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 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev, bool lock_wait);
82 
83 /* for handling bnxt_en callbacks later */
84 static void bnxt_re_stop(void *p)
85 {
86 }
87 
88 static void bnxt_re_start(void *p)
89 {
90 }
91 
92 static void bnxt_re_sriov_config(void *p, int num_vfs)
93 {
94 }
95 
96 static void bnxt_re_shutdown(void *p)
97 {
98 	struct bnxt_re_dev *rdev = p;
99 
100 	if (!rdev)
101 		return;
102 
103 	bnxt_re_ib_unreg(rdev, false);
104 }
105 
106 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
107 	.ulp_async_notifier = NULL,
108 	.ulp_stop = bnxt_re_stop,
109 	.ulp_start = bnxt_re_start,
110 	.ulp_sriov_config = bnxt_re_sriov_config,
111 	.ulp_shutdown = bnxt_re_shutdown
112 };
113 
114 /* RoCE -> Net driver */
115 
116 /* Driver registration routines used to let the networking driver (bnxt_en)
117  * to know that the RoCE driver is now installed
118  */
119 static int bnxt_re_unregister_netdev(struct bnxt_re_dev *rdev, bool lock_wait)
120 {
121 	struct bnxt_en_dev *en_dev;
122 	int rc;
123 
124 	if (!rdev)
125 		return -EINVAL;
126 
127 	en_dev = rdev->en_dev;
128 	/* Acquire rtnl lock if it is not invokded from netdev event */
129 	if (lock_wait)
130 		rtnl_lock();
131 
132 	rc = en_dev->en_ops->bnxt_unregister_device(rdev->en_dev,
133 						    BNXT_ROCE_ULP);
134 	if (lock_wait)
135 		rtnl_unlock();
136 	return rc;
137 }
138 
139 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
140 {
141 	struct bnxt_en_dev *en_dev;
142 	int rc = 0;
143 
144 	if (!rdev)
145 		return -EINVAL;
146 
147 	en_dev = rdev->en_dev;
148 
149 	rtnl_lock();
150 	rc = en_dev->en_ops->bnxt_register_device(en_dev, BNXT_ROCE_ULP,
151 						  &bnxt_re_ulp_ops, rdev);
152 	rtnl_unlock();
153 	return rc;
154 }
155 
156 static int bnxt_re_free_msix(struct bnxt_re_dev *rdev, bool lock_wait)
157 {
158 	struct bnxt_en_dev *en_dev;
159 	int rc;
160 
161 	if (!rdev)
162 		return -EINVAL;
163 
164 	en_dev = rdev->en_dev;
165 
166 	if (lock_wait)
167 		rtnl_lock();
168 
169 	rc = en_dev->en_ops->bnxt_free_msix(rdev->en_dev, BNXT_ROCE_ULP);
170 
171 	if (lock_wait)
172 		rtnl_unlock();
173 	return rc;
174 }
175 
176 static int bnxt_re_request_msix(struct bnxt_re_dev *rdev)
177 {
178 	int rc = 0, num_msix_want = BNXT_RE_MAX_MSIX, num_msix_got;
179 	struct bnxt_en_dev *en_dev;
180 
181 	if (!rdev)
182 		return -EINVAL;
183 
184 	en_dev = rdev->en_dev;
185 
186 	num_msix_want = min_t(u32, BNXT_RE_MAX_MSIX, num_online_cpus());
187 
188 	rtnl_lock();
189 	num_msix_got = en_dev->en_ops->bnxt_request_msix(en_dev, BNXT_ROCE_ULP,
190 							 rdev->msix_entries,
191 							 num_msix_want);
192 	if (num_msix_got < BNXT_RE_MIN_MSIX) {
193 		rc = -EINVAL;
194 		goto done;
195 	}
196 	if (num_msix_got != num_msix_want) {
197 		dev_warn(rdev_to_dev(rdev),
198 			 "Requested %d MSI-X vectors, got %d\n",
199 			 num_msix_want, num_msix_got);
200 	}
201 	rdev->num_msix = num_msix_got;
202 done:
203 	rtnl_unlock();
204 	return rc;
205 }
206 
207 static void bnxt_re_init_hwrm_hdr(struct bnxt_re_dev *rdev, struct input *hdr,
208 				  u16 opcd, u16 crid, u16 trid)
209 {
210 	hdr->req_type = cpu_to_le16(opcd);
211 	hdr->cmpl_ring = cpu_to_le16(crid);
212 	hdr->target_id = cpu_to_le16(trid);
213 }
214 
215 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
216 				int msg_len, void *resp, int resp_max_len,
217 				int timeout)
218 {
219 	fw_msg->msg = msg;
220 	fw_msg->msg_len = msg_len;
221 	fw_msg->resp = resp;
222 	fw_msg->resp_max_len = resp_max_len;
223 	fw_msg->timeout = timeout;
224 }
225 
226 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev, u16 fw_ring_id,
227 				 bool lock_wait)
228 {
229 	struct bnxt_en_dev *en_dev = rdev->en_dev;
230 	struct hwrm_ring_free_input req = {0};
231 	struct hwrm_ring_free_output resp;
232 	struct bnxt_fw_msg fw_msg;
233 	bool do_unlock = false;
234 	int rc = -EINVAL;
235 
236 	if (!en_dev)
237 		return rc;
238 
239 	memset(&fw_msg, 0, sizeof(fw_msg));
240 	if (lock_wait) {
241 		rtnl_lock();
242 		do_unlock = true;
243 	}
244 
245 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1);
246 	req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
247 	req.ring_id = cpu_to_le16(fw_ring_id);
248 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
249 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
250 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
251 	if (rc)
252 		dev_err(rdev_to_dev(rdev),
253 			"Failed to free HW ring:%d :%#x", req.ring_id, rc);
254 	if (do_unlock)
255 		rtnl_unlock();
256 	return rc;
257 }
258 
259 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev, dma_addr_t *dma_arr,
260 				  int pages, int type, u32 ring_mask,
261 				  u32 map_index, u16 *fw_ring_id)
262 {
263 	struct bnxt_en_dev *en_dev = rdev->en_dev;
264 	struct hwrm_ring_alloc_input req = {0};
265 	struct hwrm_ring_alloc_output resp;
266 	struct bnxt_fw_msg fw_msg;
267 	int rc = -EINVAL;
268 
269 	if (!en_dev)
270 		return rc;
271 
272 	memset(&fw_msg, 0, sizeof(fw_msg));
273 	rtnl_lock();
274 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_ALLOC, -1, -1);
275 	req.enables = 0;
276 	req.page_tbl_addr =  cpu_to_le64(dma_arr[0]);
277 	if (pages > 1) {
278 		/* Page size is in log2 units */
