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 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
337 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
338 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
339 	if (!rc)
340 		*fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
341 
342 	rtnl_unlock();
343 	return rc;
344 }
345 
346 /* Device */
347 
348 static bool is_bnxt_re_dev(struct net_device *netdev)
349 {
350 	struct ethtool_drvinfo drvinfo;
351 
352 	if (netdev->ethtool_ops && netdev->ethtool_ops->get_drvinfo) {
353 		memset(&drvinfo, 0, sizeof(drvinfo));
354 		netdev->ethtool_ops->get_drvinfo(netdev, &drvinfo);
355 
356 		if (strcmp(drvinfo.driver, "bnxt_en"))
357 			return false;
358 		return true;
359 	}
360 	return false;
361 }
362 
363 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
364 {
365 	struct bnxt_re_dev *rdev;
366 
367 	rcu_read_lock();
368 	list_for_each_entry_rcu(rdev, &bnxt_re_dev_list, list) {
369 		if (rdev->netdev == netdev) {
370 			rcu_read_unlock();
371 			return rdev;
372 		}
373 	}
374 	rcu_read_unlock();
375 	return NULL;
376 }
377 
378 static void bnxt_re_dev_unprobe(struct net_device *netdev,
379 				struct bnxt_en_dev *en_dev)
380 {
381 	dev_put(netdev);
382 	module_put(en_dev->pdev->driver->driver.owner);
383 }
384 
385 static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev)
386 {
387 	struct bnxt *bp = netdev_priv(netdev);
388 	struct bnxt_en_dev *en_dev;
389 	struct pci_dev *pdev;
390 
391 	/* Call bnxt_en's RoCE probe via indirect API */
392 	if (!bp->ulp_probe)
393 		return ERR_PTR(-EINVAL);
394 
395 	en_dev = bp->ulp_probe(netdev);
396 	if (IS_ERR(en_dev))
397 		return en_dev;
398 
399 	pdev = en_dev->pdev;
400 	if (!pdev)
401 		return ERR_PTR(-EINVAL);
402 
403 	if (!(en_dev->flags & BNXT_EN_FLAG_ROCE_CAP)) {
404 		dev_dbg(&pdev->dev,
405 			"%s: probe error: RoCE is not supported on this device",
406 			ROCE_DRV_MODULE_NAME);
407 		return ERR_PTR(-ENODEV);
408 	}
409 
410 	/* Bump net device reference count */
411 	if (!try_module_get(pdev->driver->driver.owner))
412 		return ERR_PTR(-ENODEV);
413 
414 	dev_hold(netdev);
415 
416 	return en_dev;
417 }
418 
419 static void bnxt_re_unregister_ib(struct bnxt_re_dev *rdev)
420 {
421 	ib_unregister_device(&rdev->ibdev);
422 }
423 
424 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
425 {
426 	struct ib_device *ibdev = &rdev->ibdev;
427 
428 	/* ib device init */
429 	ibdev->owner = THIS_MODULE;
430 	ibdev->node_type = RDMA_NODE_IB_CA;
431 	strlcpy(ibdev->name, "bnxt_re%d", IB_DEVICE_NAME_MAX);
432 	strlcpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
433 		strlen(BNXT_RE_DESC) + 5);
434 	ibdev->phys_port_cnt = 1;
435 
436 	bnxt_qplib_get_guid(rdev->netdev->dev_addr, (u8 *)&ibdev->node_guid);
437 
438 	ibdev->num_comp_vectors	= 1;
439 	ibdev->dev.parent = &rdev->en_dev->pdev->dev;
440 	ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
441 
442 	/* User space */
443 	ibdev->uverbs_abi_ver = BNXT_RE_ABI_VERSION;
444 	ibdev->uverbs_cmd_mask =
445 			(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
446 			(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
447 			(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
448 			(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
449 			(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
450 			(1ull << IB_USER_VERBS_CMD_REG_MR)		|
451 			(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
452 			(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
453 			(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
454 			(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
455 			(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
456 			(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
457 			(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
458 			(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
