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 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev);
85 
86 static void bnxt_re_set_drv_mode(struct bnxt_re_dev *rdev, u8 mode)
87 {
88 	struct bnxt_qplib_chip_ctx *cctx;
89 
90 	cctx = rdev->chip_ctx;
91 	cctx->modes.wqe_mode = bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx) ?
92 			       mode : BNXT_QPLIB_WQE_MODE_STATIC;
93 }
94 
95 static void bnxt_re_destroy_chip_ctx(struct bnxt_re_dev *rdev)
96 {
97 	struct bnxt_qplib_chip_ctx *chip_ctx;
98 
99 	if (!rdev->chip_ctx)
100 		return;
101 	chip_ctx = rdev->chip_ctx;
102 	rdev->chip_ctx = NULL;
103 	rdev->rcfw.res = NULL;
104 	rdev->qplib_res.cctx = NULL;
105 	rdev->qplib_res.pdev = NULL;
106 	rdev->qplib_res.netdev = NULL;
107 	kfree(chip_ctx);
108 }
109 
110 static int bnxt_re_setup_chip_ctx(struct bnxt_re_dev *rdev, u8 wqe_mode)
111 {
112 	struct bnxt_qplib_chip_ctx *chip_ctx;
113 	struct bnxt_en_dev *en_dev;
114 	struct bnxt *bp;
115 
116 	en_dev = rdev->en_dev;
117 	bp = netdev_priv(en_dev->net);
118 
119 	chip_ctx = kzalloc(sizeof(*chip_ctx), GFP_KERNEL);
120 	if (!chip_ctx)
121 		return -ENOMEM;
122 	chip_ctx->chip_num = bp->chip_num;
123 	chip_ctx->hw_stats_size = bp->hw_ring_stats_size;
124 
125 	rdev->chip_ctx = chip_ctx;
126 	/* rest members to follow eventually */
127 
128 	rdev->qplib_res.cctx = rdev->chip_ctx;
129 	rdev->rcfw.res = &rdev->qplib_res;
130 
131 	bnxt_re_set_drv_mode(rdev, wqe_mode);
132 	if (bnxt_qplib_determine_atomics(en_dev->pdev))
133 		ibdev_info(&rdev->ibdev,
134 			   "platform doesn't support global atomics.");
135 	return 0;
136 }
137 
138 /* SR-IOV helper functions */
139 
140 static void bnxt_re_get_sriov_func_type(struct bnxt_re_dev *rdev)
141 {
142 	struct bnxt *bp;
143 
144 	bp = netdev_priv(rdev->en_dev->net);
145 	if (BNXT_VF(bp))
146 		rdev->is_virtfn = 1;
147 }
148 
149 /* Set the maximum number of each resource that the driver actually wants
150  * to allocate. This may be up to the maximum number the firmware has
151  * reserved for the function. The driver may choose to allocate fewer
152  * resources than the firmware maximum.
153  */
154 static void bnxt_re_limit_pf_res(struct bnxt_re_dev *rdev)
155 {
156 	struct bnxt_qplib_dev_attr *attr;
157 	struct bnxt_qplib_ctx *ctx;
158 	int i;
159 
160 	attr = &rdev->dev_attr;
161 	ctx = &rdev->qplib_ctx;
162 
163 	ctx->qpc_count = min_t(u32, BNXT_RE_MAX_QPC_COUNT,
164 			       attr->max_qp);
165 	ctx->mrw_count = BNXT_RE_MAX_MRW_COUNT_256K;
166 	/* Use max_mr from fw since max_mrw does not get set */
167 	ctx->mrw_count = min_t(u32, ctx->mrw_count, attr->max_mr);
168 	ctx->srqc_count = min_t(u32, BNXT_RE_MAX_SRQC_COUNT,
169 				attr->max_srq);
170 	ctx->cq_count = min_t(u32, BNXT_RE_MAX_CQ_COUNT, attr->max_cq);
171 	if (!bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx))
172 		for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
173 			rdev->qplib_ctx.tqm_ctx.qcount[i] =
174 			rdev->dev_attr.tqm_alloc_reqs[i];
175 }
176 
177 static void bnxt_re_limit_vf_res(struct bnxt_qplib_ctx *qplib_ctx, u32 num_vf)
178 {
179 	struct bnxt_qplib_vf_res *vf_res;
180 	u32 mrws = 0;
181 	u32 vf_pct;
182 	u32 nvfs;
183 
184 	vf_res = &qplib_ctx->vf_res;
185 	/*
186 	 * Reserve a set of resources for the PF. Divide the remaining
187 	 * resources among the VFs
188 	 */
189 	vf_pct = 100 - BNXT_RE_PCT_RSVD_FOR_PF;
190 	nvfs = num_vf;
191 	num_vf = 100 * num_vf;
192 	vf_res->max_qp_per_vf = (qplib_ctx->qpc_count * vf_pct) / num_vf;
193 	vf_res->max_srq_per_vf = (qplib_ctx->srqc_count * vf_pct) / num_vf;
194 	vf_res->max_cq_per_vf = (qplib_ctx->cq_count * vf_pct) / num_vf;
195 	/*
196 	 * The driver allows many more MRs than other resources. If the
197 	 * firmware does also, then reserve a fixed amount for the PF and
198 	 * divide the rest among VFs. VFs may use many MRs for NFS
199 	 * mounts, ISER, NVME applications, etc. If the firmware severely
200 	 * restricts the number of MRs, then let PF have half and divide
201 	 * the rest among VFs, as for the other resource types.
202 	 */
203 	if (qplib_ctx->mrw_count < BNXT_RE_MAX_MRW_COUNT_64K) {
204 		mrws = qplib_ctx->mrw_count * vf_pct;
205 		nvfs = num_vf;
206 	} else {
207 		mrws = qplib_ctx->mrw_count - BNXT_RE_RESVD_MR_FOR_PF;
208 	}
209 	vf_res->max_mrw_per_vf = (mrws / nvfs);
210 	vf_res->max_gid_per_vf = BNXT_RE_MAX_GID_PER_VF;
211 }
212 
213 static void bnxt_re_set_resource_limits(struct bnxt_re_dev *rdev)
214 {
215 	u32 num_vfs;
216 
217 	memset(&rdev->qplib_ctx.vf_res, 0, sizeof(struct bnxt_qplib_vf_res));
218 	bnxt_re_limit_pf_res(rdev);
219 
220 	num_vfs =  bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx) ?
221 			BNXT_RE_GEN_P5_MAX_VF : rdev->num_vfs;
222 	if (num_vfs)
223 		bnxt_re_limit_vf_res(&rdev->qplib_ctx, num_vfs);
224 }
225 
226 /* for handling bnxt_en callbacks later */
227 static void bnxt_re_stop(void *p)
228 {
229 	struct bnxt_re_dev *rdev = p;
230 	struct bnxt *bp;
231 
232 	if (!rdev)
233 		return;
234 	ASSERT_RTNL();
235 
236 	/* L2 driver invokes this callback during device error/crash or device
237 	 * reset. Current RoCE driver doesn't recover the device in case of
238 	 * error. Handle the error by dispatching fatal events to all qps
239 	 * ie. by calling bnxt_re_dev_stop and release the MSIx vectors as
240 	 * L2 driver want to modify the MSIx table.
241 	 */
242 	bp = netdev_priv(rdev->netdev);
243 
244 	ibdev_info(&rdev->ibdev, "Handle device stop call from L2 driver");
245 	/* Check the current device state from L2 structure and move the
246 	 * device to detached state if FW_FATAL_COND is set.
247 	 * This prevents more commands to HW during clean-up,
248 	 * in case the device is already in error.
