xref: /openbmc/linux/drivers/infiniband/hw/bnxt_re/main.c (revision 5ef12cb4a3a78ffb331c03a795a15eea4ae35155)
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->get_netdev		= bnxt_re_get_netdev;
578 	ibdev->add_gid			= bnxt_re_add_gid;
579 	ibdev->del_gid			= bnxt_re_del_gid;
580 	ibdev->get_link_layer		= bnxt_re_get_link_layer;
581 
582 	ibdev->alloc_pd			= bnxt_re_alloc_pd;
583 	ibdev->dealloc_pd		= bnxt_re_dealloc_pd;
584 
585 	ibdev->create_ah		= bnxt_re_create_ah;
586 	ibdev->modify_ah		= bnxt_re_modify_ah;
587 	ibdev->query_ah			= bnxt_re_query_ah;
588 	ibdev->destroy_ah		= bnxt_re_destroy_ah;
589 
590 	ibdev->create_srq		= bnxt_re_create_srq;
591 	ibdev->modify_srq		= bnxt_re_modify_srq;
592 	ibdev->query_srq		= bnxt_re_query_srq;
593 	ibdev->destroy_srq		= bnxt_re_destroy_srq;
594 	ibdev->post_srq_recv		= bnxt_re_post_srq_recv;
595 
596 	ibdev->create_qp		= bnxt_re_create_qp;
597 	ibdev->modify_qp		= bnxt_re_modify_qp;
598 	ibdev->query_qp			= bnxt_re_query_qp;
599 	ibdev->destroy_qp		= bnxt_re_destroy_qp;
600 
601 	ibdev->post_send		= bnxt_re_post_send;
602 	ibdev->post_recv		= bnxt_re_post_recv;
603 
604 	ibdev->create_cq		= bnxt_re_create_cq;
605 	ibdev->destroy_cq		= bnxt_re_destroy_cq;
606 	ibdev->poll_cq			= bnxt_re_poll_cq;
607 	ibdev->req_notify_cq		= bnxt_re_req_notify_cq;
608 
609 	ibdev->get_dma_mr		= bnxt_re_get_dma_mr;
610 	ibdev->dereg_mr			= bnxt_re_dereg_mr;
611 	ibdev->alloc_mr			= bnxt_re_alloc_mr;
612 	ibdev->map_mr_sg		= bnxt_re_map_mr_sg;
613 
614 	ibdev->reg_user_mr		= bnxt_re_reg_user_mr;
615 	ibdev->alloc_ucontext		= bnxt_re_alloc_ucontext;
616 	ibdev->dealloc_ucontext		= bnxt_re_dealloc_ucontext;
617 	ibdev->mmap			= bnxt_re_mmap;
618 	ibdev->get_hw_stats             = bnxt_re_ib_get_hw_stats;
619 	ibdev->alloc_hw_stats           = bnxt_re_ib_alloc_hw_stats;
620 
621 	ibdev->driver_id = RDMA_DRIVER_BNXT_RE;
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 	unsigned int flags;
734 
735 	if (qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
736 		flags = bnxt_re_lock_cqs(qp);
737 		bnxt_qplib_add_flush_qp(&qp->qplib_qp);
738 		bnxt_re_unlock_cqs(qp, flags);
739 	}
740 
741 	memset(&event, 0, sizeof(event));
742 	if (qp->qplib_qp.srq) {
743 		event.device = &qp->rdev->ibdev;
744 		event.element.qp = &qp->ib_qp;
745 		event.event = IB_EVENT_QP_LAST_WQE_REACHED;
746 	}
747 
748 	if (event.device && qp->ib_qp.event_handler)
749 		qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
750 
751 	return 0;
752 }
753 
754 static int bnxt_re_handle_affi_async_event(struct creq_qp_event *affi_async,
755 					   void *obj)
756 {
757 	int rc = 0;
758 	u8 event;
759 
760 	if (!obj)
761 		return rc; /* QP was already dead, still return success */
762 
763 	event = affi_async->event;
764 	if (event == CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION) {
765 		struct bnxt_qplib_qp *lib_qp = obj;
766 		struct bnxt_re_qp *qp = container_of(lib_qp, struct bnxt_re_qp,
767 						     qplib_qp);
768 		rc = bnxt_re_handle_qp_async_event(affi_async, qp);
769 	}
770 	return rc;
771 }
772 
773 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
774 			       void *aeqe, void *obj)
775 {
776 	struct creq_qp_event *affi_async;
777 	struct creq_func_event *unaffi_async;
778 	u8 type;
779 	int rc;
780 
781 	type = ((struct creq_base *)aeqe)->type;
782 	if (type == CREQ_BASE_TYPE_FUNC_EVENT) {
783 		unaffi_async = aeqe;
784 		rc = bnxt_re_handle_unaffi_async_event(unaffi_async);
785 	} else {
786 		affi_async = aeqe;
