xref: /linux/drivers/infiniband/hw/bnxt_re/main.c (revision b4db9f840283caca0d904436f187ef56a9126eaa)
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 
71 static char version[] =
72 		BNXT_RE_DESC "\n";
73 
74 MODULE_AUTHOR("Eddie Wai <eddie.wai@broadcom.com>");
75 MODULE_DESCRIPTION(BNXT_RE_DESC);
76 MODULE_LICENSE("Dual BSD/GPL");
77 
78 /* globals */
79 static DEFINE_MUTEX(bnxt_re_mutex);
80 
81 static void bnxt_re_stop_irq(void *handle);
82 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev);
83 static int bnxt_re_netdev_event(struct notifier_block *notifier,
84 				unsigned long event, void *ptr);
85 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev);
86 static void bnxt_re_dev_uninit(struct bnxt_re_dev *rdev);
87 static int bnxt_re_hwrm_qcaps(struct bnxt_re_dev *rdev);
88 
89 static int bnxt_re_hwrm_qcfg(struct bnxt_re_dev *rdev, u32 *db_len,
90 			     u32 *offset);
91 static void bnxt_re_set_db_offset(struct bnxt_re_dev *rdev)
92 {
93 	struct bnxt_qplib_chip_ctx *cctx;
94 	struct bnxt_en_dev *en_dev;
95 	struct bnxt_qplib_res *res;
96 	u32 l2db_len = 0;
97 	u32 offset = 0;
98 	u32 barlen;
99 	int rc;
100 
101 	res = &rdev->qplib_res;
102 	en_dev = rdev->en_dev;
103 	cctx = rdev->chip_ctx;
104 
105 	/* Issue qcfg */
106 	rc = bnxt_re_hwrm_qcfg(rdev, &l2db_len, &offset);
107 	if (rc)
108 		dev_info(rdev_to_dev(rdev),
109 			 "Couldn't get DB bar size, Low latency framework is disabled\n");
110 	/* set register offsets for both UC and WC */
111 	if (bnxt_qplib_is_chip_gen_p7(cctx)) {
112 		res->dpi_tbl.ucreg.offset = offset;
113 		res->dpi_tbl.wcreg.offset = en_dev->l2_db_size;
114 	} else {
115 		res->dpi_tbl.ucreg.offset = res->is_vf ? BNXT_QPLIB_DBR_VF_DB_OFFSET :
116 							 BNXT_QPLIB_DBR_PF_DB_OFFSET;
117 		res->dpi_tbl.wcreg.offset = res->dpi_tbl.ucreg.offset;
118 	}
119 
120 	/* If WC mapping is disabled by L2 driver then en_dev->l2_db_size
121 	 * is equal to the DB-Bar actual size. This indicates that L2
122 	 * is mapping entire bar as UC-. RoCE driver can't enable WC mapping
123 	 * in such cases and DB-push will be disabled.
124 	 */
125 	barlen = pci_resource_len(res->pdev, RCFW_DBR_PCI_BAR_REGION);
126 	if (cctx->modes.db_push && l2db_len && en_dev->l2_db_size != barlen) {
127 		res->dpi_tbl.wcreg.offset = en_dev->l2_db_size;
128 		dev_info(rdev_to_dev(rdev),  "Low latency framework is enabled\n");
129 	}
130 }
131 
132 static void bnxt_re_set_drv_mode(struct bnxt_re_dev *rdev, u8 mode)
133 {
134 	struct bnxt_qplib_chip_ctx *cctx;
135 
136 	cctx = rdev->chip_ctx;
137 	cctx->modes.wqe_mode = bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx) ?
138 			       mode : BNXT_QPLIB_WQE_MODE_STATIC;
139 	if (bnxt_re_hwrm_qcaps(rdev))
140 		dev_err(rdev_to_dev(rdev),
141 			"Failed to query hwrm qcaps\n");
142 	if (bnxt_qplib_is_chip_gen_p7(rdev->chip_ctx))
143 		cctx->modes.toggle_bits |= BNXT_QPLIB_CQ_TOGGLE_BIT;
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 	chip_ctx = rdev->chip_ctx;
153 	rdev->chip_ctx = NULL;
154 	rdev->rcfw.res = NULL;
155 	rdev->qplib_res.cctx = NULL;
156 	rdev->qplib_res.pdev = NULL;
157 	rdev->qplib_res.netdev = NULL;
158 	kfree(chip_ctx);
159 }
160 
161 static int bnxt_re_setup_chip_ctx(struct bnxt_re_dev *rdev, u8 wqe_mode)
162 {
163 	struct bnxt_qplib_chip_ctx *chip_ctx;
164 	struct bnxt_en_dev *en_dev;
165 	int rc;
166 
167 	en_dev = rdev->en_dev;
168 
169 	chip_ctx = kzalloc(sizeof(*chip_ctx), GFP_KERNEL);
170 	if (!chip_ctx)
171 		return -ENOMEM;
172 	chip_ctx->chip_num = en_dev->chip_num;
173 	chip_ctx->hw_stats_size = en_dev->hw_ring_stats_size;
174 
175 	rdev->chip_ctx = chip_ctx;
176 	/* rest members to follow eventually */
177 
178 	rdev->qplib_res.cctx = rdev->chip_ctx;
179 	rdev->rcfw.res = &rdev->qplib_res;
180 	rdev->qplib_res.dattr = &rdev->dev_attr;
181 	rdev->qplib_res.is_vf = BNXT_EN_VF(en_dev);
182 
183 	bnxt_re_set_drv_mode(rdev, wqe_mode);
184 
185 	bnxt_re_set_db_offset(rdev);
186 	rc = bnxt_qplib_map_db_bar(&rdev->qplib_res);
187 	if (rc)
188 		return rc;
189 
190 	if (bnxt_qplib_determine_atomics(en_dev->pdev))
191 		ibdev_info(&rdev->ibdev,
192 			   "platform doesn't support global atomics.");
193 	return 0;
194 }
195 
196 /* SR-IOV helper functions */
197 
198 static void bnxt_re_get_sriov_func_type(struct bnxt_re_dev *rdev)
199 {
200 	if (BNXT_EN_VF(rdev->en_dev))
201 		rdev->is_virtfn = 1;
202 }
203 
204 /* Set the maximum number of each resource that the driver actually wants
205  * to allocate. This may be up to the maximum number the firmware has
206  * reserved for the function. The driver may choose to allocate fewer
207  * resources than the firmware maximum.
208  */
209 static void bnxt_re_limit_pf_res(struct bnxt_re_dev *rdev)
210 {
211 	struct bnxt_qplib_dev_attr *attr;
212 	struct bnxt_qplib_ctx *ctx;
213 	int i;
214 
215 	attr = &rdev->dev_attr;
216 	ctx = &rdev->qplib_ctx;
217 
218 	ctx->qpc_count = min_t(u32, BNXT_RE_MAX_QPC_COUNT,
219 			       attr->max_qp);
220 	ctx->mrw_count = BNXT_RE_MAX_MRW_COUNT_256K;
221 	/* Use max_mr from fw since max_mrw does not get set */
222 	ctx->mrw_count = min_t(u32, ctx->mrw_count, attr->max_mr);
223 	ctx->srqc_count = min_t(u32, BNXT_RE_MAX_SRQC_COUNT,
224 				attr->max_srq);
225 	ctx->cq_count = min_t(u32, BNXT_RE_MAX_CQ_COUNT, attr->max_cq);
226 	if (!bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx))
227 		for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
228 			rdev->qplib_ctx.tqm_ctx.qcount[i] =
229 			rdev->dev_attr.tqm_alloc_reqs[i];
230 }
231 
232 static void bnxt_re_limit_vf_res(struct bnxt_qplib_ctx *qplib_ctx, u32 num_vf)
233 {
234 	struct bnxt_qplib_vf_res *vf_res;
235 	u32 mrws = 0;
236 	u32 vf_pct;
237 	u32 nvfs;
238 
239 	vf_res = &qplib_ctx->vf_res;
240 	/*
241 	 * Reserve a set of resources for the PF. Divide the remaining
242 	 * resources among the VFs
243 	 */
244 	vf_pct = 100 - BNXT_RE_PCT_RSVD_FOR_PF;
245 	nvfs = num_vf;
246 	num_vf = 100 * num_vf;
247 	vf_res->max_qp_per_vf = (qplib_ctx->qpc_count * vf_pct) / num_vf;
248 	vf_res->max_srq_per_vf = (qplib_ctx->srqc_count * vf_pct) / num_vf;
249 	vf_res->max_cq_per_vf = (qplib_ctx->cq_count * vf_pct) / num_vf;
250 	/*
251 	 * The driver allows many more MRs than other resources. If the
252 	 * firmware does also, then reserve a fixed amount for the PF and
253 	 * divide the rest among VFs. VFs may use many MRs for NFS
254 	 * mounts, ISER, NVME applications, etc. If the firmware severely
255 	 * restricts the number of MRs, then let PF have half and divide
256 	 * the rest among VFs, as for the other resource types.
257 	 */
258 	if (qplib_ctx->mrw_count < BNXT_RE_MAX_MRW_COUNT_64K) {
259 		mrws = qplib_ctx->mrw_count * vf_pct;
260 		nvfs = num_vf;
261 	} else {
262 		mrws = qplib_ctx->mrw_count - BNXT_RE_RESVD_MR_FOR_PF;
263 	}
264 	vf_res->max_mrw_per_vf = (mrws / nvfs);
265 	vf_res->max_gid_per_vf = BNXT_RE_MAX_GID_PER_VF;
266 }
267 
268 static void bnxt_re_set_resource_limits(struct bnxt_re_dev *rdev)
269 {
270 	u32 num_vfs;
271 
272 	memset(&rdev->qplib_ctx.vf_res, 0, sizeof(struct bnxt_qplib_vf_res));
273 	bnxt_re_limit_pf_res(rdev);
274 
275 	num_vfs =  bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx) ?