249 	 */
250 	if (test_bit(BNXT_STATE_FW_FATAL_COND, &bp->state))
251 		set_bit(ERR_DEVICE_DETACHED, &rdev->rcfw.cmdq.flags);
252 
253 	bnxt_re_dev_stop(rdev);
254 	bnxt_re_stop_irq(rdev);
255 	/* Move the device states to detached and  avoid sending any more
256 	 * commands to HW
257 	 */
258 	set_bit(BNXT_RE_FLAG_ERR_DEVICE_DETACHED, &rdev->flags);
259 	set_bit(ERR_DEVICE_DETACHED, &rdev->rcfw.cmdq.flags);
260 }
261 
262 static void bnxt_re_start(void *p)
263 {
264 }
265 
266 static void bnxt_re_sriov_config(void *p, int num_vfs)
267 {
268 	struct bnxt_re_dev *rdev = p;
269 
270 	if (!rdev)
271 		return;
272 
273 	if (test_bit(BNXT_RE_FLAG_ERR_DEVICE_DETACHED, &rdev->flags))
274 		return;
275 	rdev->num_vfs = num_vfs;
276 	if (!bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx)) {
277 		bnxt_re_set_resource_limits(rdev);
278 		bnxt_qplib_set_func_resources(&rdev->qplib_res, &rdev->rcfw,
279 					      &rdev->qplib_ctx);
280 	}
281 }
282 
283 static void bnxt_re_shutdown(void *p)
284 {
285 	struct bnxt_re_dev *rdev = p;
286 
287 	if (!rdev)
288 		return;
289 	ASSERT_RTNL();
290 	/* Release the MSIx vectors before queuing unregister */
291 	bnxt_re_stop_irq(rdev);
292 	ib_unregister_device_queued(&rdev->ibdev);
293 }
294 
295 static void bnxt_re_stop_irq(void *handle)
296 {
297 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
298 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
299 	struct bnxt_qplib_nq *nq;
300 	int indx;
301 
302 	for (indx = BNXT_RE_NQ_IDX; indx < rdev->num_msix; indx++) {
303 		nq = &rdev->nq[indx - 1];
304 		bnxt_qplib_nq_stop_irq(nq, false);
305 	}
306 
307 	bnxt_qplib_rcfw_stop_irq(rcfw, false);
308 }
309 
310 static void bnxt_re_start_irq(void *handle, struct bnxt_msix_entry *ent)
311 {
312 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
313 	struct bnxt_msix_entry *msix_ent = rdev->msix_entries;
314 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
315 	struct bnxt_qplib_nq *nq;
316 	int indx, rc;
317 
318 	if (!ent) {
319 		/* Not setting the f/w timeout bit in rcfw.
320 		 * During the driver unload the first command
321 		 * to f/w will timeout and that will set the
322 		 * timeout bit.
323 		 */
324 		ibdev_err(&rdev->ibdev, "Failed to re-start IRQs\n");
325 		return;
326 	}
327 
328 	/* Vectors may change after restart, so update with new vectors
329 	 * in device sctructure.
330 	 */
331 	for (indx = 0; indx < rdev->num_msix; indx++)
332 		rdev->msix_entries[indx].vector = ent[indx].vector;
333 
334 	bnxt_qplib_rcfw_start_irq(rcfw, msix_ent[BNXT_RE_AEQ_IDX].vector,
335 				  false);
336 	for (indx = BNXT_RE_NQ_IDX ; indx < rdev->num_msix; indx++) {
337 		nq = &rdev->nq[indx - 1];
338 		rc = bnxt_qplib_nq_start_irq(nq, indx - 1,
339 					     msix_ent[indx].vector, false);
340 		if (rc)
341 			ibdev_warn(&rdev->ibdev, "Failed to reinit NQ index %d\n",
342 				   indx - 1);
343 	}
344 }
345 
346 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
347 	.ulp_async_notifier = NULL,
348 	.ulp_stop = bnxt_re_stop,
349 	.ulp_start = bnxt_re_start,
350 	.ulp_sriov_config = bnxt_re_sriov_config,
351 	.ulp_shutdown = bnxt_re_shutdown,
352 	.ulp_irq_stop = bnxt_re_stop_irq,
353 	.ulp_irq_restart = bnxt_re_start_irq
354 };
355 
356 /* RoCE -> Net driver */
357 
358 /* Driver registration routines used to let the networking driver (bnxt_en)
359  * to know that the RoCE driver is now installed
360  */
361 static int bnxt_re_unregister_netdev(struct bnxt_re_dev *rdev)
362 {
363 	struct bnxt_en_dev *en_dev;
364 	int rc;
365 
366 	if (!rdev)
367 		return -EINVAL;
368 
369 	en_dev = rdev->en_dev;
370 
371 	rc = en_dev->en_ops->bnxt_unregister_device(rdev->en_dev,
372 						    BNXT_ROCE_ULP);
373 	return rc;
374 }
375 
376 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
377 {
378 	struct bnxt_en_dev *en_dev;
379 	int rc = 0;
380 
381 	if (!rdev)
382 		return -EINVAL;
383 
384 	en_dev = rdev->en_dev;
385 
386 	rc = en_dev->en_ops->bnxt_register_device(en_dev, BNXT_ROCE_ULP,
387 						  &bnxt_re_ulp_ops, rdev);
388 	rdev->qplib_res.pdev = rdev->en_dev->pdev;
389 	return rc;
390 }
391 
392 static int bnxt_re_free_msix(struct bnxt_re_dev *rdev)
393 {
394 	struct bnxt_en_dev *en_dev;
395 	int rc;
396 
397 	if (!rdev)
398 		return -EINVAL;
399 
400 	en_dev = rdev->en_dev;
401 
402 
403 	rc = en_dev->en_ops->bnxt_free_msix(rdev->en_dev, BNXT_ROCE_ULP);
404 
405 	return rc;
406 }
407 
408 static int bnxt_re_request_msix(struct bnxt_re_dev *rdev)
409 {
410 	int rc = 0, num_msix_want = BNXT_RE_MAX_MSIX, num_msix_got;
411 	struct bnxt_en_dev *en_dev;
412 
413 	if (!rdev)
414 		return -EINVAL;
415 
416 	en_dev = rdev->en_dev;
417 
418 	num_msix_want = min_t(u32, BNXT_RE_MAX_MSIX, num_online_cpus());
419 
420 	num_msix_got = en_dev->en_ops->bnxt_request_msix(en_dev, BNXT_ROCE_ULP,
421 							 rdev->msix_entries,
422 							 num_msix_want);
423 	if (num_msix_got < BNXT_RE_MIN_MSIX) {
424 		rc = -EINVAL;
425 		goto done;
426 	}
427 	if (num_msix_got != num_msix_want) {
428 		ibdev_warn(&rdev->ibdev,
429 			   "Requested %d MSI-X vectors, got %d\n",
430 			   num_msix_want, num_msix_got);
431 	}
432 	rdev->num_msix = num_msix_got;
433 done:
434 	return rc;
435 }
436 
437 static void bnxt_re_init_hwrm_hdr(struct bnxt_re_dev *rdev, struct input *hdr,
438 				  u16 opcd, u16 crid, u16 trid)
439 {
440 	hdr->req_type = cpu_to_le16(opcd);
441 	hdr->cmpl_ring = cpu_to_le16(crid);
442 	hdr->target_id = cpu_to_le16(trid);
443 }
444 
445 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
446 				int msg_len, void *resp, int resp_max_len,
447 				int timeout)
448 {
449 	fw_msg->msg = msg;
450 	fw_msg->msg_len = msg_len;
451 	fw_msg->resp = resp;
452 	fw_msg->resp_max_len = resp_max_len;
453 	fw_msg->timeout = timeout;
454 }
455 
456 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev,
457 				 u16 fw_ring_id, int type)
458 {
459 	struct bnxt_en_dev *en_dev = rdev->en_dev;
460 	struct hwrm_ring_free_input req = {0};
461 	struct hwrm_ring_free_output resp;
462 	struct bnxt_fw_msg fw_msg;
463 	int rc = -EINVAL;
464 
465 	if (!en_dev)
466 		return rc;
467 
468 	if (test_bit(BNXT_RE_FLAG_ERR_DEVICE_DETACHED, &rdev->flags))
469 		return 0;
470 
471 	memset(&fw_msg, 0, sizeof(fw_msg));
472 
473 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1);
474 	req.ring_type = type;
475 	req.ring_id = cpu_to_le16(fw_ring_id);
476 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
477 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
478 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
479 	if (rc)
480 		ibdev_err(&rdev->ibdev, "Failed to free HW ring:%d :%#x",
481 			  req.ring_id, rc);
482 	return rc;
483 }
484 
485 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev,
486 				  struct bnxt_re_ring_attr *ring_attr,
487 				  u16 *fw_ring_id)
488 {
489 	struct bnxt_en_dev *en_dev = rdev->en_dev;
