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_remove_device(struct bnxt_re_dev *rdev);
82 static void bnxt_re_dealloc_driver(struct ib_device *ib_dev);
83 static void bnxt_re_stop_irq(void *handle);
84 
85 static void bnxt_re_set_drv_mode(struct bnxt_re_dev *rdev, u8 mode)
86 {
87 	struct bnxt_qplib_chip_ctx *cctx;
88 
89 	cctx = rdev->chip_ctx;
90 	cctx->modes.wqe_mode = bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx) ?
91 			       mode : BNXT_QPLIB_WQE_MODE_STATIC;
92 }
93 
94 static void bnxt_re_destroy_chip_ctx(struct bnxt_re_dev *rdev)
95 {
96 	struct bnxt_qplib_chip_ctx *chip_ctx;
97 
98 	if (!rdev->chip_ctx)
99 		return;
100 	chip_ctx = rdev->chip_ctx;
101 	rdev->chip_ctx = NULL;
102 	rdev->rcfw.res = NULL;
103 	rdev->qplib_res.cctx = NULL;
104 	rdev->qplib_res.pdev = NULL;
105 	rdev->qplib_res.netdev = NULL;
106 	kfree(chip_ctx);
107 }
108 
109 static int bnxt_re_setup_chip_ctx(struct bnxt_re_dev *rdev, u8 wqe_mode)
110 {
111 	struct bnxt_qplib_chip_ctx *chip_ctx;
112 	struct bnxt_en_dev *en_dev;
113 	struct bnxt *bp;
114 
115 	en_dev = rdev->en_dev;
116 	bp = netdev_priv(en_dev->net);
117 
118 	chip_ctx = kzalloc(sizeof(*chip_ctx), GFP_KERNEL);
119 	if (!chip_ctx)
120 		return -ENOMEM;
121 	chip_ctx->chip_num = bp->chip_num;
122 
123 	rdev->chip_ctx = chip_ctx;
124 	/* rest members to follow eventually */
125 
126 	rdev->qplib_res.cctx = rdev->chip_ctx;
127 	rdev->rcfw.res = &rdev->qplib_res;
128 
129 	bnxt_re_set_drv_mode(rdev, wqe_mode);
130 	return 0;
131 }
132 
133 /* SR-IOV helper functions */
134 
135 static void bnxt_re_get_sriov_func_type(struct bnxt_re_dev *rdev)
136 {
137 	struct bnxt *bp;
138 
139 	bp = netdev_priv(rdev->en_dev->net);
140 	if (BNXT_VF(bp))
141 		rdev->is_virtfn = 1;
142 }
143 
144 /* Set the maximum number of each resource that the driver actually wants
145  * to allocate. This may be up to the maximum number the firmware has
146  * reserved for the function. The driver may choose to allocate fewer
147  * resources than the firmware maximum.
148  */
149 static void bnxt_re_limit_pf_res(struct bnxt_re_dev *rdev)
150 {
151 	struct bnxt_qplib_dev_attr *attr;
152 	struct bnxt_qplib_ctx *ctx;
153 	int i;
154 
155 	attr = &rdev->dev_attr;
156 	ctx = &rdev->qplib_ctx;
157 
158 	ctx->qpc_count = min_t(u32, BNXT_RE_MAX_QPC_COUNT,
159 			       attr->max_qp);
160 	ctx->mrw_count = BNXT_RE_MAX_MRW_COUNT_256K;
161 	/* Use max_mr from fw since max_mrw does not get set */
162 	ctx->mrw_count = min_t(u32, ctx->mrw_count, attr->max_mr);
163 	ctx->srqc_count = min_t(u32, BNXT_RE_MAX_SRQC_COUNT,
164 				attr->max_srq);
165 	ctx->cq_count = min_t(u32, BNXT_RE_MAX_CQ_COUNT, attr->max_cq);
166 	if (!bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx))
167 		for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
168 			rdev->qplib_ctx.tqm_ctx.qcount[i] =
169 			rdev->dev_attr.tqm_alloc_reqs[i];
170 }
171 
172 static void bnxt_re_limit_vf_res(struct bnxt_qplib_ctx *qplib_ctx, u32 num_vf)
173 {
174 	struct bnxt_qplib_vf_res *vf_res;
175 	u32 mrws = 0;
176 	u32 vf_pct;
177 	u32 nvfs;
178 
179 	vf_res = &qplib_ctx->vf_res;
180 	/*
181 	 * Reserve a set of resources for the PF. Divide the remaining
182 	 * resources among the VFs
183 	 */
184 	vf_pct = 100 - BNXT_RE_PCT_RSVD_FOR_PF;
185 	nvfs = num_vf;
186 	num_vf = 100 * num_vf;
187 	vf_res->max_qp_per_vf = (qplib_ctx->qpc_count * vf_pct) / num_vf;
188 	vf_res->max_srq_per_vf = (qplib_ctx->srqc_count * vf_pct) / num_vf;
189 	vf_res->max_cq_per_vf = (qplib_ctx->cq_count * vf_pct) / num_vf;
190 	/*
191 	 * The driver allows many more MRs than other resources. If the
192 	 * firmware does also, then reserve a fixed amount for the PF and
193 	 * divide the rest among VFs. VFs may use many MRs for NFS
194 	 * mounts, ISER, NVME applications, etc. If the firmware severely
195 	 * restricts the number of MRs, then let PF have half and divide
196 	 * the rest among VFs, as for the other resource types.
197 	 */
198 	if (qplib_ctx->mrw_count < BNXT_RE_MAX_MRW_COUNT_64K) {
199 		mrws = qplib_ctx->mrw_count * vf_pct;
200 		nvfs = num_vf;
201 	} else {
202 		mrws = qplib_ctx->mrw_count - BNXT_RE_RESVD_MR_FOR_PF;
203 	}
204 	vf_res->max_mrw_per_vf = (mrws / nvfs);
205 	vf_res->max_gid_per_vf = BNXT_RE_MAX_GID_PER_VF;
206 }
207 
208 static void bnxt_re_set_resource_limits(struct bnxt_re_dev *rdev)
209 {
210 	u32 num_vfs;
211 
212 	memset(&rdev->qplib_ctx.vf_res, 0, sizeof(struct bnxt_qplib_vf_res));
213 	bnxt_re_limit_pf_res(rdev);
214 
215 	num_vfs =  bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx) ?
216 			BNXT_RE_GEN_P5_MAX_VF : rdev->num_vfs;
217 	if (num_vfs)
218 		bnxt_re_limit_vf_res(&rdev->qplib_ctx, num_vfs);
219 }
220 
221 /* for handling bnxt_en callbacks later */
222 static void bnxt_re_stop(void *p)
223 {
224 }
225 
226 static void bnxt_re_start(void *p)
227 {
228 }
229 
230 static void bnxt_re_sriov_config(void *p, int num_vfs)
231 {
232 	struct bnxt_re_dev *rdev = p;
233 
234 	if (!rdev)
235 		return;
236 
237 	rdev->num_vfs = num_vfs;
238 	if (!bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx)) {
239 		bnxt_re_set_resource_limits(rdev);
240 		bnxt_qplib_set_func_resources(&rdev->qplib_res, &rdev->rcfw,
241 					      &rdev->qplib_ctx);
242 	}
243 }
244 
245 static void bnxt_re_shutdown(void *p)
246 {
247 	struct bnxt_re_dev *rdev = p;
248 
249 	if (!rdev)
250 		return;
251 	ASSERT_RTNL();
252 	/* Release the MSIx vectors before queuing unregister */
253 	bnxt_re_stop_irq(rdev);
254 	ib_unregister_device_queued(&rdev->ibdev);
255 }
256 
257 static void bnxt_re_stop_irq(void *handle)
258 {
259 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
260 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
261 	struct bnxt_qplib_nq *nq;
262 	int indx;
263 
264 	for (indx = BNXT_RE_NQ_IDX; indx < rdev->num_msix; indx++) {
265 		nq = &rdev->nq[indx - 1];
266 		bnxt_qplib_nq_stop_irq(nq, false);
267 	}
268 
269 	bnxt_qplib_rcfw_stop_irq(rcfw, false);
270 }
271 
272 static void bnxt_re_start_irq(void *handle, struct bnxt_msix_entry *ent)
273 {
274 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
275 	struct bnxt_msix_entry *msix_ent = rdev->msix_entries;
276 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
277 	struct bnxt_qplib_nq *nq;
278 	int indx, rc;
279 
280 	if (!ent) {
281 		/* Not setting the f/w timeout bit in rcfw.
