xref: /openbmc/linux/drivers/infiniband/hw/bnxt_re/main.c (revision dce8efa0575c8d9b5f9f9ae41437200c6d3e0bf3)
1 /*
2  * Broadcom NetXtreme-E RoCE driver.
3  *
4  * Copyright (c) 2016 - 2017, Broadcom. All rights reserved.  The term
5  * Broadcom refers to Broadcom Limited and/or its subsidiaries.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * BSD license below:
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  *
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in
21  *    the documentation and/or other materials provided with the
22  *    distribution.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS''
25  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
26  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS
28  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
32  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
33  * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
34  * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Description: Main component of the bnxt_re driver
37  */
38 
39 #include <linux/module.h>
40 #include <linux/netdevice.h>
41 #include <linux/ethtool.h>
42 #include <linux/mutex.h>
43 #include <linux/list.h>
44 #include <linux/rculist.h>
45 #include <linux/spinlock.h>
46 #include <linux/pci.h>
47 #include <net/dcbnl.h>
48 #include <net/ipv6.h>
49 #include <net/addrconf.h>
50 #include <linux/if_ether.h>
51 
52 #include <rdma/ib_verbs.h>
53 #include <rdma/ib_user_verbs.h>
54 #include <rdma/ib_umem.h>
55 #include <rdma/ib_addr.h>
56 
57 #include "bnxt_ulp.h"
58 #include "roce_hsi.h"
59 #include "qplib_res.h"
60 #include "qplib_sp.h"
61 #include "qplib_fp.h"
62 #include "qplib_rcfw.h"
63 #include "bnxt_re.h"
64 #include "ib_verbs.h"
65 #include <rdma/bnxt_re-abi.h>
66 #include "bnxt.h"
67 #include "hw_counters.h"
68 
69 static char version[] =
70 		BNXT_RE_DESC " v" ROCE_DRV_MODULE_VERSION "\n";
71 
72 MODULE_AUTHOR("Eddie Wai <eddie.wai@broadcom.com>");
73 MODULE_DESCRIPTION(BNXT_RE_DESC " Driver");
74 MODULE_LICENSE("Dual BSD/GPL");
75 
76 /* globals */
77 static struct list_head bnxt_re_dev_list = LIST_HEAD_INIT(bnxt_re_dev_list);
78 /* Mutex to protect the list of bnxt_re devices added */
79 static DEFINE_MUTEX(bnxt_re_dev_lock);
80 static struct workqueue_struct *bnxt_re_wq;
81 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev, bool lock_wait);
82 
83 /* SR-IOV helper functions */
84 
85 static void bnxt_re_get_sriov_func_type(struct bnxt_re_dev *rdev)
86 {
87 	struct bnxt *bp;
88 
89 	bp = netdev_priv(rdev->en_dev->net);
90 	if (BNXT_VF(bp))
91 		rdev->is_virtfn = 1;
92 }
93 
94 /* Set the maximum number of each resource that the driver actually wants
95  * to allocate. This may be up to the maximum number the firmware has
96  * reserved for the function. The driver may choose to allocate fewer
97  * resources than the firmware maximum.
98  */
99 static void bnxt_re_set_resource_limits(struct bnxt_re_dev *rdev)
100 {
101 	u32 vf_qps = 0, vf_srqs = 0, vf_cqs = 0, vf_mrws = 0, vf_gids = 0;
102 	u32 i;
103 	u32 vf_pct;
104 	u32 num_vfs;
105 	struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
106 
107 	rdev->qplib_ctx.qpc_count = min_t(u32, BNXT_RE_MAX_QPC_COUNT,
108 					  dev_attr->max_qp);
109 
110 	rdev->qplib_ctx.mrw_count = BNXT_RE_MAX_MRW_COUNT_256K;
111 	/* Use max_mr from fw since max_mrw does not get set */
112 	rdev->qplib_ctx.mrw_count = min_t(u32, rdev->qplib_ctx.mrw_count,
113 					  dev_attr->max_mr);
114 	rdev->qplib_ctx.srqc_count = min_t(u32, BNXT_RE_MAX_SRQC_COUNT,
115 					   dev_attr->max_srq);
116 	rdev->qplib_ctx.cq_count = min_t(u32, BNXT_RE_MAX_CQ_COUNT,
117 					 dev_attr->max_cq);
118 
119 	for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
120 		rdev->qplib_ctx.tqm_count[i] =
121 		rdev->dev_attr.tqm_alloc_reqs[i];
122 
123 	if (rdev->num_vfs) {
124 		/*
125 		 * Reserve a set of resources for the PF. Divide the remaining
126 		 * resources among the VFs
127 		 */
128 		vf_pct = 100 - BNXT_RE_PCT_RSVD_FOR_PF;
129 		num_vfs = 100 * rdev->num_vfs;
130 		vf_qps = (rdev->qplib_ctx.qpc_count * vf_pct) / num_vfs;
131 		vf_srqs = (rdev->qplib_ctx.srqc_count * vf_pct) / num_vfs;
132 		vf_cqs = (rdev->qplib_ctx.cq_count * vf_pct) / num_vfs;
133 		/*
134 		 * The driver allows many more MRs than other resources. If the
135 		 * firmware does also, then reserve a fixed amount for the PF
136 		 * and divide the rest among VFs. VFs may use many MRs for NFS
137 		 * mounts, ISER, NVME applications, etc. If the firmware
138 		 * severely restricts the number of MRs, then let PF have
139 		 * half and divide the rest among VFs, as for the other
140 		 * resource types.