459 			(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
460 			(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
461 			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
462 			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
463 			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
464 			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
465 			(1ull << IB_USER_VERBS_CMD_CREATE_AH)		|
466 			(1ull << IB_USER_VERBS_CMD_MODIFY_AH)		|
467 			(1ull << IB_USER_VERBS_CMD_QUERY_AH)		|
468 			(1ull << IB_USER_VERBS_CMD_DESTROY_AH);
469 	/* POLL_CQ and REQ_NOTIFY_CQ is directly handled in libbnxt_re */
470 
471 	/* Kernel verbs */
472 	ibdev->query_device		= bnxt_re_query_device;
473 	ibdev->modify_device		= bnxt_re_modify_device;
474 
475 	ibdev->query_port		= bnxt_re_query_port;
476 	ibdev->modify_port		= bnxt_re_modify_port;
477 	ibdev->get_port_immutable	= bnxt_re_get_port_immutable;
478 	ibdev->query_pkey		= bnxt_re_query_pkey;
479 	ibdev->query_gid		= bnxt_re_query_gid;
480 	ibdev->get_netdev		= bnxt_re_get_netdev;
481 	ibdev->add_gid			= bnxt_re_add_gid;
482 	ibdev->del_gid			= bnxt_re_del_gid;
483 	ibdev->get_link_layer		= bnxt_re_get_link_layer;
484 
485 	ibdev->alloc_pd			= bnxt_re_alloc_pd;
486 	ibdev->dealloc_pd		= bnxt_re_dealloc_pd;
487 
488 	ibdev->create_ah		= bnxt_re_create_ah;
489 	ibdev->modify_ah		= bnxt_re_modify_ah;
490 	ibdev->query_ah			= bnxt_re_query_ah;
491 	ibdev->destroy_ah		= bnxt_re_destroy_ah;
492 
493 	ibdev->create_qp		= bnxt_re_create_qp;
494 	ibdev->modify_qp		= bnxt_re_modify_qp;
495 	ibdev->query_qp			= bnxt_re_query_qp;
496 	ibdev->destroy_qp		= bnxt_re_destroy_qp;
497 
498 	ibdev->post_send		= bnxt_re_post_send;
499 	ibdev->post_recv		= bnxt_re_post_recv;
500 
501 	ibdev->create_cq		= bnxt_re_create_cq;
502 	ibdev->destroy_cq		= bnxt_re_destroy_cq;
503 	ibdev->poll_cq			= bnxt_re_poll_cq;
504 	ibdev->req_notify_cq		= bnxt_re_req_notify_cq;
505 
506 	ibdev->get_dma_mr		= bnxt_re_get_dma_mr;
507 	ibdev->dereg_mr			= bnxt_re_dereg_mr;
508 	ibdev->alloc_mr			= bnxt_re_alloc_mr;
509 	ibdev->map_mr_sg		= bnxt_re_map_mr_sg;
510 	ibdev->alloc_fmr		= bnxt_re_alloc_fmr;
511 	ibdev->map_phys_fmr		= bnxt_re_map_phys_fmr;
512 	ibdev->unmap_fmr		= bnxt_re_unmap_fmr;
513 	ibdev->dealloc_fmr		= bnxt_re_dealloc_fmr;
514 
515 	ibdev->reg_user_mr		= bnxt_re_reg_user_mr;
516 	ibdev->alloc_ucontext		= bnxt_re_alloc_ucontext;
517 	ibdev->dealloc_ucontext		= bnxt_re_dealloc_ucontext;
518 	ibdev->mmap			= bnxt_re_mmap;
519 
520 	return ib_register_device(ibdev, NULL);
521 }
522 
523 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
524 			char *buf)
525 {
526 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
527 
528 	return scnprintf(buf, PAGE_SIZE, "0x%x\n", rdev->en_dev->pdev->vendor);
529 }
530 
531 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
532 			   char *buf)
533 {
534 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
535 
536 	return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->dev_attr.fw_ver);
537 }
538 
539 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
540 			char *buf)
541 {
542 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
543 
544 	return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->ibdev.node_desc);
545 }
546 
547 static DEVICE_ATTR(hw_rev, 0444, show_rev, NULL);
548 static DEVICE_ATTR(fw_rev, 0444, show_fw_ver, NULL);
549 static DEVICE_ATTR(hca_type, 0444, show_hca, NULL);
550 
551 static struct device_attribute *bnxt_re_attributes[] = {
552 	&dev_attr_hw_rev,
553 	&dev_attr_fw_rev,
554 	&dev_attr_hca_type
555 };
556 
557 static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev)
558 {
559 	dev_put(rdev->netdev);
560 	rdev->netdev = NULL;
561 
562 	mutex_lock(&bnxt_re_dev_lock);
563 	list_del_rcu(&rdev->list);
564 	mutex_unlock(&bnxt_re_dev_lock);
565 
566 	synchronize_rcu();
567 	flush_workqueue(bnxt_re_wq);
568 
