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 " v" ROCE_DRV_MODULE_VERSION "\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 
82 /* for handling bnxt_en callbacks later */
83 static void bnxt_re_stop(void *p)
84 {
85 }
86 
87 static void bnxt_re_start(void *p)
88 {
89 }
90 
91 static void bnxt_re_sriov_config(void *p, int num_vfs)
92 {
93 }
94 
95 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
96 	.ulp_async_notifier = NULL,
97 	.ulp_stop = bnxt_re_stop,
98 	.ulp_start = bnxt_re_start,
99 	.ulp_sriov_config = bnxt_re_sriov_config
100 };
101 
102 /* RoCE -> Net driver */
103 
104 /* Driver registration routines used to let the networking driver (bnxt_en)
105  * to know that the RoCE driver is now installed
106  */
107 static int bnxt_re_unregister_netdev(struct bnxt_re_dev *rdev, bool lock_wait)
108 {
109 	struct bnxt_en_dev *en_dev;
110 	int rc;
111 
112 	if (!rdev)
113 		return -EINVAL;
114 
115 	en_dev = rdev->en_dev;
116 	/* Acquire rtnl lock if it is not invokded from netdev event */
117 	if (lock_wait)
118 		rtnl_lock();
119 
120 	rc = en_dev->en_ops->bnxt_unregister_device(rdev->en_dev,
121 						    BNXT_ROCE_ULP);
122 	if (lock_wait)
123 		rtnl_unlock();
124 	return rc;
125 }
126 
127 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
128 {
129 	struct bnxt_en_dev *en_dev;
130 	int rc = 0;
131 
132 	if (!rdev)
133 		return -EINVAL;
134 
135 	en_dev = rdev->en_dev;
136 
137 	rtnl_lock();
138 	rc = en_dev->en_ops->bnxt_register_device(en_dev, BNXT_ROCE_ULP,
139 						  &bnxt_re_ulp_ops, rdev);
140 	rtnl_unlock();
141 	return rc;
142 }
143 
144 static int bnxt_re_free_msix(struct bnxt_re_dev *rdev, bool lock_wait)
145 {
146 	struct bnxt_en_dev *en_dev;
147 	int rc;
148 
149 	if (!rdev)
150 		return -EINVAL;
151 
152 	en_dev = rdev->en_dev;
153 
154 	if (lock_wait)
155 		rtnl_lock();
156 
157 	rc = en_dev->en_ops->bnxt_free_msix(rdev->en_dev, BNXT_ROCE_ULP);
158 
159 	if (lock_wait)
160 		rtnl_unlock();
161 	return rc;
162 }
163 
164 static int bnxt_re_request_msix(struct bnxt_re_dev *rdev)
165 {
166 	int rc = 0, num_msix_want = BNXT_RE_MAX_MSIX, num_msix_got;
167 	struct bnxt_en_dev *en_dev;
168 
169 	if (!rdev)
170 		return -EINVAL;
171 
172 	en_dev = rdev->en_dev;
173 
174 	num_msix_want = min_t(u32, BNXT_RE_MAX_MSIX, num_online_cpus());
175 
176 	rtnl_lock();
177 	num_msix_got = en_dev->en_ops->bnxt_request_msix(en_dev, BNXT_ROCE_ULP,
178 							 rdev->msix_entries,
179 							 num_msix_want);
180 	if (num_msix_got < BNXT_RE_MIN_MSIX) {
181 		rc = -EINVAL;
182 		goto done;
183 	}
184 	if (num_msix_got != num_msix_want) {
185 		dev_warn(rdev_to_dev(rdev),
186 			 "Requested %d MSI-X vectors, got %d\n",
187 			 num_msix_want, num_msix_got);
188 	}
189 	rdev->num_msix = num_msix_got;
190 done:
191 	rtnl_unlock();
192 	return rc;
193 }
194 
195 static void bnxt_re_init_hwrm_hdr(struct bnxt_re_dev *rdev, struct input *hdr,
196 				  u16 opcd, u16 crid, u16 trid)
197 {
198 	hdr->req_type = cpu_to_le16(opcd);
199 	hdr->cmpl_ring = cpu_to_le16(crid);
200 	hdr->target_id = cpu_to_le16(trid);
201 }
202 
203 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
204 				int msg_len, void *resp, int resp_max_len,
205 				int timeout)
206 {
207 	fw_msg->msg = msg;
208 	fw_msg->msg_len = msg_len;
209 	fw_msg->resp = resp;
210 	fw_msg->resp_max_len = resp_max_len;
211 	fw_msg->timeout = timeout;
212 }
213 
214 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev, u16 fw_ring_id,
215 				 bool lock_wait)
216 {
217 	struct bnxt_en_dev *en_dev = rdev->en_dev;
218 	struct hwrm_ring_free_input req = {0};
219 	struct hwrm_ring_free_output resp;
220 	struct bnxt_fw_msg fw_msg;
221 	bool do_unlock = false;
222 	int rc = -EINVAL;
223 
224 	if (!en_dev)
225 		return rc;
226 
227 	memset(&fw_msg, 0, sizeof(fw_msg));
228 	if (lock_wait) {
229 		rtnl_lock();
230 		do_unlock = true;
231 	}
232 
233 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1);
234 	req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
235 	req.ring_id = cpu_to_le16(fw_ring_id);
236 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
237 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
238 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
239 	if (rc)
240 		dev_err(rdev_to_dev(rdev),
241 			"Failed to free HW ring:%d :%#x", req.ring_id, rc);
242 	if (do_unlock)
243 		rtnl_unlock();
244 	return rc;
245 }
246 
247 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev, dma_addr_t *dma_arr,
248 				  int pages, int type, u32 ring_mask,
249 				  u32 map_index, u16 *fw_ring_id)
250 {
251 	struct bnxt_en_dev *en_dev = rdev->en_dev;
252 	struct hwrm_ring_alloc_input req = {0};
253 	struct hwrm_ring_alloc_output resp;
254 	struct bnxt_fw_msg fw_msg;
255 	int rc = -EINVAL;
256 
257 	if (!en_dev)
258 		return rc;
259 
260 	memset(&fw_msg, 0, sizeof(fw_msg));
261 	rtnl_lock();
262 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_ALLOC, -1, -1);
263 	req.enables = 0;
264 	req.page_tbl_addr =  cpu_to_le64(dma_arr[0]);
265 	if (pages > 1) {
266 		/* Page size is in log2 units */
267 		req.page_size = BNXT_PAGE_SHIFT;
268 		req.page_tbl_depth = 1;
269 	}
270 	req.fbo = 0;
271 	/* Association of ring index with doorbell index and MSIX number */
272 	req.logical_id = cpu_to_le16(map_index);
