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 
52 #include <rdma/ib_verbs.h>
53 #include <rdma/ib_user_verbs.h>
54 #include <rdma/ib_umem.h>
55 #include <rdma/ib_addr.h>
56 
57 #include "bnxt_ulp.h"
58 #include "roce_hsi.h"
59 #include "qplib_res.h"
60 #include "qplib_sp.h"
61 #include "qplib_fp.h"
62 #include "qplib_rcfw.h"
63 #include "bnxt_re.h"
64 #include "ib_verbs.h"
65 #include <rdma/bnxt_re-abi.h>
66 #include "bnxt.h"
67 #include "hw_counters.h"
68 
69 static char version[] =
70 		BNXT_RE_DESC "\n";
71 
72 MODULE_AUTHOR("Eddie Wai <eddie.wai@broadcom.com>");
73 MODULE_DESCRIPTION(BNXT_RE_DESC " Driver");
74 MODULE_LICENSE("Dual BSD/GPL");
75 
76 /* globals */
77 static struct list_head bnxt_re_dev_list = LIST_HEAD_INIT(bnxt_re_dev_list);
78 /* Mutex to protect the list of bnxt_re devices added */
79 static DEFINE_MUTEX(bnxt_re_dev_lock);
80 static struct workqueue_struct *bnxt_re_wq;
81 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev);
82 
83 static void bnxt_re_destroy_chip_ctx(struct bnxt_re_dev *rdev)
84 {
85 	rdev->rcfw.res = NULL;
86 	rdev->qplib_res.cctx = NULL;
87 }
88 
89 static int bnxt_re_setup_chip_ctx(struct bnxt_re_dev *rdev)
90 {
91 	struct bnxt_en_dev *en_dev;
92 	struct bnxt *bp;
93 
94 	en_dev = rdev->en_dev;
95 	bp = netdev_priv(en_dev->net);
96 
97 	rdev->chip_ctx.chip_num = bp->chip_num;
98 	/* rest members to follow eventually */
99 
100 	rdev->qplib_res.cctx = &rdev->chip_ctx;
101 	rdev->rcfw.res = &rdev->qplib_res;
102 
103 	return 0;
104 }
105 
106 /* SR-IOV helper functions */
107 
108 static void bnxt_re_get_sriov_func_type(struct bnxt_re_dev *rdev)
109 {
110 	struct bnxt *bp;
111 
112 	bp = netdev_priv(rdev->en_dev->net);
113 	if (BNXT_VF(bp))
114 		rdev->is_virtfn = 1;
115 }
116 
117 /* Set the maximum number of each resource that the driver actually wants
118  * to allocate. This may be up to the maximum number the firmware has
119  * reserved for the function. The driver may choose to allocate fewer
120  * resources than the firmware maximum.
121  */
122 static void bnxt_re_limit_pf_res(struct bnxt_re_dev *rdev)
123 {
124 	struct bnxt_qplib_dev_attr *attr;
125 	struct bnxt_qplib_ctx *ctx;
126 	int i;
127 
128 	attr = &rdev->dev_attr;
129 	ctx = &rdev->qplib_ctx;
130 
131 	ctx->qpc_count = min_t(u32, BNXT_RE_MAX_QPC_COUNT,
132 			       attr->max_qp);
133 	ctx->mrw_count = BNXT_RE_MAX_MRW_COUNT_256K;
134 	/* Use max_mr from fw since max_mrw does not get set */
135 	ctx->mrw_count = min_t(u32, ctx->mrw_count, attr->max_mr);
136 	ctx->srqc_count = min_t(u32, BNXT_RE_MAX_SRQC_COUNT,
137 				attr->max_srq);
138 	ctx->cq_count = min_t(u32, BNXT_RE_MAX_CQ_COUNT, attr->max_cq);
139 	if (!bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx))
140 		for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
141 			rdev->qplib_ctx.tqm_count[i] =
142 			rdev->dev_attr.tqm_alloc_reqs[i];
143 }
144 
145 static void bnxt_re_limit_vf_res(struct bnxt_qplib_ctx *qplib_ctx, u32 num_vf)
146 {
147 	struct bnxt_qplib_vf_res *vf_res;
148 	u32 mrws = 0;
149 	u32 vf_pct;
150 	u32 nvfs;
151 
152 	vf_res = &qplib_ctx->vf_res;
153 	/*
154 	 * Reserve a set of resources for the PF. Divide the remaining
155 	 * resources among the VFs
156 	 */
157 	vf_pct = 100 - BNXT_RE_PCT_RSVD_FOR_PF;
158 	nvfs = num_vf;
159 	num_vf = 100 * num_vf;
160 	vf_res->max_qp_per_vf = (qplib_ctx->qpc_count * vf_pct) / num_vf;
161 	vf_res->max_srq_per_vf = (qplib_ctx->srqc_count * vf_pct) / num_vf;
162 	vf_res->max_cq_per_vf = (qplib_ctx->cq_count * vf_pct) / num_vf;
163 	/*
164 	 * The driver allows many more MRs than other resources. If the
165 	 * firmware does also, then reserve a fixed amount for the PF and
166 	 * divide the rest among VFs. VFs may use many MRs for NFS
167 	 * mounts, ISER, NVME applications, etc. If the firmware severely
168 	 * restricts the number of MRs, then let PF have half and divide
169 	 * the rest among VFs, as for the other resource types.
170 	 */
171 	if (qplib_ctx->mrw_count < BNXT_RE_MAX_MRW_COUNT_64K) {
172 		mrws = qplib_ctx->mrw_count * vf_pct;
173 		nvfs = num_vf;
174 	} else {
175 		mrws = qplib_ctx->mrw_count - BNXT_RE_RESVD_MR_FOR_PF;
176 	}
177 	vf_res->max_mrw_per_vf = (mrws / nvfs);
178 	vf_res->max_gid_per_vf = BNXT_RE_MAX_GID_PER_VF;
179 }
180 
181 static void bnxt_re_set_resource_limits(struct bnxt_re_dev *rdev)
182 {
183 	u32 num_vfs;
184 
185 	memset(&rdev->qplib_ctx.vf_res, 0, sizeof(struct bnxt_qplib_vf_res));
186 	bnxt_re_limit_pf_res(rdev);
187 
188 	num_vfs =  bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx) ?
189 			BNXT_RE_GEN_P5_MAX_VF : rdev->num_vfs;
190 	if (num_vfs)
191 		bnxt_re_limit_vf_res(&rdev->qplib_ctx, num_vfs);
192 }
193 
194 /* for handling bnxt_en callbacks later */
195 static void bnxt_re_stop(void *p)
196 {
197 }
198 
199 static void bnxt_re_start(void *p)
200 {
201 }
202 
203 static void bnxt_re_sriov_config(void *p, int num_vfs)
204 {
205 	struct bnxt_re_dev *rdev = p;
206 
207 	if (!rdev)
208 		return;
209 
210 	rdev->num_vfs = num_vfs;
211 	if (!bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx)) {
212 		bnxt_re_set_resource_limits(rdev);
213 		bnxt_qplib_set_func_resources(&rdev->qplib_res, &rdev->rcfw,
214 					      &rdev->qplib_ctx);
215 	}
216 }
217 
218 static void bnxt_re_shutdown(void *p)
219 {
220 	struct bnxt_re_dev *rdev = p;
221 
222 	if (!rdev)
223 		return;
224 
225 	bnxt_re_ib_unreg(rdev);
226 }
227 
228 static void bnxt_re_stop_irq(void *handle)
229 {
230 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
231 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
232 	struct bnxt_qplib_nq *nq;
233 	int indx;
234 
235 	for (indx = BNXT_RE_NQ_IDX; indx < rdev->num_msix; indx++) {
236 		nq = &rdev->nq[indx - 1];
237 		bnxt_qplib_nq_stop_irq(nq, false);
238 	}
239 
240 	bnxt_qplib_rcfw_stop_irq(rcfw, false);
241 }
242 
243 static void bnxt_re_start_irq(void *handle, struct bnxt_msix_entry *ent)
244 {
245 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
246 	struct bnxt_msix_entry *msix_ent = rdev->msix_entries;
247 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
248 	struct bnxt_qplib_nq *nq;
249 	int indx, rc;
250 
251 	if (!ent) {
252 		/* Not setting the f/w timeout bit in rcfw.
