xref: /linux/drivers/infiniband/hw/bnxt_re/main.c (revision 68a052239fc4b351e961f698b824f7654a346091)
1 /*
2  * Broadcom NetXtreme-E RoCE driver.
3  *
4  * Copyright (c) 2016 - 2017, Broadcom. All rights reserved.  The term
5  * Broadcom refers to Broadcom Limited and/or its subsidiaries.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * BSD license below:
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  *
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in
21  *    the documentation and/or other materials provided with the
22  *    distribution.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS''
25  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
26  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS
28  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
32  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
33  * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
34  * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Description: Main component of the bnxt_re driver
37  */
38 
39 #include <linux/module.h>
40 #include <linux/netdevice.h>
41 #include <linux/ethtool.h>
42 #include <linux/mutex.h>
43 #include <linux/list.h>
44 #include <linux/rculist.h>
45 #include <linux/spinlock.h>
46 #include <linux/pci.h>
47 #include <net/dcbnl.h>
48 #include <net/ipv6.h>
49 #include <net/addrconf.h>
50 #include <linux/if_ether.h>
51 #include <linux/auxiliary_bus.h>
52 
53 #include <rdma/ib_verbs.h>
54 #include <rdma/ib_user_verbs.h>
55 #include <rdma/ib_umem.h>
56 #include <rdma/ib_addr.h>
57 #include <linux/hashtable.h>
58 
59 #include "bnxt_ulp.h"
60 #include "roce_hsi.h"
61 #include "qplib_res.h"
62 #include "qplib_sp.h"
63 #include "qplib_fp.h"
64 #include "qplib_rcfw.h"
65 #include "bnxt_re.h"
66 #include "ib_verbs.h"
67 #include <rdma/bnxt_re-abi.h>
68 #include "bnxt.h"
69 #include "hw_counters.h"
70 #include "debugfs.h"
71 
72 static char version[] =
73 		BNXT_RE_DESC "\n";
74 
75 MODULE_AUTHOR("Eddie Wai <eddie.wai@broadcom.com>");
76 MODULE_DESCRIPTION(BNXT_RE_DESC);
77 MODULE_LICENSE("Dual BSD/GPL");
78 
79 /* globals */
80 static DEFINE_MUTEX(bnxt_re_mutex);
81 
82 static int bnxt_re_hwrm_qcaps(struct bnxt_re_dev *rdev);
83 static int bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev);
84 
85 static int bnxt_re_hwrm_qcfg(struct bnxt_re_dev *rdev, u32 *db_len,
86 			     u32 *offset);
87 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
88 				   u8 port_num, enum ib_event_type event);
89 static void bnxt_re_set_db_offset(struct bnxt_re_dev *rdev)
90 {
91 	struct bnxt_qplib_chip_ctx *cctx;
92 	struct bnxt_en_dev *en_dev;
93 	struct bnxt_qplib_res *res;
94 	u32 l2db_len = 0;
95 	u32 offset = 0;
96 	u32 barlen;
97 	int rc;
98 
99 	res = &rdev->qplib_res;
100 	en_dev = rdev->en_dev;
101 	cctx = rdev->chip_ctx;
102 
103 	/* Issue qcfg */
104 	rc = bnxt_re_hwrm_qcfg(rdev, &l2db_len, &offset);
105 	if (rc)
106 		dev_info(rdev_to_dev(rdev),
107 			 "Couldn't get DB bar size, Low latency framework is disabled\n");
108 	/* set register offsets for both UC and WC */
109 	if (bnxt_qplib_is_chip_gen_p7(cctx)) {
110 		res->dpi_tbl.ucreg.offset = offset;
111 		res->dpi_tbl.wcreg.offset = en_dev->l2_db_size;
112 	} else {
113 		res->dpi_tbl.ucreg.offset = res->is_vf ? BNXT_QPLIB_DBR_VF_DB_OFFSET :
114 							 BNXT_QPLIB_DBR_PF_DB_OFFSET;
115 		res->dpi_tbl.wcreg.offset = res->dpi_tbl.ucreg.offset;
116 	}
117 
118 	/* If WC mapping is disabled by L2 driver then en_dev->l2_db_size
119 	 * is equal to the DB-Bar actual size. This indicates that L2
120 	 * is mapping entire bar as UC-. RoCE driver can't enable WC mapping
121 	 * in such cases and DB-push will be disabled.
122 	 */
123 	barlen = pci_resource_len(res->pdev, RCFW_DBR_PCI_BAR_REGION);
124 	if (cctx->modes.db_push && l2db_len && en_dev->l2_db_size != barlen) {
125 		res->dpi_tbl.wcreg.offset = en_dev->l2_db_size;
126 		dev_info(rdev_to_dev(rdev),  "Low latency framework is enabled\n");
127 	}
128 }
129 
130 static void bnxt_re_set_drv_mode(struct bnxt_re_dev *rdev)
131 {
132 	struct bnxt_qplib_chip_ctx *cctx;
133 
134 	cctx = rdev->chip_ctx;
135 	cctx->modes.wqe_mode = bnxt_qplib_is_chip_gen_p7(rdev->chip_ctx) ?
136 			       BNXT_QPLIB_WQE_MODE_VARIABLE : BNXT_QPLIB_WQE_MODE_STATIC;
137 	if (bnxt_re_hwrm_qcaps(rdev))
138 		dev_err(rdev_to_dev(rdev),
139 			"Failed to query hwrm qcaps\n");
140 	if (bnxt_qplib_is_chip_gen_p7(rdev->chip_ctx)) {
141 		cctx->modes.toggle_bits |= BNXT_QPLIB_CQ_TOGGLE_BIT;
142 		cctx->modes.toggle_bits |= BNXT_QPLIB_SRQ_TOGGLE_BIT;
143 	}
144 }
145 
146 static void bnxt_re_destroy_chip_ctx(struct bnxt_re_dev *rdev)
147 {
148 	struct bnxt_qplib_chip_ctx *chip_ctx;
149 
150 	if (!rdev->chip_ctx)
151 		return;
152 
153 	kfree(rdev->dev_attr);
154 	rdev->dev_attr = NULL;
155 
156 	chip_ctx = rdev->chip_ctx;
157 	rdev->chip_ctx = NULL;
158 	rdev->rcfw.res = NULL;
159 	rdev->qplib_res.cctx = NULL;
160 	rdev->qplib_res.pdev = NULL;
161 	rdev->qplib_res.netdev = NULL;
162 	kfree(chip_ctx);
163 }
164 
165 static int bnxt_re_setup_chip_ctx(struct bnxt_re_dev *rdev)
166 {
167 	struct bnxt_qplib_chip_ctx *chip_ctx;
168 	struct bnxt_en_dev *en_dev;
169 	int rc = -ENOMEM;
170 
171 	en_dev = rdev->en_dev;
172 
173 	rdev->qplib_res.pdev = en_dev->pdev;
174 	chip_ctx = kzalloc(sizeof(*chip_ctx), GFP_KERNEL);
175 	if (!chip_ctx)
176 		return -ENOMEM;
177 	chip_ctx->chip_num = en_dev->chip_num;
178 	chip_ctx->hw_stats_size = en_dev->hw_ring_stats_size;
179 
180 	rdev->chip_ctx = chip_ctx;
181 	/* rest members to follow eventually */
182 
183 	rdev->qplib_res.cctx = rdev->chip_ctx;
184 	rdev->rcfw.res = &rdev->qplib_res;
185 	rdev->dev_attr = kzalloc(sizeof(*rdev->dev_attr), GFP_KERNEL);
186 	if (!rdev->dev_attr)
187 		goto free_chip_ctx;
188 	rdev->qplib_res.dattr = rdev->dev_attr;
189 	rdev->qplib_res.is_vf = BNXT_EN_VF(en_dev);
190 	rdev->qplib_res.en_dev = en_dev;
191 
192 	rc = bnxt_re_query_hwrm_intf_version(rdev);
193 	if (rc)
194 		goto free_dev_attr;
195 
196 	bnxt_re_set_drv_mode(rdev);
197 
198 	bnxt_re_set_db_offset(rdev);
199 	rc = bnxt_qplib_map_db_bar(&rdev->qplib_res);
200 	if (rc)
201 		goto free_dev_attr;
202 
203 	if (bnxt_qplib_determine_atomics(en_dev->pdev))
204 		ibdev_info(&rdev->ibdev,
205 			   "platform doesn't support global atomics.");
206 	return 0;
207 free_dev_attr:
208 	kfree(rdev->dev_attr);
209 	rdev->dev_attr = NULL;
210 free_chip_ctx:
211 	kfree(rdev->chip_ctx);
212 	rdev->chip_ctx = NULL;
213 	return rc;
214 }
215 
216 /* SR-IOV helper functions */
217 
218 static void bnxt_re_get_sriov_func_type(struct bnxt_re_dev *rdev)
219 {
220 	if (BNXT_EN_VF(rdev->en_dev))
221 		rdev->is_virtfn = 1;
222 }
223 
224 /* Set the maximum number of each resource that the driver actually wants
225  * to allocate. This may be up to the maximum number the firmware has
226  * reserved for the function. The driver may choose to allocate fewer
227  * resources than the firmware maximum.
228  */
229 static void bnxt_re_limit_pf_res(struct bnxt_re_dev *rdev)
230 {
231 	struct bnxt_qplib_dev_attr *attr;
232 	struct bnxt_qplib_ctx *ctx;
233 	int i;
234 
235 	attr = rdev->dev_attr;
236 	ctx = &rdev->qplib_ctx;
237 
238 	ctx->qpc_count = min_t(u32, BNXT_RE_MAX_QPC_COUNT,
239 			       attr->max_qp);
240 	ctx->mrw_count = BNXT_RE_MAX_MRW_COUNT_256K;
241 	/* Use max_mr from fw since max_mrw does not get set */
242 	ctx->mrw_count = min_t(u32, ctx->mrw_count, attr->max_mr);
243 	ctx->srqc_count = min_t(u32, BNXT_RE_MAX_SRQC_COUNT,
244 				attr->max_srq);
245 	ctx->cq_count = min_t(u32, BNXT_RE_MAX_CQ_COUNT, attr->max_cq);
246 	if (!bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx))
247 		for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
248 			rdev->qplib_ctx.tqm_ctx.qcount[i] =
249 			rdev->dev_attr->tqm_alloc_reqs[i];
250 }
251 
252 static void bnxt_re_limit_vf_res(struct bnxt_qplib_ctx *qplib_ctx, u32 num_vf)
253 {
254 	struct bnxt_qplib_vf_res *vf_res;
255 	u32 mrws = 0;
256 	u32 vf_pct;
257 	u32 nvfs;
258 
259 	vf_res = &qplib_ctx->vf_res;
260 	/*
261 	 * Reserve a set of resources for the PF. Divide the remaining
262 	 * resources among the VFs
263 	 */
264 	vf_pct = 100 - BNXT_RE_PCT_RSVD_FOR_PF;
265 	nvfs = num_vf;
266 	num_vf = 100 * num_vf;
267 	vf_res->max_qp_per_vf = (qplib_ctx->qpc_count * vf_pct) / num_vf;
268 	vf_res->max_srq_per_vf = (qplib_ctx->srqc_count * vf_pct) / num_vf;
269 	vf_res->max_cq_per_vf = (qplib_ctx->cq_count * vf_pct) / num_vf;
270 	/*
271 	 * The driver allows many more MRs than other resources. If the
272 	 * firmware does also, then reserve a fixed amount for the PF and
273 	 * divide the rest among VFs. VFs may use many MRs for NFS
274 	 * mounts, ISER, NVME applications, etc. If the firmware severely
275 	 * restricts the number of MRs, then let PF have half and divide
276 	 * the rest among VFs, as for the other resource types.
277 	 */
278 	if (qplib_ctx->mrw_count < BNXT_RE_MAX_MRW_COUNT_64K) {
279 		mrws = qplib_ctx->mrw_count * vf_pct;
280 		nvfs = num_vf;
281 	} else {
282 		mrws = qplib_ctx->mrw_count - BNXT_RE_RESVD_MR_FOR_PF;
283 	}
284 	vf_res->max_mrw_per_vf = (mrws / nvfs);
285 	vf_res->max_gid_per_vf = BNXT_RE_MAX_GID_PER_VF;
286 }
287 
288 static void bnxt_re_set_resource_limits(struct bnxt_re_dev *rdev)
289 {
290 	u32 num_vfs;
291 
292 	memset(&rdev->qplib_ctx.vf_res, 0, sizeof(struct bnxt_qplib_vf_res));
293 	bnxt_re_limit_pf_res(rdev);
294 
295 	num_vfs =  bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx) ?
296 			BNXT_RE_GEN_P5_MAX_VF : rdev->num_vfs;
297 	if (num_vfs)
298 		bnxt_re_limit_vf_res(&rdev->qplib_ctx, num_vfs);
299 }
300 
301 static void bnxt_re_vf_res_config(struct bnxt_re_dev *rdev)
302 {
303 	/*
304 	 * Use the total VF count since the actual VF count may not be
305 	 * available at this point.