490 	struct hwrm_ring_alloc_input req = {0};
491 	struct hwrm_ring_alloc_output resp;
492 	struct bnxt_fw_msg fw_msg;
493 	int rc = -EINVAL;
494 
495 	if (!en_dev)
496 		return rc;
497 
498 	memset(&fw_msg, 0, sizeof(fw_msg));
499 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_ALLOC, -1, -1);
500 	req.enables = 0;
501 	req.page_tbl_addr =  cpu_to_le64(ring_attr->dma_arr[0]);
502 	if (ring_attr->pages > 1) {
503 		/* Page size is in log2 units */
504 		req.page_size = BNXT_PAGE_SHIFT;
505 		req.page_tbl_depth = 1;
506 	}
507 	req.fbo = 0;
508 	/* Association of ring index with doorbell index and MSIX number */
509 	req.logical_id = cpu_to_le16(ring_attr->lrid);
510 	req.length = cpu_to_le32(ring_attr->depth + 1);
511 	req.ring_type = ring_attr->type;
512 	req.int_mode = ring_attr->mode;
513 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
514 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
515 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
516 	if (!rc)
517 		*fw_ring_id = le16_to_cpu(resp.ring_id);
518 
519 	return rc;
520 }
521 
522 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
523 				      u32 fw_stats_ctx_id)
524 {
525 	struct bnxt_en_dev *en_dev = rdev->en_dev;
526 	struct hwrm_stat_ctx_free_input req = {0};
527 	struct bnxt_fw_msg fw_msg;
528 	int rc = -EINVAL;
529 
530 	if (!en_dev)
531 		return rc;
532 
533 	if (test_bit(BNXT_RE_FLAG_ERR_DEVICE_DETACHED, &rdev->flags))
534 		return 0;
535 
536 	memset(&fw_msg, 0, sizeof(fw_msg));
537 
538 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_FREE, -1, -1);
539 	req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
540 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&req,
541 			    sizeof(req), DFLT_HWRM_CMD_TIMEOUT);
542 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
543 	if (rc)
544 		ibdev_err(&rdev->ibdev, "Failed to free HW stats context %#x",
545 			  rc);
546 
547 	return rc;
548 }
549 
550 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
551 				       dma_addr_t dma_map,
552 				       u32 *fw_stats_ctx_id)
553 {
554 	struct bnxt_qplib_chip_ctx *chip_ctx = rdev->chip_ctx;
555 	struct hwrm_stat_ctx_alloc_output resp = {0};
556 	struct hwrm_stat_ctx_alloc_input req = {0};
557 	struct bnxt_en_dev *en_dev = rdev->en_dev;
558 	struct bnxt_fw_msg fw_msg;
559 	int rc = -EINVAL;
560 
561 	*fw_stats_ctx_id = INVALID_STATS_CTX_ID;
562 
563 	if (!en_dev)
564 		return rc;
565 
566 	memset(&fw_msg, 0, sizeof(fw_msg));
567 
568 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1);
569 	req.update_period_ms = cpu_to_le32(1000);
570 	req.stats_dma_addr = cpu_to_le64(dma_map);
571 	req.stats_dma_length = cpu_to_le16(chip_ctx->hw_stats_size);
572 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
573 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
574 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
575 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
576 	if (!rc)
577 		*fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
578 
579 	return rc;
580 }
581 
582 /* Device */
583 
584 static bool is_bnxt_re_dev(struct net_device *netdev)
585 {
586 	struct ethtool_drvinfo drvinfo;
587 
588 	if (netdev->ethtool_ops && netdev->ethtool_ops->get_drvinfo) {
589 		memset(&drvinfo, 0, sizeof(drvinfo));
590 		netdev->ethtool_ops->get_drvinfo(netdev, &drvinfo);
591 
592 		if (strcmp(drvinfo.driver, "bnxt_en"))
593 			return false;
594 		return true;
595 	}
596 	return false;
597 }
598 
599 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
600 {
601 	struct ib_device *ibdev =
602 		ib_device_get_by_netdev(netdev, RDMA_DRIVER_BNXT_RE);
603 	if (!ibdev)
604 		return NULL;
605 
606 	return container_of(ibdev, struct bnxt_re_dev, ibdev);
607 }
608 
609 static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev)
610 {
611 	struct bnxt_en_dev *en_dev;
612 	struct pci_dev *pdev;
613 
614 	en_dev = bnxt_ulp_probe(netdev);
615 	if (IS_ERR(en_dev))
616 		return en_dev;
617 
618 	pdev = en_dev->pdev;
619 	if (!pdev)
620 		return ERR_PTR(-EINVAL);
621 
622 	if (!(en_dev->flags & BNXT_EN_FLAG_ROCE_CAP)) {
623 		dev_info(&pdev->dev,
624 			"%s: probe error: RoCE is not supported on this device",
625 			ROCE_DRV_MODULE_NAME);
626 		return ERR_PTR(-ENODEV);
627 	}
628 
629 	dev_hold(netdev);
630 
631 	return en_dev;
632 }
633 
634 static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr,
635 			   char *buf)
636 {
637 	struct bnxt_re_dev *rdev =
638 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
639 
640 	return sysfs_emit(buf, "0x%x\n", rdev->en_dev->pdev->vendor);
641 }
642 static DEVICE_ATTR_RO(hw_rev);
643 
644 static ssize_t hca_type_show(struct device *device,
645 			     struct device_attribute *attr, char *buf)
646 {
647 	struct bnxt_re_dev *rdev =
648 		rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
649 
650 	return sysfs_emit(buf, "%s\n", rdev->ibdev.node_desc);
651 }
652 static DEVICE_ATTR_RO(hca_type);
653 
654 static struct attribute *bnxt_re_attributes[] = {
655 	&dev_attr_hw_rev.attr,
656 	&dev_attr_hca_type.attr,
657 	NULL
658 };
659 
660 static const struct attribute_group bnxt_re_dev_attr_group = {
661 	.attrs = bnxt_re_attributes,
662 };
663 
664 static const struct ib_device_ops bnxt_re_dev_ops = {
665 	.owner = THIS_MODULE,
666 	.driver_id = RDMA_DRIVER_BNXT_RE,
667 	.uverbs_abi_ver = BNXT_RE_ABI_VERSION,
668 
669 	.add_gid = bnxt_re_add_gid,
670 	.alloc_hw_port_stats = bnxt_re_ib_alloc_hw_port_stats,
671 	.alloc_mr = bnxt_re_alloc_mr,
672 	.alloc_pd = bnxt_re_alloc_pd,
673 	.alloc_ucontext = bnxt_re_alloc_ucontext,
674 	.create_ah = bnxt_re_create_ah,
675 	.create_cq = bnxt_re_create_cq,
676 	.create_qp = bnxt_re_create_qp,
677 	.create_srq = bnxt_re_create_srq,
678 	.create_user_ah = bnxt_re_create_ah,
679 	.dealloc_driver = bnxt_re_dealloc_driver,
680 	.dealloc_pd = bnxt_re_dealloc_pd,
681 	.dealloc_ucontext = bnxt_re_dealloc_ucontext,
682 	.del_gid = bnxt_re_del_gid,
683 	.dereg_mr = bnxt_re_dereg_mr,
684 	.destroy_ah = bnxt_re_destroy_ah,
685 	.destroy_cq = bnxt_re_destroy_cq,
686 	.destroy_qp = bnxt_re_destroy_qp,
687 	.destroy_srq = bnxt_re_destroy_srq,
688 	.device_group = &bnxt_re_dev_attr_group,
689 	.get_dev_fw_str = bnxt_re_query_fw_str,
690 	.get_dma_mr = bnxt_re_get_dma_mr,
691 	.get_hw_stats = bnxt_re_ib_get_hw_stats,
692 	.get_link_layer = bnxt_re_get_link_layer,
693 	.get_port_immutable = bnxt_re_get_port_immutable,
694 	.map_mr_sg = bnxt_re_map_mr_sg,
695 	.mmap = bnxt_re_mmap,
696 	.modify_ah = bnxt_re_modify_ah,
697 	.modify_qp = bnxt_re_modify_qp,
698 	.modify_srq = bnxt_re_modify_srq,
699 	.poll_cq = bnxt_re_poll_cq,
700 	.post_recv = bnxt_re_post_recv,
701 	.post_send = bnxt_re_post_send,
702 	.post_srq_recv = bnxt_re_post_srq_recv,
703 	.query_ah = bnxt_re_query_ah,
704 	.query_device = bnxt_re_query_device,
705 	.query_pkey = bnxt_re_query_pkey,
706 	.query_port = bnxt_re_query_port,