253 		 * During the driver unload the first command
254 		 * to f/w will timeout and that will set the
255 		 * timeout bit.
256 		 */
257 		dev_err(rdev_to_dev(rdev), "Failed to re-start IRQs\n");
258 		return;
259 	}
260 
261 	/* Vectors may change after restart, so update with new vectors
262 	 * in device sctructure.
263 	 */
264 	for (indx = 0; indx < rdev->num_msix; indx++)
265 		rdev->msix_entries[indx].vector = ent[indx].vector;
266 
267 	bnxt_qplib_rcfw_start_irq(rcfw, msix_ent[BNXT_RE_AEQ_IDX].vector,
268 				  false);
269 	for (indx = BNXT_RE_NQ_IDX ; indx < rdev->num_msix; indx++) {
270 		nq = &rdev->nq[indx - 1];
271 		rc = bnxt_qplib_nq_start_irq(nq, indx - 1,
272 					     msix_ent[indx].vector, false);
273 		if (rc)
274 			dev_warn(rdev_to_dev(rdev),
275 				 "Failed to reinit NQ index %d\n", indx - 1);
276 	}
277 }
278 
279 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
280 	.ulp_async_notifier = NULL,
281 	.ulp_stop = bnxt_re_stop,
282 	.ulp_start = bnxt_re_start,
283 	.ulp_sriov_config = bnxt_re_sriov_config,
284 	.ulp_shutdown = bnxt_re_shutdown,
285 	.ulp_irq_stop = bnxt_re_stop_irq,
286 	.ulp_irq_restart = bnxt_re_start_irq
287 };
288 
289 /* RoCE -> Net driver */
290 
291 /* Driver registration routines used to let the networking driver (bnxt_en)
292  * to know that the RoCE driver is now installed
293  */
294 static int bnxt_re_unregister_netdev(struct bnxt_re_dev *rdev)
295 {
296 	struct bnxt_en_dev *en_dev;
297 	int rc;
298 
299 	if (!rdev)
300 		return -EINVAL;
301 
302 	en_dev = rdev->en_dev;
303 
304 	rc = en_dev->en_ops->bnxt_unregister_device(rdev->en_dev,
305 						    BNXT_ROCE_ULP);
306 	return rc;
307 }
308 
309 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
310 {
311 	struct bnxt_en_dev *en_dev;
312 	int rc = 0;
313 
314 	if (!rdev)
315 		return -EINVAL;
316 
317 	en_dev = rdev->en_dev;
318 
319 	rc = en_dev->en_ops->bnxt_register_device(en_dev, BNXT_ROCE_ULP,
320 						  &bnxt_re_ulp_ops, rdev);
321 	rdev->qplib_res.pdev = rdev->en_dev->pdev;
322 	return rc;
323 }
324 
325 static int bnxt_re_free_msix(struct bnxt_re_dev *rdev)
326 {
327 	struct bnxt_en_dev *en_dev;
328 	int rc;
329 
330 	if (!rdev)
331 		return -EINVAL;
332 
333 	en_dev = rdev->en_dev;
334 
335 
336 	rc = en_dev->en_ops->bnxt_free_msix(rdev->en_dev, BNXT_ROCE_ULP);
337 
338 	return rc;
339 }
340 
341 static int bnxt_re_request_msix(struct bnxt_re_dev *rdev)
342 {
343 	int rc = 0, num_msix_want = BNXT_RE_MAX_MSIX, num_msix_got;
344 	struct bnxt_en_dev *en_dev;
345 
346 	if (!rdev)
347 		return -EINVAL;
348 
349 	en_dev = rdev->en_dev;
350 
351 	num_msix_want = min_t(u32, BNXT_RE_MAX_MSIX, num_online_cpus());
352 
353 	num_msix_got = en_dev->en_ops->bnxt_request_msix(en_dev, BNXT_ROCE_ULP,
354 							 rdev->msix_entries,
355 							 num_msix_want);
356 	if (num_msix_got < BNXT_RE_MIN_MSIX) {
357 		rc = -EINVAL;
358 		goto done;
359 	}
360 	if (num_msix_got != num_msix_want) {
361 		dev_warn(rdev_to_dev(rdev),
362 			 "Requested %d MSI-X vectors, got %d\n",
363 			 num_msix_want, num_msix_got);
364 	}
365 	rdev->num_msix = num_msix_got;
366 done:
367 	return rc;
368 }
369 
370 static void bnxt_re_init_hwrm_hdr(struct bnxt_re_dev *rdev, struct input *hdr,
371 				  u16 opcd, u16 crid, u16 trid)
372 {
373 	hdr->req_type = cpu_to_le16(opcd);
374 	hdr->cmpl_ring = cpu_to_le16(crid);
375 	hdr->target_id = cpu_to_le16(trid);
376 }
377 
378 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
379 				int msg_len, void *resp, int resp_max_len,
380 				int timeout)
381 {
382 	fw_msg->msg = msg;
383 	fw_msg->msg_len = msg_len;
384 	fw_msg->resp = resp;
385 	fw_msg->resp_max_len = resp_max_len;
386 	fw_msg->timeout = timeout;
387 }
388 
389 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev,
390 				 u16 fw_ring_id, int type)
391 {
392 	struct bnxt_en_dev *en_dev = rdev->en_dev;
393 	struct hwrm_ring_free_input req = {0};
394 	struct hwrm_ring_free_output resp;
395 	struct bnxt_fw_msg fw_msg;
396 	int rc = -EINVAL;
397 
398 	if (!en_dev)
399 		return rc;
400 
401 	memset(&fw_msg, 0, sizeof(fw_msg));
402 
403 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1);
404 	req.ring_type = type;
405 	req.ring_id = cpu_to_le16(fw_ring_id);
406 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
407 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
408 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
409 	if (rc)
410 		dev_err(rdev_to_dev(rdev),
411 			"Failed to free HW ring:%d :%#x", req.ring_id, rc);
412 	return rc;
413 }
414 
415 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev, dma_addr_t *dma_arr,
416 				  int pages, int type, u32 ring_mask,
417 				  u32 map_index, u16 *fw_ring_id)
418 {
419 	struct bnxt_en_dev *en_dev = rdev->en_dev;
420 	struct hwrm_ring_alloc_input req = {0};
421 	struct hwrm_ring_alloc_output resp;
422 	struct bnxt_fw_msg fw_msg;
423 	int rc = -EINVAL;
424 
425 	if (!en_dev)
426 		return rc;
427 
428 	memset(&fw_msg, 0, sizeof(fw_msg));
429 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_ALLOC, -1, -1);
430 	req.enables = 0;
431 	req.page_tbl_addr =  cpu_to_le64(dma_arr[0]);
432 	if (pages > 1) {
433 		/* Page size is in log2 units */
434 		req.page_size = BNXT_PAGE_SHIFT;
435 		req.page_tbl_depth = 1;
436 	}
437 	req.fbo = 0;
438 	/* Association of ring index with doorbell index and MSIX number */
439 	req.logical_id = cpu_to_le16(map_index);
440 	req.length = cpu_to_le32(ring_mask + 1);
441 	req.ring_type = type;
442 	req.int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
443 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
444 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
445 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
446 	if (!rc)
447 		*fw_ring_id = le16_to_cpu(resp.ring_id);
448 
449 	return rc;
450 }
451 
452 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
453 				      u32 fw_stats_ctx_id)
454 {
455 	struct bnxt_en_dev *en_dev = rdev->en_dev;
456 	struct hwrm_stat_ctx_free_input req = {0};
457 	struct bnxt_fw_msg fw_msg;
458 	int rc = -EINVAL;
459 
460 	if (!en_dev)
461 		return rc;
462 
463 	memset(&fw_msg, 0, sizeof(fw_msg));
464 
465 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_FREE, -1, -1);
466 	req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
467 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&req,
468 			    sizeof(req), DFLT_HWRM_CMD_TIMEOUT);
469 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
470 	if (rc)
471 		dev_err(rdev_to_dev(rdev),
472 			"Failed to free HW stats context %#x", rc);
473 
474 	return rc;
475 }
476 
477 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
478 				       dma_addr_t dma_map,
479 				       u32 *fw_stats_ctx_id)
480 {
481 	struct hwrm_stat_ctx_alloc_output resp = {0};
482 	struct hwrm_stat_ctx_alloc_input req = {0};
483 	struct bnxt_en_dev *en_dev = rdev->en_dev;
484 	struct bnxt_fw_msg fw_msg;
