1 /*
2  * Broadcom NetXtreme-E RoCE driver.
3  *
4  * Copyright (c) 2016 - 2017, Broadcom. All rights reserved.  The term
5  * Broadcom refers to Broadcom Limited and/or its subsidiaries.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * BSD license below:
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  *
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in
21  *    the documentation and/or other materials provided with the
22  *    distribution.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS''
25  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
26  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS
28  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
32  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
33  * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
34  * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Description: Main component of the bnxt_re driver
37  */
38 
39 #include <linux/module.h>
40 #include <linux/netdevice.h>
41 #include <linux/ethtool.h>
42 #include <linux/mutex.h>
43 #include <linux/list.h>
44 #include <linux/rculist.h>
45 #include <linux/spinlock.h>
46 #include <linux/pci.h>
47 #include <net/dcbnl.h>
48 #include <net/ipv6.h>
49 #include <net/addrconf.h>
50 #include <linux/if_ether.h>
51 
52 #include <rdma/ib_verbs.h>
53 #include <rdma/ib_user_verbs.h>
54 #include <rdma/ib_umem.h>
55 #include <rdma/ib_addr.h>
56 
57 #include "bnxt_ulp.h"
58 #include "roce_hsi.h"
59 #include "qplib_res.h"
60 #include "qplib_sp.h"
61 #include "qplib_fp.h"
62 #include "qplib_rcfw.h"
63 #include "bnxt_re.h"
64 #include "ib_verbs.h"
65 #include <rdma/bnxt_re-abi.h>
66 #include "bnxt.h"
67 #include "hw_counters.h"
68 
69 static char version[] =
70 		BNXT_RE_DESC "\n";
71 
72 MODULE_AUTHOR("Eddie Wai <eddie.wai@broadcom.com>");
73 MODULE_DESCRIPTION(BNXT_RE_DESC " Driver");
74 MODULE_LICENSE("Dual BSD/GPL");
75 
76 /* globals */
77 static struct list_head bnxt_re_dev_list = LIST_HEAD_INIT(bnxt_re_dev_list);
78 /* Mutex to protect the list of bnxt_re devices added */
79 static DEFINE_MUTEX(bnxt_re_dev_lock);
80 static struct workqueue_struct *bnxt_re_wq;
81 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev);
82 
83 static void bnxt_re_destroy_chip_ctx(struct bnxt_re_dev *rdev)
84 {
85 	rdev->rcfw.res = NULL;
86 	rdev->qplib_res.cctx = NULL;
87 }
88 
89 static int bnxt_re_setup_chip_ctx(struct bnxt_re_dev *rdev)
90 {
91 	struct bnxt_en_dev *en_dev;
92 	struct bnxt *bp;
93 
94 	en_dev = rdev->en_dev;
95 	bp = netdev_priv(en_dev->net);
96 
97 	rdev->chip_ctx.chip_num = bp->chip_num;
98 	/* rest members to follow eventually */
99 
100 	rdev->qplib_res.cctx = &rdev->chip_ctx;
101 	rdev->rcfw.res = &rdev->qplib_res;
102 
103 	return 0;
104 }
105 
106 /* SR-IOV helper functions */
107 
108 static void bnxt_re_get_sriov_func_type(struct bnxt_re_dev *rdev)
109 {
110 	struct bnxt *bp;
111 
112 	bp = netdev_priv(rdev->en_dev->net);
113 	if (BNXT_VF(bp))
114 		rdev->is_virtfn = 1;
115 }
116 
117 /* Set the maximum number of each resource that the driver actually wants
118  * to allocate. This may be up to the maximum number the firmware has
119  * reserved for the function. The driver may choose to allocate fewer
120  * resources than the firmware maximum.
121  */
122 static void bnxt_re_set_resource_limits(struct bnxt_re_dev *rdev)
123 {
124 	u32 vf_qps = 0, vf_srqs = 0, vf_cqs = 0, vf_mrws = 0, vf_gids = 0;
125 	u32 i;
126 	u32 vf_pct;
127 	u32 num_vfs;
128 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
129 
130 	rdev->qplib_ctx.qpc_count = min_t(u32, BNXT_RE_MAX_QPC_COUNT,
131 					  dev_attr->max_qp);
132 
133 	rdev->qplib_ctx.mrw_count = BNXT_RE_MAX_MRW_COUNT_256K;
134 	/* Use max_mr from fw since max_mrw does not get set */
135 	rdev->qplib_ctx.mrw_count = min_t(u32, rdev->qplib_ctx.mrw_count,
136 					  dev_attr->max_mr);
137 	rdev->qplib_ctx.srqc_count = min_t(u32, BNXT_RE_MAX_SRQC_COUNT,
138 					   dev_attr->max_srq);
139 	rdev->qplib_ctx.cq_count = min_t(u32, BNXT_RE_MAX_CQ_COUNT,
140 					 dev_attr->max_cq);
141 
142 	for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
143 		rdev->qplib_ctx.tqm_count[i] =
144 		rdev->dev_attr.tqm_alloc_reqs[i];
145 
146 	if (rdev->num_vfs) {
147 		/*
148 		 * Reserve a set of resources for the PF. Divide the remaining
149 		 * resources among the VFs
150 		 */
151 		vf_pct = 100 - BNXT_RE_PCT_RSVD_FOR_PF;
152 		num_vfs = 100 * rdev->num_vfs;
153 		vf_qps = (rdev->qplib_ctx.qpc_count * vf_pct) / num_vfs;
154 		vf_srqs = (rdev->qplib_ctx.srqc_count * vf_pct) / num_vfs;
155 		vf_cqs = (rdev->qplib_ctx.cq_count * vf_pct) / num_vfs;
156 		/*
157 		 * The driver allows many more MRs than other resources. If the
158 		 * firmware does also, then reserve a fixed amount for the PF
159 		 * and divide the rest among VFs. VFs may use many MRs for NFS
160 		 * mounts, ISER, NVME applications, etc. If the firmware
161 		 * severely restricts the number of MRs, then let PF have
162 		 * half and divide the rest among VFs, as for the other
163 		 * resource types.
164 		 */
165 		if (rdev->qplib_ctx.mrw_count < BNXT_RE_MAX_MRW_COUNT_64K)
166 			vf_mrws = rdev->qplib_ctx.mrw_count * vf_pct / num_vfs;
167 		else
168 			vf_mrws = (rdev->qplib_ctx.mrw_count -
169 				   BNXT_RE_RESVD_MR_FOR_PF) / rdev->num_vfs;
170 		vf_gids = BNXT_RE_MAX_GID_PER_VF;
171 	}
172 	rdev->qplib_ctx.vf_res.max_mrw_per_vf = vf_mrws;
173 	rdev->qplib_ctx.vf_res.max_gid_per_vf = vf_gids;
174 	rdev->qplib_ctx.vf_res.max_qp_per_vf = vf_qps;
175 	rdev->qplib_ctx.vf_res.max_srq_per_vf = vf_srqs;
176 	rdev->qplib_ctx.vf_res.max_cq_per_vf = vf_cqs;
177 }
178 
179 /* for handling bnxt_en callbacks later */
180 static void bnxt_re_stop(void *p)
181 {
182 }
183 
184 static void bnxt_re_start(void *p)
185 {
186 }
187 
188 static void bnxt_re_sriov_config(void *p, int num_vfs)
189 {
190 	struct bnxt_re_dev *rdev = p;
191 
192 	if (!rdev)
193 		return;
194 
195 	rdev->num_vfs = num_vfs;
196 	bnxt_re_set_resource_limits(rdev);
197 	bnxt_qplib_set_func_resources(&rdev->qplib_res, &rdev->rcfw,
198 				      &rdev->qplib_ctx);
199 }
200 
201 static void bnxt_re_shutdown(void *p)
202 {
203 	struct bnxt_re_dev *rdev = p;
204 
205 	if (!rdev)
206 		return;
207 
208 	bnxt_re_ib_unreg(rdev);
209 }
210 
211 static void bnxt_re_stop_irq(void *handle)
212 {
213 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
214 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
215 	struct bnxt_qplib_nq *nq;
216 	int indx;
217 
218 	for (indx = BNXT_RE_NQ_IDX; indx < rdev->num_msix; indx++) {
219 		nq = &rdev->nq[indx - 1];
220 		bnxt_qplib_nq_stop_irq(nq, false);
221 	}
222 
223 	bnxt_qplib_rcfw_stop_irq(rcfw, false);
224 }
225 
226 static void bnxt_re_start_irq(void *handle, struct bnxt_msix_entry *ent)
227 {
228 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
229 	struct bnxt_msix_entry *msix_ent = rdev->msix_entries;
230 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
231 	struct bnxt_qplib_nq *nq;
232 	int indx, rc;
233 
234 	if (!ent) {
235 		/* Not setting the f/w timeout bit in rcfw.