273 	req.length = cpu_to_le32(ring_mask + 1);
274 	req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
275 	req.int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
276 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
277 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
278 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
279 	if (!rc)
280 		*fw_ring_id = le16_to_cpu(resp.ring_id);
281 
282 	rtnl_unlock();
283 	return rc;
284 }
285 
286 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
287 				      u32 fw_stats_ctx_id, bool lock_wait)
288 {
289 	struct bnxt_en_dev *en_dev = rdev->en_dev;
290 	struct hwrm_stat_ctx_free_input req = {0};
291 	struct bnxt_fw_msg fw_msg;
292 	bool do_unlock = false;
293 	int rc = -EINVAL;
294 
295 	if (!en_dev)
296 		return rc;
297 
298 	memset(&fw_msg, 0, sizeof(fw_msg));
299 	if (lock_wait) {
300 		rtnl_lock();
301 		do_unlock = true;
302 	}
303 
304 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_FREE, -1, -1);
305 	req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
306 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&req,
307 			    sizeof(req), DFLT_HWRM_CMD_TIMEOUT);
308 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
309 	if (rc)
310 		dev_err(rdev_to_dev(rdev),
311 			"Failed to free HW stats context %#x", rc);
312 
313 	if (do_unlock)
314 		rtnl_unlock();
315 	return rc;
316 }
317 
318 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
319 				       dma_addr_t dma_map,
320 				       u32 *fw_stats_ctx_id)
321 {
322 	struct hwrm_stat_ctx_alloc_output resp = {0};
323 	struct hwrm_stat_ctx_alloc_input req = {0};
324 	struct bnxt_en_dev *en_dev = rdev->en_dev;
325 	struct bnxt_fw_msg fw_msg;
326 	int rc = -EINVAL;
327 
328 	*fw_stats_ctx_id = INVALID_STATS_CTX_ID;
329 
330 	if (!en_dev)
331 		return rc;
332 
333 	memset(&fw_msg, 0, sizeof(fw_msg));
334 	rtnl_lock();
335 
336 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1);
337 	req.update_period_ms = cpu_to_le32(1000);
338 	req.stats_dma_addr = cpu_to_le64(dma_map);
339 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
340 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
341 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
342 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
343 	if (!rc)
344 		*fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
345 
346 	rtnl_unlock();
347 	return rc;
348 }
349 
350 /* Device */
351 
352 static bool is_bnxt_re_dev(struct net_device *netdev)
353 {
354 	struct ethtool_drvinfo drvinfo;
355 
356 	if (netdev->ethtool_ops && netdev->ethtool_ops->get_drvinfo) {
357 		memset(&drvinfo, 0, sizeof(drvinfo));
358 		netdev->ethtool_ops->get_drvinfo(netdev, &drvinfo);
359 
360 		if (strcmp(drvinfo.driver, "bnxt_en"))
361 			return false;
362 		return true;
363 	}
364 	return false;
365 }
366 
367 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
368 {
369 	struct bnxt_re_dev *rdev;
370 
371 	rcu_read_lock();
372 	list_for_each_entry_rcu(rdev, &bnxt_re_dev_list, list) {
373 		if (rdev->netdev == netdev) {
374 			rcu_read_unlock();
375 			return rdev;
376 		}
377 	}
378 	rcu_read_unlock();
379 	return NULL;
380 }
381 
382 static void bnxt_re_dev_unprobe(struct net_device *netdev,
383 				struct bnxt_en_dev *en_dev)
384 {
385 	dev_put(netdev);
386 	module_put(en_dev->pdev->driver->driver.owner);
387 }
388 
389 static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev)
390 {
391 	struct bnxt *bp = netdev_priv(netdev);
392 	struct bnxt_en_dev *en_dev;
393 	struct pci_dev *pdev;
394 
395 	/* Call bnxt_en's RoCE probe via indirect API */
396 	if (!bp->ulp_probe)
397 		return ERR_PTR(-EINVAL);
398 
399 	en_dev = bp->ulp_probe(netdev);
400 	if (IS_ERR(en_dev))
401 		return en_dev;
402 
403 	pdev = en_dev->pdev;
404 	if (!pdev)
405 		return ERR_PTR(-EINVAL);
406 
407 	if (!(en_dev->flags & BNXT_EN_FLAG_ROCE_CAP)) {
408 		dev_dbg(&pdev->dev,
409 			"%s: probe error: RoCE is not supported on this device",
410 			ROCE_DRV_MODULE_NAME);
411 		return ERR_PTR(-ENODEV);
412 	}
413 
414 	/* Bump net device reference count */
415 	if (!try_module_get(pdev->driver->driver.owner))
416 		return ERR_PTR(-ENODEV);
417 
418 	dev_hold(netdev);
419 
420 	return en_dev;
421 }
422 
423 static void bnxt_re_unregister_ib(struct bnxt_re_dev *rdev)
424 {
425 	ib_unregister_device(&rdev->ibdev);
426 }
427 
428 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
429 {
430 	struct ib_device *ibdev = &rdev->ibdev;
431 
432 	/* ib device init */
433 	ibdev->owner = THIS_MODULE;
434 	ibdev->node_type = RDMA_NODE_IB_CA;
435 	strlcpy(ibdev->name, "bnxt_re%d", IB_DEVICE_NAME_MAX);
436 	strlcpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
437 		strlen(BNXT_RE_DESC) + 5);
438 	ibdev->phys_port_cnt = 1;
439 
440 	bnxt_qplib_get_guid(rdev->netdev->dev_addr, (u8 *)&ibdev->node_guid);
441 
442 	ibdev->num_comp_vectors	= 1;
443 	ibdev->dev.parent = &rdev->en_dev->pdev->dev;
444 	ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
445 
446 	/* User space */
447 	ibdev->uverbs_abi_ver = BNXT_RE_ABI_VERSION;
448 	ibdev->uverbs_cmd_mask =
449 			(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
450 			(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