787 		rc = bnxt_re_handle_affi_async_event(affi_async, obj);
788 	}
789 
790 	return rc;
791 }
792 
793 static int bnxt_re_srqn_handler(struct bnxt_qplib_nq *nq,
794 				struct bnxt_qplib_srq *handle, u8 event)
795 {
796 	struct bnxt_re_srq *srq = container_of(handle, struct bnxt_re_srq,
797 					       qplib_srq);
798 	struct ib_event ib_event;
799 	int rc = 0;
800 
801 	if (!srq) {
802 		dev_err(NULL, "%s: SRQ is NULL, SRQN not handled",
803 			ROCE_DRV_MODULE_NAME);
804 		rc = -EINVAL;
805 		goto done;
806 	}
807 	ib_event.device = &srq->rdev->ibdev;
808 	ib_event.element.srq = &srq->ib_srq;
809 	if (event == NQ_SRQ_EVENT_EVENT_SRQ_THRESHOLD_EVENT)
810 		ib_event.event = IB_EVENT_SRQ_LIMIT_REACHED;
811 	else
812 		ib_event.event = IB_EVENT_SRQ_ERR;
813 
814 	if (srq->ib_srq.event_handler) {
815 		/* Lock event_handler? */
816 		(*srq->ib_srq.event_handler)(&ib_event,
817 					     srq->ib_srq.srq_context);
818 	}
819 done:
820 	return rc;
821 }
822 
823 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
824 			       struct bnxt_qplib_cq *handle)
825 {
826 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
827 					     qplib_cq);
828 
829 	if (!cq) {
830 		dev_err(NULL, "%s: CQ is NULL, CQN not handled",
831 			ROCE_DRV_MODULE_NAME);
832 		return -EINVAL;
833 	}
834 	if (cq->ib_cq.comp_handler) {
835 		/* Lock comp_handler? */
836 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
837 	}
838 
839 	return 0;
840 }
841 
842 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
843 {
844 	int i;
845 
846 	if (rdev->nq[0].hwq.max_elements) {
847 		for (i = 1; i < rdev->num_msix; i++)
848 			bnxt_qplib_disable_nq(&rdev->nq[i - 1]);
849 	}
850 
851 	if (rdev->qplib_res.rcfw)
852 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
853 }
854 
855 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
856 {
857 	int rc = 0, i;
858 
859 	bnxt_qplib_init_res(&rdev->qplib_res);
860 
861 	for (i = 1; i < rdev->num_msix ; i++) {
862 		rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
863 					  i - 1, rdev->msix_entries[i].vector,
864 					  rdev->msix_entries[i].db_offset,
865 					  &bnxt_re_cqn_handler,
866 					  &bnxt_re_srqn_handler);
867 
868 		if (rc) {
869 			dev_err(rdev_to_dev(rdev),
870 				"Failed to enable NQ with rc = 0x%x", rc);
871 			goto fail;
872 		}
873 	}
874 	return 0;
875 fail:
876 	return rc;
877 }
878 
879 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev, bool lock_wait)
880 {
881 	int i;
882 
883 	for (i = 0; i < rdev->num_msix - 1; i++) {
884 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, lock_wait);
885 		bnxt_qplib_free_nq(&rdev->nq[i]);
886 	}
887 }
888 
889 static void bnxt_re_free_res(struct bnxt_re_dev *rdev, bool lock_wait)
890 {
891 	bnxt_re_free_nq_res(rdev, lock_wait);
892 
893 	if (rdev->qplib_res.dpi_tbl.max) {
894 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
895 				       &rdev->qplib_res.dpi_tbl,
896 				       &rdev->dpi_privileged);
897 	}
898 	if (rdev->qplib_res.rcfw) {
899 		bnxt_qplib_free_res(&rdev->qplib_res);
900 		rdev->qplib_res.rcfw = NULL;
901 	}
902 }
903 
904 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
905 {
906 	int rc = 0, i;
907 
908 	/* Configure and allocate resources for qplib */
909 	rdev->qplib_res.rcfw = &rdev->rcfw;
910 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
911 				     rdev->is_virtfn);
912 	if (rc)
913 		goto fail;
914 
915 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
916 				  rdev->netdev, &rdev->dev_attr);
917 	if (rc)
918 		goto fail;
919 