236 		 * During the driver unload the first command
237 		 * to f/w will timeout and that will set the
238 		 * timeout bit.
239 		 */
240 		dev_err(rdev_to_dev(rdev), "Failed to re-start IRQs\n");
241 		return;
242 	}
243 
244 	/* Vectors may change after restart, so update with new vectors
245 	 * in device sctructure.
246 	 */
247 	for (indx = 0; indx < rdev->num_msix; indx++)
248 		rdev->msix_entries[indx].vector = ent[indx].vector;
249 
250 	bnxt_qplib_rcfw_start_irq(rcfw, msix_ent[BNXT_RE_AEQ_IDX].vector,
251 				  false);
252 	for (indx = BNXT_RE_NQ_IDX ; indx < rdev->num_msix; indx++) {
253 		nq = &rdev->nq[indx - 1];
254 		rc = bnxt_qplib_nq_start_irq(nq, indx - 1,
255 					     msix_ent[indx].vector, false);
256 		if (rc)
257 			dev_warn(rdev_to_dev(rdev),
258 				 "Failed to reinit NQ index %d\n", indx - 1);
259 	}
260 }
261 
262 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
263 	.ulp_async_notifier = NULL,
264 	.ulp_stop = bnxt_re_stop,
265 	.ulp_start = bnxt_re_start,
266 	.ulp_sriov_config = bnxt_re_sriov_config,
267 	.ulp_shutdown = bnxt_re_shutdown,
268 	.ulp_irq_stop = bnxt_re_stop_irq,
269 	.ulp_irq_restart = bnxt_re_start_irq
270 };
271 
272 /* RoCE -> Net driver */
273 
274 /* Driver registration routines used to let the networking driver (bnxt_en)
275  * to know that the RoCE driver is now installed
276  */
277 static int bnxt_re_unregister_netdev(struct bnxt_re_dev *rdev)
278 {
279 	struct bnxt_en_dev *en_dev;
280 	int rc;
281 
282 	if (!rdev)
283 		return -EINVAL;
284 
285 	en_dev = rdev->en_dev;
286 
287 	rc = en_dev->en_ops->bnxt_unregister_device(rdev->en_dev,
288 						    BNXT_ROCE_ULP);
289 	return rc;
290 }
291 
292 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
293 {
294 	struct bnxt_en_dev *en_dev;
295 	int rc = 0;
296 
297 	if (!rdev)
298 		return -EINVAL;
299 
300 	en_dev = rdev->en_dev;
301 
302 	rc = en_dev->en_ops->bnxt_register_device(en_dev, BNXT_ROCE_ULP,
303 						  &bnxt_re_ulp_ops, rdev);
304 	rdev->qplib_res.pdev = rdev->en_dev->pdev;
305 	return rc;
306 }
307 
308 static int bnxt_re_free_msix(struct bnxt_re_dev *rdev)
309 {
310 	struct bnxt_en_dev *en_dev;
311 	int rc;
312 
313 	if (!rdev)
314 		return -EINVAL;
315 
316 	en_dev = rdev->en_dev;
317 
318 
319 	rc = en_dev->en_ops->bnxt_free_msix(rdev->en_dev, BNXT_ROCE_ULP);
320 
321 	return rc;
322 }
323 
324 static int bnxt_re_request_msix(struct bnxt_re_dev *rdev)
325 {
326 	int rc = 0, num_msix_want = BNXT_RE_MAX_MSIX, num_msix_got;
327 	struct bnxt_en_dev *en_dev;
328 
329 	if (!rdev)
330 		return -EINVAL;
331 
332 	en_dev = rdev->en_dev;
333 
334 	num_msix_want = min_t(u32, BNXT_RE_MAX_MSIX, num_online_cpus());
335 
336 	num_msix_got = en_dev->en_ops->bnxt_request_msix(en_dev, BNXT_ROCE_ULP,
337 							 rdev->msix_entries,
338 							 num_msix_want);
339 	if (num_msix_got < BNXT_RE_MIN_MSIX) {
340 		rc = -EINVAL;
341 		goto done;
342 	}
343 	if (num_msix_got != num_msix_want) {
344 		dev_warn(rdev_to_dev(rdev),
345 			 "Requested %d MSI-X vectors, got %d\n",
346 			 num_msix_want, num_msix_got);
347 	}
348 	rdev->num_msix = num_msix_got;
349 done:
350 	return rc;
351 }
352 
353 static void bnxt_re_init_hwrm_hdr(struct bnxt_re_dev *rdev, struct input *hdr,
354 				  u16 opcd, u16 crid, u16 trid)
355 {
356 	hdr->req_type = cpu_to_le16(opcd);
357 	hdr->cmpl_ring = cpu_to_le16(crid);
358 	hdr->target_id = cpu_to_le16(trid);
359 }
360 
361 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
362 				int msg_len, void *resp, int resp_max_len,
363 				int timeout)
364 {
365 	fw_msg->msg = msg;
366 	fw_msg->msg_len = msg_len;
367 	fw_msg->resp = resp;
368 	fw_msg->resp_max_len = resp_max_len;
369 	fw_msg->timeout = timeout;
370 }
371 
372 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev,
373 				 u16 fw_ring_id, int type)
374 {
375 	struct bnxt_en_dev *en_dev = rdev->en_dev;
376 	struct hwrm_ring_free_input req = {0};
377 	struct hwrm_ring_free_output resp;
378 	struct bnxt_fw_msg fw_msg;
379 	int rc = -EINVAL;
380 
381 	if (!en_dev)
382 		return rc;
383 
384 	memset(&fw_msg, 0, sizeof(fw_msg));
385 
386 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1);
387 	req.ring_type = type;
388 	req.ring_id = cpu_to_le16(fw_ring_id);
389 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
390 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
391 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
392 	if (rc)
393 		dev_err(rdev_to_dev(rdev),
394 			"Failed to free HW ring:%d :%#x", req.ring_id, rc);
395 	return rc;
396 }
397 
398 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev, dma_addr_t *dma_arr,
399 				  int pages, int type, u32 ring_mask,
400 				  u32 map_index, u16 *fw_ring_id)
401 {
402 	struct bnxt_en_dev *en_dev = rdev->en_dev;
403 	struct hwrm_ring_alloc_input req = {0};
404 	struct hwrm_ring_alloc_output resp;
405 	struct bnxt_fw_msg fw_msg;
406 	int rc = -EINVAL;
407 
408 	if (!en_dev)
409 		return rc;
410 
411 	memset(&fw_msg, 0, sizeof(fw_msg));
412 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_ALLOC, -1, -1);
413 	req.enables = 0;
414 	req.page_tbl_addr =  cpu_to_le64(dma_arr[0]);
415 	if (pages > 1) {
416 		/* Page size is in log2 units */
417 		req.page_size = BNXT_PAGE_SHIFT;
418 		req.page_tbl_depth = 1;
419 	}
420 	req.fbo = 0;
421 	/* Association of ring index with doorbell index and MSIX number */
422 	req.logical_id = cpu_to_le16(map_index);
423 	req.length = cpu_to_le32(ring_mask + 1);
424 	req.ring_type = type;
425 	req.int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
426 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
427 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
428 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
429 	if (!rc)
430 		*fw_ring_id = le16_to_cpu(resp.ring_id);
431 
432 	return rc;
433 }
434 
435 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
436 				      u32 fw_stats_ctx_id)
437 {
438 	struct bnxt_en_dev *en_dev = rdev->en_dev;
439 	struct hwrm_stat_ctx_free_input req = {0};
440 	struct bnxt_fw_msg fw_msg;
441 	int rc = -EINVAL;
442 
443 	if (!en_dev)
444 		return rc;
445 
446 	memset(&fw_msg, 0, sizeof(fw_msg));
447 
448 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_FREE, -1, -1);
449 	req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
450 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&req,
451 			    sizeof(req), DFLT_HWRM_CMD_TIMEOUT);
452 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
453 	if (rc)
454 		dev_err(rdev_to_dev(rdev),
455 			"Failed to free HW stats context %#x", rc);
456 
457 	return rc;
458 }
459 
460 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
461 				       dma_addr_t dma_map,
462 				       u32 *fw_stats_ctx_id)
463 {
464 	struct hwrm_stat_ctx_alloc_output resp = {0};
