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 static char version[] =
68 		BNXT_RE_DESC " v" ROCE_DRV_MODULE_VERSION "\n";
69 
70 MODULE_AUTHOR("Eddie Wai <eddie.wai@broadcom.com>");
71 MODULE_DESCRIPTION(BNXT_RE_DESC " Driver");
72 MODULE_LICENSE("Dual BSD/GPL");
73 MODULE_VERSION(ROCE_DRV_MODULE_VERSION);
74 
75 /* globals */
76 static struct list_head bnxt_re_dev_list = LIST_HEAD_INIT(bnxt_re_dev_list);
77 /* Mutex to protect the list of bnxt_re devices added */
78 static DEFINE_MUTEX(bnxt_re_dev_lock);
79 static struct workqueue_struct *bnxt_re_wq;
80 
81 /* for handling bnxt_en callbacks later */
82 static void bnxt_re_stop(void *p)
83 {
84 }
85 
86 static void bnxt_re_start(void *p)
87 {
88 }
89 
90 static void bnxt_re_sriov_config(void *p, int num_vfs)
91 {
92 }
93 
94 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
95 	.ulp_async_notifier = NULL,
96 	.ulp_stop = bnxt_re_stop,
97 	.ulp_start = bnxt_re_start,
98 	.ulp_sriov_config = bnxt_re_sriov_config
99 };
100 
101 /* RoCE -> Net driver */
102 
103 /* Driver registration routines used to let the networking driver (bnxt_en)
104  * to know that the RoCE driver is now installed
105  */
106 static int bnxt_re_unregister_netdev(struct bnxt_re_dev *rdev, bool lock_wait)
107 {
108 	struct bnxt_en_dev *en_dev;
109 	int rc;
110 
111 	if (!rdev)
112 		return -EINVAL;
113 
114 	en_dev = rdev->en_dev;
115 	/* Acquire rtnl lock if it is not invokded from netdev event */
116 	if (lock_wait)
117 		rtnl_lock();
118 
119 	rc = en_dev->en_ops->bnxt_unregister_device(rdev->en_dev,
120 						    BNXT_ROCE_ULP);
121 	if (lock_wait)
122 		rtnl_unlock();
123 	return rc;
124 }
125 
126 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
127 {
128 	struct bnxt_en_dev *en_dev;
129 	int rc = 0;
130 
131 	if (!rdev)
132 		return -EINVAL;
133 
134 	en_dev = rdev->en_dev;
135 
136 	rtnl_lock();
137 	rc = en_dev->en_ops->bnxt_register_device(en_dev, BNXT_ROCE_ULP,
138 						  &bnxt_re_ulp_ops, rdev);
139 	rtnl_unlock();
140 	return rc;
141 }
142 
143 static int bnxt_re_free_msix(struct bnxt_re_dev *rdev, bool lock_wait)
144 {
145 	struct bnxt_en_dev *en_dev;
146 	int rc;
147 
148 	if (!rdev)
149 		return -EINVAL;
150 
151 	en_dev = rdev->en_dev;
152 
153 	if (lock_wait)
154 		rtnl_lock();
155 
156 	rc = en_dev->en_ops->bnxt_free_msix(rdev->en_dev, BNXT_ROCE_ULP);
157 
158 	if (lock_wait)
159 		rtnl_unlock();
160 	return rc;
161 }
162 
163 static int bnxt_re_request_msix(struct bnxt_re_dev *rdev)
164 {
165 	int rc = 0, num_msix_want = BNXT_RE_MIN_MSIX, num_msix_got;
166 	struct bnxt_en_dev *en_dev;
167 
168 	if (!rdev)
169 		return -EINVAL;
170 
171 	en_dev = rdev->en_dev;
172 
173 	rtnl_lock();
174 	num_msix_got = en_dev->en_ops->bnxt_request_msix(en_dev, BNXT_ROCE_ULP,
175 							 rdev->msix_entries,
176 							 num_msix_want);
177 	if (num_msix_got < BNXT_RE_MIN_MSIX) {
178 		rc = -EINVAL;
179 		goto done;
180 	}
181 	if (num_msix_got != num_msix_want) {
182 		dev_warn(rdev_to_dev(rdev),
183 			 "Requested %d MSI-X vectors, got %d\n",
184 			 num_msix_want, num_msix_got);
185 	}
186 	rdev->num_msix = num_msix_got;
187 done:
188 	rtnl_unlock();
189 	return rc;
190 }
191 
192 static void bnxt_re_init_hwrm_hdr(struct bnxt_re_dev *rdev, struct input *hdr,
193 				  u16 opcd, u16 crid, u16 trid)
194 {
195 	hdr->req_type = cpu_to_le16(opcd);
196 	hdr->cmpl_ring = cpu_to_le16(crid);
197 	hdr->target_id = cpu_to_le16(trid);
198 }
199 
200 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
201 				int msg_len, void *resp, int resp_max_len,
202 				int timeout)
203 {
204 	fw_msg->msg = msg;
205 	fw_msg->msg_len = msg_len;
206 	fw_msg->resp = resp;
207 	fw_msg->resp_max_len = resp_max_len;
208 	fw_msg->timeout = timeout;
209 }
210 
211 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev, u16 fw_ring_id,
212 				 bool lock_wait)
213 {
214 	struct bnxt_en_dev *en_dev = rdev->en_dev;
215 	struct hwrm_ring_free_input req = {0};
216 	struct hwrm_ring_free_output resp;
217 	struct bnxt_fw_msg fw_msg;
218 	bool do_unlock = false;
219 	int rc = -EINVAL;
220 
221 	if (!en_dev)
222 		return rc;
223 
224 	memset(&fw_msg, 0, sizeof(fw_msg));
225 	if (lock_wait) {
226 		rtnl_lock();
227 		do_unlock = true;
228 	}
229 
230 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1);
231 	req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
232 	req.ring_id = cpu_to_le16(fw_ring_id);
233 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
234 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
235 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
236 	if (rc)
237 		dev_err(rdev_to_dev(rdev),
238 			"Failed to free HW ring:%d :%#x", req.ring_id, rc);
239 	if (do_unlock)
240 		rtnl_unlock();
241 	return rc;