920 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
921 				  &rdev->dpi_privileged,
922 				  rdev);
923 	if (rc)
924 		goto dealloc_res;
925 
926 	for (i = 0; i < rdev->num_msix - 1; i++) {
927 		rdev->nq[i].hwq.max_elements = BNXT_RE_MAX_CQ_COUNT +
928 			BNXT_RE_MAX_SRQC_COUNT + 2;
929 		rc = bnxt_qplib_alloc_nq(rdev->en_dev->pdev, &rdev->nq[i]);
930 		if (rc) {
931 			dev_err(rdev_to_dev(rdev), "Alloc Failed NQ%d rc:%#x",
932 				i, rc);
933 			goto dealloc_dpi;
934 		}
935 		rc = bnxt_re_net_ring_alloc
936 			(rdev, rdev->nq[i].hwq.pbl[PBL_LVL_0].pg_map_arr,
937 			 rdev->nq[i].hwq.pbl[rdev->nq[i].hwq.level].pg_count,
938 			 HWRM_RING_ALLOC_CMPL,
939 			 BNXT_QPLIB_NQE_MAX_CNT - 1,
940 			 rdev->msix_entries[i + 1].ring_idx,
941 			 &rdev->nq[i].ring_id);
942 		if (rc) {
943 			dev_err(rdev_to_dev(rdev),
944 				"Failed to allocate NQ fw id with rc = 0x%x",
945 				rc);
946 			goto free_nq;
947 		}
948 	}
949 	return 0;
950 free_nq:
951 	for (i = 0; i < rdev->num_msix - 1; i++)
952 		bnxt_qplib_free_nq(&rdev->nq[i]);
953 dealloc_dpi:
954 	bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
955 			       &rdev->qplib_res.dpi_tbl,
956 			       &rdev->dpi_privileged);
957 dealloc_res:
958 	bnxt_qplib_free_res(&rdev->qplib_res);
959 
960 fail:
961 	rdev->qplib_res.rcfw = NULL;
962 	return rc;
963 }
964 
965 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
966 				   u8 port_num, enum ib_event_type event)
967 {
968 	struct ib_event ib_event;
969 
970 	ib_event.device = ibdev;
971 	if (qp)
972 		ib_event.element.qp = qp;
973 	else
974 		ib_event.element.port_num = port_num;
975 	ib_event.event = event;
976 	ib_dispatch_event(&ib_event);
977 }
978 
979 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN      0x02
980 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
981 				      u64 *cid_map)
982 {
983 	struct hwrm_queue_pri2cos_qcfg_input req = {0};
984 	struct bnxt *bp = netdev_priv(rdev->netdev);
985 	struct hwrm_queue_pri2cos_qcfg_output resp;
986 	struct bnxt_en_dev *en_dev = rdev->en_dev;
987 	struct bnxt_fw_msg fw_msg;
988 	u32 flags = 0;
989 	u8 *qcfgmap, *tmp_map;
990 	int rc = 0, i;
991 
992 	if (!cid_map)
993 		return -EINVAL;
994 
995 	memset(&fw_msg, 0, sizeof(fw_msg));
996 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
997 			      HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
998 	flags |= (dir & 0x01);
999 	flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
1000 	req.flags = cpu_to_le32(flags);
1001 	req.port_id = bp->pf.port_id;
1002 
1003 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1004 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1005 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1006 	if (rc)
1007 		return rc;
1008 
1009 	if (resp.queue_cfg_info) {
1010 		dev_warn(rdev_to_dev(rdev),
1011 			 "Asymmetric cos queue configuration detected");
1012 		dev_warn(rdev_to_dev(rdev),
1013 			 " on device, QoS may not be fully functional\n");
1014 	}
1015 	qcfgmap = &resp.pri0_cos_queue_id;
1016 	tmp_map = (u8 *)cid_map;
1017 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
1018 		tmp_map[i] = qcfgmap[i];
1019 
1020 	return rc;
1021 }
1022 
1023 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
1024 					struct bnxt_re_qp *qp)
1025 {
1026 	return (qp->ib_qp.qp_type == IB_QPT_GSI) || (qp == rdev->qp1_sqp);
1027 }
1028 
1029 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
1030 {
1031 	int mask = IB_QP_STATE;
1032 	struct ib_qp_attr qp_attr;
1033 	struct bnxt_re_qp *qp;
1034 
1035 	qp_attr.qp_state = IB_QPS_ERR;
