sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq);
1507 		sge->addr = (dma_addr_t)(qp->rq_hdr_buf_map +
1508 					 sw_prod * qp->rq_hdr_buf_size);
1509 		sge->lkey = 0xFFFFFFFF;
1510 		sge->size = qp->rq_hdr_buf_size;
1511 		return qp->rq_hdr_buf + sw_prod * sge->size;
1512 	}
1513 	return NULL;
1514 }
1515 
1516 void bnxt_qplib_post_send_db(struct bnxt_qplib_qp *qp)
1517 {
1518 	struct bnxt_qplib_q *sq = &qp->sq;
1519 	struct dbr_dbr db_msg = { 0 };
1520 	u32 sw_prod;
1521 
1522 	sw_prod = HWQ_CMP(sq->hwq.prod, &sq->hwq);
1523 
1524 	db_msg.index = cpu_to_le32((sw_prod << DBR_DBR_INDEX_SFT) &
1525 				   DBR_DBR_INDEX_MASK);
1526 	db_msg.type_xid =
1527 		cpu_to_le32(((qp->id << DBR_DBR_XID_SFT) & DBR_DBR_XID_MASK) |
1528 			    DBR_DBR_TYPE_SQ);
1529 	/* Flush all the WQE writes to HW */
1530 	wmb();
1531 	__iowrite64_copy(qp->dpi->dbr, &db_msg, sizeof(db_msg) / sizeof(u64));
1532 }
1533 
1534 int bnxt_qplib_post_send(struct bnxt_qplib_qp *qp,
1535 			 struct bnxt_qplib_swqe *wqe)
1536 {
1537 	struct bnxt_qplib_q *sq = &qp->sq;
1538 	struct bnxt_qplib_swq *swq;
1539 	struct sq_send *hw_sq_send_hdr, **hw_sq_send_ptr;
1540 	struct sq_sge *hw_sge;
1541 	struct bnxt_qplib_nq_work *nq_work = NULL;
1542 	bool sch_handler = false;
1543 	u32 sw_prod;
1544 	u8 wqe_size16;
1545 	int i, rc = 0, data_len = 0, pkt_num = 0;
1546 	__le32 temp32;
1547 
1548 	if (qp->state != CMDQ_MODIFY_QP_NEW_STATE_RTS) {
1549 		if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1550 			sch_handler = true;
1551 			dev_dbg(&sq->hwq.pdev->dev,
1552 				"%s Error QP. Scheduling for poll_cq\n",
1553 				__func__);
1554 			goto queue_err;
1555 		}
1556 	}
1557 
1558 	if (bnxt_qplib_queue_full(sq)) {
1559 		dev_err(&sq->hwq.pdev->dev,
1560 			"QPLIB: prod = %#x cons = %#x qdepth = %#x delta = %#x",
1561 			sq->hwq.prod, sq->hwq.cons, sq->hwq.max_elements,
1562 			sq->q_full_delta);
1563 		rc = -ENOMEM;
1564 		goto done;
1565 	}
1566 	sw_prod = HWQ_CMP(sq->hwq.prod, &sq->hwq);
1567 	swq = &sq->swq[sw_prod];
1568 	swq->wr_id = wqe->wr_id;
1569 	swq->type = wqe->type;
1570 	swq->flags = wqe->flags;
1571 	if (qp->sig_type)
1572 		swq->flags |= SQ_SEND_FLAGS_SIGNAL_COMP;
1573 	swq->start_psn = sq->psn & BTH_PSN_MASK;
1574 
1575 	hw_sq_send_ptr = (struct sq_send **)sq->hwq.pbl_ptr;
1576 	hw_sq_send_hdr = &hw_sq_send_ptr[get_sqe_pg(sw_prod)]
1577 					[get_sqe_idx(sw_prod)];
1578 
1579 	memset(hw_sq_send_hdr, 0, BNXT_QPLIB_MAX_SQE_ENTRY_SIZE);
1580 
1581 	if (wqe->flags & BNXT_QPLIB_SWQE_FLAGS_INLINE) {
1582 		/* Copy the inline data */
1583 		if (wqe->inline_len > BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH) {
1584 			dev_warn(&sq->hwq.pdev->dev,
1585 				 "QPLIB: Inline data length > 96 detected");
1586 			data_len = BNXT_QPLIB_SWQE_MAX_INLINE_LENGTH;
1587 		} else {
1588 			data_len = wqe->inline_len;
1589 		}
1590 		memcpy(hw_sq_send_hdr->data, wqe->inline_data, data_len);
1591 		wqe_size16 = (data_len + 15) >> 4;
1592 	} else {
1593 		for (i = 0, hw_sge = (struct sq_sge *)hw_sq_send_hdr->data;
1594 		     i < wqe->num_sge; i++, hw_sge++) {
1595 			hw_sge->va_or_pa = cpu_to_le64(wqe->sg_list[i].addr);
1596 			hw_sge->l_key = cpu_to_le32(wqe->sg_list[i].lkey);
1597 			hw_sge->size = cpu_to_le32(wqe->sg_list[i].size);
1598 			data_len += wqe->sg_list[i].size;
1599 		}
1600 		/* Each SGE entry = 1 WQE size16 */
1601 		wqe_size16 = wqe->num_sge;
1602 		/* HW requires wqe size has room for atleast one SGE even if
1603 		 * none was supplied by ULP
1604 		 */
1605 		if (!wqe->num_sge)
1606 			wqe_size16++;
1607 	}
1608 
1609 	/* Specifics */
1610 	switch (wqe->type) {
1611 	case BNXT_QPLIB_SWQE_TYPE_SEND:
1612 		if (qp->type == CMDQ_CREATE_QP1_TYPE_GSI) {
1613 			/* Assemble info for Raw Ethertype QPs */
1614 			struct sq_send_raweth_qp1 *sqe =
1615 				(struct sq_send_raweth_qp1 *)hw_sq_send_hdr;
1616 
1617 			sqe->wqe_type = wqe->type;
1618 			sqe->flags = wqe->flags;
1619 			sqe->wqe_size = wqe_size16 +
1620 				((offsetof(typeof(*sqe), data) + 15) >> 4);
1621 			sqe->cfa_action = cpu_to_le16(wqe->rawqp1.cfa_action);
1622 			sqe->lflags = cpu_to_le16(wqe->rawqp1.lflags);
1623 			sqe->length = cpu_to_le32(data_len);
1624 			sqe->cfa_meta = cpu_to_le32((wqe->rawqp1.cfa_meta &
1625 				SQ_SEND_RAWETH_QP1_CFA_META_VLAN_VID_MASK) <<
1626 				SQ_SEND_RAWETH_QP1_CFA_META_VLAN_VID_SFT);
1627 
1628 			break;
1629 		}
1630 		/* fall thru */
1631 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_IMM:
1632 	case BNXT_QPLIB_SWQE_TYPE_SEND_WITH_INV:
1633 	{
1634 		struct sq_send *sqe = (struct sq_send *)hw_sq_send_hdr;
1635 
1636 		sqe->wqe_type = wqe->type;
1637 		sqe->flags = wqe->flags;
1638 		sqe->wqe_size = wqe_size16 +
1639 				((offsetof(typeof(*sqe), data) + 15) >> 4);
1640 		sqe->inv_key_or_imm_data = cpu_to_le32(
1641 						wqe->send.inv_key);
1642 		if (qp->type == CMDQ_CREATE_QP_TYPE_UD) {
1643 			sqe->q_key = cpu_to_le32(wqe->send.q_key);
1644 			sqe->dst_qp = cpu_to_le32(
1645 					wqe->send.dst_qp & SQ_SEND_DST_QP_MASK);
1646 			sqe->length = cpu_to_le32(data_len);
1647 			sqe->avid = cpu_to_le32(wqe->send.avid &
1648 						SQ_SEND_AVID_MASK);
1649 			sq->psn = (sq->psn + 1) & BTH_PSN_MASK;
1650 		} else {
1651 			sqe->length = cpu_to_le32(data_len);
1652 			sqe->dst_qp = 0;
1653 			sqe->avid = 0;
1654 			if (qp->mtu)
1655 				pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
1656 			if (!pkt_num)
1657 				pkt_num = 1;
1658 			sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
1659 		}
1660 		break;
