xref: /linux/drivers/infiniband/hw/bnxt_re/main.c (revision b8d312aa075f33282565467662c4628dae0a2aff)
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_set_resource_limits(struct bnxt_re_dev *rdev)
123 {
124 	u32 vf_qps = 0, vf_srqs = 0, vf_cqs = 0, vf_mrws = 0, vf_gids = 0;
125 	u32 i;
126 	u32 vf_pct;
127 	u32 num_vfs;
128 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
129 
130 	rdev->qplib_ctx.qpc_count = min_t(u32, BNXT_RE_MAX_QPC_COUNT,
131 					  dev_attr->max_qp);
132 
133 	rdev->qplib_ctx.mrw_count = BNXT_RE_MAX_MRW_COUNT_256K;
134 	/* Use max_mr from fw since max_mrw does not get set */
135 	rdev->qplib_ctx.mrw_count = min_t(u32, rdev->qplib_ctx.mrw_count,
136 					  dev_attr->max_mr);
137 	rdev->qplib_ctx.srqc_count = min_t(u32, BNXT_RE_MAX_SRQC_COUNT,
138 					   dev_attr->max_srq);
139 	rdev->qplib_ctx.cq_count = min_t(u32, BNXT_RE_MAX_CQ_COUNT,
140 					 dev_attr->max_cq);
141 
142 	for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
143 		rdev->qplib_ctx.tqm_count[i] =
144 		rdev->dev_attr.tqm_alloc_reqs[i];
145 
146 	if (rdev->num_vfs) {
147 		/*
148 		 * Reserve a set of resources for the PF. Divide the remaining
149 		 * resources among the VFs
150 		 */
151 		vf_pct = 100 - BNXT_RE_PCT_RSVD_FOR_PF;
152 		num_vfs = 100 * rdev->num_vfs;
153 		vf_qps = (rdev->qplib_ctx.qpc_count * vf_pct) / num_vfs;
154 		vf_srqs = (rdev->qplib_ctx.srqc_count * vf_pct) / num_vfs;
155 		vf_cqs = (rdev->qplib_ctx.cq_count * vf_pct) / num_vfs;
156 		/*
157 		 * The driver allows many more MRs than other resources. If the
158 		 * firmware does also, then reserve a fixed amount for the PF
159 		 * and divide the rest among VFs. VFs may use many MRs for NFS
160 		 * mounts, ISER, NVME applications, etc. If the firmware
161 		 * severely restricts the number of MRs, then let PF have
162 		 * half and divide the rest among VFs, as for the other
163 		 * resource types.
164 		 */
165 		if (rdev->qplib_ctx.mrw_count < BNXT_RE_MAX_MRW_COUNT_64K)
166 			vf_mrws = rdev->qplib_ctx.mrw_count * vf_pct / num_vfs;
167 		else
168 			vf_mrws = (rdev->qplib_ctx.mrw_count -
169 				   BNXT_RE_RESVD_MR_FOR_PF) / rdev->num_vfs;
170 		vf_gids = BNXT_RE_MAX_GID_PER_VF;
171 	}
172 	rdev->qplib_ctx.vf_res.max_mrw_per_vf = vf_mrws;
173 	rdev->qplib_ctx.vf_res.max_gid_per_vf = vf_gids;
174 	rdev->qplib_ctx.vf_res.max_qp_per_vf = vf_qps;
175 	rdev->qplib_ctx.vf_res.max_srq_per_vf = vf_srqs;
176 	rdev->qplib_ctx.vf_res.max_cq_per_vf = vf_cqs;
177 }
178 
179 /* for handling bnxt_en callbacks later */
180 static void bnxt_re_stop(void *p)
181 {
182 }
183 
184 static void bnxt_re_start(void *p)
185 {
186 }
187 
188 static void bnxt_re_sriov_config(void *p, int num_vfs)
189 {
190 	struct bnxt_re_dev *rdev = p;
191 
192 	if (!rdev)
193 		return;
194 
195 	rdev->num_vfs = num_vfs;
196 	bnxt_re_set_resource_limits(rdev);
197 	bnxt_qplib_set_func_resources(&rdev->qplib_res, &rdev->rcfw,
198 				      &rdev->qplib_ctx);
199 }
200 
201 static void bnxt_re_shutdown(void *p)
202 {
203 	struct bnxt_re_dev *rdev = p;
204 
205 	if (!rdev)
206 		return;
207 
208 	bnxt_re_ib_unreg(rdev);
209 }
210 
211 static void bnxt_re_stop_irq(void *handle)
212 {
213 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
214 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
215 	struct bnxt_qplib_nq *nq;
216 	int indx;
217 
218 	for (indx = BNXT_RE_NQ_IDX; indx < rdev->num_msix; indx++) {
219 		nq = &rdev->nq[indx - 1];
220 		bnxt_qplib_nq_stop_irq(nq, false);
221 	}
222 
223 	bnxt_qplib_rcfw_stop_irq(rcfw, false);
224 }
225 
226 static void bnxt_re_start_irq(void *handle, struct bnxt_msix_entry *ent)
227 {
228 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
229 	struct bnxt_msix_entry *msix_ent = rdev->msix_entries;
230 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
231 	struct bnxt_qplib_nq *nq;
232 	int indx, rc;
233 
234 	if (!ent) {
235 		/* Not setting the f/w timeout bit in rcfw.
236 		 * During the driver unload the first command
237 		 * to f/w will timeout and that will set the
238 		 * timeout bit.
239 		 */
240 		dev_err(rdev_to_dev(rdev), "Failed to re-start IRQs\n");
241 		return;
242 	}
243 
244 	/* Vectors may change after restart, so update with new vectors
245 	 * in device sctructure.