465 	struct hwrm_stat_ctx_alloc_input req = {0};
466 	struct bnxt_en_dev *en_dev = rdev->en_dev;
467 	struct bnxt_fw_msg fw_msg;
468 	int rc = -EINVAL;
469 
470 	*fw_stats_ctx_id = INVALID_STATS_CTX_ID;
471 
472 	if (!en_dev)
473 		return rc;
474 
475 	memset(&fw_msg, 0, sizeof(fw_msg));
476 
477 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1);
478 	req.update_period_ms = cpu_to_le32(1000);
479 	req.stats_dma_addr = cpu_to_le64(dma_map);
480 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
481 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
482 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
483 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
484 	if (!rc)
485 		*fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
486 
487 	return rc;
488 }
489 
490 /* Device */
491 
492 static bool is_bnxt_re_dev(struct net_device *netdev)
493 {
494 	struct ethtool_drvinfo drvinfo;
495 
496 	if (netdev->ethtool_ops && netdev->ethtool_ops->get_drvinfo) {
497 		memset(&drvinfo, 0, sizeof(drvinfo));
498 		netdev->ethtool_ops->get_drvinfo(netdev, &drvinfo);
499 
500 		if (strcmp(drvinfo.driver, "bnxt_en"))
501 			return false;
502 		return true;
503 	}
504 	return false;
505 }
506 
507 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
508 {
509 	struct bnxt_re_dev *rdev;
510 
511 	rcu_read_lock();
512 	list_for_each_entry_rcu(rdev, &bnxt_re_dev_list, list) {
513 		if (rdev->netdev == netdev) {
514 			rcu_read_unlock();
515 			return rdev;
516 		}
517 	}
518 	rcu_read_unlock();
519 	return NULL;
520 }
521 
522 static void bnxt_re_dev_unprobe(struct net_device *netdev,
523 				struct bnxt_en_dev *en_dev)
524 {
525 	dev_put(netdev);
526 	module_put(en_dev->pdev->driver->driver.owner);
527 }
528 
529 static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev)
530 {
531 	struct bnxt *bp = netdev_priv(netdev);
532 	struct bnxt_en_dev *en_dev;
533 	struct pci_dev *pdev;
534 
535 	/* Call bnxt_en's RoCE probe via indirect API */
536 	if (!bp->ulp_probe)
537 		return ERR_PTR(-EINVAL);
538 
539 	en_dev = bp->ulp_probe(netdev);
540 	if (IS_ERR(en_dev))
541 		return en_dev;
542 
543 	pdev = en_dev->pdev;
544 	if (!pdev)
545 		return ERR_PTR(-EINVAL);
546 
547 	if (!(en_dev->flags & BNXT_EN_FLAG_ROCE_CAP)) {
548 		dev_info(&pdev->dev,
549 			"%s: probe error: RoCE is not supported on this device",
550 			ROCE_DRV_MODULE_NAME);
551 		return ERR_PTR(-ENODEV);
552 	}
553 
554 	/* Bump net device reference count */
555 	if (!try_module_get(pdev->driver->driver.owner))
556 		return ERR_PTR(-ENODEV);
557 
558 	dev_hold(netdev);
559 
560 	return en_dev;
561 }
562 
563 static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr,
564 			   char *buf)
565 {
566 	struct bnxt_re_dev *rdev =
567 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
568 
569 	return scnprintf(buf, PAGE_SIZE, "0x%x\n", rdev->en_dev->pdev->vendor);
570 }
571 static DEVICE_ATTR_RO(hw_rev);
572 
573 static ssize_t hca_type_show(struct device *device,
574 			     struct device_attribute *attr, char *buf)
575 {
576 	struct bnxt_re_dev *rdev =
577 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
578 
579 	return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->ibdev.node_desc);
580 }
581 static DEVICE_ATTR_RO(hca_type);
582 
583 static struct attribute *bnxt_re_attributes[] = {
584 	&dev_attr_hw_rev.attr,
585 	&dev_attr_hca_type.attr,
586 	NULL
587 };
588 
589 static const struct attribute_group bnxt_re_dev_attr_group = {
590 	.attrs = bnxt_re_attributes,
591 };
592 
593 static void bnxt_re_unregister_ib(struct bnxt_re_dev *rdev)
594 {
595 	ib_unregister_device(&rdev->ibdev);
596 }
597 
598 static const struct ib_device_ops bnxt_re_dev_ops = {
599 	.add_gid = bnxt_re_add_gid,
600 	.alloc_hw_stats = bnxt_re_ib_alloc_hw_stats,
601 	.alloc_mr = bnxt_re_alloc_mr,
602 	.alloc_pd = bnxt_re_alloc_pd,
603 	.alloc_ucontext = bnxt_re_alloc_ucontext,
604 	.create_ah = bnxt_re_create_ah,
605 	.create_cq = bnxt_re_create_cq,
606 	.create_qp = bnxt_re_create_qp,
607 	.create_srq = bnxt_re_create_srq,
608 	.dealloc_pd = bnxt_re_dealloc_pd,
609 	.dealloc_ucontext = bnxt_re_dealloc_ucontext,
610 	.del_gid = bnxt_re_del_gid,
611 	.dereg_mr = bnxt_re_dereg_mr,
612 	.destroy_ah = bnxt_re_destroy_ah,
613 	.destroy_cq = bnxt_re_destroy_cq,
614 	.destroy_qp = bnxt_re_destroy_qp,
615 	.destroy_srq = bnxt_re_destroy_srq,
616 	.get_dev_fw_str = bnxt_re_query_fw_str,
617 	.get_dma_mr = bnxt_re_get_dma_mr,
618 	.get_hw_stats = bnxt_re_ib_get_hw_stats,
619 	.get_link_layer = bnxt_re_get_link_layer,
620 	.get_netdev = bnxt_re_get_netdev,
621 	.get_port_immutable = bnxt_re_get_port_immutable,
622 	.map_mr_sg = bnxt_re_map_mr_sg,
623 	.mmap = bnxt_re_mmap,
624 	.modify_ah = bnxt_re_modify_ah,
625 	.modify_device = bnxt_re_modify_device,
626 	.modify_qp = bnxt_re_modify_qp,
627 	.modify_srq = bnxt_re_modify_srq,
628 	.poll_cq = bnxt_re_poll_cq,
629 	.post_recv = bnxt_re_post_recv,
630 	.post_send = bnxt_re_post_send,
631 	.post_srq_recv = bnxt_re_post_srq_recv,
632 	.query_ah = bnxt_re_query_ah,
633 	.query_device = bnxt_re_query_device,
634 	.query_pkey = bnxt_re_query_pkey,
635 	.query_port = bnxt_re_query_port,
636 	.query_qp = bnxt_re_query_qp,
637 	.query_srq = bnxt_re_query_srq,
638 	.reg_user_mr = bnxt_re_reg_user_mr,
639 	.req_notify_cq = bnxt_re_req_notify_cq,
640 	INIT_RDMA_OBJ_SIZE(ib_pd, bnxt_re_pd, ib_pd),
641 	INIT_RDMA_OBJ_SIZE(ib_ucontext, bnxt_re_ucontext, ib_uctx),
642 };
643 
644 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
645 {
646 	struct ib_device *ibdev = &rdev->ibdev;
647 
648 	/* ib device init */
649 	ibdev->owner = THIS_MODULE;
650 	ibdev->node_type = RDMA_NODE_IB_CA;
651 	strlcpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
652 		strlen(BNXT_RE_DESC) + 5);
653 	ibdev->phys_port_cnt = 1;
654 
655 	bnxt_qplib_get_guid(rdev->netdev->dev_addr, (u8 *)&ibdev->node_guid);
656 
657 	ibdev->num_comp_vectors	= 1;
658 	ibdev->dev.parent = &rdev->en_dev->pdev->dev;
659 	ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
660 
661 	/* User space */
662 	ibdev->uverbs_abi_ver = BNXT_RE_ABI_VERSION;
663 	ibdev->uverbs_cmd_mask =
664 			(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
665 			(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
666 			(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
667 			(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
668 			(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
669 			(1ull << IB_USER_VERBS_CMD_REG_MR)		|
670 			(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
671 			(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
672 			(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
673 			(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
674 			(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