279 		req.page_size = BNXT_PAGE_SHIFT;
280 		req.page_tbl_depth = 1;
281 	}
282 	req.fbo = 0;
283 	/* Association of ring index with doorbell index and MSIX number */
284 	req.logical_id = cpu_to_le16(map_index);
285 	req.length = cpu_to_le32(ring_mask + 1);
286 	req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
287 	req.int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
288 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
289 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
290 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
291 	if (!rc)
292 		*fw_ring_id = le16_to_cpu(resp.ring_id);
293 
294 	rtnl_unlock();
295 	return rc;
296 }
297 
298 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
299 				      u32 fw_stats_ctx_id, bool lock_wait)
300 {
301 	struct bnxt_en_dev *en_dev = rdev->en_dev;
302 	struct hwrm_stat_ctx_free_input req = {0};
303 	struct bnxt_fw_msg fw_msg;
304 	bool do_unlock = false;
305 	int rc = -EINVAL;
306 
307 	if (!en_dev)
308 		return rc;
309 
310 	memset(&fw_msg, 0, sizeof(fw_msg));
311 	if (lock_wait) {
312 		rtnl_lock();
313 		do_unlock = true;
314 	}
315 
316 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_FREE, -1, -1);
317 	req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
318 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&req,
319 			    sizeof(req), DFLT_HWRM_CMD_TIMEOUT);
320 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
321 	if (rc)
322 		dev_err(rdev_to_dev(rdev),
323 			"Failed to free HW stats context %#x", rc);
324 
325 	if (do_unlock)
326 		rtnl_unlock();
327 	return rc;
328 }
329 
330 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
331 				       dma_addr_t dma_map,
332 				       u32 *fw_stats_ctx_id)
333 {
334 	struct hwrm_stat_ctx_alloc_output resp = {0};
335 	struct hwrm_stat_ctx_alloc_input req = {0};
336 	struct bnxt_en_dev *en_dev = rdev->en_dev;
337 	struct bnxt_fw_msg fw_msg;
338 	int rc = -EINVAL;
339 
340 	*fw_stats_ctx_id = INVALID_STATS_CTX_ID;
341 
342 	if (!en_dev)
343 		return rc;
344 
345 	memset(&fw_msg, 0, sizeof(fw_msg));
346 	rtnl_lock();
347 
348 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1);
349 	req.update_period_ms = cpu_to_le32(1000);
350 	req.stats_dma_addr = cpu_to_le64(dma_map);
351 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
352 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
353 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
354 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
355 	if (!rc)
356 		*fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
357 
358 	rtnl_unlock();
359 	return rc;
360 }
361 
362 /* Device */
363 
364 static bool is_bnxt_re_dev(struct net_device *netdev)
365 {
366 	struct ethtool_drvinfo drvinfo;
367 
368 	if (netdev->ethtool_ops && netdev->ethtool_ops->get_drvinfo) {
369 		memset(&drvinfo, 0, sizeof(drvinfo));
370 		netdev->ethtool_ops->get_drvinfo(netdev, &drvinfo);
371 
372 		if (strcmp(drvinfo.driver, "bnxt_en"))
373 			return false;
374 		return true;
375 	}
376 	return false;
377 }
378 
379 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
380 {
381 	struct bnxt_re_dev *rdev;
382 
383 	rcu_read_lock();
384 	list_for_each_entry_rcu(rdev, &bnxt_re_dev_list, list) {
385 		if (rdev->netdev == netdev) {
386 			rcu_read_unlock();
387 			return rdev;
388 		}
389 	}
390 	rcu_read_unlock();
391 	return NULL;
392 }
393 
394 static void bnxt_re_dev_unprobe(struct net_device *netdev,
395 				struct bnxt_en_dev *en_dev)
396 {
397 	dev_put(netdev);
398 	module_put(en_dev->pdev->driver->driver.owner);
399 }
400 
401 static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev)
402 {
403 	struct bnxt *bp = netdev_priv(netdev);
404 	struct bnxt_en_dev *en_dev;
405 	struct pci_dev *pdev;
406 
407 	/* Call bnxt_en's RoCE probe via indirect API */
408 	if (!bp->ulp_probe)
409 		return ERR_PTR(-EINVAL);
410 
411 	en_dev = bp->ulp_probe(netdev);
412 	if (IS_ERR(en_dev))
413 		return en_dev;
414 
415 	pdev = en_dev->pdev;
416 	if (!pdev)
417 		return ERR_PTR(-EINVAL);
418 
419 	if (!(en_dev->flags & BNXT_EN_FLAG_ROCE_CAP)) {
420 		dev_dbg(&pdev->dev,
421 			"%s: probe error: RoCE is not supported on this device",
422 			ROCE_DRV_MODULE_NAME);
423 		return ERR_PTR(-ENODEV);
424 	}
425 
426 	/* Bump net device reference count */
427 	if (!try_module_get(pdev->driver->driver.owner))
428 		return ERR_PTR(-ENODEV);
429 
430 	dev_hold(netdev);
431 
432 	return en_dev;
433 }
434 
435 static void bnxt_re_unregister_ib(struct bnxt_re_dev *rdev)
436 {
437 	ib_unregister_device(&rdev->ibdev);
438 }
439 
440 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
441 {
442 	struct ib_device *ibdev = &rdev->ibdev;
443 
444 	/* ib device init */
445 	ibdev->owner = THIS_MODULE;
446 	ibdev->node_type = RDMA_NODE_IB_CA;
447 	strlcpy(ibdev->name, "bnxt_re%d", IB_DEVICE_NAME_MAX);
448 	strlcpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
449 		strlen(BNXT_RE_DESC) + 5);
450 	ibdev->phys_port_cnt = 1;
451 
452 	bnxt_qplib_get_guid(rdev->netdev->dev_addr, (u8 *)&ibdev->node_guid);
453 
454 	ibdev->num_comp_vectors	= 1;
455 	ibdev->dev.parent = &rdev->en_dev->pdev->dev;
456 	ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
457 
458 	/* User space */
459 	ibdev->uverbs_abi_ver = BNXT_RE_ABI_VERSION;
460 	ibdev->uverbs_cmd_mask =