707 	.query_qp = bnxt_re_query_qp,
708 	.query_srq = bnxt_re_query_srq,
709 	.reg_user_mr = bnxt_re_reg_user_mr,
710 	.req_notify_cq = bnxt_re_req_notify_cq,
711 	INIT_RDMA_OBJ_SIZE(ib_ah, bnxt_re_ah, ib_ah),
712 	INIT_RDMA_OBJ_SIZE(ib_cq, bnxt_re_cq, ib_cq),
713 	INIT_RDMA_OBJ_SIZE(ib_pd, bnxt_re_pd, ib_pd),
714 	INIT_RDMA_OBJ_SIZE(ib_qp, bnxt_re_qp, ib_qp),
715 	INIT_RDMA_OBJ_SIZE(ib_srq, bnxt_re_srq, ib_srq),
716 	INIT_RDMA_OBJ_SIZE(ib_ucontext, bnxt_re_ucontext, ib_uctx),
717 };
718 
719 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
720 {
721 	struct ib_device *ibdev = &rdev->ibdev;
722 	int ret;
723 
724 	/* ib device init */
725 	ibdev->node_type = RDMA_NODE_IB_CA;
726 	strlcpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
727 		strlen(BNXT_RE_DESC) + 5);
728 	ibdev->phys_port_cnt = 1;
729 
730 	bnxt_qplib_get_guid(rdev->netdev->dev_addr, (u8 *)&ibdev->node_guid);
731 
732 	ibdev->num_comp_vectors	= rdev->num_msix - 1;
733 	ibdev->dev.parent = &rdev->en_dev->pdev->dev;
734 	ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
735 
736 	ib_set_device_ops(ibdev, &bnxt_re_dev_ops);
737 	ret = ib_device_set_netdev(&rdev->ibdev, rdev->netdev, 1);
738 	if (ret)
739 		return ret;
740 
741 	dma_set_max_seg_size(&rdev->en_dev->pdev->dev, UINT_MAX);
742 	return ib_register_device(ibdev, "bnxt_re%d", &rdev->en_dev->pdev->dev);
743 }
744 
745 static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev)
746 {
747 	dev_put(rdev->netdev);
748 	rdev->netdev = NULL;
749 	mutex_lock(&bnxt_re_dev_lock);
750 	list_del_rcu(&rdev->list);
751 	mutex_unlock(&bnxt_re_dev_lock);
752 
753 	synchronize_rcu();
754 }
755 
756 static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev,
757 					   struct bnxt_en_dev *en_dev)
758 {
759 	struct bnxt_re_dev *rdev;
760 
761 	/* Allocate bnxt_re_dev instance here */
762 	rdev = ib_alloc_device(bnxt_re_dev, ibdev);
763 	if (!rdev) {
764 		ibdev_err(NULL, "%s: bnxt_re_dev allocation failure!",
765 			  ROCE_DRV_MODULE_NAME);
766 		return NULL;
767 	}
768 	/* Default values */
769 	rdev->netdev = netdev;
770 	dev_hold(rdev->netdev);
771 	rdev->en_dev = en_dev;
772 	rdev->id = rdev->en_dev->pdev->devfn;
773 	INIT_LIST_HEAD(&rdev->qp_list);
774 	mutex_init(&rdev->qp_lock);
775 	atomic_set(&rdev->qp_count, 0);
776 	atomic_set(&rdev->cq_count, 0);
777 	atomic_set(&rdev->srq_count, 0);
778 	atomic_set(&rdev->mr_count, 0);
779 	atomic_set(&rdev->mw_count, 0);
780 	rdev->cosq[0] = 0xFFFF;
781 	rdev->cosq[1] = 0xFFFF;
782 
783 	mutex_lock(&bnxt_re_dev_lock);
784 	list_add_tail_rcu(&rdev->list, &bnxt_re_dev_list);
785 	mutex_unlock(&bnxt_re_dev_lock);
786 	return rdev;
787 }
788 
789 static int bnxt_re_handle_unaffi_async_event(struct creq_func_event
790 					     *unaffi_async)
791 {
792 	switch (unaffi_async->event) {
793 	case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
794 		break;
795 	case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
796 		break;
797 	case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
798 		break;
799 	case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
800 		break;
801 	case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
802 		break;
803 	case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
804 		break;
805 	case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
806 		break;
807 	case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
808 		break;
809 	case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
810 		break;
811 	case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
812 		break;
813 	case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
814 		break;
815 	default:
816 		return -EINVAL;
817 	}
818 	return 0;
819 }
820 
821 static int bnxt_re_handle_qp_async_event(struct creq_qp_event *qp_event,
822 					 struct bnxt_re_qp *qp)
823 {
824 	struct ib_event event;
825 	unsigned int flags;
826 
827 	if (qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_ERR &&
828 	    rdma_is_kernel_res(&qp->ib_qp.res)) {
829 		flags = bnxt_re_lock_cqs(qp);
830 		bnxt_qplib_add_flush_qp(&qp->qplib_qp);
831 		bnxt_re_unlock_cqs(qp, flags);
832 	}
833 
834 	memset(&event, 0, sizeof(event));
835 	if (qp->qplib_qp.srq) {
836 		event.device = &qp->rdev->ibdev;
837 		event.element.qp = &qp->ib_qp;
838 		event.event = IB_EVENT_QP_LAST_WQE_REACHED;
839 	}
840 
841 	if (event.device && qp->ib_qp.event_handler)
842 		qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
843 
844 	return 0;
845 }
846 
847 static int bnxt_re_handle_affi_async_event(struct creq_qp_event *affi_async,
848 					   void *obj)
849 {
850 	int rc = 0;
851 	u8 event;
852 
853 	if (!obj)
854 		return rc; /* QP was already dead, still return success */
855 
856 	event = affi_async->event;
857 	if (event == CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION) {
858 		struct bnxt_qplib_qp *lib_qp = obj;
859 		struct bnxt_re_qp *qp = container_of(lib_qp, struct bnxt_re_qp,
860 						     qplib_qp);
861 		rc = bnxt_re_handle_qp_async_event(affi_async, qp);
862 	}
863 	return rc;
864 }
865 
866 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
867 			       void *aeqe, void *obj)
868 {
869 	struct creq_qp_event *affi_async;
870 	struct creq_func_event *unaffi_async;
871 	u8 type;
872 	int rc;
873 
874 	type = ((struct creq_base *)aeqe)->type;
875 	if (type == CREQ_BASE_TYPE_FUNC_EVENT) {
876 		unaffi_async = aeqe;
877 		rc = bnxt_re_handle_unaffi_async_event(unaffi_async);
878 	} else {
879 		affi_async = aeqe;
880 		rc = bnxt_re_handle_affi_async_event(affi_async, obj);
881 	}
882 
883 	return rc;
884 }
885 
886 static int bnxt_re_srqn_handler(struct bnxt_qplib_nq *nq,
887 				struct bnxt_qplib_srq *handle, u8 event)
888 {
889 	struct bnxt_re_srq *srq = container_of(handle, struct bnxt_re_srq,
890 					       qplib_srq);
891 	struct ib_event ib_event;
892 	int rc = 0;
893 
894 	ib_event.device = &srq->rdev->ibdev;
895 	ib_event.element.srq = &srq->ib_srq;
896 	if (event == NQ_SRQ_EVENT_EVENT_SRQ_THRESHOLD_EVENT)
897 		ib_event.event = IB_EVENT_SRQ_LIMIT_REACHED;
898 	else
899 		ib_event.event = IB_EVENT_SRQ_ERR;
900 
901 	if (srq->ib_srq.event_handler) {
902 		/* Lock event_handler? */
903 		(*srq->ib_srq.event_handler)(&ib_event,
904 					     srq->ib_srq.srq_context);
905 	}
906 	return rc;
907 }
908 
909 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
910 			       struct bnxt_qplib_cq *handle)
911 {
912 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
913 					     qplib_cq);
914 
915 	if (cq->ib_cq.comp_handler) {
916 		/* Lock comp_handler? */
917 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
918 	}
919 
920 	return 0;
921 }
922 
923 #define BNXT_RE_GEN_P5_PF_NQ_DB		0x10000
924 #define BNXT_RE_GEN_P5_VF_NQ_DB		0x4000
925 static u32 bnxt_re_get_nqdb_offset(struct bnxt_re_dev *rdev, u16 indx)
926 {
927 	return bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx) ?