242 }
243 
244 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev, dma_addr_t *dma_arr,
245 				  int pages, int type, u32 ring_mask,
246 				  u32 map_index, u16 *fw_ring_id)
247 {
248 	struct bnxt_en_dev *en_dev = rdev->en_dev;
249 	struct hwrm_ring_alloc_input req = {0};
250 	struct hwrm_ring_alloc_output resp;
251 	struct bnxt_fw_msg fw_msg;
252 	int rc = -EINVAL;
253 
254 	if (!en_dev)
255 		return rc;
256 
257 	memset(&fw_msg, 0, sizeof(fw_msg));
258 	rtnl_lock();
259 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_ALLOC, -1, -1);
260 	req.enables = 0;
261 	req.page_tbl_addr =  cpu_to_le64(dma_arr[0]);
262 	if (pages > 1) {
263 		/* Page size is in log2 units */
264 		req.page_size = BNXT_PAGE_SHIFT;
265 		req.page_tbl_depth = 1;
266 	}
267 	req.fbo = 0;
268 	/* Association of ring index with doorbell index and MSIX number */
269 	req.logical_id = cpu_to_le16(map_index);
270 	req.length = cpu_to_le32(ring_mask + 1);
271 	req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
272 	req.int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
273 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
274 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
275 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
276 	if (!rc)
277 		*fw_ring_id = le16_to_cpu(resp.ring_id);
278 
279 	rtnl_unlock();
280 	return rc;
281 }
282 
283 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
284 				      u32 fw_stats_ctx_id, bool lock_wait)
285 {
286 	struct bnxt_en_dev *en_dev = rdev->en_dev;
287 	struct hwrm_stat_ctx_free_input req = {0};
288 	struct bnxt_fw_msg fw_msg;
289 	bool do_unlock = false;
290 	int rc = -EINVAL;
291 
292 	if (!en_dev)
293 		return rc;
294 
295 	memset(&fw_msg, 0, sizeof(fw_msg));
296 	if (lock_wait) {
297 		rtnl_lock();
298 		do_unlock = true;
299 	}
300 
301 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_FREE, -1, -1);
302 	req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
303 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&req,
304 			    sizeof(req), DFLT_HWRM_CMD_TIMEOUT);
305 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
306 	if (rc)
307 		dev_err(rdev_to_dev(rdev),
308 			"Failed to free HW stats context %#x", rc);
309 
310 	if (do_unlock)
311 		rtnl_unlock();
312 	return rc;
313 }
314 
315 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
316 				       dma_addr_t dma_map,
317 				       u32 *fw_stats_ctx_id)
318 {
319 	struct hwrm_stat_ctx_alloc_output resp = {0};
320 	struct hwrm_stat_ctx_alloc_input req = {0};
321 	struct bnxt_en_dev *en_dev = rdev->en_dev;
322 	struct bnxt_fw_msg fw_msg;
323 	int rc = -EINVAL;
324 
325 	*fw_stats_ctx_id = INVALID_STATS_CTX_ID;
326 
327 	if (!en_dev)
328 		return rc;
329 
330 	memset(&fw_msg, 0, sizeof(fw_msg));
331 	rtnl_lock();
332 
333 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1);
334 	req.update_period_ms = cpu_to_le32(1000);
335 	req.stats_dma_addr = cpu_to_le64(dma_map);
336 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
337 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
338 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
339 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
340 	if (!rc)
341 		*fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
342 
343 	rtnl_unlock();
344 	return rc;
345 }
346 
347 /* Device */
348 
349 static bool is_bnxt_re_dev(struct net_device *netdev)
350 {
351 	struct ethtool_drvinfo drvinfo;
352 
353 	if (netdev->ethtool_ops && netdev->ethtool_ops->get_drvinfo) {
354 		memset(&drvinfo, 0, sizeof(drvinfo));
355 		netdev->ethtool_ops->get_drvinfo(netdev, &drvinfo);
356 
357 		if (strcmp(drvinfo.driver, "bnxt_en"))
358 			return false;
359 		return true;
360 	}
361 	return false;
362 }
363 
364 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
365 {
366 	struct bnxt_re_dev *rdev;
367 
368 	rcu_read_lock();
369 	list_for_each_entry_rcu(rdev, &bnxt_re_dev_list, list) {
370 		if (rdev->netdev == netdev) {
371 			rcu_read_unlock();
372 			return rdev;
373 		}
374 	}
375 	rcu_read_unlock();
376 	return NULL;
377 }
378 
379 static void bnxt_re_dev_unprobe(struct net_device *netdev,
380 				struct bnxt_en_dev *en_dev)
381 {
382 	dev_put(netdev);
383 	module_put(en_dev->pdev->driver->driver.owner);
384 }
385 
386 static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev)
387 {
388 	struct bnxt *bp = netdev_priv(netdev);
389 	struct bnxt_en_dev *en_dev;
390 	struct pci_dev *pdev;
391 
392 	/* Call bnxt_en's RoCE probe via indirect API */
393 	if (!bp->ulp_probe)
394 		return ERR_PTR(-EINVAL);
395 
396 	en_dev = bp->ulp_probe(netdev);
397 	if (IS_ERR(en_dev))
398 		return en_dev;
399 
400 	pdev = en_dev->pdev;
401 	if (!pdev)
402 		return ERR_PTR(-EINVAL);
403 