1661 	}
1662 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE:
1663 	case BNXT_QPLIB_SWQE_TYPE_RDMA_WRITE_WITH_IMM:
1664 	case BNXT_QPLIB_SWQE_TYPE_RDMA_READ:
1665 	{
1666 		struct sq_rdma *sqe = (struct sq_rdma *)hw_sq_send_hdr;
1667 
1668 		sqe->wqe_type = wqe->type;
1669 		sqe->flags = wqe->flags;
1670 		sqe->wqe_size = wqe_size16 +
1671 				((offsetof(typeof(*sqe), data) + 15) >> 4);
1672 		sqe->imm_data = cpu_to_le32(wqe->rdma.inv_key);
1673 		sqe->length = cpu_to_le32((u32)data_len);
1674 		sqe->remote_va = cpu_to_le64(wqe->rdma.remote_va);
1675 		sqe->remote_key = cpu_to_le32(wqe->rdma.r_key);
1676 		if (qp->mtu)
1677 			pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
1678 		if (!pkt_num)
1679 			pkt_num = 1;
1680 		sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
1681 		break;
1682 	}
1683 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_CMP_AND_SWP:
1684 	case BNXT_QPLIB_SWQE_TYPE_ATOMIC_FETCH_AND_ADD:
1685 	{
1686 		struct sq_atomic *sqe = (struct sq_atomic *)hw_sq_send_hdr;
1687 
1688 		sqe->wqe_type = wqe->type;
1689 		sqe->flags = wqe->flags;
1690 		sqe->remote_key = cpu_to_le32(wqe->atomic.r_key);
1691 		sqe->remote_va = cpu_to_le64(wqe->atomic.remote_va);
1692 		sqe->swap_data = cpu_to_le64(wqe->atomic.swap_data);
1693 		sqe->cmp_data = cpu_to_le64(wqe->atomic.cmp_data);
1694 		if (qp->mtu)
1695 			pkt_num = (data_len + qp->mtu - 1) / qp->mtu;
1696 		if (!pkt_num)
1697 			pkt_num = 1;
1698 		sq->psn = (sq->psn + pkt_num) & BTH_PSN_MASK;
1699 		break;
1700 	}
1701 	case BNXT_QPLIB_SWQE_TYPE_LOCAL_INV:
1702 	{
1703 		struct sq_localinvalidate *sqe =
1704 				(struct sq_localinvalidate *)hw_sq_send_hdr;
1705 
1706 		sqe->wqe_type = wqe->type;
1707 		sqe->flags = wqe->flags;
1708 		sqe->inv_l_key = cpu_to_le32(wqe->local_inv.inv_l_key);
1709 
1710 		break;
1711 	}
1712 	case BNXT_QPLIB_SWQE_TYPE_FAST_REG_MR:
1713 	{
1714 		struct sq_fr_pmr *sqe = (struct sq_fr_pmr *)hw_sq_send_hdr;
1715 
1716 		sqe->wqe_type = wqe->type;
1717 		sqe->flags = wqe->flags;
1718 		sqe->access_cntl = wqe->frmr.access_cntl |
1719 				   SQ_FR_PMR_ACCESS_CNTL_LOCAL_WRITE;
1720 		sqe->zero_based_page_size_log =
1721 			(wqe->frmr.pg_sz_log & SQ_FR_PMR_PAGE_SIZE_LOG_MASK) <<
1722 			SQ_FR_PMR_PAGE_SIZE_LOG_SFT |
1723 			(wqe->frmr.zero_based ? SQ_FR_PMR_ZERO_BASED : 0);
1724 		sqe->l_key = cpu_to_le32(wqe->frmr.l_key);
1725 		temp32 = cpu_to_le32(wqe->frmr.length);
1726 		memcpy(sqe->length, &temp32, sizeof(wqe->frmr.length));
1727 		sqe->numlevels_pbl_page_size_log =
1728 			((wqe->frmr.pbl_pg_sz_log <<
1729 					SQ_FR_PMR_PBL_PAGE_SIZE_LOG_SFT) &
1730 					SQ_FR_PMR_PBL_PAGE_SIZE_LOG_MASK) |
1731 			((wqe->frmr.levels << SQ_FR_PMR_NUMLEVELS_SFT) &
1732 					SQ_FR_PMR_NUMLEVELS_MASK);
1733 
1734 		for (i = 0; i < wqe->frmr.page_list_len; i++)
1735 			wqe->frmr.pbl_ptr[i] = cpu_to_le64(
1736 						wqe->frmr.page_list[i] |
1737 						PTU_PTE_VALID);
1738 		sqe->pblptr = cpu_to_le64(wqe->frmr.pbl_dma_ptr);
1739 		sqe->va = cpu_to_le64(wqe->frmr.va);
1740 
1741 		break;
1742 	}
1743 	case BNXT_QPLIB_SWQE_TYPE_BIND_MW:
1744 	{
1745 		struct sq_bind *sqe = (struct sq_bind *)hw_sq_send_hdr;
1746 
1747 		sqe->wqe_type = wqe->type;
1748 		sqe->flags = wqe->flags;
1749 		sqe->access_cntl = wqe->bind.access_cntl;
1750 		sqe->mw_type_zero_based = wqe->bind.mw_type |
1751 			(wqe->bind.zero_based ? SQ_BIND_ZERO_BASED : 0);
1752 		sqe->parent_l_key = cpu_to_le32(wqe->bind.parent_l_key);
1753 		sqe->l_key = cpu_to_le32(wqe->bind.r_key);
1754 		sqe->va = cpu_to_le64(wqe->bind.va);
1755 		temp32 = cpu_to_le32(wqe->bind.length);
1756 		memcpy(&sqe->length, &temp32, sizeof(wqe->bind.length));
1757 		break;
1758 	}
1759 	default:
1760 		/* Bad wqe, return error */
1761 		rc = -EINVAL;
1762 		goto done;
1763 	}
1764 	swq->next_psn = sq->psn & BTH_PSN_MASK;
1765 	if (swq->psn_search) {
1766 		swq->psn_search->opcode_start_psn = cpu_to_le32(
1767 			((swq->start_psn << SQ_PSN_SEARCH_START_PSN_SFT) &
1768 			 SQ_PSN_SEARCH_START_PSN_MASK) |
1769 			((wqe->type << SQ_PSN_SEARCH_OPCODE_SFT) &
1770 			 SQ_PSN_SEARCH_OPCODE_MASK));
1771 		swq->psn_search->flags_next_psn = cpu_to_le32(
1772 			((swq->next_psn << SQ_PSN_SEARCH_NEXT_PSN_SFT) &
1773 			 SQ_PSN_SEARCH_NEXT_PSN_MASK));
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 				"QPLIB: FP: Failed to allocate SQ nq_work!");
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 	struct dbr_dbr db_msg = { 0 };
1811 	u32 sw_prod;
1812 
1813 	sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq);
1814 	db_msg.index = cpu_to_le32((sw_prod << DBR_DBR_INDEX_SFT) &
1815 				   DBR_DBR_INDEX_MASK);
1816 	db_msg.type_xid =
1817 		cpu_to_le32(((qp->id << DBR_DBR_XID_SFT) & DBR_DBR_XID_MASK) |
1818 			    DBR_DBR_TYPE_RQ);
1819 
1820 	/* Flush the writes to HW Rx WQE before the ringing Rx DB */
1821 	wmb();
1822 	__iowrite64_copy(qp->dpi->dbr, &db_msg, sizeof(db_msg) / sizeof(u64));
1823 }
1824 
1825 int bnxt_qplib_post_recv(struct bnxt_qplib_qp *qp,
1826 			 struct bnxt_qplib_swqe *wqe)
1827 {
1828 	struct bnxt_qplib_q *rq = &qp->rq;
1829 	struct rq_wqe *rqe, **rqe_ptr;
1830 	struct sq_sge *hw_sge;
1831 	struct bnxt_qplib_nq_work *nq_work = NULL;
1832 	bool sch_handler = false;
1833 	u32 sw_prod;
1834 	int i, rc = 0;
1835 
1836 	if (qp->state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1837 		sch_handler = true;
1838 		dev_dbg(&rq->hwq.pdev->dev,
1839 			"%s Error QP. Scheduling for poll_cq\n",
1840 			__func__);
1841 		goto queue_err;
1842 	}