485 	int rc = -EINVAL;
486 
487 	*fw_stats_ctx_id = INVALID_STATS_CTX_ID;
488 
489 	if (!en_dev)
490 		return rc;
491 
492 	memset(&fw_msg, 0, sizeof(fw_msg));
493 
494 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1);
495 	req.update_period_ms = cpu_to_le32(1000);
496 	req.stats_dma_addr = cpu_to_le64(dma_map);
497 	req.stats_dma_length = cpu_to_le16(sizeof(struct ctx_hw_stats_ext));
498 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
499 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
500 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
501 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
502 	if (!rc)
503 		*fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
504 
505 	return rc;
506 }
507 
508 /* Device */
509 
510 static bool is_bnxt_re_dev(struct net_device *netdev)
511 {
512 	struct ethtool_drvinfo drvinfo;
513 
514 	if (netdev->ethtool_ops && netdev->ethtool_ops->get_drvinfo) {
515 		memset(&drvinfo, 0, sizeof(drvinfo));
516 		netdev->ethtool_ops->get_drvinfo(netdev, &drvinfo);
517 
518 		if (strcmp(drvinfo.driver, "bnxt_en"))
519 			return false;
520 		return true;
521 	}
522 	return false;
523 }
524 
525 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
526 {
527 	struct bnxt_re_dev *rdev;
528 
529 	rcu_read_lock();
530 	list_for_each_entry_rcu(rdev, &bnxt_re_dev_list, list) {
531 		if (rdev->netdev == netdev) {
532 			rcu_read_unlock();
533 			return rdev;
534 		}
535 	}
536 	rcu_read_unlock();
537 	return NULL;
538 }
539 
540 static void bnxt_re_dev_unprobe(struct net_device *netdev,
541 				struct bnxt_en_dev *en_dev)
542 {
543 	dev_put(netdev);
544 	module_put(en_dev->pdev->driver->driver.owner);
545 }
546 
547 static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev)
548 {
549 	struct bnxt *bp = netdev_priv(netdev);
550 	struct bnxt_en_dev *en_dev;
551 	struct pci_dev *pdev;
552 
553 	/* Call bnxt_en's RoCE probe via indirect API */
554 	if (!bp->ulp_probe)
555 		return ERR_PTR(-EINVAL);
556 
557 	en_dev = bp->ulp_probe(netdev);
558 	if (IS_ERR(en_dev))
559 		return en_dev;
560 
561 	pdev = en_dev->pdev;
562 	if (!pdev)
563 		return ERR_PTR(-EINVAL);
564 
565 	if (!(en_dev->flags & BNXT_EN_FLAG_ROCE_CAP)) {
566 		dev_info(&pdev->dev,
567 			"%s: probe error: RoCE is not supported on this device",
568 			ROCE_DRV_MODULE_NAME);
569 		return ERR_PTR(-ENODEV);
570 	}
571 
572 	/* Bump net device reference count */
573 	if (!try_module_get(pdev->driver->driver.owner))
574 		return ERR_PTR(-ENODEV);
575 
576 	dev_hold(netdev);
577 
578 	return en_dev;
579 }
580 
581 static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr,
582 			   char *buf)
583 {
584 	struct bnxt_re_dev *rdev =
585 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
586 
587 	return scnprintf(buf, PAGE_SIZE, "0x%x\n", rdev->en_dev->pdev->vendor);
588 }
589 static DEVICE_ATTR_RO(hw_rev);
590 
591 static ssize_t hca_type_show(struct device *device,
592 			     struct device_attribute *attr, char *buf)
593 {
594 	struct bnxt_re_dev *rdev =
595 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
596 
597 	return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->ibdev.node_desc);
598 }
599 static DEVICE_ATTR_RO(hca_type);
600 
601 static struct attribute *bnxt_re_attributes[] = {
602 	&dev_attr_hw_rev.attr,
603 	&dev_attr_hca_type.attr,
604 	NULL
605 };
606 
607 static const struct attribute_group bnxt_re_dev_attr_group = {
608 	.attrs = bnxt_re_attributes,
609 };
610 
611 static void bnxt_re_unregister_ib(struct bnxt_re_dev *rdev)
612 {
613 	ib_unregister_device(&rdev->ibdev);
614 }
615 
616 static const struct ib_device_ops bnxt_re_dev_ops = {
617 	.owner = THIS_MODULE,
618 	.driver_id = RDMA_DRIVER_BNXT_RE,
619 	.uverbs_abi_ver = BNXT_RE_ABI_VERSION,
620 
621 	.add_gid = bnxt_re_add_gid,
622 	.alloc_hw_stats = bnxt_re_ib_alloc_hw_stats,
623 	.alloc_mr = bnxt_re_alloc_mr,
624 	.alloc_pd = bnxt_re_alloc_pd,
625 	.alloc_ucontext = bnxt_re_alloc_ucontext,
626 	.create_ah = bnxt_re_create_ah,
627 	.create_cq = bnxt_re_create_cq,
628 	.create_qp = bnxt_re_create_qp,
629 	.create_srq = bnxt_re_create_srq,
630 	.dealloc_pd = bnxt_re_dealloc_pd,
631 	.dealloc_ucontext = bnxt_re_dealloc_ucontext,
632 	.del_gid = bnxt_re_del_gid,
633 	.dereg_mr = bnxt_re_dereg_mr,
634 	.destroy_ah = bnxt_re_destroy_ah,
635 	.destroy_cq = bnxt_re_destroy_cq,
636 	.destroy_qp = bnxt_re_destroy_qp,
637 	.destroy_srq = bnxt_re_destroy_srq,
638 	.get_dev_fw_str = bnxt_re_query_fw_str,
639 	.get_dma_mr = bnxt_re_get_dma_mr,
640 	.get_hw_stats = bnxt_re_ib_get_hw_stats,
641 	.get_link_layer = bnxt_re_get_link_layer,
642 	.get_port_immutable = bnxt_re_get_port_immutable,
643 	.map_mr_sg = bnxt_re_map_mr_sg,
644 	.mmap = bnxt_re_mmap,
645 	.modify_ah = bnxt_re_modify_ah,
646 	.modify_qp = bnxt_re_modify_qp,
647 	.modify_srq = bnxt_re_modify_srq,
648 	.poll_cq = bnxt_re_poll_cq,
649 	.post_recv = bnxt_re_post_recv,
650 	.post_send = bnxt_re_post_send,
651 	.post_srq_recv = bnxt_re_post_srq_recv,
652 	.query_ah = bnxt_re_query_ah,
653 	.query_device = bnxt_re_query_device,
654 	.query_pkey = bnxt_re_query_pkey,
655 	.query_port = bnxt_re_query_port,
656 	.query_qp = bnxt_re_query_qp,
657 	.query_srq = bnxt_re_query_srq,
658 	.reg_user_mr = bnxt_re_reg_user_mr,
659 	.req_notify_cq = bnxt_re_req_notify_cq,
660 	INIT_RDMA_OBJ_SIZE(ib_ah, bnxt_re_ah, ib_ah),
661 	INIT_RDMA_OBJ_SIZE(ib_cq, bnxt_re_cq, ib_cq),
662 	INIT_RDMA_OBJ_SIZE(ib_pd, bnxt_re_pd, ib_pd),
663 	INIT_RDMA_OBJ_SIZE(ib_srq, bnxt_re_srq, ib_srq),
664 	INIT_RDMA_OBJ_SIZE(ib_ucontext, bnxt_re_ucontext, ib_uctx),
665 };
666 
667 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
668 {
669 	struct ib_device *ibdev = &rdev->ibdev;
670 	int ret;
671 
672 	/* ib device init */
673 	ibdev->node_type = RDMA_NODE_IB_CA;
674 	strlcpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
675 		strlen(BNXT_RE_DESC) + 5);
676 	ibdev->phys_port_cnt = 1;
677 
678 	bnxt_qplib_get_guid(rdev->netdev->dev_addr, (u8 *)&ibdev->node_guid);
679 
680 	ibdev->num_comp_vectors	= rdev->num_msix - 1;
681 	ibdev->dev.parent = &rdev->en_dev->pdev->dev;
682 	ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
683 
684 	/* User space */
685 	ibdev->uverbs_cmd_mask =
686 			(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
687 			(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
688 			(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
689 			(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
690 			(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
691 			(1ull << IB_USER_VERBS_CMD_REG_MR)		|
692 			(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
693 			(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
694 			(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
695 			(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
696 			(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