246 	 */
247 	for (indx = 0; indx < rdev->num_msix; indx++)
248 		rdev->msix_entries[indx].vector = ent[indx].vector;
249 
250 	bnxt_qplib_rcfw_start_irq(rcfw, msix_ent[BNXT_RE_AEQ_IDX].vector,
251 				  false);
252 	for (indx = BNXT_RE_NQ_IDX ; indx < rdev->num_msix; indx++) {
253 		nq = &rdev->nq[indx - 1];
254 		rc = bnxt_qplib_nq_start_irq(nq, indx - 1,
255 					     msix_ent[indx].vector, false);
256 		if (rc)
257 			dev_warn(rdev_to_dev(rdev),
258 				 "Failed to reinit NQ index %d\n", indx - 1);
259 	}
260 }
261 
262 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
263 	.ulp_async_notifier = NULL,
264 	.ulp_stop = bnxt_re_stop,
265 	.ulp_start = bnxt_re_start,
266 	.ulp_sriov_config = bnxt_re_sriov_config,
267 	.ulp_shutdown = bnxt_re_shutdown,
268 	.ulp_irq_stop = bnxt_re_stop_irq,
269 	.ulp_irq_restart = bnxt_re_start_irq
270 };
271 
272 /* RoCE -> Net driver */
273 
274 /* Driver registration routines used to let the networking driver (bnxt_en)
275  * to know that the RoCE driver is now installed
276  */
277 static int bnxt_re_unregister_netdev(struct bnxt_re_dev *rdev)
278 {
279 	struct bnxt_en_dev *en_dev;
280 	int rc;
281 
282 	if (!rdev)
283 		return -EINVAL;
284 
285 	en_dev = rdev->en_dev;
286 
287 	rc = en_dev->en_ops->bnxt_unregister_device(rdev->en_dev,
288 						    BNXT_ROCE_ULP);
289 	return rc;
290 }
291 
292 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
293 {
294 	struct bnxt_en_dev *en_dev;
295 	int rc = 0;
296 
297 	if (!rdev)
298 		return -EINVAL;
299 
300 	en_dev = rdev->en_dev;
301 
302 	rc = en_dev->en_ops->bnxt_register_device(en_dev, BNXT_ROCE_ULP,
303 						  &bnxt_re_ulp_ops, rdev);
304 	rdev->qplib_res.pdev = rdev->en_dev->pdev;
305 	return rc;
306 }
307 
308 static int bnxt_re_free_msix(struct bnxt_re_dev *rdev)
309 {
310 	struct bnxt_en_dev *en_dev;
311 	int rc;
312 
313 	if (!rdev)
314 		return -EINVAL;
315 
316 	en_dev = rdev->en_dev;
317 
318 
319 	rc = en_dev->en_ops->bnxt_free_msix(rdev->en_dev, BNXT_ROCE_ULP);
320 
321 	return rc;
322 }
323 
324 static int bnxt_re_request_msix(struct bnxt_re_dev *rdev)
325 {
326 	int rc = 0, num_msix_want = BNXT_RE_MAX_MSIX, num_msix_got;
327 	struct bnxt_en_dev *en_dev;
328 
329 	if (!rdev)
330 		return -EINVAL;
331 
332 	en_dev = rdev->en_dev;
333 
334 	num_msix_want = min_t(u32, BNXT_RE_MAX_MSIX, num_online_cpus());
335 
336 	num_msix_got = en_dev->en_ops->bnxt_request_msix(en_dev, BNXT_ROCE_ULP,
337 							 rdev->msix_entries,
338 							 num_msix_want);
339 	if (num_msix_got < BNXT_RE_MIN_MSIX) {
340 		rc = -EINVAL;
341 		goto done;
342 	}
343 	if (num_msix_got != num_msix_want) {
344 		dev_warn(rdev_to_dev(rdev),
345 			 "Requested %d MSI-X vectors, got %d\n",
346 			 num_msix_want, num_msix_got);
347 	}
348 	rdev->num_msix = num_msix_got;
349 done:
350 	return rc;
351 }
352 
353 static void bnxt_re_init_hwrm_hdr(struct bnxt_re_dev *rdev, struct input *hdr,
354 				  u16 opcd, u16 crid, u16 trid)
355 {
356 	hdr->req_type = cpu_to_le16(opcd);
357 	hdr->cmpl_ring = cpu_to_le16(crid);
358 	hdr->target_id = cpu_to_le16(trid);
359 }
360 
361 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
362 				int msg_len, void *resp, int resp_max_len,
363 				int timeout)
364 {
365 	fw_msg->msg = msg;
366 	fw_msg->msg_len = msg_len;
367 	fw_msg->resp = resp;
368 	fw_msg->resp_max_len = resp_max_len;
369 	fw_msg->timeout = timeout;
370 }
371 
372 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev,
373 				 u16 fw_ring_id, int type)
374 {
375 	struct bnxt_en_dev *en_dev = rdev->en_dev;
376 	struct hwrm_ring_free_input req = {0};
377 	struct hwrm_ring_free_output resp;
378 	struct bnxt_fw_msg fw_msg;
379 	int rc = -EINVAL;
380 
381 	if (!en_dev)
382 		return rc;
383 
384 	memset(&fw_msg, 0, sizeof(fw_msg));
385 
386 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1);
387 	req.ring_type = type;
388 	req.ring_id = cpu_to_le16(fw_ring_id);
389 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
390 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
391 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
392 	if (rc)
393 		dev_err(rdev_to_dev(rdev),
394 			"Failed to free HW ring:%d :%#x", req.ring_id, rc);
395 	return rc;
396 }
397 
398 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev, dma_addr_t *dma_arr,
399 				  int pages, int type, u32 ring_mask,
400 				  u32 map_index, u16 *fw_ring_id)
401 {
402 	struct bnxt_en_dev *en_dev = rdev->en_dev;
403 	struct hwrm_ring_alloc_input req = {0};
404 	struct hwrm_ring_alloc_output resp;
405 	struct bnxt_fw_msg fw_msg;
406 	int rc = -EINVAL;
407 
408 	if (!en_dev)
409 		return rc;
410 
411 	memset(&fw_msg, 0, sizeof(fw_msg));
412 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_ALLOC, -1, -1);
413 	req.enables = 0;
414 	req.page_tbl_addr =  cpu_to_le64(dma_arr[0]);
415 	if (pages > 1) {
416 		/* Page size is in log2 units */
417 		req.page_size = BNXT_PAGE_SHIFT;
418 		req.page_tbl_depth = 1;
419 	}
420 	req.fbo = 0;
421 	/* Association of ring index with doorbell index and MSIX number */
422 	req.logical_id = cpu_to_le16(map_index);
423 	req.length = cpu_to_le32(ring_mask + 1);
424 	req.ring_type = type;
425 	req.int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
426 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
427 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
428 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
429 	if (!rc)
430 		*fw_ring_id = le16_to_cpu(resp.ring_id);
431 
432 	return rc;
433 }
434 
435 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
436 				      u32 fw_stats_ctx_id)
437 {
438 	struct bnxt_en_dev *en_dev = rdev->en_dev;
439 	struct hwrm_stat_ctx_free_input req = {0};
440 	struct bnxt_fw_msg fw_msg;
441 	int rc = -EINVAL;
442 
443 	if (!en_dev)
444 		return rc;
445 
446 	memset(&fw_msg, 0, sizeof(fw_msg));
447 
448 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_FREE, -1, -1);
449 	req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
450 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&req,
451 			    sizeof(req), DFLT_HWRM_CMD_TIMEOUT);
452 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
453 	if (rc)
454 		dev_err(rdev_to_dev(rdev),
455 			"Failed to free HW stats context %#x", rc);
456 
457 	return rc;
458 }
459 
460 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
461 				       dma_addr_t dma_map,
462 				       u32 *fw_stats_ctx_id)
463 {
464 	struct hwrm_stat_ctx_alloc_output resp = {0};
465 	struct hwrm_stat_ctx_alloc_input req = {0};
466 	struct bnxt_en_dev *en_dev = rdev->en_dev;
467 	struct bnxt_fw_msg fw_msg;
468 	int rc = -EINVAL;
469 
470 	*fw_stats_ctx_id = INVALID_STATS_CTX_ID;
471 
472 	if (!en_dev)
473 		return rc;
474 
475 	memset(&fw_msg, 0, sizeof(fw_msg));
476 