404 	if (!(en_dev->flags & BNXT_EN_FLAG_ROCE_CAP)) {
405 		dev_dbg(&pdev->dev,
406 			"%s: probe error: RoCE is not supported on this device",
407 			ROCE_DRV_MODULE_NAME);
408 		return ERR_PTR(-ENODEV);
409 	}
410 
411 	/* Bump net device reference count */
412 	if (!try_module_get(pdev->driver->driver.owner))
413 		return ERR_PTR(-ENODEV);
414 
415 	dev_hold(netdev);
416 
417 	return en_dev;
418 }
419 
420 static void bnxt_re_unregister_ib(struct bnxt_re_dev *rdev)
421 {
422 	ib_unregister_device(&rdev->ibdev);
423 }
424 
425 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
426 {
427 	struct ib_device *ibdev = &rdev->ibdev;
428 
429 	/* ib device init */
430 	ibdev->owner = THIS_MODULE;
431 	ibdev->node_type = RDMA_NODE_IB_CA;
432 	strlcpy(ibdev->name, "bnxt_re%d", IB_DEVICE_NAME_MAX);
433 	strlcpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
434 		strlen(BNXT_RE_DESC) + 5);
435 	ibdev->phys_port_cnt = 1;
436 
437 	bnxt_qplib_get_guid(rdev->netdev->dev_addr, (u8 *)&ibdev->node_guid);
438 
439 	ibdev->num_comp_vectors	= 1;
440 	ibdev->dev.parent = &rdev->en_dev->pdev->dev;
441 	ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
442 
443 	/* User space */
444 	ibdev->uverbs_abi_ver = BNXT_RE_ABI_VERSION;
445 	ibdev->uverbs_cmd_mask =
446 			(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
447 			(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
448 			(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
449 			(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
450 			(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
451 			(1ull << IB_USER_VERBS_CMD_REG_MR)		|
452 			(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
453 			(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
454 			(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
455 			(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
456 			(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
457 			(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
458 			(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
459 			(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
460 			(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
461 			(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
462 			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
463 			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
464 			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
465 			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
466 			(1ull << IB_USER_VERBS_CMD_CREATE_AH)		|
467 			(1ull << IB_USER_VERBS_CMD_MODIFY_AH)		|
468 			(1ull << IB_USER_VERBS_CMD_QUERY_AH)		|
469 			(1ull << IB_USER_VERBS_CMD_DESTROY_AH);
470 	/* POLL_CQ and REQ_NOTIFY_CQ is directly handled in libbnxt_re */
471 
472 	/* Kernel verbs */
473 	ibdev->query_device		= bnxt_re_query_device;
474 	ibdev->modify_device		= bnxt_re_modify_device;
475 
476 	ibdev->query_port		= bnxt_re_query_port;
477 	ibdev->modify_port		= bnxt_re_modify_port;
478 	ibdev->get_port_immutable	= bnxt_re_get_port_immutable;
479 	ibdev->query_pkey		= bnxt_re_query_pkey;
480 	ibdev->query_gid		= bnxt_re_query_gid;
481 	ibdev->get_netdev		= bnxt_re_get_netdev;
482 	ibdev->add_gid			= bnxt_re_add_gid;
483 	ibdev->del_gid			= bnxt_re_del_gid;
484 	ibdev->get_link_layer		= bnxt_re_get_link_layer;
485 
486 	ibdev->alloc_pd			= bnxt_re_alloc_pd;
487 	ibdev->dealloc_pd		= bnxt_re_dealloc_pd;
488 
489 	ibdev->create_ah		= bnxt_re_create_ah;
490 	ibdev->modify_ah		= bnxt_re_modify_ah;
491 	ibdev->query_ah			= bnxt_re_query_ah;
492 	ibdev->destroy_ah		= bnxt_re_destroy_ah;
493 
494 	ibdev->create_qp		= bnxt_re_create_qp;
495 	ibdev->modify_qp		= bnxt_re_modify_qp;
496 	ibdev->query_qp			= bnxt_re_query_qp;
497 	ibdev->destroy_qp		= bnxt_re_destroy_qp;
498 
499 	ibdev->post_send		= bnxt_re_post_send;
500 	ibdev->post_recv		= bnxt_re_post_recv;
501 
502 	ibdev->create_cq		= bnxt_re_create_cq;
503 	ibdev->destroy_cq		= bnxt_re_destroy_cq;
504 	ibdev->poll_cq			= bnxt_re_poll_cq;
505 	ibdev->req_notify_cq		= bnxt_re_req_notify_cq;
506 
507 	ibdev->get_dma_mr		= bnxt_re_get_dma_mr;
508 	ibdev->dereg_mr			= bnxt_re_dereg_mr;
509 	ibdev->alloc_mr			= bnxt_re_alloc_mr;
510 	ibdev->map_mr_sg		= bnxt_re_map_mr_sg;
511 
512 	ibdev->reg_user_mr		= bnxt_re_reg_user_mr;
513 	ibdev->alloc_ucontext		= bnxt_re_alloc_ucontext;
514 	ibdev->dealloc_ucontext		= bnxt_re_dealloc_ucontext;
515 	ibdev->mmap			= bnxt_re_mmap;
516 
517 	return ib_register_device(ibdev, NULL);
518 }
519 
520 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
521 			char *buf)
522 {
523 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
524 
525 	return scnprintf(buf, PAGE_SIZE, "0x%x\n", rdev->en_dev->pdev->vendor);
526 }
527 