282 		 * During the driver unload the first command
283 		 * to f/w will timeout and that will set the
284 		 * timeout bit.
285 		 */
286 		ibdev_err(&rdev->ibdev, "Failed to re-start IRQs\n");
287 		return;
288 	}
289 
290 	/* Vectors may change after restart, so update with new vectors
291 	 * in device sctructure.
292 	 */
293 	for (indx = 0; indx < rdev->num_msix; indx++)
294 		rdev->msix_entries[indx].vector = ent[indx].vector;
295 
296 	bnxt_qplib_rcfw_start_irq(rcfw, msix_ent[BNXT_RE_AEQ_IDX].vector,
297 				  false);
298 	for (indx = BNXT_RE_NQ_IDX ; indx < rdev->num_msix; indx++) {
299 		nq = &rdev->nq[indx - 1];
300 		rc = bnxt_qplib_nq_start_irq(nq, indx - 1,
301 					     msix_ent[indx].vector, false);
302 		if (rc)
303 			ibdev_warn(&rdev->ibdev, "Failed to reinit NQ index %d\n",
304 				   indx - 1);
305 	}
306 }
307 
308 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
309 	.ulp_async_notifier = NULL,
310 	.ulp_stop = bnxt_re_stop,
311 	.ulp_start = bnxt_re_start,
312 	.ulp_sriov_config = bnxt_re_sriov_config,
313 	.ulp_shutdown = bnxt_re_shutdown,
314 	.ulp_irq_stop = bnxt_re_stop_irq,
315 	.ulp_irq_restart = bnxt_re_start_irq
316 };
317 
318 /* RoCE -> Net driver */
319 
320 /* Driver registration routines used to let the networking driver (bnxt_en)
321  * to know that the RoCE driver is now installed
322  */
323 static int bnxt_re_unregister_netdev(struct bnxt_re_dev *rdev)
324 {
325 	struct bnxt_en_dev *en_dev;
326 	int rc;
327 
328 	if (!rdev)
329 		return -EINVAL;
330 
331 	en_dev = rdev->en_dev;
332 
333 	rc = en_dev->en_ops->bnxt_unregister_device(rdev->en_dev,
334 						    BNXT_ROCE_ULP);
335 	return rc;
336 }
337 
338 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
339 {
340 	struct bnxt_en_dev *en_dev;
341 	int rc = 0;
342 
343 	if (!rdev)
344 		return -EINVAL;
345 
346 	en_dev = rdev->en_dev;
347 
348 	rc = en_dev->en_ops->bnxt_register_device(en_dev, BNXT_ROCE_ULP,
349 						  &bnxt_re_ulp_ops, rdev);
350 	rdev->qplib_res.pdev = rdev->en_dev->pdev;
351 	return rc;
352 }
353 
354 static int bnxt_re_free_msix(struct bnxt_re_dev *rdev)
355 {
356 	struct bnxt_en_dev *en_dev;
357 	int rc;
358 
359 	if (!rdev)
360 		return -EINVAL;
361 
362 	en_dev = rdev->en_dev;
363 
364 
365 	rc = en_dev->en_ops->bnxt_free_msix(rdev->en_dev, BNXT_ROCE_ULP);
366 
367 	return rc;
368 }
369 
370 static int bnxt_re_request_msix(struct bnxt_re_dev *rdev)
371 {
372 	int rc = 0, num_msix_want = BNXT_RE_MAX_MSIX, num_msix_got;
373 	struct bnxt_en_dev *en_dev;
374 
375 	if (!rdev)
376 		return -EINVAL;
377 
378 	en_dev = rdev->en_dev;
379 
380 	num_msix_want = min_t(u32, BNXT_RE_MAX_MSIX, num_online_cpus());
381 
382 	num_msix_got = en_dev->en_ops->bnxt_request_msix(en_dev, BNXT_ROCE_ULP,
383 							 rdev->msix_entries,
384 							 num_msix_want);
385 	if (num_msix_got < BNXT_RE_MIN_MSIX) {
386 		rc = -EINVAL;
387 		goto done;
388 	}
389 	if (num_msix_got != num_msix_want) {
390 		ibdev_warn(&rdev->ibdev,
391 			   "Requested %d MSI-X vectors, got %d\n",
392 			   num_msix_want, num_msix_got);
393 	}
394 	rdev->num_msix = num_msix_got;
395 done:
396 	return rc;
397 }
398 
399 static void bnxt_re_init_hwrm_hdr(struct bnxt_re_dev *rdev, struct input *hdr,
400 				  u16 opcd, u16 crid, u16 trid)
401 {
402 	hdr->req_type = cpu_to_le16(opcd);
403 	hdr->cmpl_ring = cpu_to_le16(crid);
404 	hdr->target_id = cpu_to_le16(trid);
405 }
406 
407 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
408 				int msg_len, void *resp, int resp_max_len,
409 				int timeout)
410 {
411 	fw_msg->msg = msg;
412 	fw_msg->msg_len = msg_len;
413 	fw_msg->resp = resp;
414 	fw_msg->resp_max_len = resp_max_len;
415 	fw_msg->timeout = timeout;
416 }
417 
418 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev,
419 				 u16 fw_ring_id, int type)
420 {
421 	struct bnxt_en_dev *en_dev = rdev->en_dev;
422 	struct hwrm_ring_free_input req = {0};
423 	struct hwrm_ring_free_output resp;
424 	struct bnxt_fw_msg fw_msg;
425 	int rc = -EINVAL;
426 
427 	if (!en_dev)
428 		return rc;
429 
430 	memset(&fw_msg, 0, sizeof(fw_msg));
431 
432 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1);
433 	req.ring_type = type;
434 	req.ring_id = cpu_to_le16(fw_ring_id);
435 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
436 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
437 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
438 	if (rc)
439 		ibdev_err(&rdev->ibdev, "Failed to free HW ring:%d :%#x",
440 			  req.ring_id, rc);
441 	return rc;
442 }
443 
444 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev,
445 				  struct bnxt_re_ring_attr *ring_attr,
446 				  u16 *fw_ring_id)
447 {
448 	struct bnxt_en_dev *en_dev = rdev->en_dev;
449 	struct hwrm_ring_alloc_input req = {0};
450 	struct hwrm_ring_alloc_output resp;
451 	struct bnxt_fw_msg fw_msg;
452 	int rc = -EINVAL;
453 
454 	if (!en_dev)
455 		return rc;
456 
457 	memset(&fw_msg, 0, sizeof(fw_msg));
458 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_ALLOC, -1, -1);
459 	req.enables = 0;
460 	req.page_tbl_addr =  cpu_to_le64(ring_attr->dma_arr[0]);
461 	if (ring_attr->pages > 1) {
462 		/* Page size is in log2 units */
463 		req.page_size = BNXT_PAGE_SHIFT;
464 		req.page_tbl_depth = 1;
465 	}
466 	req.fbo = 0;
467 	/* Association of ring index with doorbell index and MSIX number */
468 	req.logical_id = cpu_to_le16(ring_attr->lrid);
469 	req.length = cpu_to_le32(ring_attr->depth + 1);
470 	req.ring_type = ring_attr->type;
471 	req.int_mode = ring_attr->mode;
472 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
473 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
474 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
475 	if (!rc)
476 		*fw_ring_id = le16_to_cpu(resp.ring_id);
477 
478 	return rc;
479 }
480 
481 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
482 				      u32 fw_stats_ctx_id)
483 {
484 	struct bnxt_en_dev *en_dev = rdev->en_dev;
485 	struct hwrm_stat_ctx_free_input req = {0};
486 	struct bnxt_fw_msg fw_msg;
487 	int rc = -EINVAL;
488 
489 	if (!en_dev)
490 		return rc;
491 
492 	memset(&fw_msg, 0, sizeof(fw_msg));
493 
494 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_FREE, -1, -1);
495 	req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
496 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&req,
497 			    sizeof(req), DFLT_HWRM_CMD_TIMEOUT);