306 	 */
307 	rdev->num_vfs = pci_sriov_get_totalvfs(rdev->en_dev->pdev);
308 	if (!rdev->num_vfs)
309 		return;
310 
311 	bnxt_re_set_resource_limits(rdev);
312 	bnxt_qplib_set_func_resources(&rdev->qplib_res, &rdev->rcfw,
313 				      &rdev->qplib_ctx);
314 }
315 
316 struct bnxt_re_dcb_work {
317 	struct work_struct work;
318 	struct bnxt_re_dev *rdev;
319 	struct hwrm_async_event_cmpl cmpl;
320 };
321 
322 static bool bnxt_re_is_qp1_qp(struct bnxt_re_qp *qp)
323 {
324 	return qp->ib_qp.qp_type == IB_QPT_GSI;
325 }
326 
327 static struct bnxt_re_qp *bnxt_re_get_qp1_qp(struct bnxt_re_dev *rdev)
328 {
329 	struct bnxt_re_qp *qp;
330 
331 	mutex_lock(&rdev->qp_lock);
332 	list_for_each_entry(qp, &rdev->qp_list, list) {
333 		if (bnxt_re_is_qp1_qp(qp)) {
334 			mutex_unlock(&rdev->qp_lock);
335 			return qp;
336 		}
337 	}
338 	mutex_unlock(&rdev->qp_lock);
339 	return NULL;
340 }
341 
342 static int bnxt_re_update_qp1_tos_dscp(struct bnxt_re_dev *rdev)
343 {
344 	struct bnxt_re_qp *qp;
345 
346 	if (!bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx))
347 		return 0;
348 
349 	qp = bnxt_re_get_qp1_qp(rdev);
350 	if (!qp)
351 		return 0;
352 
353 	qp->qplib_qp.modify_flags = CMDQ_MODIFY_QP_MODIFY_MASK_TOS_DSCP;
354 	qp->qplib_qp.tos_dscp = rdev->cc_param.qp1_tos_dscp;
355 
356 	return bnxt_qplib_modify_qp(&rdev->qplib_res, &qp->qplib_qp);
357 }
358 
359 static void bnxt_re_init_dcb_wq(struct bnxt_re_dev *rdev)
360 {
361 	rdev->dcb_wq = create_singlethread_workqueue("bnxt_re_dcb_wq");
362 }
363 
364 static void bnxt_re_uninit_dcb_wq(struct bnxt_re_dev *rdev)
365 {
366 	if (!rdev->dcb_wq)
367 		return;
368 	destroy_workqueue(rdev->dcb_wq);
369 }
370 
371 static void bnxt_re_dcb_wq_task(struct work_struct *work)
372 {
373 	struct bnxt_re_dcb_work *dcb_work =
374 		container_of(work, struct bnxt_re_dcb_work, work);
375 	struct bnxt_re_dev *rdev = dcb_work->rdev;
376 	struct bnxt_qplib_cc_param *cc_param;
377 	int rc;
378 
379 	if (!rdev)
380 		goto free_dcb;
381 
382 	cc_param = &rdev->cc_param;
383 	rc = bnxt_qplib_query_cc_param(&rdev->qplib_res, cc_param);
384 	if (rc) {
385 		ibdev_dbg(&rdev->ibdev, "Failed to query ccparam rc:%d", rc);
386 		goto free_dcb;
387 	}
388 	if (cc_param->qp1_tos_dscp != cc_param->tos_dscp) {
389 		cc_param->qp1_tos_dscp = cc_param->tos_dscp;
390 		rc = bnxt_re_update_qp1_tos_dscp(rdev);
391 		if (rc) {
392 			ibdev_dbg(&rdev->ibdev, "%s: Failed to modify QP1 rc:%d",
393 				  __func__, rc);
394 			goto free_dcb;
395 		}
396 	}
397 
398 free_dcb:
399 	kfree(dcb_work);
400 }
401 
402 static void bnxt_re_async_notifier(void *handle, struct hwrm_async_event_cmpl *cmpl)
403 {
404 	struct bnxt_re_en_dev_info *en_info = auxiliary_get_drvdata(handle);
405 	struct bnxt_re_dcb_work *dcb_work;
406 	struct bnxt_re_dev *rdev;
407 	u32 data1, data2;
408 	u16 event_id;
409 
410 	rdev = en_info->rdev;
411 	if (!rdev)
412 		return;
413 
414 	event_id = le16_to_cpu(cmpl->event_id);
415 	data1 = le32_to_cpu(cmpl->event_data1);
416 	data2 = le32_to_cpu(cmpl->event_data2);
417 
418 	ibdev_dbg(&rdev->ibdev, "Async event_id = %d data1 = %d data2 = %d",
419 		  event_id, data1, data2);
420 
421 	switch (event_id) {
422 	case ASYNC_EVENT_CMPL_EVENT_ID_DCB_CONFIG_CHANGE:
423 		dcb_work = kzalloc(sizeof(*dcb_work), GFP_ATOMIC);
424 		if (!dcb_work)
425 			break;
426 
427 		dcb_work->rdev = rdev;
428 		memcpy(&dcb_work->cmpl, cmpl, sizeof(*cmpl));
429 		INIT_WORK(&dcb_work->work, bnxt_re_dcb_wq_task);
430 		queue_work(rdev->dcb_wq, &dcb_work->work);
431 		break;
432 	default:
433 		break;
434 	}
435 }
436 
437 static void bnxt_re_stop_irq(void *handle, bool reset)
438 {
439 	struct bnxt_re_en_dev_info *en_info = auxiliary_get_drvdata(handle);
440 	struct bnxt_qplib_rcfw *rcfw;
441 	struct bnxt_re_dev *rdev;
442 	struct bnxt_qplib_nq *nq;
443 	int indx;
444 
445 	rdev = en_info->rdev;
446 	if (!rdev)
447 		return;
448 	rcfw = &rdev->rcfw;
449 
450 	if (reset) {
451 		set_bit(ERR_DEVICE_DETACHED, &rdev->rcfw.cmdq.flags);
452 		set_bit(BNXT_RE_FLAG_ERR_DEVICE_DETACHED, &rdev->flags);
453 		wake_up_all(&rdev->rcfw.cmdq.waitq);
454 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
455 				       IB_EVENT_DEVICE_FATAL);
456 	}
457 
458 	for (indx = BNXT_RE_NQ_IDX; indx < rdev->nqr->num_msix; indx++) {
459 		nq = &rdev->nqr->nq[indx - 1];
460 		bnxt_qplib_nq_stop_irq(nq, false);
461 	}
462 
463 	bnxt_qplib_rcfw_stop_irq(rcfw, false);
464 }
465 
466 static void bnxt_re_start_irq(void *handle, struct bnxt_msix_entry *ent)
467 {
468 	struct bnxt_re_en_dev_info *en_info = auxiliary_get_drvdata(handle);
469 	struct bnxt_msix_entry *msix_ent;
470 	struct bnxt_qplib_rcfw *rcfw;
471 	struct bnxt_re_dev *rdev;
472 	struct bnxt_qplib_nq *nq;
473 	int indx, rc;
474 
475 	rdev = en_info->rdev;
476 	if (!rdev)
477 		return;
478 	msix_ent = rdev->nqr->msix_entries;
479 	rcfw = &rdev->rcfw;
480 	if (!ent) {
481 		/* Not setting the f/w timeout bit in rcfw.
482 		 * During the driver unload the first command
483 		 * to f/w will timeout and that will set the
484 		 * timeout bit.
485 		 */
486 		ibdev_err(&rdev->ibdev, "Failed to re-start IRQs\n");
487 		return;
488 	}
489 
490 	/* Vectors may change after restart, so update with new vectors
491 	 * in device sctructure.
492 	 */
493 	for (indx = 0; indx < rdev->nqr->num_msix; indx++)
494 		rdev->nqr->msix_entries[indx].vector = ent[indx].vector;
495 
496 	rc = bnxt_qplib_rcfw_start_irq(rcfw, msix_ent[BNXT_RE_AEQ_IDX].vector,
497 				       false);
498 	if (rc) {
499 		ibdev_warn(&rdev->ibdev, "Failed to reinit CREQ\n");
500 		return;
501 	}
502 	for (indx = BNXT_RE_NQ_IDX ; indx < rdev->nqr->num_msix; indx++) {
503 		nq = &rdev->nqr->nq[indx - 1];
504 		rc = bnxt_qplib_nq_start_irq(nq, indx - 1,
505 					     msix_ent[indx].vector, false);
506 		if (rc) {
507 			ibdev_warn(&rdev->ibdev, "Failed to reinit NQ index %d\n",
508 				   indx - 1);
509 			return;
510 		}
511 	}
512 }
513 
514 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
515 	.ulp_async_notifier = bnxt_re_async_notifier,
516 	.ulp_irq_stop = bnxt_re_stop_irq,
517 	.ulp_irq_restart = bnxt_re_start_irq
518 };
519 
520 /* RoCE -> Net driver */
521 
522 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
523 {
524 	struct bnxt_en_dev *en_dev;
525 
526 	en_dev = rdev->en_dev;
527 	return bnxt_register_dev(en_dev, &bnxt_re_ulp_ops, rdev->adev);
528 }
529 
530 static void bnxt_re_init_hwrm_hdr(struct input *hdr, u16 opcd)
531 {
532 	hdr->req_type = cpu_to_le16(opcd);
533 	hdr->cmpl_ring = cpu_to_le16(-1);
534 	hdr->target_id = cpu_to_le16(-1);
535 }
536 
537 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
538 				int msg_len, void *resp, int resp_max_len,
539 				int timeout)
540 {
541 	fw_msg->msg = msg;
542 	fw_msg->msg_len = msg_len;
543 	fw_msg->resp = resp;
544 	fw_msg->resp_max_len = resp_max_len;
545 	fw_msg->timeout = timeout;
546 }
547 
548 void bnxt_re_hwrm_free_vnic(struct bnxt_re_dev *rdev)
549 {
550 	struct bnxt_en_dev *en_dev = rdev->en_dev;
551 	struct hwrm_vnic_free_input req = {};
552 	struct bnxt_fw_msg fw_msg = {};
553 	int rc;
554 
555 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_VNIC_FREE);
556 
557 	req.vnic_id = cpu_to_le32(rdev->mirror_vnic_id);
558 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), NULL,
559 			    0, BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
560 	rc = bnxt_send_msg(en_dev, &fw_msg);
561 	if (rc)
562 		ibdev_dbg(&rdev->ibdev,
563 			  "Failed to free vnic, rc = %d\n", rc);
564 }
565 
566 int bnxt_re_hwrm_alloc_vnic(struct bnxt_re_dev *rdev)
567 {
568 	struct bnxt_en_dev *en_dev = rdev->en_dev;
569 	struct hwrm_vnic_alloc_output resp = {};
570 	struct hwrm_vnic_alloc_input req = {};
571 	struct bnxt_fw_msg fw_msg = {};
572 	int rc;
573 
574 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_VNIC_ALLOC);
575 
576 	req.vnic_id = cpu_to_le16(rdev->mirror_vnic_id);
577 	req.flags = cpu_to_le32(VNIC_ALLOC_REQ_FLAGS_VNIC_ID_VALID);
578 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
579 			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
580 	rc = bnxt_send_msg(en_dev, &fw_msg);
581 	if (rc)
582 		ibdev_dbg(&rdev->ibdev,
583 			  "Failed to alloc vnic, rc = %d\n", rc);
584 
585 	return rc;
586 }
587 
588 int bnxt_re_hwrm_cfg_vnic(struct bnxt_re_dev *rdev, u32 qp_id)
589 {
590 	struct bnxt_en_dev *en_dev = rdev->en_dev;
591 	struct hwrm_vnic_cfg_input req = {};
592 	struct bnxt_fw_msg fw_msg = {};
593 	int rc;
594 
595 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_VNIC_CFG);
596 
597 	req.flags = cpu_to_le32(VNIC_CFG_REQ_FLAGS_ROCE_ONLY_VNIC_MODE);
598 	req.enables = cpu_to_le32(VNIC_CFG_REQ_ENABLES_RAW_QP_ID |
599 				  VNIC_CFG_REQ_ENABLES_MRU);
600 	req.vnic_id = cpu_to_le16(rdev->mirror_vnic_id);
601 	req.raw_qp_id = cpu_to_le32(qp_id);
602 	req.mru = cpu_to_le16(rdev->netdev->mtu + VLAN_ETH_HLEN);
603 
604 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), NULL,
605 			    0, BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
606 	rc = bnxt_send_msg(en_dev, &fw_msg);
607 	if (rc)
608 		ibdev_dbg(&rdev->ibdev,
609 			  "Failed to cfg vnic, rc = %d\n", rc);
610 
611 	return rc;
612 }
613 
614 /* Query device config using common hwrm */
615 static int bnxt_re_hwrm_qcfg(struct bnxt_re_dev *rdev, u32 *db_len,
616 			     u32 *offset)
617 {
618 	struct bnxt_en_dev *en_dev = rdev->en_dev;
619 	struct hwrm_func_qcfg_output resp = {0};
620 	struct hwrm_func_qcfg_input req = {0};
621 	struct bnxt_fw_msg fw_msg = {};
622 	int rc;
623 
624 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_FUNC_QCFG);
625 	req.fid = cpu_to_le16(0xffff);
626 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
627 			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
628 	rc = bnxt_send_msg(en_dev, &fw_msg);
629 	if (!rc) {
630 		*db_len = PAGE_ALIGN(le16_to_cpu(resp.l2_doorbell_bar_size_kb) * 1024);
631 		*offset = PAGE_ALIGN(le16_to_cpu(resp.legacy_l2_db_size_kb) * 1024);
632 		rdev->mirror_vnic_id = le16_to_cpu(resp.mirror_vnic_id);
633 	}
634 	return rc;
635 }
636 
637 /* Query function capabilities using common hwrm */
638 int bnxt_re_hwrm_qcaps(struct bnxt_re_dev *rdev)
639 {
640 	struct bnxt_en_dev *en_dev = rdev->en_dev;
641 	struct hwrm_func_qcaps_output resp = {};
642 	struct hwrm_func_qcaps_input req = {};
643 	struct bnxt_qplib_chip_ctx *cctx;
644 	struct bnxt_fw_msg fw_msg = {};
645 	u32 flags_ext2;
646 	int rc;
647 
648 	cctx = rdev->chip_ctx;
649 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_FUNC_QCAPS);
650 	req.fid = cpu_to_le16(0xffff);
651 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
652 			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
653 
654 	rc = bnxt_send_msg(en_dev, &fw_msg);
655 	if (rc)
656 		return rc;
657 	cctx->modes.db_push = le32_to_cpu(resp.flags) & FUNC_QCAPS_RESP_FLAGS_WCB_PUSH_MODE;
658 
659 	flags_ext2 = le32_to_cpu(resp.flags_ext2);
660 	cctx->modes.dbr_pacing = flags_ext2 & FUNC_QCAPS_RESP_FLAGS_EXT2_DBR_PACING_EXT_SUPPORTED ||
661 				 flags_ext2 & FUNC_QCAPS_RESP_FLAGS_EXT2_DBR_PACING_V0_SUPPORTED;
662 	cctx->modes.roce_mirror = !!(le32_to_cpu(resp.flags_ext3) &
663 				     FUNC_QCAPS_RESP_FLAGS_EXT3_MIRROR_ON_ROCE_SUPPORTED);
664 	return 0;
665 }
666 
667 static int bnxt_re_hwrm_dbr_pacing_qcfg(struct bnxt_re_dev *rdev)
668 {
669 	struct bnxt_qplib_db_pacing_data *pacing_data = rdev->qplib_res.pacing_data;
670 	struct hwrm_func_dbr_pacing_qcfg_output resp = {};
671 	struct hwrm_func_dbr_pacing_qcfg_input req = {};
672 	struct bnxt_en_dev *en_dev = rdev->en_dev;
673 	struct bnxt_qplib_chip_ctx *cctx;
674 	struct bnxt_fw_msg fw_msg = {};
675 	int rc;
676 
677 	cctx = rdev->chip_ctx;
678 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_FUNC_DBR_PACING_QCFG);
679 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
680 			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
681 	rc = bnxt_send_msg(en_dev, &fw_msg);
682 	if (rc)
683 		return rc;
684 
685 	if ((le32_to_cpu(resp.dbr_stat_db_fifo_reg) &
686 	    FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_MASK) ==
687 		FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_GRC)
688 		cctx->dbr_stat_db_fifo =
689 			le32_to_cpu(resp.dbr_stat_db_fifo_reg) &
690 			~FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_MASK;
691 
692 	pacing_data->fifo_max_depth = le32_to_cpu(resp.dbr_stat_db_max_fifo_depth);
693 	if (!pacing_data->fifo_max_depth)
694 		pacing_data->fifo_max_depth = BNXT_RE_MAX_FIFO_DEPTH(cctx);
695 	pacing_data->fifo_room_mask = le32_to_cpu(resp.dbr_stat_db_fifo_reg_fifo_room_mask);
696 	pacing_data->fifo_room_shift = resp.dbr_stat_db_fifo_reg_fifo_room_shift;
697 
698 	return 0;
699 }
700 
701 /* Update the pacing tunable parameters to the default values */
702 static void bnxt_re_set_default_pacing_data(struct bnxt_re_dev *rdev)
703 {
704 	struct bnxt_qplib_db_pacing_data *pacing_data = rdev->qplib_res.pacing_data;
705 
706 	pacing_data->do_pacing = rdev->pacing.dbr_def_do_pacing;
707 	pacing_data->pacing_th = rdev->pacing.pacing_algo_th;
708 	pacing_data->alarm_th =
709 		pacing_data->pacing_th * BNXT_RE_PACING_ALARM_TH_MULTIPLE;
710 }
711 
712 static u32 __get_fifo_occupancy(struct bnxt_re_dev *rdev)
713 {
714 	struct bnxt_qplib_db_pacing_data *pacing_data = rdev->qplib_res.pacing_data;
715 	u32 read_val, fifo_occup;
716 
717 	read_val = readl(rdev->en_dev->bar0 + rdev->pacing.dbr_db_fifo_reg_off);
718 	fifo_occup = pacing_data->fifo_max_depth -
719 		     ((read_val & pacing_data->fifo_room_mask) >>
720 		      pacing_data->fifo_room_shift);
721 	return fifo_occup;
722 }
723 
724 static bool is_dbr_fifo_full(struct bnxt_re_dev *rdev)
725 {
726 	u32 max_occup, fifo_occup;
727 
728 	fifo_occup = __get_fifo_occupancy(rdev);
729 	max_occup = BNXT_RE_MAX_FIFO_DEPTH(rdev->chip_ctx) - 1;
730 	if (fifo_occup == max_occup)
731 		return true;
732 
733 	return false;
734 }
735 
736 static void __wait_for_fifo_occupancy_below_th(struct bnxt_re_dev *rdev)
737 {
738 	struct bnxt_qplib_db_pacing_data *pacing_data = rdev->qplib_res.pacing_data;
739 	u32 retry_fifo_check = 1000;
740 	u32 fifo_occup;
741 
742 	/* loop shouldn't run infintely as the occupancy usually goes
743 	 * below pacing algo threshold as soon as pacing kicks in.