528 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
529 			   char *buf)
530 {
531 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
532 
533 	return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->dev_attr.fw_ver);
534 }
535 
536 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
537 			char *buf)
538 {
539 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
540 
541 	return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->ibdev.node_desc);
542 }
543 
544 static DEVICE_ATTR(hw_rev, 0444, show_rev, NULL);
545 static DEVICE_ATTR(fw_rev, 0444, show_fw_ver, NULL);
546 static DEVICE_ATTR(hca_type, 0444, show_hca, NULL);
547 
548 static struct device_attribute *bnxt_re_attributes[] = {
549 	&dev_attr_hw_rev,
550 	&dev_attr_fw_rev,
551 	&dev_attr_hca_type
552 };
553 
554 static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev)
555 {
556 	dev_put(rdev->netdev);
557 	rdev->netdev = NULL;
558 
559 	mutex_lock(&bnxt_re_dev_lock);
560 	list_del_rcu(&rdev->list);
561 	mutex_unlock(&bnxt_re_dev_lock);
562 
563 	synchronize_rcu();
564 	flush_workqueue(bnxt_re_wq);
565 
566 	ib_dealloc_device(&rdev->ibdev);
567 	/* rdev is gone */
568 }
569 
570 static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev,
571 					   struct bnxt_en_dev *en_dev)
572 {
573 	struct bnxt_re_dev *rdev;
574 
575 	/* Allocate bnxt_re_dev instance here */
576 	rdev = (struct bnxt_re_dev *)ib_alloc_device(sizeof(*rdev));
577 	if (!rdev) {
578 		dev_err(NULL, "%s: bnxt_re_dev allocation failure!",
579 			ROCE_DRV_MODULE_NAME);
580 		return NULL;
581 	}
582 	/* Default values */
583 	rdev->netdev = netdev;
584 	dev_hold(rdev->netdev);
585 	rdev->en_dev = en_dev;
586 	rdev->id = rdev->en_dev->pdev->devfn;
587 	INIT_LIST_HEAD(&rdev->qp_list);
588 	mutex_init(&rdev->qp_lock);
589 	atomic_set(&rdev->qp_count, 0);
590 	atomic_set(&rdev->cq_count, 0);
591 	atomic_set(&rdev->srq_count, 0);
592 	atomic_set(&rdev->mr_count, 0);
593 	atomic_set(&rdev->mw_count, 0);
594 	rdev->cosq[0] = 0xFFFF;
595 	rdev->cosq[1] = 0xFFFF;
596 
597 	mutex_lock(&bnxt_re_dev_lock);
598 	list_add_tail_rcu(&rdev->list, &bnxt_re_dev_list);
599 	mutex_unlock(&bnxt_re_dev_lock);
600 	return rdev;
601 }
602 
603 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
604 			       struct creq_func_event *aeqe)
605 {
606 	switch (aeqe->event) {
607 	case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
608 		break;
609 	case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
610 		break;
611 	case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
612 		break;
613 	case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
614 		break;
615 	case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
616 		break;
617 	case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
618 		break;
619 	case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
620 		break;
621 	case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
622 		break;
623 	case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
624 		break;
625 	case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
626 		break;
627 	case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
628 		break;
629 	default:
630 		return -EINVAL;
631 	}
632 	return 0;
633 }
634 
635 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
636 			       struct bnxt_qplib_cq *handle)
637 {
638 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
639 					     qplib_cq);
640 
641 	if (!cq) {
642 		dev_err(NULL, "%s: CQ is NULL, CQN not handled",
643 			ROCE_DRV_MODULE_NAME);
644 		return -EINVAL;
645 	}
646 	if (cq->ib_cq.comp_handler) {
647 		/* Lock comp_handler? */
648 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
649 	}
650 
651 	return 0;
652 }
653 
654 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
655 {
656 	if (rdev->nq.hwq.max_elements)
657 		bnxt_qplib_disable_nq(&rdev->nq);
658 
659 	if (rdev->qplib_res.rcfw)
660 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
661 }
662 
663 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
664 {
665 	int rc = 0;
666 
667 	bnxt_qplib_init_res(&rdev->qplib_res);
668 
669 	if (rdev->msix_entries[BNXT_RE_NQ_IDX].vector <= 0)
670 		return -EINVAL;
671 
672 	rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq,
673 				  rdev->msix_entries[BNXT_RE_NQ_IDX].vector,
674 				  rdev->msix_entries[BNXT_RE_NQ_IDX].db_offset,
675 				  &bnxt_re_cqn_handler,
676 				  NULL);
677 
678 	if (rc)
679 		dev_err(rdev_to_dev(rdev), "Failed to enable NQ: %#x", rc);
680 
681 	return rc;
682 }
683 
684 static void bnxt_re_free_res(struct bnxt_re_dev *rdev, bool lock_wait)
685 {
686 	if (rdev->nq.hwq.max_elements) {
687 		bnxt_re_net_ring_free(rdev, rdev->nq.ring_id, lock_wait);