1036 	mutex_lock(&rdev->qp_lock);
1037 	list_for_each_entry(qp, &rdev->qp_list, list) {
1038 		/* Modify the state of all QPs except QP1/Shadow QP */
1039 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
1040 			if (qp->qplib_qp.state !=
1041 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
1042 			    qp->qplib_qp.state !=
1043 			    CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1044 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
1045 						       1, IB_EVENT_QP_FATAL);
1046 				bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
1047 						  NULL);
1048 			}
1049 		}
1050 	}
1051 	mutex_unlock(&rdev->qp_lock);
1052 }
1053 
1054 static int bnxt_re_update_gid(struct bnxt_re_dev *rdev)
1055 {
1056 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
1057 	struct bnxt_qplib_gid gid;
1058 	u16 gid_idx, index;
1059 	int rc = 0;
1060 
1061 	if (!test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1062 		return 0;
1063 
1064 	if (!sgid_tbl) {
1065 		dev_err(rdev_to_dev(rdev), "QPLIB: SGID table not allocated");
1066 		return -EINVAL;
1067 	}
1068 
1069 	for (index = 0; index < sgid_tbl->active; index++) {
1070 		gid_idx = sgid_tbl->hw_id[index];
1071 
1072 		if (!memcmp(&sgid_tbl->tbl[index], &bnxt_qplib_gid_zero,
1073 			    sizeof(bnxt_qplib_gid_zero)))
1074 			continue;
1075 		/* need to modify the VLAN enable setting of non VLAN GID only
1076 		 * as setting is done for VLAN GID while adding GID
1077 		 */
1078 		if (sgid_tbl->vlan[index])
1079 			continue;
1080 
1081 		memcpy(&gid, &sgid_tbl->tbl[index], sizeof(gid));
1082 
1083 		rc = bnxt_qplib_update_sgid(sgid_tbl, &gid, gid_idx,
1084 					    rdev->qplib_res.netdev->dev_addr);
1085 	}
1086 
1087 	return rc;
1088 }
1089 
1090 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
1091 {
1092 	u32 prio_map = 0, tmp_map = 0;
1093 	struct net_device *netdev;
1094 	struct dcb_app app;
1095 
1096 	netdev = rdev->netdev;
1097 
1098 	memset(&app, 0, sizeof(app));
1099 	app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
1100 	app.protocol = ETH_P_IBOE;
1101 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1102 	prio_map = tmp_map;
1103 
1104 	app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
1105 	app.protocol = ROCE_V2_UDP_DPORT;
1106 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1107 	prio_map |= tmp_map;
1108 
1109 	return prio_map;
1110 }
1111 
1112 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
1113 {
1114 	u16 prio;
1115 	u8 id;
1116 
1117 	for (prio = 0, id = 0; prio < 8; prio++) {
1118 		if (prio_map & (1 << prio)) {
1119 			cosq[id] = cid_map[prio];
1120 			id++;
1121 			if (id == 2) /* Max 2 tcs supported */
1122 				break;
1123 		}
1124 	}
1125 }
1126 
1127 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
1128 {
1129 	u8 prio_map = 0;
1130 	u64 cid_map;
1131 	int rc;
1132 
1133 	/* Get priority for roce */
1134 	prio_map = bnxt_re_get_priority_mask(rdev);
1135 
1136 	if (prio_map == rdev->cur_prio_map)
1137 		return 0;
1138 	rdev->cur_prio_map = prio_map;
1139 	/* Get cosq id for this priority */
1140 	rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
1141 	if (rc) {
1142 		dev_warn(rdev_to_dev(rdev), "no cos for p_mask %x\n", prio_map);
1143 		return rc;
1144 	}
1145 	/* Parse CoS IDs for app priority */
1146 	bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
1147 
1148 	/* Config BONO. */
1149 	rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
1150 	if (rc) {
1151 		dev_warn(rdev_to_dev(rdev), "no tc for cos{%x, %x}\n",