928 		(rdev->is_virtfn ? BNXT_RE_GEN_P5_VF_NQ_DB :
929 				   BNXT_RE_GEN_P5_PF_NQ_DB) :
930 				   rdev->msix_entries[indx].db_offset;
931 }
932 
933 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
934 {
935 	int i;
936 
937 	for (i = 1; i < rdev->num_msix; i++)
938 		bnxt_qplib_disable_nq(&rdev->nq[i - 1]);
939 
940 	if (rdev->qplib_res.rcfw)
941 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
942 }
943 
944 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
945 {
946 	int num_vec_enabled = 0;
947 	int rc = 0, i;
948 	u32 db_offt;
949 
950 	bnxt_qplib_init_res(&rdev->qplib_res);
951 
952 	for (i = 1; i < rdev->num_msix ; i++) {
953 		db_offt = bnxt_re_get_nqdb_offset(rdev, i);
954 		rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
955 					  i - 1, rdev->msix_entries[i].vector,
956 					  db_offt, &bnxt_re_cqn_handler,
957 					  &bnxt_re_srqn_handler);
958 		if (rc) {
959 			ibdev_err(&rdev->ibdev,
960 				  "Failed to enable NQ with rc = 0x%x", rc);
961 			goto fail;
962 		}
963 		num_vec_enabled++;
964 	}
965 	return 0;
966 fail:
967 	for (i = num_vec_enabled; i >= 0; i--)
968 		bnxt_qplib_disable_nq(&rdev->nq[i]);
969 	return rc;
970 }
971 
972 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev)
973 {
974 	u8 type;
975 	int i;
976 
977 	for (i = 0; i < rdev->num_msix - 1; i++) {
978 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
979 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
980 		bnxt_qplib_free_nq(&rdev->nq[i]);
981 		rdev->nq[i].res = NULL;
982 	}
983 }
984 
985 static void bnxt_re_free_res(struct bnxt_re_dev *rdev)
986 {
987 	bnxt_re_free_nq_res(rdev);
988 
989 	if (rdev->qplib_res.dpi_tbl.max) {
990 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
991 				       &rdev->qplib_res.dpi_tbl,
992 				       &rdev->dpi_privileged);
993 	}
994 	if (rdev->qplib_res.rcfw) {
995 		bnxt_qplib_free_res(&rdev->qplib_res);
996 		rdev->qplib_res.rcfw = NULL;
997 	}
998 }
999 
1000 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
1001 {
1002 	struct bnxt_re_ring_attr rattr = {};
1003 	int num_vec_created = 0;
1004 	int rc = 0, i;
1005 	u8 type;
1006 
1007 	/* Configure and allocate resources for qplib */
1008 	rdev->qplib_res.rcfw = &rdev->rcfw;
1009 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
1010 				     rdev->is_virtfn);
1011 	if (rc)
1012 		goto fail;
1013 
1014 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
1015 				  rdev->netdev, &rdev->dev_attr);
1016 	if (rc)
1017 		goto fail;
1018 
1019 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
1020 				  &rdev->dpi_privileged,
1021 				  rdev);
1022 	if (rc)
1023 		goto dealloc_res;
1024 
1025 	for (i = 0; i < rdev->num_msix - 1; i++) {
1026 		struct bnxt_qplib_nq *nq;
1027 
1028 		nq = &rdev->nq[i];
1029 		nq->hwq.max_elements = BNXT_QPLIB_NQE_MAX_CNT;
1030 		rc = bnxt_qplib_alloc_nq(&rdev->qplib_res, &rdev->nq[i]);
1031 		if (rc) {
1032 			ibdev_err(&rdev->ibdev, "Alloc Failed NQ%d rc:%#x",
1033 				  i, rc);
1034 			goto free_nq;
1035 		}
1036 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1037 		rattr.dma_arr = nq->hwq.pbl[PBL_LVL_0].pg_map_arr;
1038 		rattr.pages = nq->hwq.pbl[rdev->nq[i].hwq.level].pg_count;
1039 		rattr.type = type;
1040 		rattr.mode = RING_ALLOC_REQ_INT_MODE_MSIX;
1041 		rattr.depth = BNXT_QPLIB_NQE_MAX_CNT - 1;
1042 		rattr.lrid = rdev->msix_entries[i + 1].ring_idx;
1043 		rc = bnxt_re_net_ring_alloc(rdev, &rattr, &nq->ring_id);
1044 		if (rc) {
1045 			ibdev_err(&rdev->ibdev,
1046 				  "Failed to allocate NQ fw id with rc = 0x%x",
1047 				  rc);
1048 			bnxt_qplib_free_nq(&rdev->nq[i]);
1049 			goto free_nq;
1050 		}
1051 		num_vec_created++;
1052 	}
1053 	return 0;
1054 free_nq:
1055 	for (i = num_vec_created - 1; i >= 0; i--) {
1056 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1057 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
1058 		bnxt_qplib_free_nq(&rdev->nq[i]);
1059 	}
1060 	bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
1061 			       &rdev->qplib_res.dpi_tbl,
1062 			       &rdev->dpi_privileged);
1063 dealloc_res:
1064 	bnxt_qplib_free_res(&rdev->qplib_res);
1065 
1066 fail:
1067 	rdev->qplib_res.rcfw = NULL;
1068 	return rc;
1069 }
1070 
1071 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
1072 				   u8 port_num, enum ib_event_type event)
1073 {
1074 	struct ib_event ib_event;
1075 
1076 	ib_event.device = ibdev;
1077 	if (qp) {
1078 		ib_event.element.qp = qp;
1079 		ib_event.event = event;
1080 		if (qp->event_handler)
1081 			qp->event_handler(&ib_event, qp->qp_context);
1082 
1083 	} else {
1084 		ib_event.element.port_num = port_num;
1085 		ib_event.event = event;
1086 		ib_dispatch_event(&ib_event);
1087 	}
1088 }
1089 
1090 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN      0x02
1091 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
1092 				      u64 *cid_map)
1093 {
1094 	struct hwrm_queue_pri2cos_qcfg_input req = {0};
1095 	struct bnxt *bp = netdev_priv(rdev->netdev);
1096 	struct hwrm_queue_pri2cos_qcfg_output resp;
1097 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1098 	struct bnxt_fw_msg fw_msg;
1099 	u32 flags = 0;
1100 	u8 *qcfgmap, *tmp_map;
1101 	int rc = 0, i;
1102 
1103 	if (!cid_map)
1104 		return -EINVAL;
1105 
1106 	memset(&fw_msg, 0, sizeof(fw_msg));
1107 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1108 			      HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
1109 	flags |= (dir & 0x01);
1110 	flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
1111 	req.flags = cpu_to_le32(flags);
1112 	req.port_id = bp->pf.port_id;
1113 
1114 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1115 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1116 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1117 	if (rc)
1118 		return rc;
1119 
1120 	if (resp.queue_cfg_info) {
1121 		ibdev_warn(&rdev->ibdev,
1122 			   "Asymmetric cos queue configuration detected");
1123 		ibdev_warn(&rdev->ibdev,
1124 			   " on device, QoS may not be fully functional\n");
1125 	}
1126 	qcfgmap = &resp.pri0_cos_queue_id;
1127 	tmp_map = (u8 *)cid_map;
1128 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