451 			(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
452 			(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
453 			(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
454 			(1ull << IB_USER_VERBS_CMD_REG_MR)		|
455 			(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
456 			(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
457 			(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
458 			(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
459 			(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
460 			(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
461 			(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
462 			(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
463 			(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
464 			(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
465 			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
466 			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
467 			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
468 			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
469 			(1ull << IB_USER_VERBS_CMD_CREATE_AH)		|
470 			(1ull << IB_USER_VERBS_CMD_MODIFY_AH)		|
471 			(1ull << IB_USER_VERBS_CMD_QUERY_AH)		|
472 			(1ull << IB_USER_VERBS_CMD_DESTROY_AH);
473 	/* POLL_CQ and REQ_NOTIFY_CQ is directly handled in libbnxt_re */
474 
475 	/* Kernel verbs */
476 	ibdev->query_device		= bnxt_re_query_device;
477 	ibdev->modify_device		= bnxt_re_modify_device;
478 
479 	ibdev->query_port		= bnxt_re_query_port;
480 	ibdev->get_port_immutable	= bnxt_re_get_port_immutable;
481 	ibdev->query_pkey		= bnxt_re_query_pkey;
482 	ibdev->query_gid		= bnxt_re_query_gid;
483 	ibdev->get_netdev		= bnxt_re_get_netdev;
484 	ibdev->add_gid			= bnxt_re_add_gid;
485 	ibdev->del_gid			= bnxt_re_del_gid;
486 	ibdev->get_link_layer		= bnxt_re_get_link_layer;
487 
488 	ibdev->alloc_pd			= bnxt_re_alloc_pd;
489 	ibdev->dealloc_pd		= bnxt_re_dealloc_pd;
490 
491 	ibdev->create_ah		= bnxt_re_create_ah;
492 	ibdev->modify_ah		= bnxt_re_modify_ah;
493 	ibdev->query_ah			= bnxt_re_query_ah;
494 	ibdev->destroy_ah		= bnxt_re_destroy_ah;
495 
496 	ibdev->create_qp		= bnxt_re_create_qp;
497 	ibdev->modify_qp		= bnxt_re_modify_qp;
498 	ibdev->query_qp			= bnxt_re_query_qp;
499 	ibdev->destroy_qp		= bnxt_re_destroy_qp;
500 
501 	ibdev->post_send		= bnxt_re_post_send;
502 	ibdev->post_recv		= bnxt_re_post_recv;
503 
504 	ibdev->create_cq		= bnxt_re_create_cq;
505 	ibdev->destroy_cq		= bnxt_re_destroy_cq;
506 	ibdev->poll_cq			= bnxt_re_poll_cq;
507 	ibdev->req_notify_cq		= bnxt_re_req_notify_cq;
508 
509 	ibdev->get_dma_mr		= bnxt_re_get_dma_mr;
510 	ibdev->dereg_mr			= bnxt_re_dereg_mr;
511 	ibdev->alloc_mr			= bnxt_re_alloc_mr;
512 	ibdev->map_mr_sg		= bnxt_re_map_mr_sg;
513 
514 	ibdev->reg_user_mr		= bnxt_re_reg_user_mr;
515 	ibdev->alloc_ucontext		= bnxt_re_alloc_ucontext;
516 	ibdev->dealloc_ucontext		= bnxt_re_dealloc_ucontext;
517 	ibdev->mmap			= bnxt_re_mmap;
518 	ibdev->get_hw_stats             = bnxt_re_ib_get_hw_stats;
519 	ibdev->alloc_hw_stats           = bnxt_re_ib_alloc_hw_stats;
520 
521 	return ib_register_device(ibdev, NULL);
522 }
523 
524 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
525 			char *buf)
526 {
527 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
528 
529 	return scnprintf(buf, PAGE_SIZE, "0x%x\n", rdev->en_dev->pdev->vendor);
530 }
531 
532 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
533 			   char *buf)
534 {
535 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
536 
537 	return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->dev_attr.fw_ver);
538 }
539 
540 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
541 			char *buf)
542 {
543 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
544 
545 	return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->ibdev.node_desc);
546 }
547 
548 static DEVICE_ATTR(hw_rev, 0444, show_rev, NULL);
549 static DEVICE_ATTR(fw_rev, 0444, show_fw_ver, NULL);
550 static DEVICE_ATTR(hca_type, 0444, show_hca, NULL);
551 
552 static struct device_attribute *bnxt_re_attributes[] = {
553 	&dev_attr_hw_rev,
554 	&dev_attr_fw_rev,
555 	&dev_attr_hca_type
556 };
557 
558 static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev)
559 {
560 	dev_put(rdev->netdev);
561 	rdev->netdev = NULL;
562 
563 	mutex_lock(&bnxt_re_dev_lock);
564 	list_del_rcu(&rdev->list);
565 	mutex_unlock(&bnxt_re_dev_lock);
566 
567 	synchronize_rcu();
568 	flush_workqueue(bnxt_re_wq);
569 
570 	ib_dealloc_device(&rdev->ibdev);
571 	/* rdev is gone */
572 }
573 
574 static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev,
575 					   struct bnxt_en_dev *en_dev)
576 {
577 	struct bnxt_re_dev *rdev;
578 
579 	/* Allocate bnxt_re_dev instance here */
580 	rdev = (struct bnxt_re_dev *)ib_alloc_device(sizeof(*rdev));
581 	if (!rdev) {
582 		dev_err(NULL, "%s: bnxt_re_dev allocation failure!",
583 			ROCE_DRV_MODULE_NAME);
584 		return NULL;
585 	}
586 	/* Default values */
587 	rdev->netdev = netdev;
588 	dev_hold(rdev->netdev);
589 	rdev->en_dev = en_dev;
590 	rdev->id = rdev->en_dev->pdev->devfn;
591 	INIT_LIST_HEAD(&rdev->qp_list);
592 	mutex_init(&rdev->qp_lock);
593 	atomic_set(&rdev->qp_count, 0);
594 	atomic_set(&rdev->cq_count, 0);