1843 	if (bnxt_qplib_queue_full(rq)) {
1844 		dev_err(&rq->hwq.pdev->dev,
1845 			"QPLIB: FP: QP (0x%x) RQ is full!", qp->id);
1846 		rc = -EINVAL;
1847 		goto done;
1848 	}
1849 	sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq);
1850 	rq->swq[sw_prod].wr_id = wqe->wr_id;
1851 
1852 	rqe_ptr = (struct rq_wqe **)rq->hwq.pbl_ptr;
1853 	rqe = &rqe_ptr[RQE_PG(sw_prod)][RQE_IDX(sw_prod)];
1854 
1855 	memset(rqe, 0, BNXT_QPLIB_MAX_RQE_ENTRY_SIZE);
1856 
1857 	/* Calculate wqe_size16 and data_len */
1858 	for (i = 0, hw_sge = (struct sq_sge *)rqe->data;
1859 	     i < wqe->num_sge; i++, hw_sge++) {
1860 		hw_sge->va_or_pa = cpu_to_le64(wqe->sg_list[i].addr);
1861 		hw_sge->l_key = cpu_to_le32(wqe->sg_list[i].lkey);
1862 		hw_sge->size = cpu_to_le32(wqe->sg_list[i].size);
1863 	}
1864 	rqe->wqe_type = wqe->type;
1865 	rqe->flags = wqe->flags;
1866 	rqe->wqe_size = wqe->num_sge +
1867 			((offsetof(typeof(*rqe), data) + 15) >> 4);
1868 	/* HW requires wqe size has room for atleast one SGE even if none
1869 	 * was supplied by ULP
1870 	 */
1871 	if (!wqe->num_sge)
1872 		rqe->wqe_size++;
1873 
1874 	/* Supply the rqe->wr_id index to the wr_id_tbl for now */
1875 	rqe->wr_id[0] = cpu_to_le32(sw_prod);
1876 
1877 queue_err:
1878 	if (sch_handler) {
1879 		/* Store the ULP info in the software structures */
1880 		sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq);
1881 		rq->swq[sw_prod].wr_id = wqe->wr_id;
1882 	}
1883 
1884 	rq->hwq.prod++;
1885 	if (sch_handler) {
1886 		nq_work = kzalloc(sizeof(*nq_work), GFP_ATOMIC);
1887 		if (nq_work) {
1888 			nq_work->cq = qp->rcq;
1889 			nq_work->nq = qp->rcq->nq;
1890 			INIT_WORK(&nq_work->work, bnxt_qpn_cqn_sched_task);
1891 			queue_work(qp->rcq->nq->cqn_wq, &nq_work->work);
1892 		} else {
1893 			dev_err(&rq->hwq.pdev->dev,
1894 				"QPLIB: FP: Failed to allocate RQ nq_work!");
1895 			rc = -ENOMEM;
1896 		}
1897 	}
1898 done:
1899 	return rc;
1900 }
1901 
1902 /* CQ */
1903 
1904 /* Spinlock must be held */
1905 static void bnxt_qplib_arm_cq_enable(struct bnxt_qplib_cq *cq)
1906 {
1907 	struct dbr_dbr db_msg = { 0 };
1908 
1909 	db_msg.type_xid =
1910 		cpu_to_le32(((cq->id << DBR_DBR_XID_SFT) & DBR_DBR_XID_MASK) |
1911 			    DBR_DBR_TYPE_CQ_ARMENA);
1912 	/* Flush memory writes before enabling the CQ */
1913 	wmb();
1914 	__iowrite64_copy(cq->dbr_base, &db_msg, sizeof(db_msg) / sizeof(u64));
1915 }
1916 
1917 static void bnxt_qplib_arm_cq(struct bnxt_qplib_cq *cq, u32 arm_type)
1918 {
1919 	struct bnxt_qplib_hwq *cq_hwq = &cq->hwq;
1920 	struct dbr_dbr db_msg = { 0 };
1921 	u32 sw_cons;
1922 
1923 	/* Ring DB */
1924 	sw_cons = HWQ_CMP(cq_hwq->cons, cq_hwq);
1925 	db_msg.index = cpu_to_le32((sw_cons << DBR_DBR_INDEX_SFT) &
1926 				    DBR_DBR_INDEX_MASK);
1927 	db_msg.type_xid =
1928 		cpu_to_le32(((cq->id << DBR_DBR_XID_SFT) & DBR_DBR_XID_MASK) |
1929 			    arm_type);
1930 	/* flush memory writes before arming the CQ */
1931 	wmb();
1932 	__iowrite64_copy(cq->dpi->dbr, &db_msg, sizeof(db_msg) / sizeof(u64));
1933 }
1934 
1935 int bnxt_qplib_create_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq)
1936 {
1937 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
1938 	struct cmdq_create_cq req;
1939 	struct creq_create_cq_resp resp;
1940 	struct bnxt_qplib_pbl *pbl;
1941 	u16 cmd_flags = 0;
1942 	int rc;
1943 
1944 	cq->hwq.max_elements = cq->max_wqe;
1945 	rc = bnxt_qplib_alloc_init_hwq(res->pdev, &cq->hwq, cq->sghead,
1946 				       cq->nmap, &cq->hwq.max_elements,
1947 				       BNXT_QPLIB_MAX_CQE_ENTRY_SIZE, 0,
1948 				       PAGE_SIZE, HWQ_TYPE_QUEUE);
1949 	if (rc)
1950 		goto exit;
1951 
1952 	RCFW_CMD_PREP(req, CREATE_CQ, cmd_flags);
1953 
1954 	if (!cq->dpi) {
1955 		dev_err(&rcfw->pdev->dev,
1956 			"QPLIB: FP: CREATE_CQ failed due to NULL DPI");
1957 		return -EINVAL;
1958 	}
1959 	req.dpi = cpu_to_le32(cq->dpi->dpi);
1960 	req.cq_handle = cpu_to_le64(cq->cq_handle);
1961 
1962 	req.cq_size = cpu_to_le32(cq->hwq.max_elements);
1963 	pbl = &cq->hwq.pbl[PBL_LVL_0];
1964 	req.pg_size_lvl = cpu_to_le32(
1965 	    ((cq->hwq.level & CMDQ_CREATE_CQ_LVL_MASK) <<
1966 						CMDQ_CREATE_CQ_LVL_SFT) |
1967 	    (pbl->pg_size == ROCE_PG_SIZE_4K ? CMDQ_CREATE_CQ_PG_SIZE_PG_4K :
1968 	     pbl->pg_size == ROCE_PG_SIZE_8K ? CMDQ_CREATE_CQ_PG_SIZE_PG_8K :
1969 	     pbl->pg_size == ROCE_PG_SIZE_64K ? CMDQ_CREATE_CQ_PG_SIZE_PG_64K :
1970 	     pbl->pg_size == ROCE_PG_SIZE_2M ? CMDQ_CREATE_CQ_PG_SIZE_PG_2M :
1971 	     pbl->pg_size == ROCE_PG_SIZE_8M ? CMDQ_CREATE_CQ_PG_SIZE_PG_8M :
1972 	     pbl->pg_size == ROCE_PG_SIZE_1G ? CMDQ_CREATE_CQ_PG_SIZE_PG_1G :
1973 	     CMDQ_CREATE_CQ_PG_SIZE_PG_4K));
1974 
1975 	req.pbl = cpu_to_le64(pbl->pg_map_arr[0]);
1976 
1977 	req.cq_fco_cnq_id = cpu_to_le32(
1978 			(cq->cnq_hw_ring_id & CMDQ_CREATE_CQ_CNQ_ID_MASK) <<
1979 			 CMDQ_CREATE_CQ_CNQ_ID_SFT);
1980 
1981 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
1982 					  (void *)&resp, NULL, 0);
1983 	if (rc)
1984 		goto fail;
1985 
1986 	cq->id = le32_to_cpu(resp.xid);
1987 	cq->dbr_base = res->dpi_tbl.dbr_bar_reg_iomem;
1988 	cq->period = BNXT_QPLIB_QUEUE_START_PERIOD;
1989 	init_waitqueue_head(&cq->waitq);
1990 	INIT_LIST_HEAD(&cq->sqf_head);
1991 	INIT_LIST_HEAD(&cq->rqf_head);
1992 	spin_lock_init(&cq->compl_lock);
1993 
1994 	bnxt_qplib_arm_cq_enable(cq);
1995 	return 0;
1996 
1997 fail:
1998 	bnxt_qplib_free_hwq(res->pdev, &cq->hwq);
1999 exit:
2000 	return rc;
2001 }
2002 
2003 int bnxt_qplib_destroy_cq(struct bnxt_qplib_res *res, struct bnxt_qplib_cq *cq)