1129 		tmp_map[i] = qcfgmap[i];
1130 
1131 	return rc;
1132 }
1133 
1134 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
1135 					struct bnxt_re_qp *qp)
1136 {
1137 	return (qp->ib_qp.qp_type == IB_QPT_GSI) ||
1138 	       (qp == rdev->gsi_ctx.gsi_sqp);
1139 }
1140 
1141 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
1142 {
1143 	int mask = IB_QP_STATE;
1144 	struct ib_qp_attr qp_attr;
1145 	struct bnxt_re_qp *qp;
1146 
1147 	qp_attr.qp_state = IB_QPS_ERR;
1148 	mutex_lock(&rdev->qp_lock);
1149 	list_for_each_entry(qp, &rdev->qp_list, list) {
1150 		/* Modify the state of all QPs except QP1/Shadow QP */
1151 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
1152 			if (qp->qplib_qp.state !=
1153 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
1154 			    qp->qplib_qp.state !=
1155 			    CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1156 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
1157 						       1, IB_EVENT_QP_FATAL);
1158 				bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
1159 						  NULL);
1160 			}
1161 		}
1162 	}
1163 	mutex_unlock(&rdev->qp_lock);
1164 }
1165 
1166 static int bnxt_re_update_gid(struct bnxt_re_dev *rdev)
1167 {
1168 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
1169 	struct bnxt_qplib_gid gid;
1170 	u16 gid_idx, index;
1171 	int rc = 0;
1172 
1173 	if (!ib_device_try_get(&rdev->ibdev))
1174 		return 0;
1175 
1176 	if (!sgid_tbl) {
1177 		ibdev_err(&rdev->ibdev, "QPLIB: SGID table not allocated");
1178 		rc = -EINVAL;
1179 		goto out;
1180 	}
1181 
1182 	for (index = 0; index < sgid_tbl->active; index++) {
1183 		gid_idx = sgid_tbl->hw_id[index];
1184 
1185 		if (!memcmp(&sgid_tbl->tbl[index], &bnxt_qplib_gid_zero,
1186 			    sizeof(bnxt_qplib_gid_zero)))
1187 			continue;
1188 		/* need to modify the VLAN enable setting of non VLAN GID only
1189 		 * as setting is done for VLAN GID while adding GID
1190 		 */
1191 		if (sgid_tbl->vlan[index])
1192 			continue;
1193 
1194 		memcpy(&gid, &sgid_tbl->tbl[index], sizeof(gid));
1195 
1196 		rc = bnxt_qplib_update_sgid(sgid_tbl, &gid, gid_idx,
1197 					    rdev->qplib_res.netdev->dev_addr);
1198 	}
1199 out:
1200 	ib_device_put(&rdev->ibdev);
1201 	return rc;
1202 }
1203 
1204 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
1205 {
1206 	u32 prio_map = 0, tmp_map = 0;
1207 	struct net_device *netdev;
1208 	struct dcb_app app;
1209 
1210 	netdev = rdev->netdev;
1211 
1212 	memset(&app, 0, sizeof(app));
1213 	app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
1214 	app.protocol = ETH_P_IBOE;
1215 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1216 	prio_map = tmp_map;
1217 
1218 	app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
1219 	app.protocol = ROCE_V2_UDP_DPORT;
1220 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1221 	prio_map |= tmp_map;
1222 
1223 	return prio_map;
1224 }
1225 
1226 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
1227 {
1228 	u16 prio;
1229 	u8 id;
1230 
1231 	for (prio = 0, id = 0; prio < 8; prio++) {
1232 		if (prio_map & (1 << prio)) {
1233 			cosq[id] = cid_map[prio];
1234 			id++;
1235 			if (id == 2) /* Max 2 tcs supported */
1236 				break;
1237 		}
1238 	}
1239 }
1240 
1241 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
1242 {
1243 	u8 prio_map = 0;
1244 	u64 cid_map;
1245 	int rc;
1246 
1247 	/* Get priority for roce */
1248 	prio_map = bnxt_re_get_priority_mask(rdev);
1249 
1250 	if (prio_map == rdev->cur_prio_map)
1251 		return 0;
1252 	rdev->cur_prio_map = prio_map;
1253 	/* Get cosq id for this priority */
1254 	rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
1255 	if (rc) {
1256 		ibdev_warn(&rdev->ibdev, "no cos for p_mask %x\n", prio_map);
1257 		return rc;
1258 	}
1259 	/* Parse CoS IDs for app priority */
1260 	bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
1261 
1262 	/* Config BONO. */
1263 	rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
1264 	if (rc) {
1265 		ibdev_warn(&rdev->ibdev, "no tc for cos{%x, %x}\n",
1266 			   rdev->cosq[0], rdev->cosq[1]);
1267 		return rc;
1268 	}
1269 
1270 	/* Actual priorities are not programmed as they are already
1271 	 * done by L2 driver; just enable or disable priority vlan tagging
1272 	 */
1273 	if ((prio_map == 0 && rdev->qplib_res.prio) ||
1274 	    (prio_map != 0 && !rdev->qplib_res.prio)) {
1275 		rdev->qplib_res.prio = prio_map ? true : false;
1276 
1277 		bnxt_re_update_gid(rdev);
1278 	}
1279 
1280 	return 0;
1281 }
1282 
1283 static void bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev)
1284 {
1285 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1286 	struct hwrm_ver_get_output resp = {0};
1287 	struct hwrm_ver_get_input req = {0};
1288 	struct bnxt_fw_msg fw_msg;
1289 	int rc = 0;
1290 
1291 	memset(&fw_msg, 0, sizeof(fw_msg));
1292 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1293 			      HWRM_VER_GET, -1, -1);
1294 	req.hwrm_intf_maj = HWRM_VERSION_MAJOR;
1295 	req.hwrm_intf_min = HWRM_VERSION_MINOR;
1296 	req.hwrm_intf_upd = HWRM_VERSION_UPDATE;
1297 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1298 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1299 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1300 	if (rc) {
1301 		ibdev_err(&rdev->ibdev, "Failed to query HW version, rc = 0x%x",
1302 			  rc);
1303 		return;
1304 	}
1305 	rdev->qplib_ctx.hwrm_intf_ver =
1306 		(u64)le16_to_cpu(resp.hwrm_intf_major) << 48 |
1307 		(u64)le16_to_cpu(resp.hwrm_intf_minor) << 32 |
1308 		(u64)le16_to_cpu(resp.hwrm_intf_build) << 16 |
1309 		le16_to_cpu(resp.hwrm_intf_patch);
1310 }
1311 
1312 static int bnxt_re_ib_init(struct bnxt_re_dev *rdev)
1313 {
1314 	int rc = 0;
1315 	u32 event;
1316 
1317 	/* Register ib dev */
1318 	rc = bnxt_re_register_ib(rdev);
1319 	if (rc) {
1320 		pr_err("Failed to register with IB: %#x\n", rc);
1321 		return rc;
1322 	}
1323 	dev_info(rdev_to_dev(rdev), "Device registered successfully");
1324 	ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1325 			 &rdev->active_width);
1326 	set_bit(BNXT_RE_FLAG_ISSUE_ROCE_STATS, &rdev->flags);
1327 
1328 	event = netif_running(rdev->netdev) && netif_carrier_ok(rdev->netdev) ?