595 	atomic_set(&rdev->srq_count, 0);
596 	atomic_set(&rdev->mr_count, 0);
597 	atomic_set(&rdev->mw_count, 0);
598 	rdev->cosq[0] = 0xFFFF;
599 	rdev->cosq[1] = 0xFFFF;
600 
601 	mutex_lock(&bnxt_re_dev_lock);
602 	list_add_tail_rcu(&rdev->list, &bnxt_re_dev_list);
603 	mutex_unlock(&bnxt_re_dev_lock);
604 	return rdev;
605 }
606 
607 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
608 			       struct creq_func_event *aeqe)
609 {
610 	switch (aeqe->event) {
611 	case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
612 		break;
613 	case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
614 		break;
615 	case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
616 		break;
617 	case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
618 		break;
619 	case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
620 		break;
621 	case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
622 		break;
623 	case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
624 		break;
625 	case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
626 		break;
627 	case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
628 		break;
629 	case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
630 		break;
631 	case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
632 		break;
633 	default:
634 		return -EINVAL;
635 	}
636 	return 0;
637 }
638 
639 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
640 			       struct bnxt_qplib_cq *handle)
641 {
642 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
643 					     qplib_cq);
644 
645 	if (!cq) {
646 		dev_err(NULL, "%s: CQ is NULL, CQN not handled",
647 			ROCE_DRV_MODULE_NAME);
648 		return -EINVAL;
649 	}
650 	if (cq->ib_cq.comp_handler) {
651 		/* Lock comp_handler? */
652 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
653 	}
654 
655 	return 0;
656 }
657 
658 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
659 {
660 	int i;
661 
662 	if (rdev->nq[0].hwq.max_elements) {
663 		for (i = 1; i < rdev->num_msix; i++)
664 			bnxt_qplib_disable_nq(&rdev->nq[i - 1]);
665 	}
666 
667 	if (rdev->qplib_res.rcfw)
668 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
669 }
670 
671 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
672 {
673 	int rc = 0, i;
674 
675 	bnxt_qplib_init_res(&rdev->qplib_res);
676 
677 	for (i = 1; i < rdev->num_msix ; i++) {
678 		rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
679 					  i - 1, rdev->msix_entries[i].vector,
680 					  rdev->msix_entries[i].db_offset,
681 					  &bnxt_re_cqn_handler, NULL);
682 
683 		if (rc) {
684 			dev_err(rdev_to_dev(rdev),
685 				"Failed to enable NQ with rc = 0x%x", rc);
686 			goto fail;
687 		}
688 	}
689 	return 0;
690 fail:
691 	return rc;
692 }
693 
694 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev, bool lock_wait)
695 {
696 	int i;
697 
698 	for (i = 0; i < rdev->num_msix - 1; i++) {
699 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, lock_wait);
700 		bnxt_qplib_free_nq(&rdev->nq[i]);
701 	}
702 }
703 
704 static void bnxt_re_free_res(struct bnxt_re_dev *rdev, bool lock_wait)
705 {
706 	bnxt_re_free_nq_res(rdev, lock_wait);
707 
708 	if (rdev->qplib_res.dpi_tbl.max) {
709 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
710 				       &rdev->qplib_res.dpi_tbl,
711 				       &rdev->dpi_privileged);
712 	}
713 	if (rdev->qplib_res.rcfw) {
714 		bnxt_qplib_free_res(&rdev->qplib_res);
715 		rdev->qplib_res.rcfw = NULL;
716 	}
717 }
718 
719 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
720 {
721 	int rc = 0, i;
722 
723 	/* Configure and allocate resources for qplib */
724 	rdev->qplib_res.rcfw = &rdev->rcfw;
725 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr);
726 	if (rc)
727 		goto fail;
728 
729 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
730 				  rdev->netdev, &rdev->dev_attr);
731 	if (rc)
732 		goto fail;
733 
734 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
735 				  &rdev->dpi_privileged,
736 				  rdev);
737 	if (rc)
738 		goto dealloc_res;
739 
740 	for (i = 0; i < rdev->num_msix - 1; i++) {
741 		rdev->nq[i].hwq.max_elements = BNXT_RE_MAX_CQ_COUNT +
742 			BNXT_RE_MAX_SRQC_COUNT + 2;
743 		rc = bnxt_qplib_alloc_nq(rdev->en_dev->pdev, &rdev->nq[i]);
744 		if (rc) {
745 			dev_err(rdev_to_dev(rdev), "Alloc Failed NQ%d rc:%#x",
746 				i, rc);
747 			goto dealloc_dpi;
748 		}
749 		rc = bnxt_re_net_ring_alloc
750 			(rdev, rdev->nq[i].hwq.pbl[PBL_LVL_0].pg_map_arr,
751 			 rdev->nq[i].hwq.pbl[rdev->nq[i].hwq.level].pg_count,
752 			 HWRM_RING_ALLOC_CMPL,
753 			 BNXT_QPLIB_NQE_MAX_CNT - 1,
754 			 rdev->msix_entries[i + 1].ring_idx,
755 			 &rdev->nq[i].ring_id);
756 		if (rc) {
757 			dev_err(rdev_to_dev(rdev),
758 				"Failed to allocate NQ fw id with rc = 0x%x",
759 				rc);
760 			goto free_nq;
761 		}
762 	}
763 	return 0;
764 free_nq:
765 	for (i = 0; i < rdev->num_msix - 1; i++)
766 		bnxt_qplib_free_nq(&rdev->nq[i]);
767 dealloc_dpi:
768 	bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
769 			       &rdev->qplib_res.dpi_tbl,
770 			       &rdev->dpi_privileged);
771 dealloc_res:
772 	bnxt_qplib_free_res(&rdev->qplib_res);
773 
774 fail:
775 	rdev->qplib_res.rcfw = NULL;
776 	return rc;
777 }
778 