498 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
499 	if (rc)
500 		ibdev_err(&rdev->ibdev, "Failed to free HW stats context %#x",
501 			  rc);
502 
503 	return rc;
504 }
505 
506 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
507 				       dma_addr_t dma_map,
508 				       u32 *fw_stats_ctx_id)
509 {
510 	struct hwrm_stat_ctx_alloc_output resp = {0};
511 	struct hwrm_stat_ctx_alloc_input req = {0};
512 	struct bnxt_en_dev *en_dev = rdev->en_dev;
513 	struct bnxt_fw_msg fw_msg;
514 	int rc = -EINVAL;
515 
516 	*fw_stats_ctx_id = INVALID_STATS_CTX_ID;
517 
518 	if (!en_dev)
519 		return rc;
520 
521 	memset(&fw_msg, 0, sizeof(fw_msg));
522 
523 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1);
524 	req.update_period_ms = cpu_to_le32(1000);
525 	req.stats_dma_addr = cpu_to_le64(dma_map);
526 	req.stats_dma_length = cpu_to_le16(sizeof(struct ctx_hw_stats_ext));
527 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
528 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
529 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
530 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
531 	if (!rc)
532 		*fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
533 
534 	return rc;
535 }
536 
537 /* Device */
538 
539 static bool is_bnxt_re_dev(struct net_device *netdev)
540 {
541 	struct ethtool_drvinfo drvinfo;
542 
543 	if (netdev->ethtool_ops && netdev->ethtool_ops->get_drvinfo) {
544 		memset(&drvinfo, 0, sizeof(drvinfo));
545 		netdev->ethtool_ops->get_drvinfo(netdev, &drvinfo);
546 
547 		if (strcmp(drvinfo.driver, "bnxt_en"))
548 			return false;
549 		return true;
550 	}
551 	return false;
552 }
553 
554 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
555 {
556 	struct ib_device *ibdev =
557 		ib_device_get_by_netdev(netdev, RDMA_DRIVER_BNXT_RE);
558 	if (!ibdev)
559 		return NULL;
560 
561 	return container_of(ibdev, struct bnxt_re_dev, ibdev);
562 }
563 
564 static void bnxt_re_dev_unprobe(struct net_device *netdev,
565 				struct bnxt_en_dev *en_dev)
566 {
567 	dev_put(netdev);
568 	module_put(en_dev->pdev->driver->driver.owner);
569 }
570 
571 static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev)
572 {
573 	struct bnxt *bp = netdev_priv(netdev);
574 	struct bnxt_en_dev *en_dev;
575 	struct pci_dev *pdev;
576 
577 	/* Call bnxt_en's RoCE probe via indirect API */
578 	if (!bp->ulp_probe)
579 		return ERR_PTR(-EINVAL);
580 
581 	en_dev = bp->ulp_probe(netdev);
582 	if (IS_ERR(en_dev))
583 		return en_dev;
584 
585 	pdev = en_dev->pdev;
586 	if (!pdev)
587 		return ERR_PTR(-EINVAL);
588 
589 	if (!(en_dev->flags & BNXT_EN_FLAG_ROCE_CAP)) {
590 		dev_info(&pdev->dev,
591 			"%s: probe error: RoCE is not supported on this device",
592 			ROCE_DRV_MODULE_NAME);
593 		return ERR_PTR(-ENODEV);
594 	}
595 
596 	/* Bump net device reference count */
597 	if (!try_module_get(pdev->driver->driver.owner))
598 		return ERR_PTR(-ENODEV);
599 
600 	dev_hold(netdev);
601 
602 	return en_dev;
603 }
604 
605 static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr,
606 			   char *buf)
607 {
608 	struct bnxt_re_dev *rdev =
609 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
610 
611 	return sysfs_emit(buf, "0x%x\n", rdev->en_dev->pdev->vendor);
612 }
613 static DEVICE_ATTR_RO(hw_rev);
614 
615 static ssize_t hca_type_show(struct device *device,
616 			     struct device_attribute *attr, char *buf)
617 {
618 	struct bnxt_re_dev *rdev =
619 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
620 
621 	return sysfs_emit(buf, "%s\n", rdev->ibdev.node_desc);
622 }
623 static DEVICE_ATTR_RO(hca_type);
624 
625 static struct attribute *bnxt_re_attributes[] = {
626 	&dev_attr_hw_rev.attr,
627 	&dev_attr_hca_type.attr,
628 	NULL
629 };
630 
631 static const struct attribute_group bnxt_re_dev_attr_group = {
632 	.attrs = bnxt_re_attributes,
633 };
634 
635 static const struct ib_device_ops bnxt_re_dev_ops = {
636 	.owner = THIS_MODULE,
637 	.driver_id = RDMA_DRIVER_BNXT_RE,
638 	.uverbs_abi_ver = BNXT_RE_ABI_VERSION,
639 
640 	.add_gid = bnxt_re_add_gid,
641 	.alloc_hw_stats = bnxt_re_ib_alloc_hw_stats,
642 	.alloc_mr = bnxt_re_alloc_mr,
643 	.alloc_pd = bnxt_re_alloc_pd,
644 	.alloc_ucontext = bnxt_re_alloc_ucontext,
645 	.create_ah = bnxt_re_create_ah,
646 	.create_cq = bnxt_re_create_cq,
647 	.create_qp = bnxt_re_create_qp,
648 	.create_srq = bnxt_re_create_srq,
649 	.create_user_ah = bnxt_re_create_ah,
650 	.dealloc_driver = bnxt_re_dealloc_driver,
651 	.dealloc_pd = bnxt_re_dealloc_pd,
652 	.dealloc_ucontext = bnxt_re_dealloc_ucontext,
653 	.del_gid = bnxt_re_del_gid,
654 	.dereg_mr = bnxt_re_dereg_mr,
655 	.destroy_ah = bnxt_re_destroy_ah,
656 	.destroy_cq = bnxt_re_destroy_cq,
657 	.destroy_qp = bnxt_re_destroy_qp,
658 	.destroy_srq = bnxt_re_destroy_srq,
659 	.get_dev_fw_str = bnxt_re_query_fw_str,
660 	.get_dma_mr = bnxt_re_get_dma_mr,
661 	.get_hw_stats = bnxt_re_ib_get_hw_stats,
662 	.get_link_layer = bnxt_re_get_link_layer,
663 	.get_port_immutable = bnxt_re_get_port_immutable,
664 	.map_mr_sg = bnxt_re_map_mr_sg,
665 	.mmap = bnxt_re_mmap,
666 	.modify_ah = bnxt_re_modify_ah,
667 	.modify_qp = bnxt_re_modify_qp,
668 	.modify_srq = bnxt_re_modify_srq,
669 	.poll_cq = bnxt_re_poll_cq,
670 	.post_recv = bnxt_re_post_recv,
671 	.post_send = bnxt_re_post_send,
672 	.post_srq_recv = bnxt_re_post_srq_recv,
673 	.query_ah = bnxt_re_query_ah,
674 	.query_device = bnxt_re_query_device,
675 	.query_pkey = bnxt_re_query_pkey,
676 	.query_port = bnxt_re_query_port,
677 	.query_qp = bnxt_re_query_qp,
678 	.query_srq = bnxt_re_query_srq,
679 	.reg_user_mr = bnxt_re_reg_user_mr,
680 	.req_notify_cq = bnxt_re_req_notify_cq,
681 	INIT_RDMA_OBJ_SIZE(ib_ah, bnxt_re_ah, ib_ah),
682 	INIT_RDMA_OBJ_SIZE(ib_cq, bnxt_re_cq, ib_cq),
683 	INIT_RDMA_OBJ_SIZE(ib_pd, bnxt_re_pd, ib_pd),
684 	INIT_RDMA_OBJ_SIZE(ib_srq, bnxt_re_srq, ib_srq),
685 	INIT_RDMA_OBJ_SIZE(ib_ucontext, bnxt_re_ucontext, ib_uctx),
686 };
687 
688 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
689 {
690 	struct ib_device *ibdev = &rdev->ibdev;
691 	int ret;
692 
693 	/* ib device init */
694 	ibdev->node_type = RDMA_NODE_IB_CA;
695 	strlcpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
696 		strlen(BNXT_RE_DESC) + 5);
697 	ibdev->phys_port_cnt = 1;
698 
699 	bnxt_qplib_get_guid(rdev->netdev->dev_addr, (u8 *)&ibdev->node_guid);