1329 		IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
1330 
1331 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, event);
1332 
1333 	return rc;
1334 }
1335 
1336 static void bnxt_re_dev_uninit(struct bnxt_re_dev *rdev)
1337 {
1338 	u8 type;
1339 	int rc;
1340 
1341 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
1342 		cancel_delayed_work_sync(&rdev->worker);
1343 
1344 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED,
1345 			       &rdev->flags))
1346 		bnxt_re_cleanup_res(rdev);
1347 	if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags))
1348 		bnxt_re_free_res(rdev);
1349 
1350 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
1351 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
1352 		if (rc)
1353 			ibdev_warn(&rdev->ibdev,
1354 				   "Failed to deinitialize RCFW: %#x", rc);
1355 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1356 		bnxt_qplib_free_ctx(&rdev->qplib_res, &rdev->qplib_ctx);
1357 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1358 		type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1359 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id, type);
1360 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1361 	}
1362 	if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
1363 		rc = bnxt_re_free_msix(rdev);
1364 		if (rc)
1365 			ibdev_warn(&rdev->ibdev,
1366 				   "Failed to free MSI-X vectors: %#x", rc);
1367 	}
1368 
1369 	bnxt_re_destroy_chip_ctx(rdev);
1370 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
1371 		rc = bnxt_re_unregister_netdev(rdev);
1372 		if (rc)
1373 			ibdev_warn(&rdev->ibdev,
1374 				   "Failed to unregister with netdev: %#x", rc);
1375 	}
1376 }
1377 
1378 /* worker thread for polling periodic events. Now used for QoS programming*/
1379 static void bnxt_re_worker(struct work_struct *work)
1380 {
1381 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
1382 						worker.work);
1383 
1384 	bnxt_re_setup_qos(rdev);
1385 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1386 }
1387 
1388 static int bnxt_re_dev_init(struct bnxt_re_dev *rdev, u8 wqe_mode)
1389 {
1390 	struct bnxt_qplib_creq_ctx *creq;
1391 	struct bnxt_re_ring_attr rattr;
1392 	u32 db_offt;
1393 	int vid;
1394 	u8 type;
1395 	int rc;
1396 
1397 	/* Registered a new RoCE device instance to netdev */
1398 	memset(&rattr, 0, sizeof(rattr));
1399 	rc = bnxt_re_register_netdev(rdev);
1400 	if (rc) {
1401 		ibdev_err(&rdev->ibdev,
1402 			  "Failed to register with netedev: %#x\n", rc);
1403 		return -EINVAL;
1404 	}
1405 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1406 
1407 	rc = bnxt_re_setup_chip_ctx(rdev, wqe_mode);
1408 	if (rc) {
1409 		ibdev_err(&rdev->ibdev, "Failed to get chip context\n");
1410 		return -EINVAL;
1411 	}
1412 
1413 	/* Check whether VF or PF */
1414 	bnxt_re_get_sriov_func_type(rdev);
1415 
1416 	rc = bnxt_re_request_msix(rdev);
1417 	if (rc) {
1418 		ibdev_err(&rdev->ibdev,
1419 			  "Failed to get MSI-X vectors: %#x\n", rc);
1420 		rc = -EINVAL;
1421 		goto fail;
1422 	}
1423 	set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
1424 
1425 	bnxt_re_query_hwrm_intf_version(rdev);
1426 
1427 	/* Establish RCFW Communication Channel to initialize the context
1428 	 * memory for the function and all child VFs
1429 	 */
1430 	rc = bnxt_qplib_alloc_rcfw_channel(&rdev->qplib_res, &rdev->rcfw,
1431 					   &rdev->qplib_ctx,
1432 					   BNXT_RE_MAX_QPC_COUNT);
1433 	if (rc) {
1434 		ibdev_err(&rdev->ibdev,
1435 			  "Failed to allocate RCFW Channel: %#x\n", rc);
1436 		goto fail;
1437 	}
1438 
1439 	type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1440 	creq = &rdev->rcfw.creq;
1441 	rattr.dma_arr = creq->hwq.pbl[PBL_LVL_0].pg_map_arr;
1442 	rattr.pages = creq->hwq.pbl[creq->hwq.level].pg_count;
1443 	rattr.type = type;
1444 	rattr.mode = RING_ALLOC_REQ_INT_MODE_MSIX;
1445 	rattr.depth = BNXT_QPLIB_CREQE_MAX_CNT - 1;
1446 	rattr.lrid = rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx;
1447 	rc = bnxt_re_net_ring_alloc(rdev, &rattr, &creq->ring_id);
1448 	if (rc) {
1449 		ibdev_err(&rdev->ibdev, "Failed to allocate CREQ: %#x\n", rc);
1450 		goto free_rcfw;
1451 	}
1452 	db_offt = bnxt_re_get_nqdb_offset(rdev, BNXT_RE_AEQ_IDX);
1453 	vid = rdev->msix_entries[BNXT_RE_AEQ_IDX].vector;
1454 	rc = bnxt_qplib_enable_rcfw_channel(&rdev->rcfw,
1455 					    vid, db_offt, rdev->is_virtfn,
1456 					    &bnxt_re_aeq_handler);
1457 	if (rc) {
1458 		ibdev_err(&rdev->ibdev, "Failed to enable RCFW channel: %#x\n",
1459 			  rc);
1460 		goto free_ring;
1461 	}
1462 
1463 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
1464 				     rdev->is_virtfn);
1465 	if (rc)
1466 		goto disable_rcfw;
1467 
1468 	bnxt_re_set_resource_limits(rdev);
1469 
1470 	rc = bnxt_qplib_alloc_ctx(&rdev->qplib_res, &rdev->qplib_ctx, 0,
1471 				  bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx));
1472 	if (rc) {
1473 		ibdev_err(&rdev->ibdev,
1474 			  "Failed to allocate QPLIB context: %#x\n", rc);
1475 		goto disable_rcfw;
1476 	}
1477 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
1478 					 rdev->qplib_ctx.stats.dma_map,
1479 					 &rdev->qplib_ctx.stats.fw_id);
1480 	if (rc) {
1481 		ibdev_err(&rdev->ibdev,
1482 			  "Failed to allocate stats context: %#x\n", rc);
1483 		goto free_ctx;
1484 	}
1485 
1486 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx,
1487 				  rdev->is_virtfn);
1488 	if (rc) {
1489 		ibdev_err(&rdev->ibdev,
1490 			  "Failed to initialize RCFW: %#x\n", rc);
1491 		goto free_sctx;
1492 	}
1493 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1494 
1495 	/* Resources based on the 'new' device caps */
1496 	rc = bnxt_re_alloc_res(rdev);
1497 	if (rc) {
1498 		ibdev_err(&rdev->ibdev,
1499 			  "Failed to allocate resources: %#x\n", rc);
1500 		goto fail;
1501 	}
1502 	set_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags);
1503 	rc = bnxt_re_init_res(rdev);
1504 	if (rc) {
1505 		ibdev_err(&rdev->ibdev,
1506 			  "Failed to initialize resources: %#x\n", rc);
1507 		goto fail;
1508 	}
1509 
1510 	set_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED, &rdev->flags);
1511 
1512 	if (!rdev->is_virtfn) {
1513 		rc = bnxt_re_setup_qos(rdev);
1514 		if (rc)
1515 			ibdev_info(&rdev->ibdev,
1516 				   "RoCE priority not yet configured\n");
1517 
1518 		INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1519 		set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1520 		schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1521 	}
1522 
1523 	return 0;
1524 free_sctx:
1525 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1526 free_ctx:
1527 	bnxt_qplib_free_ctx(&rdev->qplib_res, &rdev->qplib_ctx);
1528 disable_rcfw:
1529 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1530 free_ring:
1531 	type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1532 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id, type);
1533 free_rcfw:
1534 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1535 fail:
1536 	bnxt_re_dev_uninit(rdev);