744 	 */
745 	while (1) {
746 		fifo_occup = __get_fifo_occupancy(rdev);
747 		/* Fifo occupancy cannot be greater the MAX FIFO depth */
748 		if (fifo_occup > pacing_data->fifo_max_depth)
749 			break;
750 
751 		if (fifo_occup < pacing_data->pacing_th)
752 			break;
753 		if (!retry_fifo_check--) {
754 			dev_info_once(rdev_to_dev(rdev),
755 				      "%s: fifo_occup = 0x%xfifo_max_depth = 0x%x pacing_th = 0x%x\n",
756 				      __func__, fifo_occup, pacing_data->fifo_max_depth,
757 					pacing_data->pacing_th);
758 			break;
759 		}
760 
761 	}
762 }
763 
764 static void bnxt_re_db_fifo_check(struct work_struct *work)
765 {
766 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
767 			dbq_fifo_check_work);
768 	struct bnxt_qplib_db_pacing_data *pacing_data;
769 	u32 pacing_save;
770 
771 	if (!mutex_trylock(&rdev->pacing.dbq_lock))
772 		return;
773 	pacing_data = rdev->qplib_res.pacing_data;
774 	pacing_save = rdev->pacing.do_pacing_save;
775 	__wait_for_fifo_occupancy_below_th(rdev);
776 	cancel_delayed_work_sync(&rdev->dbq_pacing_work);
777 	if (pacing_save > rdev->pacing.dbr_def_do_pacing) {
778 		/* Double the do_pacing value during the congestion */
779 		pacing_save = pacing_save << 1;
780 	} else {
781 		/*
782 		 * when a new congestion is detected increase the do_pacing
783 		 * by 8 times. And also increase the pacing_th by 4 times. The
784 		 * reason to increase pacing_th is to give more space for the
785 		 * queue to oscillate down without getting empty, but also more
786 		 * room for the queue to increase without causing another alarm.
787 		 */
788 		pacing_save = pacing_save << 3;
789 		pacing_data->pacing_th = rdev->pacing.pacing_algo_th * 4;
790 	}
791 
792 	if (pacing_save > BNXT_RE_MAX_DBR_DO_PACING)
793 		pacing_save = BNXT_RE_MAX_DBR_DO_PACING;
794 
795 	pacing_data->do_pacing = pacing_save;
796 	rdev->pacing.do_pacing_save = pacing_data->do_pacing;
797 	pacing_data->alarm_th =
798 		pacing_data->pacing_th * BNXT_RE_PACING_ALARM_TH_MULTIPLE;
799 	schedule_delayed_work(&rdev->dbq_pacing_work,
800 			      msecs_to_jiffies(rdev->pacing.dbq_pacing_time));
801 	rdev->stats.pacing.alerts++;
802 	mutex_unlock(&rdev->pacing.dbq_lock);
803 }
804 
805 static void bnxt_re_pacing_timer_exp(struct work_struct *work)
806 {
807 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
808 			dbq_pacing_work.work);
809 	struct bnxt_qplib_db_pacing_data *pacing_data;
810 	u32 fifo_occup;
811 
812 	if (!mutex_trylock(&rdev->pacing.dbq_lock))
813 		return;
814 
815 	pacing_data = rdev->qplib_res.pacing_data;
816 	fifo_occup = __get_fifo_occupancy(rdev);
817 
818 	if (fifo_occup > pacing_data->pacing_th)
819 		goto restart_timer;
820 
821 	/*
822 	 * Instead of immediately going back to the default do_pacing
823 	 * reduce it by 1/8 times and restart the timer.
824 	 */
825 	pacing_data->do_pacing = pacing_data->do_pacing - (pacing_data->do_pacing >> 3);
826 	pacing_data->do_pacing = max_t(u32, rdev->pacing.dbr_def_do_pacing, pacing_data->do_pacing);
827 	if (pacing_data->do_pacing <= rdev->pacing.dbr_def_do_pacing) {
828 		bnxt_re_set_default_pacing_data(rdev);
829 		rdev->stats.pacing.complete++;
830 		goto dbq_unlock;
831 	}
832 
833 restart_timer:
834 	schedule_delayed_work(&rdev->dbq_pacing_work,
835 			      msecs_to_jiffies(rdev->pacing.dbq_pacing_time));
836 	rdev->stats.pacing.resched++;
837 dbq_unlock:
838 	rdev->pacing.do_pacing_save = pacing_data->do_pacing;
839 	mutex_unlock(&rdev->pacing.dbq_lock);
840 }
841 
842 void bnxt_re_pacing_alert(struct bnxt_re_dev *rdev)
843 {
844 	struct bnxt_qplib_db_pacing_data *pacing_data;
845 
846 	if (!rdev->pacing.dbr_pacing)
847 		return;
848 	mutex_lock(&rdev->pacing.dbq_lock);
849 	pacing_data = rdev->qplib_res.pacing_data;
850 
851 	/*
852 	 * Increase the alarm_th to max so that other user lib instances do not
853 	 * keep alerting the driver.
854 	 */
855 	pacing_data->alarm_th = pacing_data->fifo_max_depth;
856 	pacing_data->do_pacing = BNXT_RE_MAX_DBR_DO_PACING;
857 	cancel_work_sync(&rdev->dbq_fifo_check_work);
858 	schedule_work(&rdev->dbq_fifo_check_work);
859 	mutex_unlock(&rdev->pacing.dbq_lock);
860 }
861 
862 static int bnxt_re_initialize_dbr_pacing(struct bnxt_re_dev *rdev)
863 {
864 	/* Allocate a page for app use */
865 	rdev->pacing.dbr_page = (void *)__get_free_page(GFP_KERNEL);
866 	if (!rdev->pacing.dbr_page)
867 		return -ENOMEM;
868 
869 	memset((u8 *)rdev->pacing.dbr_page, 0, PAGE_SIZE);
870 	rdev->qplib_res.pacing_data = (struct bnxt_qplib_db_pacing_data *)rdev->pacing.dbr_page;
871 
872 	if (bnxt_re_hwrm_dbr_pacing_qcfg(rdev)) {
873 		free_page((u64)rdev->pacing.dbr_page);
874 		rdev->pacing.dbr_page = NULL;
875 		return -EIO;
876 	}
877 
878 	/* MAP HW window 2 for reading db fifo depth */
879 	writel(rdev->chip_ctx->dbr_stat_db_fifo & BNXT_GRC_BASE_MASK,
880 	       rdev->en_dev->bar0 + BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
881 	rdev->pacing.dbr_db_fifo_reg_off =
882 		(rdev->chip_ctx->dbr_stat_db_fifo & BNXT_GRC_OFFSET_MASK) +
883 		 BNXT_RE_GRC_FIFO_REG_BASE;
884 	rdev->pacing.dbr_bar_addr =
885 		pci_resource_start(rdev->qplib_res.pdev, 0) + rdev->pacing.dbr_db_fifo_reg_off;
886 
887 	if (is_dbr_fifo_full(rdev)) {
888 		free_page((u64)rdev->pacing.dbr_page);
889 		rdev->pacing.dbr_page = NULL;
890 		return -EIO;
891 	}
892 
893 	rdev->pacing.pacing_algo_th = BNXT_RE_PACING_ALGO_THRESHOLD;
894 	rdev->pacing.dbq_pacing_time = BNXT_RE_DBR_PACING_TIME;
895 	rdev->pacing.dbr_def_do_pacing = BNXT_RE_DBR_DO_PACING_NO_CONGESTION;
896 	rdev->pacing.do_pacing_save = rdev->pacing.dbr_def_do_pacing;
897 	rdev->qplib_res.pacing_data->grc_reg_offset = rdev->pacing.dbr_db_fifo_reg_off;
898 	bnxt_re_set_default_pacing_data(rdev);
899 	/* Initialize worker for DBR Pacing */
900 	INIT_WORK(&rdev->dbq_fifo_check_work, bnxt_re_db_fifo_check);
901 	INIT_DELAYED_WORK(&rdev->dbq_pacing_work, bnxt_re_pacing_timer_exp);
902 	return 0;
903 }
904 
905 static void bnxt_re_deinitialize_dbr_pacing(struct bnxt_re_dev *rdev)
906 {
907 	cancel_work_sync(&rdev->dbq_fifo_check_work);
908 	cancel_delayed_work_sync(&rdev->dbq_pacing_work);
909 	if (rdev->pacing.dbr_page)
910 		free_page((u64)rdev->pacing.dbr_page);
911 
912 	rdev->pacing.dbr_page = NULL;
913 	rdev->pacing.dbr_pacing = false;
914 }
915 
916 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev,
917 				 u16 fw_ring_id, int type)
918 {
919 	struct bnxt_en_dev *en_dev = rdev->en_dev;
920 	struct hwrm_ring_free_input req = {};
921 	struct hwrm_ring_free_output resp;
922 	struct bnxt_fw_msg fw_msg = {};
923 	int rc = -EINVAL;
924 
925 	if (test_bit(BNXT_RE_FLAG_ERR_DEVICE_DETACHED, &rdev->flags))
926 		return 0;
927 
928 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_RING_FREE);
929 	req.ring_type = type;
930 	req.ring_id = cpu_to_le16(fw_ring_id);
931 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
932 			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
933 	rc = bnxt_send_msg(en_dev, &fw_msg);
934 	if (rc)
935 		ibdev_err(&rdev->ibdev, "Failed to free HW ring:%d :%#x",
936 			  req.ring_id, rc);
937 	return rc;
938 }
939 
940 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev,
941 				  struct bnxt_re_ring_attr *ring_attr,
942 				  u16 *fw_ring_id)
943 {
944 	struct bnxt_en_dev *en_dev = rdev->en_dev;
945 	struct hwrm_ring_alloc_input req = {};
946 	struct hwrm_ring_alloc_output resp;
947 	struct bnxt_fw_msg fw_msg = {};
948 	int rc = -EINVAL;
949 
950 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_RING_ALLOC);
951 	req.enables = 0;
952 	req.page_tbl_addr =  cpu_to_le64(ring_attr->dma_arr[0]);
953 	if (ring_attr->pages > 1) {
954 		/* Page size is in log2 units */
955 		req.page_size = BNXT_PAGE_SHIFT;
956 		req.page_tbl_depth = 1;
957 	}
958 	req.fbo = 0;
959 	/* Association of ring index with doorbell index and MSIX number */
960 	req.logical_id = cpu_to_le16(ring_attr->lrid);
961 	req.length = cpu_to_le32(ring_attr->depth + 1);
962 	req.ring_type = ring_attr->type;
963 	req.int_mode = ring_attr->mode;
964 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
965 			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
966 	rc = bnxt_send_msg(en_dev, &fw_msg);
967 	if (!rc)
968 		*fw_ring_id = le16_to_cpu(resp.ring_id);
969 
970 	return rc;
971 }
972 
973 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
974 				      u32 fw_stats_ctx_id)
975 {
976 	struct bnxt_en_dev *en_dev = rdev->en_dev;
977 	struct hwrm_stat_ctx_free_input req = {};
978 	struct hwrm_stat_ctx_free_output resp = {};
979 	struct bnxt_fw_msg fw_msg = {};
980 	int rc = -EINVAL;
981 
982 	if (test_bit(BNXT_RE_FLAG_ERR_DEVICE_DETACHED, &rdev->flags))
983 		return 0;
984 
985 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_STAT_CTX_FREE);
986 	req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
987 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
988 			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
989 	rc = bnxt_send_msg(en_dev, &fw_msg);
990 	if (rc)
991 		ibdev_err(&rdev->ibdev, "Failed to free HW stats context %#x",
992 			  rc);
993 
994 	return rc;
995 }
996 
997 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
998 				       struct bnxt_qplib_stats *stats)
999 {
1000 	struct bnxt_qplib_chip_ctx *chip_ctx = rdev->chip_ctx;
1001 	struct hwrm_stat_ctx_alloc_output resp = {};
1002 	struct hwrm_stat_ctx_alloc_input req = {};
1003 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1004 	struct bnxt_fw_msg fw_msg = {};
1005 	int rc = -EINVAL;
1006 
1007 	stats->fw_id = INVALID_STATS_CTX_ID;
1008 
1009 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_STAT_CTX_ALLOC);
1010 	req.update_period_ms = cpu_to_le32(1000);
1011 	req.stats_dma_addr = cpu_to_le64(stats->dma_map);
1012 	req.stats_dma_length = cpu_to_le16(chip_ctx->hw_stats_size);
1013 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
1014 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1015 			    sizeof(resp), BNXT_RE_HWRM_CMD_TIMEOUT(rdev));
1016 	rc = bnxt_send_msg(en_dev, &fw_msg);
1017 	if (!rc)
1018 		stats->fw_id = le32_to_cpu(resp.stat_ctx_id);
1019 
1020 	return rc;
1021 }
1022 
1023 static void bnxt_re_disassociate_ucontext(struct ib_ucontext *ibcontext)
1024 {
1025 }
1026 
1027 /* Device */
1028 static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr,
1029 			   char *buf)
1030 {
1031 	struct bnxt_re_dev *rdev =
1032 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
1033 
1034 	return sysfs_emit(buf, "0x%x\n", rdev->en_dev->pdev->revision);
1035 }
1036 static DEVICE_ATTR_RO(hw_rev);
1037 
1038 static ssize_t hca_type_show(struct device *device,
1039 			     struct device_attribute *attr, char *buf)
1040 {
1041 	struct bnxt_re_dev *rdev =
1042 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
1043 
1044 	return sysfs_emit(buf, "0x%x\n", rdev->en_dev->pdev->device);
1045 }
1046 static DEVICE_ATTR_RO(hca_type);
1047 
1048 static ssize_t board_id_show(struct device *device, struct device_attribute *attr,
1049 			     char *buf)
1050 {
1051 	struct bnxt_re_dev *rdev = rdma_device_to_drv_device(device,
1052 							     struct bnxt_re_dev, ibdev);
1053 	char buffer[BNXT_VPD_FLD_LEN] = {};
1054 
1055 	if (!rdev->is_virtfn)
1056 		memcpy(buffer, rdev->board_partno, BNXT_VPD_FLD_LEN - 1);
1057 	else
1058 		scnprintf(buffer, BNXT_VPD_FLD_LEN, "0x%x-VF",
1059 			  rdev->en_dev->pdev->device);
1060 