461 			(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
462 			(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
463 			(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
464 			(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
465 			(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
466 			(1ull << IB_USER_VERBS_CMD_REG_MR)		|
467 			(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
468 			(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
469 			(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
470 			(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
471 			(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
472 			(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
473 			(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
474 			(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
475 			(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
476 			(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
477 			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
478 			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
479 			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
480 			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
481 			(1ull << IB_USER_VERBS_CMD_CREATE_AH)		|
482 			(1ull << IB_USER_VERBS_CMD_MODIFY_AH)		|
483 			(1ull << IB_USER_VERBS_CMD_QUERY_AH)		|
484 			(1ull << IB_USER_VERBS_CMD_DESTROY_AH);
485 	/* POLL_CQ and REQ_NOTIFY_CQ is directly handled in libbnxt_re */
486 
487 	/* Kernel verbs */
488 	ibdev->query_device		= bnxt_re_query_device;
489 	ibdev->modify_device		= bnxt_re_modify_device;
490 
491 	ibdev->query_port		= bnxt_re_query_port;
492 	ibdev->get_port_immutable	= bnxt_re_get_port_immutable;
493 	ibdev->query_pkey		= bnxt_re_query_pkey;
494 	ibdev->query_gid		= bnxt_re_query_gid;
495 	ibdev->get_netdev		= bnxt_re_get_netdev;
496 	ibdev->add_gid			= bnxt_re_add_gid;
497 	ibdev->del_gid			= bnxt_re_del_gid;
498 	ibdev->get_link_layer		= bnxt_re_get_link_layer;
499 
500 	ibdev->alloc_pd			= bnxt_re_alloc_pd;
501 	ibdev->dealloc_pd		= bnxt_re_dealloc_pd;
502 
503 	ibdev->create_ah		= bnxt_re_create_ah;
504 	ibdev->modify_ah		= bnxt_re_modify_ah;
505 	ibdev->query_ah			= bnxt_re_query_ah;
506 	ibdev->destroy_ah		= bnxt_re_destroy_ah;
507 
508 	ibdev->create_qp		= bnxt_re_create_qp;
509 	ibdev->modify_qp		= bnxt_re_modify_qp;
510 	ibdev->query_qp			= bnxt_re_query_qp;
511 	ibdev->destroy_qp		= bnxt_re_destroy_qp;
512 
513 	ibdev->post_send		= bnxt_re_post_send;
514 	ibdev->post_recv		= bnxt_re_post_recv;
515 
516 	ibdev->create_cq		= bnxt_re_create_cq;
517 	ibdev->destroy_cq		= bnxt_re_destroy_cq;
518 	ibdev->poll_cq			= bnxt_re_poll_cq;
519 	ibdev->req_notify_cq		= bnxt_re_req_notify_cq;
520 
521 	ibdev->get_dma_mr		= bnxt_re_get_dma_mr;
522 	ibdev->dereg_mr			= bnxt_re_dereg_mr;
523 	ibdev->alloc_mr			= bnxt_re_alloc_mr;
524 	ibdev->map_mr_sg		= bnxt_re_map_mr_sg;
525 
526 	ibdev->reg_user_mr		= bnxt_re_reg_user_mr;
527 	ibdev->alloc_ucontext		= bnxt_re_alloc_ucontext;
528 	ibdev->dealloc_ucontext		= bnxt_re_dealloc_ucontext;
529 	ibdev->mmap			= bnxt_re_mmap;
530 	ibdev->get_hw_stats             = bnxt_re_ib_get_hw_stats;
531 	ibdev->alloc_hw_stats           = bnxt_re_ib_alloc_hw_stats;
532 
533 	return ib_register_device(ibdev, NULL);
534 }
535 
536 static ssize_t show_rev(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, "0x%x\n", rdev->en_dev->pdev->vendor);
542 }
543 
544 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
545 			   char *buf)
546 {
547 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
548 
549 	return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->dev_attr.fw_ver);
550 }
551 
552 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
553 			char *buf)
554 {
555 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
556 
557 	return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->ibdev.node_desc);
558 }
559 
560 static DEVICE_ATTR(hw_rev, 0444, show_rev, NULL);
561 static DEVICE_ATTR(fw_rev, 0444, show_fw_ver, NULL);
562 static DEVICE_ATTR(hca_type, 0444, show_hca, NULL);
563 
564 static struct device_attribute *bnxt_re_attributes[] = {
565 	&dev_attr_hw_rev,
566 	&dev_attr_fw_rev,
567 	&dev_attr_hca_type
568 };
569 
570 static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev)
571 {
572 	dev_put(rdev->netdev);
573 	rdev->netdev = NULL;
574 
575 	mutex_lock(&bnxt_re_dev_lock);
576 	list_del_rcu(&rdev->list);
577 	mutex_unlock(&bnxt_re_dev_lock);
578 
579 	synchronize_rcu();
580 	flush_workqueue(bnxt_re_wq);
581 
582 	ib_dealloc_device(&rdev->ibdev);
583 	/* rdev is gone */
584 }
585 
586 static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev,
587 					   struct bnxt_en_dev *en_dev)
588 {
589 	struct bnxt_re_dev *rdev;
590 
591 	/* Allocate bnxt_re_dev instance here */
592 	rdev = (struct bnxt_re_dev *)ib_alloc_device(sizeof(*rdev));
593 	if (!rdev) {
594 		dev_err(NULL, "%s: bnxt_re_dev allocation failure!",
595 			ROCE_DRV_MODULE_NAME);
596 		return NULL;
597 	}
598 	/* Default values */
599 	rdev->netdev = netdev;
600 	dev_hold(rdev->netdev);
601 	rdev->en_dev = en_dev;
602 	rdev->id = rdev->en_dev->pdev->devfn;
603 	INIT_LIST_HEAD(&rdev->qp_list);
604 	mutex_init(&rdev->qp_lock);
605 	atomic_set(&rdev->qp_count, 0);