276 			BNXT_RE_GEN_P5_MAX_VF : rdev->num_vfs;
277 	if (num_vfs)
278 		bnxt_re_limit_vf_res(&rdev->qplib_ctx, num_vfs);
279 }
280 
281 static void bnxt_re_vf_res_config(struct bnxt_re_dev *rdev)
282 {
283 	rdev->num_vfs = pci_sriov_get_totalvfs(rdev->en_dev->pdev);
284 	if (!bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx)) {
285 		bnxt_re_set_resource_limits(rdev);
286 		bnxt_qplib_set_func_resources(&rdev->qplib_res, &rdev->rcfw,
287 					      &rdev->qplib_ctx);
288 	}
289 }
290 
291 static void bnxt_re_shutdown(struct auxiliary_device *adev)
292 {
293 	struct bnxt_re_dev *rdev = auxiliary_get_drvdata(adev);
294 
295 	if (!rdev)
296 		return;
297 	ib_unregister_device(&rdev->ibdev);
298 	bnxt_re_dev_uninit(rdev);
299 }
300 
301 static void bnxt_re_stop_irq(void *handle)
302 {
303 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
304 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
305 	struct bnxt_qplib_nq *nq;
306 	int indx;
307 
308 	for (indx = BNXT_RE_NQ_IDX; indx < rdev->num_msix; indx++) {
309 		nq = &rdev->nq[indx - 1];
310 		bnxt_qplib_nq_stop_irq(nq, false);
311 	}
312 
313 	bnxt_qplib_rcfw_stop_irq(rcfw, false);
314 }
315 
316 static void bnxt_re_start_irq(void *handle, struct bnxt_msix_entry *ent)
317 {
318 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
319 	struct bnxt_msix_entry *msix_ent = rdev->en_dev->msix_entries;
320 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
321 	struct bnxt_qplib_nq *nq;
322 	int indx, rc;
323 
324 	if (!ent) {
325 		/* Not setting the f/w timeout bit in rcfw.
326 		 * During the driver unload the first command
327 		 * to f/w will timeout and that will set the
328 		 * timeout bit.
329 		 */
330 		ibdev_err(&rdev->ibdev, "Failed to re-start IRQs\n");
331 		return;
332 	}
333 
334 	/* Vectors may change after restart, so update with new vectors
335 	 * in device sctructure.
336 	 */
337 	for (indx = 0; indx < rdev->num_msix; indx++)
338 		rdev->en_dev->msix_entries[indx].vector = ent[indx].vector;
339 
340 	rc = bnxt_qplib_rcfw_start_irq(rcfw, msix_ent[BNXT_RE_AEQ_IDX].vector,
341 				       false);
342 	if (rc) {
343 		ibdev_warn(&rdev->ibdev, "Failed to reinit CREQ\n");
344 		return;
345 	}
346 	for (indx = BNXT_RE_NQ_IDX ; indx < rdev->num_msix; indx++) {
347 		nq = &rdev->nq[indx - 1];
348 		rc = bnxt_qplib_nq_start_irq(nq, indx - 1,
349 					     msix_ent[indx].vector, false);
350 		if (rc) {
351 			ibdev_warn(&rdev->ibdev, "Failed to reinit NQ index %d\n",
352 				   indx - 1);
353 			return;
354 		}
355 	}
356 }
357 
358 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
359 	.ulp_irq_stop = bnxt_re_stop_irq,
360 	.ulp_irq_restart = bnxt_re_start_irq
361 };
362 
363 /* RoCE -> Net driver */
364 
365 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
366 {
367 	struct bnxt_en_dev *en_dev;
368 	int rc;
369 
370 	en_dev = rdev->en_dev;
371 
372 	rc = bnxt_register_dev(en_dev, &bnxt_re_ulp_ops, rdev);
373 	if (!rc)
374 		rdev->qplib_res.pdev = rdev->en_dev->pdev;
375 	return rc;
376 }
377 
378 static void bnxt_re_init_hwrm_hdr(struct input *hdr, u16 opcd)
379 {
380 	hdr->req_type = cpu_to_le16(opcd);
381 	hdr->cmpl_ring = cpu_to_le16(-1);
382 	hdr->target_id = cpu_to_le16(-1);
383 }
384 
385 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
386 				int msg_len, void *resp, int resp_max_len,
387 				int timeout)
388 {
389 	fw_msg->msg = msg;
390 	fw_msg->msg_len = msg_len;
391 	fw_msg->resp = resp;
392 	fw_msg->resp_max_len = resp_max_len;
393 	fw_msg->timeout = timeout;
394 }
395 
396 /* Query device config using common hwrm */
397 static int bnxt_re_hwrm_qcfg(struct bnxt_re_dev *rdev, u32 *db_len,
398 			     u32 *offset)
399 {
400 	struct bnxt_en_dev *en_dev = rdev->en_dev;
401 	struct hwrm_func_qcfg_output resp = {0};
402 	struct hwrm_func_qcfg_input req = {0};
403 	struct bnxt_fw_msg fw_msg = {};
404 	int rc;
405 
406 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_FUNC_QCFG);
407 	req.fid = cpu_to_le16(0xffff);
408 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
409 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
410 	rc = bnxt_send_msg(en_dev, &fw_msg);
411 	if (!rc) {
412 		*db_len = PAGE_ALIGN(le16_to_cpu(resp.l2_doorbell_bar_size_kb) * 1024);
413 		*offset = PAGE_ALIGN(le16_to_cpu(resp.legacy_l2_db_size_kb) * 1024);
414 	}
415 	return rc;
416 }
417 
418 /* Query function capabilities using common hwrm */
419 int bnxt_re_hwrm_qcaps(struct bnxt_re_dev *rdev)
420 {
421 	struct bnxt_en_dev *en_dev = rdev->en_dev;
422 	struct hwrm_func_qcaps_output resp = {};
423 	struct hwrm_func_qcaps_input req = {};
424 	struct bnxt_qplib_chip_ctx *cctx;
425 	struct bnxt_fw_msg fw_msg = {};
426 	int rc;
427 
428 	cctx = rdev->chip_ctx;
429 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_FUNC_QCAPS);
430 	req.fid = cpu_to_le16(0xffff);
431 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
432 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
433 
434 	rc = bnxt_send_msg(en_dev, &fw_msg);
435 	if (rc)
436 		return rc;
437 	cctx->modes.db_push = le32_to_cpu(resp.flags) & FUNC_QCAPS_RESP_FLAGS_WCB_PUSH_MODE;
438 
439 	cctx->modes.dbr_pacing =
440 		le32_to_cpu(resp.flags_ext2) &
441 		FUNC_QCAPS_RESP_FLAGS_EXT2_DBR_PACING_EXT_SUPPORTED;
442 	return 0;
443 }
444 
445 static int bnxt_re_hwrm_dbr_pacing_qcfg(struct bnxt_re_dev *rdev)
446 {
447 	struct hwrm_func_dbr_pacing_qcfg_output resp = {};
448 	struct hwrm_func_dbr_pacing_qcfg_input req = {};
449 	struct bnxt_en_dev *en_dev = rdev->en_dev;
450 	struct bnxt_qplib_chip_ctx *cctx;
451 	struct bnxt_fw_msg fw_msg = {};
452 	int rc;
453 
454 	cctx = rdev->chip_ctx;
455 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_FUNC_DBR_PACING_QCFG);
456 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
457 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
458 	rc = bnxt_send_msg(en_dev, &fw_msg);
459 	if (rc)
460 		return rc;
461 
462 	if ((le32_to_cpu(resp.dbr_stat_db_fifo_reg) &
463 	    FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_MASK) ==
464 		FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_GRC)
465 		cctx->dbr_stat_db_fifo =
466 			le32_to_cpu(resp.dbr_stat_db_fifo_reg) &
467 			~FUNC_DBR_PACING_QCFG_RESP_DBR_STAT_DB_FIFO_REG_ADDR_SPACE_MASK;
468 	return 0;
469 }
470 
471 /* Update the pacing tunable parameters to the default values */
472 static void bnxt_re_set_default_pacing_data(struct bnxt_re_dev *rdev)
473 {
474 	struct bnxt_qplib_db_pacing_data *pacing_data = rdev->qplib_res.pacing_data;
475 
476 	pacing_data->do_pacing = rdev->pacing.dbr_def_do_pacing;
477 	pacing_data->pacing_th = rdev->pacing.pacing_algo_th;
478 	pacing_data->alarm_th =
479 		pacing_data->pacing_th * BNXT_RE_PACING_ALARM_TH_MULTIPLE;
480 }
481 
482 static void __wait_for_fifo_occupancy_below_th(struct bnxt_re_dev *rdev)
483 {
484 	u32 read_val, fifo_occup;
485 
486 	/* loop shouldn't run infintely as the occupancy usually goes
487 	 * below pacing algo threshold as soon as pacing kicks in.
488 	 */
489 	while (1) {
490 		read_val = readl(rdev->en_dev->bar0 + rdev->pacing.dbr_db_fifo_reg_off);
491 		fifo_occup = BNXT_RE_MAX_FIFO_DEPTH -
492 			((read_val & BNXT_RE_DB_FIFO_ROOM_MASK) >>
493 			 BNXT_RE_DB_FIFO_ROOM_SHIFT);
494 		/* Fifo occupancy cannot be greater the MAX FIFO depth */
495 		if (fifo_occup > BNXT_RE_MAX_FIFO_DEPTH)
496 			break;
497 
498 		if (fifo_occup < rdev->qplib_res.pacing_data->pacing_th)
499 			break;
500 	}
501 }
502 
503 static void bnxt_re_db_fifo_check(struct work_struct *work)
504 {
505 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
506 			dbq_fifo_check_work);
507 	struct bnxt_qplib_db_pacing_data *pacing_data;
508 	u32 pacing_save;
509 
510 	if (!mutex_trylock(&rdev->pacing.dbq_lock))
511 		return;
512 	pacing_data = rdev->qplib_res.pacing_data;
513 	pacing_save = rdev->pacing.do_pacing_save;
514 	__wait_for_fifo_occupancy_below_th(rdev);
515 	cancel_delayed_work_sync(&rdev->dbq_pacing_work);
516 	if (pacing_save > rdev->pacing.dbr_def_do_pacing) {
517 		/* Double the do_pacing value during the congestion */
518 		pacing_save = pacing_save << 1;
519 	} else {
520 		/*
521 		 * when a new congestion is detected increase the do_pacing
522 		 * by 8 times. And also increase the pacing_th by 4 times. The
523 		 * reason to increase pacing_th is to give more space for the
524 		 * queue to oscillate down without getting empty, but also more
525 		 * room for the queue to increase without causing another alarm.