569 	ib_dealloc_device(&rdev->ibdev);
570 	/* rdev is gone */
571 }
572 
573 static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev,
574 					   struct bnxt_en_dev *en_dev)
575 {
576 	struct bnxt_re_dev *rdev;
577 
578 	/* Allocate bnxt_re_dev instance here */
579 	rdev = (struct bnxt_re_dev *)ib_alloc_device(sizeof(*rdev));
580 	if (!rdev) {
581 		dev_err(NULL, "%s: bnxt_re_dev allocation failure!",
582 			ROCE_DRV_MODULE_NAME);
583 		return NULL;
584 	}
585 	/* Default values */
586 	rdev->netdev = netdev;
587 	dev_hold(rdev->netdev);
588 	rdev->en_dev = en_dev;
589 	rdev->id = rdev->en_dev->pdev->devfn;
590 	INIT_LIST_HEAD(&rdev->qp_list);
591 	mutex_init(&rdev->qp_lock);
592 	atomic_set(&rdev->qp_count, 0);
593 	atomic_set(&rdev->cq_count, 0);
594 	atomic_set(&rdev->srq_count, 0);
595 	atomic_set(&rdev->mr_count, 0);
596 	atomic_set(&rdev->mw_count, 0);
597 	rdev->cosq[0] = 0xFFFF;
598 	rdev->cosq[1] = 0xFFFF;
599 
600 	mutex_lock(&bnxt_re_dev_lock);
601 	list_add_tail_rcu(&rdev->list, &bnxt_re_dev_list);
602 	mutex_unlock(&bnxt_re_dev_lock);
603 	return rdev;
604 }
605 
606 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
607 			       struct creq_func_event *aeqe)
608 {
609 	switch (aeqe->event) {
610 	case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
611 		break;
612 	case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
613 		break;
614 	case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
615 		break;
616 	case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
617 		break;
618 	case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
619 		break;
620 	case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
621 		break;
622 	case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
623 		break;
624 	case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
625 		break;
626 	case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
627 		break;
628 	case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
629 		break;
630 	case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
631 		break;
632 	default:
633 		return -EINVAL;
634 	}
635 	return 0;
636 }
637 
638 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
639 			       struct bnxt_qplib_cq *handle)
640 {
641 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
642 					     qplib_cq);
643 
644 	if (!cq) {
645 		dev_err(NULL, "%s: CQ is NULL, CQN not handled",
646 			ROCE_DRV_MODULE_NAME);
647 		return -EINVAL;
648 	}
649 	if (cq->ib_cq.comp_handler) {
650 		/* Lock comp_handler? */
651 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
652 	}
653 
654 	return 0;
655 }
656 
657 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
658 {
659 	if (rdev->nq.hwq.max_elements)
660 		bnxt_qplib_disable_nq(&rdev->nq);
661 
662 	if (rdev->qplib_res.rcfw)
663 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
664 }
665 
666 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
667 {
668 	int rc = 0;
669 
670 	bnxt_qplib_init_res(&rdev->qplib_res);
671 
672 	if (rdev->msix_entries[BNXT_RE_NQ_IDX].vector <= 0)
673 		return -EINVAL;
674 
675 	rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq,
676 				  rdev->msix_entries[BNXT_RE_NQ_IDX].vector,
677 				  rdev->msix_entries[BNXT_RE_NQ_IDX].db_offset,
678 				  &bnxt_re_cqn_handler,
679 				  NULL);
680 
681 	if (rc)
682 		dev_err(rdev_to_dev(rdev), "Failed to enable NQ: %#x", rc);
683 
684 	return rc;
685 }
686 
687 static void bnxt_re_free_res(struct bnxt_re_dev *rdev, bool lock_wait)
688 {
689 	if (rdev->nq.hwq.max_elements) {
690 		bnxt_re_net_ring_free(rdev, rdev->nq.ring_id, lock_wait);
691 		bnxt_qplib_free_nq(&rdev->nq);
692 	}
693 	if (rdev->qplib_res.dpi_tbl.max) {
694 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
695 				       &rdev->qplib_res.dpi_tbl,
696 				       &rdev->dpi_privileged);
697 	}
698 	if (rdev->qplib_res.rcfw) {