477 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1);
478 	req.update_period_ms = cpu_to_le32(1000);
479 	req.stats_dma_addr = cpu_to_le64(dma_map);
480 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
481 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
482 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
483 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
484 	if (!rc)
485 		*fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
486 
487 	return rc;
488 }
489 
490 /* Device */
491 
492 static bool is_bnxt_re_dev(struct net_device *netdev)
493 {
494 	struct ethtool_drvinfo drvinfo;
495 
496 	if (netdev->ethtool_ops && netdev->ethtool_ops->get_drvinfo) {
497 		memset(&drvinfo, 0, sizeof(drvinfo));
498 		netdev->ethtool_ops->get_drvinfo(netdev, &drvinfo);
499 
500 		if (strcmp(drvinfo.driver, "bnxt_en"))
501 			return false;
502 		return true;
503 	}
504 	return false;
505 }
506 
507 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
508 {
509 	struct bnxt_re_dev *rdev;
510 
511 	rcu_read_lock();
512 	list_for_each_entry_rcu(rdev, &bnxt_re_dev_list, list) {
513 		if (rdev->netdev == netdev) {
514 			rcu_read_unlock();
515 			return rdev;
516 		}
517 	}
518 	rcu_read_unlock();
519 	return NULL;
520 }
521 
522 static void bnxt_re_dev_unprobe(struct net_device *netdev,
523 				struct bnxt_en_dev *en_dev)
524 {
525 	dev_put(netdev);
526 	module_put(en_dev->pdev->driver->driver.owner);
527 }
528 
529 static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev)
530 {
531 	struct bnxt *bp = netdev_priv(netdev);
532 	struct bnxt_en_dev *en_dev;
533 	struct pci_dev *pdev;
534 
535 	/* Call bnxt_en's RoCE probe via indirect API */
536 	if (!bp->ulp_probe)
537 		return ERR_PTR(-EINVAL);
538 
539 	en_dev = bp->ulp_probe(netdev);
540 	if (IS_ERR(en_dev))
541 		return en_dev;
542 
543 	pdev = en_dev->pdev;
544 	if (!pdev)
545 		return ERR_PTR(-EINVAL);
546 
547 	if (!(en_dev->flags & BNXT_EN_FLAG_ROCE_CAP)) {
548 		dev_info(&pdev->dev,
549 			"%s: probe error: RoCE is not supported on this device",
550 			ROCE_DRV_MODULE_NAME);
551 		return ERR_PTR(-ENODEV);
552 	}
553 
554 	/* Bump net device reference count */
555 	if (!try_module_get(pdev->driver->driver.owner))
556 		return ERR_PTR(-ENODEV);
557 
558 	dev_hold(netdev);
559 
560 	return en_dev;
561 }
562 
563 static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr,
564 			   char *buf)
565 {
566 	struct bnxt_re_dev *rdev =
567 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
568 
569 	return scnprintf(buf, PAGE_SIZE, "0x%x\n", rdev->en_dev->pdev->vendor);
570 }
571 static DEVICE_ATTR_RO(hw_rev);
572 
573 static ssize_t hca_type_show(struct device *device,
574 			     struct device_attribute *attr, char *buf)
575 {
576 	struct bnxt_re_dev *rdev =
577 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
578 
579 	return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->ibdev.node_desc);
580 }
581 static DEVICE_ATTR_RO(hca_type);
582 
583 static struct attribute *bnxt_re_attributes[] = {
584 	&dev_attr_hw_rev.attr,
585 	&dev_attr_hca_type.attr,
586 	NULL
587 };
588 
589 static const struct attribute_group bnxt_re_dev_attr_group = {
590 	.attrs = bnxt_re_attributes,
591 };
592 
593 static void bnxt_re_unregister_ib(struct bnxt_re_dev *rdev)
594 {
595 	ib_unregister_device(&rdev->ibdev);
596 }
597 
598 static const struct ib_device_ops bnxt_re_dev_ops = {
599 	.owner = THIS_MODULE,
600 	.driver_id = RDMA_DRIVER_BNXT_RE,
601 	.uverbs_abi_ver = BNXT_RE_ABI_VERSION,
602 
603 	.add_gid = bnxt_re_add_gid,
604 	.alloc_hw_stats = bnxt_re_ib_alloc_hw_stats,
605 	.alloc_mr = bnxt_re_alloc_mr,
606 	.alloc_pd = bnxt_re_alloc_pd,
607 	.alloc_ucontext = bnxt_re_alloc_ucontext,
608 	.create_ah = bnxt_re_create_ah,
609 	.create_cq = bnxt_re_create_cq,
610 	.create_qp = bnxt_re_create_qp,
611 	.create_srq = bnxt_re_create_srq,
612 	.dealloc_pd = bnxt_re_dealloc_pd,
613 	.dealloc_ucontext = bnxt_re_dealloc_ucontext,
614 	.del_gid = bnxt_re_del_gid,
615 	.dereg_mr = bnxt_re_dereg_mr,
616 	.destroy_ah = bnxt_re_destroy_ah,
617 	.destroy_cq = bnxt_re_destroy_cq,
618 	.destroy_qp = bnxt_re_destroy_qp,
619 	.destroy_srq = bnxt_re_destroy_srq,
620 	.get_dev_fw_str = bnxt_re_query_fw_str,
621 	.get_dma_mr = bnxt_re_get_dma_mr,
622 	.get_hw_stats = bnxt_re_ib_get_hw_stats,
623 	.get_link_layer = bnxt_re_get_link_layer,
624 	.get_port_immutable = bnxt_re_get_port_immutable,
625 	.map_mr_sg = bnxt_re_map_mr_sg,
626 	.mmap = bnxt_re_mmap,
627 	.modify_ah = bnxt_re_modify_ah,
628 	.modify_device = bnxt_re_modify_device,
629 	.modify_qp = bnxt_re_modify_qp,
630 	.modify_srq = bnxt_re_modify_srq,
631 	.poll_cq = bnxt_re_poll_cq,
632 	.post_recv = bnxt_re_post_recv,
633 	.post_send = bnxt_re_post_send,
634 	.post_srq_recv = bnxt_re_post_srq_recv,
635 	.query_ah = bnxt_re_query_ah,
636 	.query_device = bnxt_re_query_device,
637 	.query_pkey = bnxt_re_query_pkey,
638 	.query_port = bnxt_re_query_port,
639 	.query_qp = bnxt_re_query_qp,
640 	.query_srq = bnxt_re_query_srq,
641 	.reg_user_mr = bnxt_re_reg_user_mr,
642 	.req_notify_cq = bnxt_re_req_notify_cq,
643 	INIT_RDMA_OBJ_SIZE(ib_ah, bnxt_re_ah, ib_ah),
644 	INIT_RDMA_OBJ_SIZE(ib_cq, bnxt_re_cq, ib_cq),
645 	INIT_RDMA_OBJ_SIZE(ib_pd, bnxt_re_pd, ib_pd),
646 	INIT_RDMA_OBJ_SIZE(ib_srq, bnxt_re_srq, ib_srq),
647 	INIT_RDMA_OBJ_SIZE(ib_ucontext, bnxt_re_ucontext, ib_uctx),
648 };
649 
650 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
651 {
652 	struct ib_device *ibdev = &rdev->ibdev;
653 	int ret;
654 
655 	/* ib device init */
656 	ibdev->node_type = RDMA_NODE_IB_CA;
657 	strlcpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
658 		strlen(BNXT_RE_DESC) + 5);
659 	ibdev->phys_port_cnt = 1;
660 
661 	bnxt_qplib_get_guid(rdev->netdev->dev_addr, (u8 *)&ibdev->node_guid);
662 
663 	ibdev->num_comp_vectors	= 1;
664 	ibdev->dev.parent = &rdev->en_dev->pdev->dev;
665 	ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
666 
667 	/* User space */
668 	ibdev->uverbs_cmd_mask =
669 			(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
670 			(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
671 			(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
672 			(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
673 			(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
674 			(1ull << IB_USER_VERBS_CMD_REG_MR)		|
675 			(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
676 			(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
677 			(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
678 			(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
679 			(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
680 			(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