141 		 */
142 		if (rdev->qplib_ctx.mrw_count < BNXT_RE_MAX_MRW_COUNT_64K)
143 			vf_mrws = rdev->qplib_ctx.mrw_count * vf_pct / num_vfs;
144 		else
145 			vf_mrws = (rdev->qplib_ctx.mrw_count -
146 				   BNXT_RE_RESVD_MR_FOR_PF) / rdev->num_vfs;
147 		vf_gids = BNXT_RE_MAX_GID_PER_VF;
148 	}
149 	rdev->qplib_ctx.vf_res.max_mrw_per_vf = vf_mrws;
150 	rdev->qplib_ctx.vf_res.max_gid_per_vf = vf_gids;
151 	rdev->qplib_ctx.vf_res.max_qp_per_vf = vf_qps;
152 	rdev->qplib_ctx.vf_res.max_srq_per_vf = vf_srqs;
153 	rdev->qplib_ctx.vf_res.max_cq_per_vf = vf_cqs;
154 }
155 
156 /* for handling bnxt_en callbacks later */
157 static void bnxt_re_stop(void *p)
158 {
159 }
160 
161 static void bnxt_re_start(void *p)
162 {
163 }
164 
165 static void bnxt_re_sriov_config(void *p, int num_vfs)
166 {
167 	struct bnxt_re_dev *rdev = p;
168 
169 	if (!rdev)
170 		return;
171 
172 	rdev->num_vfs = num_vfs;
173 	bnxt_re_set_resource_limits(rdev);
174 	bnxt_qplib_set_func_resources(&rdev->qplib_res, &rdev->rcfw,
175 				      &rdev->qplib_ctx);
176 }
177 
178 static void bnxt_re_shutdown(void *p)
179 {
180 	struct bnxt_re_dev *rdev = p;
181 
182 	if (!rdev)
183 		return;
184 
185 	bnxt_re_ib_unreg(rdev, false);
186 }
187 
188 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
189 	.ulp_async_notifier = NULL,
190 	.ulp_stop = bnxt_re_stop,
191 	.ulp_start = bnxt_re_start,
192 	.ulp_sriov_config = bnxt_re_sriov_config,
193 	.ulp_shutdown = bnxt_re_shutdown
194 };
195 
196 /* RoCE -> Net driver */
197 
198 /* Driver registration routines used to let the networking driver (bnxt_en)
199  * to know that the RoCE driver is now installed
200  */
201 static int bnxt_re_unregister_netdev(struct bnxt_re_dev *rdev, bool lock_wait)
202 {
203 	struct bnxt_en_dev *en_dev;
204 	int rc;
205 
206 	if (!rdev)
207 		return -EINVAL;
208 
209 	en_dev = rdev->en_dev;
210 	/* Acquire rtnl lock if it is not invokded from netdev event */
211 	if (lock_wait)
212 		rtnl_lock();
213 
214 	rc = en_dev->en_ops->bnxt_unregister_device(rdev->en_dev,
215 						    BNXT_ROCE_ULP);
216 	if (lock_wait)
217 		rtnl_unlock();
218 	return rc;
219 }
220 
221 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
222 {
223 	struct bnxt_en_dev *en_dev;
224 	int rc = 0;
225 
226 	if (!rdev)
227 		return -EINVAL;
228 
229 	en_dev = rdev->en_dev;
230 
231 	rtnl_lock();
232 	rc = en_dev->en_ops->bnxt_register_device(en_dev, BNXT_ROCE_ULP,
233 						  &bnxt_re_ulp_ops, rdev);
234 	rtnl_unlock();
235 	return rc;
236 }
237 
238 static int bnxt_re_free_msix(struct bnxt_re_dev *rdev, bool lock_wait)
239 {
240 	struct bnxt_en_dev *en_dev;
241 	int rc;
242 
243 	if (!rdev)
244 		return -EINVAL;
245 
246 	en_dev = rdev->en_dev;
247 
248 	if (lock_wait)
249 		rtnl_lock();
250 
251 	rc = en_dev->en_ops->bnxt_free_msix(rdev->en_dev, BNXT_ROCE_ULP);
252 
253 	if (lock_wait)
254 		rtnl_unlock();
255 	return rc;
256 }
257 
258 static int bnxt_re_request_msix(struct bnxt_re_dev *rdev)
259 {
260 	int rc = 0, num_msix_want = BNXT_RE_MAX_MSIX, num_msix_got;
261 	struct bnxt_en_dev *en_dev;
262 
263 	if (!rdev)
264 		return -EINVAL;
265 
266 	en_dev = rdev->en_dev;
267 
268 	num_msix_want = min_t(u32, BNXT_RE_MAX_MSIX, num_online_cpus());
269 
270 	rtnl_lock();
271 	num_msix_got = en_dev->en_ops->bnxt_request_msix(en_dev, BNXT_ROCE_ULP,
272 							 rdev->msix_entries,
273 							 num_msix_want);
274 	if (num_msix_got < BNXT_RE_MIN_MSIX) {
275 		rc = -EINVAL;
276 		goto done;
277 	}
278 	if (num_msix_got != num_msix_want) {
279 		dev_warn(rdev_to_dev(rdev),
280 			 "Requested %d MSI-X vectors, got %d\n",
281 			 num_msix_want, num_msix_got);
282 	}
283 	rdev->num_msix = num_msix_got;
284 done:
285 	rtnl_unlock();
286 	return rc;
287 }
288 
289 static void bnxt_re_init_hwrm_hdr(struct bnxt_re_dev *rdev, struct input *hdr,
290 				  u16 opcd, u16 crid, u16 trid)
291 {
292 	hdr->req_type = cpu_to_le16(opcd);
293 	hdr->cmpl_ring = cpu_to_le16(crid);
294 	hdr->target_id = cpu_to_le16(trid);
295 }
296 
297 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
298 				int msg_len, void *resp, int resp_max_len,
299 				int timeout)
300 {
301 	fw_msg->msg = msg;
302 	fw_msg->msg_len = msg_len;
303 	fw_msg->resp = resp;
304 	fw_msg->resp_max_len = resp_max_len;
305 	fw_msg->timeout = timeout;
306 }
307 
308 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev, u16 fw_ring_id,
309 				 bool lock_wait)
310 {
311 	struct bnxt_en_dev *en_dev = rdev->en_dev;
312 	struct hwrm_ring_free_input req = {0};
313 	struct hwrm_ring_free_output resp;
314 	struct bnxt_fw_msg fw_msg;
315 	bool do_unlock = false;
316 	int rc = -EINVAL;
317 
318 	if (!en_dev)
319 		return rc;
320 
321 	memset(&fw_msg, 0, sizeof(fw_msg));
322 	if (lock_wait) {
323 		rtnl_lock();
324 		do_unlock = true;
325 	}
326 
327 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1);
328 	req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
329 	req.ring_id = cpu_to_le16(fw_ring_id);
330 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
331 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
332 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
333 	if (rc)
334 		dev_err(rdev_to_dev(rdev),
335 			"Failed to free HW ring:%d :%#x", req.ring_id, rc);
336 	if (do_unlock)
337 		rtnl_unlock();
338 	return rc;
339 }
340 
341 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev, dma_addr_t *dma_arr,
342 				  int pages, int type, u32 ring_mask,
343 				  u32 map_index, u16 *fw_ring_id)
344 {
345 	struct bnxt_en_dev *en_dev = rdev->en_dev;
346 	struct hwrm_ring_alloc_input req = {0};
347 	struct hwrm_ring_alloc_output resp;
348 	struct bnxt_fw_msg fw_msg;
349 	int rc = -EINVAL;
350 
351 	if (!en_dev)
352 		return rc;
353 
354 	memset(&fw_msg, 0, sizeof(fw_msg));
355 	rtnl_lock();
356 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_ALLOC, -1, -1);
357 	req.enables = 0;
358 	req.page_tbl_addr =  cpu_to_le64(dma_arr[0]);
359 	if (pages > 1) {
360 		/* Page size is in log2 units */
361 		req.page_size = BNXT_PAGE_SHIFT;
362 		req.page_tbl_depth = 1;
363 	}
364 	req.fbo = 0;
365 	/* Association of ring index with doorbell index and MSIX number */
366 	req.logical_id = cpu_to_le16(map_index);
367 	req.length = cpu_to_le32(ring_mask + 1);
368 	req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
369 	req.int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
370 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
371 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
372 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
373 	if (!rc)
374 		*fw_ring_id = le16_to_cpu(resp.ring_id);
375 
376 	rtnl_unlock();
377 	return rc;
378 }
379 
380 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