526 		 */
527 		pacing_save = pacing_save << 3;
528 		pacing_data->pacing_th = rdev->pacing.pacing_algo_th * 4;
529 	}
530 
531 	if (pacing_save > BNXT_RE_MAX_DBR_DO_PACING)
532 		pacing_save = BNXT_RE_MAX_DBR_DO_PACING;
533 
534 	pacing_data->do_pacing = pacing_save;
535 	rdev->pacing.do_pacing_save = pacing_data->do_pacing;
536 	pacing_data->alarm_th =
537 		pacing_data->pacing_th * BNXT_RE_PACING_ALARM_TH_MULTIPLE;
538 	schedule_delayed_work(&rdev->dbq_pacing_work,
539 			      msecs_to_jiffies(rdev->pacing.dbq_pacing_time));
540 	rdev->stats.pacing.alerts++;
541 	mutex_unlock(&rdev->pacing.dbq_lock);
542 }
543 
544 static void bnxt_re_pacing_timer_exp(struct work_struct *work)
545 {
546 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
547 			dbq_pacing_work.work);
548 	struct bnxt_qplib_db_pacing_data *pacing_data;
549 	u32 read_val, fifo_occup;
550 
551 	if (!mutex_trylock(&rdev->pacing.dbq_lock))
552 		return;
553 
554 	pacing_data = rdev->qplib_res.pacing_data;
555 	read_val = readl(rdev->en_dev->bar0 + rdev->pacing.dbr_db_fifo_reg_off);
556 	fifo_occup = BNXT_RE_MAX_FIFO_DEPTH -
557 		((read_val & BNXT_RE_DB_FIFO_ROOM_MASK) >>
558 		 BNXT_RE_DB_FIFO_ROOM_SHIFT);
559 
560 	if (fifo_occup > pacing_data->pacing_th)
561 		goto restart_timer;
562 
563 	/*
564 	 * Instead of immediately going back to the default do_pacing
565 	 * reduce it by 1/8 times and restart the timer.
566 	 */
567 	pacing_data->do_pacing = pacing_data->do_pacing - (pacing_data->do_pacing >> 3);
568 	pacing_data->do_pacing = max_t(u32, rdev->pacing.dbr_def_do_pacing, pacing_data->do_pacing);
569 	if (pacing_data->do_pacing <= rdev->pacing.dbr_def_do_pacing) {
570 		bnxt_re_set_default_pacing_data(rdev);
571 		rdev->stats.pacing.complete++;
572 		goto dbq_unlock;
573 	}
574 
575 restart_timer:
576 	schedule_delayed_work(&rdev->dbq_pacing_work,
577 			      msecs_to_jiffies(rdev->pacing.dbq_pacing_time));
578 	rdev->stats.pacing.resched++;
579 dbq_unlock:
580 	rdev->pacing.do_pacing_save = pacing_data->do_pacing;
581 	mutex_unlock(&rdev->pacing.dbq_lock);
582 }
583 
584 void bnxt_re_pacing_alert(struct bnxt_re_dev *rdev)
585 {
586 	struct bnxt_qplib_db_pacing_data *pacing_data;
587 
588 	if (!rdev->pacing.dbr_pacing)
589 		return;
590 	mutex_lock(&rdev->pacing.dbq_lock);
591 	pacing_data = rdev->qplib_res.pacing_data;
592 
593 	/*
594 	 * Increase the alarm_th to max so that other user lib instances do not
595 	 * keep alerting the driver.
596 	 */
597 	pacing_data->alarm_th = BNXT_RE_MAX_FIFO_DEPTH;
598 	pacing_data->do_pacing = BNXT_RE_MAX_DBR_DO_PACING;
599 	cancel_work_sync(&rdev->dbq_fifo_check_work);
600 	schedule_work(&rdev->dbq_fifo_check_work);
601 	mutex_unlock(&rdev->pacing.dbq_lock);
602 }
603 
604 static int bnxt_re_initialize_dbr_pacing(struct bnxt_re_dev *rdev)
605 {
606 	if (bnxt_re_hwrm_dbr_pacing_qcfg(rdev))
607 		return -EIO;
608 
609 	/* Allocate a page for app use */
610 	rdev->pacing.dbr_page = (void *)__get_free_page(GFP_KERNEL);
611 	if (!rdev->pacing.dbr_page)
612 		return -ENOMEM;
613 
614 	memset((u8 *)rdev->pacing.dbr_page, 0, PAGE_SIZE);
615 	rdev->qplib_res.pacing_data = (struct bnxt_qplib_db_pacing_data *)rdev->pacing.dbr_page;
616 
617 	/* MAP HW window 2 for reading db fifo depth */
618 	writel(rdev->chip_ctx->dbr_stat_db_fifo & BNXT_GRC_BASE_MASK,
619 	       rdev->en_dev->bar0 + BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
620 	rdev->pacing.dbr_db_fifo_reg_off =
621 		(rdev->chip_ctx->dbr_stat_db_fifo & BNXT_GRC_OFFSET_MASK) +
622 		 BNXT_RE_GRC_FIFO_REG_BASE;
623 	rdev->pacing.dbr_bar_addr =
624 		pci_resource_start(rdev->qplib_res.pdev, 0) + rdev->pacing.dbr_db_fifo_reg_off;
625 
626 	rdev->pacing.pacing_algo_th = BNXT_RE_PACING_ALGO_THRESHOLD;
627 	rdev->pacing.dbq_pacing_time = BNXT_RE_DBR_PACING_TIME;
628 	rdev->pacing.dbr_def_do_pacing = BNXT_RE_DBR_DO_PACING_NO_CONGESTION;
629 	rdev->pacing.do_pacing_save = rdev->pacing.dbr_def_do_pacing;
630 	rdev->qplib_res.pacing_data->fifo_max_depth = BNXT_RE_MAX_FIFO_DEPTH;
631 	rdev->qplib_res.pacing_data->fifo_room_mask = BNXT_RE_DB_FIFO_ROOM_MASK;
632 	rdev->qplib_res.pacing_data->fifo_room_shift = BNXT_RE_DB_FIFO_ROOM_SHIFT;
633 	rdev->qplib_res.pacing_data->grc_reg_offset = rdev->pacing.dbr_db_fifo_reg_off;
634 	bnxt_re_set_default_pacing_data(rdev);
635 	/* Initialize worker for DBR Pacing */
636 	INIT_WORK(&rdev->dbq_fifo_check_work, bnxt_re_db_fifo_check);
637 	INIT_DELAYED_WORK(&rdev->dbq_pacing_work, bnxt_re_pacing_timer_exp);
638 	return 0;
639 }
640 
641 static void bnxt_re_deinitialize_dbr_pacing(struct bnxt_re_dev *rdev)
642 {
643 	cancel_work_sync(&rdev->dbq_fifo_check_work);
644 	cancel_delayed_work_sync(&rdev->dbq_pacing_work);
645 	if (rdev->pacing.dbr_page)
646 		free_page((u64)rdev->pacing.dbr_page);
647 
648 	rdev->pacing.dbr_page = NULL;
649 	rdev->pacing.dbr_pacing = false;
650 }
651 
652 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev,
653 				 u16 fw_ring_id, int type)
654 {
655 	struct bnxt_en_dev *en_dev;
656 	struct hwrm_ring_free_input req = {};
657 	struct hwrm_ring_free_output resp;
658 	struct bnxt_fw_msg fw_msg = {};
659 	int rc = -EINVAL;
660 
661 	if (!rdev)
662 		return rc;
663 
664 	en_dev = rdev->en_dev;
665 
666 	if (!en_dev)
667 		return rc;
668 
669 	if (test_bit(BNXT_RE_FLAG_ERR_DEVICE_DETACHED, &rdev->flags))
670 		return 0;
671 
672 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_RING_FREE);
673 	req.ring_type = type;
674 	req.ring_id = cpu_to_le16(fw_ring_id);
675 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
676 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
677 	rc = bnxt_send_msg(en_dev, &fw_msg);
678 	if (rc)
679 		ibdev_err(&rdev->ibdev, "Failed to free HW ring:%d :%#x",
680 			  req.ring_id, rc);
681 	return rc;
682 }
683 
684 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev,
685 				  struct bnxt_re_ring_attr *ring_attr,
686 				  u16 *fw_ring_id)
687 {
688 	struct bnxt_en_dev *en_dev = rdev->en_dev;
689 	struct hwrm_ring_alloc_input req = {};
690 	struct hwrm_ring_alloc_output resp;
691 	struct bnxt_fw_msg fw_msg = {};
692 	int rc = -EINVAL;
693 
694 	if (!en_dev)
695 		return rc;
696 
697 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_RING_ALLOC);
698 	req.enables = 0;
699 	req.page_tbl_addr =  cpu_to_le64(ring_attr->dma_arr[0]);
700 	if (ring_attr->pages > 1) {
701 		/* Page size is in log2 units */
702 		req.page_size = BNXT_PAGE_SHIFT;
703 		req.page_tbl_depth = 1;
704 	}
705 	req.fbo = 0;
706 	/* Association of ring index with doorbell index and MSIX number */
707 	req.logical_id = cpu_to_le16(ring_attr->lrid);
708 	req.length = cpu_to_le32(ring_attr->depth + 1);
709 	req.ring_type = ring_attr->type;
710 	req.int_mode = ring_attr->mode;
711 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
712 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
713 	rc = bnxt_send_msg(en_dev, &fw_msg);
714 	if (!rc)
715 		*fw_ring_id = le16_to_cpu(resp.ring_id);
716 
717 	return rc;
718 }
719 
720 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
721 				      u32 fw_stats_ctx_id)
722 {
723 	struct bnxt_en_dev *en_dev = rdev->en_dev;
724 	struct hwrm_stat_ctx_free_input req = {};
725 	struct hwrm_stat_ctx_free_output resp = {};
726 	struct bnxt_fw_msg fw_msg = {};
727 	int rc = -EINVAL;
728 
729 	if (!en_dev)
730 		return rc;
731 
732 	if (test_bit(BNXT_RE_FLAG_ERR_DEVICE_DETACHED, &rdev->flags))
733 		return 0;
734 
735 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_STAT_CTX_FREE);
736 	req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
737 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
738 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
739 	rc = bnxt_send_msg(en_dev, &fw_msg);
740 	if (rc)
741 		ibdev_err(&rdev->ibdev, "Failed to free HW stats context %#x",
742 			  rc);
743 
744 	return rc;
745 }
746 
747 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
748 				       dma_addr_t dma_map,
749 				       u32 *fw_stats_ctx_id)
750 {
751 	struct bnxt_qplib_chip_ctx *chip_ctx = rdev->chip_ctx;
752 	struct hwrm_stat_ctx_alloc_output resp = {};
753 	struct hwrm_stat_ctx_alloc_input req = {};
754 	struct bnxt_en_dev *en_dev = rdev->en_dev;
755 	struct bnxt_fw_msg fw_msg = {};
756 	int rc = -EINVAL;
757 
758 	*fw_stats_ctx_id = INVALID_STATS_CTX_ID;
759 
760 	if (!en_dev)
761 		return rc;
762 
763 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_STAT_CTX_ALLOC);
764 	req.update_period_ms = cpu_to_le32(1000);
765 	req.stats_dma_addr = cpu_to_le64(dma_map);
766 	req.stats_dma_length = cpu_to_le16(chip_ctx->hw_stats_size);
767 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
768 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
769 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
770 	rc = bnxt_send_msg(en_dev, &fw_msg);
771 	if (!rc)
772 		*fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
773 
774 	return rc;
775 }
776 