1061 	return sysfs_emit(buf, "%s\n", buffer);
1062 }
1063 static DEVICE_ATTR_RO(board_id);
1064 
1065 static struct attribute *bnxt_re_attributes[] = {
1066 	&dev_attr_hw_rev.attr,
1067 	&dev_attr_hca_type.attr,
1068 	&dev_attr_board_id.attr,
1069 	NULL
1070 };
1071 
1072 static const struct attribute_group bnxt_re_dev_attr_group = {
1073 	.attrs = bnxt_re_attributes,
1074 };
1075 
1076 static int bnxt_re_fill_res_mr_entry(struct sk_buff *msg, struct ib_mr *ib_mr)
1077 {
1078 	struct bnxt_qplib_hwq *mr_hwq;
1079 	struct nlattr *table_attr;
1080 	struct bnxt_re_mr *mr;
1081 
1082 	table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_DRIVER);
1083 	if (!table_attr)
1084 		return -EMSGSIZE;
1085 
1086 	mr = container_of(ib_mr, struct bnxt_re_mr, ib_mr);
1087 	mr_hwq = &mr->qplib_mr.hwq;
1088 
1089 	if (rdma_nl_put_driver_u32(msg, "page_size",
1090 				   mr_hwq->qe_ppg * mr_hwq->element_size))
1091 		goto err;
1092 	if (rdma_nl_put_driver_u32(msg, "max_elements", mr_hwq->max_elements))
1093 		goto err;
1094 	if (rdma_nl_put_driver_u32(msg, "element_size", mr_hwq->element_size))
1095 		goto err;
1096 	if (rdma_nl_put_driver_u64_hex(msg, "hwq", (unsigned long)mr_hwq))
1097 		goto err;
1098 	if (rdma_nl_put_driver_u64_hex(msg, "va", mr->qplib_mr.va))
1099 		goto err;
1100 
1101 	nla_nest_end(msg, table_attr);
1102 	return 0;
1103 
1104 err:
1105 	nla_nest_cancel(msg, table_attr);
1106 	return -EMSGSIZE;
1107 }
1108 
1109 static int bnxt_re_fill_res_mr_entry_raw(struct sk_buff *msg, struct ib_mr *ib_mr)
1110 {
1111 	struct bnxt_re_dev *rdev;
1112 	struct bnxt_re_mr *mr;
1113 	int err, len;
1114 	void *data;
1115 
1116 	mr = container_of(ib_mr, struct bnxt_re_mr, ib_mr);
1117 	rdev = mr->rdev;
1118 
1119 	err = bnxt_re_read_context_allowed(rdev);
1120 	if (err)
1121 		return err;
1122 
1123 	len = bnxt_qplib_is_chip_gen_p7(rdev->chip_ctx) ? BNXT_RE_CONTEXT_TYPE_MRW_SIZE_P7 :
1124 							  BNXT_RE_CONTEXT_TYPE_MRW_SIZE_P5;
1125 	data = kzalloc(len, GFP_KERNEL);
1126 	if (!data)
1127 		return -ENOMEM;
1128 
1129 	err = bnxt_qplib_read_context(&rdev->rcfw, CMDQ_READ_CONTEXT_TYPE_MRW,
1130 				      mr->qplib_mr.lkey, len, data);
1131 	if (!err)
1132 		err = nla_put(msg, RDMA_NLDEV_ATTR_RES_RAW, len, data);
1133 
1134 	kfree(data);
1135 	return err;
1136 }
1137 
1138 static int bnxt_re_fill_res_cq_entry(struct sk_buff *msg, struct ib_cq *ib_cq)
1139 {
1140 	struct bnxt_qplib_hwq *cq_hwq;
1141 	struct nlattr *table_attr;
1142 	struct bnxt_re_cq *cq;
1143 
1144 	cq = container_of(ib_cq, struct bnxt_re_cq, ib_cq);
1145 	cq_hwq = &cq->qplib_cq.hwq;
1146 
1147 	table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_DRIVER);
1148 	if (!table_attr)
1149 		return -EMSGSIZE;
1150 
1151 	if (rdma_nl_put_driver_u32(msg, "cq_depth", cq_hwq->depth))
1152 		goto err;
1153 	if (rdma_nl_put_driver_u32(msg, "max_elements", cq_hwq->max_elements))
1154 		goto err;
1155 	if (rdma_nl_put_driver_u32(msg, "element_size", cq_hwq->element_size))
1156 		goto err;
1157 	if (rdma_nl_put_driver_u32(msg, "max_wqe", cq->qplib_cq.max_wqe))
1158 		goto err;
1159 
1160 	nla_nest_end(msg, table_attr);
1161 	return 0;
1162 
1163 err:
1164 	nla_nest_cancel(msg, table_attr);
1165 	return -EMSGSIZE;
1166 }
1167 
1168 static int bnxt_re_fill_res_cq_entry_raw(struct sk_buff *msg, struct ib_cq *ib_cq)
1169 {
1170 	struct bnxt_re_dev *rdev;
1171 	struct bnxt_re_cq *cq;
1172 	int err, len;
1173 	void *data;
1174 
1175 	cq = container_of(ib_cq, struct bnxt_re_cq, ib_cq);
1176 	rdev = cq->rdev;
1177 
1178 	err = bnxt_re_read_context_allowed(rdev);
1179 	if (err)
1180 		return err;
1181 
1182 	len = bnxt_qplib_is_chip_gen_p7(rdev->chip_ctx) ? BNXT_RE_CONTEXT_TYPE_CQ_SIZE_P7 :
1183 					BNXT_RE_CONTEXT_TYPE_CQ_SIZE_P5;
1184 	data = kzalloc(len, GFP_KERNEL);
1185 	if (!data)
1186 		return -ENOMEM;
1187 
1188 	err = bnxt_qplib_read_context(&rdev->rcfw,
1189 				      CMDQ_READ_CONTEXT_TYPE_CQ,
1190 				      cq->qplib_cq.id, len, data);
1191 	if (!err)
1192 		err = nla_put(msg, RDMA_NLDEV_ATTR_RES_RAW, len, data);
1193 
1194 	kfree(data);
1195 	return err;
1196 }
1197 
1198 static int bnxt_re_fill_res_qp_entry(struct sk_buff *msg, struct ib_qp *ib_qp)
1199 {
1200 	struct bnxt_qplib_qp *qplib_qp;
1201 	struct nlattr *table_attr;
1202 	struct bnxt_re_qp *qp;
1203 
1204 	table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_DRIVER);
1205 	if (!table_attr)
1206 		return -EMSGSIZE;
1207 
1208 	qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
1209 	qplib_qp = &qp->qplib_qp;
1210 
1211 	if (rdma_nl_put_driver_u32(msg, "sq_max_wqe", qplib_qp->sq.max_wqe))
1212 		goto err;
1213 	if (rdma_nl_put_driver_u32(msg, "sq_max_sge", qplib_qp->sq.max_sge))
1214 		goto err;
1215 	if (rdma_nl_put_driver_u32(msg, "sq_wqe_size", qplib_qp->sq.wqe_size))
1216 		goto err;
1217 	if (rdma_nl_put_driver_u32(msg, "sq_swq_start", qplib_qp->sq.swq_start))
1218 		goto err;
1219 	if (rdma_nl_put_driver_u32(msg, "sq_swq_last", qplib_qp->sq.swq_last))
1220 		goto err;
1221 	if (rdma_nl_put_driver_u32(msg, "rq_max_wqe", qplib_qp->rq.max_wqe))
1222 		goto err;
1223 	if (rdma_nl_put_driver_u32(msg, "rq_max_sge", qplib_qp->rq.max_sge))
1224 		goto err;
1225 	if (rdma_nl_put_driver_u32(msg, "rq_wqe_size", qplib_qp->rq.wqe_size))
1226 		goto err;
1227 	if (rdma_nl_put_driver_u32(msg, "rq_swq_start", qplib_qp->rq.swq_start))
1228 		goto err;
1229 	if (rdma_nl_put_driver_u32(msg, "rq_swq_last", qplib_qp->rq.swq_last))
1230 		goto err;
1231 	if (rdma_nl_put_driver_u32(msg, "timeout", qplib_qp->timeout))
1232 		goto err;
1233 
1234 	nla_nest_end(msg, table_attr);
1235 	return 0;
1236 
1237 err:
1238 	nla_nest_cancel(msg, table_attr);
1239 	return -EMSGSIZE;
1240 }
1241 
1242 static int bnxt_re_fill_res_qp_entry_raw(struct sk_buff *msg, struct ib_qp *ibqp)
1243 {
1244 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(ibqp->device, ibdev);
1245 	int err, len;
1246 	void *data;
1247 
1248 	err = bnxt_re_read_context_allowed(rdev);
1249 	if (err)
1250 		return err;
1251 
1252 	len = bnxt_qplib_is_chip_gen_p7(rdev->chip_ctx) ? BNXT_RE_CONTEXT_TYPE_QPC_SIZE_P7 :
1253 							  BNXT_RE_CONTEXT_TYPE_QPC_SIZE_P5;
1254 	data = kzalloc(len, GFP_KERNEL);
1255 	if (!data)
1256 		return -ENOMEM;
1257 
1258 	err = bnxt_qplib_read_context(&rdev->rcfw, CMDQ_READ_CONTEXT_TYPE_QPC,
1259 				      ibqp->qp_num, len, data);
1260 	if (!err)
1261 		err = nla_put(msg, RDMA_NLDEV_ATTR_RES_RAW, len, data);
1262 
1263 	kfree(data);
1264 	return err;
1265 }
1266 
1267 static int bnxt_re_fill_res_srq_entry(struct sk_buff *msg, struct ib_srq *ib_srq)
1268 {
1269 	struct nlattr *table_attr;
1270 	struct bnxt_re_srq *srq;
1271 
1272 	table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_DRIVER);
1273 	if (!table_attr)
1274 		return -EMSGSIZE;
1275 
1276 	srq = container_of(ib_srq, struct bnxt_re_srq, ib_srq);
1277 
1278 	if (rdma_nl_put_driver_u32_hex(msg, "wqe_size", srq->qplib_srq.wqe_size))
1279 		goto err;
1280 	if (rdma_nl_put_driver_u32_hex(msg, "max_wqe", srq->qplib_srq.max_wqe))
1281 		goto err;
1282 	if (rdma_nl_put_driver_u32_hex(msg, "max_sge", srq->qplib_srq.max_sge))
1283 		goto err;
1284 	if (rdma_nl_put_driver_u32_hex(msg, "srq_limit", srq->qplib_srq.threshold))
1285 		goto err;
1286 
1287 	nla_nest_end(msg, table_attr);
1288 	return 0;
1289 
1290 err:
1291 	nla_nest_cancel(msg, table_attr);
1292 	return -EMSGSIZE;
1293 }
1294 
1295 static int bnxt_re_fill_res_srq_entry_raw(struct sk_buff *msg, struct ib_srq *ib_srq)
1296 {
1297 	struct bnxt_re_dev *rdev;
1298 	struct bnxt_re_srq *srq;
1299 	int err, len;
1300 	void *data;
1301 
1302 	srq = container_of(ib_srq, struct bnxt_re_srq, ib_srq);
1303 	rdev = srq->rdev;
1304 
1305 	err = bnxt_re_read_context_allowed(rdev);
1306 	if (err)
1307 		return err;
1308 
1309 	len = bnxt_qplib_is_chip_gen_p7(rdev->chip_ctx) ? BNXT_RE_CONTEXT_TYPE_SRQ_SIZE_P7 :
1310 							  BNXT_RE_CONTEXT_TYPE_SRQ_SIZE_P5;
1311 
1312 	data = kzalloc(len, GFP_KERNEL);
1313 	if (!data)
1314 		return -ENOMEM;
1315 
1316 	err = bnxt_qplib_read_context(&rdev->rcfw, CMDQ_READ_CONTEXT_TYPE_SRQ,
1317 				      srq->qplib_srq.id, len, data);
1318 	if (!err)
1319 		err = nla_put(msg, RDMA_NLDEV_ATTR_RES_RAW, len, data);
1320 
1321 	kfree(data);
1322 	return err;
1323 }
1324 
1325 static const struct ib_device_ops bnxt_re_dev_ops = {
1326 	.owner = THIS_MODULE,
1327 	.driver_id = RDMA_DRIVER_BNXT_RE,
1328 	.uverbs_abi_ver = BNXT_RE_ABI_VERSION,
1329 
1330 	.add_gid = bnxt_re_add_gid,
1331 	.alloc_hw_port_stats = bnxt_re_ib_alloc_hw_port_stats,
1332 	.alloc_mr = bnxt_re_alloc_mr,
1333 	.alloc_pd = bnxt_re_alloc_pd,
1334 	.alloc_ucontext = bnxt_re_alloc_ucontext,
1335 	.create_ah = bnxt_re_create_ah,
1336 	.create_cq = bnxt_re_create_cq,
1337 	.create_qp = bnxt_re_create_qp,
1338 	.create_srq = bnxt_re_create_srq,
1339 	.create_user_ah = bnxt_re_create_ah,
1340 	.dealloc_pd = bnxt_re_dealloc_pd,
1341 	.dealloc_ucontext = bnxt_re_dealloc_ucontext,
1342 	.del_gid = bnxt_re_del_gid,
1343 	.dereg_mr = bnxt_re_dereg_mr,
1344 	.destroy_ah = bnxt_re_destroy_ah,
1345 	.destroy_cq = bnxt_re_destroy_cq,
1346 	.destroy_qp = bnxt_re_destroy_qp,
1347 	.destroy_srq = bnxt_re_destroy_srq,
1348 	.device_group = &bnxt_re_dev_attr_group,
1349 	.disassociate_ucontext = bnxt_re_disassociate_ucontext,
1350 	.get_dev_fw_str = bnxt_re_query_fw_str,
1351 	.get_dma_mr = bnxt_re_get_dma_mr,
1352 	.get_hw_stats = bnxt_re_ib_get_hw_stats,
1353 	.get_link_layer = bnxt_re_get_link_layer,
1354 	.get_port_immutable = bnxt_re_get_port_immutable,
1355 	.map_mr_sg = bnxt_re_map_mr_sg,
1356 	.mmap = bnxt_re_mmap,
1357 	.mmap_free = bnxt_re_mmap_free,
1358 	.modify_qp = bnxt_re_modify_qp,
1359 	.modify_srq = bnxt_re_modify_srq,
1360 	.poll_cq = bnxt_re_poll_cq,
1361 	.post_recv = bnxt_re_post_recv,
1362 	.post_send = bnxt_re_post_send,
1363 	.post_srq_recv = bnxt_re_post_srq_recv,
1364 	.process_mad = bnxt_re_process_mad,
1365 	.query_ah = bnxt_re_query_ah,
1366 	.query_device = bnxt_re_query_device,
1367 	.modify_device = bnxt_re_modify_device,
1368 	.query_pkey = bnxt_re_query_pkey,
1369 	.query_port = bnxt_re_query_port,
1370 	.query_qp = bnxt_re_query_qp,
1371 	.query_srq = bnxt_re_query_srq,
1372 	.reg_user_mr = bnxt_re_reg_user_mr,
1373 	.reg_user_mr_dmabuf = bnxt_re_reg_user_mr_dmabuf,
1374 	.req_notify_cq = bnxt_re_req_notify_cq,
1375 	.resize_cq = bnxt_re_resize_cq,
1376 	.create_flow = bnxt_re_create_flow,
1377 	.destroy_flow = bnxt_re_destroy_flow,
1378 	INIT_RDMA_OBJ_SIZE(ib_ah, bnxt_re_ah, ib_ah),
1379 	INIT_RDMA_OBJ_SIZE(ib_cq, bnxt_re_cq, ib_cq),
1380 	INIT_RDMA_OBJ_SIZE(ib_pd, bnxt_re_pd, ib_pd),
1381 	INIT_RDMA_OBJ_SIZE(ib_qp, bnxt_re_qp, ib_qp),
1382 	INIT_RDMA_OBJ_SIZE(ib_srq, bnxt_re_srq, ib_srq),
1383 	INIT_RDMA_OBJ_SIZE(ib_ucontext, bnxt_re_ucontext, ib_uctx),
1384 };
1385 
1386 static const struct ib_device_ops restrack_ops = {
1387 	.fill_res_cq_entry = bnxt_re_fill_res_cq_entry,
1388 	.fill_res_cq_entry_raw = bnxt_re_fill_res_cq_entry_raw,
1389 	.fill_res_qp_entry = bnxt_re_fill_res_qp_entry,
1390 	.fill_res_qp_entry_raw = bnxt_re_fill_res_qp_entry_raw,
1391 	.fill_res_mr_entry = bnxt_re_fill_res_mr_entry,
1392 	.fill_res_mr_entry_raw = bnxt_re_fill_res_mr_entry_raw,
1393 	.fill_res_srq_entry = bnxt_re_fill_res_srq_entry,
1394 	.fill_res_srq_entry_raw = bnxt_re_fill_res_srq_entry_raw,
1395 };
1396 
1397 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