700 
701 	ibdev->num_comp_vectors	= rdev->num_msix - 1;
702 	ibdev->dev.parent = &rdev->en_dev->pdev->dev;
703 	ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
704 
705 	rdma_set_device_sysfs_group(ibdev, &bnxt_re_dev_attr_group);
706 	ib_set_device_ops(ibdev, &bnxt_re_dev_ops);
707 	ret = ib_device_set_netdev(&rdev->ibdev, rdev->netdev, 1);
708 	if (ret)
709 		return ret;
710 
711 	dma_set_max_seg_size(&rdev->en_dev->pdev->dev, UINT_MAX);
712 	return ib_register_device(ibdev, "bnxt_re%d", &rdev->en_dev->pdev->dev);
713 }
714 
715 static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev)
716 {
717 	dev_put(rdev->netdev);
718 	rdev->netdev = NULL;
719 	mutex_lock(&bnxt_re_dev_lock);
720 	list_del_rcu(&rdev->list);
721 	mutex_unlock(&bnxt_re_dev_lock);
722 
723 	synchronize_rcu();
724 }
725 
726 static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev,
727 					   struct bnxt_en_dev *en_dev)
728 {
729 	struct bnxt_re_dev *rdev;
730 
731 	/* Allocate bnxt_re_dev instance here */
732 	rdev = ib_alloc_device(bnxt_re_dev, ibdev);
733 	if (!rdev) {
734 		ibdev_err(NULL, "%s: bnxt_re_dev allocation failure!",
735 			  ROCE_DRV_MODULE_NAME);
736 		return NULL;
737 	}
738 	/* Default values */
739 	rdev->netdev = netdev;
740 	dev_hold(rdev->netdev);
741 	rdev->en_dev = en_dev;
742 	rdev->id = rdev->en_dev->pdev->devfn;
743 	INIT_LIST_HEAD(&rdev->qp_list);
744 	mutex_init(&rdev->qp_lock);
745 	atomic_set(&rdev->qp_count, 0);
746 	atomic_set(&rdev->cq_count, 0);
747 	atomic_set(&rdev->srq_count, 0);
748 	atomic_set(&rdev->mr_count, 0);
749 	atomic_set(&rdev->mw_count, 0);
750 	rdev->cosq[0] = 0xFFFF;
751 	rdev->cosq[1] = 0xFFFF;
752 
753 	mutex_lock(&bnxt_re_dev_lock);
754 	list_add_tail_rcu(&rdev->list, &bnxt_re_dev_list);
755 	mutex_unlock(&bnxt_re_dev_lock);
756 	return rdev;
757 }
758 
759 static int bnxt_re_handle_unaffi_async_event(struct creq_func_event
760 					     *unaffi_async)
761 {
762 	switch (unaffi_async->event) {
763 	case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
764 		break;
765 	case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
766 		break;
767 	case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
768 		break;
769 	case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
770 		break;
771 	case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
772 		break;
773 	case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
774 		break;
775 	case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
776 		break;
777 	case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
778 		break;
779 	case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
780 		break;
781 	case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
782 		break;
783 	case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
784 		break;
785 	default:
786 		return -EINVAL;
787 	}
788 	return 0;
789 }
790 
791 static int bnxt_re_handle_qp_async_event(struct creq_qp_event *qp_event,
792 					 struct bnxt_re_qp *qp)
793 {
794 	struct ib_event event;
795 	unsigned int flags;
796 
797 	if (qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_ERR &&
798 	    rdma_is_kernel_res(&qp->ib_qp.res)) {
799 		flags = bnxt_re_lock_cqs(qp);
800 		bnxt_qplib_add_flush_qp(&qp->qplib_qp);
801 		bnxt_re_unlock_cqs(qp, flags);
802 	}
803 
804 	memset(&event, 0, sizeof(event));
805 	if (qp->qplib_qp.srq) {
806 		event.device = &qp->rdev->ibdev;
807 		event.element.qp = &qp->ib_qp;
808 		event.event = IB_EVENT_QP_LAST_WQE_REACHED;
809 	}
810 
811 	if (event.device && qp->ib_qp.event_handler)
812 		qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
813 
814 	return 0;
815 }
816 
817 static int bnxt_re_handle_affi_async_event(struct creq_qp_event *affi_async,
818 					   void *obj)
819 {
820 	int rc = 0;
821 	u8 event;
822 
823 	if (!obj)
824 		return rc; /* QP was already dead, still return success */
825 
826 	event = affi_async->event;
827 	if (event == CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION) {
828 		struct bnxt_qplib_qp *lib_qp = obj;
829 		struct bnxt_re_qp *qp = container_of(lib_qp, struct bnxt_re_qp,
830 						     qplib_qp);
831 		rc = bnxt_re_handle_qp_async_event(affi_async, qp);
832 	}
833 	return rc;
834 }
835 
836 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
837 			       void *aeqe, void *obj)
838 {
839 	struct creq_qp_event *affi_async;
840 	struct creq_func_event *unaffi_async;
841 	u8 type;
842 	int rc;
843 
844 	type = ((struct creq_base *)aeqe)->type;
845 	if (type == CREQ_BASE_TYPE_FUNC_EVENT) {
846 		unaffi_async = aeqe;
847 		rc = bnxt_re_handle_unaffi_async_event(unaffi_async);
848 	} else {
849 		affi_async = aeqe;
850 		rc = bnxt_re_handle_affi_async_event(affi_async, obj);
851 	}
852 
853 	return rc;
854 }
855 
856 static int bnxt_re_srqn_handler(struct bnxt_qplib_nq *nq,
857 				struct bnxt_qplib_srq *handle, u8 event)
858 {
859 	struct bnxt_re_srq *srq = container_of(handle, struct bnxt_re_srq,
860 					       qplib_srq);
861 	struct ib_event ib_event;
862 	int rc = 0;
863 
864 	if (!srq) {
865 		ibdev_err(NULL, "%s: SRQ is NULL, SRQN not handled",
866 			  ROCE_DRV_MODULE_NAME);
867 		rc = -EINVAL;
868 		goto done;
869 	}
870 	ib_event.device = &srq->rdev->ibdev;
871 	ib_event.element.srq = &srq->ib_srq;
872 	if (event == NQ_SRQ_EVENT_EVENT_SRQ_THRESHOLD_EVENT)
873 		ib_event.event = IB_EVENT_SRQ_LIMIT_REACHED;
874 	else
875 		ib_event.event = IB_EVENT_SRQ_ERR;
876 
877 	if (srq->ib_srq.event_handler) {
878 		/* Lock event_handler? */
879 		(*srq->ib_srq.event_handler)(&ib_event,
880 					     srq->ib_srq.srq_context);
881 	}
882 done:
883 	return rc;
884 }
885 
886 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
887 			       struct bnxt_qplib_cq *handle)
888 {
889 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
890 					     qplib_cq);
891 
892 	if (!cq) {
893 		ibdev_err(NULL, "%s: CQ is NULL, CQN not handled",
894 			  ROCE_DRV_MODULE_NAME);
895 		return -EINVAL;
896 	}
897 	if (cq->ib_cq.comp_handler) {
898 		/* Lock comp_handler? */
899 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
900 	}
901 
902 	return 0;
903 }
904 
905 #define BNXT_RE_GEN_P5_PF_NQ_DB		0x10000
906 #define BNXT_RE_GEN_P5_VF_NQ_DB		0x4000
907 static u32 bnxt_re_get_nqdb_offset(struct bnxt_re_dev *rdev, u16 indx)
908 {
909 	return bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx) ?