681 			(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
682 			(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
683 			(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
684 			(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
685 			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
686 			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
687 			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
688 			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
689 			(1ull << IB_USER_VERBS_CMD_CREATE_AH)		|
690 			(1ull << IB_USER_VERBS_CMD_MODIFY_AH)		|
691 			(1ull << IB_USER_VERBS_CMD_QUERY_AH)		|
692 			(1ull << IB_USER_VERBS_CMD_DESTROY_AH);
693 	/* POLL_CQ and REQ_NOTIFY_CQ is directly handled in libbnxt_re */
694 
695 
696 	rdma_set_device_sysfs_group(ibdev, &bnxt_re_dev_attr_group);
697 	ib_set_device_ops(ibdev, &bnxt_re_dev_ops);
698 	ret = ib_device_set_netdev(&rdev->ibdev, rdev->netdev, 1);
699 	if (ret)
700 		return ret;
701 
702 	return ib_register_device(ibdev, "bnxt_re%d");
703 }
704 
705 static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev)
706 {
707 	dev_put(rdev->netdev);
708 	rdev->netdev = NULL;
709 
710 	mutex_lock(&bnxt_re_dev_lock);
711 	list_del_rcu(&rdev->list);
712 	mutex_unlock(&bnxt_re_dev_lock);
713 
714 	synchronize_rcu();
715 
716 	ib_dealloc_device(&rdev->ibdev);
717 	/* rdev is gone */
718 }
719 
720 static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev,
721 					   struct bnxt_en_dev *en_dev)
722 {
723 	struct bnxt_re_dev *rdev;
724 
725 	/* Allocate bnxt_re_dev instance here */
726 	rdev = ib_alloc_device(bnxt_re_dev, ibdev);
727 	if (!rdev) {
728 		dev_err(NULL, "%s: bnxt_re_dev allocation failure!",
729 			ROCE_DRV_MODULE_NAME);
730 		return NULL;
731 	}
732 	/* Default values */
733 	rdev->netdev = netdev;
734 	dev_hold(rdev->netdev);
735 	rdev->en_dev = en_dev;
736 	rdev->id = rdev->en_dev->pdev->devfn;
737 	INIT_LIST_HEAD(&rdev->qp_list);
738 	mutex_init(&rdev->qp_lock);
739 	atomic_set(&rdev->qp_count, 0);
740 	atomic_set(&rdev->cq_count, 0);
741 	atomic_set(&rdev->srq_count, 0);
742 	atomic_set(&rdev->mr_count, 0);
743 	atomic_set(&rdev->mw_count, 0);
744 	rdev->cosq[0] = 0xFFFF;
745 	rdev->cosq[1] = 0xFFFF;
746 
747 	mutex_lock(&bnxt_re_dev_lock);
748 	list_add_tail_rcu(&rdev->list, &bnxt_re_dev_list);
749 	mutex_unlock(&bnxt_re_dev_lock);
750 	return rdev;
751 }
752 
753 static int bnxt_re_handle_unaffi_async_event(struct creq_func_event
754 					     *unaffi_async)
755 {
756 	switch (unaffi_async->event) {
757 	case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
758 		break;
759 	case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
760 		break;
761 	case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
762 		break;
763 	case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
764 		break;
765 	case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
766 		break;
767 	case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
768 		break;
769 	case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
770 		break;
771 	case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
772 		break;
773 	case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
774 		break;
775 	case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
776 		break;
777 	case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
778 		break;
779 	default:
780 		return -EINVAL;
781 	}
782 	return 0;
783 }
784 
785 static int bnxt_re_handle_qp_async_event(struct creq_qp_event *qp_event,
786 					 struct bnxt_re_qp *qp)
787 {
788 	struct ib_event event;
789 	unsigned int flags;
790 
791 	if (qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
792 		flags = bnxt_re_lock_cqs(qp);
793 		bnxt_qplib_add_flush_qp(&qp->qplib_qp);
794 		bnxt_re_unlock_cqs(qp, flags);
795 	}
796 
797 	memset(&event, 0, sizeof(event));
798 	if (qp->qplib_qp.srq) {
799 		event.device = &qp->rdev->ibdev;
800 		event.element.qp = &qp->ib_qp;
801 		event.event = IB_EVENT_QP_LAST_WQE_REACHED;
802 	}
803 
804 	if (event.device && qp->ib_qp.event_handler)
805 		qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
806 
807 	return 0;
808 }
809 
810 static int bnxt_re_handle_affi_async_event(struct creq_qp_event *affi_async,
811 					   void *obj)
812 {
813 	int rc = 0;
814 	u8 event;
815 
816 	if (!obj)
817 		return rc; /* QP was already dead, still return success */
818 
819 	event = affi_async->event;
820 	if (event == CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION) {
821 		struct bnxt_qplib_qp *lib_qp = obj;
822 		struct bnxt_re_qp *qp = container_of(lib_qp, struct bnxt_re_qp,
823 						     qplib_qp);
824 		rc = bnxt_re_handle_qp_async_event(affi_async, qp);
825 	}
826 	return rc;
827 }
828 
829 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
830 			       void *aeqe, void *obj)
831 {
832 	struct creq_qp_event *affi_async;
833 	struct creq_func_event *unaffi_async;
834 	u8 type;
835 	int rc;
836 
837 	type = ((struct creq_base *)aeqe)->type;
838 	if (type == CREQ_BASE_TYPE_FUNC_EVENT) {
839 		unaffi_async = aeqe;
840 		rc = bnxt_re_handle_unaffi_async_event(unaffi_async);
841 	} else {
842 		affi_async = aeqe;
843 		rc = bnxt_re_handle_affi_async_event(affi_async, obj);
844 	}
845 
846 	return rc;
847 }
848 
849 static int bnxt_re_srqn_handler(struct bnxt_qplib_nq *nq,
850 				struct bnxt_qplib_srq *handle, u8 event)
851 {
852 	struct bnxt_re_srq *srq = container_of(handle, struct bnxt_re_srq,
853 					       qplib_srq);
854 	struct ib_event ib_event;
855 	int rc = 0;
856 
857 	if (!srq) {
858 		dev_err(NULL, "%s: SRQ is NULL, SRQN not handled",
859 			ROCE_DRV_MODULE_NAME);
860 		rc = -EINVAL;
861 		goto done;
862 	}
863 	ib_event.device = &srq->rdev->ibdev;
864 	ib_event.element.srq = &srq->ib_srq;
865 	if (event == NQ_SRQ_EVENT_EVENT_SRQ_THRESHOLD_EVENT)
866 		ib_event.event = IB_EVENT_SRQ_LIMIT_REACHED;
867 	else
868 		ib_event.event = IB_EVENT_SRQ_ERR;
869 
870 	if (srq->ib_srq.event_handler) {
871 		/* Lock event_handler? */
872 		(*srq->ib_srq.event_handler)(&ib_event,
873 					     srq->ib_srq.srq_context);
874 	}
875 done:
876 	return rc;
877 }
878 
879 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
880 			       struct bnxt_qplib_cq *handle)
881 {
882 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
883 					     qplib_cq);
884 
885 	if (!cq) {
886 		dev_err(NULL, "%s: CQ is NULL, CQN not handled",
887 			ROCE_DRV_MODULE_NAME);
888 		return -EINVAL;
889 	}
890 	if (cq->ib_cq.comp_handler) {
891 		/* Lock comp_handler? */
892 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
893 	}
894 
895 	return 0;
896 }
897 
898 static u32 bnxt_re_get_nqdb_offset(struct bnxt_re_dev *rdev, u16 indx)
899 {
900 	return bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx) ?