699 		bnxt_qplib_free_res(&rdev->qplib_res);
700 		rdev->qplib_res.rcfw = NULL;
701 	}
702 }
703 
704 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
705 {
706 	int rc = 0;
707 
708 	/* Configure and allocate resources for qplib */
709 	rdev->qplib_res.rcfw = &rdev->rcfw;
710 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr);
711 	if (rc)
712 		goto fail;
713 
714 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
715 				  rdev->netdev, &rdev->dev_attr);
716 	if (rc)
717 		goto fail;
718 
719 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
720 				  &rdev->dpi_privileged,
721 				  rdev);
722 	if (rc)
723 		goto fail;
724 
725 	rdev->nq.hwq.max_elements = BNXT_RE_MAX_CQ_COUNT +
726 				    BNXT_RE_MAX_SRQC_COUNT + 2;
727 	rc = bnxt_qplib_alloc_nq(rdev->en_dev->pdev, &rdev->nq);
728 	if (rc) {
729 		dev_err(rdev_to_dev(rdev),
730 			"Failed to allocate NQ memory: %#x", rc);
731 		goto fail;
732 	}
733 	rc = bnxt_re_net_ring_alloc
734 			(rdev, rdev->nq.hwq.pbl[PBL_LVL_0].pg_map_arr,
735 			 rdev->nq.hwq.pbl[rdev->nq.hwq.level].pg_count,
736 			 HWRM_RING_ALLOC_CMPL, BNXT_QPLIB_NQE_MAX_CNT - 1,
737 			 rdev->msix_entries[BNXT_RE_NQ_IDX].ring_idx,
738 			 &rdev->nq.ring_id);
739 	if (rc) {
740 		dev_err(rdev_to_dev(rdev),
741 			"Failed to allocate NQ ring: %#x", rc);
742 		goto free_nq;
743 	}
744 	return 0;
745 free_nq:
746 	bnxt_qplib_free_nq(&rdev->nq);
747 fail:
748 	rdev->qplib_res.rcfw = NULL;
749 	return rc;
750 }
751 
752 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
753 				   u8 port_num, enum ib_event_type event)
754 {
755 	struct ib_event ib_event;
756 
757 	ib_event.device = ibdev;
758 	if (qp)
759 		ib_event.element.qp = qp;
760 	else
761 		ib_event.element.port_num = port_num;
762 	ib_event.event = event;
763 	ib_dispatch_event(&ib_event);
764 }
765 
766 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN      0x02
767 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
768 				      u64 *cid_map)
769 {
770 	struct hwrm_queue_pri2cos_qcfg_input req = {0};
771 	struct bnxt *bp = netdev_priv(rdev->netdev);
772 	struct hwrm_queue_pri2cos_qcfg_output resp;
773 	struct bnxt_en_dev *en_dev = rdev->en_dev;
774 	struct bnxt_fw_msg fw_msg;
775 	u32 flags = 0;
776 	u8 *qcfgmap, *tmp_map;
777 	int rc = 0, i;
778 
779 	if (!cid_map)
780 		return -EINVAL;
781 
782 	memset(&fw_msg, 0, sizeof(fw_msg));
783 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
784 			      HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
785 	flags |= (dir & 0x01);
786 	flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
787 	req.flags = cpu_to_le32(flags);
788 	req.port_id = bp->pf.port_id;
789 
790 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
791 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
792 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
793 	if (rc)
794 		return rc;
795 
796 	if (resp.queue_cfg_info) {
797 		dev_warn(rdev_to_dev(rdev),
798 			 "Asymmetric cos queue configuration detected");
799 		dev_warn(rdev_to_dev(rdev),
800 			 " on device, QoS may not be fully functional\n");
801 	}
802 	qcfgmap = &resp.pri0_cos_queue_id;
803 	tmp_map = (u8 *)cid_map;
804 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
805 		tmp_map[i] = qcfgmap[i];
806 
807 	return rc;
808 }
809 
810 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
811 					struct bnxt_re_qp *qp)
812 {
813 	return (qp->ib_qp.qp_type == IB_QPT_GSI) || (qp == rdev->qp1_sqp);
814 }
815 
816 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
817 {
818 	int mask = IB_QP_STATE;
819 	struct ib_qp_attr qp_attr;
820 	struct bnxt_re_qp *qp;
821 
822 	qp_attr.qp_state = IB_QPS_ERR;
823 	mutex_lock(&rdev->qp_lock);
824 	list_for_each_entry(qp, &rdev->qp_list, list) {
825 		/* Modify the state of all QPs except QP1/Shadow QP */
826 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
827 			if (qp->qplib_qp.state !=