675 			(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
676 			(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
677 			(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
678 			(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
679 			(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
680 			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
681 			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
682 			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
683 			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
684 			(1ull << IB_USER_VERBS_CMD_CREATE_AH)		|
685 			(1ull << IB_USER_VERBS_CMD_MODIFY_AH)		|
686 			(1ull << IB_USER_VERBS_CMD_QUERY_AH)		|
687 			(1ull << IB_USER_VERBS_CMD_DESTROY_AH);
688 	/* POLL_CQ and REQ_NOTIFY_CQ is directly handled in libbnxt_re */
689 
690 
691 	rdma_set_device_sysfs_group(ibdev, &bnxt_re_dev_attr_group);
692 	ibdev->driver_id = RDMA_DRIVER_BNXT_RE;
693 	ib_set_device_ops(ibdev, &bnxt_re_dev_ops);
694 	return ib_register_device(ibdev, "bnxt_re%d");
695 }
696 
697 static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev)
698 {
699 	dev_put(rdev->netdev);
700 	rdev->netdev = NULL;
701 
702 	mutex_lock(&bnxt_re_dev_lock);
703 	list_del_rcu(&rdev->list);
704 	mutex_unlock(&bnxt_re_dev_lock);
705 
706 	synchronize_rcu();
707 
708 	ib_dealloc_device(&rdev->ibdev);
709 	/* rdev is gone */
710 }
711 
712 static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev,
713 					   struct bnxt_en_dev *en_dev)
714 {
715 	struct bnxt_re_dev *rdev;
716 
717 	/* Allocate bnxt_re_dev instance here */
718 	rdev = ib_alloc_device(bnxt_re_dev, ibdev);
719 	if (!rdev) {
720 		dev_err(NULL, "%s: bnxt_re_dev allocation failure!",
721 			ROCE_DRV_MODULE_NAME);
722 		return NULL;
723 	}
724 	/* Default values */
725 	rdev->netdev = netdev;
726 	dev_hold(rdev->netdev);
727 	rdev->en_dev = en_dev;
728 	rdev->id = rdev->en_dev->pdev->devfn;
729 	INIT_LIST_HEAD(&rdev->qp_list);
730 	mutex_init(&rdev->qp_lock);
731 	atomic_set(&rdev->qp_count, 0);
732 	atomic_set(&rdev->cq_count, 0);
733 	atomic_set(&rdev->srq_count, 0);
734 	atomic_set(&rdev->mr_count, 0);
735 	atomic_set(&rdev->mw_count, 0);
736 	rdev->cosq[0] = 0xFFFF;
737 	rdev->cosq[1] = 0xFFFF;
738 
739 	mutex_lock(&bnxt_re_dev_lock);
740 	list_add_tail_rcu(&rdev->list, &bnxt_re_dev_list);
741 	mutex_unlock(&bnxt_re_dev_lock);
742 	return rdev;
743 }
744 
745 static int bnxt_re_handle_unaffi_async_event(struct creq_func_event
746 					     *unaffi_async)
747 {
748 	switch (unaffi_async->event) {
749 	case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
750 		break;
751 	case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
752 		break;
753 	case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
754 		break;
755 	case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
756 		break;
757 	case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
758 		break;
759 	case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
760 		break;
761 	case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
762 		break;
763 	case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
764 		break;
765 	case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
766 		break;
767 	case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
768 		break;
769 	case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
770 		break;
771 	default:
772 		return -EINVAL;
773 	}
774 	return 0;
775 }
776 
777 static int bnxt_re_handle_qp_async_event(struct creq_qp_event *qp_event,
778 					 struct bnxt_re_qp *qp)
779 {
780 	struct ib_event event;
781 	unsigned int flags;
782 
783 	if (qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
784 		flags = bnxt_re_lock_cqs(qp);
785 		bnxt_qplib_add_flush_qp(&qp->qplib_qp);
786 		bnxt_re_unlock_cqs(qp, flags);
787 	}
788 
789 	memset(&event, 0, sizeof(event));
790 	if (qp->qplib_qp.srq) {
791 		event.device = &qp->rdev->ibdev;
792 		event.element.qp = &qp->ib_qp;
793 		event.event = IB_EVENT_QP_LAST_WQE_REACHED;
794 	}
795 
796 	if (event.device && qp->ib_qp.event_handler)
797 		qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
798 
799 	return 0;
800 }
801 
802 static int bnxt_re_handle_affi_async_event(struct creq_qp_event *affi_async,
803 					   void *obj)
804 {
805 	int rc = 0;
806 	u8 event;
807 
808 	if (!obj)
809 		return rc; /* QP was already dead, still return success */
810 
811 	event = affi_async->event;
812 	if (event == CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION) {
813 		struct bnxt_qplib_qp *lib_qp = obj;
814 		struct bnxt_re_qp *qp = container_of(lib_qp, struct bnxt_re_qp,
815 						     qplib_qp);
816 		rc = bnxt_re_handle_qp_async_event(affi_async, qp);
817 	}
818 	return rc;
819 }
820 
821 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
822 			       void *aeqe, void *obj)
823 {
824 	struct creq_qp_event *affi_async;
825 	struct creq_func_event *unaffi_async;
826 	u8 type;
827 	int rc;
828 
829 	type = ((struct creq_base *)aeqe)->type;
830 	if (type == CREQ_BASE_TYPE_FUNC_EVENT) {
831 		unaffi_async = aeqe;
832 		rc = bnxt_re_handle_unaffi_async_event(unaffi_async);
833 	} else {
834 		affi_async = aeqe;
835 		rc = bnxt_re_handle_affi_async_event(affi_async, obj);
836 	}
837 
838 	return rc;
839 }
840 
841 static int bnxt_re_srqn_handler(struct bnxt_qplib_nq *nq,
842 				struct bnxt_qplib_srq *handle, u8 event)
843 {
844 	struct bnxt_re_srq *srq = container_of(handle, struct bnxt_re_srq,
845 					       qplib_srq);
846 	struct ib_event ib_event;
847 	int rc = 0;
848 
849 	if (!srq) {
850 		dev_err(NULL, "%s: SRQ is NULL, SRQN not handled",
851 			ROCE_DRV_MODULE_NAME);
852 		rc = -EINVAL;
853 		goto done;
854 	}
855 	ib_event.device = &srq->rdev->ibdev;
856 	ib_event.element.srq = &srq->ib_srq;
857 	if (event == NQ_SRQ_EVENT_EVENT_SRQ_THRESHOLD_EVENT)
858 		ib_event.event = IB_EVENT_SRQ_LIMIT_REACHED;
859 	else
860 		ib_event.event = IB_EVENT_SRQ_ERR;
861 
862 	if (srq->ib_srq.event_handler) {
863 		/* Lock event_handler? */
864 		(*srq->ib_srq.event_handler)(&ib_event,
865 					     srq->ib_srq.srq_context);
866 	}
867 done:
868 	return rc;
869 }
870 
871 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
872 			       struct bnxt_qplib_cq *handle)
873 {
874 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
875 					     qplib_cq);
876 
877 	if (!cq) {
878 		dev_err(NULL, "%s: CQ is NULL, CQN not handled",
879 			ROCE_DRV_MODULE_NAME);
880 		return -EINVAL;
881 	}
882 	if (cq->ib_cq.comp_handler) {
883 		/* Lock comp_handler? */
884 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
885 	}
886 
887 	return 0;
888 }
889 
890 static u32 bnxt_re_get_nqdb_offset(struct bnxt_re_dev *rdev, u16 indx)
891 {
892 	return bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx) ?