901 				0x10000 : rdev->msix_entries[indx].db_offset;
902 }
903 
904 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
905 {
906 	int i;
907 
908 	for (i = 1; i < rdev->num_msix; i++)
909 		bnxt_qplib_disable_nq(&rdev->nq[i - 1]);
910 
911 	if (rdev->qplib_res.rcfw)
912 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
913 }
914 
915 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
916 {
917 	int num_vec_enabled = 0;
918 	int rc = 0, i;
919 	u32 db_offt;
920 
921 	bnxt_qplib_init_res(&rdev->qplib_res);
922 
923 	for (i = 1; i < rdev->num_msix ; i++) {
924 		db_offt = bnxt_re_get_nqdb_offset(rdev, i);
925 		rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
926 					  i - 1, rdev->msix_entries[i].vector,
927 					  db_offt, &bnxt_re_cqn_handler,
928 					  &bnxt_re_srqn_handler);
929 		if (rc) {
930 			dev_err(rdev_to_dev(rdev),
931 				"Failed to enable NQ with rc = 0x%x", rc);
932 			goto fail;
933 		}
934 		num_vec_enabled++;
935 	}
936 	return 0;
937 fail:
938 	for (i = num_vec_enabled; i >= 0; i--)
939 		bnxt_qplib_disable_nq(&rdev->nq[i]);
940 	return rc;
941 }
942 
943 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev)
944 {
945 	u8 type;
946 	int i;
947 
948 	for (i = 0; i < rdev->num_msix - 1; i++) {
949 		type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
950 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
951 		rdev->nq[i].res = NULL;
952 		bnxt_qplib_free_nq(&rdev->nq[i]);
953 	}
954 }
955 
956 static void bnxt_re_free_res(struct bnxt_re_dev *rdev)
957 {
958 	bnxt_re_free_nq_res(rdev);
959 
960 	if (rdev->qplib_res.dpi_tbl.max) {
961 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
962 				       &rdev->qplib_res.dpi_tbl,
963 				       &rdev->dpi_privileged);
964 	}
965 	if (rdev->qplib_res.rcfw) {
966 		bnxt_qplib_free_res(&rdev->qplib_res);
967 		rdev->qplib_res.rcfw = NULL;
968 	}
969 }
970 
971 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
972 {
973 	int num_vec_created = 0;
974 	dma_addr_t *pg_map;
975 	int rc = 0, i;
976 	int pages;
977 	u8 type;
978 
979 	/* Configure and allocate resources for qplib */
980 	rdev->qplib_res.rcfw = &rdev->rcfw;
981 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
982 				     rdev->is_virtfn);
983 	if (rc)
984 		goto fail;
985 
986 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
987 				  rdev->netdev, &rdev->dev_attr);
988 	if (rc)
989 		goto fail;
990 
991 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
992 				  &rdev->dpi_privileged,
993 				  rdev);
994 	if (rc)
995 		goto dealloc_res;
996 
997 	for (i = 0; i < rdev->num_msix - 1; i++) {
998 		rdev->nq[i].res = &rdev->qplib_res;
999 		rdev->nq[i].hwq.max_elements = BNXT_RE_MAX_CQ_COUNT +
1000 			BNXT_RE_MAX_SRQC_COUNT + 2;
1001 		rc = bnxt_qplib_alloc_nq(rdev->en_dev->pdev, &rdev->nq[i]);
1002 		if (rc) {
1003 			dev_err(rdev_to_dev(rdev), "Alloc Failed NQ%d rc:%#x",
1004 				i, rc);
1005 			goto free_nq;
1006 		}
1007 		type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1008 		pg_map = rdev->nq[i].hwq.pbl[PBL_LVL_0].pg_map_arr;
1009 		pages = rdev->nq[i].hwq.pbl[rdev->nq[i].hwq.level].pg_count;
1010 		rc = bnxt_re_net_ring_alloc(rdev, pg_map, pages, type,
1011 					    BNXT_QPLIB_NQE_MAX_CNT - 1,
1012 					    rdev->msix_entries[i + 1].ring_idx,
1013 					    &rdev->nq[i].ring_id);
1014 		if (rc) {
1015 			dev_err(rdev_to_dev(rdev),
1016 				"Failed to allocate NQ fw id with rc = 0x%x",
1017 				rc);
1018 			bnxt_qplib_free_nq(&rdev->nq[i]);
1019 			goto free_nq;
1020 		}
1021 		num_vec_created++;
1022 	}
1023 	return 0;
1024 free_nq:
1025 	for (i = num_vec_created; i >= 0; i--) {
1026 		type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1027 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
1028 		bnxt_qplib_free_nq(&rdev->nq[i]);
1029 	}
1030 	bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
1031 			       &rdev->qplib_res.dpi_tbl,
1032 			       &rdev->dpi_privileged);
1033 dealloc_res:
1034 	bnxt_qplib_free_res(&rdev->qplib_res);
1035 
1036 fail:
1037 	rdev->qplib_res.rcfw = NULL;
1038 	return rc;
1039 }
1040 
1041 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
1042 				   u8 port_num, enum ib_event_type event)
1043 {
1044 	struct ib_event ib_event;
1045 
1046 	ib_event.device = ibdev;
1047 	if (qp) {
1048 		ib_event.element.qp = qp;
1049 		ib_event.event = event;
1050 		if (qp->event_handler)
1051 			qp->event_handler(&ib_event, qp->qp_context);
1052 
1053 	} else {
1054 		ib_event.element.port_num = port_num;
1055 		ib_event.event = event;
1056 		ib_dispatch_event(&ib_event);
1057 	}
1058 }
1059 
1060 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN      0x02
1061 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
1062 				      u64 *cid_map)
1063 {
1064 	struct hwrm_queue_pri2cos_qcfg_input req = {0};
1065 	struct bnxt *bp = netdev_priv(rdev->netdev);
1066 	struct hwrm_queue_pri2cos_qcfg_output resp;
1067 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1068 	struct bnxt_fw_msg fw_msg;
1069 	u32 flags = 0;
1070 	u8 *qcfgmap, *tmp_map;
1071 	int rc = 0, i;
1072 
1073 	if (!cid_map)
1074 		return -EINVAL;
1075 
1076 	memset(&fw_msg, 0, sizeof(fw_msg));
1077 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1078 			      HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
1079 	flags |= (dir & 0x01);
1080 	flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
1081 	req.flags = cpu_to_le32(flags);
1082 	req.port_id = bp->pf.port_id;
1083 
1084 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1085 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1086 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1087 	if (rc)
1088 		return rc;
1089 
1090 	if (resp.queue_cfg_info) {
1091 		dev_warn(rdev_to_dev(rdev),
1092 			 "Asymmetric cos queue configuration detected");
1093 		dev_warn(rdev_to_dev(rdev),
1094 			 " on device, QoS may not be fully functional\n");
1095 	}
1096 	qcfgmap = &resp.pri0_cos_queue_id;
1097 	tmp_map = (u8 *)cid_map;
1098 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
1099 		tmp_map[i] = qcfgmap[i];
1100 
1101 	return rc;
1102 }
1103 
1104 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
1105 					struct bnxt_re_qp *qp)
1106 {
1107 	return (qp->ib_qp.qp_type == IB_QPT_GSI) || (qp == rdev->qp1_sqp);
1108 }
1109 
1110 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
1111 {
1112 	int mask = IB_QP_STATE;
1113 	struct ib_qp_attr qp_attr;
1114 	struct bnxt_re_qp *qp;
1115 
1116 	qp_attr.qp_state = IB_QPS_ERR;
1117 	mutex_lock(&rdev->qp_lock);
1118 	list_for_each_entry(qp, &rdev->qp_list, list) {
1119 		/* Modify the state of all QPs except QP1/Shadow QP */
1120 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
1121 			if (qp->qplib_qp.state !=
1122 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
1123 			    qp->qplib_qp.state !=
1124 			    CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1125 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
1126 						       1, IB_EVENT_QP_FATAL);
1127 				bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
1128 						  NULL);
1129 			}
1130 		}
1131 	}
1132 	mutex_unlock(&rdev->qp_lock);
1133 }
1134 
1135 static int bnxt_re_update_gid(struct bnxt_re_dev *rdev)
1136 {
1137 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