1398 {
1399 	struct ib_device *ibdev = &rdev->ibdev;
1400 	int ret;
1401 
1402 	/* ib device init */
1403 	ibdev->node_type = RDMA_NODE_IB_CA;
1404 	strscpy(ibdev->node_desc, BNXT_RE_DESC " HCA");
1405 	ibdev->phys_port_cnt = 1;
1406 
1407 	addrconf_addr_eui48((u8 *)&ibdev->node_guid, rdev->netdev->dev_addr);
1408 
1409 	ibdev->num_comp_vectors	= rdev->nqr->num_msix - 1;
1410 	ibdev->dev.parent = &rdev->en_dev->pdev->dev;
1411 	ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
1412 
1413 	if (IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS))
1414 		ibdev->driver_def = bnxt_re_uapi_defs;
1415 
1416 	ib_set_device_ops(ibdev, &bnxt_re_dev_ops);
1417 	ib_set_device_ops(ibdev, &restrack_ops);
1418 	ret = ib_device_set_netdev(&rdev->ibdev, rdev->netdev, 1);
1419 	if (ret)
1420 		return ret;
1421 
1422 	dma_set_max_seg_size(&rdev->en_dev->pdev->dev, UINT_MAX);
1423 	ibdev->uverbs_cmd_mask |= BIT_ULL(IB_USER_VERBS_CMD_POLL_CQ);
1424 	return ib_register_device(ibdev, "bnxt_re%d", &rdev->en_dev->pdev->dev);
1425 }
1426 
1427 static struct bnxt_re_dev *bnxt_re_dev_add(struct auxiliary_device *adev,
1428 					   struct bnxt_en_dev *en_dev)
1429 {
1430 	struct bnxt_re_dev *rdev;
1431 
1432 	/* Allocate bnxt_re_dev instance here */
1433 	rdev = ib_alloc_device(bnxt_re_dev, ibdev);
1434 	if (!rdev) {
1435 		ibdev_err(NULL, "%s: bnxt_re_dev allocation failure!",
1436 			  ROCE_DRV_MODULE_NAME);
1437 		return NULL;
1438 	}
1439 	/* Default values */
1440 	rdev->netdev = en_dev->net;
1441 	rdev->en_dev = en_dev;
1442 	rdev->adev = adev;
1443 	rdev->id = rdev->en_dev->pdev->devfn;
1444 	INIT_LIST_HEAD(&rdev->qp_list);
1445 	mutex_init(&rdev->qp_lock);
1446 	mutex_init(&rdev->pacing.dbq_lock);
1447 	atomic_set(&rdev->stats.res.qp_count, 0);
1448 	atomic_set(&rdev->stats.res.cq_count, 0);
1449 	atomic_set(&rdev->stats.res.srq_count, 0);
1450 	atomic_set(&rdev->stats.res.mr_count, 0);
1451 	atomic_set(&rdev->stats.res.mw_count, 0);
1452 	atomic_set(&rdev->stats.res.ah_count, 0);
1453 	atomic_set(&rdev->stats.res.pd_count, 0);
1454 	rdev->cosq[0] = 0xFFFF;
1455 	rdev->cosq[1] = 0xFFFF;
1456 	rdev->cq_coalescing.buf_maxtime = BNXT_QPLIB_CQ_COAL_DEF_BUF_MAXTIME;
1457 	if (bnxt_re_chip_gen_p7(en_dev->chip_num)) {
1458 		rdev->cq_coalescing.normal_maxbuf = BNXT_QPLIB_CQ_COAL_DEF_NORMAL_MAXBUF_P7;
1459 		rdev->cq_coalescing.during_maxbuf = BNXT_QPLIB_CQ_COAL_DEF_DURING_MAXBUF_P7;
1460 	} else {
1461 		rdev->cq_coalescing.normal_maxbuf = BNXT_QPLIB_CQ_COAL_DEF_NORMAL_MAXBUF_P5;
1462 		rdev->cq_coalescing.during_maxbuf = BNXT_QPLIB_CQ_COAL_DEF_DURING_MAXBUF_P5;
1463 	}
1464 	rdev->cq_coalescing.en_ring_idle_mode = BNXT_QPLIB_CQ_COAL_DEF_EN_RING_IDLE_MODE;
1465 
1466 	return rdev;
1467 }
1468 
1469 static int bnxt_re_handle_unaffi_async_event(struct creq_func_event
1470 					     *unaffi_async)
1471 {
1472 	switch (unaffi_async->event) {
1473 	case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
1474 		break;
1475 	case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
1476 		break;
1477 	case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
1478 		break;
1479 	case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
1480 		break;
1481 	case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
1482 		break;
1483 	case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
1484 		break;
1485 	case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
1486 		break;
1487 	case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
1488 		break;
1489 	case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
1490 		break;
1491 	case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
1492 		break;
1493 	case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
1494 		break;
1495 	default:
1496 		return -EINVAL;
1497 	}
1498 	return 0;
1499 }
1500 
1501 static int bnxt_re_handle_qp_async_event(struct creq_qp_event *qp_event,
1502 					 struct bnxt_re_qp *qp)
1503 {
1504 	struct creq_qp_error_notification *err_event;
1505 	struct bnxt_re_srq *srq = NULL;
1506 	struct ib_event event = {};
1507 	unsigned int flags;
1508 
1509 	if (qp->qplib_qp.srq)
1510 		srq =  container_of(qp->qplib_qp.srq, struct bnxt_re_srq,
1511 				    qplib_srq);
1512 
1513 	if (qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_ERR &&
1514 	    rdma_is_kernel_res(&qp->ib_qp.res)) {
1515 		flags = bnxt_re_lock_cqs(qp);
1516 		bnxt_qplib_add_flush_qp(&qp->qplib_qp);
1517 		bnxt_re_unlock_cqs(qp, flags);
1518 	}
1519 
1520 	event.device = &qp->rdev->ibdev;
1521 	event.element.qp = &qp->ib_qp;
1522 	event.event = IB_EVENT_QP_FATAL;
1523 
1524 	err_event = (struct creq_qp_error_notification *)qp_event;
1525 
1526 	switch (err_event->req_err_state_reason) {
1527 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_OPCODE_ERROR:
1528 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_TIMEOUT_RETRY_LIMIT:
1529 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_RNR_TIMEOUT_RETRY_LIMIT:
1530 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_NAK_ARRIVAL_2:
1531 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_NAK_ARRIVAL_3:
1532 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_INVALID_READ_RESP:
1533 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_ILLEGAL_BIND:
1534 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_ILLEGAL_FAST_REG:
1535 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_ILLEGAL_INVALIDATE:
1536 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_RETRAN_LOCAL_ERROR:
1537 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_AV_DOMAIN_ERROR:
1538 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_PROD_WQE_MSMTCH_ERROR:
1539 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_PSN_RANGE_CHECK_ERROR:
1540 		event.event = IB_EVENT_QP_ACCESS_ERR;
1541 		break;
1542 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_NAK_ARRIVAL_1:
1543 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_NAK_ARRIVAL_4:
1544 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_READ_RESP_LENGTH:
1545 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_WQE_FORMAT_ERROR:
1546 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_ORRQ_FORMAT_ERROR:
1547 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_INVALID_AVID_ERROR:
1548 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_SERV_TYPE_ERROR:
1549 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_INVALID_OP_ERROR:
1550 		event.event = IB_EVENT_QP_REQ_ERR;
1551 		break;
1552 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_RX_MEMORY_ERROR:
1553 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_TX_MEMORY_ERROR:
1554 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_CMP_ERROR:
1555 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_CQ_LOAD_ERROR:
1556 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_TX_PCI_ERROR:
1557 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_RX_PCI_ERROR:
1558 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_RETX_SETUP_ERROR:
1559 		event.event = IB_EVENT_QP_FATAL;
1560 		break;
1561 
1562 	default:
1563 		break;
1564 	}
1565 
1566 	switch (err_event->res_err_state_reason) {
1567 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_EXCEED_MAX:
1568 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_PAYLOAD_LENGTH_MISMATCH:
1569 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_PSN_SEQ_ERROR_RETRY_LIMIT:
1570 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_RX_INVALID_R_KEY:
1571 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_RX_DOMAIN_ERROR:
1572 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_RX_NO_PERMISSION:
1573 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_RX_RANGE_ERROR:
1574 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_TX_INVALID_R_KEY:
1575 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_TX_DOMAIN_ERROR:
1576 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_TX_NO_PERMISSION:
1577 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_TX_RANGE_ERROR:
1578 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_UNALIGN_ATOMIC:
1579 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_PSN_NOT_FOUND:
1580 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_INVALID_DUP_RKEY:
1581 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_IRRQ_FORMAT_ERROR:
1582 		event.event = IB_EVENT_QP_ACCESS_ERR;
1583 		break;
1584 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_EXCEEDS_WQE:
1585 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_WQE_FORMAT_ERROR:
1586 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_UNSUPPORTED_OPCODE:
1587 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_REM_INVALIDATE:
1588 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_OPCODE_ERROR:
1589 		event.event = IB_EVENT_QP_REQ_ERR;
1590 		break;
1591 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_IRRQ_OFLOW:
1592 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_CMP_ERROR:
1593 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_CQ_LOAD_ERROR:
1594 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_TX_PCI_ERROR:
1595 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_RX_PCI_ERROR:
1596 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_MEMORY_ERROR:
1597 		event.event = IB_EVENT_QP_FATAL;
1598 		break;
1599 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_SRQ_LOAD_ERROR:
1600 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_SRQ_ERROR:
1601 		if (srq)
1602 			event.event = IB_EVENT_SRQ_ERR;
1603 		break;
1604 	default:
1605 		break;
1606 	}
1607 
1608 	if (err_event->res_err_state_reason || err_event->req_err_state_reason) {
1609 		ibdev_dbg(&qp->rdev->ibdev,
1610 			  "%s %s qp_id: %d cons (%d %d) req (%d %d) res (%d %d)\n",
1611 			   __func__, rdma_is_kernel_res(&qp->ib_qp.res) ? "kernel" : "user",
1612 			   qp->qplib_qp.id,
1613 			   err_event->sq_cons_idx,
1614 			   err_event->rq_cons_idx,
1615 			   err_event->req_slow_path_state,
1616 			   err_event->req_err_state_reason,
1617 			   err_event->res_slow_path_state,
1618 			   err_event->res_err_state_reason);
1619 	} else {
1620 		if (srq)
1621 			event.event = IB_EVENT_QP_LAST_WQE_REACHED;
1622 	}
1623 
1624 	if (event.event == IB_EVENT_SRQ_ERR && srq->ib_srq.event_handler)  {
1625 		(*srq->ib_srq.event_handler)(&event,
1626 				srq->ib_srq.srq_context);
1627 	} else if (event.device && qp->ib_qp.event_handler) {
1628 		qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
1629 	}
1630 
1631 	return 0;
1632 }
1633 
1634 static int bnxt_re_handle_cq_async_error(void *event, struct bnxt_re_cq *cq)
1635 {
1636 	struct creq_cq_error_notification *cqerr;
1637 	struct ib_event ibevent = {};
1638 
1639 	cqerr = event;
1640 	switch (cqerr->cq_err_reason) {
1641 	case CREQ_CQ_ERROR_NOTIFICATION_CQ_ERR_REASON_REQ_CQ_INVALID_ERROR:
1642 	case CREQ_CQ_ERROR_NOTIFICATION_CQ_ERR_REASON_REQ_CQ_OVERFLOW_ERROR:
1643 	case CREQ_CQ_ERROR_NOTIFICATION_CQ_ERR_REASON_REQ_CQ_LOAD_ERROR:
1644 	case CREQ_CQ_ERROR_NOTIFICATION_CQ_ERR_REASON_RES_CQ_INVALID_ERROR:
1645 	case CREQ_CQ_ERROR_NOTIFICATION_CQ_ERR_REASON_RES_CQ_OVERFLOW_ERROR:
1646 	case CREQ_CQ_ERROR_NOTIFICATION_CQ_ERR_REASON_RES_CQ_LOAD_ERROR:
1647 		ibevent.event = IB_EVENT_CQ_ERR;
1648 		break;
1649 	default:
1650 		break;
1651 	}
1652 
1653 	if (ibevent.event == IB_EVENT_CQ_ERR && cq->ib_cq.event_handler) {
1654 		ibevent.element.cq = &cq->ib_cq;
1655 		ibevent.device = &cq->rdev->ibdev;