697 			(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
698 			(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
699 			(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
700 			(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
701 			(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
702 			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
703 			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
704 			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
705 			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
706 			(1ull << IB_USER_VERBS_CMD_CREATE_AH)		|
707 			(1ull << IB_USER_VERBS_CMD_MODIFY_AH)		|
708 			(1ull << IB_USER_VERBS_CMD_QUERY_AH)		|
709 			(1ull << IB_USER_VERBS_CMD_DESTROY_AH);
710 	/* POLL_CQ and REQ_NOTIFY_CQ is directly handled in libbnxt_re */
711 
712 
713 	rdma_set_device_sysfs_group(ibdev, &bnxt_re_dev_attr_group);
714 	ib_set_device_ops(ibdev, &bnxt_re_dev_ops);
715 	ret = ib_device_set_netdev(&rdev->ibdev, rdev->netdev, 1);
716 	if (ret)
717 		return ret;
718 
719 	return ib_register_device(ibdev, "bnxt_re%d");
720 }
721 
722 static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev)
723 {
724 	dev_put(rdev->netdev);
725 	rdev->netdev = NULL;
726 
727 	mutex_lock(&bnxt_re_dev_lock);
728 	list_del_rcu(&rdev->list);
729 	mutex_unlock(&bnxt_re_dev_lock);
730 
731 	synchronize_rcu();
732 
733 	ib_dealloc_device(&rdev->ibdev);
734 	/* rdev is gone */
735 }
736 
737 static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev,
738 					   struct bnxt_en_dev *en_dev)
739 {
740 	struct bnxt_re_dev *rdev;
741 
742 	/* Allocate bnxt_re_dev instance here */
743 	rdev = ib_alloc_device(bnxt_re_dev, ibdev);
744 	if (!rdev) {
745 		dev_err(NULL, "%s: bnxt_re_dev allocation failure!",
746 			ROCE_DRV_MODULE_NAME);
747 		return NULL;
748 	}
749 	/* Default values */
750 	rdev->netdev = netdev;
751 	dev_hold(rdev->netdev);
752 	rdev->en_dev = en_dev;
753 	rdev->id = rdev->en_dev->pdev->devfn;
754 	INIT_LIST_HEAD(&rdev->qp_list);
755 	mutex_init(&rdev->qp_lock);
756 	atomic_set(&rdev->qp_count, 0);
757 	atomic_set(&rdev->cq_count, 0);
758 	atomic_set(&rdev->srq_count, 0);
759 	atomic_set(&rdev->mr_count, 0);
760 	atomic_set(&rdev->mw_count, 0);
761 	rdev->cosq[0] = 0xFFFF;
762 	rdev->cosq[1] = 0xFFFF;
763 
764 	mutex_lock(&bnxt_re_dev_lock);
765 	list_add_tail_rcu(&rdev->list, &bnxt_re_dev_list);
766 	mutex_unlock(&bnxt_re_dev_lock);
767 	return rdev;
768 }
769 
770 static int bnxt_re_handle_unaffi_async_event(struct creq_func_event
771 					     *unaffi_async)
772 {
773 	switch (unaffi_async->event) {
774 	case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
775 		break;
776 	case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
777 		break;
778 	case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
779 		break;
780 	case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
781 		break;
782 	case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
783 		break;
784 	case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
785 		break;
786 	case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
787 		break;
788 	case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
789 		break;
790 	case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
791 		break;
792 	case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
793 		break;
794 	case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
795 		break;
796 	default:
797 		return -EINVAL;
798 	}
799 	return 0;
800 }
801 
802 static int bnxt_re_handle_qp_async_event(struct creq_qp_event *qp_event,
803 					 struct bnxt_re_qp *qp)
804 {
805 	struct ib_event event;
806 	unsigned int flags;
807 
808 	if (qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
809 		flags = bnxt_re_lock_cqs(qp);
810 		bnxt_qplib_add_flush_qp(&qp->qplib_qp);
811 		bnxt_re_unlock_cqs(qp, flags);
812 	}
813 
814 	memset(&event, 0, sizeof(event));
815 	if (qp->qplib_qp.srq) {
816 		event.device = &qp->rdev->ibdev;
817 		event.element.qp = &qp->ib_qp;
818 		event.event = IB_EVENT_QP_LAST_WQE_REACHED;
819 	}
820 
821 	if (event.device && qp->ib_qp.event_handler)
822 		qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
823 
824 	return 0;
825 }
826 
827 static int bnxt_re_handle_affi_async_event(struct creq_qp_event *affi_async,
828 					   void *obj)
829 {
830 	int rc = 0;
831 	u8 event;
832 
833 	if (!obj)
834 		return rc; /* QP was already dead, still return success */
835 
836 	event = affi_async->event;
837 	if (event == CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION) {
838 		struct bnxt_qplib_qp *lib_qp = obj;
839 		struct bnxt_re_qp *qp = container_of(lib_qp, struct bnxt_re_qp,
840 						     qplib_qp);
841 		rc = bnxt_re_handle_qp_async_event(affi_async, qp);
842 	}
843 	return rc;
844 }
845 
846 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
847 			       void *aeqe, void *obj)
848 {
849 	struct creq_qp_event *affi_async;
850 	struct creq_func_event *unaffi_async;
851 	u8 type;
852 	int rc;
853 
854 	type = ((struct creq_base *)aeqe)->type;
855 	if (type == CREQ_BASE_TYPE_FUNC_EVENT) {
856 		unaffi_async = aeqe;
857 		rc = bnxt_re_handle_unaffi_async_event(unaffi_async);
858 	} else {
859 		affi_async = aeqe;
860 		rc = bnxt_re_handle_affi_async_event(affi_async, obj);
861 	}
862 
863 	return rc;
864 }
865 
866 static int bnxt_re_srqn_handler(struct bnxt_qplib_nq *nq,
867 				struct bnxt_qplib_srq *handle, u8 event)
868 {
869 	struct bnxt_re_srq *srq = container_of(handle, struct bnxt_re_srq,
870 					       qplib_srq);
871 	struct ib_event ib_event;
872 	int rc = 0;
873 
874 	if (!srq) {
875 		dev_err(NULL, "%s: SRQ is NULL, SRQN not handled",
876 			ROCE_DRV_MODULE_NAME);
877 		rc = -EINVAL;
878 		goto done;
879 	}
880 	ib_event.device = &srq->rdev->ibdev;
881 	ib_event.element.srq = &srq->ib_srq;
882 	if (event == NQ_SRQ_EVENT_EVENT_SRQ_THRESHOLD_EVENT)
883 		ib_event.event = IB_EVENT_SRQ_LIMIT_REACHED;
884 	else
885 		ib_event.event = IB_EVENT_SRQ_ERR;
886 
887 	if (srq->ib_srq.event_handler) {
888 		/* Lock event_handler? */
889 		(*srq->ib_srq.event_handler)(&ib_event,
890 					     srq->ib_srq.srq_context);
891 	}
892 done:
893 	return rc;
894 }
895 
896 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
897 			       struct bnxt_qplib_cq *handle)
898 {
899 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
900 					     qplib_cq);
901 
902 	if (!cq) {
903 		dev_err(NULL, "%s: CQ is NULL, CQN not handled",
904 			ROCE_DRV_MODULE_NAME);
905 		return -EINVAL;
906 	}
907 	if (cq->ib_cq.comp_handler) {
908 		/* Lock comp_handler? */
909 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
910 	}
911 
912 	return 0;
913 }
914 
915 #define BNXT_RE_GEN_P5_PF_NQ_DB		0x10000
916 #define BNXT_RE_GEN_P5_VF_NQ_DB		0x4000
917 static u32 bnxt_re_get_nqdb_offset(struct bnxt_re_dev *rdev, u16 indx)
918 {
919 	return bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx) ?