1138 	struct bnxt_qplib_gid gid;
1139 	u16 gid_idx, index;
1140 	int rc = 0;
1141 
1142 	if (!test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1143 		return 0;
1144 
1145 	if (!sgid_tbl) {
1146 		dev_err(rdev_to_dev(rdev), "QPLIB: SGID table not allocated");
1147 		return -EINVAL;
1148 	}
1149 
1150 	for (index = 0; index < sgid_tbl->active; index++) {
1151 		gid_idx = sgid_tbl->hw_id[index];
1152 
1153 		if (!memcmp(&sgid_tbl->tbl[index], &bnxt_qplib_gid_zero,
1154 			    sizeof(bnxt_qplib_gid_zero)))
1155 			continue;
1156 		/* need to modify the VLAN enable setting of non VLAN GID only
1157 		 * as setting is done for VLAN GID while adding GID
1158 		 */
1159 		if (sgid_tbl->vlan[index])
1160 			continue;
1161 
1162 		memcpy(&gid, &sgid_tbl->tbl[index], sizeof(gid));
1163 
1164 		rc = bnxt_qplib_update_sgid(sgid_tbl, &gid, gid_idx,
1165 					    rdev->qplib_res.netdev->dev_addr);
1166 	}
1167 
1168 	return rc;
1169 }
1170 
1171 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
1172 {
1173 	u32 prio_map = 0, tmp_map = 0;
1174 	struct net_device *netdev;
1175 	struct dcb_app app;
1176 
1177 	netdev = rdev->netdev;
1178 
1179 	memset(&app, 0, sizeof(app));
1180 	app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
1181 	app.protocol = ETH_P_IBOE;
1182 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1183 	prio_map = tmp_map;
1184 
1185 	app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
1186 	app.protocol = ROCE_V2_UDP_DPORT;
1187 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1188 	prio_map |= tmp_map;
1189 
1190 	return prio_map;
1191 }
1192 
1193 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
1194 {
1195 	u16 prio;
1196 	u8 id;
1197 
1198 	for (prio = 0, id = 0; prio < 8; prio++) {
1199 		if (prio_map & (1 << prio)) {
1200 			cosq[id] = cid_map[prio];
1201 			id++;
1202 			if (id == 2) /* Max 2 tcs supported */
1203 				break;
1204 		}
1205 	}
1206 }
1207 
1208 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
1209 {
1210 	u8 prio_map = 0;
1211 	u64 cid_map;
1212 	int rc;
1213 
1214 	/* Get priority for roce */
1215 	prio_map = bnxt_re_get_priority_mask(rdev);
1216 
1217 	if (prio_map == rdev->cur_prio_map)
1218 		return 0;
1219 	rdev->cur_prio_map = prio_map;
1220 	/* Get cosq id for this priority */
1221 	rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
1222 	if (rc) {
1223 		dev_warn(rdev_to_dev(rdev), "no cos for p_mask %x\n", prio_map);
1224 		return rc;
1225 	}
1226 	/* Parse CoS IDs for app priority */
1227 	bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
1228 
1229 	/* Config BONO. */
1230 	rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
1231 	if (rc) {
1232 		dev_warn(rdev_to_dev(rdev), "no tc for cos{%x, %x}\n",
1233 			 rdev->cosq[0], rdev->cosq[1]);
1234 		return rc;
1235 	}
1236 
1237 	/* Actual priorities are not programmed as they are already
1238 	 * done by L2 driver; just enable or disable priority vlan tagging
1239 	 */
1240 	if ((prio_map == 0 && rdev->qplib_res.prio) ||
1241 	    (prio_map != 0 && !rdev->qplib_res.prio)) {
1242 		rdev->qplib_res.prio = prio_map ? true : false;
1243 
1244 		bnxt_re_update_gid(rdev);
1245 	}
1246 
1247 	return 0;
1248 }
1249 
1250 static void bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev)
1251 {
1252 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1253 	struct hwrm_ver_get_output resp = {0};
1254 	struct hwrm_ver_get_input req = {0};
1255 	struct bnxt_fw_msg fw_msg;
1256 	int rc = 0;
1257 
1258 	memset(&fw_msg, 0, sizeof(fw_msg));
1259 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1260 			      HWRM_VER_GET, -1, -1);
1261 	req.hwrm_intf_maj = HWRM_VERSION_MAJOR;
1262 	req.hwrm_intf_min = HWRM_VERSION_MINOR;
1263 	req.hwrm_intf_upd = HWRM_VERSION_UPDATE;
1264 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1265 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1266 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1267 	if (rc) {
1268 		dev_err(rdev_to_dev(rdev),
1269 			"Failed to query HW version, rc = 0x%x", rc);
1270 		return;
1271 	}
1272 	rdev->qplib_ctx.hwrm_intf_ver =
1273 		(u64)resp.hwrm_intf_major << 48 |
1274 		(u64)resp.hwrm_intf_minor << 32 |
1275 		(u64)resp.hwrm_intf_build << 16 |
1276 		resp.hwrm_intf_patch;
1277 }
1278 
1279 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev)
1280 {
1281 	u8 type;
1282 	int rc;
1283 
1284 	if (test_and_clear_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags)) {
1285 		/* Cleanup ib dev */
1286 		bnxt_re_unregister_ib(rdev);
1287 	}
1288 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
1289 		cancel_delayed_work_sync(&rdev->worker);
1290 
1291 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED,
1292 			       &rdev->flags))
1293 		bnxt_re_cleanup_res(rdev);
1294 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags))
1295 		bnxt_re_free_res(rdev);
1296 
1297 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
1298 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
1299 		if (rc)
1300 			dev_warn(rdev_to_dev(rdev),
1301 				 "Failed to deinitialize RCFW: %#x", rc);
1302 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1303 		bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1304 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1305 		type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1306 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, type);
1307 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1308 	}
1309 	if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
1310 		rc = bnxt_re_free_msix(rdev);
1311 		if (rc)
1312 			dev_warn(rdev_to_dev(rdev),
1313 				 "Failed to free MSI-X vectors: %#x", rc);
1314 	}
1315 
1316 	bnxt_re_destroy_chip_ctx(rdev);
1317 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
1318 		rc = bnxt_re_unregister_netdev(rdev);
1319 		if (rc)
1320 			dev_warn(rdev_to_dev(rdev),
1321 				 "Failed to unregister with netdev: %#x", rc);
1322 	}
1323 }
1324 
1325 /* worker thread for polling periodic events. Now used for QoS programming*/
1326 static void bnxt_re_worker(struct work_struct *work)
1327 {
1328 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
1329 						worker.work);
1330 
1331 	bnxt_re_setup_qos(rdev);
1332 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1333 }
1334 
1335 static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
1336 {
1337 	dma_addr_t *pg_map;
1338 	u32 db_offt, ridx;
1339 	int pages, vid;
1340 	bool locked;
1341 	u8 type;
1342 	int rc;
1343 
1344 	/* Acquire rtnl lock through out this function */
1345 	rtnl_lock();
1346 	locked = true;
1347 
1348 	/* Registered a new RoCE device instance to netdev */
1349 	rc = bnxt_re_register_netdev(rdev);
1350 	if (rc) {
1351 		rtnl_unlock();
1352 		pr_err("Failed to register with netedev: %#x\n", rc);
1353 		return -EINVAL;
1354 	}
1355 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1356 
1357 	rc = bnxt_re_setup_chip_ctx(rdev);
1358 	if (rc) {
1359 		dev_err(rdev_to_dev(rdev), "Failed to get chip context\n");
1360 		return -EINVAL;
1361 	}
1362 
1363 	/* Check whether VF or PF */
1364 	bnxt_re_get_sriov_func_type(rdev);
1365 
1366 	rc = bnxt_re_request_msix(rdev);
1367 	if (rc) {
1368 		pr_err("Failed to get MSI-X vectors: %#x\n", rc);