1152 			 rdev->cosq[0], rdev->cosq[1]);
1153 		return rc;
1154 	}
1155 
1156 	/* Actual priorities are not programmed as they are already
1157 	 * done by L2 driver; just enable or disable priority vlan tagging
1158 	 */
1159 	if ((prio_map == 0 && rdev->qplib_res.prio) ||
1160 	    (prio_map != 0 && !rdev->qplib_res.prio)) {
1161 		rdev->qplib_res.prio = prio_map ? true : false;
1162 
1163 		bnxt_re_update_gid(rdev);
1164 	}
1165 
1166 	return 0;
1167 }
1168 
1169 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev, bool lock_wait)
1170 {
1171 	int i, rc;
1172 
1173 	if (test_and_clear_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags)) {
1174 		for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++)
1175 			device_remove_file(&rdev->ibdev.dev,
1176 					   bnxt_re_attributes[i]);
1177 		/* Cleanup ib dev */
1178 		bnxt_re_unregister_ib(rdev);
1179 	}
1180 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
1181 		cancel_delayed_work(&rdev->worker);
1182 
1183 	bnxt_re_cleanup_res(rdev);
1184 	bnxt_re_free_res(rdev, lock_wait);
1185 
1186 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
1187 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
1188 		if (rc)
1189 			dev_warn(rdev_to_dev(rdev),
1190 				 "Failed to deinitialize RCFW: %#x", rc);
1191 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id,
1192 					   lock_wait);
1193 		bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1194 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1195 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, lock_wait);
1196 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1197 	}
1198 	if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
1199 		rc = bnxt_re_free_msix(rdev, lock_wait);
1200 		if (rc)
1201 			dev_warn(rdev_to_dev(rdev),
1202 				 "Failed to free MSI-X vectors: %#x", rc);
1203 	}
1204 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
1205 		rc = bnxt_re_unregister_netdev(rdev, lock_wait);
1206 		if (rc)
1207 			dev_warn(rdev_to_dev(rdev),
1208 				 "Failed to unregister with netdev: %#x", rc);
1209 	}
1210 }
1211 
1212 /* worker thread for polling periodic events. Now used for QoS programming*/
1213 static void bnxt_re_worker(struct work_struct *work)
1214 {
1215 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
1216 						worker.work);
1217 
1218 	bnxt_re_setup_qos(rdev);
1219 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1220 }
1221 
1222 static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
1223 {
1224 	int i, j, rc;
1225 
1226 	/* Registered a new RoCE device instance to netdev */
1227 	rc = bnxt_re_register_netdev(rdev);
1228 	if (rc) {
1229 		pr_err("Failed to register with netedev: %#x\n", rc);
1230 		return -EINVAL;
1231 	}
1232 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1233 
1234 	/* Check whether VF or PF */
1235 	bnxt_re_get_sriov_func_type(rdev);
1236 
1237 	rc = bnxt_re_request_msix(rdev);
1238 	if (rc) {
1239 		pr_err("Failed to get MSI-X vectors: %#x\n", rc);
1240 		rc = -EINVAL;
1241 		goto fail;
1242 	}
1243 	set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
1244 
1245 	/* Establish RCFW Communication Channel to initialize the context
1246 	 * memory for the function and all child VFs
1247 	 */
1248 	rc = bnxt_qplib_alloc_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw,
1249 					   BNXT_RE_MAX_QPC_COUNT);
1250 	if (rc) {
1251 		pr_err("Failed to allocate RCFW Channel: %#x\n", rc);
1252 		goto fail;
1253 	}
1254 	rc = bnxt_re_net_ring_alloc