381 				      u32 fw_stats_ctx_id, bool lock_wait)
382 {
383 	struct bnxt_en_dev *en_dev = rdev->en_dev;
384 	struct hwrm_stat_ctx_free_input req = {0};
385 	struct bnxt_fw_msg fw_msg;
386 	bool do_unlock = false;
387 	int rc = -EINVAL;
388 
389 	if (!en_dev)
390 		return rc;
391 
392 	memset(&fw_msg, 0, sizeof(fw_msg));
393 	if (lock_wait) {
394 		rtnl_lock();
395 		do_unlock = true;
396 	}
397 
398 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_FREE, -1, -1);
399 	req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
400 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&req,
401 			    sizeof(req), DFLT_HWRM_CMD_TIMEOUT);
402 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
403 	if (rc)
404 		dev_err(rdev_to_dev(rdev),
405 			"Failed to free HW stats context %#x", rc);
406 
407 	if (do_unlock)
408 		rtnl_unlock();
409 	return rc;
410 }
411 
412 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
413 				       dma_addr_t dma_map,
414 				       u32 *fw_stats_ctx_id)
415 {
416 	struct hwrm_stat_ctx_alloc_output resp = {0};
417 	struct hwrm_stat_ctx_alloc_input req = {0};
418 	struct bnxt_en_dev *en_dev = rdev->en_dev;
419 	struct bnxt_fw_msg fw_msg;
420 	int rc = -EINVAL;
421 
422 	*fw_stats_ctx_id = INVALID_STATS_CTX_ID;
423 
424 	if (!en_dev)
425 		return rc;
426 
427 	memset(&fw_msg, 0, sizeof(fw_msg));
428 	rtnl_lock();
429 
430 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1);
431 	req.update_period_ms = cpu_to_le32(1000);
432 	req.stats_dma_addr = cpu_to_le64(dma_map);
433 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
434 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
435 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
436 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
437 	if (!rc)
438 		*fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
439 
440 	rtnl_unlock();
441 	return rc;
442 }
443 
444 /* Device */
445 
446 static bool is_bnxt_re_dev(struct net_device *netdev)
447 {
448 	struct ethtool_drvinfo drvinfo;
449 
450 	if (netdev->ethtool_ops && netdev->ethtool_ops->get_drvinfo) {
451 		memset(&drvinfo, 0, sizeof(drvinfo));
452 		netdev->ethtool_ops->get_drvinfo(netdev, &drvinfo);
453 
454 		if (strcmp(drvinfo.driver, "bnxt_en"))
455 			return false;
456 		return true;
457 	}
458 	return false;
459 }
460 
461 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
462 {
463 	struct bnxt_re_dev *rdev;
464 
465 	rcu_read_lock();
466 	list_for_each_entry_rcu(rdev, &bnxt_re_dev_list, list) {
467 		if (rdev->netdev == netdev) {
468 			rcu_read_unlock();
469 			return rdev;
470 		}
471 	}
472 	rcu_read_unlock();
473 	return NULL;
474 }
475 
476 static void bnxt_re_dev_unprobe(struct net_device *netdev,
477 				struct bnxt_en_dev *en_dev)
478 {
479 	dev_put(netdev);
480 	module_put(en_dev->pdev->driver->driver.owner);
481 }
482 
483 static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev)
484 {
485 	struct bnxt *bp = netdev_priv(netdev);
486 	struct bnxt_en_dev *en_dev;
487 	struct pci_dev *pdev;
488 
489 	/* Call bnxt_en's RoCE probe via indirect API */
490 	if (!bp->ulp_probe)
491 		return ERR_PTR(-EINVAL);
492 
493 	en_dev = bp->ulp_probe(netdev);
494 	if (IS_ERR(en_dev))
495 		return en_dev;
496 
497 	pdev = en_dev->pdev;
498 	if (!pdev)
499 		return ERR_PTR(-EINVAL);
500 
501 	if (!(en_dev->flags & BNXT_EN_FLAG_ROCE_CAP)) {
502 		dev_info(&pdev->dev,
503 			"%s: probe error: RoCE is not supported on this device",
504 			ROCE_DRV_MODULE_NAME);
505 		return ERR_PTR(-ENODEV);
506 	}
507 
508 	/* Bump net device reference count */
509 	if (!try_module_get(pdev->driver->driver.owner))
510 		return ERR_PTR(-ENODEV);
511 
512 	dev_hold(netdev);
513 
514 	return en_dev;
515 }
516 
517 static void bnxt_re_unregister_ib(struct bnxt_re_dev *rdev)
518 {
519 	ib_unregister_device(&rdev->ibdev);
520 }
521 
522 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
523 {
524 	struct ib_device *ibdev = &rdev->ibdev;
525 
526 	/* ib device init */
527 	ibdev->owner = THIS_MODULE;
528 	ibdev->node_type = RDMA_NODE_IB_CA;
529 	strlcpy(ibdev->name, "bnxt_re%d", IB_DEVICE_NAME_MAX);
530 	strlcpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
531 		strlen(BNXT_RE_DESC) + 5);
532 	ibdev->phys_port_cnt = 1;
533 
534 	bnxt_qplib_get_guid(rdev->netdev->dev_addr, (u8 *)&ibdev->node_guid);
535 
536 	ibdev->num_comp_vectors	= 1;
537 	ibdev->dev.parent = &rdev->en_dev->pdev->dev;
538 	ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
539 
540 	/* User space */
541 	ibdev->uverbs_abi_ver = BNXT_RE_ABI_VERSION;
542 	ibdev->uverbs_cmd_mask =
543 			(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
544 			(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
545 			(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
546 			(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
547 			(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
548 			(1ull << IB_USER_VERBS_CMD_REG_MR)		|
549 			(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
550 			(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
551 			(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
552 			(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
553 			(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
554 			(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
555 			(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
556 			(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
557 			(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
558 			(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
559 			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
560 			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
561 			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
562 			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
563 			(1ull << IB_USER_VERBS_CMD_CREATE_AH)		|
564 			(1ull << IB_USER_VERBS_CMD_MODIFY_AH)		|
565 			(1ull << IB_USER_VERBS_CMD_QUERY_AH)		|
566 			(1ull << IB_USER_VERBS_CMD_DESTROY_AH);
567 	/* POLL_CQ and REQ_NOTIFY_CQ is directly handled in libbnxt_re */
568 
569 	/* Kernel verbs */
570 	ibdev->query_device		= bnxt_re_query_device;
571 	ibdev->modify_device		= bnxt_re_modify_device;
572 
573 	ibdev->query_port		= bnxt_re_query_port;
574 	ibdev->get_port_immutable	= bnxt_re_get_port_immutable;
575 	ibdev->get_dev_fw_str           = bnxt_re_query_fw_str;
576 	ibdev->query_pkey		= bnxt_re_query_pkey;
577 	ibdev->query_gid		= bnxt_re_query_gid;
578 	ibdev->get_netdev		= bnxt_re_get_netdev;
579 	ibdev->add_gid			= bnxt_re_add_gid;
580 	ibdev->del_gid			= bnxt_re_del_gid;
581 	ibdev->get_link_layer		= bnxt_re_get_link_layer;
582 
583 	ibdev->alloc_pd			= bnxt_re_alloc_pd;