777 static void bnxt_re_disassociate_ucontext(struct ib_ucontext *ibcontext)
778 {
779 }
780 
781 /* Device */
782 
783 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
784 {
785 	struct ib_device *ibdev =
786 		ib_device_get_by_netdev(netdev, RDMA_DRIVER_BNXT_RE);
787 	if (!ibdev)
788 		return NULL;
789 
790 	return container_of(ibdev, struct bnxt_re_dev, ibdev);
791 }
792 
793 static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr,
794 			   char *buf)
795 {
796 	struct bnxt_re_dev *rdev =
797 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
798 
799 	return sysfs_emit(buf, "0x%x\n", rdev->en_dev->pdev->vendor);
800 }
801 static DEVICE_ATTR_RO(hw_rev);
802 
803 static ssize_t hca_type_show(struct device *device,
804 			     struct device_attribute *attr, char *buf)
805 {
806 	struct bnxt_re_dev *rdev =
807 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
808 
809 	return sysfs_emit(buf, "%s\n", rdev->ibdev.node_desc);
810 }
811 static DEVICE_ATTR_RO(hca_type);
812 
813 static struct attribute *bnxt_re_attributes[] = {
814 	&dev_attr_hw_rev.attr,
815 	&dev_attr_hca_type.attr,
816 	NULL
817 };
818 
819 static const struct attribute_group bnxt_re_dev_attr_group = {
820 	.attrs = bnxt_re_attributes,
821 };
822 
823 static const struct ib_device_ops bnxt_re_dev_ops = {
824 	.owner = THIS_MODULE,
825 	.driver_id = RDMA_DRIVER_BNXT_RE,
826 	.uverbs_abi_ver = BNXT_RE_ABI_VERSION,
827 
828 	.add_gid = bnxt_re_add_gid,
829 	.alloc_hw_port_stats = bnxt_re_ib_alloc_hw_port_stats,
830 	.alloc_mr = bnxt_re_alloc_mr,
831 	.alloc_pd = bnxt_re_alloc_pd,
832 	.alloc_ucontext = bnxt_re_alloc_ucontext,
833 	.create_ah = bnxt_re_create_ah,
834 	.create_cq = bnxt_re_create_cq,
835 	.create_qp = bnxt_re_create_qp,
836 	.create_srq = bnxt_re_create_srq,
837 	.create_user_ah = bnxt_re_create_ah,
838 	.dealloc_pd = bnxt_re_dealloc_pd,
839 	.dealloc_ucontext = bnxt_re_dealloc_ucontext,
840 	.del_gid = bnxt_re_del_gid,
841 	.dereg_mr = bnxt_re_dereg_mr,
842 	.destroy_ah = bnxt_re_destroy_ah,
843 	.destroy_cq = bnxt_re_destroy_cq,
844 	.destroy_qp = bnxt_re_destroy_qp,
845 	.destroy_srq = bnxt_re_destroy_srq,
846 	.device_group = &bnxt_re_dev_attr_group,
847 	.disassociate_ucontext = bnxt_re_disassociate_ucontext,
848 	.get_dev_fw_str = bnxt_re_query_fw_str,
849 	.get_dma_mr = bnxt_re_get_dma_mr,
850 	.get_hw_stats = bnxt_re_ib_get_hw_stats,
851 	.get_link_layer = bnxt_re_get_link_layer,
852 	.get_port_immutable = bnxt_re_get_port_immutable,
853 	.map_mr_sg = bnxt_re_map_mr_sg,
854 	.mmap = bnxt_re_mmap,
855 	.mmap_free = bnxt_re_mmap_free,
856 	.modify_qp = bnxt_re_modify_qp,
857 	.modify_srq = bnxt_re_modify_srq,
858 	.poll_cq = bnxt_re_poll_cq,
859 	.post_recv = bnxt_re_post_recv,
860 	.post_send = bnxt_re_post_send,
861 	.post_srq_recv = bnxt_re_post_srq_recv,
862 	.query_ah = bnxt_re_query_ah,
863 	.query_device = bnxt_re_query_device,
864 	.query_pkey = bnxt_re_query_pkey,
865 	.query_port = bnxt_re_query_port,
866 	.query_qp = bnxt_re_query_qp,
867 	.query_srq = bnxt_re_query_srq,
868 	.reg_user_mr = bnxt_re_reg_user_mr,
869 	.reg_user_mr_dmabuf = bnxt_re_reg_user_mr_dmabuf,
870 	.req_notify_cq = bnxt_re_req_notify_cq,
871 	.resize_cq = bnxt_re_resize_cq,
872 	INIT_RDMA_OBJ_SIZE(ib_ah, bnxt_re_ah, ib_ah),
873 	INIT_RDMA_OBJ_SIZE(ib_cq, bnxt_re_cq, ib_cq),
874 	INIT_RDMA_OBJ_SIZE(ib_pd, bnxt_re_pd, ib_pd),
875 	INIT_RDMA_OBJ_SIZE(ib_qp, bnxt_re_qp, ib_qp),
876 	INIT_RDMA_OBJ_SIZE(ib_srq, bnxt_re_srq, ib_srq),
877 	INIT_RDMA_OBJ_SIZE(ib_ucontext, bnxt_re_ucontext, ib_uctx),
878 };
879 
880 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
881 {
882 	struct ib_device *ibdev = &rdev->ibdev;
883 	int ret;
884 
885 	/* ib device init */
886 	ibdev->node_type = RDMA_NODE_IB_CA;
887 	strscpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
888 		strlen(BNXT_RE_DESC) + 5);
889 	ibdev->phys_port_cnt = 1;
890 
891 	addrconf_addr_eui48((u8 *)&ibdev->node_guid, rdev->netdev->dev_addr);
892 
893 	ibdev->num_comp_vectors	= rdev->num_msix - 1;
894 	ibdev->dev.parent = &rdev->en_dev->pdev->dev;
895 	ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
896 
897 	if (IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS))
898 		ibdev->driver_def = bnxt_re_uapi_defs;
899 
900 	ib_set_device_ops(ibdev, &bnxt_re_dev_ops);
901 	ret = ib_device_set_netdev(&rdev->ibdev, rdev->netdev, 1);
902 	if (ret)
903 		return ret;
904 
905 	dma_set_max_seg_size(&rdev->en_dev->pdev->dev, UINT_MAX);
906 	ibdev->uverbs_cmd_mask |= BIT_ULL(IB_USER_VERBS_CMD_POLL_CQ);
907 	return ib_register_device(ibdev, "bnxt_re%d", &rdev->en_dev->pdev->dev);
908 }
909 
910 static struct bnxt_re_dev *bnxt_re_dev_add(struct bnxt_aux_priv *aux_priv,
911 					   struct bnxt_en_dev *en_dev)
912 {
913 	struct bnxt_re_dev *rdev;
914 
915 	/* Allocate bnxt_re_dev instance here */
916 	rdev = ib_alloc_device(bnxt_re_dev, ibdev);
917 	if (!rdev) {
918 		ibdev_err(NULL, "%s: bnxt_re_dev allocation failure!",
919 			  ROCE_DRV_MODULE_NAME);
920 		return NULL;
921 	}
922 	/* Default values */
923 	rdev->nb.notifier_call = NULL;
924 	rdev->netdev = en_dev->net;
925 	rdev->en_dev = en_dev;
926 	rdev->id = rdev->en_dev->pdev->devfn;
927 	INIT_LIST_HEAD(&rdev->qp_list);
928 	mutex_init(&rdev->qp_lock);
929 	mutex_init(&rdev->pacing.dbq_lock);
930 	atomic_set(&rdev->stats.res.qp_count, 0);
931 	atomic_set(&rdev->stats.res.cq_count, 0);
932 	atomic_set(&rdev->stats.res.srq_count, 0);
933 	atomic_set(&rdev->stats.res.mr_count, 0);
934 	atomic_set(&rdev->stats.res.mw_count, 0);
935 	atomic_set(&rdev->stats.res.ah_count, 0);
936 	atomic_set(&rdev->stats.res.pd_count, 0);
937 	rdev->cosq[0] = 0xFFFF;
938 	rdev->cosq[1] = 0xFFFF;
939 
940 	return rdev;
941 }
942 
943 static int bnxt_re_handle_unaffi_async_event(struct creq_func_event
944 					     *unaffi_async)
945 {
946 	switch (unaffi_async->event) {
947 	case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
948 		break;
949 	case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
950 		break;
951 	case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
952 		break;
953 	case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
954 		break;
955 	case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
956 		break;
957 	case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
958 		break;
959 	case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
960 		break;
961 	case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
962 		break;
963 	case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
964 		break;
965 	case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
966 		break;
967 	case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
968 		break;
969 	default:
970 		return -EINVAL;
971 	}
972 	return 0;
973 }
974 
975 static int bnxt_re_handle_qp_async_event(struct creq_qp_event *qp_event,
976 					 struct bnxt_re_qp *qp)
977 {
978 	struct bnxt_re_srq *srq = container_of(qp->qplib_qp.srq, struct bnxt_re_srq,
979 					       qplib_srq);
980 	struct creq_qp_error_notification *err_event;
981 	struct ib_event event = {};
982 	unsigned int flags;
983 
984 	if (qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_ERR &&
985 	    rdma_is_kernel_res(&qp->ib_qp.res)) {
986 		flags = bnxt_re_lock_cqs(qp);
987 		bnxt_qplib_add_flush_qp(&qp->qplib_qp);
988 		bnxt_re_unlock_cqs(qp, flags);
989 	}
990 
991 	event.device = &qp->rdev->ibdev;
992 	event.element.qp = &qp->ib_qp;
993 	event.event = IB_EVENT_QP_FATAL;
994 
995 	err_event = (struct creq_qp_error_notification *)qp_event;
996 
997 	switch (err_event->req_err_state_reason) {
998 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_OPCODE_ERROR:
999 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_TIMEOUT_RETRY_LIMIT:
1000 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_RNR_TIMEOUT_RETRY_LIMIT:
1001 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_NAK_ARRIVAL_2:
1002 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_NAK_ARRIVAL_3:
1003 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_INVALID_READ_RESP:
1004 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_ILLEGAL_BIND:
1005 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_ILLEGAL_FAST_REG:
1006 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_ILLEGAL_INVALIDATE:
1007 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_RETRAN_LOCAL_ERROR:
1008 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_AV_DOMAIN_ERROR:
1009 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_PROD_WQE_MSMTCH_ERROR:
1010 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_PSN_RANGE_CHECK_ERROR:
1011 		event.event = IB_EVENT_QP_ACCESS_ERR;
1012 		break;
1013 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_NAK_ARRIVAL_1:
1014 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_NAK_ARRIVAL_4:
1015 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_READ_RESP_LENGTH:
1016 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_WQE_FORMAT_ERROR:
1017 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_ORRQ_FORMAT_ERROR:
1018 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_INVALID_AVID_ERROR:
1019 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_SERV_TYPE_ERROR:
1020 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_INVALID_OP_ERROR:
1021 		event.event = IB_EVENT_QP_REQ_ERR;
1022 		break;
1023 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_RX_MEMORY_ERROR:
1024 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_TX_MEMORY_ERROR:
1025 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_CMP_ERROR:
1026 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_CQ_LOAD_ERROR:
1027 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_TX_PCI_ERROR:
1028 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_RX_PCI_ERROR:
1029 	case CREQ_QP_ERROR_NOTIFICATION_REQ_ERR_STATE_REASON_REQ_RETX_SETUP_ERROR:
1030 		event.event = IB_EVENT_QP_FATAL;
1031 		break;
1032 
1033 	default:
1034 		break;
1035 	}
1036 
1037 	switch (err_event->res_err_state_reason) {
1038 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_EXCEED_MAX:
1039 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_PAYLOAD_LENGTH_MISMATCH:
1040 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_PSN_SEQ_ERROR_RETRY_LIMIT:
1041 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_RX_INVALID_R_KEY:
1042 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_RX_DOMAIN_ERROR:
1043 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_RX_NO_PERMISSION:
1044 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_RX_RANGE_ERROR:
1045 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_TX_INVALID_R_KEY:
1046 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_TX_DOMAIN_ERROR:
1047 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_TX_NO_PERMISSION:
1048 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_TX_RANGE_ERROR:
1049 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_UNALIGN_ATOMIC:
1050 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_PSN_NOT_FOUND:
1051 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_INVALID_DUP_RKEY:
1052 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_IRRQ_FORMAT_ERROR:
1053 		event.event = IB_EVENT_QP_ACCESS_ERR;
1054 		break;
1055 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_EXCEEDS_WQE:
1056 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_WQE_FORMAT_ERROR:
1057 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_UNSUPPORTED_OPCODE:
1058 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_REM_INVALIDATE:
1059 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_OPCODE_ERROR:
1060 		event.event = IB_EVENT_QP_REQ_ERR;
1061 		break;
1062 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_IRRQ_OFLOW:
1063 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_CMP_ERROR:
1064 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_CQ_LOAD_ERROR:
1065 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_TX_PCI_ERROR:
1066 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_RX_PCI_ERROR:
1067 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_MEMORY_ERROR:
1068 		event.event = IB_EVENT_QP_FATAL;
1069 		break;
1070 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_SRQ_LOAD_ERROR:
1071 	case CREQ_QP_ERROR_NOTIFICATION_RES_ERR_STATE_REASON_RES_SRQ_ERROR:
1072 		if (srq)
1073 			event.event = IB_EVENT_SRQ_ERR;
1074 		break;
1075 	default:
1076 		break;
1077 	}
1078 
1079 	if (err_event->res_err_state_reason || err_event->req_err_state_reason) {
1080 		ibdev_dbg(&qp->rdev->ibdev,
1081 			  "%s %s qp_id: %d cons (%d %d) req (%d %d) res (%d %d)\n",
1082 			   __func__, rdma_is_kernel_res(&qp->ib_qp.res) ? "kernel" : "user",
1083 			   qp->qplib_qp.id,
1084 			   err_event->sq_cons_idx,
1085 			   err_event->rq_cons_idx,
1086 			   err_event->req_slow_path_state,
1087 			   err_event->req_err_state_reason,
1088 			   err_event->res_slow_path_state,
1089 			   err_event->res_err_state_reason);
1090 	} else {
1091 		if (srq)
1092 			event.event = IB_EVENT_QP_LAST_WQE_REACHED;
1093 	}
1094 
1095 	if (event.event == IB_EVENT_SRQ_ERR && srq->ib_srq.event_handler)  {
1096 		(*srq->ib_srq.event_handler)(&event,
1097 				srq->ib_srq.srq_context);
1098 	} else if (event.device && qp->ib_qp.event_handler) {
1099 		qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
1100 	}
1101 
1102 	return 0;
1103 }
1104 
1105 static int bnxt_re_handle_cq_async_error(void *event, struct bnxt_re_cq *cq)
1106 {
1107 	struct creq_cq_error_notification *cqerr;
1108 	struct ib_event ibevent = {};
1109 
1110 	cqerr = event;
1111 	switch (cqerr->cq_err_reason) {
1112 	case CREQ_CQ_ERROR_NOTIFICATION_CQ_ERR_REASON_REQ_CQ_INVALID_ERROR:
1113 	case CREQ_CQ_ERROR_NOTIFICATION_CQ_ERR_REASON_REQ_CQ_OVERFLOW_ERROR:
1114 	case CREQ_CQ_ERROR_NOTIFICATION_CQ_ERR_REASON_REQ_CQ_LOAD_ERROR:
1115 	case CREQ_CQ_ERROR_NOTIFICATION_CQ_ERR_REASON_RES_CQ_INVALID_ERROR:
1116 	case CREQ_CQ_ERROR_NOTIFICATION_CQ_ERR_REASON_RES_CQ_OVERFLOW_ERROR:
1117 	case CREQ_CQ_ERROR_NOTIFICATION_CQ_ERR_REASON_RES_CQ_LOAD_ERROR:
1118 		ibevent.event = IB_EVENT_CQ_ERR;
1119 		break;
1120 	default:
1121 		break;
1122 	}
1123 
1124 	if (ibevent.event == IB_EVENT_CQ_ERR && cq->ib_cq.event_handler) {
1125 		ibevent.element.cq = &cq->ib_cq;
1126 		ibevent.device = &cq->rdev->ibdev;
1127 
1128 		ibdev_dbg(&cq->rdev->ibdev,
1129 			  "%s err reason %d\n", __func__, cqerr->cq_err_reason);
1130 		cq->ib_cq.event_handler(&ibevent, cq->ib_cq.cq_context);
1131 	}
1132 
1133 	return 0;
1134 }
1135 
1136 static int bnxt_re_handle_affi_async_event(struct creq_qp_event *affi_async,
1137 					   void *obj)
1138 {
1139 	struct bnxt_qplib_qp *lib_qp;
1140 	struct bnxt_qplib_cq *lib_cq;
1141 	struct bnxt_re_qp *qp;
1142 	struct bnxt_re_cq *cq;
1143 	int rc = 0;
1144 	u8 event;
1145 
1146 	if (!obj)
1147 		return rc; /* QP was already dead, still return success */
1148 
1149 	event = affi_async->event;
1150 	switch (event) {
1151 	case CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION:
1152 		lib_qp = obj;
1153 		qp = container_of(lib_qp, struct bnxt_re_qp, qplib_qp);
1154 		rc = bnxt_re_handle_qp_async_event(affi_async, qp);
1155 		break;
1156 	case CREQ_QP_EVENT_EVENT_CQ_ERROR_NOTIFICATION:
1157 		lib_cq = obj;
1158 		cq = container_of(lib_cq, struct bnxt_re_cq, qplib_cq);
1159 		rc = bnxt_re_handle_cq_async_error(affi_async, cq);
1160 		break;
1161 	default:
1162 		rc = -EINVAL;
1163 	}
1164 	return rc;
1165 }
1166 
1167 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
1168 			       void *aeqe, void *obj)
1169 {
1170 	struct creq_qp_event *affi_async;
1171 	struct creq_func_event *unaffi_async;
1172 	u8 type;
1173 	int rc;
1174 
1175 	type = ((struct creq_base *)aeqe)->type;
1176 	if (type == CREQ_BASE_TYPE_FUNC_EVENT) {
1177 		unaffi_async = aeqe;
1178 		rc = bnxt_re_handle_unaffi_async_event(unaffi_async);
1179 	} else {
1180 		affi_async = aeqe;
1181 		rc = bnxt_re_handle_affi_async_event(affi_async, obj);
1182 	}
1183 
1184 	return rc;
1185 }
1186 
1187 static int bnxt_re_srqn_handler(struct bnxt_qplib_nq *nq,
1188 				struct bnxt_qplib_srq *handle, u8 event)
1189 {
1190 	struct bnxt_re_srq *srq = container_of(handle, struct bnxt_re_srq,
1191 					       qplib_srq);
1192 	struct ib_event ib_event;
1193 
1194 	ib_event.device = &srq->rdev->ibdev;
1195 	ib_event.element.srq = &srq->ib_srq;
1196 
1197 	if (srq->ib_srq.event_handler) {
1198 		if (event == NQ_SRQ_EVENT_EVENT_SRQ_THRESHOLD_EVENT)
1199 			ib_event.event = IB_EVENT_SRQ_LIMIT_REACHED;
1200 		(*srq->ib_srq.event_handler)(&ib_event,
1201 					     srq->ib_srq.srq_context);
1202 	}
1203 	return 0;
1204 }
1205 
1206 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
1207 			       struct bnxt_qplib_cq *handle)
1208 {
1209 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
1210 					     qplib_cq);
1211 	u32 *cq_ptr;
1212 
1213 	if (cq->ib_cq.comp_handler) {
1214 		if (cq->uctx_cq_page) {
1215 			cq_ptr = (u32 *)cq->uctx_cq_page;
1216 			*cq_ptr = cq->qplib_cq.toggle;
1217 		}
1218 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
1219 	}
1220 
1221 	return 0;
1222 }
1223 
1224 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
1225 {
1226 	int i;
1227 
1228 	for (i = 1; i < rdev->num_msix; i++)
1229 		bnxt_qplib_disable_nq(&rdev->nq[i - 1]);
1230 
1231 	if (rdev->qplib_res.rcfw)
1232 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
1233 }
1234 
1235 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
1236 {
1237 	int num_vec_enabled = 0;
1238 	int rc = 0, i;
1239 	u32 db_offt;
1240 
1241 	bnxt_qplib_init_res(&rdev->qplib_res);
1242 
1243 	for (i = 1; i < rdev->num_msix ; i++) {
1244 		db_offt = rdev->en_dev->msix_entries[i].db_offset;
1245 		rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
1246 					  i - 1, rdev->en_dev->msix_entries[i].vector,
1247 					  db_offt, &bnxt_re_cqn_handler,