1255 			(rdev, rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr,
1256 			 rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count,
1257 			 HWRM_RING_ALLOC_CMPL, BNXT_QPLIB_CREQE_MAX_CNT - 1,
1258 			 rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx,
1259 			 &rdev->rcfw.creq_ring_id);
1260 	if (rc) {
1261 		pr_err("Failed to allocate CREQ: %#x\n", rc);
1262 		goto free_rcfw;
1263 	}
1264 	rc = bnxt_qplib_enable_rcfw_channel
1265 				(rdev->en_dev->pdev, &rdev->rcfw,
1266 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].vector,
1267 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].db_offset,
1268 				 rdev->is_virtfn, &bnxt_re_aeq_handler);
1269 	if (rc) {
1270 		pr_err("Failed to enable RCFW channel: %#x\n", rc);
1271 		goto free_ring;
1272 	}
1273 
1274 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
1275 				     rdev->is_virtfn);
1276 	if (rc)
1277 		goto disable_rcfw;
1278 	if (!rdev->is_virtfn)
1279 		bnxt_re_set_resource_limits(rdev);
1280 
1281 	rc = bnxt_qplib_alloc_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx, 0);
1282 	if (rc) {
1283 		pr_err("Failed to allocate QPLIB context: %#x\n", rc);
1284 		goto disable_rcfw;
1285 	}
1286 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
1287 					 rdev->qplib_ctx.stats.dma_map,
1288 					 &rdev->qplib_ctx.stats.fw_id);
1289 	if (rc) {
1290 		pr_err("Failed to allocate stats context: %#x\n", rc);
1291 		goto free_ctx;
1292 	}
1293 
1294 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx,
1295 				  rdev->is_virtfn);
1296 	if (rc) {
1297 		pr_err("Failed to initialize RCFW: %#x\n", rc);
1298 		goto free_sctx;
1299 	}
1300 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1301 
1302 	/* Resources based on the 'new' device caps */
1303 	rc = bnxt_re_alloc_res(rdev);
1304 	if (rc) {
1305 		pr_err("Failed to allocate resources: %#x\n", rc);
1306 		goto fail;
1307 	}
1308 	rc = bnxt_re_init_res(rdev);
1309 	if (rc) {
1310 		pr_err("Failed to initialize resources: %#x\n", rc);
1311 		goto fail;
1312 	}
1313 
1314 	if (!rdev->is_virtfn) {
1315 		rc = bnxt_re_setup_qos(rdev);
1316 		if (rc)
1317 			pr_info("RoCE priority not yet configured\n");
1318 
1319 		INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1320 		set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1321 		schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1322 	}
1323 
1324 	/* Register ib dev */
1325 	rc = bnxt_re_register_ib(rdev);
1326 	if (rc) {
1327 		pr_err("Failed to register with IB: %#x\n", rc);
1328 		goto fail;
1329 	}
1330 	dev_info(rdev_to_dev(rdev), "Device registered successfully");
1331 	for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++) {
1332 		rc = device_create_file(&rdev->ibdev.dev,
1333 					bnxt_re_attributes[i]);
1334 		if (rc) {
1335 			dev_err(rdev_to_dev(rdev),
1336 				"Failed to create IB sysfs: %#x", rc);
1337 			/* Must clean up all created device files */
1338 			for (j = 0; j < i; j++)
1339 				device_remove_file(&rdev->ibdev.dev,
1340 						   bnxt_re_attributes[j]);
1341 			bnxt_re_unregister_ib(rdev);
1342 			goto fail;
1343 		}
1344 	}
1345 	set_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags);
1346 	ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1347 			 &rdev->active_width);
1348 	set_bit(BNXT_RE_FLAG_ISSUE_ROCE_STATS, &rdev->flags);
1349 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_PORT_ACTIVE);
1350 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_GID_CHANGE);
1351 
1352 	return 0;
1353 free_sctx:
1354 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id, true);