1656 
1657 		ibdev_dbg(&cq->rdev->ibdev,
1658 			  "%s err reason %d\n", __func__, cqerr->cq_err_reason);
1659 		cq->ib_cq.event_handler(&ibevent, cq->ib_cq.cq_context);
1660 	}
1661 
1662 	return 0;
1663 }
1664 
1665 static int bnxt_re_handle_affi_async_event(struct creq_qp_event *affi_async,
1666 					   void *obj)
1667 {
1668 	struct bnxt_qplib_qp *lib_qp;
1669 	struct bnxt_qplib_cq *lib_cq;
1670 	struct bnxt_re_qp *qp;
1671 	struct bnxt_re_cq *cq;
1672 	int rc = 0;
1673 	u8 event;
1674 
1675 	if (!obj)
1676 		return rc; /* QP was already dead, still return success */
1677 
1678 	event = affi_async->event;
1679 	switch (event) {
1680 	case CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION:
1681 		lib_qp = obj;
1682 		qp = container_of(lib_qp, struct bnxt_re_qp, qplib_qp);
1683 		rc = bnxt_re_handle_qp_async_event(affi_async, qp);
1684 		break;
1685 	case CREQ_QP_EVENT_EVENT_CQ_ERROR_NOTIFICATION:
1686 		lib_cq = obj;
1687 		cq = container_of(lib_cq, struct bnxt_re_cq, qplib_cq);
1688 		rc = bnxt_re_handle_cq_async_error(affi_async, cq);
1689 		break;
1690 	default:
1691 		rc = -EINVAL;
1692 	}
1693 	return rc;
1694 }
1695 
1696 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
1697 			       void *aeqe, void *obj)
1698 {
1699 	struct creq_qp_event *affi_async;
1700 	struct creq_func_event *unaffi_async;
1701 	u8 type;
1702 	int rc;
1703 
1704 	type = ((struct creq_base *)aeqe)->type;
1705 	if (type == CREQ_BASE_TYPE_FUNC_EVENT) {
1706 		unaffi_async = aeqe;
1707 		rc = bnxt_re_handle_unaffi_async_event(unaffi_async);
1708 	} else {
1709 		affi_async = aeqe;
1710 		rc = bnxt_re_handle_affi_async_event(affi_async, obj);
1711 	}
1712 
1713 	return rc;
1714 }
1715 
1716 static int bnxt_re_srqn_handler(struct bnxt_qplib_nq *nq,
1717 				struct bnxt_qplib_srq *handle, u8 event)
1718 {
1719 	struct bnxt_re_srq *srq = container_of(handle, struct bnxt_re_srq,
1720 					       qplib_srq);
1721 	struct ib_event ib_event;
1722 
1723 	ib_event.device = &srq->rdev->ibdev;
1724 	ib_event.element.srq = &srq->ib_srq;
1725 
1726 	if (srq->ib_srq.event_handler) {
1727 		if (event == NQ_SRQ_EVENT_EVENT_SRQ_THRESHOLD_EVENT)
1728 			ib_event.event = IB_EVENT_SRQ_LIMIT_REACHED;
1729 		(*srq->ib_srq.event_handler)(&ib_event,
1730 					     srq->ib_srq.srq_context);
1731 	}
1732 	return 0;
1733 }
1734 
1735 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
1736 			       struct bnxt_qplib_cq *handle)
1737 {
1738 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
1739 					     qplib_cq);
1740 
1741 	if (cq->ib_cq.comp_handler)
1742 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
1743 
1744 	return 0;
1745 }
1746 
1747 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
1748 {
1749 	int i;
1750 
1751 	for (i = 1; i < rdev->nqr->num_msix; i++)
1752 		bnxt_qplib_disable_nq(&rdev->nqr->nq[i - 1]);
1753 
1754 	if (rdev->qplib_res.rcfw)
1755 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
1756 }
1757 
1758 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
1759 {
1760 	int num_vec_enabled = 0;
1761 	int rc = 0, i;
1762 	u32 db_offt;
1763 
1764 	bnxt_qplib_init_res(&rdev->qplib_res);
1765 
1766 	mutex_init(&rdev->nqr->load_lock);
1767 
1768 	for (i = 1; i < rdev->nqr->num_msix ; i++) {
1769 		db_offt = rdev->nqr->msix_entries[i].db_offset;
1770 		rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nqr->nq[i - 1],
1771 					  i - 1, rdev->nqr->msix_entries[i].vector,
1772 					  db_offt, &bnxt_re_cqn_handler,
1773 					  &bnxt_re_srqn_handler);
1774 		if (rc) {
1775 			ibdev_err(&rdev->ibdev,
1776 				  "Failed to enable NQ with rc = 0x%x", rc);
1777 			goto fail;
1778 		}
1779 		num_vec_enabled++;
1780 	}
1781 	return 0;
1782 fail:
1783 	for (i = num_vec_enabled; i >= 0; i--)
1784 		bnxt_qplib_disable_nq(&rdev->nqr->nq[i]);
1785 	return rc;
1786 }
1787 
1788 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev)
1789 {
1790 	struct bnxt_qplib_nq *nq;
1791 	u8 type;
1792 	int i;
1793 
1794 	for (i = 0; i < rdev->nqr->num_msix - 1; i++) {
1795 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1796 		nq = &rdev->nqr->nq[i];
1797 		bnxt_re_net_ring_free(rdev, nq->ring_id, type);
1798 		bnxt_qplib_free_nq(nq);
1799 		nq->res = NULL;
1800 	}
1801 }
1802 
1803 static void bnxt_re_free_res(struct bnxt_re_dev *rdev)
1804 {
1805 	bnxt_re_free_nq_res(rdev);
1806 
1807 	if (rdev->qplib_res.dpi_tbl.max) {
1808 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
1809 				       &rdev->dpi_privileged);
1810 	}
1811 	if (rdev->qplib_res.rcfw) {
1812 		bnxt_qplib_free_res(&rdev->qplib_res);
1813 		rdev->qplib_res.rcfw = NULL;
1814 	}
1815 }
1816 
1817 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
1818 {
1819 	struct bnxt_re_ring_attr rattr = {};
1820 	int num_vec_created = 0;
1821 	int rc, i;
1822 	u8 type;
1823 
1824 	/* Configure and allocate resources for qplib */
1825 	rdev->qplib_res.rcfw = &rdev->rcfw;
1826 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw);
1827 	if (rc)
1828 		goto fail;
1829 
1830 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->netdev);
1831 	if (rc)
1832 		goto fail;
1833 
1834 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res,
1835 				  &rdev->dpi_privileged,
1836 				  rdev, BNXT_QPLIB_DPI_TYPE_KERNEL);
1837 	if (rc)
1838 		goto dealloc_res;
1839 
1840 	for (i = 0; i < rdev->nqr->num_msix - 1; i++) {
1841 		struct bnxt_qplib_nq *nq;
1842 
1843 		nq = &rdev->nqr->nq[i];
1844 		nq->hwq.max_elements = BNXT_QPLIB_NQE_MAX_CNT;
1845 		rc = bnxt_qplib_alloc_nq(&rdev->qplib_res, nq);
1846 		if (rc) {
1847 			ibdev_err(&rdev->ibdev, "Alloc Failed NQ%d rc:%#x",
1848 				  i, rc);
1849 			goto free_nq;
1850 		}
1851 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1852 		rattr.dma_arr = nq->hwq.pbl[PBL_LVL_0].pg_map_arr;
1853 		rattr.pages = nq->hwq.pbl[rdev->nqr->nq[i].hwq.level].pg_count;
1854 		rattr.type = type;
1855 		rattr.mode = RING_ALLOC_REQ_INT_MODE_MSIX;
1856 		rattr.depth = BNXT_QPLIB_NQE_MAX_CNT - 1;
1857 		rattr.lrid = rdev->nqr->msix_entries[i + 1].ring_idx;
1858 		rc = bnxt_re_net_ring_alloc(rdev, &rattr, &nq->ring_id);
1859 		if (rc) {
1860 			ibdev_err(&rdev->ibdev,
1861 				  "Failed to allocate NQ fw id with rc = 0x%x",
1862 				  rc);
1863 			bnxt_qplib_free_nq(nq);
1864 			goto free_nq;
1865 		}
1866 		num_vec_created++;
1867 	}
1868 	return 0;
1869 free_nq:
1870 	for (i = num_vec_created - 1; i >= 0; i--) {
1871 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1872 		bnxt_re_net_ring_free(rdev, rdev->nqr->nq[i].ring_id, type);
1873 		bnxt_qplib_free_nq(&rdev->nqr->nq[i]);
1874 	}
1875 	bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
1876 			       &rdev->dpi_privileged);
1877 dealloc_res:
1878 	bnxt_qplib_free_res(&rdev->qplib_res);
1879 
1880 fail:
1881 	rdev->qplib_res.rcfw = NULL;
1882 	return rc;
1883 }
1884 
1885 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
1886 				   u8 port_num, enum ib_event_type event)
1887 {
1888 	struct ib_event ib_event;
1889 
1890 	ib_event.device = ibdev;
1891 	if (qp) {
1892 		ib_event.element.qp = qp;
1893 		ib_event.event = event;
1894 		if (qp->event_handler)
1895 			qp->event_handler(&ib_event, qp->qp_context);
1896 
1897 	} else {
1898 		ib_event.element.port_num = port_num;
1899 		ib_event.event = event;
1900 		ib_dispatch_event(&ib_event);
1901 	}
1902 }
1903 
1904 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
1905 					struct bnxt_re_qp *qp)
1906 {
1907 	return (qp->ib_qp.qp_type == IB_QPT_GSI) ||
1908 	       (qp == rdev->gsi_ctx.gsi_sqp);
1909 }
1910 
1911 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
1912 {
1913 	struct bnxt_re_qp *qp;
1914 
1915 	mutex_lock(&rdev->qp_lock);
1916 	list_for_each_entry(qp, &rdev->qp_list, list) {
1917 		/* Modify the state of all QPs except QP1/Shadow QP */
1918 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
1919 			if (qp->qplib_qp.state !=
1920 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
1921 			    qp->qplib_qp.state !=
1922 			    CMDQ_MODIFY_QP_NEW_STATE_ERR)
1923 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
1924 						       1, IB_EVENT_QP_FATAL);
1925 		}
1926 	}
1927 	mutex_unlock(&rdev->qp_lock);
1928 }
1929 
1930 static void bnxt_re_net_unregister_async_event(struct bnxt_re_dev *rdev)
1931 {
1932 	if (rdev->is_virtfn)
1933 		return;
1934 
1935 	memset(&rdev->event_bitmap, 0, sizeof(rdev->event_bitmap));
1936 	bnxt_register_async_events(rdev->en_dev, &rdev->event_bitmap,
1937 				   ASYNC_EVENT_CMPL_EVENT_ID_DCB_CONFIG_CHANGE);
1938 }
1939 
1940 static void bnxt_re_net_register_async_event(struct bnxt_re_dev *rdev)
1941 {
1942 	if (rdev->is_virtfn)
1943 		return;
1944 
1945 	rdev->event_bitmap |= (1 << ASYNC_EVENT_CMPL_EVENT_ID_DCB_CONFIG_CHANGE);
1946 	bnxt_register_async_events(rdev->en_dev, &rdev->event_bitmap,
1947 				   ASYNC_EVENT_CMPL_EVENT_ID_DCB_CONFIG_CHANGE);
1948 }
1949 
1950 static void bnxt_re_read_vpd_info(struct bnxt_re_dev *rdev)
1951 {
1952 	struct pci_dev *pdev = rdev->en_dev->pdev;
1953 	unsigned int vpd_size, kw_len;
1954 	int pos, size;
1955 	u8 *vpd_data;
1956 
1957 	vpd_data = pci_vpd_alloc(pdev, &vpd_size);
1958 	if (IS_ERR(vpd_data)) {
1959 		pci_warn(pdev, "Unable to read VPD, err=%pe\n", vpd_data);
1960 		return;
1961 	}
1962 
1963 	pos = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size,
1964 					   PCI_VPD_RO_KEYWORD_PARTNO, &kw_len);
1965 	if (pos < 0)
1966 		goto free;
1967 
1968 	size = min_t(int, kw_len, BNXT_VPD_FLD_LEN - 1);
1969 	memcpy(rdev->board_partno, &vpd_data[pos], size);
1970 free:
1971 	kfree(vpd_data);
1972 }
1973 
1974 static int bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev)
1975 {
1976 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1977 	struct hwrm_ver_get_output resp = {};
1978 	struct hwrm_ver_get_input req = {};
1979 	struct bnxt_qplib_chip_ctx *cctx;
1980 	struct bnxt_fw_msg fw_msg = {};
1981 	int rc;
1982 
1983 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_VER_GET);
1984 	req.hwrm_intf_maj = HWRM_VERSION_MAJOR;
1985 	req.hwrm_intf_min = HWRM_VERSION_MINOR;
1986 	req.hwrm_intf_upd = HWRM_VERSION_UPDATE;
1987 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1988 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1989 	rc = bnxt_send_msg(en_dev, &fw_msg);
1990 	if (rc) {
1991 		ibdev_err(&rdev->ibdev, "Failed to query HW version, rc = 0x%x",
1992 			  rc);
1993 		return rc;
1994 	}
1995 
1996 	cctx = rdev->chip_ctx;
1997 	cctx->hwrm_intf_ver =
1998 		(u64)le16_to_cpu(resp.hwrm_intf_major) << 48 |
1999 		(u64)le16_to_cpu(resp.hwrm_intf_minor) << 32 |
2000 		(u64)le16_to_cpu(resp.hwrm_intf_build) << 16 |
2001 		le16_to_cpu(resp.hwrm_intf_patch);
2002 
2003 	cctx->hwrm_cmd_max_timeout = le16_to_cpu(resp.max_req_timeout);
2004 
2005 	if (!cctx->hwrm_cmd_max_timeout)
2006 		cctx->hwrm_cmd_max_timeout = RCFW_FW_STALL_MAX_TIMEOUT;
2007 
2008 	return 0;
2009 }
2010 
2011 static int bnxt_re_ib_init(struct bnxt_re_dev *rdev)
2012 {
2013 	int rc;
2014 	u32 event;
2015 
2016 	/* Register ib dev */
2017 	rc = bnxt_re_register_ib(rdev);
2018 	if (rc) {
2019 		pr_err("Failed to register with IB: %#x\n", rc);
2020 		return rc;
2021 	}
2022 	dev_info(rdev_to_dev(rdev), "Device registered with IB successfully");
2023 	set_bit(BNXT_RE_FLAG_ISSUE_ROCE_STATS, &rdev->flags);
2024 
2025 	event = netif_running(rdev->netdev) && netif_carrier_ok(rdev->netdev) ?