779 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
780 				   u8 port_num, enum ib_event_type event)
781 {
782 	struct ib_event ib_event;
783 
784 	ib_event.device = ibdev;
785 	if (qp)
786 		ib_event.element.qp = qp;
787 	else
788 		ib_event.element.port_num = port_num;
789 	ib_event.event = event;
790 	ib_dispatch_event(&ib_event);
791 }
792 
793 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN      0x02
794 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
795 				      u64 *cid_map)
796 {
797 	struct hwrm_queue_pri2cos_qcfg_input req = {0};
798 	struct bnxt *bp = netdev_priv(rdev->netdev);
799 	struct hwrm_queue_pri2cos_qcfg_output resp;
800 	struct bnxt_en_dev *en_dev = rdev->en_dev;
801 	struct bnxt_fw_msg fw_msg;
802 	u32 flags = 0;
803 	u8 *qcfgmap, *tmp_map;
804 	int rc = 0, i;
805 
806 	if (!cid_map)
807 		return -EINVAL;
808 
809 	memset(&fw_msg, 0, sizeof(fw_msg));
810 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
811 			      HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
812 	flags |= (dir & 0x01);
813 	flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
814 	req.flags = cpu_to_le32(flags);
815 	req.port_id = bp->pf.port_id;
816 
817 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
818 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
819 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
820 	if (rc)
821 		return rc;
822 
823 	if (resp.queue_cfg_info) {
824 		dev_warn(rdev_to_dev(rdev),
825 			 "Asymmetric cos queue configuration detected");
826 		dev_warn(rdev_to_dev(rdev),
827 			 " on device, QoS may not be fully functional\n");
828 	}
829 	qcfgmap = &resp.pri0_cos_queue_id;
830 	tmp_map = (u8 *)cid_map;
831 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
832 		tmp_map[i] = qcfgmap[i];
833 
834 	return rc;
835 }
836 
837 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
838 					struct bnxt_re_qp *qp)
839 {
840 	return (qp->ib_qp.qp_type == IB_QPT_GSI) || (qp == rdev->qp1_sqp);
841 }
842 
843 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
844 {
845 	int mask = IB_QP_STATE;
846 	struct ib_qp_attr qp_attr;
847 	struct bnxt_re_qp *qp;
848 
849 	qp_attr.qp_state = IB_QPS_ERR;
850 	mutex_lock(&rdev->qp_lock);
851 	list_for_each_entry(qp, &rdev->qp_list, list) {
852 		/* Modify the state of all QPs except QP1/Shadow QP */
853 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
854 			if (qp->qplib_qp.state !=
855 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
856 			    qp->qplib_qp.state !=
857 			    CMDQ_MODIFY_QP_NEW_STATE_ERR) {
858 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
859 						       1, IB_EVENT_QP_FATAL);
860 				bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
861 						  NULL);
862 			}
863 		}
864 	}
865 	mutex_unlock(&rdev->qp_lock);
866 }
867 
868 static int bnxt_re_update_gid(struct bnxt_re_dev *rdev)
869 {
870 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
871 	struct bnxt_qplib_gid gid;
872 	u16 gid_idx, index;
873 	int rc = 0;
874 
875 	if (!test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
876 		return 0;
877 
878 	if (!sgid_tbl) {
879 		dev_err(rdev_to_dev(rdev), "QPLIB: SGID table not allocated");
880 		return -EINVAL;
881 	}
882 
883 	for (index = 0; index < sgid_tbl->active; index++) {
884 		gid_idx = sgid_tbl->hw_id[index];
885 
886 		if (!memcmp(&sgid_tbl->tbl[index], &bnxt_qplib_gid_zero,
887 			    sizeof(bnxt_qplib_gid_zero)))
888 			continue;
889 		/* need to modify the VLAN enable setting of non VLAN GID only
890 		 * as setting is done for VLAN GID while adding GID
891 		 */
892 		if (sgid_tbl->vlan[index])
893 			continue;
894 
895 		memcpy(&gid, &sgid_tbl->tbl[index], sizeof(gid));
896 
897 		rc = bnxt_qplib_update_sgid(sgid_tbl, &gid, gid_idx,
898 					    rdev->qplib_res.netdev->dev_addr);
899 	}
900 
901 	return rc;
902 }
903 
904 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
905 {
906 	u32 prio_map = 0, tmp_map = 0;
907 	struct net_device *netdev;
908 	struct dcb_app app;
909 
910 	netdev = rdev->netdev;
911 
912 	memset(&app, 0, sizeof(app));
913 	app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
914 	app.protocol = ETH_P_IBOE;
915 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
916 	prio_map = tmp_map;
917 
918 	app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
919 	app.protocol = ROCE_V2_UDP_DPORT;
920 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
921 	prio_map |= tmp_map;
922 
923 	return prio_map;
924 }
925 
926 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
927 {
928 	u16 prio;
929 	u8 id;
930 
931 	for (prio = 0, id = 0; prio < 8; prio++) {
932 		if (prio_map & (1 << prio)) {
933 			cosq[id] = cid_map[prio];
934 			id++;
935 			if (id == 2) /* Max 2 tcs supported */
936 				break;
937 		}
938 	}
939 }
940 
941 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
942 {
943 	u8 prio_map = 0;
944 	u64 cid_map;
945 	int rc;
946 
947 	/* Get priority for roce */
948 	prio_map = bnxt_re_get_priority_mask(rdev);
949 
950 	if (prio_map == rdev->cur_prio_map)
951 		return 0;
952 	rdev->cur_prio_map = prio_map;
953 	/* Get cosq id for this priority */
954 	rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
955 	if (rc) {
956 		dev_warn(rdev_to_dev(rdev), "no cos for p_mask %x\n", prio_map);