828 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
829 			    qp->qplib_qp.state !=
830 			    CMDQ_MODIFY_QP_NEW_STATE_ERR) {
831 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
832 						       1, IB_EVENT_QP_FATAL);
833 				bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
834 						  NULL);
835 			}
836 		}
837 	}
838 	mutex_unlock(&rdev->qp_lock);
839 }
840 
841 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
842 {
843 	u32 prio_map = 0, tmp_map = 0;
844 	struct net_device *netdev;
845 	struct dcb_app app;
846 
847 	netdev = rdev->netdev;
848 
849 	memset(&app, 0, sizeof(app));
850 	app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
851 	app.protocol = ETH_P_IBOE;
852 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
853 	prio_map = tmp_map;
854 
855 	app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
856 	app.protocol = ROCE_V2_UDP_DPORT;
857 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
858 	prio_map |= tmp_map;
859 
860 	if (!prio_map)
861 		prio_map = -EFAULT;
862 	return prio_map;
863 }
864 
865 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
866 {
867 	u16 prio;
868 	u8 id;
869 
870 	for (prio = 0, id = 0; prio < 8; prio++) {
871 		if (prio_map & (1 << prio)) {
872 			cosq[id] = cid_map[prio];
873 			id++;
874 			if (id == 2) /* Max 2 tcs supported */
875 				break;
876 		}
877 	}
878 }
879 
880 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
881 {
882 	u8 prio_map = 0;
883 	u64 cid_map;
884 	int rc;
885 
886 	/* Get priority for roce */
887 	rc = bnxt_re_get_priority_mask(rdev);
888 	if (rc < 0)
889 		return rc;
890 	prio_map = (u8)rc;
891 
892 	if (prio_map == rdev->cur_prio_map)
893 		return 0;
894 	rdev->cur_prio_map = prio_map;
895 	/* Get cosq id for this priority */
896 	rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
897 	if (rc) {
898 		dev_warn(rdev_to_dev(rdev), "no cos for p_mask %x\n", prio_map);
899 		return rc;
900 	}
901 	/* Parse CoS IDs for app priority */
902 	bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
903 
904 	/* Config BONO. */
905 	rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
906 	if (rc) {
907 		dev_warn(rdev_to_dev(rdev), "no tc for cos{%x, %x}\n",
908 			 rdev->cosq[0], rdev->cosq[1]);
909 		return rc;
910 	}
911 
912 	return 0;
913 }
914 
915 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev, bool lock_wait)
916 {
917 	int i, rc;
918 
919 	if (test_and_clear_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags)) {
920 		for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++)
921 			device_remove_file(&rdev->ibdev.dev,
922 					   bnxt_re_attributes[i]);
923 		/* Cleanup ib dev */
924 		bnxt_re_unregister_ib(rdev);
925 	}
926 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
927 		cancel_delayed_work(&rdev->worker);
928 
929 	bnxt_re_cleanup_res(rdev);
930 	bnxt_re_free_res(rdev, lock_wait);
931 
932 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
933 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
934 		if (rc)
935 			dev_warn(rdev_to_dev(rdev),
936 				 "Failed to deinitialize RCFW: %#x", rc);
937 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id,
938 					   lock_wait);
939 		bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
940 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
941 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, lock_wait);
942 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
943 	}
944 	if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
945 		rc = bnxt_re_free_msix(rdev, lock_wait);
946 		if (rc)
947 			dev_warn(rdev_to_dev(rdev),
948 				 "Failed to free MSI-X vectors: %#x", rc);
949 	}
950 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
951 		rc = bnxt_re_unregister_netdev(rdev, lock_wait);
952 		if (rc)
953 			dev_warn(rdev_to_dev(rdev),
954 				 "Failed to unregister with netdev: %#x", rc);
955 	}
956 }
957 