584 	ibdev->dealloc_pd		= bnxt_re_dealloc_pd;
585 
586 	ibdev->create_ah		= bnxt_re_create_ah;
587 	ibdev->modify_ah		= bnxt_re_modify_ah;
588 	ibdev->query_ah			= bnxt_re_query_ah;
589 	ibdev->destroy_ah		= bnxt_re_destroy_ah;
590 
591 	ibdev->create_srq		= bnxt_re_create_srq;
592 	ibdev->modify_srq		= bnxt_re_modify_srq;
593 	ibdev->query_srq		= bnxt_re_query_srq;
594 	ibdev->destroy_srq		= bnxt_re_destroy_srq;
595 	ibdev->post_srq_recv		= bnxt_re_post_srq_recv;
596 
597 	ibdev->create_qp		= bnxt_re_create_qp;
598 	ibdev->modify_qp		= bnxt_re_modify_qp;
599 	ibdev->query_qp			= bnxt_re_query_qp;
600 	ibdev->destroy_qp		= bnxt_re_destroy_qp;
601 
602 	ibdev->post_send		= bnxt_re_post_send;
603 	ibdev->post_recv		= bnxt_re_post_recv;
604 
605 	ibdev->create_cq		= bnxt_re_create_cq;
606 	ibdev->destroy_cq		= bnxt_re_destroy_cq;
607 	ibdev->poll_cq			= bnxt_re_poll_cq;
608 	ibdev->req_notify_cq		= bnxt_re_req_notify_cq;
609 
610 	ibdev->get_dma_mr		= bnxt_re_get_dma_mr;
611 	ibdev->dereg_mr			= bnxt_re_dereg_mr;
612 	ibdev->alloc_mr			= bnxt_re_alloc_mr;
613 	ibdev->map_mr_sg		= bnxt_re_map_mr_sg;
614 
615 	ibdev->reg_user_mr		= bnxt_re_reg_user_mr;
616 	ibdev->alloc_ucontext		= bnxt_re_alloc_ucontext;
617 	ibdev->dealloc_ucontext		= bnxt_re_dealloc_ucontext;
618 	ibdev->mmap			= bnxt_re_mmap;
619 	ibdev->get_hw_stats             = bnxt_re_ib_get_hw_stats;
620 	ibdev->alloc_hw_stats           = bnxt_re_ib_alloc_hw_stats;
621 
622 	return ib_register_device(ibdev, NULL);
623 }
624 
625 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
626 			char *buf)
627 {
628 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
629 
630 	return scnprintf(buf, PAGE_SIZE, "0x%x\n", rdev->en_dev->pdev->vendor);
631 }
632 
633 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
634 			char *buf)
635 {
636 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
637 
638 	return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->ibdev.node_desc);
639 }
640 
641 static DEVICE_ATTR(hw_rev, 0444, show_rev, NULL);
642 static DEVICE_ATTR(hca_type, 0444, show_hca, NULL);
643 
644 static struct device_attribute *bnxt_re_attributes[] = {
645 	&dev_attr_hw_rev,
646 	&dev_attr_hca_type
647 };
648 
649 static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev)
650 {
651 	dev_put(rdev->netdev);
652 	rdev->netdev = NULL;
653 
654 	mutex_lock(&bnxt_re_dev_lock);
655 	list_del_rcu(&rdev->list);
656 	mutex_unlock(&bnxt_re_dev_lock);
657 
658 	synchronize_rcu();
659 
660 	ib_dealloc_device(&rdev->ibdev);
661 	/* rdev is gone */
662 }
663 
664 static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev,
665 					   struct bnxt_en_dev *en_dev)
666 {
667 	struct bnxt_re_dev *rdev;
668 
669 	/* Allocate bnxt_re_dev instance here */
670 	rdev = (struct bnxt_re_dev *)ib_alloc_device(sizeof(*rdev));
671 	if (!rdev) {
672 		dev_err(NULL, "%s: bnxt_re_dev allocation failure!",
673 			ROCE_DRV_MODULE_NAME);
674 		return NULL;
675 	}
676 	/* Default values */
677 	rdev->netdev = netdev;
678 	dev_hold(rdev->netdev);
679 	rdev->en_dev = en_dev;
680 	rdev->id = rdev->en_dev->pdev->devfn;
681 	INIT_LIST_HEAD(&rdev->qp_list);
682 	mutex_init(&rdev->qp_lock);
683 	atomic_set(&rdev->qp_count, 0);
684 	atomic_set(&rdev->cq_count, 0);
685 	atomic_set(&rdev->srq_count, 0);
686 	atomic_set(&rdev->mr_count, 0);
687 	atomic_set(&rdev->mw_count, 0);
688 	rdev->cosq[0] = 0xFFFF;
689 	rdev->cosq[1] = 0xFFFF;
690 
691 	mutex_lock(&bnxt_re_dev_lock);
692 	list_add_tail_rcu(&rdev->list, &bnxt_re_dev_list);
693 	mutex_unlock(&bnxt_re_dev_lock);
694 	return rdev;
695 }
696 
697 static int bnxt_re_handle_unaffi_async_event(struct creq_func_event
698 					     *unaffi_async)
699 {
700 	switch (unaffi_async->event) {
701 	case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
702 		break;
703 	case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
704 		break;
705 	case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
706 		break;
707 	case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
708 		break;
709 	case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
710 		break;
711 	case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
712 		break;
713 	case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
714 		break;
715 	case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
716 		break;
717 	case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
718 		break;
719 	case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
720 		break;
721 	case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
722 		break;
723 	default:
724 		return -EINVAL;
725 	}
726 	return 0;
727 }
728 
729 static int bnxt_re_handle_qp_async_event(struct creq_qp_event *qp_event,
730 					 struct bnxt_re_qp *qp)
731 {
732 	struct ib_event event;
733 
734 	memset(&event, 0, sizeof(event));
735 	if (qp->qplib_qp.srq) {
736 		event.device = &qp->rdev->ibdev;
737 		event.element.qp = &qp->ib_qp;
738 		event.event = IB_EVENT_QP_LAST_WQE_REACHED;
739 	}
740 
741 	if (event.device && qp->ib_qp.event_handler)
742 		qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
743 
744 	return 0;
745 }
746 
747 static int bnxt_re_handle_affi_async_event(struct creq_qp_event *affi_async,
748 					   void *obj)
749 {
750 	int rc = 0;
751 	u8 event;
752 
753 	if (!obj)
754 		return rc; /* QP was already dead, still return success */
755 
756 	event = affi_async->event;
757 	if (event == CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION) {
758 		struct bnxt_qplib_qp *lib_qp = obj;
759 		struct bnxt_re_qp *qp = container_of(lib_qp, struct bnxt_re_qp,
760 						     qplib_qp);
761 		rc = bnxt_re_handle_qp_async_event(affi_async, qp);
762 	}
763 	return rc;
764 }
765 
766 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
767 			       void *aeqe, void *obj)
768 {
769 	struct creq_qp_event *affi_async;
770 	struct creq_func_event *unaffi_async;
771 	u8 type;
772 	int rc;
773 
774 	type = ((struct creq_base *)aeqe)->type;
775 	if (type == CREQ_BASE_TYPE_FUNC_EVENT) {
776 		unaffi_async = aeqe;
777 		rc = bnxt_re_handle_unaffi_async_event(unaffi_async);
778 	} else {
779 		affi_async = aeqe;
780 		rc = bnxt_re_handle_affi_async_event(affi_async, obj);
781 	}
782 
783 	return rc;
784 }
785 
786 static int bnxt_re_srqn_handler(struct bnxt_qplib_nq *nq,
787 				struct bnxt_qplib_srq *handle, u8 event)
788 {
789 	struct bnxt_re_srq *srq = container_of(handle, struct bnxt_re_srq,
790 					       qplib_srq);
791 	struct ib_event ib_event;
792 	int rc = 0;
793 
794 	if (!srq) {
795 		dev_err(NULL, "%s: SRQ is NULL, SRQN not handled",
796 			ROCE_DRV_MODULE_NAME);
797 		rc = -EINVAL;
798 		goto done;
799 	}
800 	ib_event.device = &srq->rdev->ibdev;