893 				0x10000 : rdev->msix_entries[indx].db_offset;
894 }
895 
896 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
897 {
898 	int i;
899 
900 	for (i = 1; i < rdev->num_msix; i++)
901 		bnxt_qplib_disable_nq(&rdev->nq[i - 1]);
902 
903 	if (rdev->qplib_res.rcfw)
904 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
905 }
906 
907 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
908 {
909 	int num_vec_enabled = 0;
910 	int rc = 0, i;
911 	u32 db_offt;
912 
913 	bnxt_qplib_init_res(&rdev->qplib_res);
914 
915 	for (i = 1; i < rdev->num_msix ; i++) {
916 		db_offt = bnxt_re_get_nqdb_offset(rdev, i);
917 		rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
918 					  i - 1, rdev->msix_entries[i].vector,
919 					  db_offt, &bnxt_re_cqn_handler,
920 					  &bnxt_re_srqn_handler);
921 		if (rc) {
922 			dev_err(rdev_to_dev(rdev),
923 				"Failed to enable NQ with rc = 0x%x", rc);
924 			goto fail;
925 		}
926 		num_vec_enabled++;
927 	}
928 	return 0;
929 fail:
930 	for (i = num_vec_enabled; i >= 0; i--)
931 		bnxt_qplib_disable_nq(&rdev->nq[i]);
932 	return rc;
933 }
934 
935 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev)
936 {
937 	u8 type;
938 	int i;
939 
940 	for (i = 0; i < rdev->num_msix - 1; i++) {
941 		type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
942 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
943 		rdev->nq[i].res = NULL;
944 		bnxt_qplib_free_nq(&rdev->nq[i]);
945 	}
946 }
947 
948 static void bnxt_re_free_res(struct bnxt_re_dev *rdev)
949 {
950 	bnxt_re_free_nq_res(rdev);
951 
952 	if (rdev->qplib_res.dpi_tbl.max) {
953 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
954 				       &rdev->qplib_res.dpi_tbl,
955 				       &rdev->dpi_privileged);
956 	}
957 	if (rdev->qplib_res.rcfw) {
958 		bnxt_qplib_free_res(&rdev->qplib_res);
959 		rdev->qplib_res.rcfw = NULL;
960 	}
961 }
962 
963 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
964 {
965 	int num_vec_created = 0;
966 	dma_addr_t *pg_map;
967 	int rc = 0, i;
968 	int pages;
969 	u8 type;
970 
971 	/* Configure and allocate resources for qplib */
972 	rdev->qplib_res.rcfw = &rdev->rcfw;
973 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
974 				     rdev->is_virtfn);
975 	if (rc)
976 		goto fail;
977 
978 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
979 				  rdev->netdev, &rdev->dev_attr);
980 	if (rc)
981 		goto fail;
982 
983 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
984 				  &rdev->dpi_privileged,
985 				  rdev);
986 	if (rc)
987 		goto dealloc_res;
988 
989 	for (i = 0; i < rdev->num_msix - 1; i++) {
990 		rdev->nq[i].res = &rdev->qplib_res;
991 		rdev->nq[i].hwq.max_elements = BNXT_RE_MAX_CQ_COUNT +
992 			BNXT_RE_MAX_SRQC_COUNT + 2;
993 		rc = bnxt_qplib_alloc_nq(rdev->en_dev->pdev, &rdev->nq[i]);
994 		if (rc) {
995 			dev_err(rdev_to_dev(rdev), "Alloc Failed NQ%d rc:%#x",
996 				i, rc);
997 			goto free_nq;
998 		}
999 		type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1000 		pg_map = rdev->nq[i].hwq.pbl[PBL_LVL_0].pg_map_arr;
1001 		pages = rdev->nq[i].hwq.pbl[rdev->nq[i].hwq.level].pg_count;
1002 		rc = bnxt_re_net_ring_alloc(rdev, pg_map, pages, type,
1003 					    BNXT_QPLIB_NQE_MAX_CNT - 1,
1004 					    rdev->msix_entries[i + 1].ring_idx,
1005 					    &rdev->nq[i].ring_id);
1006 		if (rc) {
1007 			dev_err(rdev_to_dev(rdev),
1008 				"Failed to allocate NQ fw id with rc = 0x%x",
1009 				rc);
1010 			bnxt_qplib_free_nq(&rdev->nq[i]);
1011 			goto free_nq;
1012 		}
1013 		num_vec_created++;
1014 	}
1015 	return 0;
1016 free_nq:
1017 	for (i = num_vec_created; i >= 0; i--) {
1018 		type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1019 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
1020 		bnxt_qplib_free_nq(&rdev->nq[i]);
1021 	}
1022 	bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
1023 			       &rdev->qplib_res.dpi_tbl,
1024 			       &rdev->dpi_privileged);
1025 dealloc_res:
1026 	bnxt_qplib_free_res(&rdev->qplib_res);
1027 
1028 fail:
1029 	rdev->qplib_res.rcfw = NULL;
1030 	return rc;
1031 }
1032 
1033 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
1034 				   u8 port_num, enum ib_event_type event)
1035 {
1036 	struct ib_event ib_event;
1037 
1038 	ib_event.device = ibdev;
1039 	if (qp) {
1040 		ib_event.element.qp = qp;
1041 		ib_event.event = event;
1042 		if (qp->event_handler)
1043 			qp->event_handler(&ib_event, qp->qp_context);
1044 
1045 	} else {
1046 		ib_event.element.port_num = port_num;
1047 		ib_event.event = event;
1048 		ib_dispatch_event(&ib_event);
1049 	}
1050 }
1051 
1052 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN      0x02
1053 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
1054 				      u64 *cid_map)
1055 {
1056 	struct hwrm_queue_pri2cos_qcfg_input req = {0};
1057 	struct bnxt *bp = netdev_priv(rdev->netdev);
1058 	struct hwrm_queue_pri2cos_qcfg_output resp;
1059 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1060 	struct bnxt_fw_msg fw_msg;
1061 	u32 flags = 0;
1062 	u8 *qcfgmap, *tmp_map;
1063 	int rc = 0, i;
1064 
1065 	if (!cid_map)
1066 		return -EINVAL;
1067 
1068 	memset(&fw_msg, 0, sizeof(fw_msg));
1069 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1070 			      HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
1071 	flags |= (dir & 0x01);
1072 	flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
1073 	req.flags = cpu_to_le32(flags);
1074 	req.port_id = bp->pf.port_id;
1075 
1076 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1077 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1078 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1079 	if (rc)
1080 		return rc;
1081 
1082 	if (resp.queue_cfg_info) {
1083 		dev_warn(rdev_to_dev(rdev),
1084 			 "Asymmetric cos queue configuration detected");
1085 		dev_warn(rdev_to_dev(rdev),
1086 			 " on device, QoS may not be fully functional\n");
1087 	}
1088 	qcfgmap = &resp.pri0_cos_queue_id;
1089 	tmp_map = (u8 *)cid_map;
1090 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
1091 		tmp_map[i] = qcfgmap[i];
1092 
1093 	return rc;
1094 }
1095 
1096 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
1097 					struct bnxt_re_qp *qp)
1098 {
1099 	return (qp->ib_qp.qp_type == IB_QPT_GSI) || (qp == rdev->qp1_sqp);
1100 }
1101 
1102 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
1103 {
1104 	int mask = IB_QP_STATE;
1105 	struct ib_qp_attr qp_attr;
1106 	struct bnxt_re_qp *qp;
1107 
1108 	qp_attr.qp_state = IB_QPS_ERR;
1109 	mutex_lock(&rdev->qp_lock);