2026 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2027 
2028 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, event);
2029 
2030 	return rc;
2031 }
2032 
2033 static int bnxt_re_alloc_nqr_mem(struct bnxt_re_dev *rdev)
2034 {
2035 	rdev->nqr = kzalloc(sizeof(*rdev->nqr), GFP_KERNEL);
2036 	if (!rdev->nqr)
2037 		return -ENOMEM;
2038 
2039 	return 0;
2040 }
2041 
2042 static void bnxt_re_free_nqr_mem(struct bnxt_re_dev *rdev)
2043 {
2044 	kfree(rdev->nqr);
2045 	rdev->nqr = NULL;
2046 }
2047 
2048 /* When DEL_GID fails, driver is not freeing GID ctx memory.
2049  * To avoid the memory leak, free the memory during unload
2050  */
2051 static void bnxt_re_free_gid_ctx(struct bnxt_re_dev *rdev)
2052 {
2053 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
2054 	struct bnxt_re_gid_ctx *ctx, **ctx_tbl;
2055 	int i;
2056 
2057 	if (!sgid_tbl->active)
2058 		return;
2059 
2060 	ctx_tbl = sgid_tbl->ctx;
2061 	for (i = 0; i < sgid_tbl->max; i++) {
2062 		if (sgid_tbl->hw_id[i] == 0xFFFF)
2063 			continue;
2064 
2065 		ctx = ctx_tbl[i];
2066 		kfree(ctx);
2067 	}
2068 }
2069 
2070 static int bnxt_re_get_stats_ctx(struct bnxt_re_dev *rdev)
2071 {
2072 	struct bnxt_qplib_ctx *hctx = &rdev->qplib_ctx;
2073 	struct bnxt_qplib_res *res = &rdev->qplib_res;
2074 	int rc;
2075 
2076 	rc = bnxt_qplib_alloc_stats_ctx(res->pdev, res->cctx, &hctx->stats);
2077 	if (rc)
2078 		return rc;
2079 
2080 	rc = bnxt_re_net_stats_ctx_alloc(rdev, &hctx->stats);
2081 	if (rc)
2082 		goto free_stat_mem;
2083 
2084 	return 0;
2085 free_stat_mem:
2086 	bnxt_qplib_free_stats_ctx(res->pdev, &hctx->stats);
2087 
2088 	return rc;
2089 }
2090 
2091 static int bnxt_re_get_stats3_ctx(struct bnxt_re_dev *rdev)
2092 {
2093 	struct bnxt_qplib_ctx *hctx = &rdev->qplib_ctx;
2094 	struct bnxt_qplib_res *res = &rdev->qplib_res;
2095 	int rc;
2096 
2097 	if (!rdev->rcfw.roce_mirror)
2098 		return 0;
2099 
2100 	rc = bnxt_qplib_alloc_stats_ctx(res->pdev, res->cctx, &hctx->stats3);
2101 	if (rc)
2102 		return rc;
2103 
2104 	rc = bnxt_re_net_stats_ctx_alloc(rdev, &hctx->stats3);
2105 	if (rc)
2106 		goto free_stat_mem;
2107 
2108 	return 0;
2109 free_stat_mem:
2110 	bnxt_qplib_free_stats_ctx(res->pdev, &hctx->stats3);
2111 
2112 	return rc;
2113 }
2114 
2115 static void bnxt_re_put_stats3_ctx(struct bnxt_re_dev *rdev)
2116 {
2117 	struct bnxt_qplib_ctx *hctx = &rdev->qplib_ctx;
2118 	struct bnxt_qplib_res *res = &rdev->qplib_res;
2119 
2120 	if (!rdev->rcfw.roce_mirror)
2121 		return;
2122 
2123 	bnxt_re_net_stats_ctx_free(rdev, hctx->stats3.fw_id);
2124 	bnxt_qplib_free_stats_ctx(res->pdev, &hctx->stats3);
2125 }
2126 
2127 static void bnxt_re_put_stats_ctx(struct bnxt_re_dev *rdev)
2128 {
2129 	struct bnxt_qplib_ctx *hctx = &rdev->qplib_ctx;
2130 	struct bnxt_qplib_res *res = &rdev->qplib_res;
2131 
2132 	bnxt_re_net_stats_ctx_free(rdev, hctx->stats.fw_id);
2133 	bnxt_qplib_free_stats_ctx(res->pdev, &hctx->stats);
2134 }
2135 
2136 static void bnxt_re_dev_uninit(struct bnxt_re_dev *rdev, u8 op_type)
2137 {
2138 	u8 type;
2139 	int rc;
2140 
2141 	bnxt_re_debugfs_rem_pdev(rdev);
2142 
2143 	bnxt_re_net_unregister_async_event(rdev);
2144 	bnxt_re_uninit_dcb_wq(rdev);
2145 
2146 	bnxt_re_put_stats3_ctx(rdev);
2147 
2148 	bnxt_re_free_gid_ctx(rdev);
2149 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED,
2150 			       &rdev->flags))
2151 		bnxt_re_cleanup_res(rdev);
2152 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags))
2153 		bnxt_re_free_res(rdev);
2154 
2155 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
2156 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
2157 		if (rc)
2158 			ibdev_warn(&rdev->ibdev,
2159 				   "Failed to deinitialize RCFW: %#x", rc);
2160 		bnxt_re_put_stats_ctx(rdev);
2161 		bnxt_qplib_free_hwctx(&rdev->qplib_res, &rdev->qplib_ctx);
2162 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
2163 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
2164 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id, type);
2165 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
2166 	}
2167 
2168 	rdev->nqr->num_msix = 0;
2169 
2170 	if (rdev->pacing.dbr_pacing)
2171 		bnxt_re_deinitialize_dbr_pacing(rdev);
2172 
2173 	bnxt_re_free_nqr_mem(rdev);
2174 	bnxt_re_destroy_chip_ctx(rdev);
2175 	if (op_type == BNXT_RE_COMPLETE_REMOVE) {
2176 		if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags))
2177 			bnxt_unregister_dev(rdev->en_dev);
2178 	}
2179 }
2180 
2181 static int bnxt_re_dev_init(struct bnxt_re_dev *rdev, u8 op_type)
2182 {
2183 	struct bnxt_re_ring_attr rattr = {};
2184 	struct bnxt_qplib_creq_ctx *creq;
2185 	u32 db_offt;
2186 	int vid;
2187 	u8 type;
2188 	int rc;
2189 
2190 	if (op_type == BNXT_RE_COMPLETE_INIT) {
2191 		/* Registered a new RoCE device instance to netdev */
2192 		rc = bnxt_re_register_netdev(rdev);
2193 		if (rc) {
2194 			ibdev_err(&rdev->ibdev,
2195 				  "Failed to register with Ethernet driver, rc %d\n",
2196 				  rc);
2197 			return rc;
2198 		}
2199 	}
2200 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
2201 
2202 	if (rdev->en_dev->ulp_tbl->msix_requested < BNXT_RE_MIN_MSIX) {
2203 		ibdev_err(&rdev->ibdev,
2204 			  "RoCE requires minimum 2 MSI-X vectors, but only %d reserved\n",
2205 			  rdev->en_dev->ulp_tbl->msix_requested);
2206 		bnxt_unregister_dev(rdev->en_dev);
2207 		clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
2208 		return -EINVAL;
2209 	}
2210 	ibdev_dbg(&rdev->ibdev, "Got %d MSI-X vectors\n",
2211 		  rdev->en_dev->ulp_tbl->msix_requested);
2212 
2213 	rc = bnxt_re_setup_chip_ctx(rdev);
2214 	if (rc) {
2215 		bnxt_unregister_dev(rdev->en_dev);
2216 		clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
2217 		ibdev_err(&rdev->ibdev, "Failed to get chip context\n");
2218 		return -EINVAL;
2219 	}
2220 
2221 	rc = bnxt_re_alloc_nqr_mem(rdev);
2222 	if (rc) {
2223 		bnxt_re_destroy_chip_ctx(rdev);
2224 		bnxt_unregister_dev(rdev->en_dev);
2225 		clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
2226 		return rc;
2227 	}
2228 	rdev->nqr->num_msix = rdev->en_dev->ulp_tbl->msix_requested;
2229 	memcpy(rdev->nqr->msix_entries, rdev->en_dev->msix_entries,
2230 	       sizeof(struct bnxt_msix_entry) * rdev->nqr->num_msix);
2231 
2232 	/* Check whether VF or PF */
2233 	bnxt_re_get_sriov_func_type(rdev);
2234 
2235 	/* Establish RCFW Communication Channel to initialize the context
2236 	 * memory for the function and all child VFs
2237 	 */
2238 	rc = bnxt_qplib_alloc_rcfw_channel(&rdev->qplib_res, &rdev->rcfw,
2239 					   &rdev->qplib_ctx);
2240 	if (rc) {
2241 		ibdev_err(&rdev->ibdev,
2242 			  "Failed to allocate RCFW Channel: %#x\n", rc);
2243 		goto fail;
2244 	}
2245 
2246 	type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
2247 	creq = &rdev->rcfw.creq;
2248 	rattr.dma_arr = creq->hwq.pbl[PBL_LVL_0].pg_map_arr;
2249 	rattr.pages = creq->hwq.pbl[creq->hwq.level].pg_count;
2250 	rattr.type = type;
2251 	rattr.mode = RING_ALLOC_REQ_INT_MODE_MSIX;
2252 	rattr.depth = BNXT_QPLIB_CREQE_MAX_CNT - 1;
2253 	rattr.lrid = rdev->nqr->msix_entries[BNXT_RE_AEQ_IDX].ring_idx;
2254 	rc = bnxt_re_net_ring_alloc(rdev, &rattr, &creq->ring_id);
2255 	if (rc) {
2256 		ibdev_err(&rdev->ibdev, "Failed to allocate CREQ: %#x\n", rc);
2257 		goto free_rcfw;
2258 	}
2259 	db_offt = rdev->nqr->msix_entries[BNXT_RE_AEQ_IDX].db_offset;
2260 	vid = rdev->nqr->msix_entries[BNXT_RE_AEQ_IDX].vector;
2261 	rc = bnxt_qplib_enable_rcfw_channel(&rdev->rcfw,
2262 					    vid, db_offt,
2263 					    &bnxt_re_aeq_handler);
2264 	if (rc) {
2265 		ibdev_err(&rdev->ibdev, "Failed to enable RCFW channel: %#x\n",
2266 			  rc);
2267 		goto free_ring;
2268 	}
2269 
2270 	if (bnxt_qplib_dbr_pacing_en(rdev->chip_ctx)) {
2271 		rc = bnxt_re_initialize_dbr_pacing(rdev);
2272 		if (!rc) {
2273 			rdev->pacing.dbr_pacing = true;
2274 		} else {
2275 			ibdev_err(&rdev->ibdev,
2276 				  "DBR pacing disabled with error : %d\n", rc);
2277 			rdev->pacing.dbr_pacing = false;
2278 		}
2279 	}
2280 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw);
2281 	if (rc)
2282 		goto disable_rcfw;
2283 
2284 	bnxt_qplib_query_version(&rdev->rcfw);
2285 	bnxt_re_set_resource_limits(rdev);
2286 
2287 	if (!rdev->is_virtfn &&
2288 	    !bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx)) {
2289 		rc = bnxt_qplib_alloc_hwctx(&rdev->qplib_res, &rdev->qplib_ctx);
2290 		if (rc) {
2291 			ibdev_err(&rdev->ibdev,
2292 				  "Failed to allocate hw context: %#x\n", rc);
2293 			goto disable_rcfw;
2294 		}
2295 	}
2296 
2297 	rc = bnxt_re_get_stats_ctx(rdev);
2298 	if (rc) {
2299 		ibdev_err(&rdev->ibdev,
2300 			  "Failed to allocate stats context: %#x\n", rc);
2301 		goto free_ctx;
2302 	}
2303 
2304 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx,
2305 				  rdev->is_virtfn);
2306 	if (rc) {
2307 		ibdev_err(&rdev->ibdev,
2308 			  "Failed to initialize RCFW: %#x\n", rc);
2309 		goto free_sctx;
2310 	}
2311 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
2312 
2313 	/* Resources based on the 'new' device caps */
2314 	rc = bnxt_re_alloc_res(rdev);