801 	ib_event.element.srq = &srq->ib_srq;
802 	if (event == NQ_SRQ_EVENT_EVENT_SRQ_THRESHOLD_EVENT)
803 		ib_event.event = IB_EVENT_SRQ_LIMIT_REACHED;
804 	else
805 		ib_event.event = IB_EVENT_SRQ_ERR;
806 
807 	if (srq->ib_srq.event_handler) {
808 		/* Lock event_handler? */
809 		(*srq->ib_srq.event_handler)(&ib_event,
810 					     srq->ib_srq.srq_context);
811 	}
812 done:
813 	return rc;
814 }
815 
816 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
817 			       struct bnxt_qplib_cq *handle)
818 {
819 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
820 					     qplib_cq);
821 
822 	if (!cq) {
823 		dev_err(NULL, "%s: CQ is NULL, CQN not handled",
824 			ROCE_DRV_MODULE_NAME);
825 		return -EINVAL;
826 	}
827 	if (cq->ib_cq.comp_handler) {
828 		/* Lock comp_handler? */
829 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
830 	}
831 
832 	return 0;
833 }
834 
835 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
836 {
837 	int i;
838 
839 	if (rdev->nq[0].hwq.max_elements) {
840 		for (i = 1; i < rdev->num_msix; i++)
841 			bnxt_qplib_disable_nq(&rdev->nq[i - 1]);
842 	}
843 
844 	if (rdev->qplib_res.rcfw)
845 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
846 }
847 
848 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
849 {
850 	int rc = 0, i;
851 
852 	bnxt_qplib_init_res(&rdev->qplib_res);
853 
854 	for (i = 1; i < rdev->num_msix ; i++) {
855 		rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
856 					  i - 1, rdev->msix_entries[i].vector,
857 					  rdev->msix_entries[i].db_offset,
858 					  &bnxt_re_cqn_handler,
859 					  &bnxt_re_srqn_handler);
860 
861 		if (rc) {
862 			dev_err(rdev_to_dev(rdev),
863 				"Failed to enable NQ with rc = 0x%x", rc);
864 			goto fail;
865 		}
866 	}
867 	return 0;
868 fail:
869 	return rc;
870 }
871 
872 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev, bool lock_wait)
873 {
874 	int i;
875 
876 	for (i = 0; i < rdev->num_msix - 1; i++) {
877 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, lock_wait);
878 		bnxt_qplib_free_nq(&rdev->nq[i]);
879 	}
880 }
881 
882 static void bnxt_re_free_res(struct bnxt_re_dev *rdev, bool lock_wait)
883 {
884 	bnxt_re_free_nq_res(rdev, lock_wait);
885 
886 	if (rdev->qplib_res.dpi_tbl.max) {
887 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
888 				       &rdev->qplib_res.dpi_tbl,
889 				       &rdev->dpi_privileged);
890 	}
891 	if (rdev->qplib_res.rcfw) {
892 		bnxt_qplib_free_res(&rdev->qplib_res);
893 		rdev->qplib_res.rcfw = NULL;
894 	}
895 }
896 
897 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
898 {
899 	int rc = 0, i;
900 
901 	/* Configure and allocate resources for qplib */
902 	rdev->qplib_res.rcfw = &rdev->rcfw;
903 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
904 				     rdev->is_virtfn);
905 	if (rc)
906 		goto fail;
907 
908 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
909 				  rdev->netdev, &rdev->dev_attr);
910 	if (rc)
911 		goto fail;
912 
913 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
914 				  &rdev->dpi_privileged,
915 				  rdev);
916 	if (rc)
917 		goto dealloc_res;
918 
919 	for (i = 0; i < rdev->num_msix - 1; i++) {
920 		rdev->nq[i].hwq.max_elements = BNXT_RE_MAX_CQ_COUNT +
921 			BNXT_RE_MAX_SRQC_COUNT + 2;
922 		rc = bnxt_qplib_alloc_nq(rdev->en_dev->pdev, &rdev->nq[i]);
923 		if (rc) {
924 			dev_err(rdev_to_dev(rdev), "Alloc Failed NQ%d rc:%#x",
925 				i, rc);
926 			goto dealloc_dpi;
927 		}
928 		rc = bnxt_re_net_ring_alloc
929 			(rdev, rdev->nq[i].hwq.pbl[PBL_LVL_0].pg_map_arr,
930 			 rdev->nq[i].hwq.pbl[rdev->nq[i].hwq.level].pg_count,
931 			 HWRM_RING_ALLOC_CMPL,
932 			 BNXT_QPLIB_NQE_MAX_CNT - 1,
933 			 rdev->msix_entries[i + 1].ring_idx,
934 			 &rdev->nq[i].ring_id);
935 		if (rc) {
936 			dev_err(rdev_to_dev(rdev),
937 				"Failed to allocate NQ fw id with rc = 0x%x",
938 				rc);
939 			goto free_nq;
940 		}
941 	}
942 	return 0;
943 free_nq:
944 	for (i = 0; i < rdev->num_msix - 1; i++)
945 		bnxt_qplib_free_nq(&rdev->nq[i]);
946 dealloc_dpi:
947 	bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
948 			       &rdev->qplib_res.dpi_tbl,
949 			       &rdev->dpi_privileged);
950 dealloc_res:
951 	bnxt_qplib_free_res(&rdev->qplib_res);
952 
953 fail:
954 	rdev->qplib_res.rcfw = NULL;
955 	return rc;
956 }
957 
958 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
959 				   u8 port_num, enum ib_event_type event)
960 {
961 	struct ib_event ib_event;
962 
963 	ib_event.device = ibdev;
964 	if (qp)
965 		ib_event.element.qp = qp;
966 	else
967 		ib_event.element.port_num = port_num;
968 	ib_event.event = event;
969 	ib_dispatch_event(&ib_event);
970 }
971 
972 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN      0x02
973 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
974 				      u64 *cid_map)
975 {
976 	struct hwrm_queue_pri2cos_qcfg_input req = {0};
977 	struct bnxt *bp = netdev_priv(rdev->netdev);
978 	struct hwrm_queue_pri2cos_qcfg_output resp;
979 	struct bnxt_en_dev *en_dev = rdev->en_dev;
980 	struct bnxt_fw_msg fw_msg;
981 	u32 flags = 0;
982 	u8 *qcfgmap, *tmp_map;
983 	int rc = 0, i;
984 
985 	if (!cid_map)
986 		return -EINVAL;
987 
988 	memset(&fw_msg, 0, sizeof(fw_msg));
989 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
990 			      HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
991 	flags |= (dir & 0x01);
992 	flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
993 	req.flags = cpu_to_le32(flags);
994 	req.port_id = bp->pf.port_id;
995 
996 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
997 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
998 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
999 	if (rc)
1000 		return rc;
1001 
1002 	if (resp.queue_cfg_info) {
1003 		dev_warn(rdev_to_dev(rdev),
1004 			 "Asymmetric cos queue configuration detected");
1005 		dev_warn(rdev_to_dev(rdev),
1006 			 " on device, QoS may not be fully functional\n");
1007 	}
1008 	qcfgmap = &resp.pri0_cos_queue_id;
1009 	tmp_map = (u8 *)cid_map;
1010 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
1011 		tmp_map[i] = qcfgmap[i];
1012 
1013 	return rc;
1014 }
1015 
1016 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
1017 					struct bnxt_re_qp *qp)
1018 {
1019 	return (qp->ib_qp.qp_type == IB_QPT_GSI) || (qp == rdev->qp1_sqp);
1020 }
1021 
1022 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
1023 {
1024 	int mask = IB_QP_STATE;
1025 	struct ib_qp_attr qp_attr;
1026 	struct bnxt_re_qp *qp;
1027 
1028 	qp_attr.qp_state = IB_QPS_ERR;
1029 	mutex_lock(&rdev->qp_lock);
1030 	list_for_each_entry(qp, &rdev->qp_list, list) {
1031 		/* Modify the state of all QPs except QP1/Shadow QP */