1248 					  &bnxt_re_srqn_handler);
1249 		if (rc) {
1250 			ibdev_err(&rdev->ibdev,
1251 				  "Failed to enable NQ with rc = 0x%x", rc);
1252 			goto fail;
1253 		}
1254 		num_vec_enabled++;
1255 	}
1256 	return 0;
1257 fail:
1258 	for (i = num_vec_enabled; i >= 0; i--)
1259 		bnxt_qplib_disable_nq(&rdev->nq[i]);
1260 	return rc;
1261 }
1262 
1263 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev)
1264 {
1265 	u8 type;
1266 	int i;
1267 
1268 	for (i = 0; i < rdev->num_msix - 1; i++) {
1269 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1270 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
1271 		bnxt_qplib_free_nq(&rdev->nq[i]);
1272 		rdev->nq[i].res = NULL;
1273 	}
1274 }
1275 
1276 static void bnxt_re_free_res(struct bnxt_re_dev *rdev)
1277 {
1278 	bnxt_re_free_nq_res(rdev);
1279 
1280 	if (rdev->qplib_res.dpi_tbl.max) {
1281 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
1282 				       &rdev->dpi_privileged);
1283 	}
1284 	if (rdev->qplib_res.rcfw) {
1285 		bnxt_qplib_free_res(&rdev->qplib_res);
1286 		rdev->qplib_res.rcfw = NULL;
1287 	}
1288 }
1289 
1290 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
1291 {
1292 	struct bnxt_re_ring_attr rattr = {};
1293 	int num_vec_created = 0;
1294 	int rc, i;
1295 	u8 type;
1296 
1297 	/* Configure and allocate resources for qplib */
1298 	rdev->qplib_res.rcfw = &rdev->rcfw;
1299 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr);
1300 	if (rc)
1301 		goto fail;
1302 
1303 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
1304 				  rdev->netdev, &rdev->dev_attr);
1305 	if (rc)
1306 		goto fail;
1307 
1308 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res,
1309 				  &rdev->dpi_privileged,
1310 				  rdev, BNXT_QPLIB_DPI_TYPE_KERNEL);
1311 	if (rc)
1312 		goto dealloc_res;
1313 
1314 	for (i = 0; i < rdev->num_msix - 1; i++) {
1315 		struct bnxt_qplib_nq *nq;
1316 
1317 		nq = &rdev->nq[i];
1318 		nq->hwq.max_elements = BNXT_QPLIB_NQE_MAX_CNT;
1319 		rc = bnxt_qplib_alloc_nq(&rdev->qplib_res, &rdev->nq[i]);
1320 		if (rc) {
1321 			ibdev_err(&rdev->ibdev, "Alloc Failed NQ%d rc:%#x",
1322 				  i, rc);
1323 			goto free_nq;
1324 		}
1325 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1326 		rattr.dma_arr = nq->hwq.pbl[PBL_LVL_0].pg_map_arr;
1327 		rattr.pages = nq->hwq.pbl[rdev->nq[i].hwq.level].pg_count;
1328 		rattr.type = type;
1329 		rattr.mode = RING_ALLOC_REQ_INT_MODE_MSIX;
1330 		rattr.depth = BNXT_QPLIB_NQE_MAX_CNT - 1;
1331 		rattr.lrid = rdev->en_dev->msix_entries[i + 1].ring_idx;
1332 		rc = bnxt_re_net_ring_alloc(rdev, &rattr, &nq->ring_id);
1333 		if (rc) {
1334 			ibdev_err(&rdev->ibdev,
1335 				  "Failed to allocate NQ fw id with rc = 0x%x",
1336 				  rc);
1337 			bnxt_qplib_free_nq(&rdev->nq[i]);
1338 			goto free_nq;
1339 		}
1340 		num_vec_created++;
1341 	}
1342 	return 0;
1343 free_nq:
1344 	for (i = num_vec_created - 1; i >= 0; i--) {
1345 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1346 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
1347 		bnxt_qplib_free_nq(&rdev->nq[i]);
1348 	}
1349 	bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
1350 			       &rdev->dpi_privileged);
1351 dealloc_res:
1352 	bnxt_qplib_free_res(&rdev->qplib_res);
1353 
1354 fail:
1355 	rdev->qplib_res.rcfw = NULL;
1356 	return rc;
1357 }
1358 
1359 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
1360 				   u8 port_num, enum ib_event_type event)
1361 {
1362 	struct ib_event ib_event;
1363 
1364 	ib_event.device = ibdev;
1365 	if (qp) {
1366 		ib_event.element.qp = qp;
1367 		ib_event.event = event;
1368 		if (qp->event_handler)
1369 			qp->event_handler(&ib_event, qp->qp_context);
1370 
1371 	} else {
1372 		ib_event.element.port_num = port_num;
1373 		ib_event.event = event;
1374 		ib_dispatch_event(&ib_event);
1375 	}
1376 }
1377 
1378 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
1379 					struct bnxt_re_qp *qp)
1380 {
1381 	return (qp->ib_qp.qp_type == IB_QPT_GSI) ||
1382 	       (qp == rdev->gsi_ctx.gsi_sqp);
1383 }
1384 
1385 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
1386 {
1387 	int mask = IB_QP_STATE;
1388 	struct ib_qp_attr qp_attr;
1389 	struct bnxt_re_qp *qp;
1390 
1391 	qp_attr.qp_state = IB_QPS_ERR;
1392 	mutex_lock(&rdev->qp_lock);
1393 	list_for_each_entry(qp, &rdev->qp_list, list) {
1394 		/* Modify the state of all QPs except QP1/Shadow QP */
1395 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
1396 			if (qp->qplib_qp.state !=
1397 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
1398 			    qp->qplib_qp.state !=
1399 			    CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1400 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
1401 						       1, IB_EVENT_QP_FATAL);
1402 				bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
1403 						  NULL);
1404 			}
1405 		}
1406 	}
1407 	mutex_unlock(&rdev->qp_lock);
1408 }
1409 
1410 static int bnxt_re_update_gid(struct bnxt_re_dev *rdev)
1411 {
1412 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
1413 	struct bnxt_qplib_gid gid;
1414 	u16 gid_idx, index;
1415 	int rc = 0;
1416 
1417 	if (!ib_device_try_get(&rdev->ibdev))
1418 		return 0;
1419 
1420 	for (index = 0; index < sgid_tbl->active; index++) {
1421 		gid_idx = sgid_tbl->hw_id[index];
1422 
1423 		if (!memcmp(&sgid_tbl->tbl[index], &bnxt_qplib_gid_zero,
1424 			    sizeof(bnxt_qplib_gid_zero)))
1425 			continue;
1426 		/* need to modify the VLAN enable setting of non VLAN GID only
1427 		 * as setting is done for VLAN GID while adding GID
1428 		 */
1429 		if (sgid_tbl->vlan[index])
1430 			continue;
1431 
1432 		memcpy(&gid, &sgid_tbl->tbl[index], sizeof(gid));
1433 
1434 		rc = bnxt_qplib_update_sgid(sgid_tbl, &gid, gid_idx,
1435 					    rdev->qplib_res.netdev->dev_addr);
1436 	}
1437 
1438 	ib_device_put(&rdev->ibdev);
1439 	return rc;
1440 }
1441 
1442 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
1443 {
1444 	u32 prio_map = 0, tmp_map = 0;
1445 	struct net_device *netdev;
1446 	struct dcb_app app = {};
1447 
1448 	netdev = rdev->netdev;
1449 
1450 	app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
1451 	app.protocol = ETH_P_IBOE;
1452 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1453 	prio_map = tmp_map;
1454 
1455 	app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
1456 	app.protocol = ROCE_V2_UDP_DPORT;
1457 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1458 	prio_map |= tmp_map;
1459 
1460 	return prio_map;
1461 }
1462 
1463 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
1464 {
1465 	u8 prio_map = 0;
1466 
1467 	/* Get priority for roce */
1468 	prio_map = bnxt_re_get_priority_mask(rdev);
1469 
1470 	if (prio_map == rdev->cur_prio_map)
1471 		return 0;
1472 	rdev->cur_prio_map = prio_map;
1473 	/* Actual priorities are not programmed as they are already
1474 	 * done by L2 driver; just enable or disable priority vlan tagging
1475 	 */
1476 	if ((prio_map == 0 && rdev->qplib_res.prio) ||
1477 	    (prio_map != 0 && !rdev->qplib_res.prio)) {
1478 		rdev->qplib_res.prio = prio_map;
1479 		bnxt_re_update_gid(rdev);
1480 	}
1481 
1482 	return 0;
1483 }
1484 
1485 static void bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev)
1486 {
1487 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1488 	struct hwrm_ver_get_output resp = {};
1489 	struct hwrm_ver_get_input req = {};
1490 	struct bnxt_qplib_chip_ctx *cctx;
1491 	struct bnxt_fw_msg fw_msg = {};
1492 	int rc;
1493 
1494 	bnxt_re_init_hwrm_hdr((void *)&req, HWRM_VER_GET);
1495 	req.hwrm_intf_maj = HWRM_VERSION_MAJOR;
1496 	req.hwrm_intf_min = HWRM_VERSION_MINOR;
1497 	req.hwrm_intf_upd = HWRM_VERSION_UPDATE;
1498 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1499 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1500 	rc = bnxt_send_msg(en_dev, &fw_msg);
1501 	if (rc) {
1502 		ibdev_err(&rdev->ibdev, "Failed to query HW version, rc = 0x%x",
1503 			  rc);
1504 		return;
1505 	}
1506 
1507 	cctx = rdev->chip_ctx;
1508 	cctx->hwrm_intf_ver =
1509 		(u64)le16_to_cpu(resp.hwrm_intf_major) << 48 |
1510 		(u64)le16_to_cpu(resp.hwrm_intf_minor) << 32 |
1511 		(u64)le16_to_cpu(resp.hwrm_intf_build) << 16 |
1512 		le16_to_cpu(resp.hwrm_intf_patch);
1513 
1514 	cctx->hwrm_cmd_max_timeout = le16_to_cpu(resp.max_req_timeout);
1515 
1516 	if (!cctx->hwrm_cmd_max_timeout)
1517 		cctx->hwrm_cmd_max_timeout = RCFW_FW_STALL_MAX_TIMEOUT;
1518 }
1519 
1520 static int bnxt_re_ib_init(struct bnxt_re_dev *rdev)
1521 {
1522 	int rc;
1523 	u32 event;
1524 
1525 	/* Register ib dev */
1526 	rc = bnxt_re_register_ib(rdev);
1527 	if (rc) {
1528 		pr_err("Failed to register with IB: %#x\n", rc);
1529 		return rc;
1530 	}
1531 	dev_info(rdev_to_dev(rdev), "Device registered with IB successfully");
1532 	set_bit(BNXT_RE_FLAG_ISSUE_ROCE_STATS, &rdev->flags);
1533 
1534 	event = netif_running(rdev->netdev) && netif_carrier_ok(rdev->netdev) ?