1110 	list_for_each_entry(qp, &rdev->qp_list, list) {
1111 		/* Modify the state of all QPs except QP1/Shadow QP */
1112 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
1113 			if (qp->qplib_qp.state !=
1114 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
1115 			    qp->qplib_qp.state !=
1116 			    CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1117 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
1118 						       1, IB_EVENT_QP_FATAL);
1119 				bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
1120 						  NULL);
1121 			}
1122 		}
1123 	}
1124 	mutex_unlock(&rdev->qp_lock);
1125 }
1126 
1127 static int bnxt_re_update_gid(struct bnxt_re_dev *rdev)
1128 {
1129 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
1130 	struct bnxt_qplib_gid gid;
1131 	u16 gid_idx, index;
1132 	int rc = 0;
1133 
1134 	if (!test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1135 		return 0;
1136 
1137 	if (!sgid_tbl) {
1138 		dev_err(rdev_to_dev(rdev), "QPLIB: SGID table not allocated");
1139 		return -EINVAL;
1140 	}
1141 
1142 	for (index = 0; index < sgid_tbl->active; index++) {
1143 		gid_idx = sgid_tbl->hw_id[index];
1144 
1145 		if (!memcmp(&sgid_tbl->tbl[index], &bnxt_qplib_gid_zero,
1146 			    sizeof(bnxt_qplib_gid_zero)))
1147 			continue;
1148 		/* need to modify the VLAN enable setting of non VLAN GID only
1149 		 * as setting is done for VLAN GID while adding GID
1150 		 */
1151 		if (sgid_tbl->vlan[index])
1152 			continue;
1153 
1154 		memcpy(&gid, &sgid_tbl->tbl[index], sizeof(gid));
1155 
1156 		rc = bnxt_qplib_update_sgid(sgid_tbl, &gid, gid_idx,
1157 					    rdev->qplib_res.netdev->dev_addr);
1158 	}
1159 
1160 	return rc;
1161 }
1162 
1163 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
1164 {
1165 	u32 prio_map = 0, tmp_map = 0;
1166 	struct net_device *netdev;
1167 	struct dcb_app app;
1168 
1169 	netdev = rdev->netdev;
1170 
1171 	memset(&app, 0, sizeof(app));
1172 	app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
1173 	app.protocol = ETH_P_IBOE;
1174 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1175 	prio_map = tmp_map;
1176 
1177 	app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
1178 	app.protocol = ROCE_V2_UDP_DPORT;
1179 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1180 	prio_map |= tmp_map;
1181 
1182 	return prio_map;
1183 }
1184 
1185 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
1186 {
1187 	u16 prio;
1188 	u8 id;
1189 
1190 	for (prio = 0, id = 0; prio < 8; prio++) {
1191 		if (prio_map & (1 << prio)) {
1192 			cosq[id] = cid_map[prio];
1193 			id++;
1194 			if (id == 2) /* Max 2 tcs supported */
1195 				break;
1196 		}
1197 	}
1198 }
1199 
1200 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
1201 {
1202 	u8 prio_map = 0;
1203 	u64 cid_map;
1204 	int rc;
1205 
1206 	/* Get priority for roce */
1207 	prio_map = bnxt_re_get_priority_mask(rdev);
1208 
1209 	if (prio_map == rdev->cur_prio_map)
1210 		return 0;
1211 	rdev->cur_prio_map = prio_map;
1212 	/* Get cosq id for this priority */
1213 	rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
1214 	if (rc) {
1215 		dev_warn(rdev_to_dev(rdev), "no cos for p_mask %x\n", prio_map);
1216 		return rc;
1217 	}
1218 	/* Parse CoS IDs for app priority */
1219 	bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
1220 
1221 	/* Config BONO. */
1222 	rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
1223 	if (rc) {
1224 		dev_warn(rdev_to_dev(rdev), "no tc for cos{%x, %x}\n",
1225 			 rdev->cosq[0], rdev->cosq[1]);
1226 		return rc;
1227 	}
1228 
1229 	/* Actual priorities are not programmed as they are already
1230 	 * done by L2 driver; just enable or disable priority vlan tagging
1231 	 */
1232 	if ((prio_map == 0 && rdev->qplib_res.prio) ||
1233 	    (prio_map != 0 && !rdev->qplib_res.prio)) {
1234 		rdev->qplib_res.prio = prio_map ? true : false;
1235 
1236 		bnxt_re_update_gid(rdev);
1237 	}
1238 
1239 	return 0;
1240 }
1241 
1242 static void bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev)
1243 {
1244 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1245 	struct hwrm_ver_get_output resp = {0};
1246 	struct hwrm_ver_get_input req = {0};
1247 	struct bnxt_fw_msg fw_msg;
1248 	int rc = 0;
1249 
1250 	memset(&fw_msg, 0, sizeof(fw_msg));
1251 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1252 			      HWRM_VER_GET, -1, -1);
1253 	req.hwrm_intf_maj = HWRM_VERSION_MAJOR;
1254 	req.hwrm_intf_min = HWRM_VERSION_MINOR;
1255 	req.hwrm_intf_upd = HWRM_VERSION_UPDATE;
1256 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1257 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1258 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1259 	if (rc) {
1260 		dev_err(rdev_to_dev(rdev),
1261 			"Failed to query HW version, rc = 0x%x", rc);
1262 		return;
1263 	}
1264 	rdev->qplib_ctx.hwrm_intf_ver =
1265 		(u64)resp.hwrm_intf_major << 48 |
1266 		(u64)resp.hwrm_intf_minor << 32 |
1267 		(u64)resp.hwrm_intf_build << 16 |
1268 		resp.hwrm_intf_patch;
1269 }
1270 
1271 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev)
1272 {
1273 	u8 type;
1274 	int rc;
1275 
1276 	if (test_and_clear_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags)) {
1277 		/* Cleanup ib dev */
1278 		bnxt_re_unregister_ib(rdev);
1279 	}
1280 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
1281 		cancel_delayed_work_sync(&rdev->worker);
1282 
1283 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED,
1284 			       &rdev->flags))
1285 		bnxt_re_cleanup_res(rdev);
1286 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags))
1287 		bnxt_re_free_res(rdev);
1288 
1289 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
1290 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
1291 		if (rc)
1292 			dev_warn(rdev_to_dev(rdev),
1293 				 "Failed to deinitialize RCFW: %#x", rc);
1294 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1295 		bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1296 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1297 		type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1298 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, type);
1299 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1300 	}
1301 	if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
1302 		rc = bnxt_re_free_msix(rdev);
1303 		if (rc)
1304 			dev_warn(rdev_to_dev(rdev),
1305 				 "Failed to free MSI-X vectors: %#x", rc);
1306 	}
1307 
1308 	bnxt_re_destroy_chip_ctx(rdev);