1032 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
1033 			if (qp->qplib_qp.state !=
1034 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
1035 			    qp->qplib_qp.state !=
1036 			    CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1037 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
1038 						       1, IB_EVENT_QP_FATAL);
1039 				bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
1040 						  NULL);
1041 			}
1042 		}
1043 	}
1044 	mutex_unlock(&rdev->qp_lock);
1045 }
1046 
1047 static int bnxt_re_update_gid(struct bnxt_re_dev *rdev)
1048 {
1049 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
1050 	struct bnxt_qplib_gid gid;
1051 	u16 gid_idx, index;
1052 	int rc = 0;
1053 
1054 	if (!test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1055 		return 0;
1056 
1057 	if (!sgid_tbl) {
1058 		dev_err(rdev_to_dev(rdev), "QPLIB: SGID table not allocated");
1059 		return -EINVAL;
1060 	}
1061 
1062 	for (index = 0; index < sgid_tbl->active; index++) {
1063 		gid_idx = sgid_tbl->hw_id[index];
1064 
1065 		if (!memcmp(&sgid_tbl->tbl[index], &bnxt_qplib_gid_zero,
1066 			    sizeof(bnxt_qplib_gid_zero)))
1067 			continue;
1068 		/* need to modify the VLAN enable setting of non VLAN GID only
1069 		 * as setting is done for VLAN GID while adding GID
1070 		 */
1071 		if (sgid_tbl->vlan[index])
1072 			continue;
1073 
1074 		memcpy(&gid, &sgid_tbl->tbl[index], sizeof(gid));
1075 
1076 		rc = bnxt_qplib_update_sgid(sgid_tbl, &gid, gid_idx,
1077 					    rdev->qplib_res.netdev->dev_addr);
1078 	}
1079 
1080 	return rc;
1081 }
1082 
1083 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
1084 {
1085 	u32 prio_map = 0, tmp_map = 0;
1086 	struct net_device *netdev;
1087 	struct dcb_app app;
1088 
1089 	netdev = rdev->netdev;
1090 
1091 	memset(&app, 0, sizeof(app));
1092 	app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
1093 	app.protocol = ETH_P_IBOE;
1094 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1095 	prio_map = tmp_map;
1096 
1097 	app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
1098 	app.protocol = ROCE_V2_UDP_DPORT;
1099 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1100 	prio_map |= tmp_map;
1101 
1102 	return prio_map;
1103 }
1104 
1105 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
1106 {
1107 	u16 prio;
1108 	u8 id;
1109 
1110 	for (prio = 0, id = 0; prio < 8; prio++) {
1111 		if (prio_map & (1 << prio)) {
1112 			cosq[id] = cid_map[prio];
1113 			id++;
1114 			if (id == 2) /* Max 2 tcs supported */
1115 				break;
1116 		}
1117 	}
1118 }
1119 
1120 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
1121 {
1122 	u8 prio_map = 0;
1123 	u64 cid_map;
1124 	int rc;
1125 
1126 	/* Get priority for roce */
1127 	prio_map = bnxt_re_get_priority_mask(rdev);
1128 
1129 	if (prio_map == rdev->cur_prio_map)
1130 		return 0;
1131 	rdev->cur_prio_map = prio_map;
1132 	/* Get cosq id for this priority */
1133 	rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
1134 	if (rc) {
1135 		dev_warn(rdev_to_dev(rdev), "no cos for p_mask %x\n", prio_map);
1136 		return rc;
1137 	}
1138 	/* Parse CoS IDs for app priority */
1139 	bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
1140 
1141 	/* Config BONO. */
1142 	rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
1143 	if (rc) {
1144 		dev_warn(rdev_to_dev(rdev), "no tc for cos{%x, %x}\n",
1145 			 rdev->cosq[0], rdev->cosq[1]);
1146 		return rc;
1147 	}
1148 
1149 	/* Actual priorities are not programmed as they are already
1150 	 * done by L2 driver; just enable or disable priority vlan tagging
1151 	 */
1152 	if ((prio_map == 0 && rdev->qplib_res.prio) ||
1153 	    (prio_map != 0 && !rdev->qplib_res.prio)) {
1154 		rdev->qplib_res.prio = prio_map ? true : false;
1155 
1156 		bnxt_re_update_gid(rdev);
1157 	}
1158 
1159 	return 0;
1160 }
1161 
1162 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev, bool lock_wait)
1163 {
1164 	int i, rc;
1165 
1166 	if (test_and_clear_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags)) {
1167 		for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++)
1168 			device_remove_file(&rdev->ibdev.dev,
1169 					   bnxt_re_attributes[i]);
1170 		/* Cleanup ib dev */
1171 		bnxt_re_unregister_ib(rdev);
1172 	}
1173 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
1174 		cancel_delayed_work(&rdev->worker);
1175 
1176 	bnxt_re_cleanup_res(rdev);
1177 	bnxt_re_free_res(rdev, lock_wait);
1178 
1179 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
1180 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
1181 		if (rc)
1182 			dev_warn(rdev_to_dev(rdev),
1183 				 "Failed to deinitialize RCFW: %#x", rc);
1184 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id,
1185 					   lock_wait);
1186 		bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1187 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1188 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, lock_wait);
1189 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1190 	}
1191 	if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
1192 		rc = bnxt_re_free_msix(rdev, lock_wait);
1193 		if (rc)
1194 			dev_warn(rdev_to_dev(rdev),
1195 				 "Failed to free MSI-X vectors: %#x", rc);
1196 	}
1197 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
1198 		rc = bnxt_re_unregister_netdev(rdev, lock_wait);
1199 		if (rc)
1200 			dev_warn(rdev_to_dev(rdev),
1201 				 "Failed to unregister with netdev: %#x", rc);
1202 	}
1203 }
1204 
1205 /* worker thread for polling periodic events. Now used for QoS programming*/
1206 static void bnxt_re_worker(struct work_struct *work)
1207 {
1208 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
1209 						worker.work);
1210 
1211 	bnxt_re_setup_qos(rdev);
1212 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1213 }
1214 
1215 static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
1216 {
1217 	int i, j, rc;
1218 
1219 	/* Registered a new RoCE device instance to netdev */
1220 	rc = bnxt_re_register_netdev(rdev);
1221 	if (rc) {
1222 		pr_err("Failed to register with netedev: %#x\n", rc);
1223 		return -EINVAL;
1224 	}
1225 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1226 
1227 	/* Check whether VF or PF */
1228 	bnxt_re_get_sriov_func_type(rdev);
1229 
1230 	rc = bnxt_re_request_msix(rdev);
1231 	if (rc) {
1232 		pr_err("Failed to get MSI-X vectors: %#x\n", rc);
1233 		rc = -EINVAL;
1234 		goto fail;
1235 	}
1236 	set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
1237 
1238 	/* Establish RCFW Communication Channel to initialize the context
1239 	 * memory for the function and all child VFs
1240 	 */
1241 	rc = bnxt_qplib_alloc_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw,