606 	atomic_set(&rdev->cq_count, 0);
607 	atomic_set(&rdev->srq_count, 0);
608 	atomic_set(&rdev->mr_count, 0);
609 	atomic_set(&rdev->mw_count, 0);
610 	rdev->cosq[0] = 0xFFFF;
611 	rdev->cosq[1] = 0xFFFF;
612 
613 	mutex_lock(&bnxt_re_dev_lock);
614 	list_add_tail_rcu(&rdev->list, &bnxt_re_dev_list);
615 	mutex_unlock(&bnxt_re_dev_lock);
616 	return rdev;
617 }
618 
619 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
620 			       struct creq_func_event *aeqe)
621 {
622 	switch (aeqe->event) {
623 	case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
624 		break;
625 	case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
626 		break;
627 	case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
628 		break;
629 	case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
630 		break;
631 	case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
632 		break;
633 	case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
634 		break;
635 	case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
636 		break;
637 	case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
638 		break;
639 	case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
640 		break;
641 	case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
642 		break;
643 	case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
644 		break;
645 	default:
646 		return -EINVAL;
647 	}
648 	return 0;
649 }
650 
651 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
652 			       struct bnxt_qplib_cq *handle)
653 {
654 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
655 					     qplib_cq);
656 
657 	if (!cq) {
658 		dev_err(NULL, "%s: CQ is NULL, CQN not handled",
659 			ROCE_DRV_MODULE_NAME);
660 		return -EINVAL;
661 	}
662 	if (cq->ib_cq.comp_handler) {
663 		/* Lock comp_handler? */
664 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
665 	}
666 
667 	return 0;
668 }
669 
670 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
671 {
672 	int i;
673 
674 	if (rdev->nq[0].hwq.max_elements) {
675 		for (i = 1; i < rdev->num_msix; i++)
676 			bnxt_qplib_disable_nq(&rdev->nq[i - 1]);
677 	}
678 
679 	if (rdev->qplib_res.rcfw)
680 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
681 }
682 
683 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
684 {
685 	int rc = 0, i;
686 
687 	bnxt_qplib_init_res(&rdev->qplib_res);
688 
689 	for (i = 1; i < rdev->num_msix ; i++) {
690 		rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
691 					  i - 1, rdev->msix_entries[i].vector,
692 					  rdev->msix_entries[i].db_offset,
693 					  &bnxt_re_cqn_handler, NULL);
694 
695 		if (rc) {
696 			dev_err(rdev_to_dev(rdev),
697 				"Failed to enable NQ with rc = 0x%x", rc);
698 			goto fail;
699 		}
700 	}
701 	return 0;
702 fail:
703 	return rc;
704 }
705 
706 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev, bool lock_wait)
707 {
708 	int i;
709 
710 	for (i = 0; i < rdev->num_msix - 1; i++) {
711 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, lock_wait);
712 		bnxt_qplib_free_nq(&rdev->nq[i]);
713 	}
714 }
715 
716 static void bnxt_re_free_res(struct bnxt_re_dev *rdev, bool lock_wait)
717 {
718 	bnxt_re_free_nq_res(rdev, lock_wait);
719 
720 	if (rdev->qplib_res.dpi_tbl.max) {
721 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
722 				       &rdev->qplib_res.dpi_tbl,
723 				       &rdev->dpi_privileged);
724 	}
725 	if (rdev->qplib_res.rcfw) {
726 		bnxt_qplib_free_res(&rdev->qplib_res);
727 		rdev->qplib_res.rcfw = NULL;
728 	}
729 }
730 
731 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
732 {
733 	int rc = 0, i;
734 
735 	/* Configure and allocate resources for qplib */
736 	rdev->qplib_res.rcfw = &rdev->rcfw;
737 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr);
738 	if (rc)
739 		goto fail;
740 
741 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
742 				  rdev->netdev, &rdev->dev_attr);
743 	if (rc)
744 		goto fail;
745 
746 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
747 				  &rdev->dpi_privileged,
748 				  rdev);
749 	if (rc)
750 		goto dealloc_res;
751 
752 	for (i = 0; i < rdev->num_msix - 1; i++) {
753 		rdev->nq[i].hwq.max_elements = BNXT_RE_MAX_CQ_COUNT +
754 			BNXT_RE_MAX_SRQC_COUNT + 2;
755 		rc = bnxt_qplib_alloc_nq(rdev->en_dev->pdev, &rdev->nq[i]);
756 		if (rc) {
757 			dev_err(rdev_to_dev(rdev), "Alloc Failed NQ%d rc:%#x",
758 				i, rc);
759 			goto dealloc_dpi;
760 		}
761 		rc = bnxt_re_net_ring_alloc
762 			(rdev, rdev->nq[i].hwq.pbl[PBL_LVL_0].pg_map_arr,
763 			 rdev->nq[i].hwq.pbl[rdev->nq[i].hwq.level].pg_count,
764 			 HWRM_RING_ALLOC_CMPL,
765 			 BNXT_QPLIB_NQE_MAX_CNT - 1,
766 			 rdev->msix_entries[i + 1].ring_idx,
767 			 &rdev->nq[i].ring_id);
768 		if (rc) {
769 			dev_err(rdev_to_dev(rdev),
770 				"Failed to allocate NQ fw id with rc = 0x%x",
771 				rc);
772 			goto free_nq;
773 		}
774 	}
775 	return 0;
776 free_nq:
777 	for (i = 0; i < rdev->num_msix - 1; i++)
778 		bnxt_qplib_free_nq(&rdev->nq[i]);
779 dealloc_dpi:
780 	bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
781 			       &rdev->qplib_res.dpi_tbl,
782 			       &rdev->dpi_privileged);
783 dealloc_res:
784 	bnxt_qplib_free_res(&rdev->qplib_res);
785 
786 fail:
787 	rdev->qplib_res.rcfw = NULL;
788 	return rc;
789 }
790 