920 		(rdev->is_virtfn ? BNXT_RE_GEN_P5_VF_NQ_DB :
921 				   BNXT_RE_GEN_P5_PF_NQ_DB) :
922 				   rdev->msix_entries[indx].db_offset;
923 }
924 
925 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
926 {
927 	int i;
928 
929 	for (i = 1; i < rdev->num_msix; i++)
930 		bnxt_qplib_disable_nq(&rdev->nq[i - 1]);
931 
932 	if (rdev->qplib_res.rcfw)
933 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
934 }
935 
936 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
937 {
938 	int num_vec_enabled = 0;
939 	int rc = 0, i;
940 	u32 db_offt;
941 
942 	bnxt_qplib_init_res(&rdev->qplib_res);
943 
944 	for (i = 1; i < rdev->num_msix ; i++) {
945 		db_offt = bnxt_re_get_nqdb_offset(rdev, i);
946 		rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
947 					  i - 1, rdev->msix_entries[i].vector,
948 					  db_offt, &bnxt_re_cqn_handler,
949 					  &bnxt_re_srqn_handler);
950 		if (rc) {
951 			dev_err(rdev_to_dev(rdev),
952 				"Failed to enable NQ with rc = 0x%x", rc);
953 			goto fail;
954 		}
955 		num_vec_enabled++;
956 	}
957 	return 0;
958 fail:
959 	for (i = num_vec_enabled; i >= 0; i--)
960 		bnxt_qplib_disable_nq(&rdev->nq[i]);
961 	return rc;
962 }
963 
964 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev)
965 {
966 	u8 type;
967 	int i;
968 
969 	for (i = 0; i < rdev->num_msix - 1; i++) {
970 		type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
971 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
972 		rdev->nq[i].res = NULL;
973 		bnxt_qplib_free_nq(&rdev->nq[i]);
974 	}
975 }
976 
977 static void bnxt_re_free_res(struct bnxt_re_dev *rdev)
978 {
979 	bnxt_re_free_nq_res(rdev);
980 
981 	if (rdev->qplib_res.dpi_tbl.max) {
982 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
983 				       &rdev->qplib_res.dpi_tbl,
984 				       &rdev->dpi_privileged);
985 	}
986 	if (rdev->qplib_res.rcfw) {
987 		bnxt_qplib_free_res(&rdev->qplib_res);
988 		rdev->qplib_res.rcfw = NULL;
989 	}
990 }
991 
992 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
993 {
994 	int num_vec_created = 0;
995 	dma_addr_t *pg_map;
996 	int rc = 0, i;
997 	int pages;
998 	u8 type;
999 
1000 	/* Configure and allocate resources for qplib */
1001 	rdev->qplib_res.rcfw = &rdev->rcfw;
1002 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
1003 				     rdev->is_virtfn);
1004 	if (rc)
1005 		goto fail;
1006 
1007 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
1008 				  rdev->netdev, &rdev->dev_attr);
1009 	if (rc)
1010 		goto fail;
1011 
1012 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
1013 				  &rdev->dpi_privileged,
1014 				  rdev);
1015 	if (rc)
1016 		goto dealloc_res;
1017 
1018 	for (i = 0; i < rdev->num_msix - 1; i++) {
1019 		rdev->nq[i].res = &rdev->qplib_res;
1020 		rdev->nq[i].hwq.max_elements = BNXT_RE_MAX_CQ_COUNT +
1021 			BNXT_RE_MAX_SRQC_COUNT + 2;
1022 		rc = bnxt_qplib_alloc_nq(rdev->en_dev->pdev, &rdev->nq[i]);
1023 		if (rc) {
1024 			dev_err(rdev_to_dev(rdev), "Alloc Failed NQ%d rc:%#x",
1025 				i, rc);
1026 			goto free_nq;
1027 		}
1028 		type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1029 		pg_map = rdev->nq[i].hwq.pbl[PBL_LVL_0].pg_map_arr;
1030 		pages = rdev->nq[i].hwq.pbl[rdev->nq[i].hwq.level].pg_count;
1031 		rc = bnxt_re_net_ring_alloc(rdev, pg_map, pages, type,
1032 					    BNXT_QPLIB_NQE_MAX_CNT - 1,
1033 					    rdev->msix_entries[i + 1].ring_idx,
1034 					    &rdev->nq[i].ring_id);
1035 		if (rc) {
1036 			dev_err(rdev_to_dev(rdev),
1037 				"Failed to allocate NQ fw id with rc = 0x%x",
1038 				rc);
1039 			bnxt_qplib_free_nq(&rdev->nq[i]);
1040 			goto free_nq;
1041 		}
1042 		num_vec_created++;
1043 	}
1044 	return 0;
1045 free_nq:
1046 	for (i = num_vec_created; i >= 0; i--) {
1047 		type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1048 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
1049 		bnxt_qplib_free_nq(&rdev->nq[i]);
1050 	}
1051 	bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
1052 			       &rdev->qplib_res.dpi_tbl,
1053 			       &rdev->dpi_privileged);
1054 dealloc_res:
1055 	bnxt_qplib_free_res(&rdev->qplib_res);
1056 
1057 fail:
1058 	rdev->qplib_res.rcfw = NULL;
1059 	return rc;
1060 }
1061 
1062 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
1063 				   u8 port_num, enum ib_event_type event)
1064 {
1065 	struct ib_event ib_event;
1066 
1067 	ib_event.device = ibdev;
1068 	if (qp) {
1069 		ib_event.element.qp = qp;
1070 		ib_event.event = event;
1071 		if (qp->event_handler)
1072 			qp->event_handler(&ib_event, qp->qp_context);
1073 
1074 	} else {
1075 		ib_event.element.port_num = port_num;
1076 		ib_event.event = event;
1077 		ib_dispatch_event(&ib_event);
1078 	}
1079 }
1080 
1081 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN      0x02
1082 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
1083 				      u64 *cid_map)
1084 {
1085 	struct hwrm_queue_pri2cos_qcfg_input req = {0};
1086 	struct bnxt *bp = netdev_priv(rdev->netdev);
1087 	struct hwrm_queue_pri2cos_qcfg_output resp;
1088 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1089 	struct bnxt_fw_msg fw_msg;
1090 	u32 flags = 0;
1091 	u8 *qcfgmap, *tmp_map;
1092 	int rc = 0, i;
1093 
1094 	if (!cid_map)
1095 		return -EINVAL;
1096 
1097 	memset(&fw_msg, 0, sizeof(fw_msg));
1098 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1099 			      HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
1100 	flags |= (dir & 0x01);
1101 	flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
1102 	req.flags = cpu_to_le32(flags);
1103 	req.port_id = bp->pf.port_id;
1104 
1105 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1106 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1107 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1108 	if (rc)
1109 		return rc;
1110 
1111 	if (resp.queue_cfg_info) {
1112 		dev_warn(rdev_to_dev(rdev),
1113 			 "Asymmetric cos queue configuration detected");
1114 		dev_warn(rdev_to_dev(rdev),
1115 			 " on device, QoS may not be fully functional\n");
1116 	}
1117 	qcfgmap = &resp.pri0_cos_queue_id;
1118 	tmp_map = (u8 *)cid_map;
1119 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
1120 		tmp_map[i] = qcfgmap[i];
1121 
1122 	return rc;
1123 }
1124 
1125 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
1126 					struct bnxt_re_qp *qp)
1127 {
1128 	return (qp->ib_qp.qp_type == IB_QPT_GSI) || (qp == rdev->qp1_sqp);
1129 }
1130 
1131 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
1132 {
1133 	int mask = IB_QP_STATE;
1134 	struct ib_qp_attr qp_attr;