1242 					   BNXT_RE_MAX_QPC_COUNT);
1243 	if (rc) {
1244 		pr_err("Failed to allocate RCFW Channel: %#x\n", rc);
1245 		goto fail;
1246 	}
1247 	rc = bnxt_re_net_ring_alloc
1248 			(rdev, rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr,
1249 			 rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count,
1250 			 HWRM_RING_ALLOC_CMPL, BNXT_QPLIB_CREQE_MAX_CNT - 1,
1251 			 rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx,
1252 			 &rdev->rcfw.creq_ring_id);
1253 	if (rc) {
1254 		pr_err("Failed to allocate CREQ: %#x\n", rc);
1255 		goto free_rcfw;
1256 	}
1257 	rc = bnxt_qplib_enable_rcfw_channel
1258 				(rdev->en_dev->pdev, &rdev->rcfw,
1259 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].vector,
1260 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].db_offset,
1261 				 rdev->is_virtfn, &bnxt_re_aeq_handler);
1262 	if (rc) {
1263 		pr_err("Failed to enable RCFW channel: %#x\n", rc);
1264 		goto free_ring;
1265 	}
1266 
1267 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
1268 				     rdev->is_virtfn);
1269 	if (rc)
1270 		goto disable_rcfw;
1271 	if (!rdev->is_virtfn)
1272 		bnxt_re_set_resource_limits(rdev);
1273 
1274 	rc = bnxt_qplib_alloc_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx, 0);
1275 	if (rc) {
1276 		pr_err("Failed to allocate QPLIB context: %#x\n", rc);
1277 		goto disable_rcfw;
1278 	}
1279 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
1280 					 rdev->qplib_ctx.stats.dma_map,
1281 					 &rdev->qplib_ctx.stats.fw_id);
1282 	if (rc) {
1283 		pr_err("Failed to allocate stats context: %#x\n", rc);
1284 		goto free_ctx;
1285 	}
1286 
1287 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx,
1288 				  rdev->is_virtfn);
1289 	if (rc) {
1290 		pr_err("Failed to initialize RCFW: %#x\n", rc);
1291 		goto free_sctx;
1292 	}
1293 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1294 
1295 	/* Resources based on the 'new' device caps */
1296 	rc = bnxt_re_alloc_res(rdev);
1297 	if (rc) {
1298 		pr_err("Failed to allocate resources: %#x\n", rc);
1299 		goto fail;
1300 	}
1301 	rc = bnxt_re_init_res(rdev);
1302 	if (rc) {
1303 		pr_err("Failed to initialize resources: %#x\n", rc);
1304 		goto fail;
1305 	}
1306 
1307 	if (!rdev->is_virtfn) {
1308 		rc = bnxt_re_setup_qos(rdev);
1309 		if (rc)
1310 			pr_info("RoCE priority not yet configured\n");
1311 
1312 		INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1313 		set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1314 		schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1315 	}
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 		goto fail;
1322 	}
1323 	dev_info(rdev_to_dev(rdev), "Device registered successfully");
1324 	for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++) {
1325 		rc = device_create_file(&rdev->ibdev.dev,
1326 					bnxt_re_attributes[i]);
1327 		if (rc) {
1328 			dev_err(rdev_to_dev(rdev),
1329 				"Failed to create IB sysfs: %#x", rc);
1330 			/* Must clean up all created device files */
1331 			for (j = 0; j < i; j++)
1332 				device_remove_file(&rdev->ibdev.dev,
1333 						   bnxt_re_attributes[j]);
1334 			bnxt_re_unregister_ib(rdev);
1335 			goto fail;
1336 		}
1337 	}
1338 	set_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags);
1339 	ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1340 			 &rdev->active_width);
1341 	set_bit(BNXT_RE_FLAG_ISSUE_ROCE_STATS, &rdev->flags);
1342 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_PORT_ACTIVE);
1343 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_GID_CHANGE);
1344 
1345 	return 0;
1346 free_sctx:
1347 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id, true);
1348 free_ctx:
1349 	bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1350 disable_rcfw:
1351 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1352 free_ring:
1353 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, true);
1354 free_rcfw:
1355 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1356 fail:
1357 	bnxt_re_ib_unreg(rdev, true);
1358 	return rc;
1359 }
1360 
1361 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1362 {
1363 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1364 	struct net_device *netdev = rdev->netdev;
1365 
1366 	bnxt_re_dev_remove(rdev);
1367 
1368 	if (netdev)
1369 		bnxt_re_dev_unprobe(netdev, en_dev);
1370 }
1371 
1372 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1373 {
1374 	struct bnxt_en_dev *en_dev;
1375 	int rc = 0;
1376 
1377 	if (!is_bnxt_re_dev(netdev))
1378 		return -ENODEV;
1379 
1380 	en_dev = bnxt_re_dev_probe(netdev);
1381 	if (IS_ERR(en_dev)) {
1382 		if (en_dev != ERR_PTR(-ENODEV))
1383 			pr_err("%s: Failed to probe\n", ROCE_DRV_MODULE_NAME);
1384 		rc = PTR_ERR(en_dev);
1385 		goto exit;
1386 	}
1387 	*rdev = bnxt_re_dev_add(netdev, en_dev);
1388 	if (!*rdev) {
1389 		rc = -ENOMEM;
1390 		bnxt_re_dev_unprobe(netdev, en_dev);
1391 		goto exit;
1392 	}
1393 exit:
1394 	return rc;
1395 }
1396 
1397 static void bnxt_re_remove_one(struct bnxt_re_dev *rdev)
1398 {
1399 	pci_dev_put(rdev->en_dev->pdev);
1400 }
1401 
1402 /* Handle all deferred netevents tasks */
1403 static void bnxt_re_task(struct work_struct *work)
1404 {
1405 	struct bnxt_re_work *re_work;
1406 	struct bnxt_re_dev *rdev;
1407 	int rc = 0;
1408 
1409 	re_work = container_of(work, struct bnxt_re_work, work);
1410 	rdev = re_work->rdev;
1411 
1412 	if (re_work->event != NETDEV_REGISTER &&
1413 	    !test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1414 		return;
1415 
1416 	switch (re_work->event) {
1417 	case NETDEV_REGISTER:
1418 		rc = bnxt_re_ib_reg(rdev);
1419 		if (rc)
1420 			dev_err(rdev_to_dev(rdev),
1421 				"Failed to register with IB: %#x", rc);
1422 		break;
1423 	case NETDEV_UP:
1424 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1425 				       IB_EVENT_PORT_ACTIVE);
1426 		break;
1427 	case NETDEV_DOWN:
1428 		bnxt_re_dev_stop(rdev);
1429 		break;
1430 	case NETDEV_CHANGE:
1431 		if (!netif_carrier_ok(rdev->netdev))
1432 			bnxt_re_dev_stop(rdev);
1433 		else if (netif_carrier_ok(rdev->netdev))
1434 			bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1435 					       IB_EVENT_PORT_ACTIVE);
1436 		ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1437 				 &rdev->active_width);
1438 		break;
1439 	default:
1440 		break;
1441 	}
1442 	smp_mb__before_atomic();
1443 	atomic_dec(&rdev->sched_count);
1444 	kfree(re_work);
1445 }
1446 
1447 static void bnxt_re_init_one(struct bnxt_re_dev *rdev)
1448 {
1449 	pci_dev_get(rdev->en_dev->pdev);
1450 }
1451 
1452 /*
1453  * "Notifier chain callback can be invoked for the same chain from
1454  * different CPUs at the same time".