2004 {
2005 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
2006 	struct cmdq_destroy_cq req;
2007 	struct creq_destroy_cq_resp resp;
2008 	u16 cmd_flags = 0;
2009 	int rc;
2010 
2011 	RCFW_CMD_PREP(req, DESTROY_CQ, cmd_flags);
2012 
2013 	req.cq_cid = cpu_to_le32(cq->id);
2014 	rc = bnxt_qplib_rcfw_send_message(rcfw, (void *)&req,
2015 					  (void *)&resp, NULL, 0);
2016 	if (rc)
2017 		return rc;
2018 	bnxt_qplib_free_hwq(res->pdev, &cq->hwq);
2019 	return 0;
2020 }
2021 
2022 static int __flush_sq(struct bnxt_qplib_q *sq, struct bnxt_qplib_qp *qp,
2023 		      struct bnxt_qplib_cqe **pcqe, int *budget)
2024 {
2025 	u32 sw_prod, sw_cons;
2026 	struct bnxt_qplib_cqe *cqe;
2027 	int rc = 0;
2028 
2029 	/* Now complete all outstanding SQEs with FLUSHED_ERR */
2030 	sw_prod = HWQ_CMP(sq->hwq.prod, &sq->hwq);
2031 	cqe = *pcqe;
2032 	while (*budget) {
2033 		sw_cons = HWQ_CMP(sq->hwq.cons, &sq->hwq);
2034 		if (sw_cons == sw_prod) {
2035 			break;
2036 		}
2037 		/* Skip the FENCE WQE completions */
2038 		if (sq->swq[sw_cons].wr_id == BNXT_QPLIB_FENCE_WRID) {
2039 			bnxt_qplib_cancel_phantom_processing(qp);
2040 			goto skip_compl;
2041 		}
2042 		memset(cqe, 0, sizeof(*cqe));
2043 		cqe->status = CQ_REQ_STATUS_WORK_REQUEST_FLUSHED_ERR;
2044 		cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
2045 		cqe->qp_handle = (u64)(unsigned long)qp;
2046 		cqe->wr_id = sq->swq[sw_cons].wr_id;
2047 		cqe->src_qp = qp->id;
2048 		cqe->type = sq->swq[sw_cons].type;
2049 		cqe++;
2050 		(*budget)--;
2051 skip_compl:
2052 		sq->hwq.cons++;
2053 	}
2054 	*pcqe = cqe;
2055 	if (!(*budget) && HWQ_CMP(sq->hwq.cons, &sq->hwq) != sw_prod)
2056 		/* Out of budget */
2057 		rc = -EAGAIN;
2058 
2059 	return rc;
2060 }
2061 
2062 static int __flush_rq(struct bnxt_qplib_q *rq, struct bnxt_qplib_qp *qp,
2063 		      struct bnxt_qplib_cqe **pcqe, int *budget)
2064 {
2065 	struct bnxt_qplib_cqe *cqe;
2066 	u32 sw_prod, sw_cons;
2067 	int rc = 0;
2068 	int opcode = 0;
2069 
2070 	switch (qp->type) {
2071 	case CMDQ_CREATE_QP1_TYPE_GSI:
2072 		opcode = CQ_BASE_CQE_TYPE_RES_RAWETH_QP1;
2073 		break;
2074 	case CMDQ_CREATE_QP_TYPE_RC:
2075 		opcode = CQ_BASE_CQE_TYPE_RES_RC;
2076 		break;
2077 	case CMDQ_CREATE_QP_TYPE_UD:
2078 		opcode = CQ_BASE_CQE_TYPE_RES_UD;
2079 		break;
2080 	}
2081 
2082 	/* Flush the rest of the RQ */
2083 	sw_prod = HWQ_CMP(rq->hwq.prod, &rq->hwq);
2084 	cqe = *pcqe;
2085 	while (*budget) {
2086 		sw_cons = HWQ_CMP(rq->hwq.cons, &rq->hwq);
2087 		if (sw_cons == sw_prod)
2088 			break;
2089 		memset(cqe, 0, sizeof(*cqe));
2090 		cqe->status =
2091 		    CQ_RES_RC_STATUS_WORK_REQUEST_FLUSHED_ERR;
2092 		cqe->opcode = opcode;
2093 		cqe->qp_handle = (unsigned long)qp;
2094 		cqe->wr_id = rq->swq[sw_cons].wr_id;
2095 		cqe++;
2096 		(*budget)--;
2097 		rq->hwq.cons++;
2098 	}
2099 	*pcqe = cqe;
2100 	if (!*budget && HWQ_CMP(rq->hwq.cons, &rq->hwq) != sw_prod)
2101 		/* Out of budget */
2102 		rc = -EAGAIN;
2103 
2104 	return rc;
2105 }
2106 
2107 void bnxt_qplib_mark_qp_error(void *qp_handle)
2108 {
2109 	struct bnxt_qplib_qp *qp = qp_handle;
2110 
2111 	if (!qp)
2112 		return;
2113 
2114 	/* Must block new posting of SQ and RQ */
2115 	qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2116 	bnxt_qplib_cancel_phantom_processing(qp);
2117 
2118 	/* Add qp to flush list of the CQ */
2119 	__bnxt_qplib_add_flush_qp(qp);
2120 }
2121 
2122 /* Note: SQE is valid from sw_sq_cons up to cqe_sq_cons (exclusive)
2123  *       CQE is track from sw_cq_cons to max_element but valid only if VALID=1
2124  */
2125 static int do_wa9060(struct bnxt_qplib_qp *qp, struct bnxt_qplib_cq *cq,
2126 		     u32 cq_cons, u32 sw_sq_cons, u32 cqe_sq_cons)
2127 {
2128 	struct bnxt_qplib_q *sq = &qp->sq;
2129 	struct bnxt_qplib_swq *swq;
2130 	u32 peek_sw_cq_cons, peek_raw_cq_cons, peek_sq_cons_idx;
2131 	struct cq_base *peek_hwcqe, **peek_hw_cqe_ptr;
2132 	struct cq_req *peek_req_hwcqe;
2133 	struct bnxt_qplib_qp *peek_qp;
2134 	struct bnxt_qplib_q *peek_sq;
2135 	int i, rc = 0;
2136 
2137 	/* Normal mode */
2138 	/* Check for the psn_search marking before completing */
2139 	swq = &sq->swq[sw_sq_cons];
2140 	if (swq->psn_search &&
2141 	    le32_to_cpu(swq->psn_search->flags_next_psn) & 0x80000000) {
2142 		/* Unmark */
2143 		swq->psn_search->flags_next_psn = cpu_to_le32
2144 			(le32_to_cpu(swq->psn_search->flags_next_psn)
2145 				     & ~0x80000000);
2146 		dev_dbg(&cq->hwq.pdev->dev,
2147 			"FP: Process Req cq_cons=0x%x qp=0x%x sq cons sw=0x%x cqe=0x%x marked!\n",
2148 			cq_cons, qp->id, sw_sq_cons, cqe_sq_cons);
2149 		sq->condition = true;
2150 		sq->send_phantom = true;
2151 
2152 		/* TODO: Only ARM if the previous SQE is ARMALL */
2153 		bnxt_qplib_arm_cq(cq, DBR_DBR_TYPE_CQ_ARMALL);
2154 
2155 		rc = -EAGAIN;
2156 		goto out;
2157 	}
2158 	if (sq->condition) {
2159 		/* Peek at the completions */
2160 		peek_raw_cq_cons = cq->hwq.cons;
2161 		peek_sw_cq_cons = cq_cons;
2162 		i = cq->hwq.max_elements;
2163 		while (i--) {
2164 			peek_sw_cq_cons = HWQ_CMP((peek_sw_cq_cons), &cq->hwq);
2165 			peek_hw_cqe_ptr = (struct cq_base **)cq->hwq.pbl_ptr;
2166 			peek_hwcqe = &peek_hw_cqe_ptr[CQE_PG(peek_sw_cq_cons)]
2167 						     [CQE_IDX(peek_sw_cq_cons)];
2168 			/* If the next hwcqe is VALID */
2169 			if (CQE_CMP_VALID(peek_hwcqe, peek_raw_cq_cons,
2170 					  cq->hwq.max_elements)) {
2171 			/*
2172 			 * The valid test of the entry must be done first before
2173 			 * reading any further.