1135 	struct bnxt_re_qp *qp;
1136 
1137 	qp_attr.qp_state = IB_QPS_ERR;
1138 	mutex_lock(&rdev->qp_lock);
1139 	list_for_each_entry(qp, &rdev->qp_list, list) {
1140 		/* Modify the state of all QPs except QP1/Shadow QP */
1141 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
1142 			if (qp->qplib_qp.state !=
1143 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
1144 			    qp->qplib_qp.state !=
1145 			    CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1146 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
1147 						       1, IB_EVENT_QP_FATAL);
1148 				bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
1149 						  NULL);
1150 			}
1151 		}
1152 	}
1153 	mutex_unlock(&rdev->qp_lock);
1154 }
1155 
1156 static int bnxt_re_update_gid(struct bnxt_re_dev *rdev)
1157 {
1158 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
1159 	struct bnxt_qplib_gid gid;
1160 	u16 gid_idx, index;
1161 	int rc = 0;
1162 
1163 	if (!test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1164 		return 0;
1165 
1166 	if (!sgid_tbl) {
1167 		dev_err(rdev_to_dev(rdev), "QPLIB: SGID table not allocated");
1168 		return -EINVAL;
1169 	}
1170 
1171 	for (index = 0; index < sgid_tbl->active; index++) {
1172 		gid_idx = sgid_tbl->hw_id[index];
1173 
1174 		if (!memcmp(&sgid_tbl->tbl[index], &bnxt_qplib_gid_zero,
1175 			    sizeof(bnxt_qplib_gid_zero)))
1176 			continue;
1177 		/* need to modify the VLAN enable setting of non VLAN GID only
1178 		 * as setting is done for VLAN GID while adding GID
1179 		 */
1180 		if (sgid_tbl->vlan[index])
1181 			continue;
1182 
1183 		memcpy(&gid, &sgid_tbl->tbl[index], sizeof(gid));
1184 
1185 		rc = bnxt_qplib_update_sgid(sgid_tbl, &gid, gid_idx,
1186 					    rdev->qplib_res.netdev->dev_addr);
1187 	}
1188 
1189 	return rc;
1190 }
1191 
1192 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
1193 {
1194 	u32 prio_map = 0, tmp_map = 0;
1195 	struct net_device *netdev;
1196 	struct dcb_app app;
1197 
1198 	netdev = rdev->netdev;
1199 
1200 	memset(&app, 0, sizeof(app));
1201 	app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
1202 	app.protocol = ETH_P_IBOE;
1203 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1204 	prio_map = tmp_map;
1205 
1206 	app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
1207 	app.protocol = ROCE_V2_UDP_DPORT;
1208 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1209 	prio_map |= tmp_map;
1210 
1211 	return prio_map;
1212 }
1213 
1214 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
1215 {
1216 	u16 prio;
1217 	u8 id;
1218 
1219 	for (prio = 0, id = 0; prio < 8; prio++) {
1220 		if (prio_map & (1 << prio)) {
1221 			cosq[id] = cid_map[prio];
1222 			id++;
1223 			if (id == 2) /* Max 2 tcs supported */
1224 				break;
1225 		}
1226 	}
1227 }
1228 
1229 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
1230 {
1231 	u8 prio_map = 0;
1232 	u64 cid_map;
1233 	int rc;
1234 
1235 	/* Get priority for roce */
1236 	prio_map = bnxt_re_get_priority_mask(rdev);
1237 
1238 	if (prio_map == rdev->cur_prio_map)
1239 		return 0;
1240 	rdev->cur_prio_map = prio_map;
1241 	/* Get cosq id for this priority */
1242 	rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
1243 	if (rc) {
1244 		dev_warn(rdev_to_dev(rdev), "no cos for p_mask %x\n", prio_map);
1245 		return rc;
1246 	}
1247 	/* Parse CoS IDs for app priority */
1248 	bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
1249 
1250 	/* Config BONO. */
1251 	rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
1252 	if (rc) {
1253 		dev_warn(rdev_to_dev(rdev), "no tc for cos{%x, %x}\n",
1254 			 rdev->cosq[0], rdev->cosq[1]);
1255 		return rc;
1256 	}
1257 
1258 	/* Actual priorities are not programmed as they are already
1259 	 * done by L2 driver; just enable or disable priority vlan tagging
1260 	 */
1261 	if ((prio_map == 0 && rdev->qplib_res.prio) ||
1262 	    (prio_map != 0 && !rdev->qplib_res.prio)) {
1263 		rdev->qplib_res.prio = prio_map ? true : false;
1264 
1265 		bnxt_re_update_gid(rdev);
1266 	}
1267 
1268 	return 0;
1269 }
1270 
1271 static void bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev)
1272 {
1273 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1274 	struct hwrm_ver_get_output resp = {0};
1275 	struct hwrm_ver_get_input req = {0};
1276 	struct bnxt_fw_msg fw_msg;
1277 	int rc = 0;
1278 
1279 	memset(&fw_msg, 0, sizeof(fw_msg));
1280 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1281 			      HWRM_VER_GET, -1, -1);
1282 	req.hwrm_intf_maj = HWRM_VERSION_MAJOR;
1283 	req.hwrm_intf_min = HWRM_VERSION_MINOR;
1284 	req.hwrm_intf_upd = HWRM_VERSION_UPDATE;
1285 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1286 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1287 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1288 	if (rc) {
1289 		dev_err(rdev_to_dev(rdev),
1290 			"Failed to query HW version, rc = 0x%x", rc);
1291 		return;
1292 	}
1293 	rdev->qplib_ctx.hwrm_intf_ver =
1294 		(u64)le16_to_cpu(resp.hwrm_intf_major) << 48 |
1295 		(u64)le16_to_cpu(resp.hwrm_intf_minor) << 32 |
1296 		(u64)le16_to_cpu(resp.hwrm_intf_build) << 16 |
1297 		le16_to_cpu(resp.hwrm_intf_patch);
1298 }
1299 
1300 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev)
1301 {
1302 	u8 type;
1303 	int rc;
1304 
1305 	if (test_and_clear_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags)) {
1306 		/* Cleanup ib dev */
1307 		bnxt_re_unregister_ib(rdev);
1308 	}
1309 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
1310 		cancel_delayed_work_sync(&rdev->worker);
1311 
1312 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED,
1313 			       &rdev->flags))
1314 		bnxt_re_cleanup_res(rdev);
1315 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags))
1316 		bnxt_re_free_res(rdev);
1317 
1318 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
1319 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
1320 		if (rc)
1321 			dev_warn(rdev_to_dev(rdev),
1322 				 "Failed to deinitialize RCFW: %#x", rc);
1323 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1324 		bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1325 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1326 		type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1327 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, type);
1328 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1329 	}
1330 	if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
1331 		rc = bnxt_re_free_msix(rdev);
1332 		if (rc)
1333 			dev_warn(rdev_to_dev(rdev),