1309 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
1310 		rc = bnxt_re_unregister_netdev(rdev);
1311 		if (rc)
1312 			dev_warn(rdev_to_dev(rdev),
1313 				 "Failed to unregister with netdev: %#x", rc);
1314 	}
1315 }
1316 
1317 /* worker thread for polling periodic events. Now used for QoS programming*/
1318 static void bnxt_re_worker(struct work_struct *work)
1319 {
1320 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
1321 						worker.work);
1322 
1323 	bnxt_re_setup_qos(rdev);
1324 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1325 }
1326 
1327 static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
1328 {
1329 	dma_addr_t *pg_map;
1330 	u32 db_offt, ridx;
1331 	int pages, vid;
1332 	bool locked;
1333 	u8 type;
1334 	int rc;
1335 
1336 	/* Acquire rtnl lock through out this function */
1337 	rtnl_lock();
1338 	locked = true;
1339 
1340 	/* Registered a new RoCE device instance to netdev */
1341 	rc = bnxt_re_register_netdev(rdev);
1342 	if (rc) {
1343 		rtnl_unlock();
1344 		pr_err("Failed to register with netedev: %#x\n", rc);
1345 		return -EINVAL;
1346 	}
1347 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1348 
1349 	rc = bnxt_re_setup_chip_ctx(rdev);
1350 	if (rc) {
1351 		dev_err(rdev_to_dev(rdev), "Failed to get chip context\n");
1352 		return -EINVAL;
1353 	}
1354 
1355 	/* Check whether VF or PF */
1356 	bnxt_re_get_sriov_func_type(rdev);
1357 
1358 	rc = bnxt_re_request_msix(rdev);
1359 	if (rc) {
1360 		pr_err("Failed to get MSI-X vectors: %#x\n", rc);
1361 		rc = -EINVAL;
1362 		goto fail;
1363 	}
1364 	set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
1365 
1366 	bnxt_re_query_hwrm_intf_version(rdev);
1367 
1368 	/* Establish RCFW Communication Channel to initialize the context
1369 	 * memory for the function and all child VFs
1370 	 */
1371 	rc = bnxt_qplib_alloc_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw,
1372 					   &rdev->qplib_ctx,
1373 					   BNXT_RE_MAX_QPC_COUNT);
1374 	if (rc) {
1375 		pr_err("Failed to allocate RCFW Channel: %#x\n", rc);
1376 		goto fail;
1377 	}
1378 	type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1379 	pg_map = rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr;
1380 	pages = rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count;
1381 	ridx = rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx;
1382 	rc = bnxt_re_net_ring_alloc(rdev, pg_map, pages, type,
1383 				    BNXT_QPLIB_CREQE_MAX_CNT - 1,
1384 				    ridx, &rdev->rcfw.creq_ring_id);
1385 	if (rc) {
1386 		pr_err("Failed to allocate CREQ: %#x\n", rc);
1387 		goto free_rcfw;
1388 	}
1389 	db_offt = bnxt_re_get_nqdb_offset(rdev, BNXT_RE_AEQ_IDX);
1390 	vid = rdev->msix_entries[BNXT_RE_AEQ_IDX].vector;
1391 	rc = bnxt_qplib_enable_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw,
1392 					    vid, db_offt, rdev->is_virtfn,
1393 					    &bnxt_re_aeq_handler);
1394 	if (rc) {
1395 		pr_err("Failed to enable RCFW channel: %#x\n", rc);
1396 		goto free_ring;
1397 	}
1398 
1399 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
1400 				     rdev->is_virtfn);
1401 	if (rc)
1402 		goto disable_rcfw;
1403 	if (!rdev->is_virtfn)
1404 		bnxt_re_set_resource_limits(rdev);
1405 
1406 	rc = bnxt_qplib_alloc_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx, 0,
1407 				  bnxt_qplib_is_chip_gen_p5(&rdev->chip_ctx));
1408 	if (rc) {
1409 		pr_err("Failed to allocate QPLIB context: %#x\n", rc);
1410 		goto disable_rcfw;
1411 	}
1412 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
1413 					 rdev->qplib_ctx.stats.dma_map,
1414 					 &rdev->qplib_ctx.stats.fw_id);
1415 	if (rc) {
1416 		pr_err("Failed to allocate stats context: %#x\n", rc);
1417 		goto free_ctx;
1418 	}
1419 
1420 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx,
1421 				  rdev->is_virtfn);
1422 	if (rc) {
1423 		pr_err("Failed to initialize RCFW: %#x\n", rc);
1424 		goto free_sctx;
1425 	}
1426 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1427 
1428 	/* Resources based on the 'new' device caps */
1429 	rc = bnxt_re_alloc_res(rdev);
1430 	if (rc) {
1431 		pr_err("Failed to allocate resources: %#x\n", rc);
1432 		goto fail;
1433 	}
1434 	set_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags);
1435 	rc = bnxt_re_init_res(rdev);
1436 	if (rc) {
1437 		pr_err("Failed to initialize resources: %#x\n", rc);
1438 		goto fail;
1439 	}
1440 
1441 	set_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED, &rdev->flags);
1442 
1443 	if (!rdev->is_virtfn) {
1444 		rc = bnxt_re_setup_qos(rdev);
1445 		if (rc)
1446 			pr_info("RoCE priority not yet configured\n");
1447 
1448 		INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1449 		set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1450 		schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1451 	}
1452 
1453 	rtnl_unlock();
1454 	locked = false;
1455 
1456 	/* Register ib dev */
1457 	rc = bnxt_re_register_ib(rdev);
1458 	if (rc) {
1459 		pr_err("Failed to register with IB: %#x\n", rc);
1460 		goto fail;
1461 	}
1462 	set_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags);
1463 	dev_info(rdev_to_dev(rdev), "Device registered successfully");
1464 	ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1465 			 &rdev->active_width);
1466 	set_bit(BNXT_RE_FLAG_ISSUE_ROCE_STATS, &rdev->flags);
1467 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_PORT_ACTIVE);
1468 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_GID_CHANGE);
1469 
1470 	return 0;
1471 free_sctx:
1472 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1473 free_ctx:
1474 	bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1475 disable_rcfw:
1476 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1477 free_ring:
1478 	type = bnxt_qplib_get_ring_type(&rdev->chip_ctx);
1479 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, type);
1480 free_rcfw:
1481 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1482 fail:
1483 	if (!locked)
1484 		rtnl_lock();
1485 	bnxt_re_ib_unreg(rdev);
1486 	rtnl_unlock();
1487 
1488 	return rc;
1489 }
1490 
1491 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1492 {
1493 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1494 	struct net_device *netdev = rdev->netdev;
1495 
1496 	bnxt_re_dev_remove(rdev);
1497 
1498 	if (netdev)
1499 		bnxt_re_dev_unprobe(netdev, en_dev);
1500 }
1501 
1502 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1503 {
1504 	struct bnxt_en_dev *en_dev;
1505 	int rc = 0;
1506 
1507 	if (!is_bnxt_re_dev(netdev))
1508 		return -ENODEV;
1509 