910 		(rdev->is_virtfn ? BNXT_RE_GEN_P5_VF_NQ_DB :
911 				   BNXT_RE_GEN_P5_PF_NQ_DB) :
912 				   rdev->msix_entries[indx].db_offset;
913 }
914 
915 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
916 {
917 	int i;
918 
919 	for (i = 1; i < rdev->num_msix; i++)
920 		bnxt_qplib_disable_nq(&rdev->nq[i - 1]);
921 
922 	if (rdev->qplib_res.rcfw)
923 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
924 }
925 
926 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
927 {
928 	int num_vec_enabled = 0;
929 	int rc = 0, i;
930 	u32 db_offt;
931 
932 	bnxt_qplib_init_res(&rdev->qplib_res);
933 
934 	for (i = 1; i < rdev->num_msix ; i++) {
935 		db_offt = bnxt_re_get_nqdb_offset(rdev, i);
936 		rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
937 					  i - 1, rdev->msix_entries[i].vector,
938 					  db_offt, &bnxt_re_cqn_handler,
939 					  &bnxt_re_srqn_handler);
940 		if (rc) {
941 			ibdev_err(&rdev->ibdev,
942 				  "Failed to enable NQ with rc = 0x%x", rc);
943 			goto fail;
944 		}
945 		num_vec_enabled++;
946 	}
947 	return 0;
948 fail:
949 	for (i = num_vec_enabled; i >= 0; i--)
950 		bnxt_qplib_disable_nq(&rdev->nq[i]);
951 	return rc;
952 }
953 
954 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev)
955 {
956 	u8 type;
957 	int i;
958 
959 	for (i = 0; i < rdev->num_msix - 1; i++) {
960 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
961 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
962 		bnxt_qplib_free_nq(&rdev->nq[i]);
963 		rdev->nq[i].res = NULL;
964 	}
965 }
966 
967 static void bnxt_re_free_res(struct bnxt_re_dev *rdev)
968 {
969 	bnxt_re_free_nq_res(rdev);
970 
971 	if (rdev->qplib_res.dpi_tbl.max) {
972 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
973 				       &rdev->qplib_res.dpi_tbl,
974 				       &rdev->dpi_privileged);
975 	}
976 	if (rdev->qplib_res.rcfw) {
977 		bnxt_qplib_free_res(&rdev->qplib_res);
978 		rdev->qplib_res.rcfw = NULL;
979 	}
980 }
981 
982 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
983 {
984 	struct bnxt_re_ring_attr rattr = {};
985 	int num_vec_created = 0;
986 	int rc = 0, i;
987 	u8 type;
988 
989 	/* Configure and allocate resources for qplib */
990 	rdev->qplib_res.rcfw = &rdev->rcfw;
991 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
992 				     rdev->is_virtfn);
993 	if (rc)
994 		goto fail;
995 
996 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
997 				  rdev->netdev, &rdev->dev_attr);
998 	if (rc)
999 		goto fail;
1000 
1001 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
1002 				  &rdev->dpi_privileged,
1003 				  rdev);
1004 	if (rc)
1005 		goto dealloc_res;
1006 
1007 	for (i = 0; i < rdev->num_msix - 1; i++) {
1008 		struct bnxt_qplib_nq *nq;
1009 
1010 		nq = &rdev->nq[i];
1011 		nq->hwq.max_elements = BNXT_QPLIB_NQE_MAX_CNT;
1012 		rc = bnxt_qplib_alloc_nq(&rdev->qplib_res, &rdev->nq[i]);
1013 		if (rc) {
1014 			ibdev_err(&rdev->ibdev, "Alloc Failed NQ%d rc:%#x",
1015 				  i, rc);
1016 			goto free_nq;
1017 		}
1018 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1019 		rattr.dma_arr = nq->hwq.pbl[PBL_LVL_0].pg_map_arr;
1020 		rattr.pages = nq->hwq.pbl[rdev->nq[i].hwq.level].pg_count;
1021 		rattr.type = type;
1022 		rattr.mode = RING_ALLOC_REQ_INT_MODE_MSIX;
1023 		rattr.depth = BNXT_QPLIB_NQE_MAX_CNT - 1;
1024 		rattr.lrid = rdev->msix_entries[i + 1].ring_idx;
1025 		rc = bnxt_re_net_ring_alloc(rdev, &rattr, &nq->ring_id);
1026 		if (rc) {
1027 			ibdev_err(&rdev->ibdev,
1028 				  "Failed to allocate NQ fw id with rc = 0x%x",
1029 				  rc);
1030 			bnxt_qplib_free_nq(&rdev->nq[i]);
1031 			goto free_nq;
1032 		}
1033 		num_vec_created++;
1034 	}
1035 	return 0;
1036 free_nq:
1037 	for (i = num_vec_created - 1; i >= 0; i--) {
1038 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1039 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
1040 		bnxt_qplib_free_nq(&rdev->nq[i]);
1041 	}
1042 	bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
1043 			       &rdev->qplib_res.dpi_tbl,
1044 			       &rdev->dpi_privileged);
1045 dealloc_res:
1046 	bnxt_qplib_free_res(&rdev->qplib_res);
1047 
1048 fail:
1049 	rdev->qplib_res.rcfw = NULL;
1050 	return rc;
1051 }
1052 
1053 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
1054 				   u8 port_num, enum ib_event_type event)
1055 {
1056 	struct ib_event ib_event;
1057 
1058 	ib_event.device = ibdev;
1059 	if (qp) {
1060 		ib_event.element.qp = qp;
1061 		ib_event.event = event;
1062 		if (qp->event_handler)
1063 			qp->event_handler(&ib_event, qp->qp_context);
1064 
1065 	} else {
1066 		ib_event.element.port_num = port_num;
1067 		ib_event.event = event;
1068 		ib_dispatch_event(&ib_event);
1069 	}
1070 }
1071 
1072 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN      0x02
1073 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
1074 				      u64 *cid_map)
1075 {
1076 	struct hwrm_queue_pri2cos_qcfg_input req = {0};
1077 	struct bnxt *bp = netdev_priv(rdev->netdev);
1078 	struct hwrm_queue_pri2cos_qcfg_output resp;
1079 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1080 	struct bnxt_fw_msg fw_msg;
1081 	u32 flags = 0;
1082 	u8 *qcfgmap, *tmp_map;
1083 	int rc = 0, i;
1084 
1085 	if (!cid_map)
1086 		return -EINVAL;
1087 
1088 	memset(&fw_msg, 0, sizeof(fw_msg));
1089 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1090 			      HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
1091 	flags |= (dir & 0x01);
1092 	flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
1093 	req.flags = cpu_to_le32(flags);
1094 	req.port_id = bp->pf.port_id;
1095 
1096 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1097 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1098 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1099 	if (rc)
1100 		return rc;
1101 
1102 	if (resp.queue_cfg_info) {
1103 		ibdev_warn(&rdev->ibdev,
1104 			   "Asymmetric cos queue configuration detected");
1105 		ibdev_warn(&rdev->ibdev,
1106 			   " on device, QoS may not be fully functional\n");
1107 	}
1108 	qcfgmap = &resp.pri0_cos_queue_id;
1109 	tmp_map = (u8 *)cid_map;
1110 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
1111 		tmp_map[i] = qcfgmap[i];