791 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
792 				   u8 port_num, enum ib_event_type event)
793 {
794 	struct ib_event ib_event;
795 
796 	ib_event.device = ibdev;
797 	if (qp)
798 		ib_event.element.qp = qp;
799 	else
800 		ib_event.element.port_num = port_num;
801 	ib_event.event = event;
802 	ib_dispatch_event(&ib_event);
803 }
804 
805 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN      0x02
806 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
807 				      u64 *cid_map)
808 {
809 	struct hwrm_queue_pri2cos_qcfg_input req = {0};
810 	struct bnxt *bp = netdev_priv(rdev->netdev);
811 	struct hwrm_queue_pri2cos_qcfg_output resp;
812 	struct bnxt_en_dev *en_dev = rdev->en_dev;
813 	struct bnxt_fw_msg fw_msg;
814 	u32 flags = 0;
815 	u8 *qcfgmap, *tmp_map;
816 	int rc = 0, i;
817 
818 	if (!cid_map)
819 		return -EINVAL;
820 
821 	memset(&fw_msg, 0, sizeof(fw_msg));
822 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
823 			      HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
824 	flags |= (dir & 0x01);
825 	flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
826 	req.flags = cpu_to_le32(flags);
827 	req.port_id = bp->pf.port_id;
828 
829 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
830 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
831 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
832 	if (rc)
833 		return rc;
834 
835 	if (resp.queue_cfg_info) {
836 		dev_warn(rdev_to_dev(rdev),
837 			 "Asymmetric cos queue configuration detected");
838 		dev_warn(rdev_to_dev(rdev),
839 			 " on device, QoS may not be fully functional\n");
840 	}
841 	qcfgmap = &resp.pri0_cos_queue_id;
842 	tmp_map = (u8 *)cid_map;
843 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
844 		tmp_map[i] = qcfgmap[i];
845 
846 	return rc;
847 }
848 
849 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
850 					struct bnxt_re_qp *qp)
851 {
852 	return (qp->ib_qp.qp_type == IB_QPT_GSI) || (qp == rdev->qp1_sqp);
853 }
854 
855 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
856 {
857 	int mask = IB_QP_STATE;
858 	struct ib_qp_attr qp_attr;
859 	struct bnxt_re_qp *qp;
860 
861 	qp_attr.qp_state = IB_QPS_ERR;
862 	mutex_lock(&rdev->qp_lock);
863 	list_for_each_entry(qp, &rdev->qp_list, list) {
864 		/* Modify the state of all QPs except QP1/Shadow QP */
865 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
866 			if (qp->qplib_qp.state !=
867 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
868 			    qp->qplib_qp.state !=
869 			    CMDQ_MODIFY_QP_NEW_STATE_ERR) {
870 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
871 						       1, IB_EVENT_QP_FATAL);
872 				bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
873 						  NULL);
874 			}
875 		}
876 	}
877 	mutex_unlock(&rdev->qp_lock);
878 }
879 
880 static int bnxt_re_update_gid(struct bnxt_re_dev *rdev)
881 {
882 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
883 	struct bnxt_qplib_gid gid;
884 	u16 gid_idx, index;
885 	int rc = 0;
886 
887 	if (!test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
888 		return 0;
889 
890 	if (!sgid_tbl) {
891 		dev_err(rdev_to_dev(rdev), "QPLIB: SGID table not allocated");
892 		return -EINVAL;
893 	}
894 
895 	for (index = 0; index < sgid_tbl->active; index++) {
896 		gid_idx = sgid_tbl->hw_id[index];
897 
898 		if (!memcmp(&sgid_tbl->tbl[index], &bnxt_qplib_gid_zero,
899 			    sizeof(bnxt_qplib_gid_zero)))
900 			continue;
901 		/* need to modify the VLAN enable setting of non VLAN GID only
902 		 * as setting is done for VLAN GID while adding GID
903 		 */
904 		if (sgid_tbl->vlan[index])
905 			continue;
906 
907 		memcpy(&gid, &sgid_tbl->tbl[index], sizeof(gid));
908 
909 		rc = bnxt_qplib_update_sgid(sgid_tbl, &gid, gid_idx,
910 					    rdev->qplib_res.netdev->dev_addr);
911 	}
912 
913 	return rc;
914 }
915 
916 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
917 {
918 	u32 prio_map = 0, tmp_map = 0;
919 	struct net_device *netdev;
920 	struct dcb_app app;
921 
922 	netdev = rdev->netdev;
923 
924 	memset(&app, 0, sizeof(app));
925 	app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
926 	app.protocol = ETH_P_IBOE;
927 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
928 	prio_map = tmp_map;
929 
930 	app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
931 	app.protocol = ROCE_V2_UDP_DPORT;
932 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
933 	prio_map |= tmp_map;
934 
935 	return prio_map;
936 }
937 
938 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
939 {
940 	u16 prio;
941 	u8 id;
942 
943 	for (prio = 0, id = 0; prio < 8; prio++) {
944 		if (prio_map & (1 << prio)) {
945 			cosq[id] = cid_map[prio];
946 			id++;
947 			if (id == 2) /* Max 2 tcs supported */
948 				break;
949 		}
950 	}
951 }
952 
953 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
954 {
955 	u8 prio_map = 0;
956 	u64 cid_map;
957 	int rc;
958 
959 	/* Get priority for roce */
960 	prio_map = bnxt_re_get_priority_mask(rdev);
961 
962 	if (prio_map == rdev->cur_prio_map)
963 		return 0;
964 	rdev->cur_prio_map = prio_map;
965 	/* Get cosq id for this priority */
966 	rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
967 	if (rc) {
968 		dev_warn(rdev_to_dev(rdev), "no cos for p_mask %x\n", prio_map);