1355 free_ctx:
1356 	bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1357 disable_rcfw:
1358 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1359 free_ring:
1360 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id, true);
1361 free_rcfw:
1362 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1363 fail:
1364 	bnxt_re_ib_unreg(rdev, true);
1365 	return rc;
1366 }
1367 
1368 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1369 {
1370 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1371 	struct net_device *netdev = rdev->netdev;
1372 
1373 	bnxt_re_dev_remove(rdev);
1374 
1375 	if (netdev)
1376 		bnxt_re_dev_unprobe(netdev, en_dev);
1377 }
1378 
1379 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1380 {
1381 	struct bnxt_en_dev *en_dev;
1382 	int rc = 0;
1383 
1384 	if (!is_bnxt_re_dev(netdev))
1385 		return -ENODEV;
1386 
1387 	en_dev = bnxt_re_dev_probe(netdev);
1388 	if (IS_ERR(en_dev)) {
1389 		if (en_dev != ERR_PTR(-ENODEV))
1390 			pr_err("%s: Failed to probe\n", ROCE_DRV_MODULE_NAME);
1391 		rc = PTR_ERR(en_dev);
1392 		goto exit;
1393 	}
1394 	*rdev = bnxt_re_dev_add(netdev, en_dev);
1395 	if (!*rdev) {
1396 		rc = -ENOMEM;
1397 		bnxt_re_dev_unprobe(netdev, en_dev);
1398 		goto exit;
1399 	}
1400 exit:
1401 	return rc;
1402 }
1403 
1404 static void bnxt_re_remove_one(struct bnxt_re_dev *rdev)
1405 {
1406 	pci_dev_put(rdev->en_dev->pdev);
1407 }
1408 
1409 /* Handle all deferred netevents tasks */
1410 static void bnxt_re_task(struct work_struct *work)
1411 {
1412 	struct bnxt_re_work *re_work;
1413 	struct bnxt_re_dev *rdev;
1414 	int rc = 0;
1415 
1416 	re_work = container_of(work, struct bnxt_re_work, work);
1417 	rdev = re_work->rdev;
1418 
1419 	if (re_work->event != NETDEV_REGISTER &&
1420 	    !test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1421 		return;
1422 
1423 	switch (re_work->event) {
1424 	case NETDEV_REGISTER:
1425 		rc = bnxt_re_ib_reg(rdev);
1426 		if (rc) {
1427 			dev_err(rdev_to_dev(rdev),
1428 				"Failed to register with IB: %#x", rc);
1429 			bnxt_re_remove_one(rdev);
1430 			bnxt_re_dev_unreg(rdev);
1431 		}
1432 		break;
1433 	case NETDEV_UP:
1434 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1435 				       IB_EVENT_PORT_ACTIVE);
1436 		break;
1437 	case NETDEV_DOWN:
1438 		bnxt_re_dev_stop(rdev);
1439 		break;
1440 	case NETDEV_CHANGE:
1441 		if (!netif_carrier_ok(rdev->netdev))
1442 			bnxt_re_dev_stop(rdev);
1443 		else if (netif_carrier_ok(rdev->netdev))
1444 			bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1445 					       IB_EVENT_PORT_ACTIVE);
1446 		ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1447 				 &rdev->active_width);
1448 		break;
1449 	default:
1450 		break;
1451 	}
1452 	smp_mb__before_atomic();
1453 	atomic_dec(&rdev->sched_count);
1454 	kfree(re_work);
1455 }
1456 
1457 static void bnxt_re_init_one(struct bnxt_re_dev *rdev)
1458 {
1459 	pci_dev_get(rdev->en_dev->pdev);
1460 }
1461 
1462 /*
1463  * "Notifier chain callback can be invoked for the same chain from
1464  * different CPUs at the same time".
1465  *
1466  * For cases when the netdev is already present, our call to the
1467  * register_netdevice_notifier() will actually get the rtnl_lock()
1468  * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1469  * events.
1470  *
1471  * But for cases when the netdev is not already present, the notifier
1472  * chain is subjected to be invoked from different CPUs simultaneously.
1473  *
1474  * This is protected by the netdev_mutex.