688 		bnxt_qplib_free_nq(&rdev->nq);
689 	}
690 	if (rdev->qplib_res.dpi_tbl.max) {
691 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
692 				       &rdev->qplib_res.dpi_tbl,
693 				       &rdev->dpi_privileged);
694 	}
695 	if (rdev->qplib_res.rcfw) {
696 		bnxt_qplib_free_res(&rdev->qplib_res);
697 		rdev->qplib_res.rcfw = NULL;
698 	}
699 }
700 
701 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
702 {
703 	int rc = 0;
704 
705 	/* Configure and allocate resources for qplib */
706 	rdev->qplib_res.rcfw = &rdev->rcfw;
707 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr);
708 	if (rc)
709 		goto fail;
710 
711 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
712 				  rdev->netdev, &rdev->dev_attr);
713 	if (rc)
714 		goto fail;
715 
716 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
717 				  &rdev->dpi_privileged,
718 				  rdev);
719 	if (rc)
720 		goto fail;
721 
722 	rdev->nq.hwq.max_elements = BNXT_RE_MAX_CQ_COUNT +
723 				    BNXT_RE_MAX_SRQC_COUNT + 2;
724 	rc = bnxt_qplib_alloc_nq(rdev->en_dev->pdev, &rdev->nq);
725 	if (rc) {
726 		dev_err(rdev_to_dev(rdev),
727 			"Failed to allocate NQ memory: %#x", rc);
728 		goto fail;
729 	}
730 	rc = bnxt_re_net_ring_alloc
731 			(rdev, rdev->nq.hwq.pbl[PBL_LVL_0].pg_map_arr,
732 			 rdev->nq.hwq.pbl[rdev->nq.hwq.level].pg_count,
733 			 HWRM_RING_ALLOC_CMPL, BNXT_QPLIB_NQE_MAX_CNT - 1,
734 			 rdev->msix_entries[BNXT_RE_NQ_IDX].ring_idx,
735 			 &rdev->nq.ring_id);
736 	if (rc) {
737 		dev_err(rdev_to_dev(rdev),
738 			"Failed to allocate NQ ring: %#x", rc);
739 		goto free_nq;
740 	}
741 	return 0;
742 free_nq:
743 	bnxt_qplib_free_nq(&rdev->nq);
744 fail:
745 	rdev->qplib_res.rcfw = NULL;
746 	return rc;
747 }
748 
749 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
750 				   u8 port_num, enum ib_event_type event)
751 {
752 	struct ib_event ib_event;
753 
754 	ib_event.device = ibdev;
755 	if (qp)
756 		ib_event.element.qp = qp;
757 	else
758 		ib_event.element.port_num = port_num;
759 	ib_event.event = event;
760 	ib_dispatch_event(&ib_event);
761 }
762 
763 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN      0x02
764 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
765 				      u64 *cid_map)
766 {
767 	struct hwrm_queue_pri2cos_qcfg_input req = {0};
768 	struct bnxt *bp = netdev_priv(rdev->netdev);
769 	struct hwrm_queue_pri2cos_qcfg_output resp;
770 	struct bnxt_en_dev *en_dev = rdev->en_dev;
771 	struct bnxt_fw_msg fw_msg;
772 	u32 flags = 0;
773 	u8 *qcfgmap, *tmp_map;
774 	int rc = 0, i;
775 
776 	if (!cid_map)
777 		return -EINVAL;
778 
779 	memset(&fw_msg, 0, sizeof(fw_msg));
780 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
781 			      HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
782 	flags |= (dir & 0x01);
783 	flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
784 	req.flags = cpu_to_le32(flags);
785 	req.port_id = bp->pf.port_id;
786 
787 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
788 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
789 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
790 	if (rc)
791 		return rc;
792 
793 	if (resp.queue_cfg_info) {
794 		dev_warn(rdev_to_dev(rdev),
795 			 "Asymmetric cos queue configuration detected");
796 		dev_warn(rdev_to_dev(rdev),
797 			 " on device, QoS may not be fully functional\n");
798 	}
799 	qcfgmap = &resp.pri0_cos_queue_id;
800 	tmp_map = (u8 *)cid_map;
801 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
802 		tmp_map[i] = qcfgmap[i];
803 
804 	return rc;
805 }
806 
807 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
808 					struct bnxt_re_qp *qp)
809 {
810 	return (qp->ib_qp.qp_type == IB_QPT_GSI) || (qp == rdev->qp1_sqp);
811 }
812 
813 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
814 {
815 	int mask = IB_QP_STATE;
816 	struct ib_qp_attr qp_attr;
817 	struct bnxt_re_qp *qp;
818 
819 	qp_attr.qp_state = IB_QPS_ERR;
820 	mutex_lock(&rdev->qp_lock);
821 	list_for_each_entry(qp, &rdev->qp_list, list) {
822 		/* Modify the state of all QPs except QP1/Shadow QP */
823 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
824 			if (qp->qplib_qp.state !=
825 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
826 			    qp->qplib_qp.state !=
827 			    CMDQ_MODIFY_QP_NEW_STATE_ERR) {
828 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
829 						       1, IB_EVENT_QP_FATAL);
830 				bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
831 						  NULL);
832 			}
833 		}
834 	}
835 	mutex_unlock(&rdev->qp_lock);
836 }
837 
838 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
839 {
840 	u32 prio_map = 0, tmp_map = 0;
841 	struct net_device *netdev;
842 	struct dcb_app app;
843 
844 	netdev = rdev->netdev;