1537 
1538 	return rc;
1539 }
1540 
1541 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1542 {
1543 	struct net_device *netdev = rdev->netdev;
1544 
1545 	bnxt_re_dev_remove(rdev);
1546 
1547 	if (netdev)
1548 		dev_put(netdev);
1549 }
1550 
1551 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1552 {
1553 	struct bnxt_en_dev *en_dev;
1554 	int rc = 0;
1555 
1556 	if (!is_bnxt_re_dev(netdev))
1557 		return -ENODEV;
1558 
1559 	en_dev = bnxt_re_dev_probe(netdev);
1560 	if (IS_ERR(en_dev)) {
1561 		if (en_dev != ERR_PTR(-ENODEV))
1562 			ibdev_err(&(*rdev)->ibdev, "%s: Failed to probe\n",
1563 				  ROCE_DRV_MODULE_NAME);
1564 		rc = PTR_ERR(en_dev);
1565 		goto exit;
1566 	}
1567 	*rdev = bnxt_re_dev_add(netdev, en_dev);
1568 	if (!*rdev) {
1569 		rc = -ENOMEM;
1570 		dev_put(netdev);
1571 		goto exit;
1572 	}
1573 exit:
1574 	return rc;
1575 }
1576 
1577 static void bnxt_re_remove_device(struct bnxt_re_dev *rdev)
1578 {
1579 	bnxt_re_dev_uninit(rdev);
1580 	pci_dev_put(rdev->en_dev->pdev);
1581 	bnxt_re_dev_unreg(rdev);
1582 }
1583 
1584 static int bnxt_re_add_device(struct bnxt_re_dev **rdev,
1585 			      struct net_device *netdev, u8 wqe_mode)
1586 {
1587 	int rc;
1588 
1589 	rc = bnxt_re_dev_reg(rdev, netdev);
1590 	if (rc == -ENODEV)
1591 		return rc;
1592 	if (rc) {
1593 		pr_err("Failed to register with the device %s: %#x\n",
1594 		       netdev->name, rc);
1595 		return rc;
1596 	}
1597 
1598 	pci_dev_get((*rdev)->en_dev->pdev);
1599 	rc = bnxt_re_dev_init(*rdev, wqe_mode);
1600 	if (rc) {
1601 		pci_dev_put((*rdev)->en_dev->pdev);
1602 		bnxt_re_dev_unreg(*rdev);
1603 	}
1604 
1605 	return rc;
1606 }
1607 
1608 static void bnxt_re_dealloc_driver(struct ib_device *ib_dev)
1609 {
1610 	struct bnxt_re_dev *rdev =
1611 		container_of(ib_dev, struct bnxt_re_dev, ibdev);
1612 
1613 	dev_info(rdev_to_dev(rdev), "Unregistering Device");
1614 
1615 	rtnl_lock();
1616 	bnxt_re_remove_device(rdev);
1617 	rtnl_unlock();
1618 }
1619 
1620 /* Handle all deferred netevents tasks */
1621 static void bnxt_re_task(struct work_struct *work)
1622 {
1623 	struct bnxt_re_work *re_work;
1624 	struct bnxt_re_dev *rdev;
1625 	int rc = 0;
1626 
1627 	re_work = container_of(work, struct bnxt_re_work, work);
1628 	rdev = re_work->rdev;
1629 
1630 	if (re_work->event == NETDEV_REGISTER) {
1631 		rc = bnxt_re_ib_init(rdev);
1632 		if (rc) {
1633 			ibdev_err(&rdev->ibdev,
1634 				  "Failed to register with IB: %#x", rc);
1635 			rtnl_lock();
1636 			bnxt_re_remove_device(rdev);
1637 			rtnl_unlock();
1638 			goto exit;
1639 		}
1640 		goto exit;
1641 	}
1642 
1643 	if (!ib_device_try_get(&rdev->ibdev))
1644 		goto exit;
1645 
1646 	switch (re_work->event) {
1647 	case NETDEV_UP:
1648 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1649 				       IB_EVENT_PORT_ACTIVE);
1650 		break;
1651 	case NETDEV_DOWN:
1652 		bnxt_re_dev_stop(rdev);
1653 		break;
1654 	case NETDEV_CHANGE:
1655 		if (!netif_carrier_ok(rdev->netdev))
1656 			bnxt_re_dev_stop(rdev);
1657 		else if (netif_carrier_ok(rdev->netdev))
1658 			bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1659 					       IB_EVENT_PORT_ACTIVE);
1660 		ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1661 				 &rdev->active_width);
1662 		break;
1663 	default:
1664 		break;
1665 	}
1666 	ib_device_put(&rdev->ibdev);
1667 exit:
1668 	put_device(&rdev->ibdev.dev);
1669 	kfree(re_work);
1670 }
1671 
1672 /*
1673  * "Notifier chain callback can be invoked for the same chain from
1674  * different CPUs at the same time".
1675  *
1676  * For cases when the netdev is already present, our call to the
1677  * register_netdevice_notifier() will actually get the rtnl_lock()
1678  * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1679  * events.
1680  *
1681  * But for cases when the netdev is not already present, the notifier
1682  * chain is subjected to be invoked from different CPUs simultaneously.
1683  *
1684  * This is protected by the netdev_mutex.
1685  */
1686 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1687 				unsigned long event, void *ptr)
1688 {
1689 	struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1690 	struct bnxt_re_work *re_work;
1691 	struct bnxt_re_dev *rdev;
1692 	int rc = 0;
1693 	bool sch_work = false;
1694 	bool release = true;
1695 
1696 	real_dev = rdma_vlan_dev_real_dev(netdev);
1697 	if (!real_dev)
1698 		real_dev = netdev;
1699 
1700 	rdev = bnxt_re_from_netdev(real_dev);
1701 	if (!rdev && event != NETDEV_REGISTER)
1702 		return NOTIFY_OK;
1703 
1704 	if (real_dev != netdev)
1705 		goto exit;
1706 
1707 	switch (event) {
1708 	case NETDEV_REGISTER:
1709 		if (rdev)
1710 			break;
1711 		rc = bnxt_re_add_device(&rdev, real_dev,
1712 					BNXT_QPLIB_WQE_MODE_STATIC);
1713 		if (!rc)
1714 			sch_work = true;
1715 		release = false;
1716 		break;
1717 
1718 	case NETDEV_UNREGISTER:
1719 		ib_unregister_device_queued(&rdev->ibdev);
1720 		break;
1721 
1722 	default:
1723 		sch_work = true;
1724 		break;
1725 	}
1726 	if (sch_work) {
1727 		/* Allocate for the deferred task */
1728 		re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1729 		if (re_work) {
1730 			get_device(&rdev->ibdev.dev);
1731 			re_work->rdev = rdev;
1732 			re_work->event = event;
1733 			re_work->vlan_dev = (real_dev == netdev ?
1734 					     NULL : netdev);
1735 			INIT_WORK(&re_work->work, bnxt_re_task);
1736 			queue_work(bnxt_re_wq, &re_work->work);
1737 		}
1738 	}
1739 
1740 exit:
1741 	if (rdev && release)
1742 		ib_device_put(&rdev->ibdev);
1743 	return NOTIFY_DONE;
1744 }
1745 
1746 static struct notifier_block bnxt_re_netdev_notifier = {
1747 	.notifier_call = bnxt_re_netdev_event
1748 };
1749 
1750 static int __init bnxt_re_mod_init(void)
1751 {
1752 	int rc = 0;
1753 
1754 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1755 
1756 	bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1757 	if (!bnxt_re_wq)
1758 		return -ENOMEM;
1759 
1760 	INIT_LIST_HEAD(&bnxt_re_dev_list);
1761 
1762 	rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1763 	if (rc) {
1764 		pr_err("%s: Cannot register to netdevice_notifier",
1765 		       ROCE_DRV_MODULE_NAME);
1766 		goto err_netdev;
1767 	}
1768 	return 0;
1769 
1770 err_netdev:
1771 	destroy_workqueue(bnxt_re_wq);
1772 
1773 	return rc;
1774 }
1775 
1776 static void __exit bnxt_re_mod_exit(void)
1777 {
1778 	struct bnxt_re_dev *rdev;
1779 
1780 	unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1781 	if (bnxt_re_wq)
1782 		destroy_workqueue(bnxt_re_wq);
1783 	list_for_each_entry(rdev, &bnxt_re_dev_list, list) {
1784 		/* VF device removal should be called before the removal
1785 		 * of PF device. Queue VFs unregister first, so that VFs
1786 		 * shall be removed before the PF during the call of
1787 		 * ib_unregister_driver.
1788 		 */
1789 		if (rdev->is_virtfn)
1790 			ib_unregister_device(&rdev->ibdev);
1791 	}
1792 	ib_unregister_driver(RDMA_DRIVER_BNXT_RE);
1793 }
1794 
1795 module_init(bnxt_re_mod_init);
1796 module_exit(bnxt_re_mod_exit);
1797