958 static void bnxt_re_set_resource_limits(struct bnxt_re_dev *rdev)
959 {
960 	u32 i;
961 
962 	rdev->qplib_ctx.qpc_count = BNXT_RE_MAX_QPC_COUNT;
963 	rdev->qplib_ctx.mrw_count = BNXT_RE_MAX_MRW_COUNT;
964 	rdev->qplib_ctx.srqc_count = BNXT_RE_MAX_SRQC_COUNT;
965 	rdev->qplib_ctx.cq_count = BNXT_RE_MAX_CQ_COUNT;
966 	for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
967 		rdev->qplib_ctx.tqm_count[i] =
968 		rdev->dev_attr.tqm_alloc_reqs[i];
969 }
970 
971 /* worker thread for polling periodic events. Now used for QoS programming*/
972 static void bnxt_re_worker(struct work_struct *work)
973 {
974 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
975 						worker.work);
976 
977 	bnxt_re_setup_qos(rdev);
978 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
979 }
980 
981 static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
982 {
983 	int i, j, rc;
984 
985 	/* Registered a new RoCE device instance to netdev */
986 	rc = bnxt_re_register_netdev(rdev);
987 	if (rc) {
988 		pr_err("Failed to register with netedev: %#x\n", rc);
989 		return -EINVAL;
990 	}
991 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
992 
993 	rc = bnxt_re_request_msix(rdev);
994 	if (rc) {
995 		pr_err("Failed to get MSI-X vectors: %#x\n", rc);
996 		rc = -EINVAL;
997 		goto fail;
998 	}
999 	set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
1000 
1001 	/* Establish RCFW Communication Channel to initialize the context
1002 	 * memory for the function and all child VFs
1003 	 */
1004 	rc = bnxt_qplib_alloc_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw);
1005 	if (rc)
1006 		goto fail;
1007 
1008 	rc = bnxt_re_net_ring_alloc
1009 			(rdev, rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr,
1010 			 rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count,
1011 			 HWRM_RING_ALLOC_CMPL, BNXT_QPLIB_CREQE_MAX_CNT - 1,
1012 			 rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx,
1013 			 &rdev->rcfw.creq_ring_id);
1014 	if (rc) {
1015 		pr_err("Failed to allocate CREQ: %#x\n", rc);
1016 		goto free_rcfw;
1017 	}
1018 	rc = bnxt_qplib_enable_rcfw_channel
1019 				(rdev->en_dev->pdev, &rdev->rcfw,
1020 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].vector,
1021 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].db_offset,
1022 				 0, &bnxt_re_aeq_handler);
1023 	if (rc) {
1024 		pr_err("Failed to enable RCFW channel: %#x\n", rc);
1025 		goto free_ring;
1026 	}
1027 
1028 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr);
1029 	if (rc)
1030 		goto disable_rcfw;
1031 	bnxt_re_set_resource_limits(rdev);
1032 
1033 	rc = bnxt_qplib_alloc_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx, 0);
1034 	if (rc) {
1035 		pr_err("Failed to allocate QPLIB context: %#x\n", rc);
1036 		goto disable_rcfw;
1037 	}
1038 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
1039 					 rdev->qplib_ctx.stats.dma_map,
1040 					 &rdev->qplib_ctx.stats.fw_id);
1041 	if (rc) {
1042 		pr_err("Failed to allocate stats context: %#x\n", rc);
1043 		goto free_ctx;
1044 	}
1045 
1046 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx, 0);
1047 	if (rc) {
1048 		pr_err("Failed to initialize RCFW: %#x\n", rc);
1049 		goto free_sctx;
1050 	}
1051 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1052 
1053 	/* Resources based on the 'new' device caps */
1054 	rc = bnxt_re_alloc_res(rdev);
1055 	if (rc) {
1056 		pr_err("Failed to allocate resources: %#x\n", rc);
1057 		goto fail;
1058 	}
1059 	rc = bnxt_re_init_res(rdev);
1060 	if (rc) {
1061 		pr_err("Failed to initialize resources: %#x\n", rc);
1062 		goto fail;
1063 	}
1064 
1065 	rc = bnxt_re_setup_qos(rdev);
1066 	if (rc)
1067 		pr_info("RoCE priority not yet configured\n");
1068 
1069 	INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1070 	set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1071 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1072 