969 		return rc;
970 	}
971 	/* Parse CoS IDs for app priority */
972 	bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
973 
974 	/* Config BONO. */
975 	rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
976 	if (rc) {
977 		dev_warn(rdev_to_dev(rdev), "no tc for cos{%x, %x}\n",
978 			 rdev->cosq[0], rdev->cosq[1]);
979 		return rc;
980 	}
981 
982 	/* Actual priorities are not programmed as they are already
983 	 * done by L2 driver; just enable or disable priority vlan tagging
984 	 */
985 	if ((prio_map == 0 && rdev->qplib_res.prio) ||
986 	    (prio_map != 0 && !rdev->qplib_res.prio)) {
987 		rdev->qplib_res.prio = prio_map ? true : false;
988 
989 		bnxt_re_update_gid(rdev);
990 	}
991 
992 	return 0;
993 }
994 
995 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev, bool lock_wait)
996 {
997 	int i, rc;
998 
999 	if (test_and_clear_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags)) {
1000 		for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++)
1001 			device_remove_file(&rdev->ibdev.dev,
1002 					   bnxt_re_attributes[i]);
1003 		/* Cleanup ib dev */
1004 		bnxt_re_unregister_ib(rdev);
1005 	}
1006 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
1007 		cancel_delayed_work(&rdev->worker);
1008 
1009 	bnxt_re_cleanup_res(rdev);
1010 	bnxt_re_free_res(rdev, lock_wait);
1011 
1012 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
1013 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
1014 		if (rc)
1015 			dev_warn(rdev_to_dev(rdev),
1016 				 "Failed to deinitialize RCFW: %#x", rc);
1017 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id,
1018 					   lock_wait);
1019 		bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1020 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1021 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, lock_wait);
1022 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1023 	}
1024 	if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
1025 		rc = bnxt_re_free_msix(rdev, lock_wait);
1026 		if (rc)
1027 			dev_warn(rdev_to_dev(rdev),
1028 				 "Failed to free MSI-X vectors: %#x", rc);
1029 	}
1030 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
1031 		rc = bnxt_re_unregister_netdev(rdev, lock_wait);
1032 		if (rc)
1033 			dev_warn(rdev_to_dev(rdev),
1034 				 "Failed to unregister with netdev: %#x", rc);
1035 	}
1036 }
1037 
1038 static void bnxt_re_set_resource_limits(struct bnxt_re_dev *rdev)
1039 {
1040 	u32 i;
1041 
1042 	rdev->qplib_ctx.qpc_count = BNXT_RE_MAX_QPC_COUNT;
1043 	rdev->qplib_ctx.mrw_count = BNXT_RE_MAX_MRW_COUNT;
1044 	rdev->qplib_ctx.srqc_count = BNXT_RE_MAX_SRQC_COUNT;
1045 	rdev->qplib_ctx.cq_count = BNXT_RE_MAX_CQ_COUNT;
1046 	for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
1047 		rdev->qplib_ctx.tqm_count[i] =
1048 		rdev->dev_attr.tqm_alloc_reqs[i];
1049 }
1050 
1051 /* worker thread for polling periodic events. Now used for QoS programming*/
1052 static void bnxt_re_worker(struct work_struct *work)
1053 {
1054 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
1055 						worker.work);
1056 
1057 	bnxt_re_setup_qos(rdev);
1058 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1059 }
1060 
1061 static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
1062 {
1063 	int i, j, rc;
1064 
1065 	/* Registered a new RoCE device instance to netdev */
1066 	rc = bnxt_re_register_netdev(rdev);
1067 	if (rc) {
1068 		pr_err("Failed to register with netedev: %#x\n", rc);
1069 		return -EINVAL;
1070 	}
1071 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1072 
1073 	rc = bnxt_re_request_msix(rdev);
1074 	if (rc) {
1075 		pr_err("Failed to get MSI-X vectors: %#x\n", rc);
1076 		rc = -EINVAL;
1077 		goto fail;
1078 	}
1079 	set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
1080 
1081 	/* Establish RCFW Communication Channel to initialize the context
1082 	 * memory for the function and all child VFs
1083 	 */
1084 	rc = bnxt_qplib_alloc_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw,
1085 					   BNXT_RE_MAX_QPC_COUNT);
1086 	if (rc) {
1087 		pr_err("Failed to allocate RCFW Channel: %#x\n", rc);
1088 		goto fail;
1089 	}
1090 	rc = bnxt_re_net_ring_alloc
1091 			(rdev, rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr,
1092 			 rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count,
1093 			 HWRM_RING_ALLOC_CMPL, BNXT_QPLIB_CREQE_MAX_CNT - 1,
1094 			 rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx,
1095 			 &rdev->rcfw.creq_ring_id);
1096 	if (rc) {
1097 		pr_err("Failed to allocate CREQ: %#x\n", rc);
1098 		goto free_rcfw;
1099 	}
1100 	rc = bnxt_qplib_enable_rcfw_channel
1101 				(rdev->en_dev->pdev, &rdev->rcfw,
1102 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].vector,
1103 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].db_offset,
1104 				 0, &bnxt_re_aeq_handler);
1105 	if (rc) {
1106 		pr_err("Failed to enable RCFW channel: %#x\n", rc);
1107 		goto free_ring;
1108 	}
1109 
1110 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr);
1111 	if (rc)
1112 		goto disable_rcfw;
1113 	bnxt_re_set_resource_limits(rdev);
1114 
1115 	rc = bnxt_qplib_alloc_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx, 0);
1116 	if (rc) {
1117 		pr_err("Failed to allocate QPLIB context: %#x\n", rc);
1118 		goto disable_rcfw;