1112 
1113 	return rc;
1114 }
1115 
1116 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
1117 					struct bnxt_re_qp *qp)
1118 {
1119 	return (qp->ib_qp.qp_type == IB_QPT_GSI) ||
1120 	       (qp == rdev->gsi_ctx.gsi_sqp);
1121 }
1122 
1123 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
1124 {
1125 	int mask = IB_QP_STATE;
1126 	struct ib_qp_attr qp_attr;
1127 	struct bnxt_re_qp *qp;
1128 
1129 	qp_attr.qp_state = IB_QPS_ERR;
1130 	mutex_lock(&rdev->qp_lock);
1131 	list_for_each_entry(qp, &rdev->qp_list, list) {
1132 		/* Modify the state of all QPs except QP1/Shadow QP */
1133 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
1134 			if (qp->qplib_qp.state !=
1135 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
1136 			    qp->qplib_qp.state !=
1137 			    CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1138 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
1139 						       1, IB_EVENT_QP_FATAL);
1140 				bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
1141 						  NULL);
1142 			}
1143 		}
1144 	}
1145 	mutex_unlock(&rdev->qp_lock);
1146 }
1147 
1148 static int bnxt_re_update_gid(struct bnxt_re_dev *rdev)
1149 {
1150 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
1151 	struct bnxt_qplib_gid gid;
1152 	u16 gid_idx, index;
1153 	int rc = 0;
1154 
1155 	if (!ib_device_try_get(&rdev->ibdev))
1156 		return 0;
1157 
1158 	if (!sgid_tbl) {
1159 		ibdev_err(&rdev->ibdev, "QPLIB: SGID table not allocated");
1160 		rc = -EINVAL;
1161 		goto out;
1162 	}
1163 
1164 	for (index = 0; index < sgid_tbl->active; index++) {
1165 		gid_idx = sgid_tbl->hw_id[index];
1166 
1167 		if (!memcmp(&sgid_tbl->tbl[index], &bnxt_qplib_gid_zero,
1168 			    sizeof(bnxt_qplib_gid_zero)))
1169 			continue;
1170 		/* need to modify the VLAN enable setting of non VLAN GID only
1171 		 * as setting is done for VLAN GID while adding GID
1172 		 */
1173 		if (sgid_tbl->vlan[index])
1174 			continue;
1175 
1176 		memcpy(&gid, &sgid_tbl->tbl[index], sizeof(gid));
1177 
1178 		rc = bnxt_qplib_update_sgid(sgid_tbl, &gid, gid_idx,
1179 					    rdev->qplib_res.netdev->dev_addr);
1180 	}
1181 out:
1182 	ib_device_put(&rdev->ibdev);
1183 	return rc;
1184 }
1185 
1186 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
1187 {
1188 	u32 prio_map = 0, tmp_map = 0;
1189 	struct net_device *netdev;
1190 	struct dcb_app app;
1191 
1192 	netdev = rdev->netdev;
1193 
1194 	memset(&app, 0, sizeof(app));
1195 	app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
1196 	app.protocol = ETH_P_IBOE;
1197 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1198 	prio_map = tmp_map;
1199 
1200 	app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
1201 	app.protocol = ROCE_V2_UDP_DPORT;
1202 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1203 	prio_map |= tmp_map;
1204 
1205 	return prio_map;
1206 }
1207 
1208 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
1209 {
1210 	u16 prio;
1211 	u8 id;
1212 
1213 	for (prio = 0, id = 0; prio < 8; prio++) {
1214 		if (prio_map & (1 << prio)) {
1215 			cosq[id] = cid_map[prio];
1216 			id++;
1217 			if (id == 2) /* Max 2 tcs supported */
1218 				break;
1219 		}
1220 	}
1221 }
1222 
1223 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
1224 {
1225 	u8 prio_map = 0;
1226 	u64 cid_map;
1227 	int rc;
1228 
1229 	/* Get priority for roce */
1230 	prio_map = bnxt_re_get_priority_mask(rdev);
1231 
1232 	if (prio_map == rdev->cur_prio_map)
1233 		return 0;
1234 	rdev->cur_prio_map = prio_map;
1235 	/* Get cosq id for this priority */
1236 	rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
1237 	if (rc) {
1238 		ibdev_warn(&rdev->ibdev, "no cos for p_mask %x\n", prio_map);
1239 		return rc;
1240 	}
1241 	/* Parse CoS IDs for app priority */
1242 	bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
1243 
1244 	/* Config BONO. */
1245 	rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
1246 	if (rc) {
1247 		ibdev_warn(&rdev->ibdev, "no tc for cos{%x, %x}\n",
1248 			   rdev->cosq[0], rdev->cosq[1]);
1249 		return rc;
1250 	}
1251 
1252 	/* Actual priorities are not programmed as they are already
1253 	 * done by L2 driver; just enable or disable priority vlan tagging
1254 	 */
1255 	if ((prio_map == 0 && rdev->qplib_res.prio) ||
1256 	    (prio_map != 0 && !rdev->qplib_res.prio)) {
1257 		rdev->qplib_res.prio = prio_map ? true : false;
1258 
1259 		bnxt_re_update_gid(rdev);
1260 	}
1261 
1262 	return 0;
1263 }
1264 
1265 static void bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev)
1266 {
1267 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1268 	struct hwrm_ver_get_output resp = {0};
1269 	struct hwrm_ver_get_input req = {0};
1270 	struct bnxt_fw_msg fw_msg;
1271 	int rc = 0;
1272 
1273 	memset(&fw_msg, 0, sizeof(fw_msg));
1274 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1275 			      HWRM_VER_GET, -1, -1);
1276 	req.hwrm_intf_maj = HWRM_VERSION_MAJOR;
1277 	req.hwrm_intf_min = HWRM_VERSION_MINOR;
1278 	req.hwrm_intf_upd = HWRM_VERSION_UPDATE;
1279 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1280 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1281 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1282 	if (rc) {
1283 		ibdev_err(&rdev->ibdev, "Failed to query HW version, rc = 0x%x",
1284 			  rc);
1285 		return;
1286 	}
1287 	rdev->qplib_ctx.hwrm_intf_ver =
1288 		(u64)le16_to_cpu(resp.hwrm_intf_major) << 48 |
1289 		(u64)le16_to_cpu(resp.hwrm_intf_minor) << 32 |
1290 		(u64)le16_to_cpu(resp.hwrm_intf_build) << 16 |
1291 		le16_to_cpu(resp.hwrm_intf_patch);
1292 }
1293 
1294 static int bnxt_re_ib_init(struct bnxt_re_dev *rdev)
1295 {
1296 	int rc = 0;
1297 	u32 event;
1298 
1299 	/* Register ib dev */
1300 	rc = bnxt_re_register_ib(rdev);
1301 	if (rc) {
1302 		pr_err("Failed to register with IB: %#x\n", rc);
1303 		return rc;
1304 	}
1305 	dev_info(rdev_to_dev(rdev), "Device registered successfully");
1306 	ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1307 			 &rdev->active_width);
1308 	set_bit(BNXT_RE_FLAG_ISSUE_ROCE_STATS, &rdev->flags);
1309 
1310 	event = netif_running(rdev->netdev) && netif_carrier_ok(rdev->netdev) ?