1369 		rc = -EINVAL;
1370 		goto fail;
1371 	}
1372 	set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
1373 
1374 	bnxt_re_query_hwrm_intf_version(rdev);
1375 
1376 	/* Establish RCFW Communication Channel to initialize the context
1377 	 * memory for the function and all child VFs
1378 	 */
1379 	rc = bnxt_qplib_alloc_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw,
1380 					   &rdev->qplib_ctx,
1381 					   BNXT_RE_MAX_QPC_COUNT);
1382 	if (rc) {
1383 		pr_err("Failed to allocate RCFW Channel: %#x\n", rc);
1384 		goto fail;
1385 	}
1386 	type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1387 	pg_map = rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr;
1388 	pages = rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count;
1389 	ridx = rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx;
1390 	rc = bnxt_re_net_ring_alloc(rdev, pg_map, pages, type,
1391 				    BNXT_QPLIB_CREQE_MAX_CNT - 1,
1392 				    ridx, &rdev->rcfw.creq_ring_id);
1393 	if (rc) {
1394 		pr_err("Failed to allocate CREQ: %#x\n", rc);
1395 		goto free_rcfw;
1396 	}
1397 	db_offt = bnxt_re_get_nqdb_offset(rdev, BNXT_RE_AEQ_IDX);
1398 	vid = rdev->msix_entries[BNXT_RE_AEQ_IDX].vector;
1399 	rc = bnxt_qplib_enable_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw,
1400 					    vid, db_offt, rdev->is_virtfn,
1401 					    &bnxt_re_aeq_handler);
1402 	if (rc) {
1403 		pr_err("Failed to enable RCFW channel: %#x\n", rc);
1404 		goto free_ring;
1405 	}
1406 
1407 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
1408 				     rdev->is_virtfn);
1409 	if (rc)
1410 		goto disable_rcfw;
1411 	if (!rdev->is_virtfn)
1412 		bnxt_re_set_resource_limits(rdev);
1413 
1414 	rc = bnxt_qplib_alloc_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx, 0,
1415 				  bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx));
1416 	if (rc) {
1417 		pr_err("Failed to allocate QPLIB context: %#x\n", rc);
1418 		goto disable_rcfw;
1419 	}
1420 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
1421 					 rdev->qplib_ctx.stats.dma_map,
1422 					 &rdev->qplib_ctx.stats.fw_id);
1423 	if (rc) {
1424 		pr_err("Failed to allocate stats context: %#x\n", rc);
1425 		goto free_ctx;
1426 	}
1427 
1428 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx,
1429 				  rdev->is_virtfn);
1430 	if (rc) {
1431 		pr_err("Failed to initialize RCFW: %#x\n", rc);
1432 		goto free_sctx;
1433 	}
1434 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1435 
1436 	/* Resources based on the 'new' device caps */
1437 	rc = bnxt_re_alloc_res(rdev);
1438 	if (rc) {
1439 		pr_err("Failed to allocate resources: %#x\n", rc);
1440 		goto fail;
1441 	}
1442 	set_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags);
1443 	rc = bnxt_re_init_res(rdev);
1444 	if (rc) {
1445 		pr_err("Failed to initialize resources: %#x\n", rc);
1446 		goto fail;
1447 	}
1448 
1449 	set_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED, &rdev->flags);
1450 
1451 	if (!rdev->is_virtfn) {
1452 		rc = bnxt_re_setup_qos(rdev);
1453 		if (rc)
1454 			pr_info("RoCE priority not yet configured\n");
1455 
1456 		INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1457 		set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1458 		schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1459 	}
1460 
1461 	rtnl_unlock();
1462 	locked = false;
1463 
1464 	/* Register ib dev */
1465 	rc = bnxt_re_register_ib(rdev);
1466 	if (rc) {
1467 		pr_err("Failed to register with IB: %#x\n", rc);
1468 		goto fail;
1469 	}
1470 	set_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags);
1471 	dev_info(rdev_to_dev(rdev), "Device registered successfully");
1472 	ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1473 			 &rdev->active_width);
1474 	set_bit(BNXT_RE_FLAG_ISSUE_ROCE_STATS, &rdev->flags);
1475 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_PORT_ACTIVE);
1476 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_GID_CHANGE);
1477 
1478 	return 0;
1479 free_sctx:
1480 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1481 free_ctx:
1482 	bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1483 disable_rcfw:
1484 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1485 free_ring:
1486 	type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1487 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, type);
1488 free_rcfw:
1489 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1490 fail:
1491 	if (!locked)
1492 		rtnl_lock();
1493 	bnxt_re_ib_unreg(rdev);
1494 	rtnl_unlock();
1495 
1496 	return rc;
1497 }
1498 
1499 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1500 {
1501 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1502 	struct net_device *netdev = rdev->netdev;
1503 
1504 	bnxt_re_dev_remove(rdev);
1505 
1506 	if (netdev)
1507 		bnxt_re_dev_unprobe(netdev, en_dev);
1508 }
1509 
1510 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1511 {
1512 	struct bnxt_en_dev *en_dev;
1513 	int rc = 0;
1514 
1515 	if (!is_bnxt_re_dev(netdev))
1516 		return -ENODEV;
1517 
1518 	en_dev = bnxt_re_dev_probe(netdev);
1519 	if (IS_ERR(en_dev)) {
1520 		if (en_dev != ERR_PTR(-ENODEV))
1521 			pr_err("%s: Failed to probe\n", ROCE_DRV_MODULE_NAME);
1522 		rc = PTR_ERR(en_dev);
1523 		goto exit;
1524 	}
1525 	*rdev = bnxt_re_dev_add(netdev, en_dev);
1526 	if (!*rdev) {
1527 		rc = -ENOMEM;
1528 		bnxt_re_dev_unprobe(netdev, en_dev);
1529 		goto exit;
1530 	}
1531 exit:
1532 	return rc;
1533 }
1534 
1535 static void bnxt_re_remove_one(struct bnxt_re_dev *rdev)
1536 {
1537 	pci_dev_put(rdev->en_dev->pdev);
1538 }
1539 
1540 /* Handle all deferred netevents tasks */
1541 static void bnxt_re_task(struct work_struct *work)
1542 {
1543 	struct bnxt_re_work *re_work;
1544 	struct bnxt_re_dev *rdev;
1545 	int rc = 0;
1546 
1547 	re_work = container_of(work, struct bnxt_re_work, work);
1548 	rdev = re_work->rdev;
1549 
1550 	if (re_work->event != NETDEV_REGISTER &&
1551 	    !test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1552 		return;
1553 
1554 	switch (re_work->event) {
1555 	case NETDEV_REGISTER:
1556 		rc = bnxt_re_ib_reg(rdev);
1557 		if (rc) {
1558 			dev_err(rdev_to_dev(rdev),
1559 				"Failed to register with IB: %#x", rc);
1560 			bnxt_re_remove_one(rdev);
1561 			bnxt_re_dev_unreg(rdev);
1562 			goto exit;
1563 		}
1564 		break;
1565 	case NETDEV_UP:
1566 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1567 				       IB_EVENT_PORT_ACTIVE);
1568 		break;
1569 	case NETDEV_DOWN:
1570 		bnxt_re_dev_stop(rdev);
1571 		break;
1572 	case NETDEV_CHANGE:
1573 		if (!netif_carrier_ok(rdev->netdev))
1574 			bnxt_re_dev_stop(rdev);
1575 		else if (netif_carrier_ok(rdev->netdev))
1576 			bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1577 					       IB_EVENT_PORT_ACTIVE);
1578 		ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1579 				 &rdev->active_width);
1580 		break;
1581 	default:
1582 		break;
1583 	}
1584 	smp_mb__before_atomic();
1585 	atomic_dec(&rdev->sched_count);
1586 exit:
1587 	kfree(re_work);
1588 }
1589 
1590 static void bnxt_re_init_one(struct bnxt_re_dev *rdev)
1591 {
1592 	pci_dev_get(rdev->en_dev->pdev);
1593 }
1594 
1595 /*
1596  * "Notifier chain callback can be invoked for the same chain from
1597  * different CPUs at the same time".