1334 				 "Failed to free MSI-X vectors: %#x", rc);
1335 	}
1336 
1337 	bnxt_re_destroy_chip_ctx(rdev);
1338 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
1339 		rc = bnxt_re_unregister_netdev(rdev);
1340 		if (rc)
1341 			dev_warn(rdev_to_dev(rdev),
1342 				 "Failed to unregister with netdev: %#x", rc);
1343 	}
1344 }
1345 
1346 /* worker thread for polling periodic events. Now used for QoS programming*/
1347 static void bnxt_re_worker(struct work_struct *work)
1348 {
1349 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
1350 						worker.work);
1351 
1352 	bnxt_re_setup_qos(rdev);
1353 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1354 }
1355 
1356 static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
1357 {
1358 	dma_addr_t *pg_map;
1359 	u32 db_offt, ridx;
1360 	int pages, vid;
1361 	bool locked;
1362 	u8 type;
1363 	int rc;
1364 
1365 	/* Acquire rtnl lock through out this function */
1366 	rtnl_lock();
1367 	locked = true;
1368 
1369 	/* Registered a new RoCE device instance to netdev */
1370 	rc = bnxt_re_register_netdev(rdev);
1371 	if (rc) {
1372 		rtnl_unlock();
1373 		pr_err("Failed to register with netedev: %#x\n", rc);
1374 		return -EINVAL;
1375 	}
1376 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1377 
1378 	rc = bnxt_re_setup_chip_ctx(rdev);
1379 	if (rc) {
1380 		dev_err(rdev_to_dev(rdev), "Failed to get chip context\n");
1381 		return -EINVAL;
1382 	}
1383 
1384 	/* Check whether VF or PF */
1385 	bnxt_re_get_sriov_func_type(rdev);
1386 
1387 	rc = bnxt_re_request_msix(rdev);
1388 	if (rc) {
1389 		pr_err("Failed to get MSI-X vectors: %#x\n", rc);
1390 		rc = -EINVAL;
1391 		goto fail;
1392 	}
1393 	set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
1394 
1395 	bnxt_re_query_hwrm_intf_version(rdev);
1396 
1397 	/* Establish RCFW Communication Channel to initialize the context
1398 	 * memory for the function and all child VFs
1399 	 */
1400 	rc = bnxt_qplib_alloc_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw,
1401 					   &rdev->qplib_ctx,
1402 					   BNXT_RE_MAX_QPC_COUNT);
1403 	if (rc) {
1404 		pr_err("Failed to allocate RCFW Channel: %#x\n", rc);
1405 		goto fail;
1406 	}
1407 	type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1408 	pg_map = rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr;
1409 	pages = rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count;
1410 	ridx = rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx;
1411 	rc = bnxt_re_net_ring_alloc(rdev, pg_map, pages, type,
1412 				    BNXT_QPLIB_CREQE_MAX_CNT - 1,
1413 				    ridx, &rdev->rcfw.creq_ring_id);
1414 	if (rc) {
1415 		pr_err("Failed to allocate CREQ: %#x\n", rc);
1416 		goto free_rcfw;
1417 	}
1418 	db_offt = bnxt_re_get_nqdb_offset(rdev, BNXT_RE_AEQ_IDX);
1419 	vid = rdev->msix_entries[BNXT_RE_AEQ_IDX].vector;
1420 	rc = bnxt_qplib_enable_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw,
1421 					    vid, db_offt, rdev->is_virtfn,
1422 					    &bnxt_re_aeq_handler);
1423 	if (rc) {
1424 		pr_err("Failed to enable RCFW channel: %#x\n", rc);
1425 		goto free_ring;
1426 	}
1427 
1428 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
1429 				     rdev->is_virtfn);
1430 	if (rc)
1431 		goto disable_rcfw;
1432 
1433 	bnxt_re_set_resource_limits(rdev);
1434 
1435 	rc = bnxt_qplib_alloc_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx, 0,
1436 				  bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx));
1437 	if (rc) {
1438 		pr_err("Failed to allocate QPLIB context: %#x\n", rc);
1439 		goto disable_rcfw;
1440 	}
1441 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
1442 					 rdev->qplib_ctx.stats.dma_map,
1443 					 &rdev->qplib_ctx.stats.fw_id);
1444 	if (rc) {
1445 		pr_err("Failed to allocate stats context: %#x\n", rc);
1446 		goto free_ctx;
1447 	}
1448 
1449 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx,
1450 				  rdev->is_virtfn);
1451 	if (rc) {
1452 		pr_err("Failed to initialize RCFW: %#x\n", rc);
1453 		goto free_sctx;
1454 	}
1455 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1456 
1457 	/* Resources based on the 'new' device caps */
1458 	rc = bnxt_re_alloc_res(rdev);
1459 	if (rc) {
1460 		pr_err("Failed to allocate resources: %#x\n", rc);
1461 		goto fail;
1462 	}
1463 	set_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags);
1464 	rc = bnxt_re_init_res(rdev);
1465 	if (rc) {
1466 		pr_err("Failed to initialize resources: %#x\n", rc);
1467 		goto fail;
1468 	}
1469 
1470 	set_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED, &rdev->flags);
1471 
1472 	if (!rdev->is_virtfn) {
1473 		rc = bnxt_re_setup_qos(rdev);
1474 		if (rc)
1475 			pr_info("RoCE priority not yet configured\n");
1476 
1477 		INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1478 		set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1479 		schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1480 	}
1481 
1482 	rtnl_unlock();
1483 	locked = false;
1484 
1485 	/* Register ib dev */
1486 	rc = bnxt_re_register_ib(rdev);
1487 	if (rc) {
1488 		pr_err("Failed to register with IB: %#x\n", rc);
1489 		goto fail;
1490 	}
1491 	set_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags);
1492 	dev_info(rdev_to_dev(rdev), "Device registered successfully");
1493 	ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1494 			 &rdev->active_width);
1495 	set_bit(BNXT_RE_FLAG_ISSUE_ROCE_STATS, &rdev->flags);
1496 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_PORT_ACTIVE);
1497 
1498 	return 0;
1499 free_sctx:
1500 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1501 free_ctx:
1502 	bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1503 disable_rcfw:
1504 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1505 free_ring:
1506 	type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1507 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, type);
1508 free_rcfw:
1509 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1510 fail:
1511 	if (!locked)
1512 		rtnl_lock();
1513 	bnxt_re_ib_unreg(rdev);
1514 	rtnl_unlock();
1515 
1516 	return rc;
1517 }
1518 
1519 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1520 {
1521 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1522 	struct net_device *netdev = rdev->netdev;
1523 
1524 	bnxt_re_dev_remove(rdev);
1525 
1526 	if (netdev)
1527 		bnxt_re_dev_unprobe(netdev, en_dev);
1528 }
1529 
1530 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1531 {
1532 	struct bnxt_en_dev *en_dev;
1533 	int rc = 0;
1534 
1535 	if (!is_bnxt_re_dev(netdev))