2174 			 */
2175 				dma_rmb();
2176 				/* If the next hwcqe is a REQ */
2177 				if ((peek_hwcqe->cqe_type_toggle &
2178 				    CQ_BASE_CQE_TYPE_MASK) ==
2179 				    CQ_BASE_CQE_TYPE_REQ) {
2180 					peek_req_hwcqe = (struct cq_req *)
2181 							 peek_hwcqe;
2182 					peek_qp = (struct bnxt_qplib_qp *)
2183 						((unsigned long)
2184 						 le64_to_cpu
2185 						 (peek_req_hwcqe->qp_handle));
2186 					peek_sq = &peek_qp->sq;
2187 					peek_sq_cons_idx = HWQ_CMP(le16_to_cpu(
2188 						peek_req_hwcqe->sq_cons_idx) - 1
2189 						, &sq->hwq);
2190 					/* If the hwcqe's sq's wr_id matches */
2191 					if (peek_sq == sq &&
2192 					    sq->swq[peek_sq_cons_idx].wr_id ==
2193 					    BNXT_QPLIB_FENCE_WRID) {
2194 						/*
2195 						 *  Unbreak only if the phantom
2196 						 *  comes back
2197 						 */
2198 						dev_dbg(&cq->hwq.pdev->dev,
2199 							"FP:Got Phantom CQE");
2200 						sq->condition = false;
2201 						sq->single = true;
2202 						rc = 0;
2203 						goto out;
2204 					}
2205 				}
2206 				/* Valid but not the phantom, so keep looping */
2207 			} else {
2208 				/* Not valid yet, just exit and wait */
2209 				rc = -EINVAL;
2210 				goto out;
2211 			}
2212 			peek_sw_cq_cons++;
2213 			peek_raw_cq_cons++;
2214 		}
2215 		dev_err(&cq->hwq.pdev->dev,
2216 			"Should not have come here! cq_cons=0x%x qp=0x%x sq cons sw=0x%x hw=0x%x",
2217 			cq_cons, qp->id, sw_sq_cons, cqe_sq_cons);
2218 		rc = -EINVAL;
2219 	}
2220 out:
2221 	return rc;
2222 }
2223 
2224 static int bnxt_qplib_cq_process_req(struct bnxt_qplib_cq *cq,
2225 				     struct cq_req *hwcqe,
2226 				     struct bnxt_qplib_cqe **pcqe, int *budget,
2227 				     u32 cq_cons, struct bnxt_qplib_qp **lib_qp)
2228 {
2229 	struct bnxt_qplib_qp *qp;
2230 	struct bnxt_qplib_q *sq;
2231 	struct bnxt_qplib_cqe *cqe;
2232 	u32 sw_sq_cons, cqe_sq_cons;
2233 	struct bnxt_qplib_swq *swq;
2234 	int rc = 0;
2235 
2236 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2237 				      le64_to_cpu(hwcqe->qp_handle));
2238 	if (!qp) {
2239 		dev_err(&cq->hwq.pdev->dev,
2240 			"QPLIB: FP: Process Req qp is NULL");
2241 		return -EINVAL;
2242 	}
2243 	sq = &qp->sq;
2244 
2245 	cqe_sq_cons = HWQ_CMP(le16_to_cpu(hwcqe->sq_cons_idx), &sq->hwq);
2246 	if (cqe_sq_cons > sq->hwq.max_elements) {
2247 		dev_err(&cq->hwq.pdev->dev,
2248 			"QPLIB: FP: CQ Process req reported ");
2249 		dev_err(&cq->hwq.pdev->dev,
2250 			"QPLIB: sq_cons_idx 0x%x which exceeded max 0x%x",
2251 			cqe_sq_cons, sq->hwq.max_elements);
2252 		return -EINVAL;
2253 	}
2254 
2255 	if (qp->sq.flushed) {
2256 		dev_dbg(&cq->hwq.pdev->dev,
2257 			"%s: QPLIB: QP in Flush QP = %p\n", __func__, qp);
2258 		goto done;
2259 	}
2260 	/* Require to walk the sq's swq to fabricate CQEs for all previously
2261 	 * signaled SWQEs due to CQE aggregation from the current sq cons
2262 	 * to the cqe_sq_cons
2263 	 */
2264 	cqe = *pcqe;
2265 	while (*budget) {
2266 		sw_sq_cons = HWQ_CMP(sq->hwq.cons, &sq->hwq);
2267 		if (sw_sq_cons == cqe_sq_cons)
2268 			/* Done */
2269 			break;
2270 
2271 		swq = &sq->swq[sw_sq_cons];
2272 		memset(cqe, 0, sizeof(*cqe));
2273 		cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
2274 		cqe->qp_handle = (u64)(unsigned long)qp;
2275 		cqe->src_qp = qp->id;
2276 		cqe->wr_id = swq->wr_id;
2277 		if (cqe->wr_id == BNXT_QPLIB_FENCE_WRID)
2278 			goto skip;
2279 		cqe->type = swq->type;
2280 
2281 		/* For the last CQE, check for status.  For errors, regardless
2282 		 * of the request being signaled or not, it must complete with
2283 		 * the hwcqe error status
2284 		 */
2285 		if (HWQ_CMP((sw_sq_cons + 1), &sq->hwq) == cqe_sq_cons &&
2286 		    hwcqe->status != CQ_REQ_STATUS_OK) {
2287 			cqe->status = hwcqe->status;
2288 			dev_err(&cq->hwq.pdev->dev,
2289 				"QPLIB: FP: CQ Processed Req ");
2290 			dev_err(&cq->hwq.pdev->dev,
2291 				"QPLIB: wr_id[%d] = 0x%llx with status 0x%x",
2292 				sw_sq_cons, cqe->wr_id, cqe->status);
2293 			cqe++;
2294 			(*budget)--;
2295 			bnxt_qplib_lock_buddy_cq(qp, cq);
2296 			bnxt_qplib_mark_qp_error(qp);
2297 			bnxt_qplib_unlock_buddy_cq(qp, cq);
2298 		} else {
2299 			if (swq->flags & SQ_SEND_FLAGS_SIGNAL_COMP) {
2300 				/* Before we complete, do WA 9060 */
2301 				if (do_wa9060(qp, cq, cq_cons, sw_sq_cons,
2302 					      cqe_sq_cons)) {
2303 					*lib_qp = qp;
2304 					goto out;
2305 				}
2306 				cqe->status = CQ_REQ_STATUS_OK;
2307 				cqe++;
2308 				(*budget)--;
2309 			}
2310 		}
2311 skip:
2312 		sq->hwq.cons++;
2313 		if (sq->single)
2314 			break;
2315 	}
2316 out:
2317 	*pcqe = cqe;
2318 	if (HWQ_CMP(sq->hwq.cons, &sq->hwq) != cqe_sq_cons) {
2319 		/* Out of budget */
2320 		rc = -EAGAIN;
2321 		goto done;
2322 	}
2323 	/*
2324 	 * Back to normal completion mode only after it has completed all of
2325 	 * the WC for this CQE
2326 	 */
2327 	sq->single = false;
2328 done:
2329 	return rc;
2330 }
2331 
2332 static void bnxt_qplib_release_srqe(struct bnxt_qplib_srq *srq, u32 tag)
2333 {
2334 	spin_lock(&srq->hwq.lock);
2335 	srq->swq[srq->last_idx].next_idx = (int)tag;
2336 	srq->last_idx = (int)tag;
2337 	srq->swq[srq->last_idx].next_idx = -1;
2338 	srq->hwq.cons++; /* Support for SRQE counter */
2339 	spin_unlock(&srq->hwq.lock);
2340 }
2341 
2342 static int bnxt_qplib_cq_process_res_rc(struct bnxt_qplib_cq *cq,