1475  */
1476 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1477 				unsigned long event, void *ptr)
1478 {
1479 	struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1480 	struct bnxt_re_work *re_work;
1481 	struct bnxt_re_dev *rdev;
1482 	int rc = 0;
1483 	bool sch_work = false;
1484 
1485 	real_dev = rdma_vlan_dev_real_dev(netdev);
1486 	if (!real_dev)
1487 		real_dev = netdev;
1488 
1489 	rdev = bnxt_re_from_netdev(real_dev);
1490 	if (!rdev && event != NETDEV_REGISTER)
1491 		goto exit;
1492 	if (real_dev != netdev)
1493 		goto exit;
1494 
1495 	switch (event) {
1496 	case NETDEV_REGISTER:
1497 		if (rdev)
1498 			break;
1499 		rc = bnxt_re_dev_reg(&rdev, real_dev);
1500 		if (rc == -ENODEV)
1501 			break;
1502 		if (rc) {
1503 			pr_err("Failed to register with the device %s: %#x\n",
1504 			       real_dev->name, rc);
1505 			break;
1506 		}
1507 		bnxt_re_init_one(rdev);
1508 		sch_work = true;
1509 		break;
1510 
1511 	case NETDEV_UNREGISTER:
1512 		/* netdev notifier will call NETDEV_UNREGISTER again later since
1513 		 * we are still holding the reference to the netdev
1514 		 */
1515 		if (atomic_read(&rdev->sched_count) > 0)
1516 			goto exit;
1517 		bnxt_re_ib_unreg(rdev, false);
1518 		bnxt_re_remove_one(rdev);
1519 		bnxt_re_dev_unreg(rdev);
1520 		break;
1521 
1522 	default:
1523 		sch_work = true;
1524 		break;
1525 	}
1526 	if (sch_work) {
1527 		/* Allocate for the deferred task */
1528 		re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1529 		if (re_work) {
1530 			re_work->rdev = rdev;
1531 			re_work->event = event;
1532 			re_work->vlan_dev = (real_dev == netdev ?
1533 					     NULL : netdev);
1534 			INIT_WORK(&re_work->work, bnxt_re_task);
1535 			atomic_inc(&rdev->sched_count);
1536 			queue_work(bnxt_re_wq, &re_work->work);
1537 		}
1538 	}
1539 
1540 exit:
1541 	return NOTIFY_DONE;
1542 }
1543 
1544 static struct notifier_block bnxt_re_netdev_notifier = {
1545 	.notifier_call = bnxt_re_netdev_event
1546 };
1547 
1548 static int __init bnxt_re_mod_init(void)
1549 {
1550 	int rc = 0;
1551 
1552 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1553 
1554 	bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1555 	if (!bnxt_re_wq)
1556 		return -ENOMEM;
1557 
1558 	INIT_LIST_HEAD(&bnxt_re_dev_list);
1559 
1560 	rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1561 	if (rc) {
1562 		pr_err("%s: Cannot register to netdevice_notifier",
1563 		       ROCE_DRV_MODULE_NAME);
1564 		goto err_netdev;
1565 	}
1566 	return 0;
1567 
1568 err_netdev:
1569 	destroy_workqueue(bnxt_re_wq);
1570 
1571 	return rc;
1572 }
1573 
1574 static void __exit bnxt_re_mod_exit(void)
1575 {
1576 	struct bnxt_re_dev *rdev, *next;
1577 	LIST_HEAD(to_be_deleted);
1578 
1579 	mutex_lock(&bnxt_re_dev_lock);
1580 	/* Free all adapter allocated resources */
1581 	if (!list_empty(&bnxt_re_dev_list))
1582 		list_splice_init(&bnxt_re_dev_list, &to_be_deleted);
1583 	mutex_unlock(&bnxt_re_dev_lock);
1584        /*
1585 	* Cleanup the devices in reverse order so that the VF device
1586 	* cleanup is done before PF cleanup
1587 	*/
1588 	list_for_each_entry_safe_reverse(rdev, next, &to_be_deleted, list) {
1589 		dev_info(rdev_to_dev(rdev), "Unregistering Device");
1590 		/*
1591 		 * Flush out any scheduled tasks before destroying the
1592 		 * resources
1593 		 */
1594 		flush_workqueue(bnxt_re_wq);
1595 		bnxt_re_dev_stop(rdev);
1596 		bnxt_re_ib_unreg(rdev, true);
1597 		bnxt_re_remove_one(rdev);
1598 		bnxt_re_dev_unreg(rdev);
1599 	}
1600 	unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1601 	if (bnxt_re_wq)
1602 		destroy_workqueue(bnxt_re_wq);
1603 }
1604 
1605 module_init(bnxt_re_mod_init);
1606 module_exit(bnxt_re_mod_exit);
1607