957 		return rc;
958 	}
959 	/* Parse CoS IDs for app priority */
960 	bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
961 
962 	/* Config BONO. */
963 	rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
964 	if (rc) {
965 		dev_warn(rdev_to_dev(rdev), "no tc for cos{%x, %x}\n",
966 			 rdev->cosq[0], rdev->cosq[1]);
967 		return rc;
968 	}
969 
970 	/* Actual priorities are not programmed as they are already
971 	 * done by L2 driver; just enable or disable priority vlan tagging
972 	 */
973 	if ((prio_map == 0 && rdev->qplib_res.prio) ||
974 	    (prio_map != 0 && !rdev->qplib_res.prio)) {
975 		rdev->qplib_res.prio = prio_map ? true : false;
976 
977 		bnxt_re_update_gid(rdev);
978 	}
979 
980 	return 0;
981 }
982 
983 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev, bool lock_wait)
984 {
985 	int i, rc;
986 
987 	if (test_and_clear_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags)) {
988 		for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++)
989 			device_remove_file(&rdev->ibdev.dev,
990 					   bnxt_re_attributes[i]);
991 		/* Cleanup ib dev */
992 		bnxt_re_unregister_ib(rdev);
993 	}
994 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
995 		cancel_delayed_work(&rdev->worker);
996 
997 	bnxt_re_cleanup_res(rdev);
998 	bnxt_re_free_res(rdev, lock_wait);
999 
1000 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
1001 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
1002 		if (rc)
1003 			dev_warn(rdev_to_dev(rdev),
1004 				 "Failed to deinitialize RCFW: %#x", rc);
1005 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id,
1006 					   lock_wait);
1007 		bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1008 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1009 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, lock_wait);
1010 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1011 	}
1012 	if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
1013 		rc = bnxt_re_free_msix(rdev, lock_wait);
1014 		if (rc)
1015 			dev_warn(rdev_to_dev(rdev),
1016 				 "Failed to free MSI-X vectors: %#x", rc);
1017 	}
1018 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
1019 		rc = bnxt_re_unregister_netdev(rdev, lock_wait);
1020 		if (rc)
1021 			dev_warn(rdev_to_dev(rdev),
1022 				 "Failed to unregister with netdev: %#x", rc);
1023 	}
1024 }
1025 
1026 static void bnxt_re_set_resource_limits(struct bnxt_re_dev *rdev)
1027 {
1028 	u32 i;
1029 
1030 	rdev->qplib_ctx.qpc_count = BNXT_RE_MAX_QPC_COUNT;
1031 	rdev->qplib_ctx.mrw_count = BNXT_RE_MAX_MRW_COUNT;
1032 	rdev->qplib_ctx.srqc_count = BNXT_RE_MAX_SRQC_COUNT;
1033 	rdev->qplib_ctx.cq_count = BNXT_RE_MAX_CQ_COUNT;
1034 	for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
1035 		rdev->qplib_ctx.tqm_count[i] =
1036 		rdev->dev_attr.tqm_alloc_reqs[i];
1037 }
1038 
1039 /* worker thread for polling periodic events. Now used for QoS programming*/
1040 static void bnxt_re_worker(struct work_struct *work)
1041 {
1042 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
1043 						worker.work);
1044 
1045 	bnxt_re_setup_qos(rdev);
1046 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1047 }
1048 
1049 static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
1050 {
1051 	int i, j, rc;
1052 
1053 	/* Registered a new RoCE device instance to netdev */
1054 	rc = bnxt_re_register_netdev(rdev);
1055 	if (rc) {
1056 		pr_err("Failed to register with netedev: %#x\n", rc);
1057 		return -EINVAL;
1058 	}
1059 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1060 
1061 	rc = bnxt_re_request_msix(rdev);
1062 	if (rc) {
1063 		pr_err("Failed to get MSI-X vectors: %#x\n", rc);
1064 		rc = -EINVAL;
1065 		goto fail;
1066 	}
1067 	set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
1068 
1069 	/* Establish RCFW Communication Channel to initialize the context
1070 	 * memory for the function and all child VFs
1071 	 */
1072 	rc = bnxt_qplib_alloc_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw,
1073 					   BNXT_RE_MAX_QPC_COUNT);
1074 	if (rc)
1075 		goto fail;
1076 
1077 	rc = bnxt_re_net_ring_alloc
1078 			(rdev, rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr,
1079 			 rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count,
1080 			 HWRM_RING_ALLOC_CMPL, BNXT_QPLIB_CREQE_MAX_CNT - 1,
1081 			 rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx,
1082 			 &rdev->rcfw.creq_ring_id);
1083 	if (rc) {
1084 		pr_err("Failed to allocate CREQ: %#x\n", rc);
1085 		goto free_rcfw;
1086 	}
1087 	rc = bnxt_qplib_enable_rcfw_channel
1088 				(rdev->en_dev->pdev, &rdev->rcfw,
1089 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].vector,
1090 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].db_offset,
1091 				 0, &bnxt_re_aeq_handler);
1092 	if (rc) {
1093 		pr_err("Failed to enable RCFW channel: %#x\n", rc);
1094 		goto free_ring;
1095 	}
1096 
1097 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr);
1098 	if (rc)
1099 		goto disable_rcfw;
1100 	bnxt_re_set_resource_limits(rdev);
1101 
1102 	rc = bnxt_qplib_alloc_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx, 0);
1103 	if (rc) {
1104 		pr_err("Failed to allocate QPLIB context: %#x\n", rc);
1105 		goto disable_rcfw;
1106 	}