2343 					struct cq_res_rc *hwcqe,
2344 					struct bnxt_qplib_cqe **pcqe,
2345 					int *budget)
2346 {
2347 	struct bnxt_qplib_qp *qp;
2348 	struct bnxt_qplib_q *rq;
2349 	struct bnxt_qplib_srq *srq;
2350 	struct bnxt_qplib_cqe *cqe;
2351 	u32 wr_id_idx;
2352 	int rc = 0;
2353 
2354 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2355 				      le64_to_cpu(hwcqe->qp_handle));
2356 	if (!qp) {
2357 		dev_err(&cq->hwq.pdev->dev, "QPLIB: process_cq RC qp is NULL");
2358 		return -EINVAL;
2359 	}
2360 	if (qp->rq.flushed) {
2361 		dev_dbg(&cq->hwq.pdev->dev,
2362 			"%s: QPLIB: QP in Flush QP = %p\n", __func__, qp);
2363 		goto done;
2364 	}
2365 
2366 	cqe = *pcqe;
2367 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2368 	cqe->length = le32_to_cpu(hwcqe->length);
2369 	cqe->invrkey = le32_to_cpu(hwcqe->imm_data_or_inv_r_key);
2370 	cqe->mr_handle = le64_to_cpu(hwcqe->mr_handle);
2371 	cqe->flags = le16_to_cpu(hwcqe->flags);
2372 	cqe->status = hwcqe->status;
2373 	cqe->qp_handle = (u64)(unsigned long)qp;
2374 
2375 	wr_id_idx = le32_to_cpu(hwcqe->srq_or_rq_wr_id) &
2376 				CQ_RES_RC_SRQ_OR_RQ_WR_ID_MASK;
2377 	if (cqe->flags & CQ_RES_RC_FLAGS_SRQ_SRQ) {
2378 		srq = qp->srq;
2379 		if (!srq)
2380 			return -EINVAL;
2381 		if (wr_id_idx > srq->hwq.max_elements) {
2382 			dev_err(&cq->hwq.pdev->dev,
2383 				"QPLIB: FP: CQ Process RC ");
2384 			dev_err(&cq->hwq.pdev->dev,
2385 				"QPLIB: wr_id idx 0x%x exceeded SRQ max 0x%x",
2386 				wr_id_idx, srq->hwq.max_elements);
2387 			return -EINVAL;
2388 		}
2389 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2390 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2391 		cqe++;
2392 		(*budget)--;
2393 		*pcqe = cqe;
2394 	} else {
2395 		rq = &qp->rq;
2396 		if (wr_id_idx > rq->hwq.max_elements) {
2397 			dev_err(&cq->hwq.pdev->dev,
2398 				"QPLIB: FP: CQ Process RC ");
2399 			dev_err(&cq->hwq.pdev->dev,
2400 				"QPLIB: wr_id idx 0x%x exceeded RQ max 0x%x",
2401 				wr_id_idx, rq->hwq.max_elements);
2402 			return -EINVAL;
2403 		}
2404 		cqe->wr_id = rq->swq[wr_id_idx].wr_id;
2405 		cqe++;
2406 		(*budget)--;
2407 		rq->hwq.cons++;
2408 		*pcqe = cqe;
2409 
2410 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2411 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2412 			/* Add qp to flush list of the CQ */
2413 			bnxt_qplib_lock_buddy_cq(qp, cq);
2414 			__bnxt_qplib_add_flush_qp(qp);
2415 			bnxt_qplib_unlock_buddy_cq(qp, cq);
2416 		}
2417 	}
2418 
2419 done:
2420 	return rc;
2421 }
2422 
2423 static int bnxt_qplib_cq_process_res_ud(struct bnxt_qplib_cq *cq,
2424 					struct cq_res_ud *hwcqe,
2425 					struct bnxt_qplib_cqe **pcqe,
2426 					int *budget)
2427 {
2428 	struct bnxt_qplib_qp *qp;
2429 	struct bnxt_qplib_q *rq;
2430 	struct bnxt_qplib_srq *srq;
2431 	struct bnxt_qplib_cqe *cqe;
2432 	u32 wr_id_idx;
2433 	int rc = 0;
2434 
2435 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2436 				      le64_to_cpu(hwcqe->qp_handle));
2437 	if (!qp) {
2438 		dev_err(&cq->hwq.pdev->dev, "QPLIB: process_cq UD qp is NULL");
2439 		return -EINVAL;
2440 	}
2441 	if (qp->rq.flushed) {
2442 		dev_dbg(&cq->hwq.pdev->dev,
2443 			"%s: QPLIB: QP in Flush QP = %p\n", __func__, qp);
2444 		goto done;
2445 	}
2446 	cqe = *pcqe;
2447 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2448 	cqe->length = le32_to_cpu(hwcqe->length);
2449 	cqe->invrkey = le32_to_cpu(hwcqe->imm_data);
2450 	cqe->flags = le16_to_cpu(hwcqe->flags);
2451 	cqe->status = hwcqe->status;
2452 	cqe->qp_handle = (u64)(unsigned long)qp;
2453 	memcpy(cqe->smac, hwcqe->src_mac, 6);
2454 	wr_id_idx = le32_to_cpu(hwcqe->src_qp_high_srq_or_rq_wr_id)
2455 				& CQ_RES_UD_SRQ_OR_RQ_WR_ID_MASK;
2456 	cqe->src_qp = le16_to_cpu(hwcqe->src_qp_low) |
2457 				  ((le32_to_cpu(
2458 				  hwcqe->src_qp_high_srq_or_rq_wr_id) &
2459 				 CQ_RES_UD_SRC_QP_HIGH_MASK) >> 8);
2460 
2461 	if (cqe->flags & CQ_RES_RC_FLAGS_SRQ_SRQ) {
2462 		srq = qp->srq;
2463 		if (!srq)
2464 			return -EINVAL;
2465 
2466 		if (wr_id_idx > srq->hwq.max_elements) {
2467 			dev_err(&cq->hwq.pdev->dev,
2468 				"QPLIB: FP: CQ Process UD ");
2469 			dev_err(&cq->hwq.pdev->dev,
2470 				"QPLIB: wr_id idx 0x%x exceeded SRQ max 0x%x",
2471 				wr_id_idx, srq->hwq.max_elements);
2472 			return -EINVAL;
2473 		}
2474 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2475 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2476 		cqe++;
2477 		(*budget)--;
2478 		*pcqe = cqe;
2479 	} else {
2480 		rq = &qp->rq;
2481 		if (wr_id_idx > rq->hwq.max_elements) {
2482 			dev_err(&cq->hwq.pdev->dev,
2483 				"QPLIB: FP: CQ Process UD ");
2484 			dev_err(&cq->hwq.pdev->dev,
2485 				"QPLIB: wr_id idx 0x%x exceeded RQ max 0x%x",
2486 				wr_id_idx, rq->hwq.max_elements);
2487 			return -EINVAL;
2488 		}
2489 
2490 		cqe->wr_id = rq->swq[wr_id_idx].wr_id;
2491 		cqe++;
2492 		(*budget)--;
2493 		rq->hwq.cons++;
2494 		*pcqe = cqe;
2495 
2496 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2497 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2498 			/* Add qp to flush list of the CQ */
2499 			bnxt_qplib_lock_buddy_cq(qp, cq);
2500 			__bnxt_qplib_add_flush_qp(qp);
2501 			bnxt_qplib_unlock_buddy_cq(qp, cq);
2502 		}
2503 	}
2504 done:
2505 	return rc;
2506 }
2507 