1311 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
1312 
1313 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, event);
1314 
1315 	return rc;
1316 }
1317 
1318 static void bnxt_re_dev_uninit(struct bnxt_re_dev *rdev)
1319 {
1320 	u8 type;
1321 	int rc;
1322 
1323 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
1324 		cancel_delayed_work_sync(&rdev->worker);
1325 
1326 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED,
1327 			       &rdev->flags))
1328 		bnxt_re_cleanup_res(rdev);
1329 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags))
1330 		bnxt_re_free_res(rdev);
1331 
1332 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
1333 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
1334 		if (rc)
1335 			ibdev_warn(&rdev->ibdev,
1336 				   "Failed to deinitialize RCFW: %#x", rc);
1337 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1338 		bnxt_qplib_free_ctx(&rdev->qplib_res, &rdev->qplib_ctx);
1339 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1340 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1341 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id, type);
1342 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1343 	}
1344 	if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
1345 		rc = bnxt_re_free_msix(rdev);
1346 		if (rc)
1347 			ibdev_warn(&rdev->ibdev,
1348 				   "Failed to free MSI-X vectors: %#x", rc);
1349 	}
1350 
1351 	bnxt_re_destroy_chip_ctx(rdev);
1352 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
1353 		rc = bnxt_re_unregister_netdev(rdev);
1354 		if (rc)
1355 			ibdev_warn(&rdev->ibdev,
1356 				   "Failed to unregister with netdev: %#x", rc);
1357 	}
1358 }
1359 
1360 /* worker thread for polling periodic events. Now used for QoS programming*/
1361 static void bnxt_re_worker(struct work_struct *work)
1362 {
1363 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
1364 						worker.work);
1365 
1366 	bnxt_re_setup_qos(rdev);
1367 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1368 }
1369 
1370 static int bnxt_re_dev_init(struct bnxt_re_dev *rdev, u8 wqe_mode)
1371 {
1372 	struct bnxt_qplib_creq_ctx *creq;
1373 	struct bnxt_re_ring_attr rattr;
1374 	u32 db_offt;
1375 	int vid;
1376 	u8 type;
1377 	int rc;
1378 
1379 	/* Registered a new RoCE device instance to netdev */
1380 	memset(&rattr, 0, sizeof(rattr));
1381 	rc = bnxt_re_register_netdev(rdev);
1382 	if (rc) {
1383 		rtnl_unlock();
1384 		ibdev_err(&rdev->ibdev,
1385 			  "Failed to register with netedev: %#x\n", rc);
1386 		return -EINVAL;
1387 	}
1388 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1389 
1390 	rc = bnxt_re_setup_chip_ctx(rdev, wqe_mode);
1391 	if (rc) {
1392 		ibdev_err(&rdev->ibdev, "Failed to get chip context\n");
1393 		return -EINVAL;
1394 	}
1395 
1396 	/* Check whether VF or PF */
1397 	bnxt_re_get_sriov_func_type(rdev);
1398 
1399 	rc = bnxt_re_request_msix(rdev);
1400 	if (rc) {
1401 		ibdev_err(&rdev->ibdev,
1402 			  "Failed to get MSI-X vectors: %#x\n", rc);
1403 		rc = -EINVAL;
1404 		goto fail;
1405 	}
1406 	set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
1407 
1408 	bnxt_re_query_hwrm_intf_version(rdev);
1409 
1410 	/* Establish RCFW Communication Channel to initialize the context
1411 	 * memory for the function and all child VFs
1412 	 */
1413 	rc = bnxt_qplib_alloc_rcfw_channel(&rdev->qplib_res, &rdev->rcfw,
1414 					   &rdev->qplib_ctx,
1415 					   BNXT_RE_MAX_QPC_COUNT);
1416 	if (rc) {
1417 		ibdev_err(&rdev->ibdev,
1418 			  "Failed to allocate RCFW Channel: %#x\n", rc);
1419 		goto fail;
1420 	}
1421 
1422 	type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1423 	creq = &rdev->rcfw.creq;
1424 	rattr.dma_arr = creq->hwq.pbl[PBL_LVL_0].pg_map_arr;
1425 	rattr.pages = creq->hwq.pbl[creq->hwq.level].pg_count;
1426 	rattr.type = type;
1427 	rattr.mode = RING_ALLOC_REQ_INT_MODE_MSIX;
1428 	rattr.depth = BNXT_QPLIB_CREQE_MAX_CNT - 1;
1429 	rattr.lrid = rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx;
1430 	rc = bnxt_re_net_ring_alloc(rdev, &rattr, &creq->ring_id);
1431 	if (rc) {
1432 		ibdev_err(&rdev->ibdev, "Failed to allocate CREQ: %#x\n", rc);
1433 		goto free_rcfw;
1434 	}
1435 	db_offt = bnxt_re_get_nqdb_offset(rdev, BNXT_RE_AEQ_IDX);
1436 	vid = rdev->msix_entries[BNXT_RE_AEQ_IDX].vector;
1437 	rc = bnxt_qplib_enable_rcfw_channel(&rdev->rcfw,
1438 					    vid, db_offt, rdev->is_virtfn,
1439 					    &bnxt_re_aeq_handler);
1440 	if (rc) {
1441 		ibdev_err(&rdev->ibdev, "Failed to enable RCFW channel: %#x\n",
1442 			  rc);
1443 		goto free_ring;
1444 	}
1445 
1446 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
1447 				     rdev->is_virtfn);
1448 	if (rc)
1449 		goto disable_rcfw;
1450 
1451 	bnxt_re_set_resource_limits(rdev);
1452 
1453 	rc = bnxt_qplib_alloc_ctx(&rdev->qplib_res, &rdev->qplib_ctx, 0,
1454 				  bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx));
1455 	if (rc) {
1456 		ibdev_err(&rdev->ibdev,
1457 			  "Failed to allocate QPLIB context: %#x\n", rc);
1458 		goto disable_rcfw;
1459 	}
1460 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
1461 					 rdev->qplib_ctx.stats.dma_map,
1462 					 &rdev->qplib_ctx.stats.fw_id);
1463 	if (rc) {
1464 		ibdev_err(&rdev->ibdev,
1465 			  "Failed to allocate stats context: %#x\n", rc);
1466 		goto free_ctx;
1467 	}
1468 
1469 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx,
1470 				  rdev->is_virtfn);
1471 	if (rc) {
1472 		ibdev_err(&rdev->ibdev,
1473 			  "Failed to initialize RCFW: %#x\n", rc);
1474 		goto free_sctx;
1475 	}
1476 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1477 
1478 	/* Resources based on the 'new' device caps */
1479 	rc = bnxt_re_alloc_res(rdev);
1480 	if (rc) {
1481 		ibdev_err(&rdev->ibdev,
1482 			  "Failed to allocate resources: %#x\n", rc);
1483 		goto fail;
1484 	}
1485 	set_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags);
1486 	rc = bnxt_re_init_res(rdev);
1487 	if (rc) {
1488 		ibdev_err(&rdev->ibdev,
1489 			  "Failed to initialize resources: %#x\n", rc);
1490 		goto fail;
1491 	}
1492 
1493 	set_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED, &rdev->flags);
1494 
1495 	if (!rdev->is_virtfn) {
1496 		rc = bnxt_re_setup_qos(rdev);
1497 		if (rc)
1498 			ibdev_info(&rdev->ibdev,
1499 				   "RoCE priority not yet configured\n");
1500 
1501 		INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1502 		set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1503 		schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1504 	}
1505 
1506 	return 0;
1507 free_sctx:
1508 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1509 free_ctx:
1510 	bnxt_qplib_free_ctx(&rdev->qplib_res, &rdev->qplib_ctx);
1511 disable_rcfw:
1512 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1513 free_ring:
1514 	type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1515 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id, type);
1516 free_rcfw:
1517 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1518 fail:
1519 	bnxt_re_dev_uninit(rdev);
1520 
1521 	return rc;
1522 }
1523 