1107 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
1108 					 rdev->qplib_ctx.stats.dma_map,
1109 					 &rdev->qplib_ctx.stats.fw_id);
1110 	if (rc) {
1111 		pr_err("Failed to allocate stats context: %#x\n", rc);
1112 		goto free_ctx;
1113 	}
1114 
1115 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx, 0);
1116 	if (rc) {
1117 		pr_err("Failed to initialize RCFW: %#x\n", rc);
1118 		goto free_sctx;
1119 	}
1120 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1121 
1122 	/* Resources based on the 'new' device caps */
1123 	rc = bnxt_re_alloc_res(rdev);
1124 	if (rc) {
1125 		pr_err("Failed to allocate resources: %#x\n", rc);
1126 		goto fail;
1127 	}
1128 	rc = bnxt_re_init_res(rdev);
1129 	if (rc) {
1130 		pr_err("Failed to initialize resources: %#x\n", rc);
1131 		goto fail;
1132 	}
1133 
1134 	rc = bnxt_re_setup_qos(rdev);
1135 	if (rc)
1136 		pr_info("RoCE priority not yet configured\n");
1137 
1138 	INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1139 	set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1140 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1141 
1142 	/* Register ib dev */
1143 	rc = bnxt_re_register_ib(rdev);
1144 	if (rc) {
1145 		pr_err("Failed to register with IB: %#x\n", rc);
1146 		goto fail;
1147 	}
1148 	dev_info(rdev_to_dev(rdev), "Device registered successfully");
1149 	for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++) {
1150 		rc = device_create_file(&rdev->ibdev.dev,
1151 					bnxt_re_attributes[i]);
1152 		if (rc) {
1153 			dev_err(rdev_to_dev(rdev),
1154 				"Failed to create IB sysfs: %#x", rc);
1155 			/* Must clean up all created device files */
1156 			for (j = 0; j < i; j++)
1157 				device_remove_file(&rdev->ibdev.dev,
1158 						   bnxt_re_attributes[j]);
1159 			bnxt_re_unregister_ib(rdev);
1160 			goto fail;
1161 		}
1162 	}
1163 	set_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags);
1164 	ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1165 			 &rdev->active_width);
1166 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_PORT_ACTIVE);
1167 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_GID_CHANGE);
1168 
1169 	return 0;
1170 free_sctx:
1171 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id, true);
1172 free_ctx:
1173 	bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1174 disable_rcfw:
1175 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1176 free_ring:
1177 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, true);
1178 free_rcfw:
1179 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1180 fail:
1181 	bnxt_re_ib_unreg(rdev, true);
1182 	return rc;
1183 }
1184 
1185 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1186 {
1187 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1188 	struct net_device *netdev = rdev->netdev;
1189 
1190 	bnxt_re_dev_remove(rdev);
1191 
1192 	if (netdev)
1193 		bnxt_re_dev_unprobe(netdev, en_dev);
1194 }
1195 
1196 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1197 {
1198 	struct bnxt_en_dev *en_dev;
1199 	int rc = 0;
1200 
1201 	if (!is_bnxt_re_dev(netdev))
1202 		return -ENODEV;
1203 
1204 	en_dev = bnxt_re_dev_probe(netdev);
1205 	if (IS_ERR(en_dev)) {
1206 		if (en_dev != ERR_PTR(-ENODEV))
1207 			pr_err("%s: Failed to probe\n", ROCE_DRV_MODULE_NAME);
1208 		rc = PTR_ERR(en_dev);
1209 		goto exit;
1210 	}
1211 	*rdev = bnxt_re_dev_add(netdev, en_dev);
1212 	if (!*rdev) {
1213 		rc = -ENOMEM;
1214 		bnxt_re_dev_unprobe(netdev, en_dev);
1215 		goto exit;
1216 	}
1217 exit:
1218 	return rc;
1219 }
1220 
1221 static void bnxt_re_remove_one(struct bnxt_re_dev *rdev)
1222 {
1223 	pci_dev_put(rdev->en_dev->pdev);
1224 }
1225 
1226 /* Handle all deferred netevents tasks */
1227 static void bnxt_re_task(struct work_struct *work)
1228 {
1229 	struct bnxt_re_work *re_work;
1230 	struct bnxt_re_dev *rdev;
1231 	int rc = 0;
1232 
1233 	re_work = container_of(work, struct bnxt_re_work, work);
1234 	rdev = re_work->rdev;
1235 
1236 	if (re_work->event != NETDEV_REGISTER &&
1237 	    !test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1238 		return;
1239 
1240 	switch (re_work->event) {
1241 	case NETDEV_REGISTER:
1242 		rc = bnxt_re_ib_reg(rdev);
1243 		if (rc)
1244 			dev_err(rdev_to_dev(rdev),
1245 				"Failed to register with IB: %#x", rc);
1246 		break;
1247 	case NETDEV_UP:
1248 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1249 				       IB_EVENT_PORT_ACTIVE);
1250 		break;
1251 	case NETDEV_DOWN:
1252 		bnxt_re_dev_stop(rdev);
1253 		break;
1254 	case NETDEV_CHANGE:
1255 		if (!netif_carrier_ok(rdev->netdev))
1256 			bnxt_re_dev_stop(rdev);
1257 		else if (netif_carrier_ok(rdev->netdev))
1258 			bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1259 					       IB_EVENT_PORT_ACTIVE);
1260 		ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1261 				 &rdev->active_width);
1262 		break;
1263 	default:
1264 		break;
1265 	}
1266 	smp_mb__before_atomic();
1267 	clear_bit(BNXT_RE_FLAG_TASK_IN_PROG, &rdev->flags);
1268 	kfree(re_work);
1269 }
1270 
1271 static void bnxt_re_init_one(struct bnxt_re_dev *rdev)
1272 {
1273 	pci_dev_get(rdev->en_dev->pdev);
1274 }
1275 
1276 /*
1277  * "Notifier chain callback can be invoked for the same chain from
1278  * different CPUs at the same time".