2508 bool bnxt_qplib_is_cq_empty(struct bnxt_qplib_cq *cq)
2509 {
2510 	struct cq_base *hw_cqe, **hw_cqe_ptr;
2511 	unsigned long flags;
2512 	u32 sw_cons, raw_cons;
2513 	bool rc = true;
2514 
2515 	spin_lock_irqsave(&cq->hwq.lock, flags);
2516 	raw_cons = cq->hwq.cons;
2517 	sw_cons = HWQ_CMP(raw_cons, &cq->hwq);
2518 	hw_cqe_ptr = (struct cq_base **)cq->hwq.pbl_ptr;
2519 	hw_cqe = &hw_cqe_ptr[CQE_PG(sw_cons)][CQE_IDX(sw_cons)];
2520 
2521 	 /* Check for Valid bit. If the CQE is valid, return false */
2522 	rc = !CQE_CMP_VALID(hw_cqe, raw_cons, cq->hwq.max_elements);
2523 	spin_unlock_irqrestore(&cq->hwq.lock, flags);
2524 	return rc;
2525 }
2526 
2527 static int bnxt_qplib_cq_process_res_raweth_qp1(struct bnxt_qplib_cq *cq,
2528 						struct cq_res_raweth_qp1 *hwcqe,
2529 						struct bnxt_qplib_cqe **pcqe,
2530 						int *budget)
2531 {
2532 	struct bnxt_qplib_qp *qp;
2533 	struct bnxt_qplib_q *rq;
2534 	struct bnxt_qplib_srq *srq;
2535 	struct bnxt_qplib_cqe *cqe;
2536 	u32 wr_id_idx;
2537 	int rc = 0;
2538 
2539 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2540 				      le64_to_cpu(hwcqe->qp_handle));
2541 	if (!qp) {
2542 		dev_err(&cq->hwq.pdev->dev,
2543 			"QPLIB: process_cq Raw/QP1 qp is NULL");
2544 		return -EINVAL;
2545 	}
2546 	if (qp->rq.flushed) {
2547 		dev_dbg(&cq->hwq.pdev->dev,
2548 			"%s: QPLIB: QP in Flush QP = %p\n", __func__, qp);
2549 		goto done;
2550 	}
2551 	cqe = *pcqe;
2552 	cqe->opcode = hwcqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK;
2553 	cqe->flags = le16_to_cpu(hwcqe->flags);
2554 	cqe->qp_handle = (u64)(unsigned long)qp;
2555 
2556 	wr_id_idx =
2557 		le32_to_cpu(hwcqe->raweth_qp1_payload_offset_srq_or_rq_wr_id)
2558 				& CQ_RES_RAWETH_QP1_SRQ_OR_RQ_WR_ID_MASK;
2559 	cqe->src_qp = qp->id;
2560 	if (qp->id == 1 && !cqe->length) {
2561 		/* Add workaround for the length misdetection */
2562 		cqe->length = 296;
2563 	} else {
2564 		cqe->length = le16_to_cpu(hwcqe->length);
2565 	}
2566 	cqe->pkey_index = qp->pkey_index;
2567 	memcpy(cqe->smac, qp->smac, 6);
2568 
2569 	cqe->raweth_qp1_flags = le16_to_cpu(hwcqe->raweth_qp1_flags);
2570 	cqe->raweth_qp1_flags2 = le32_to_cpu(hwcqe->raweth_qp1_flags2);
2571 	cqe->raweth_qp1_metadata = le32_to_cpu(hwcqe->raweth_qp1_metadata);
2572 
2573 	if (cqe->flags & CQ_RES_RAWETH_QP1_FLAGS_SRQ_SRQ) {
2574 		srq = qp->srq;
2575 		if (!srq) {
2576 			dev_err(&cq->hwq.pdev->dev,
2577 				"QPLIB: FP: SRQ used but not defined??");
2578 			return -EINVAL;
2579 		}
2580 		if (wr_id_idx > srq->hwq.max_elements) {
2581 			dev_err(&cq->hwq.pdev->dev,
2582 				"QPLIB: FP: CQ Process Raw/QP1 ");
2583 			dev_err(&cq->hwq.pdev->dev,
2584 				"QPLIB: wr_id idx 0x%x exceeded SRQ max 0x%x",
2585 				wr_id_idx, srq->hwq.max_elements);
2586 			return -EINVAL;
2587 		}
2588 		cqe->wr_id = srq->swq[wr_id_idx].wr_id;
2589 		bnxt_qplib_release_srqe(srq, wr_id_idx);
2590 		cqe++;
2591 		(*budget)--;
2592 		*pcqe = cqe;
2593 	} else {
2594 		rq = &qp->rq;
2595 		if (wr_id_idx > rq->hwq.max_elements) {
2596 			dev_err(&cq->hwq.pdev->dev,
2597 				"QPLIB: FP: CQ Process Raw/QP1 RQ wr_id ");
2598 			dev_err(&cq->hwq.pdev->dev,
2599 				"QPLIB: ix 0x%x exceeded RQ max 0x%x",
2600 				wr_id_idx, rq->hwq.max_elements);
2601 			return -EINVAL;
2602 		}
2603 		cqe->wr_id = rq->swq[wr_id_idx].wr_id;
2604 		cqe++;
2605 		(*budget)--;
2606 		rq->hwq.cons++;
2607 		*pcqe = cqe;
2608 
2609 		if (hwcqe->status != CQ_RES_RC_STATUS_OK) {
2610 			qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2611 			/* Add qp to flush list of the CQ */
2612 			bnxt_qplib_lock_buddy_cq(qp, cq);
2613 			__bnxt_qplib_add_flush_qp(qp);
2614 			bnxt_qplib_unlock_buddy_cq(qp, cq);
2615 		}
2616 	}
2617 
2618 done:
2619 	return rc;
2620 }
2621 
2622 static int bnxt_qplib_cq_process_terminal(struct bnxt_qplib_cq *cq,
2623 					  struct cq_terminal *hwcqe,
2624 					  struct bnxt_qplib_cqe **pcqe,
2625 					  int *budget)
2626 {
2627 	struct bnxt_qplib_qp *qp;
2628 	struct bnxt_qplib_q *sq, *rq;
2629 	struct bnxt_qplib_cqe *cqe;
2630 	u32 sw_cons = 0, cqe_cons;
2631 	int rc = 0;
2632 
2633 	/* Check the Status */
2634 	if (hwcqe->status != CQ_TERMINAL_STATUS_OK)
2635 		dev_warn(&cq->hwq.pdev->dev,
2636 			 "QPLIB: FP: CQ Process Terminal Error status = 0x%x",
2637 			 hwcqe->status);
2638 
2639 	qp = (struct bnxt_qplib_qp *)((unsigned long)
2640 				      le64_to_cpu(hwcqe->qp_handle));
2641 	if (!qp) {
2642 		dev_err(&cq->hwq.pdev->dev,
2643 			"QPLIB: FP: CQ Process terminal qp is NULL");
2644 		return -EINVAL;
2645 	}
2646 
2647 	/* Must block new posting of SQ and RQ */
2648 	qp->state = CMDQ_MODIFY_QP_NEW_STATE_ERR;
2649 
2650 	sq = &qp->sq;
2651 	rq = &qp->rq;
2652 
2653 	cqe_cons = le16_to_cpu(hwcqe->sq_cons_idx);
2654 	if (cqe_cons == 0xFFFF)
2655 		goto do_rq;
2656 
2657 	if (cqe_cons > sq->hwq.max_elements) {
2658 		dev_err(&cq->hwq.pdev->dev,
2659 			"QPLIB: FP: CQ Process terminal reported ");
2660 		dev_err(&cq->hwq.pdev->dev,
2661 			"QPLIB: sq_cons_idx 0x%x which exceeded max 0x%x",
2662 			cqe_cons, sq->hwq.max_elements);
2663 		goto do_rq;
2664 	}
2665 
2666 	if (qp->sq.flushed) {
2667 		dev_dbg(&cq->hwq.pdev->dev,
2668 			"%s: QPLIB: QP in Flush QP = %p\n", __func__, qp);
2669 		goto sq_done;
2670 	}
2671 
2672 	/* Terminal CQE can also include aggregated successful CQEs prior.