1119 	}
1120 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
1121 					 rdev->qplib_ctx.stats.dma_map,
1122 					 &rdev->qplib_ctx.stats.fw_id);
1123 	if (rc) {
1124 		pr_err("Failed to allocate stats context: %#x\n", rc);
1125 		goto free_ctx;
1126 	}
1127 
1128 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx, 0);
1129 	if (rc) {
1130 		pr_err("Failed to initialize RCFW: %#x\n", rc);
1131 		goto free_sctx;
1132 	}
1133 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1134 
1135 	/* Resources based on the 'new' device caps */
1136 	rc = bnxt_re_alloc_res(rdev);
1137 	if (rc) {
1138 		pr_err("Failed to allocate resources: %#x\n", rc);
1139 		goto fail;
1140 	}
1141 	rc = bnxt_re_init_res(rdev);
1142 	if (rc) {
1143 		pr_err("Failed to initialize resources: %#x\n", rc);
1144 		goto fail;
1145 	}
1146 
1147 	rc = bnxt_re_setup_qos(rdev);
1148 	if (rc)
1149 		pr_info("RoCE priority not yet configured\n");
1150 
1151 	INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1152 	set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1153 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1154 
1155 	/* Register ib dev */
1156 	rc = bnxt_re_register_ib(rdev);
1157 	if (rc) {
1158 		pr_err("Failed to register with IB: %#x\n", rc);
1159 		goto fail;
1160 	}
1161 	dev_info(rdev_to_dev(rdev), "Device registered successfully");
1162 	for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++) {
1163 		rc = device_create_file(&rdev->ibdev.dev,
1164 					bnxt_re_attributes[i]);
1165 		if (rc) {
1166 			dev_err(rdev_to_dev(rdev),
1167 				"Failed to create IB sysfs: %#x", rc);
1168 			/* Must clean up all created device files */
1169 			for (j = 0; j < i; j++)
1170 				device_remove_file(&rdev->ibdev.dev,
1171 						   bnxt_re_attributes[j]);
1172 			bnxt_re_unregister_ib(rdev);
1173 			goto fail;
1174 		}
1175 	}
1176 	set_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags);
1177 	ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1178 			 &rdev->active_width);
1179 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_PORT_ACTIVE);
1180 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_GID_CHANGE);
1181 
1182 	return 0;
1183 free_sctx:
1184 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id, true);
1185 free_ctx:
1186 	bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1187 disable_rcfw:
1188 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1189 free_ring:
1190 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, true);
1191 free_rcfw:
1192 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1193 fail:
1194 	bnxt_re_ib_unreg(rdev, true);
1195 	return rc;
1196 }
1197 
1198 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1199 {
1200 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1201 	struct net_device *netdev = rdev->netdev;
1202 
1203 	bnxt_re_dev_remove(rdev);
1204 
1205 	if (netdev)
1206 		bnxt_re_dev_unprobe(netdev, en_dev);
1207 }
1208 
1209 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1210 {
1211 	struct bnxt_en_dev *en_dev;
1212 	int rc = 0;
1213 
1214 	if (!is_bnxt_re_dev(netdev))
1215 		return -ENODEV;
1216 
1217 	en_dev = bnxt_re_dev_probe(netdev);
1218 	if (IS_ERR(en_dev)) {
1219 		if (en_dev != ERR_PTR(-ENODEV))
1220 			pr_err("%s: Failed to probe\n", ROCE_DRV_MODULE_NAME);
1221 		rc = PTR_ERR(en_dev);
1222 		goto exit;
1223 	}
1224 	*rdev = bnxt_re_dev_add(netdev, en_dev);
1225 	if (!*rdev) {
1226 		rc = -ENOMEM;
1227 		bnxt_re_dev_unprobe(netdev, en_dev);
1228 		goto exit;
1229 	}
1230 exit:
1231 	return rc;
1232 }
1233 
1234 static void bnxt_re_remove_one(struct bnxt_re_dev *rdev)
1235 {
1236 	pci_dev_put(rdev->en_dev->pdev);
1237 }
1238 
1239 /* Handle all deferred netevents tasks */
1240 static void bnxt_re_task(struct work_struct *work)
1241 {
1242 	struct bnxt_re_work *re_work;
1243 	struct bnxt_re_dev *rdev;
1244 	int rc = 0;
1245 
1246 	re_work = container_of(work, struct bnxt_re_work, work);
1247 	rdev = re_work->rdev;
1248 
1249 	if (re_work->event != NETDEV_REGISTER &&
1250 	    !test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1251 		return;
1252 
1253 	switch (re_work->event) {
1254 	case NETDEV_REGISTER:
1255 		rc = bnxt_re_ib_reg(rdev);
1256 		if (rc)
1257 			dev_err(rdev_to_dev(rdev),
1258 				"Failed to register with IB: %#x", rc);
1259 		break;
1260 	case NETDEV_UP:
1261 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1262 				       IB_EVENT_PORT_ACTIVE);
1263 		break;
1264 	case NETDEV_DOWN:
1265 		bnxt_re_dev_stop(rdev);
1266 		break;
1267 	case NETDEV_CHANGE:
1268 		if (!netif_carrier_ok(rdev->netdev))
1269 			bnxt_re_dev_stop(rdev);
1270 		else if (netif_carrier_ok(rdev->netdev))
1271 			bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1272 					       IB_EVENT_PORT_ACTIVE);
1273 		ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1274 				 &rdev->active_width);
1275 		break;
1276 	default:
1277 		break;
1278 	}
1279 	smp_mb__before_atomic();
1280 	clear_bit(BNXT_RE_FLAG_TASK_IN_PROG, &rdev->flags);
1281 	kfree(re_work);
1282 }
1283 
1284 static void bnxt_re_init_one(struct bnxt_re_dev *rdev)
1285 {
1286 	pci_dev_get(rdev->en_dev->pdev);
1287 }
1288 
1289 /*
1290  * "Notifier chain callback can be invoked for the same chain from
1291  * different CPUs at the same time".