1535 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
1536 
1537 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, event);
1538 
1539 	return rc;
1540 }
1541 
1542 static void bnxt_re_dev_uninit(struct bnxt_re_dev *rdev)
1543 {
1544 	u8 type;
1545 	int rc;
1546 
1547 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
1548 		cancel_delayed_work_sync(&rdev->worker);
1549 
1550 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED,
1551 			       &rdev->flags))
1552 		bnxt_re_cleanup_res(rdev);
1553 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags))
1554 		bnxt_re_free_res(rdev);
1555 
1556 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
1557 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
1558 		if (rc)
1559 			ibdev_warn(&rdev->ibdev,
1560 				   "Failed to deinitialize RCFW: %#x", rc);
1561 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1562 		bnxt_qplib_free_ctx(&rdev->qplib_res, &rdev->qplib_ctx);
1563 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1564 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1565 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id, type);
1566 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1567 	}
1568 
1569 	rdev->num_msix = 0;
1570 
1571 	if (rdev->pacing.dbr_pacing)
1572 		bnxt_re_deinitialize_dbr_pacing(rdev);
1573 
1574 	bnxt_re_destroy_chip_ctx(rdev);
1575 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags))
1576 		bnxt_unregister_dev(rdev->en_dev);
1577 }
1578 
1579 /* worker thread for polling periodic events. Now used for QoS programming*/
1580 static void bnxt_re_worker(struct work_struct *work)
1581 {
1582 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
1583 						worker.work);
1584 
1585 	bnxt_re_setup_qos(rdev);
1586 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1587 }
1588 
1589 static int bnxt_re_dev_init(struct bnxt_re_dev *rdev, u8 wqe_mode)
1590 {
1591 	struct bnxt_re_ring_attr rattr = {};
1592 	struct bnxt_qplib_creq_ctx *creq;
1593 	u32 db_offt;
1594 	int vid;
1595 	u8 type;
1596 	int rc;
1597 
1598 	/* Registered a new RoCE device instance to netdev */
1599 	rc = bnxt_re_register_netdev(rdev);
1600 	if (rc) {
1601 		ibdev_err(&rdev->ibdev,
1602 			  "Failed to register with netedev: %#x\n", rc);
1603 		return -EINVAL;
1604 	}
1605 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1606 
1607 	rc = bnxt_re_setup_chip_ctx(rdev, wqe_mode);
1608 	if (rc) {
1609 		bnxt_unregister_dev(rdev->en_dev);
1610 		clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1611 		ibdev_err(&rdev->ibdev, "Failed to get chip context\n");
1612 		return -EINVAL;
1613 	}
1614 
1615 	/* Check whether VF or PF */
1616 	bnxt_re_get_sriov_func_type(rdev);
1617 
1618 	if (!rdev->en_dev->ulp_tbl->msix_requested) {
1619 		ibdev_err(&rdev->ibdev,
1620 			  "Failed to get MSI-X vectors: %#x\n", rc);
1621 		rc = -EINVAL;
1622 		goto fail;
1623 	}
1624 	ibdev_dbg(&rdev->ibdev, "Got %d MSI-X vectors\n",
1625 		  rdev->en_dev->ulp_tbl->msix_requested);
1626 	rdev->num_msix = rdev->en_dev->ulp_tbl->msix_requested;
1627 
1628 	bnxt_re_query_hwrm_intf_version(rdev);
1629 
1630 	/* Establish RCFW Communication Channel to initialize the context
1631 	 * memory for the function and all child VFs
1632 	 */
1633 	rc = bnxt_qplib_alloc_rcfw_channel(&rdev->qplib_res, &rdev->rcfw,
1634 					   &rdev->qplib_ctx,
1635 					   BNXT_RE_MAX_QPC_COUNT);
1636 	if (rc) {
1637 		ibdev_err(&rdev->ibdev,
1638 			  "Failed to allocate RCFW Channel: %#x\n", rc);
1639 		goto fail;
1640 	}
1641 
1642 	type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1643 	creq = &rdev->rcfw.creq;
1644 	rattr.dma_arr = creq->hwq.pbl[PBL_LVL_0].pg_map_arr;
1645 	rattr.pages = creq->hwq.pbl[creq->hwq.level].pg_count;
1646 	rattr.type = type;
1647 	rattr.mode = RING_ALLOC_REQ_INT_MODE_MSIX;
1648 	rattr.depth = BNXT_QPLIB_CREQE_MAX_CNT - 1;
1649 	rattr.lrid = rdev->en_dev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx;
1650 	rc = bnxt_re_net_ring_alloc(rdev, &rattr, &creq->ring_id);
1651 	if (rc) {
1652 		ibdev_err(&rdev->ibdev, "Failed to allocate CREQ: %#x\n", rc);
1653 		goto free_rcfw;
1654 	}
1655 	db_offt = rdev->en_dev->msix_entries[BNXT_RE_AEQ_IDX].db_offset;
1656 	vid = rdev->en_dev->msix_entries[BNXT_RE_AEQ_IDX].vector;
1657 	rc = bnxt_qplib_enable_rcfw_channel(&rdev->rcfw,
1658 					    vid, db_offt,
1659 					    &bnxt_re_aeq_handler);
1660 	if (rc) {
1661 		ibdev_err(&rdev->ibdev, "Failed to enable RCFW channel: %#x\n",
1662 			  rc);
1663 		goto free_ring;
1664 	}
1665 
1666 	if (bnxt_qplib_dbr_pacing_en(rdev->chip_ctx)) {
1667 		rc = bnxt_re_initialize_dbr_pacing(rdev);
1668 		if (!rc) {
1669 			rdev->pacing.dbr_pacing = true;
1670 		} else {
1671 			ibdev_err(&rdev->ibdev,
1672 				  "DBR pacing disabled with error : %d\n", rc);
1673 			rdev->pacing.dbr_pacing = false;
1674 		}
1675 	}
1676 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr);
1677 	if (rc)
1678 		goto disable_rcfw;
1679 
1680 	bnxt_re_set_resource_limits(rdev);
1681 
1682 	rc = bnxt_qplib_alloc_ctx(&rdev->qplib_res, &rdev->qplib_ctx, 0,
1683 				  bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx));
1684 	if (rc) {
1685 		ibdev_err(&rdev->ibdev,
1686 			  "Failed to allocate QPLIB context: %#x\n", rc);
1687 		goto disable_rcfw;
1688 	}
1689 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
1690 					 rdev->qplib_ctx.stats.dma_map,
1691 					 &rdev->qplib_ctx.stats.fw_id);
1692 	if (rc) {
1693 		ibdev_err(&rdev->ibdev,
1694 			  "Failed to allocate stats context: %#x\n", rc);
1695 		goto free_ctx;
1696 	}
1697 
1698 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx,
1699 				  rdev->is_virtfn);
1700 	if (rc) {
1701 		ibdev_err(&rdev->ibdev,
1702 			  "Failed to initialize RCFW: %#x\n", rc);
1703 		goto free_sctx;
1704 	}
1705 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1706 
1707 	/* Resources based on the 'new' device caps */
1708 	rc = bnxt_re_alloc_res(rdev);
1709 	if (rc) {
1710 		ibdev_err(&rdev->ibdev,
1711 			  "Failed to allocate resources: %#x\n", rc);
1712 		goto fail;
1713 	}
1714 	set_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags);
1715 	rc = bnxt_re_init_res(rdev);
1716 	if (rc) {
1717 		ibdev_err(&rdev->ibdev,
1718 			  "Failed to initialize resources: %#x\n", rc);
1719 		goto fail;
1720 	}
1721 
1722 	set_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED, &rdev->flags);
1723 
1724 	if (!rdev->is_virtfn) {
1725 		rc = bnxt_re_setup_qos(rdev);
1726 		if (rc)
1727 			ibdev_info(&rdev->ibdev,
1728 				   "RoCE priority not yet configured\n");
1729 
1730 		INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1731 		set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1732 		schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1733 		/*
1734 		 * Use the total VF count since the actual VF count may not be
1735 		 * available at this point.