1510 	en_dev = bnxt_re_dev_probe(netdev);
1511 	if (IS_ERR(en_dev)) {
1512 		if (en_dev != ERR_PTR(-ENODEV))
1513 			pr_err("%s: Failed to probe\n", ROCE_DRV_MODULE_NAME);
1514 		rc = PTR_ERR(en_dev);
1515 		goto exit;
1516 	}
1517 	*rdev = bnxt_re_dev_add(netdev, en_dev);
1518 	if (!*rdev) {
1519 		rc = -ENOMEM;
1520 		bnxt_re_dev_unprobe(netdev, en_dev);
1521 		goto exit;
1522 	}
1523 exit:
1524 	return rc;
1525 }
1526 
1527 static void bnxt_re_remove_one(struct bnxt_re_dev *rdev)
1528 {
1529 	pci_dev_put(rdev->en_dev->pdev);
1530 }
1531 
1532 /* Handle all deferred netevents tasks */
1533 static void bnxt_re_task(struct work_struct *work)
1534 {
1535 	struct bnxt_re_work *re_work;
1536 	struct bnxt_re_dev *rdev;
1537 	int rc = 0;
1538 
1539 	re_work = container_of(work, struct bnxt_re_work, work);
1540 	rdev = re_work->rdev;
1541 
1542 	if (re_work->event != NETDEV_REGISTER &&
1543 	    !test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1544 		return;
1545 
1546 	switch (re_work->event) {
1547 	case NETDEV_REGISTER:
1548 		rc = bnxt_re_ib_reg(rdev);
1549 		if (rc) {
1550 			dev_err(rdev_to_dev(rdev),
1551 				"Failed to register with IB: %#x", rc);
1552 			bnxt_re_remove_one(rdev);
1553 			bnxt_re_dev_unreg(rdev);
1554 			goto exit;
1555 		}
1556 		break;
1557 	case NETDEV_UP:
1558 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1559 				       IB_EVENT_PORT_ACTIVE);
1560 		break;
1561 	case NETDEV_DOWN:
1562 		bnxt_re_dev_stop(rdev);
1563 		break;
1564 	case NETDEV_CHANGE:
1565 		if (!netif_carrier_ok(rdev->netdev))
1566 			bnxt_re_dev_stop(rdev);
1567 		else if (netif_carrier_ok(rdev->netdev))
1568 			bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1569 					       IB_EVENT_PORT_ACTIVE);
1570 		ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1571 				 &rdev->active_width);
1572 		break;
1573 	default:
1574 		break;
1575 	}
1576 	smp_mb__before_atomic();
1577 	atomic_dec(&rdev->sched_count);
1578 exit:
1579 	kfree(re_work);
1580 }
1581 
1582 static void bnxt_re_init_one(struct bnxt_re_dev *rdev)
1583 {
1584 	pci_dev_get(rdev->en_dev->pdev);
1585 }
1586 
1587 /*
1588  * "Notifier chain callback can be invoked for the same chain from
1589  * different CPUs at the same time".
1590  *
1591  * For cases when the netdev is already present, our call to the
1592  * register_netdevice_notifier() will actually get the rtnl_lock()
1593  * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1594  * events.
1595  *
1596  * But for cases when the netdev is not already present, the notifier
1597  * chain is subjected to be invoked from different CPUs simultaneously.
1598  *
1599  * This is protected by the netdev_mutex.
1600  */
1601 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1602 				unsigned long event, void *ptr)
1603 {
1604 	struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1605 	struct bnxt_re_work *re_work;
1606 	struct bnxt_re_dev *rdev;
1607 	int rc = 0;
1608 	bool sch_work = false;
1609 
1610 	real_dev = rdma_vlan_dev_real_dev(netdev);
1611 	if (!real_dev)
1612 		real_dev = netdev;
1613 
1614 	rdev = bnxt_re_from_netdev(real_dev);
1615 	if (!rdev && event != NETDEV_REGISTER)
1616 		goto exit;
1617 	if (real_dev != netdev)
1618 		goto exit;
1619 
1620 	switch (event) {
1621 	case NETDEV_REGISTER:
1622 		if (rdev)
1623 			break;
1624 		rc = bnxt_re_dev_reg(&rdev, real_dev);
1625 		if (rc == -ENODEV)
1626 			break;
1627 		if (rc) {
1628 			pr_err("Failed to register with the device %s: %#x\n",
1629 			       real_dev->name, rc);
1630 			break;
1631 		}
1632 		bnxt_re_init_one(rdev);
1633 		sch_work = true;
1634 		break;
1635 
1636 	case NETDEV_UNREGISTER:
1637 		/* netdev notifier will call NETDEV_UNREGISTER again later since
1638 		 * we are still holding the reference to the netdev
1639 		 */
1640 		if (atomic_read(&rdev->sched_count) > 0)
1641 			goto exit;
1642 		bnxt_re_ib_unreg(rdev);
1643 		bnxt_re_remove_one(rdev);
1644 		bnxt_re_dev_unreg(rdev);
1645 		break;
1646 
1647 	default:
1648 		sch_work = true;
1649 		break;
1650 	}
1651 	if (sch_work) {
1652 		/* Allocate for the deferred task */
1653 		re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1654 		if (re_work) {
1655 			re_work->rdev = rdev;
1656 			re_work->event = event;
1657 			re_work->vlan_dev = (real_dev == netdev ?
1658 					     NULL : netdev);
1659 			INIT_WORK(&re_work->work, bnxt_re_task);
1660 			atomic_inc(&rdev->sched_count);
1661 			queue_work(bnxt_re_wq, &re_work->work);
1662 		}
1663 	}
1664 
1665 exit:
1666 	return NOTIFY_DONE;
1667 }
1668 
1669 static struct notifier_block bnxt_re_netdev_notifier = {
1670 	.notifier_call = bnxt_re_netdev_event
1671 };
1672 
1673 static int __init bnxt_re_mod_init(void)
1674 {
1675 	int rc = 0;
1676 
1677 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1678 
1679 	bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1680 	if (!bnxt_re_wq)
1681 		return -ENOMEM;
1682 
1683 	INIT_LIST_HEAD(&bnxt_re_dev_list);
1684 
1685 	rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1686 	if (rc) {
1687 		pr_err("%s: Cannot register to netdevice_notifier",
1688 		       ROCE_DRV_MODULE_NAME);
1689 		goto err_netdev;
1690 	}
1691 	return 0;
1692 
1693 err_netdev:
1694 	destroy_workqueue(bnxt_re_wq);
1695 
1696 	return rc;
1697 }
1698 
1699 static void __exit bnxt_re_mod_exit(void)
1700 {
1701 	struct bnxt_re_dev *rdev, *next;
1702 	LIST_HEAD(to_be_deleted);
1703 
1704 	mutex_lock(&bnxt_re_dev_lock);
1705 	/* Free all adapter allocated resources */
1706 	if (!list_empty(&bnxt_re_dev_list))
1707 		list_splice_init(&bnxt_re_dev_list, &to_be_deleted);
1708 	mutex_unlock(&bnxt_re_dev_lock);
1709        /*
1710 	* Cleanup the devices in reverse order so that the VF device
1711 	* cleanup is done before PF cleanup
1712 	*/
1713 	list_for_each_entry_safe_reverse(rdev, next, &to_be_deleted, list) {
1714 		dev_info(rdev_to_dev(rdev), "Unregistering Device");
1715 		/*
1716 		 * Flush out any scheduled tasks before destroying the
1717 		 * resources
1718 		 */
1719 		flush_workqueue(bnxt_re_wq);
1720 		bnxt_re_dev_stop(rdev);
1721 		/* Acquire the rtnl_lock as the L2 resources are freed here */
1722 		rtnl_lock();
1723 		bnxt_re_ib_unreg(rdev);
1724 		rtnl_unlock();
1725 		bnxt_re_remove_one(rdev);
1726 		bnxt_re_dev_unreg(rdev);
1727 	}
1728 	unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1729 	if (bnxt_re_wq)
1730 		destroy_workqueue(bnxt_re_wq);
1731 }
1732 
1733 module_init(bnxt_re_mod_init);
1734 module_exit(bnxt_re_mod_exit);
1735