1455  *
1456  * For cases when the netdev is already present, our call to the
1457  * register_netdevice_notifier() will actually get the rtnl_lock()
1458  * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1459  * events.
1460  *
1461  * But for cases when the netdev is not already present, the notifier
1462  * chain is subjected to be invoked from different CPUs simultaneously.
1463  *
1464  * This is protected by the netdev_mutex.
1465  */
1466 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1467 				unsigned long event, void *ptr)
1468 {
1469 	struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1470 	struct bnxt_re_work *re_work;
1471 	struct bnxt_re_dev *rdev;
1472 	int rc = 0;
1473 	bool sch_work = false;
1474 
1475 	real_dev = rdma_vlan_dev_real_dev(netdev);
1476 	if (!real_dev)
1477 		real_dev = netdev;
1478 
1479 	rdev = bnxt_re_from_netdev(real_dev);
1480 	if (!rdev && event != NETDEV_REGISTER)
1481 		goto exit;
1482 	if (real_dev != netdev)
1483 		goto exit;
1484 
1485 	switch (event) {
1486 	case NETDEV_REGISTER:
1487 		if (rdev)
1488 			break;
1489 		rc = bnxt_re_dev_reg(&rdev, real_dev);
1490 		if (rc == -ENODEV)
1491 			break;
1492 		if (rc) {
1493 			pr_err("Failed to register with the device %s: %#x\n",
1494 			       real_dev->name, rc);
1495 			break;
1496 		}
1497 		bnxt_re_init_one(rdev);
1498 		sch_work = true;
1499 		break;
1500 
1501 	case NETDEV_UNREGISTER:
1502 		/* netdev notifier will call NETDEV_UNREGISTER again later since
1503 		 * we are still holding the reference to the netdev
1504 		 */
1505 		if (atomic_read(&rdev->sched_count) > 0)
1506 			goto exit;
1507 		bnxt_re_ib_unreg(rdev, false);
1508 		bnxt_re_remove_one(rdev);
1509 		bnxt_re_dev_unreg(rdev);
1510 		break;
1511 
1512 	default:
1513 		sch_work = true;
1514 		break;
1515 	}
1516 	if (sch_work) {
1517 		/* Allocate for the deferred task */
1518 		re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1519 		if (re_work) {
1520 			re_work->rdev = rdev;
1521 			re_work->event = event;
1522 			re_work->vlan_dev = (real_dev == netdev ?
1523 					     NULL : netdev);
1524 			INIT_WORK(&re_work->work, bnxt_re_task);
1525 			atomic_inc(&rdev->sched_count);
1526 			queue_work(bnxt_re_wq, &re_work->work);
1527 		}
1528 	}
1529 
1530 exit:
1531 	return NOTIFY_DONE;
1532 }
1533 
1534 static struct notifier_block bnxt_re_netdev_notifier = {
1535 	.notifier_call = bnxt_re_netdev_event
1536 };
1537 
1538 static int __init bnxt_re_mod_init(void)
1539 {
1540 	int rc = 0;
1541 
1542 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1543 
1544 	bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1545 	if (!bnxt_re_wq)
1546 		return -ENOMEM;
1547 
1548 	INIT_LIST_HEAD(&bnxt_re_dev_list);
1549 
1550 	rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1551 	if (rc) {
1552 		pr_err("%s: Cannot register to netdevice_notifier",
1553 		       ROCE_DRV_MODULE_NAME);
1554 		goto err_netdev;
1555 	}
1556 	return 0;
1557 
1558 err_netdev:
1559 	destroy_workqueue(bnxt_re_wq);
1560 
1561 	return rc;
1562 }
1563 
1564 static void __exit bnxt_re_mod_exit(void)
1565 {
1566 	struct bnxt_re_dev *rdev, *next;
1567 	LIST_HEAD(to_be_deleted);
1568 
1569 	mutex_lock(&bnxt_re_dev_lock);
1570 	/* Free all adapter allocated resources */
1571 	if (!list_empty(&bnxt_re_dev_list))
1572 		list_splice_init(&bnxt_re_dev_list, &to_be_deleted);
1573 	mutex_unlock(&bnxt_re_dev_lock);
1574        /*
1575 	* Cleanup the devices in reverse order so that the VF device
1576 	* cleanup is done before PF cleanup
1577 	*/
1578 	list_for_each_entry_safe_reverse(rdev, next, &to_be_deleted, list) {
1579 		dev_info(rdev_to_dev(rdev), "Unregistering Device");
1580 		/*
1581 		 * Flush out any scheduled tasks before destroying the
1582 		 * resources
1583 		 */
1584 		flush_workqueue(bnxt_re_wq);
1585 		bnxt_re_dev_stop(rdev);
1586 		bnxt_re_ib_unreg(rdev, true);
1587 		bnxt_re_remove_one(rdev);
1588 		bnxt_re_dev_unreg(rdev);
1589 	}
1590 	unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1591 	if (bnxt_re_wq)
1592 		destroy_workqueue(bnxt_re_wq);
1593 }
1594 
1595 module_init(bnxt_re_mod_init);
1596 module_exit(bnxt_re_mod_exit);
1597