845 
846 	memset(&app, 0, sizeof(app));
847 	app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
848 	app.protocol = ETH_P_IBOE;
849 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
850 	prio_map = tmp_map;
851 
852 	app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
853 	app.protocol = ROCE_V2_UDP_DPORT;
854 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
855 	prio_map |= tmp_map;
856 
857 	if (!prio_map)
858 		prio_map = -EFAULT;
859 	return prio_map;
860 }
861 
862 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
863 {
864 	u16 prio;
865 	u8 id;
866 
867 	for (prio = 0, id = 0; prio < 8; prio++) {
868 		if (prio_map & (1 << prio)) {
869 			cosq[id] = cid_map[prio];
870 			id++;
871 			if (id == 2) /* Max 2 tcs supported */
872 				break;
873 		}
874 	}
875 }
876 
877 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
878 {
879 	u8 prio_map = 0;
880 	u64 cid_map;
881 	int rc;
882 
883 	/* Get priority for roce */
884 	rc = bnxt_re_get_priority_mask(rdev);
885 	if (rc < 0)
886 		return rc;
887 	prio_map = (u8)rc;
888 
889 	if (prio_map == rdev->cur_prio_map)
890 		return 0;
891 	rdev->cur_prio_map = prio_map;
892 	/* Get cosq id for this priority */
893 	rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
894 	if (rc) {
895 		dev_warn(rdev_to_dev(rdev), "no cos for p_mask %x\n", prio_map);
896 		return rc;
897 	}
898 	/* Parse CoS IDs for app priority */
899 	bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
900 
901 	/* Config BONO. */
902 	rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
903 	if (rc) {
904 		dev_warn(rdev_to_dev(rdev), "no tc for cos{%x, %x}\n",
905 			 rdev->cosq[0], rdev->cosq[1]);
906 		return rc;
907 	}
908 
909 	return 0;
910 }
911 
912 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev, bool lock_wait)
913 {
914 	int i, rc;
915 
916 	if (test_and_clear_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags)) {
917 		for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++)
918 			device_remove_file(&rdev->ibdev.dev,
919 					   bnxt_re_attributes[i]);
920 		/* Cleanup ib dev */
921 		bnxt_re_unregister_ib(rdev);
922 	}
923 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
924 		cancel_delayed_work(&rdev->worker);
925 
926 	bnxt_re_cleanup_res(rdev);
927 	bnxt_re_free_res(rdev, lock_wait);
928 
929 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
930 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
931 		if (rc)
932 			dev_warn(rdev_to_dev(rdev),
933 				 "Failed to deinitialize RCFW: %#x", rc);
934 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id,
935 					   lock_wait);
936 		bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
937 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
938 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, lock_wait);
939 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
940 	}
941 	if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
942 		rc = bnxt_re_free_msix(rdev, lock_wait);
943 		if (rc)
944 			dev_warn(rdev_to_dev(rdev),
945 				 "Failed to free MSI-X vectors: %#x", rc);
946 	}
947 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
948 		rc = bnxt_re_unregister_netdev(rdev, lock_wait);
949 		if (rc)
950 			dev_warn(rdev_to_dev(rdev),
951 				 "Failed to unregister with netdev: %#x", rc);
952 	}
953 }
954 
955 static void bnxt_re_set_resource_limits(struct bnxt_re_dev *rdev)
956 {
957 	u32 i;
958 
959 	rdev->qplib_ctx.qpc_count = BNXT_RE_MAX_QPC_COUNT;
960 	rdev->qplib_ctx.mrw_count = BNXT_RE_MAX_MRW_COUNT;
961 	rdev->qplib_ctx.srqc_count = BNXT_RE_MAX_SRQC_COUNT;
962 	rdev->qplib_ctx.cq_count = BNXT_RE_MAX_CQ_COUNT;
963 	for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
964 		rdev->qplib_ctx.tqm_count[i] =
965 		rdev->dev_attr.tqm_alloc_reqs[i];
966 }
967 
968 /* worker thread for polling periodic events. Now used for QoS programming*/
969 static void bnxt_re_worker(struct work_struct *work)
970 {
971 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
972 						worker.work);
973 
974 	bnxt_re_setup_qos(rdev);
975 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
976 }
977 
978 static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
979 {
980 	int i, j, rc;
981 
982 	/* Registered a new RoCE device instance to netdev */
983 	rc = bnxt_re_register_netdev(rdev);
984 	if (rc) {
985 		pr_err("Failed to register with netedev: %#x\n", rc);
986 		return -EINVAL;
987 	}
988 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
989 
990 	rc = bnxt_re_request_msix(rdev);
991 	if (rc) {
992 		pr_err("Failed to get MSI-X vectors: %#x\n", rc);
993 		rc = -EINVAL;
994 		goto fail;
995 	}
996 	set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
997 