2673 	 * So we must complete all CQEs from the current sq's cons to the
2674 	 * cq_cons with status OK
2675 	 */
2676 	cqe = *pcqe;
2677 	while (*budget) {
2678 		sw_cons = HWQ_CMP(sq->hwq.cons, &sq->hwq);
2679 		if (sw_cons == cqe_cons)
2680 			break;
2681 		if (sq->swq[sw_cons].flags & SQ_SEND_FLAGS_SIGNAL_COMP) {
2682 			memset(cqe, 0, sizeof(*cqe));
2683 			cqe->status = CQ_REQ_STATUS_OK;
2684 			cqe->opcode = CQ_BASE_CQE_TYPE_REQ;
2685 			cqe->qp_handle = (u64)(unsigned long)qp;
2686 			cqe->src_qp = qp->id;
2687 			cqe->wr_id = sq->swq[sw_cons].wr_id;
2688 			cqe->type = sq->swq[sw_cons].type;
2689 			cqe++;
2690 			(*budget)--;
2691 		}
2692 		sq->hwq.cons++;
2693 	}
2694 	*pcqe = cqe;
2695 	if (!(*budget) && sw_cons != cqe_cons) {
2696 		/* Out of budget */
2697 		rc = -EAGAIN;
2698 		goto sq_done;
2699 	}
2700 sq_done:
2701 	if (rc)
2702 		return rc;
2703 do_rq:
2704 	cqe_cons = le16_to_cpu(hwcqe->rq_cons_idx);
2705 	if (cqe_cons == 0xFFFF) {
2706 		goto done;
2707 	} else if (cqe_cons > rq->hwq.max_elements) {
2708 		dev_err(&cq->hwq.pdev->dev,
2709 			"QPLIB: FP: CQ Processed terminal ");
2710 		dev_err(&cq->hwq.pdev->dev,
2711 			"QPLIB: reported rq_cons_idx 0x%x exceeds max 0x%x",
2712 			cqe_cons, rq->hwq.max_elements);
2713 		goto done;
2714 	}
2715 
2716 	if (qp->rq.flushed) {
2717 		dev_dbg(&cq->hwq.pdev->dev,
2718 			"%s: QPLIB: QP in Flush QP = %p\n", __func__, qp);
2719 		rc = 0;
2720 		goto done;
2721 	}
2722 
2723 	/* Terminal CQE requires all posted RQEs to complete with FLUSHED_ERR
2724 	 * from the current rq->cons to the rq->prod regardless what the
2725 	 * rq->cons the terminal CQE indicates
2726 	 */
2727 
2728 	/* Add qp to flush list of the CQ */
2729 	bnxt_qplib_lock_buddy_cq(qp, cq);
2730 	__bnxt_qplib_add_flush_qp(qp);
2731 	bnxt_qplib_unlock_buddy_cq(qp, cq);
2732 done:
2733 	return rc;
2734 }
2735 
2736 static int bnxt_qplib_cq_process_cutoff(struct bnxt_qplib_cq *cq,
2737 					struct cq_cutoff *hwcqe)
2738 {
2739 	/* Check the Status */
2740 	if (hwcqe->status != CQ_CUTOFF_STATUS_OK) {
2741 		dev_err(&cq->hwq.pdev->dev,
2742 			"QPLIB: FP: CQ Process Cutoff Error status = 0x%x",
2743 			hwcqe->status);
2744 		return -EINVAL;
2745 	}
2746 	clear_bit(CQ_FLAGS_RESIZE_IN_PROG, &cq->flags);
2747 	wake_up_interruptible(&cq->waitq);
2748 
2749 	return 0;
2750 }
2751 
2752 int bnxt_qplib_process_flush_list(struct bnxt_qplib_cq *cq,
2753 				  struct bnxt_qplib_cqe *cqe,
2754 				  int num_cqes)
2755 {
2756 	struct bnxt_qplib_qp *qp = NULL;
2757 	u32 budget = num_cqes;
2758 	unsigned long flags;
2759 
2760 	spin_lock_irqsave(&cq->hwq.lock, flags);
2761 	list_for_each_entry(qp, &cq->sqf_head, sq_flush) {
2762 		dev_dbg(&cq->hwq.pdev->dev,
2763 			"QPLIB: FP: Flushing SQ QP= %p",
2764 			qp);
2765 		__flush_sq(&qp->sq, qp, &cqe, &budget);
2766 	}
2767 
2768 	list_for_each_entry(qp, &cq->rqf_head, rq_flush) {
2769 		dev_dbg(&cq->hwq.pdev->dev,
2770 			"QPLIB: FP: Flushing RQ QP= %p",
2771 			qp);
2772 		__flush_rq(&qp->rq, qp, &cqe, &budget);
2773 	}
2774 	spin_unlock_irqrestore(&cq->hwq.lock, flags);
2775 
2776 	return num_cqes - budget;
2777 }
2778 
2779 int bnxt_qplib_poll_cq(struct bnxt_qplib_cq *cq, struct bnxt_qplib_cqe *cqe,
2780 		       int num_cqes, struct bnxt_qplib_qp **lib_qp)
2781 {
2782 	struct cq_base *hw_cqe, **hw_cqe_ptr;
2783 	unsigned long flags;
2784 	u32 sw_cons, raw_cons;
2785 	int budget, rc = 0;
2786 
2787 	spin_lock_irqsave(&cq->hwq.lock, flags);
2788 	raw_cons = cq->hwq.cons;
2789 	budget = num_cqes;
2790 
2791 	while (budget) {
2792 		sw_cons = HWQ_CMP(raw_cons, &cq->hwq);
2793 		hw_cqe_ptr = (struct cq_base **)cq->hwq.pbl_ptr;
2794 		hw_cqe = &hw_cqe_ptr[CQE_PG(sw_cons)][CQE_IDX(sw_cons)];
2795 
2796 		/* Check for Valid bit */
2797 		if (!CQE_CMP_VALID(hw_cqe, raw_cons, cq->hwq.max_elements))
2798 			break;
2799 
2800 		/*
2801 		 * The valid test of the entry must be done first before
2802 		 * reading any further.
2803 		 */
2804 		dma_rmb();
2805 		/* From the device's respective CQE format to qplib_wc*/
2806 		switch (hw_cqe->cqe_type_toggle & CQ_BASE_CQE_TYPE_MASK) {
2807 		case CQ_BASE_CQE_TYPE_REQ:
2808 			rc = bnxt_qplib_cq_process_req(cq,
2809 						       (struct cq_req *)hw_cqe,
2810 						       &cqe, &budget,
2811 						       sw_cons, lib_qp);
2812 			break;
2813 		case CQ_BASE_CQE_TYPE_RES_RC:
2814 			rc = bnxt_qplib_cq_process_res_rc(cq,
2815 							  (struct cq_res_rc *)
2816 							  hw_cqe, &cqe,
2817 							  &budget);
2818 			break;
2819 		case CQ_BASE_CQE_TYPE_RES_UD:
2820 			rc = bnxt_qplib_cq_process_res_ud
2821 					(cq, (struct cq_res_ud *)hw_cqe, &cqe,
2822 					 &budget);
2823 			break;
2824 		case CQ_BASE_CQE_TYPE_RES_RAWETH_QP1:
2825 			rc = bnxt_qplib_cq_process_res_raweth_qp1
2826 					(cq, (struct cq_res_raweth_qp1 *)
2827 					 hw_cqe, &cqe, &budget);
2828 			break;
2829 		case CQ_BASE_CQE_TYPE_TERMINAL:
2830 			rc = bnxt_qplib_cq_process_terminal
2831 					(cq, (struct cq_terminal *)hw_cqe,
2832 					 &cqe, &budget);
2833 			break;
2834 		case CQ_BASE_CQE_TYPE_CUT_OFF:
2835 			bnxt_qplib_cq_process_cutoff
2836 					(cq, (struct cq_cutoff *)hw_cqe);
2837 			/* Done processing this CQ */
2838 			goto exit;
2839 		default:
2840 			dev_err(&cq->hwq.pdev->dev,
2841 				"QPLIB: process_cq unknown type 0x%lx",
2842 				hw_cqe->cqe_type_toggle &
2843 				CQ_BASE_CQE_TYPE_MASK);
2844 			rc = -EINVAL;
2845 			break;
2846 		}
2847 		if (rc < 0) {
2848 			if (rc == -EAGAIN)
2849 				break;
2850 			/* Error while processing the CQE, just skip to the
2851 			 * next one
2852 			 */
2853 			dev_err(&cq->hwq.pdev->dev,
2854 				"QPLIB: process_cqe error rc = 0x%x", rc);
2855 		}
2856 		raw_cons++;
2857 	}
2858 	if (cq->hwq.cons != raw_cons) {
2859 		cq->hwq.cons = raw_cons;
2860 		bnxt_qplib_arm_cq(cq, DBR_DBR_TYPE_CQ);
2861 	}
2862 exit:
2863 	spin_unlock_irqrestore(&cq->hwq.lock, flags);
2864 	return num_cqes - budget;
2865 }
2866 
2867 void bnxt_qplib_req_notify_cq(struct bnxt_qplib_cq *cq, u32 arm_type)
2868 {
2869 	unsigned long flags;
2870 
2871 	spin_lock_irqsave(&cq->hwq.lock, flags);
2872 	if (arm_type)
2873 		bnxt_qplib_arm_cq(cq, arm_type);
2874 	/* Using cq->arm_state variable to track whether to issue cq handler */
2875 	atomic_set(&cq->arm_state, 1);
2876 	spin_unlock_irqrestore(&cq->hwq.lock, flags);
2877 }
2878 
2879 void bnxt_qplib_flush_cqn_wq(struct bnxt_qplib_qp *qp)
2880 {
2881 	flush_workqueue(qp->scq->nq->cqn_wq);
2882 	if (qp->scq != qp->rcq)
2883 		flush_workqueue(qp->rcq->nq->cqn_wq);
2884 }
2885