1598  *
1599  * For cases when the netdev is already present, our call to the
1600  * register_netdevice_notifier() will actually get the rtnl_lock()
1601  * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1602  * events.
1603  *
1604  * But for cases when the netdev is not already present, the notifier
1605  * chain is subjected to be invoked from different CPUs simultaneously.
1606  *
1607  * This is protected by the netdev_mutex.
1608  */
1609 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1610 				unsigned long event, void *ptr)
1611 {
1612 	struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1613 	struct bnxt_re_work *re_work;
1614 	struct bnxt_re_dev *rdev;
1615 	int rc = 0;
1616 	bool sch_work = false;
1617 
1618 	real_dev = rdma_vlan_dev_real_dev(netdev);
1619 	if (!real_dev)
1620 		real_dev = netdev;
1621 
1622 	rdev = bnxt_re_from_netdev(real_dev);
1623 	if (!rdev && event != NETDEV_REGISTER)
1624 		goto exit;
1625 	if (real_dev != netdev)
1626 		goto exit;
1627 
1628 	switch (event) {
1629 	case NETDEV_REGISTER:
1630 		if (rdev)
1631 			break;
1632 		rc = bnxt_re_dev_reg(&rdev, real_dev);
1633 		if (rc == -ENODEV)
1634 			break;
1635 		if (rc) {
1636 			pr_err("Failed to register with the device %s: %#x\n",
1637 			       real_dev->name, rc);
1638 			break;
1639 		}
1640 		bnxt_re_init_one(rdev);
1641 		sch_work = true;
1642 		break;
1643 
1644 	case NETDEV_UNREGISTER:
1645 		/* netdev notifier will call NETDEV_UNREGISTER again later since
1646 		 * we are still holding the reference to the netdev
1647 		 */
1648 		if (atomic_read(&rdev->sched_count) > 0)
1649 			goto exit;
1650 		bnxt_re_ib_unreg(rdev);
1651 		bnxt_re_remove_one(rdev);
1652 		bnxt_re_dev_unreg(rdev);
1653 		break;
1654 
1655 	default:
1656 		sch_work = true;
1657 		break;
1658 	}
1659 	if (sch_work) {
1660 		/* Allocate for the deferred task */
1661 		re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1662 		if (re_work) {
1663 			re_work->rdev = rdev;
1664 			re_work->event = event;
1665 			re_work->vlan_dev = (real_dev == netdev ?
1666 					     NULL : netdev);
1667 			INIT_WORK(&re_work->work, bnxt_re_task);
1668 			atomic_inc(&rdev->sched_count);
1669 			queue_work(bnxt_re_wq, &re_work->work);
1670 		}
1671 	}
1672 
1673 exit:
1674 	return NOTIFY_DONE;
1675 }
1676 
1677 static struct notifier_block bnxt_re_netdev_notifier = {
1678 	.notifier_call = bnxt_re_netdev_event
1679 };
1680 
1681 static int __init bnxt_re_mod_init(void)
1682 {
1683 	int rc = 0;
1684 
1685 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1686 
1687 	bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1688 	if (!bnxt_re_wq)
1689 		return -ENOMEM;
1690 
1691 	INIT_LIST_HEAD(&bnxt_re_dev_list);
1692 
1693 	rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1694 	if (rc) {
1695 		pr_err("%s: Cannot register to netdevice_notifier",
1696 		       ROCE_DRV_MODULE_NAME);
1697 		goto err_netdev;
1698 	}
1699 	return 0;
1700 
1701 err_netdev:
1702 	destroy_workqueue(bnxt_re_wq);
1703 
1704 	return rc;
1705 }
1706 
1707 static void __exit bnxt_re_mod_exit(void)
1708 {
1709 	struct bnxt_re_dev *rdev, *next;
1710 	LIST_HEAD(to_be_deleted);
1711 
1712 	mutex_lock(&bnxt_re_dev_lock);
1713 	/* Free all adapter allocated resources */
1714 	if (!list_empty(&bnxt_re_dev_list))
1715 		list_splice_init(&bnxt_re_dev_list, &to_be_deleted);
1716 	mutex_unlock(&bnxt_re_dev_lock);
1717        /*
1718 	* Cleanup the devices in reverse order so that the VF device
1719 	* cleanup is done before PF cleanup
1720 	*/
1721 	list_for_each_entry_safe_reverse(rdev, next, &to_be_deleted, list) {
1722 		dev_info(rdev_to_dev(rdev), "Unregistering Device");
1723 		/*
1724 		 * Flush out any scheduled tasks before destroying the
1725 		 * resources
1726 		 */
1727 		flush_workqueue(bnxt_re_wq);
1728 		bnxt_re_dev_stop(rdev);
1729 		/* Acquire the rtnl_lock as the L2 resources are freed here */
1730 		rtnl_lock();
1731 		bnxt_re_ib_unreg(rdev);
1732 		rtnl_unlock();
1733 		bnxt_re_remove_one(rdev);
1734 		bnxt_re_dev_unreg(rdev);
1735 	}
1736 	unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1737 	if (bnxt_re_wq)
1738 		destroy_workqueue(bnxt_re_wq);
1739 }
1740 
1741 module_init(bnxt_re_mod_init);
1742 module_exit(bnxt_re_mod_exit);
1743