2315 	if (rc) {
2316 		ibdev_err(&rdev->ibdev,
2317 			  "Failed to allocate resources: %#x\n", rc);
2318 		goto fail;
2319 	}
2320 	set_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags);
2321 	rc = bnxt_re_init_res(rdev);
2322 	if (rc) {
2323 		ibdev_err(&rdev->ibdev,
2324 			  "Failed to initialize resources: %#x\n", rc);
2325 		goto fail;
2326 	}
2327 
2328 	set_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED, &rdev->flags);
2329 
2330 	if (!rdev->is_virtfn) {
2331 		/* Query f/w defaults of CC params */
2332 		rc = bnxt_qplib_query_cc_param(&rdev->qplib_res, &rdev->cc_param);
2333 		if (rc)
2334 			ibdev_warn(&rdev->ibdev, "Failed to query CC defaults\n");
2335 
2336 		if (!(rdev->qplib_res.en_dev->flags & BNXT_EN_FLAG_ROCE_VF_RES_MGMT))
2337 			bnxt_re_vf_res_config(rdev);
2338 	}
2339 	hash_init(rdev->cq_hash);
2340 	if (rdev->chip_ctx->modes.toggle_bits & BNXT_QPLIB_SRQ_TOGGLE_BIT)
2341 		hash_init(rdev->srq_hash);
2342 
2343 	bnxt_re_debugfs_add_pdev(rdev);
2344 
2345 	bnxt_re_init_dcb_wq(rdev);
2346 	bnxt_re_net_register_async_event(rdev);
2347 
2348 	if (!rdev->is_virtfn)
2349 		bnxt_re_read_vpd_info(rdev);
2350 
2351 	rc = bnxt_re_get_stats3_ctx(rdev);
2352 	if (rc)
2353 		goto fail;
2354 
2355 	return 0;
2356 free_sctx:
2357 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
2358 free_ctx:
2359 	bnxt_qplib_free_hwctx(&rdev->qplib_res, &rdev->qplib_ctx);
2360 disable_rcfw:
2361 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
2362 free_ring:
2363 	type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
2364 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id, type);
2365 free_rcfw:
2366 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
2367 fail:
2368 	bnxt_re_dev_uninit(rdev, BNXT_RE_COMPLETE_REMOVE);
2369 
2370 	return rc;
2371 }
2372 
2373 static void bnxt_re_setup_cc(struct bnxt_re_dev *rdev, bool enable)
2374 {
2375 	struct bnxt_qplib_cc_param cc_param = {};
2376 
2377 	/* Do not enable congestion control on VFs */
2378 	if (rdev->is_virtfn)
2379 		return;
2380 
2381 	/* Currently enabling only for GenP5 adapters */
2382 	if (!bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx))
2383 		return;
2384 
2385 	if (enable) {
2386 		cc_param.enable  = 1;
2387 		cc_param.tos_ecn = 1;
2388 	}
2389 
2390 	cc_param.mask = (CMDQ_MODIFY_ROCE_CC_MODIFY_MASK_ENABLE_CC |
2391 			 CMDQ_MODIFY_ROCE_CC_MODIFY_MASK_TOS_ECN);
2392 
2393 	if (bnxt_qplib_modify_cc(&rdev->qplib_res, &cc_param))
2394 		ibdev_err(&rdev->ibdev, "Failed to setup CC enable = %d\n", enable);
2395 }
2396 
2397 static void bnxt_re_update_en_info_rdev(struct bnxt_re_dev *rdev,
2398 					struct bnxt_re_en_dev_info *en_info,
2399 					struct auxiliary_device *adev)
2400 {
2401 	/* Before updating the rdev pointer in bnxt_re_en_dev_info structure,
2402 	 * take the rtnl lock to avoid accessing invalid rdev pointer from
2403 	 * L2 ULP callbacks. This is applicable in all the places where rdev
2404 	 * pointer is updated in bnxt_re_en_dev_info.
2405 	 */
2406 	rtnl_lock();
2407 	en_info->rdev = rdev;
2408 	rtnl_unlock();
2409 }
2410 
2411 static int bnxt_re_add_device(struct auxiliary_device *adev, u8 op_type)
2412 {
2413 	struct bnxt_aux_priv *aux_priv =
2414 		container_of(adev, struct bnxt_aux_priv, aux_dev);
2415 	struct bnxt_re_en_dev_info *en_info;
2416 	struct bnxt_en_dev *en_dev;
2417 	struct bnxt_re_dev *rdev;
2418 	int rc;
2419 
2420 	en_info = auxiliary_get_drvdata(adev);
2421 	en_dev = en_info->en_dev;
2422 
2423 
2424 	rdev = bnxt_re_dev_add(adev, en_dev);
2425 	if (!rdev || !rdev_to_dev(rdev)) {
2426 		rc = -ENOMEM;
2427 		goto exit;
2428 	}
2429 
2430 	bnxt_re_update_en_info_rdev(rdev, en_info, adev);
2431 
2432 	rc = bnxt_re_dev_init(rdev, op_type);
2433 	if (rc)
2434 		goto re_dev_dealloc;
2435 
2436 	rc = bnxt_re_ib_init(rdev);
2437 	if (rc) {
2438 		pr_err("Failed to register with IB: %s",
2439 			aux_priv->aux_dev.name);
2440 		goto re_dev_uninit;
2441 	}
2442 
2443 	bnxt_re_setup_cc(rdev, true);
2444 
2445 	return 0;
2446 
2447 re_dev_uninit:
2448 	bnxt_re_update_en_info_rdev(NULL, en_info, adev);
2449 	bnxt_re_dev_uninit(rdev, BNXT_RE_COMPLETE_REMOVE);
2450 re_dev_dealloc:
2451 	ib_dealloc_device(&rdev->ibdev);
2452 exit:
2453 	return rc;
2454 }
2455 
2456 #define BNXT_ADEV_NAME "bnxt_en"
2457 
2458 static void bnxt_re_remove_device(struct bnxt_re_dev *rdev, u8 op_type,
2459 				  struct auxiliary_device *aux_dev)
2460 {
2461 	bnxt_re_setup_cc(rdev, false);
2462 	ib_unregister_device(&rdev->ibdev);
2463 	bnxt_re_dev_uninit(rdev, op_type);
2464 	ib_dealloc_device(&rdev->ibdev);
2465 }
2466 
2467 static void bnxt_re_remove(struct auxiliary_device *adev)
2468 {
2469 	struct bnxt_re_en_dev_info *en_info = auxiliary_get_drvdata(adev);
2470 	struct bnxt_re_dev *rdev;
2471 
2472 	mutex_lock(&bnxt_re_mutex);
2473 	rdev = en_info->rdev;
2474 
2475 	if (rdev)
2476 		bnxt_re_remove_device(rdev, BNXT_RE_COMPLETE_REMOVE, adev);
2477 	kfree(en_info);
2478 	mutex_unlock(&bnxt_re_mutex);
2479 }
2480 
2481 static int bnxt_re_probe(struct auxiliary_device *adev,
2482 			 const struct auxiliary_device_id *id)
2483 {
2484 	struct bnxt_aux_priv *aux_priv =
2485 		container_of(adev, struct bnxt_aux_priv, aux_dev);
2486 	struct bnxt_re_en_dev_info *en_info;
2487 	struct bnxt_en_dev *en_dev;
2488 	int rc;
2489 
2490 	en_dev = aux_priv->edev;
2491 
2492 	mutex_lock(&bnxt_re_mutex);
2493 	en_info = kzalloc(sizeof(*en_info), GFP_KERNEL);
2494 	if (!en_info) {
2495 		mutex_unlock(&bnxt_re_mutex);
2496 		return -ENOMEM;
2497 	}
2498 	en_info->en_dev = en_dev;
2499 
2500 	auxiliary_set_drvdata(adev, en_info);
2501 
2502 	rc = bnxt_re_add_device(adev, BNXT_RE_COMPLETE_INIT);
2503 	if (rc)
2504 		kfree(en_info);
2505 
2506 	mutex_unlock(&bnxt_re_mutex);
2507 
2508 	return rc;
2509 }
2510 
2511 static int bnxt_re_suspend(struct auxiliary_device *adev, pm_message_t state)
2512 {
2513 	struct bnxt_re_en_dev_info *en_info = auxiliary_get_drvdata(adev);
2514 	struct bnxt_en_dev *en_dev;
2515 	struct bnxt_re_dev *rdev;
2516 
2517 	rdev = en_info->rdev;
2518 	en_dev = en_info->en_dev;
2519 	mutex_lock(&bnxt_re_mutex);
2520 
2521 	ibdev_info(&rdev->ibdev, "Handle device suspend call");
2522 	/* Check the current device state from bnxt_en_dev and move the
2523 	 * device to detached state if FW_FATAL_COND is set.
2524 	 * This prevents more commands to HW during clean-up,
2525 	 * in case the device is already in error.
2526 	 */
2527 	if (test_bit(BNXT_STATE_FW_FATAL_COND, &rdev->en_dev->en_state)) {
2528 		set_bit(ERR_DEVICE_DETACHED, &rdev->rcfw.cmdq.flags);
2529 		set_bit(BNXT_RE_FLAG_ERR_DEVICE_DETACHED, &rdev->flags);
2530 		wake_up_all(&rdev->rcfw.cmdq.waitq);
2531 		bnxt_re_dev_stop(rdev);
2532 	}
2533 
2534 	if (rdev->pacing.dbr_pacing)
2535 		bnxt_re_set_pacing_dev_state(rdev);
2536 
2537 	ibdev_info(&rdev->ibdev, "%s: L2 driver notified to stop en_state 0x%lx",
2538 		   __func__, en_dev->en_state);
2539 	bnxt_re_remove_device(rdev, BNXT_RE_PRE_RECOVERY_REMOVE, adev);
2540 	bnxt_re_update_en_info_rdev(NULL, en_info, adev);
2541 	mutex_unlock(&bnxt_re_mutex);
2542 
2543 	return 0;
2544 }
2545 
2546 static int bnxt_re_resume(struct auxiliary_device *adev)
2547 {
2548 	struct bnxt_re_en_dev_info *en_info = auxiliary_get_drvdata(adev);
2549 	struct bnxt_re_dev *rdev;
2550 
2551 	mutex_lock(&bnxt_re_mutex);
2552 	bnxt_re_add_device(adev, BNXT_RE_POST_RECOVERY_INIT);
2553 	rdev = en_info->rdev;
2554 	ibdev_info(&rdev->ibdev, "Device resume completed");
2555 	mutex_unlock(&bnxt_re_mutex);
2556 
2557 	return 0;
2558 }
2559 
2560 static void bnxt_re_shutdown(struct auxiliary_device *adev)
2561 {
2562 	struct bnxt_re_en_dev_info *en_info = auxiliary_get_drvdata(adev);
2563 	struct bnxt_re_dev *rdev;
2564 
2565 	rdev = en_info->rdev;
2566 	ib_unregister_device(&rdev->ibdev);
2567 	bnxt_re_dev_uninit(rdev, BNXT_RE_COMPLETE_REMOVE);
2568 }
2569 
2570 static const struct auxiliary_device_id bnxt_re_id_table[] = {
2571 	{ .name = BNXT_ADEV_NAME ".rdma", },
2572 	{},
2573 };
2574 
2575 MODULE_DEVICE_TABLE(auxiliary, bnxt_re_id_table);
2576 
2577 static struct auxiliary_driver bnxt_re_driver = {
2578 	.name = "rdma",
2579 	.probe = bnxt_re_probe,
2580 	.remove = bnxt_re_remove,
2581 	.shutdown = bnxt_re_shutdown,
2582 	.suspend = bnxt_re_suspend,
2583 	.resume = bnxt_re_resume,
2584 	.id_table = bnxt_re_id_table,
2585 };
2586 
2587 static int __init bnxt_re_mod_init(void)
2588 {
2589 	int rc;
2590 
2591 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
2592 	bnxt_re_register_debugfs();
2593 
2594 	rc = auxiliary_driver_register(&bnxt_re_driver);
2595 	if (rc) {
2596 		pr_err("%s: Failed to register auxiliary driver\n",
2597 			ROCE_DRV_MODULE_NAME);
2598 		goto err_debug;
2599 	}
2600 	return 0;
2601 err_debug:
2602 	bnxt_re_unregister_debugfs();
2603 	return rc;
2604 }
2605 
2606 static void __exit bnxt_re_mod_exit(void)
2607 {
2608 	auxiliary_driver_unregister(&bnxt_re_driver);
2609 	bnxt_re_unregister_debugfs();
2610 }
2611 
2612 module_init(bnxt_re_mod_init);
2613 module_exit(bnxt_re_mod_exit);
2614