1292  *
1293  * For cases when the netdev is already present, our call to the
1294  * register_netdevice_notifier() will actually get the rtnl_lock()
1295  * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1296  * events.
1297  *
1298  * But for cases when the netdev is not already present, the notifier
1299  * chain is subjected to be invoked from different CPUs simultaneously.
1300  *
1301  * This is protected by the netdev_mutex.
1302  */
1303 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1304 				unsigned long event, void *ptr)
1305 {
1306 	struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1307 	struct bnxt_re_work *re_work;
1308 	struct bnxt_re_dev *rdev;
1309 	int rc = 0;
1310 	bool sch_work = false;
1311 
1312 	real_dev = rdma_vlan_dev_real_dev(netdev);
1313 	if (!real_dev)
1314 		real_dev = netdev;
1315 
1316 	rdev = bnxt_re_from_netdev(real_dev);
1317 	if (!rdev && event != NETDEV_REGISTER)
1318 		goto exit;
1319 	if (real_dev != netdev)
1320 		goto exit;
1321 
1322 	switch (event) {
1323 	case NETDEV_REGISTER:
1324 		if (rdev)
1325 			break;
1326 		rc = bnxt_re_dev_reg(&rdev, real_dev);
1327 		if (rc == -ENODEV)
1328 			break;
1329 		if (rc) {
1330 			pr_err("Failed to register with the device %s: %#x\n",
1331 			       real_dev->name, rc);
1332 			break;
1333 		}
1334 		bnxt_re_init_one(rdev);
1335 		sch_work = true;
1336 		break;
1337 
1338 	case NETDEV_UNREGISTER:
1339 		/* netdev notifier will call NETDEV_UNREGISTER again later since
1340 		 * we are still holding the reference to the netdev
1341 		 */
1342 		if (test_bit(BNXT_RE_FLAG_TASK_IN_PROG, &rdev->flags))
1343 			goto exit;
1344 		bnxt_re_ib_unreg(rdev, false);
1345 		bnxt_re_remove_one(rdev);
1346 		bnxt_re_dev_unreg(rdev);
1347 		break;
1348 
1349 	default:
1350 		sch_work = true;
1351 		break;
1352 	}
1353 	if (sch_work) {
1354 		/* Allocate for the deferred task */
1355 		re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1356 		if (re_work) {
1357 			re_work->rdev = rdev;
1358 			re_work->event = event;
1359 			re_work->vlan_dev = (real_dev == netdev ?
1360 					     NULL : netdev);
1361 			INIT_WORK(&re_work->work, bnxt_re_task);
1362 			set_bit(BNXT_RE_FLAG_TASK_IN_PROG, &rdev->flags);
1363 			queue_work(bnxt_re_wq, &re_work->work);
1364 		}
1365 	}
1366 
1367 exit:
1368 	return NOTIFY_DONE;
1369 }
1370 
1371 static struct notifier_block bnxt_re_netdev_notifier = {
1372 	.notifier_call = bnxt_re_netdev_event
1373 };
1374 
1375 static int __init bnxt_re_mod_init(void)
1376 {
1377 	int rc = 0;
1378 
1379 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1380 
1381 	bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1382 	if (!bnxt_re_wq)
1383 		return -ENOMEM;
1384 
1385 	INIT_LIST_HEAD(&bnxt_re_dev_list);
1386 
1387 	rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1388 	if (rc) {
1389 		pr_err("%s: Cannot register to netdevice_notifier",
1390 		       ROCE_DRV_MODULE_NAME);
1391 		goto err_netdev;
1392 	}
1393 	return 0;
1394 
1395 err_netdev:
1396 	destroy_workqueue(bnxt_re_wq);
1397 
1398 	return rc;
1399 }
1400 
1401 static void __exit bnxt_re_mod_exit(void)
1402 {
1403 	struct bnxt_re_dev *rdev;
1404 	LIST_HEAD(to_be_deleted);
1405 
1406 	mutex_lock(&bnxt_re_dev_lock);
1407 	/* Free all adapter allocated resources */
1408 	if (!list_empty(&bnxt_re_dev_list))
1409 		list_splice_init(&bnxt_re_dev_list, &to_be_deleted);
1410 	mutex_unlock(&bnxt_re_dev_lock);
1411 
1412 	list_for_each_entry(rdev, &to_be_deleted, list) {
1413 		dev_info(rdev_to_dev(rdev), "Unregistering Device");
1414 		bnxt_re_dev_stop(rdev);
1415 		bnxt_re_ib_unreg(rdev, true);
1416 		bnxt_re_remove_one(rdev);
1417 		bnxt_re_dev_unreg(rdev);
1418 	}
1419 	unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1420 	if (bnxt_re_wq)
1421 		destroy_workqueue(bnxt_re_wq);
1422 }
1423 
1424 module_init(bnxt_re_mod_init);
1425 module_exit(bnxt_re_mod_exit);
1426