1524 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1525 {
1526 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1527 	struct net_device *netdev = rdev->netdev;
1528 
1529 	bnxt_re_dev_remove(rdev);
1530 
1531 	if (netdev)
1532 		bnxt_re_dev_unprobe(netdev, en_dev);
1533 }
1534 
1535 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1536 {
1537 	struct bnxt_en_dev *en_dev;
1538 	int rc = 0;
1539 
1540 	if (!is_bnxt_re_dev(netdev))
1541 		return -ENODEV;
1542 
1543 	en_dev = bnxt_re_dev_probe(netdev);
1544 	if (IS_ERR(en_dev)) {
1545 		if (en_dev != ERR_PTR(-ENODEV))
1546 			ibdev_err(&(*rdev)->ibdev, "%s: Failed to probe\n",
1547 				  ROCE_DRV_MODULE_NAME);
1548 		rc = PTR_ERR(en_dev);
1549 		goto exit;
1550 	}
1551 	*rdev = bnxt_re_dev_add(netdev, en_dev);
1552 	if (!*rdev) {
1553 		rc = -ENOMEM;
1554 		bnxt_re_dev_unprobe(netdev, en_dev);
1555 		goto exit;
1556 	}
1557 exit:
1558 	return rc;
1559 }
1560 
1561 static void bnxt_re_remove_device(struct bnxt_re_dev *rdev)
1562 {
1563 	bnxt_re_dev_uninit(rdev);
1564 	pci_dev_put(rdev->en_dev->pdev);
1565 	bnxt_re_dev_unreg(rdev);
1566 }
1567 
1568 static int bnxt_re_add_device(struct bnxt_re_dev **rdev,
1569 			      struct net_device *netdev, u8 wqe_mode)
1570 {
1571 	int rc;
1572 
1573 	rc = bnxt_re_dev_reg(rdev, netdev);
1574 	if (rc == -ENODEV)
1575 		return rc;
1576 	if (rc) {
1577 		pr_err("Failed to register with the device %s: %#x\n",
1578 		       netdev->name, rc);
1579 		return rc;
1580 	}
1581 
1582 	pci_dev_get((*rdev)->en_dev->pdev);
1583 	rc = bnxt_re_dev_init(*rdev, wqe_mode);
1584 	if (rc) {
1585 		pci_dev_put((*rdev)->en_dev->pdev);
1586 		bnxt_re_dev_unreg(*rdev);
1587 	}
1588 
1589 	return rc;
1590 }
1591 
1592 static void bnxt_re_dealloc_driver(struct ib_device *ib_dev)
1593 {
1594 	struct bnxt_re_dev *rdev =
1595 		container_of(ib_dev, struct bnxt_re_dev, ibdev);
1596 
1597 	dev_info(rdev_to_dev(rdev), "Unregistering Device");
1598 
1599 	rtnl_lock();
1600 	bnxt_re_remove_device(rdev);
1601 	rtnl_unlock();
1602 }
1603 
1604 /* Handle all deferred netevents tasks */
1605 static void bnxt_re_task(struct work_struct *work)
1606 {
1607 	struct bnxt_re_work *re_work;
1608 	struct bnxt_re_dev *rdev;
1609 	int rc = 0;
1610 
1611 	re_work = container_of(work, struct bnxt_re_work, work);
1612 	rdev = re_work->rdev;
1613 
1614 	if (re_work->event == NETDEV_REGISTER) {
1615 		rc = bnxt_re_ib_init(rdev);
1616 		if (rc) {
1617 			ibdev_err(&rdev->ibdev,
1618 				  "Failed to register with IB: %#x", rc);
1619 			rtnl_lock();
1620 			bnxt_re_remove_device(rdev);
1621 			rtnl_unlock();
1622 			goto exit;
1623 		}
1624 		goto exit;
1625 	}
1626 
1627 	if (!ib_device_try_get(&rdev->ibdev))
1628 		goto exit;
1629 
1630 	switch (re_work->event) {
1631 	case NETDEV_UP:
1632 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1633 				       IB_EVENT_PORT_ACTIVE);
1634 		break;
1635 	case NETDEV_DOWN:
1636 		bnxt_re_dev_stop(rdev);
1637 		break;
1638 	case NETDEV_CHANGE:
1639 		if (!netif_carrier_ok(rdev->netdev))
1640 			bnxt_re_dev_stop(rdev);
1641 		else if (netif_carrier_ok(rdev->netdev))
1642 			bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1643 					       IB_EVENT_PORT_ACTIVE);
1644 		ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1645 				 &rdev->active_width);
1646 		break;
1647 	default:
1648 		break;
1649 	}
1650 	ib_device_put(&rdev->ibdev);
1651 exit:
1652 	put_device(&rdev->ibdev.dev);
1653 	kfree(re_work);
1654 }
1655 
1656 /*
1657  * "Notifier chain callback can be invoked for the same chain from
1658  * different CPUs at the same time".
1659  *
1660  * For cases when the netdev is already present, our call to the
1661  * register_netdevice_notifier() will actually get the rtnl_lock()
1662  * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1663  * events.
1664  *
1665  * But for cases when the netdev is not already present, the notifier
1666  * chain is subjected to be invoked from different CPUs simultaneously.
1667  *
1668  * This is protected by the netdev_mutex.
1669  */
1670 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1671 				unsigned long event, void *ptr)
1672 {
1673 	struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1674 	struct bnxt_re_work *re_work;
1675 	struct bnxt_re_dev *rdev;
1676 	int rc = 0;
1677 	bool sch_work = false;
1678 	bool release = true;
1679 
1680 	real_dev = rdma_vlan_dev_real_dev(netdev);
1681 	if (!real_dev)
1682 		real_dev = netdev;
1683 
1684 	rdev = bnxt_re_from_netdev(real_dev);
1685 	if (!rdev && event != NETDEV_REGISTER)
1686 		return NOTIFY_OK;
1687 
1688 	if (real_dev != netdev)
1689 		goto exit;
1690 
1691 	switch (event) {
1692 	case NETDEV_REGISTER:
1693 		if (rdev)
1694 			break;
1695 		rc = bnxt_re_add_device(&rdev, real_dev,
1696 					BNXT_QPLIB_WQE_MODE_STATIC);
1697 		if (!rc)
1698 			sch_work = true;
1699 		release = false;
1700 		break;
1701 
1702 	case NETDEV_UNREGISTER:
1703 		ib_unregister_device_queued(&rdev->ibdev);
1704 		break;
1705 
1706 	default:
1707 		sch_work = true;
1708 		break;
1709 	}
1710 	if (sch_work) {
1711 		/* Allocate for the deferred task */
1712 		re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1713 		if (re_work) {
1714 			get_device(&rdev->ibdev.dev);
1715 			re_work->rdev = rdev;
1716 			re_work->event = event;
1717 			re_work->vlan_dev = (real_dev == netdev ?
1718 					     NULL : netdev);
1719 			INIT_WORK(&re_work->work, bnxt_re_task);
1720 			queue_work(bnxt_re_wq, &re_work->work);
1721 		}
1722 	}
1723 
1724 exit:
1725 	if (rdev && release)
1726 		ib_device_put(&rdev->ibdev);
1727 	return NOTIFY_DONE;
1728 }
1729 
1730 static struct notifier_block bnxt_re_netdev_notifier = {
1731 	.notifier_call = bnxt_re_netdev_event
1732 };
1733 
1734 static int __init bnxt_re_mod_init(void)
1735 {
1736 	int rc = 0;
1737 
1738 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1739 
1740 	bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1741 	if (!bnxt_re_wq)
1742 		return -ENOMEM;
1743 
1744 	INIT_LIST_HEAD(&bnxt_re_dev_list);
1745 
1746 	rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1747 	if (rc) {
1748 		pr_err("%s: Cannot register to netdevice_notifier",
1749 		       ROCE_DRV_MODULE_NAME);
1750 		goto err_netdev;
1751 	}
1752 	return 0;
1753 
1754 err_netdev:
1755 	destroy_workqueue(bnxt_re_wq);
1756 
1757 	return rc;
1758 }
1759 
1760 static void __exit bnxt_re_mod_exit(void)
1761 {
1762 	struct bnxt_re_dev *rdev;
1763 
1764 	unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1765 	if (bnxt_re_wq)
1766 		destroy_workqueue(bnxt_re_wq);
1767 	list_for_each_entry(rdev, &bnxt_re_dev_list, list) {
1768 		/* VF device removal should be called before the removal
1769 		 * of PF device. Queue VFs unregister first, so that VFs
1770 		 * shall be removed before the PF during the call of
1771 		 * ib_unregister_driver.
1772 		 */
1773 		if (rdev->is_virtfn)
1774 			ib_unregister_device(&rdev->ibdev);
1775 	}
1776 	ib_unregister_driver(RDMA_DRIVER_BNXT_RE);
1777 }
1778 
1779 module_init(bnxt_re_mod_init);
1780 module_exit(bnxt_re_mod_exit);
1781