1536 		return -ENODEV;
1537 
1538 	en_dev = bnxt_re_dev_probe(netdev);
1539 	if (IS_ERR(en_dev)) {
1540 		if (en_dev != ERR_PTR(-ENODEV))
1541 			pr_err("%s: Failed to probe\n", ROCE_DRV_MODULE_NAME);
1542 		rc = PTR_ERR(en_dev);
1543 		goto exit;
1544 	}
1545 	*rdev = bnxt_re_dev_add(netdev, en_dev);
1546 	if (!*rdev) {
1547 		rc = -ENOMEM;
1548 		bnxt_re_dev_unprobe(netdev, en_dev);
1549 		goto exit;
1550 	}
1551 exit:
1552 	return rc;
1553 }
1554 
1555 static void bnxt_re_remove_one(struct bnxt_re_dev *rdev)
1556 {
1557 	pci_dev_put(rdev->en_dev->pdev);
1558 }
1559 
1560 /* Handle all deferred netevents tasks */
1561 static void bnxt_re_task(struct work_struct *work)
1562 {
1563 	struct bnxt_re_work *re_work;
1564 	struct bnxt_re_dev *rdev;
1565 	int rc = 0;
1566 
1567 	re_work = container_of(work, struct bnxt_re_work, work);
1568 	rdev = re_work->rdev;
1569 
1570 	if (re_work->event != NETDEV_REGISTER &&
1571 	    !test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1572 		return;
1573 
1574 	switch (re_work->event) {
1575 	case NETDEV_REGISTER:
1576 		rc = bnxt_re_ib_reg(rdev);
1577 		if (rc) {
1578 			dev_err(rdev_to_dev(rdev),
1579 				"Failed to register with IB: %#x", rc);
1580 			bnxt_re_remove_one(rdev);
1581 			bnxt_re_dev_unreg(rdev);
1582 			goto exit;
1583 		}
1584 		break;
1585 	case NETDEV_UP:
1586 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1587 				       IB_EVENT_PORT_ACTIVE);
1588 		break;
1589 	case NETDEV_DOWN:
1590 		bnxt_re_dev_stop(rdev);
1591 		break;
1592 	case NETDEV_CHANGE:
1593 		if (!netif_carrier_ok(rdev->netdev))
1594 			bnxt_re_dev_stop(rdev);
1595 		else if (netif_carrier_ok(rdev->netdev))
1596 			bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1597 					       IB_EVENT_PORT_ACTIVE);
1598 		ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1599 				 &rdev->active_width);
1600 		break;
1601 	default:
1602 		break;
1603 	}
1604 	smp_mb__before_atomic();
1605 	atomic_dec(&rdev->sched_count);
1606 exit:
1607 	kfree(re_work);
1608 }
1609 
1610 static void bnxt_re_init_one(struct bnxt_re_dev *rdev)
1611 {
1612 	pci_dev_get(rdev->en_dev->pdev);
1613 }
1614 
1615 /*
1616  * "Notifier chain callback can be invoked for the same chain from
1617  * different CPUs at the same time".
1618  *
1619  * For cases when the netdev is already present, our call to the
1620  * register_netdevice_notifier() will actually get the rtnl_lock()
1621  * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1622  * events.
1623  *
1624  * But for cases when the netdev is not already present, the notifier
1625  * chain is subjected to be invoked from different CPUs simultaneously.
1626  *
1627  * This is protected by the netdev_mutex.
1628  */
1629 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1630 				unsigned long event, void *ptr)
1631 {
1632 	struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1633 	struct bnxt_re_work *re_work;
1634 	struct bnxt_re_dev *rdev;
1635 	int rc = 0;
1636 	bool sch_work = false;
1637 
1638 	real_dev = rdma_vlan_dev_real_dev(netdev);
1639 	if (!real_dev)
1640 		real_dev = netdev;
1641 
1642 	rdev = bnxt_re_from_netdev(real_dev);
1643 	if (!rdev && event != NETDEV_REGISTER)
1644 		goto exit;
1645 	if (real_dev != netdev)
1646 		goto exit;
1647 
1648 	switch (event) {
1649 	case NETDEV_REGISTER:
1650 		if (rdev)
1651 			break;
1652 		rc = bnxt_re_dev_reg(&rdev, real_dev);
1653 		if (rc == -ENODEV)
1654 			break;
1655 		if (rc) {
1656 			pr_err("Failed to register with the device %s: %#x\n",
1657 			       real_dev->name, rc);
1658 			break;
1659 		}
1660 		bnxt_re_init_one(rdev);
1661 		sch_work = true;
1662 		break;
1663 
1664 	case NETDEV_UNREGISTER:
1665 		/* netdev notifier will call NETDEV_UNREGISTER again later since
1666 		 * we are still holding the reference to the netdev
1667 		 */
1668 		if (atomic_read(&rdev->sched_count) > 0)
1669 			goto exit;
1670 		bnxt_re_ib_unreg(rdev);
1671 		bnxt_re_remove_one(rdev);
1672 		bnxt_re_dev_unreg(rdev);
1673 		break;
1674 
1675 	default:
1676 		sch_work = true;
1677 		break;
1678 	}
1679 	if (sch_work) {
1680 		/* Allocate for the deferred task */
1681 		re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1682 		if (re_work) {
1683 			re_work->rdev = rdev;
1684 			re_work->event = event;
1685 			re_work->vlan_dev = (real_dev == netdev ?
1686 					     NULL : netdev);
1687 			INIT_WORK(&re_work->work, bnxt_re_task);
1688 			atomic_inc(&rdev->sched_count);
1689 			queue_work(bnxt_re_wq, &re_work->work);
1690 		}
1691 	}
1692 
1693 exit:
1694 	return NOTIFY_DONE;
1695 }
1696 
1697 static struct notifier_block bnxt_re_netdev_notifier = {
1698 	.notifier_call = bnxt_re_netdev_event
1699 };
1700 
1701 static int __init bnxt_re_mod_init(void)
1702 {
1703 	int rc = 0;
1704 
1705 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1706 
1707 	bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1708 	if (!bnxt_re_wq)
1709 		return -ENOMEM;
1710 
1711 	INIT_LIST_HEAD(&bnxt_re_dev_list);
1712 
1713 	rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1714 	if (rc) {
1715 		pr_err("%s: Cannot register to netdevice_notifier",
1716 		       ROCE_DRV_MODULE_NAME);
1717 		goto err_netdev;
1718 	}
1719 	return 0;
1720 
1721 err_netdev:
1722 	destroy_workqueue(bnxt_re_wq);
1723 
1724 	return rc;
1725 }
1726 
1727 static void __exit bnxt_re_mod_exit(void)
1728 {
1729 	struct bnxt_re_dev *rdev, *next;
1730 	LIST_HEAD(to_be_deleted);
1731 
1732 	mutex_lock(&bnxt_re_dev_lock);
1733 	/* Free all adapter allocated resources */
1734 	if (!list_empty(&bnxt_re_dev_list))
1735 		list_splice_init(&bnxt_re_dev_list, &to_be_deleted);
1736 	mutex_unlock(&bnxt_re_dev_lock);
1737        /*
1738 	* Cleanup the devices in reverse order so that the VF device
1739 	* cleanup is done before PF cleanup
1740 	*/
1741 	list_for_each_entry_safe_reverse(rdev, next, &to_be_deleted, list) {
1742 		dev_info(rdev_to_dev(rdev), "Unregistering Device");
1743 		/*
1744 		 * Flush out any scheduled tasks before destroying the
1745 		 * resources
1746 		 */
1747 		flush_workqueue(bnxt_re_wq);
1748 		bnxt_re_dev_stop(rdev);
1749 		/* Acquire the rtnl_lock as the L2 resources are freed here */
1750 		rtnl_lock();
1751 		bnxt_re_ib_unreg(rdev);
1752 		rtnl_unlock();
1753 		bnxt_re_remove_one(rdev);
1754 		bnxt_re_dev_unreg(rdev);
1755 	}
1756 	unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1757 	if (bnxt_re_wq)
1758 		destroy_workqueue(bnxt_re_wq);
1759 }
1760 
1761 module_init(bnxt_re_mod_init);
1762 module_exit(bnxt_re_mod_exit);
1763