998 	/* Establish RCFW Communication Channel to initialize the context
999 	 * memory for the function and all child VFs
1000 	 */
1001 	rc = bnxt_qplib_alloc_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw);
1002 	if (rc)
1003 		goto fail;
1004 
1005 	rc = bnxt_re_net_ring_alloc
1006 			(rdev, rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr,
1007 			 rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count,
1008 			 HWRM_RING_ALLOC_CMPL, BNXT_QPLIB_CREQE_MAX_CNT - 1,
1009 			 rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx,
1010 			 &rdev->rcfw.creq_ring_id);
1011 	if (rc) {
1012 		pr_err("Failed to allocate CREQ: %#x\n", rc);
1013 		goto free_rcfw;
1014 	}
1015 	rc = bnxt_qplib_enable_rcfw_channel
1016 				(rdev->en_dev->pdev, &rdev->rcfw,
1017 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].vector,
1018 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].db_offset,
1019 				 0, &bnxt_re_aeq_handler);
1020 	if (rc) {
1021 		pr_err("Failed to enable RCFW channel: %#x\n", rc);
1022 		goto free_ring;
1023 	}
1024 
1025 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr);
1026 	if (rc)
1027 		goto disable_rcfw;
1028 	bnxt_re_set_resource_limits(rdev);
1029 
1030 	rc = bnxt_qplib_alloc_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx, 0);
1031 	if (rc) {
1032 		pr_err("Failed to allocate QPLIB context: %#x\n", rc);
1033 		goto disable_rcfw;
1034 	}
1035 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
1036 					 rdev->qplib_ctx.stats.dma_map,
1037 					 &rdev->qplib_ctx.stats.fw_id);
1038 	if (rc) {
1039 		pr_err("Failed to allocate stats context: %#x\n", rc);
1040 		goto free_ctx;
1041 	}
1042 
1043 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx, 0);
1044 	if (rc) {
1045 		pr_err("Failed to initialize RCFW: %#x\n", rc);
1046 		goto free_sctx;
1047 	}
1048 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1049 
1050 	/* Resources based on the 'new' device caps */
1051 	rc = bnxt_re_alloc_res(rdev);
1052 	if (rc) {
1053 		pr_err("Failed to allocate resources: %#x\n", rc);
1054 		goto fail;
1055 	}
1056 	rc = bnxt_re_init_res(rdev);
1057 	if (rc) {
1058 		pr_err("Failed to initialize resources: %#x\n", rc);
1059 		goto fail;
1060 	}
1061 
1062 	rc = bnxt_re_setup_qos(rdev);
1063 	if (rc)
1064 		pr_info("RoCE priority not yet configured\n");
1065 
1066 	INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1067 	set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1068 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1069 
1070 	/* Register ib dev */
1071 	rc = bnxt_re_register_ib(rdev);
1072 	if (rc) {
1073 		pr_err("Failed to register with IB: %#x\n", rc);
1074 		goto fail;
1075 	}
1076 	dev_info(rdev_to_dev(rdev), "Device registered successfully");
1077 	for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++) {
1078 		rc = device_create_file(&rdev->ibdev.dev,
1079 					bnxt_re_attributes[i]);
1080 		if (rc) {
1081 			dev_err(rdev_to_dev(rdev),
1082 				"Failed to create IB sysfs: %#x", rc);
1083 			/* Must clean up all created device files */
1084 			for (j = 0; j < i; j++)
1085 				device_remove_file(&rdev->ibdev.dev,
1086 						   bnxt_re_attributes[j]);
1087 			bnxt_re_unregister_ib(rdev);
1088 			goto fail;
1089 		}
1090 	}
1091 	set_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags);
1092 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_PORT_ACTIVE);
1093 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_GID_CHANGE);
1094 
1095 	return 0;
1096 free_sctx:
1097 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id, true);
1098 free_ctx:
1099 	bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1100 disable_rcfw:
1101 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1102 free_ring:
1103 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, true);
1104 free_rcfw:
1105 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1106 fail:
1107 	bnxt_re_ib_unreg(rdev, true);
1108 	return rc;
1109 }
1110 
1111 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1112 {
1113 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1114 	struct net_device *netdev = rdev->netdev;
1115 
1116 	bnxt_re_dev_remove(rdev);
1117 
1118 	if (netdev)
1119 		bnxt_re_dev_unprobe(netdev, en_dev);
1120 }
1121 
1122 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1123 {
1124 	struct bnxt_en_dev *en_dev;
1125 	int rc = 0;
1126 
1127 	if (!is_bnxt_re_dev(netdev))
1128 		return -ENODEV;
1129 
1130 	en_dev = bnxt_re_dev_probe(netdev);
1131 	if (IS_ERR(en_dev)) {
1132 		if (en_dev != ERR_PTR(-ENODEV))
1133 			pr_err("%s: Failed to probe\n", ROCE_DRV_MODULE_NAME);
1134 		rc = PTR_ERR(en_dev);
1135 		goto exit;
1136 	}
1137 	*rdev = bnxt_re_dev_add(netdev, en_dev);
1138 	if (!*rdev) {
1139 		rc = -ENOMEM;