1736 		 */
1737 		bnxt_re_vf_res_config(rdev);
1738 	}
1739 	hash_init(rdev->cq_hash);
1740 
1741 	return 0;
1742 free_sctx:
1743 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1744 free_ctx:
1745 	bnxt_qplib_free_ctx(&rdev->qplib_res, &rdev->qplib_ctx);
1746 disable_rcfw:
1747 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1748 free_ring:
1749 	type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1750 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id, type);
1751 free_rcfw:
1752 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1753 fail:
1754 	bnxt_re_dev_uninit(rdev);
1755 
1756 	return rc;
1757 }
1758 
1759 static int bnxt_re_add_device(struct auxiliary_device *adev, u8 wqe_mode)
1760 {
1761 	struct bnxt_aux_priv *aux_priv =
1762 		container_of(adev, struct bnxt_aux_priv, aux_dev);
1763 	struct bnxt_en_dev *en_dev;
1764 	struct bnxt_re_dev *rdev;
1765 	int rc;
1766 
1767 	/* en_dev should never be NULL as long as adev and aux_dev are valid. */
1768 	en_dev = aux_priv->edev;
1769 
1770 	rdev = bnxt_re_dev_add(aux_priv, en_dev);
1771 	if (!rdev || !rdev_to_dev(rdev)) {
1772 		rc = -ENOMEM;
1773 		goto exit;
1774 	}
1775 
1776 	rc = bnxt_re_dev_init(rdev, wqe_mode);
1777 	if (rc)
1778 		goto re_dev_dealloc;
1779 
1780 	rc = bnxt_re_ib_init(rdev);
1781 	if (rc) {
1782 		pr_err("Failed to register with IB: %s",
1783 			aux_priv->aux_dev.name);
1784 		goto re_dev_uninit;
1785 	}
1786 	auxiliary_set_drvdata(adev, rdev);
1787 
1788 	return 0;
1789 
1790 re_dev_uninit:
1791 	bnxt_re_dev_uninit(rdev);
1792 re_dev_dealloc:
1793 	ib_dealloc_device(&rdev->ibdev);
1794 exit:
1795 	return rc;
1796 }
1797 
1798 static void bnxt_re_setup_cc(struct bnxt_re_dev *rdev, bool enable)
1799 {
1800 	struct bnxt_qplib_cc_param cc_param = {};
1801 
1802 	/* Do not enable congestion control on VFs */
1803 	if (rdev->is_virtfn)
1804 		return;
1805 
1806 	/* Currently enabling only for GenP5 adapters */
1807 	if (!bnxt_qplib_is_chip_gen_p5_p7(rdev->chip_ctx))
1808 		return;
1809 
1810 	if (enable) {
1811 		cc_param.enable  = 1;
1812 		cc_param.cc_mode = CMDQ_MODIFY_ROCE_CC_CC_MODE_PROBABILISTIC_CC_MODE;
1813 	}
1814 
1815 	cc_param.mask = (CMDQ_MODIFY_ROCE_CC_MODIFY_MASK_CC_MODE |
1816 			 CMDQ_MODIFY_ROCE_CC_MODIFY_MASK_ENABLE_CC |
1817 			 CMDQ_MODIFY_ROCE_CC_MODIFY_MASK_TOS_ECN);
1818 
1819 	if (bnxt_qplib_modify_cc(&rdev->qplib_res, &cc_param))
1820 		ibdev_err(&rdev->ibdev, "Failed to setup CC enable = %d\n", enable);
1821 }
1822 
1823 /*
1824  * "Notifier chain callback can be invoked for the same chain from
1825  * different CPUs at the same time".
1826  *
1827  * For cases when the netdev is already present, our call to the
1828  * register_netdevice_notifier() will actually get the rtnl_lock()
1829  * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1830  * events.
1831  *
1832  * But for cases when the netdev is not already present, the notifier
1833  * chain is subjected to be invoked from different CPUs simultaneously.
1834  *
1835  * This is protected by the netdev_mutex.
1836  */
1837 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1838 				unsigned long event, void *ptr)
1839 {
1840 	struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1841 	struct bnxt_re_dev *rdev;
1842 
1843 	real_dev = rdma_vlan_dev_real_dev(netdev);
1844 	if (!real_dev)
1845 		real_dev = netdev;
1846 
1847 	if (real_dev != netdev)
1848 		goto exit;
1849 
1850 	rdev = bnxt_re_from_netdev(real_dev);
1851 	if (!rdev)
1852 		return NOTIFY_DONE;
1853 
1854 
1855 	switch (event) {
1856 	case NETDEV_UP:
1857 	case NETDEV_DOWN:
1858 	case NETDEV_CHANGE:
1859 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1860 					netif_carrier_ok(real_dev) ?
1861 					IB_EVENT_PORT_ACTIVE :
1862 					IB_EVENT_PORT_ERR);
1863 		break;
1864 	default:
1865 		break;
1866 	}
1867 	ib_device_put(&rdev->ibdev);
1868 exit:
1869 	return NOTIFY_DONE;
1870 }
1871 
1872 #define BNXT_ADEV_NAME "bnxt_en"
1873 
1874 static void bnxt_re_remove(struct auxiliary_device *adev)
1875 {
1876 	struct bnxt_re_dev *rdev = auxiliary_get_drvdata(adev);
1877 
1878 	if (!rdev)
1879 		return;
1880 
1881 	mutex_lock(&bnxt_re_mutex);
1882 	if (rdev->nb.notifier_call) {
1883 		unregister_netdevice_notifier(&rdev->nb);
1884 		rdev->nb.notifier_call = NULL;
1885 	} else {
1886 		/* If notifier is null, we should have already done a
1887 		 * clean up before coming here.
1888 		 */
1889 		goto skip_remove;
1890 	}
1891 	bnxt_re_setup_cc(rdev, false);
1892 	ib_unregister_device(&rdev->ibdev);
1893 	bnxt_re_dev_uninit(rdev);
1894 	ib_dealloc_device(&rdev->ibdev);
1895 skip_remove:
1896 	mutex_unlock(&bnxt_re_mutex);
1897 }
1898 
1899 static int bnxt_re_probe(struct auxiliary_device *adev,
1900 			 const struct auxiliary_device_id *id)
1901 {
1902 	struct bnxt_re_dev *rdev;
1903 	int rc;
1904 
1905 	mutex_lock(&bnxt_re_mutex);
1906 	rc = bnxt_re_add_device(adev, BNXT_QPLIB_WQE_MODE_STATIC);
1907 	if (rc) {
1908 		mutex_unlock(&bnxt_re_mutex);
1909 		return rc;
1910 	}
1911 
1912 	rdev = auxiliary_get_drvdata(adev);
1913 
1914 	rdev->nb.notifier_call = bnxt_re_netdev_event;
1915 	rc = register_netdevice_notifier(&rdev->nb);
1916 	if (rc) {
1917 		rdev->nb.notifier_call = NULL;
1918 		pr_err("%s: Cannot register to netdevice_notifier",
1919 		       ROCE_DRV_MODULE_NAME);
1920 		goto err;
1921 	}
1922 
1923 	bnxt_re_setup_cc(rdev, true);
1924 	mutex_unlock(&bnxt_re_mutex);
1925 	return 0;
1926 
1927 err:
1928 	mutex_unlock(&bnxt_re_mutex);
1929 	bnxt_re_remove(adev);
1930 
1931 	return rc;
1932 }
1933 
1934 static int bnxt_re_suspend(struct auxiliary_device *adev, pm_message_t state)
1935 {
1936 	struct bnxt_re_dev *rdev = auxiliary_get_drvdata(adev);
1937 
1938 	if (!rdev)
1939 		return 0;
1940 
1941 	mutex_lock(&bnxt_re_mutex);
1942 	/* L2 driver may invoke this callback during device error/crash or device
1943 	 * reset. Current RoCE driver doesn't recover the device in case of
1944 	 * error. Handle the error by dispatching fatal events to all qps
1945 	 * ie. by calling bnxt_re_dev_stop and release the MSIx vectors as
1946 	 * L2 driver want to modify the MSIx table.
1947 	 */
1948 
1949 	ibdev_info(&rdev->ibdev, "Handle device suspend call");
1950 	/* Check the current device state from bnxt_en_dev and move the
1951 	 * device to detached state if FW_FATAL_COND is set.
1952 	 * This prevents more commands to HW during clean-up,
1953 	 * in case the device is already in error.
1954 	 */
1955 	if (test_bit(BNXT_STATE_FW_FATAL_COND, &rdev->en_dev->en_state))
1956 		set_bit(ERR_DEVICE_DETACHED, &rdev->rcfw.cmdq.flags);
1957 
1958 	bnxt_re_dev_stop(rdev);
1959 	bnxt_re_stop_irq(rdev);
1960 	/* Move the device states to detached and  avoid sending any more
1961 	 * commands to HW
1962 	 */
1963 	set_bit(BNXT_RE_FLAG_ERR_DEVICE_DETACHED, &rdev->flags);
1964 	set_bit(ERR_DEVICE_DETACHED, &rdev->rcfw.cmdq.flags);
1965 	wake_up_all(&rdev->rcfw.cmdq.waitq);
1966 	mutex_unlock(&bnxt_re_mutex);
1967 
1968 	return 0;
1969 }
1970 
1971 static int bnxt_re_resume(struct auxiliary_device *adev)
1972 {
1973 	struct bnxt_re_dev *rdev = auxiliary_get_drvdata(adev);
1974 
1975 	if (!rdev)
1976 		return 0;
1977 
1978 	mutex_lock(&bnxt_re_mutex);
1979 	/* L2 driver may invoke this callback during device recovery, resume.
1980 	 * reset. Current RoCE driver doesn't recover the device in case of
1981 	 * error. Handle the error by dispatching fatal events to all qps
1982 	 * ie. by calling bnxt_re_dev_stop and release the MSIx vectors as
1983 	 * L2 driver want to modify the MSIx table.
1984 	 */
1985 
1986 	ibdev_info(&rdev->ibdev, "Handle device resume call");
1987 	mutex_unlock(&bnxt_re_mutex);
1988 
1989 	return 0;
1990 }
1991 
1992 static const struct auxiliary_device_id bnxt_re_id_table[] = {
1993 	{ .name = BNXT_ADEV_NAME ".rdma", },
1994 	{},
1995 };
1996 
1997 MODULE_DEVICE_TABLE(auxiliary, bnxt_re_id_table);
1998 
1999 static struct auxiliary_driver bnxt_re_driver = {
2000 	.name = "rdma",
2001 	.probe = bnxt_re_probe,
2002 	.remove = bnxt_re_remove,
2003 	.shutdown = bnxt_re_shutdown,
2004 	.suspend = bnxt_re_suspend,
2005 	.resume = bnxt_re_resume,
2006 	.id_table = bnxt_re_id_table,
2007 };
2008 
2009 static int __init bnxt_re_mod_init(void)
2010 {
2011 	int rc;
2012 
2013 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
2014 	rc = auxiliary_driver_register(&bnxt_re_driver);
2015 	if (rc) {
2016 		pr_err("%s: Failed to register auxiliary driver\n",
2017 			ROCE_DRV_MODULE_NAME);
2018 		return rc;
2019 	}
2020 	return 0;
2021 }
2022 
2023 static void __exit bnxt_re_mod_exit(void)
2024 {
2025 	auxiliary_driver_unregister(&bnxt_re_driver);
2026 }
2027 
2028 module_init(bnxt_re_mod_init);
2029 module_exit(bnxt_re_mod_exit);
2030