1279  *
1280  * For cases when the netdev is already present, our call to the
1281  * register_netdevice_notifier() will actually get the rtnl_lock()
1282  * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1283  * events.
1284  *
1285  * But for cases when the netdev is not already present, the notifier
1286  * chain is subjected to be invoked from different CPUs simultaneously.
1287  *
1288  * This is protected by the netdev_mutex.
1289  */
1290 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1291 				unsigned long event, void *ptr)
1292 {
1293 	struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1294 	struct bnxt_re_work *re_work;
1295 	struct bnxt_re_dev *rdev;
1296 	int rc = 0;
1297 	bool sch_work = false;
1298 
1299 	real_dev = rdma_vlan_dev_real_dev(netdev);
1300 	if (!real_dev)
1301 		real_dev = netdev;
1302 
1303 	rdev = bnxt_re_from_netdev(real_dev);
1304 	if (!rdev && event != NETDEV_REGISTER)
1305 		goto exit;
1306 	if (real_dev != netdev)
1307 		goto exit;
1308 
1309 	switch (event) {
1310 	case NETDEV_REGISTER:
1311 		if (rdev)
1312 			break;
1313 		rc = bnxt_re_dev_reg(&rdev, real_dev);
1314 		if (rc == -ENODEV)
1315 			break;
1316 		if (rc) {
1317 			pr_err("Failed to register with the device %s: %#x\n",
1318 			       real_dev->name, rc);
1319 			break;
1320 		}
1321 		bnxt_re_init_one(rdev);
1322 		sch_work = true;
1323 		break;
1324 
1325 	case NETDEV_UNREGISTER:
1326 		/* netdev notifier will call NETDEV_UNREGISTER again later since
1327 		 * we are still holding the reference to the netdev
1328 		 */
1329 		if (test_bit(BNXT_RE_FLAG_TASK_IN_PROG, &rdev->flags))
1330 			goto exit;
1331 		bnxt_re_ib_unreg(rdev, false);
1332 		bnxt_re_remove_one(rdev);
1333 		bnxt_re_dev_unreg(rdev);
1334 		break;
1335 
1336 	default:
1337 		sch_work = true;
1338 		break;
1339 	}
1340 	if (sch_work) {
1341 		/* Allocate for the deferred task */
1342 		re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1343 		if (re_work) {
1344 			re_work->rdev = rdev;
1345 			re_work->event = event;
1346 			re_work->vlan_dev = (real_dev == netdev ?
1347 					     NULL : netdev);
1348 			INIT_WORK(&re_work->work, bnxt_re_task);
1349 			set_bit(BNXT_RE_FLAG_TASK_IN_PROG, &rdev->flags);
1350 			queue_work(bnxt_re_wq, &re_work->work);
1351 		}
1352 	}
1353 
1354 exit:
1355 	return NOTIFY_DONE;
1356 }
1357 
1358 static struct notifier_block bnxt_re_netdev_notifier = {
1359 	.notifier_call = bnxt_re_netdev_event
1360 };
1361 
1362 static int __init bnxt_re_mod_init(void)
1363 {
1364 	int rc = 0;
1365 
1366 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1367 
1368 	bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1369 	if (!bnxt_re_wq)
1370 		return -ENOMEM;
1371 
1372 	INIT_LIST_HEAD(&bnxt_re_dev_list);
1373 
1374 	rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1375 	if (rc) {
1376 		pr_err("%s: Cannot register to netdevice_notifier",
1377 		       ROCE_DRV_MODULE_NAME);
1378 		goto err_netdev;
1379 	}
1380 	return 0;
1381 
1382 err_netdev:
1383 	destroy_workqueue(bnxt_re_wq);
1384 
1385 	return rc;
1386 }
1387 
1388 static void __exit bnxt_re_mod_exit(void)
1389 {
1390 	struct bnxt_re_dev *rdev;
1391 	LIST_HEAD(to_be_deleted);
1392 
1393 	mutex_lock(&bnxt_re_dev_lock);
1394 	/* Free all adapter allocated resources */
1395 	if (!list_empty(&bnxt_re_dev_list))
1396 		list_splice_init(&bnxt_re_dev_list, &to_be_deleted);
1397 	mutex_unlock(&bnxt_re_dev_lock);
1398 
1399 	list_for_each_entry(rdev, &to_be_deleted, list) {
1400 		dev_info(rdev_to_dev(rdev), "Unregistering Device");
1401 		bnxt_re_dev_stop(rdev);
1402 		bnxt_re_ib_unreg(rdev, true);
1403 		bnxt_re_remove_one(rdev);
1404 		bnxt_re_dev_unreg(rdev);
1405 	}
1406 	unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1407 	if (bnxt_re_wq)
1408 		destroy_workqueue(bnxt_re_wq);
1409 }
1410 
1411 module_init(bnxt_re_mod_init);
1412 module_exit(bnxt_re_mod_exit);
1413