1073 	/* Register ib dev */
1074 	rc = bnxt_re_register_ib(rdev);
1075 	if (rc) {
1076 		pr_err("Failed to register with IB: %#x\n", rc);
1077 		goto fail;
1078 	}
1079 	dev_info(rdev_to_dev(rdev), "Device registered successfully");
1080 	for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++) {
1081 		rc = device_create_file(&rdev->ibdev.dev,
1082 					bnxt_re_attributes[i]);
1083 		if (rc) {
1084 			dev_err(rdev_to_dev(rdev),
1085 				"Failed to create IB sysfs: %#x", rc);
1086 			/* Must clean up all created device files */
1087 			for (j = 0; j < i; j++)
1088 				device_remove_file(&rdev->ibdev.dev,
1089 						   bnxt_re_attributes[j]);
1090 			bnxt_re_unregister_ib(rdev);
1091 			goto fail;
1092 		}
1093 	}
1094 	set_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags);
1095 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_PORT_ACTIVE);
1096 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_GID_CHANGE);
1097 
1098 	return 0;
1099 free_sctx:
1100 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id, true);
1101 free_ctx:
1102 	bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1103 disable_rcfw:
1104 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1105 free_ring:
1106 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, true);
1107 free_rcfw:
1108 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1109 fail:
1110 	bnxt_re_ib_unreg(rdev, true);
1111 	return rc;
1112 }
1113 
1114 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1115 {
1116 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1117 	struct net_device *netdev = rdev->netdev;
1118 
1119 	bnxt_re_dev_remove(rdev);
1120 
1121 	if (netdev)
1122 		bnxt_re_dev_unprobe(netdev, en_dev);
1123 }
1124 
1125 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1126 {
1127 	struct bnxt_en_dev *en_dev;
1128 	int rc = 0;
1129 
1130 	if (!is_bnxt_re_dev(netdev))
1131 		return -ENODEV;
1132 
1133 	en_dev = bnxt_re_dev_probe(netdev);
1134 	if (IS_ERR(en_dev)) {
1135 		if (en_dev != ERR_PTR(-ENODEV))
1136 			pr_err("%s: Failed to probe\n", ROCE_DRV_MODULE_NAME);
1137 		rc = PTR_ERR(en_dev);
1138 		goto exit;
1139 	}
1140 	*rdev = bnxt_re_dev_add(netdev, en_dev);
1141 	if (!*rdev) {
1142 		rc = -ENOMEM;
1143 		bnxt_re_dev_unprobe(netdev, en_dev);
1144 		goto exit;
1145 	}
1146 exit:
1147 	return rc;
1148 }
1149 
1150 static void bnxt_re_remove_one(struct bnxt_re_dev *rdev)
1151 {
1152 	pci_dev_put(rdev->en_dev->pdev);
1153 }
1154 
1155 /* Handle all deferred netevents tasks */
1156 static void bnxt_re_task(struct work_struct *work)
1157 {
1158 	struct bnxt_re_work *re_work;
1159 	struct bnxt_re_dev *rdev;
1160 	int rc = 0;
1161 
1162 	re_work = container_of(work, struct bnxt_re_work, work);
1163 	rdev = re_work->rdev;
1164 
1165 	if (re_work->event != NETDEV_REGISTER &&
1166 	    !test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1167 		return;
1168 
1169 	switch (re_work->event) {
1170 	case NETDEV_REGISTER:
1171 		rc = bnxt_re_ib_reg(rdev);
1172 		if (rc)
1173 			dev_err(rdev_to_dev(rdev),
1174 				"Failed to register with IB: %#x", rc);
1175 		break;
1176 	case NETDEV_UP:
1177 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1178 				       IB_EVENT_PORT_ACTIVE);
1179 		break;
1180 	case NETDEV_DOWN:
1181 		bnxt_re_dev_stop(rdev);
1182 		break;
1183 	case NETDEV_CHANGE:
1184 		if (!netif_carrier_ok(rdev->netdev))
1185 			bnxt_re_dev_stop(rdev);
1186 		else if (netif_carrier_ok(rdev->netdev))
1187 			bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1188 					       IB_EVENT_PORT_ACTIVE);
1189 		break;
1190 	default:
1191 		break;
1192 	}
1193 	kfree(re_work);
1194 }
1195 
1196 static void bnxt_re_init_one(struct bnxt_re_dev *rdev)
1197 {
1198 	pci_dev_get(rdev->en_dev->pdev);
1199 }
1200 
1201 /*
1202  * "Notifier chain callback can be invoked for the same chain from
1203  * different CPUs at the same time".