1140 		bnxt_re_dev_unprobe(netdev, en_dev);
1141 		goto exit;
1142 	}
1143 exit:
1144 	return rc;
1145 }
1146 
1147 static void bnxt_re_remove_one(struct bnxt_re_dev *rdev)
1148 {
1149 	pci_dev_put(rdev->en_dev->pdev);
1150 }
1151 
1152 /* Handle all deferred netevents tasks */
1153 static void bnxt_re_task(struct work_struct *work)
1154 {
1155 	struct bnxt_re_work *re_work;
1156 	struct bnxt_re_dev *rdev;
1157 	int rc = 0;
1158 
1159 	re_work = container_of(work, struct bnxt_re_work, work);
1160 	rdev = re_work->rdev;
1161 
1162 	if (re_work->event != NETDEV_REGISTER &&
1163 	    !test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1164 		return;
1165 
1166 	switch (re_work->event) {
1167 	case NETDEV_REGISTER:
1168 		rc = bnxt_re_ib_reg(rdev);
1169 		if (rc)
1170 			dev_err(rdev_to_dev(rdev),
1171 				"Failed to register with IB: %#x", rc);
1172 		break;
1173 	case NETDEV_UP:
1174 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1175 				       IB_EVENT_PORT_ACTIVE);
1176 		break;
1177 	case NETDEV_DOWN:
1178 		bnxt_re_dev_stop(rdev);
1179 		break;
1180 	case NETDEV_CHANGE:
1181 		if (!netif_carrier_ok(rdev->netdev))
1182 			bnxt_re_dev_stop(rdev);
1183 		else if (netif_carrier_ok(rdev->netdev))
1184 			bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1185 					       IB_EVENT_PORT_ACTIVE);
1186 		break;
1187 	default:
1188 		break;
1189 	}
1190 	kfree(re_work);
1191 }
1192 
1193 static void bnxt_re_init_one(struct bnxt_re_dev *rdev)
1194 {
1195 	pci_dev_get(rdev->en_dev->pdev);
1196 }
1197 
1198 /*
1199  * "Notifier chain callback can be invoked for the same chain from
1200  * different CPUs at the same time".
1201  *
1202  * For cases when the netdev is already present, our call to the
1203  * register_netdevice_notifier() will actually get the rtnl_lock()
1204  * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1205  * events.
1206  *
1207  * But for cases when the netdev is not already present, the notifier
1208  * chain is subjected to be invoked from different CPUs simultaneously.
1209  *
1210  * This is protected by the netdev_mutex.
1211  */
1212 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1213 				unsigned long event, void *ptr)
1214 {
1215 	struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1216 	struct bnxt_re_work *re_work;
1217 	struct bnxt_re_dev *rdev;
1218 	int rc = 0;
1219 	bool sch_work = false;
1220 
1221 	real_dev = rdma_vlan_dev_real_dev(netdev);
1222 	if (!real_dev)
1223 		real_dev = netdev;
1224 
1225 	rdev = bnxt_re_from_netdev(real_dev);
1226 	if (!rdev && event != NETDEV_REGISTER)
1227 		goto exit;
1228 	if (real_dev != netdev)
1229 		goto exit;
1230 
1231 	switch (event) {
1232 	case NETDEV_REGISTER:
1233 		if (rdev)
1234 			break;
1235 		rc = bnxt_re_dev_reg(&rdev, real_dev);
1236 		if (rc == -ENODEV)
1237 			break;
1238 		if (rc) {
1239 			pr_err("Failed to register with the device %s: %#x\n",
1240 			       real_dev->name, rc);
1241 			break;
1242 		}
1243 		bnxt_re_init_one(rdev);
1244 		sch_work = true;
1245 		break;
1246 
1247 	case NETDEV_UNREGISTER:
1248 		bnxt_re_ib_unreg(rdev, false);
1249 		bnxt_re_remove_one(rdev);
1250 		bnxt_re_dev_unreg(rdev);
1251 		break;
1252 
1253 	default:
1254 		sch_work = true;
1255 		break;
1256 	}
1257 	if (sch_work) {
1258 		/* Allocate for the deferred task */
1259 		re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1260 		if (re_work) {
1261 			re_work->rdev = rdev;
1262 			re_work->event = event;
1263 			re_work->vlan_dev = (real_dev == netdev ?
1264 					     NULL : netdev);
1265 			INIT_WORK(&re_work->work, bnxt_re_task);
1266 			queue_work(bnxt_re_wq, &re_work->work);
1267 		}
1268 	}
1269 
1270 exit:
1271 	return NOTIFY_DONE;
1272 }
1273 
1274 static struct notifier_block bnxt_re_netdev_notifier = {
1275 	.notifier_call = bnxt_re_netdev_event
1276 };
1277 
1278 static int __init bnxt_re_mod_init(void)
1279 {
1280 	int rc = 0;
1281 
1282 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1283 
1284 	bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1285 	if (!bnxt_re_wq)
1286 		return -ENOMEM;
1287 
1288 	INIT_LIST_HEAD(&bnxt_re_dev_list);
1289 
1290 	rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1291 	if (rc) {
1292 		pr_err("%s: Cannot register to netdevice_notifier",
1293 		       ROCE_DRV_MODULE_NAME);
1294 		goto err_netdev;
1295 	}
1296 	return 0;
1297 
1298 err_netdev:
1299 	destroy_workqueue(bnxt_re_wq);
1300 
1301 	return rc;
1302 }
1303 
1304 static void __exit bnxt_re_mod_exit(void)
1305 {
1306 	unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1307 	if (bnxt_re_wq)
1308 		destroy_workqueue(bnxt_re_wq);
1309 }
1310 
1311 module_init(bnxt_re_mod_init);
1312 module_exit(bnxt_re_mod_exit);
1313