1204  *
1205  * For cases when the netdev is already present, our call to the
1206  * register_netdevice_notifier() will actually get the rtnl_lock()
1207  * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1208  * events.
1209  *
1210  * But for cases when the netdev is not already present, the notifier
1211  * chain is subjected to be invoked from different CPUs simultaneously.
1212  *
1213  * This is protected by the netdev_mutex.
1214  */
1215 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1216 				unsigned long event, void *ptr)
1217 {
1218 	struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1219 	struct bnxt_re_work *re_work;
1220 	struct bnxt_re_dev *rdev;
1221 	int rc = 0;
1222 	bool sch_work = false;
1223 
1224 	real_dev = rdma_vlan_dev_real_dev(netdev);
1225 	if (!real_dev)
1226 		real_dev = netdev;
1227 
1228 	rdev = bnxt_re_from_netdev(real_dev);
1229 	if (!rdev && event != NETDEV_REGISTER)
1230 		goto exit;
1231 	if (real_dev != netdev)
1232 		goto exit;
1233 
1234 	switch (event) {
1235 	case NETDEV_REGISTER:
1236 		if (rdev)
1237 			break;
1238 		rc = bnxt_re_dev_reg(&rdev, real_dev);
1239 		if (rc == -ENODEV)
1240 			break;
1241 		if (rc) {
1242 			pr_err("Failed to register with the device %s: %#x\n",
1243 			       real_dev->name, rc);
1244 			break;
1245 		}
1246 		bnxt_re_init_one(rdev);
1247 		sch_work = true;
1248 		break;
1249 
1250 	case NETDEV_UNREGISTER:
1251 		bnxt_re_ib_unreg(rdev, false);
1252 		bnxt_re_remove_one(rdev);
1253 		bnxt_re_dev_unreg(rdev);
1254 		break;
1255 
1256 	default:
1257 		sch_work = true;
1258 		break;
1259 	}
1260 	if (sch_work) {
1261 		/* Allocate for the deferred task */
1262 		re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1263 		if (re_work) {
1264 			re_work->rdev = rdev;
1265 			re_work->event = event;
1266 			re_work->vlan_dev = (real_dev == netdev ?
1267 					     NULL : netdev);
1268 			INIT_WORK(&re_work->work, bnxt_re_task);
1269 			queue_work(bnxt_re_wq, &re_work->work);
1270 		}
1271 	}
1272 
1273 exit:
1274 	return NOTIFY_DONE;
1275 }
1276 
1277 static struct notifier_block bnxt_re_netdev_notifier = {
1278 	.notifier_call = bnxt_re_netdev_event
1279 };
1280 
1281 static int __init bnxt_re_mod_init(void)
1282 {
1283 	int rc = 0;
1284 
1285 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1286 
1287 	bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1288 	if (!bnxt_re_wq)
1289 		return -ENOMEM;
1290 
1291 	INIT_LIST_HEAD(&bnxt_re_dev_list);
1292 
1293 	rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1294 	if (rc) {
1295 		pr_err("%s: Cannot register to netdevice_notifier",
1296 		       ROCE_DRV_MODULE_NAME);
1297 		goto err_netdev;
1298 	}
1299 	return 0;
1300 
1301 err_netdev:
1302 	destroy_workqueue(bnxt_re_wq);
1303 
1304 	return rc;
1305 }
1306 
1307 static void __exit bnxt_re_mod_exit(void)
1308 {
1309 	unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1310 	if (bnxt_re_wq)
1311 		destroy_workqueue(bnxt_re_wq);
1312 }
1313 
1314 module_init(bnxt_re_mod_init);
1315 module_exit(bnxt_re_mod_exit);
1316