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);
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);
186 }
187 
188 static void bnxt_re_stop_irq(void *handle)
189 {
190 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
191 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
192 	struct bnxt_qplib_nq *nq;
193 	int indx;
194 
195 	for (indx = BNXT_RE_NQ_IDX; indx < rdev->num_msix; indx++) {
196 		nq = &rdev->nq[indx - 1];
197 		bnxt_qplib_nq_stop_irq(nq, false);
198 	}
199 
200 	bnxt_qplib_rcfw_stop_irq(rcfw, false);
201 }
202 
203 static void bnxt_re_start_irq(void *handle, struct bnxt_msix_entry *ent)
204 {
205 	struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
206 	struct bnxt_msix_entry *msix_ent = rdev->msix_entries;
207 	struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
208 	struct bnxt_qplib_nq *nq;
209 	int indx, rc;
210 
211 	if (!ent) {
212 		/* Not setting the f/w timeout bit in rcfw.
213 		 * During the driver unload the first command
214 		 * to f/w will timeout and that will set the
215 		 * timeout bit.
216 		 */
217 		dev_err(rdev_to_dev(rdev), "Failed to re-start IRQs\n");
218 		return;
219 	}
220 
221 	/* Vectors may change after restart, so update with new vectors
222 	 * in device sctructure.
223 	 */
224 	for (indx = 0; indx < rdev->num_msix; indx++)
225 		rdev->msix_entries[indx].vector = ent[indx].vector;
226 
227 	bnxt_qplib_rcfw_start_irq(rcfw, msix_ent[BNXT_RE_AEQ_IDX].vector,
228 				  false);
229 	for (indx = BNXT_RE_NQ_IDX ; indx < rdev->num_msix; indx++) {
230 		nq = &rdev->nq[indx - 1];
231 		rc = bnxt_qplib_nq_start_irq(nq, indx - 1,
232 					     msix_ent[indx].vector, false);
233 		if (rc)
234 			dev_warn(rdev_to_dev(rdev),
235 				 "Failed to reinit NQ index %d\n", indx - 1);
236 	}
237 }
238 
239 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
240 	.ulp_async_notifier = NULL,
241 	.ulp_stop = bnxt_re_stop,
242 	.ulp_start = bnxt_re_start,
243 	.ulp_sriov_config = bnxt_re_sriov_config,
244 	.ulp_shutdown = bnxt_re_shutdown,
245 	.ulp_irq_stop = bnxt_re_stop_irq,
246 	.ulp_irq_restart = bnxt_re_start_irq
247 };
248 
249 /* RoCE -> Net driver */
250 
251 /* Driver registration routines used to let the networking driver (bnxt_en)
252  * to know that the RoCE driver is now installed
253  */
254 static int bnxt_re_unregister_netdev(struct bnxt_re_dev *rdev)
255 {
256 	struct bnxt_en_dev *en_dev;
257 	int rc;
258 
259 	if (!rdev)
260 		return -EINVAL;
261 
262 	en_dev = rdev->en_dev;
263 
264 	rc = en_dev->en_ops->bnxt_unregister_device(rdev->en_dev,
265 						    BNXT_ROCE_ULP);
266 	return rc;
267 }
268 
269 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
270 {
271 	struct bnxt_en_dev *en_dev;
272 	int rc = 0;
273 
274 	if (!rdev)
275 		return -EINVAL;
276 
277 	en_dev = rdev->en_dev;
278 
279 	rc = en_dev->en_ops->bnxt_register_device(en_dev, BNXT_ROCE_ULP,
280 						  &bnxt_re_ulp_ops, rdev);
281 	return rc;
282 }
283 
284 static int bnxt_re_free_msix(struct bnxt_re_dev *rdev)
285 {
286 	struct bnxt_en_dev *en_dev;
287 	int rc;
288 
289 	if (!rdev)
290 		return -EINVAL;
291 
292 	en_dev = rdev->en_dev;
293 
294 
295 	rc = en_dev->en_ops->bnxt_free_msix(rdev->en_dev, BNXT_ROCE_ULP);
296 
297 	return rc;
298 }
299 
300 static int bnxt_re_request_msix(struct bnxt_re_dev *rdev)
301 {
302 	int rc = 0, num_msix_want = BNXT_RE_MAX_MSIX, num_msix_got;
303 	struct bnxt_en_dev *en_dev;
304 
305 	if (!rdev)
306 		return -EINVAL;
307 
308 	en_dev = rdev->en_dev;
309 
310 	num_msix_want = min_t(u32, BNXT_RE_MAX_MSIX, num_online_cpus());
311 
312 	num_msix_got = en_dev->en_ops->bnxt_request_msix(en_dev, BNXT_ROCE_ULP,
313 							 rdev->msix_entries,
314 							 num_msix_want);
315 	if (num_msix_got < BNXT_RE_MIN_MSIX) {
316 		rc = -EINVAL;
317 		goto done;
318 	}
319 	if (num_msix_got != num_msix_want) {
320 		dev_warn(rdev_to_dev(rdev),
321 			 "Requested %d MSI-X vectors, got %d\n",
322 			 num_msix_want, num_msix_got);
323 	}
324 	rdev->num_msix = num_msix_got;
325 done:
326 	return rc;
327 }
328 
329 static void bnxt_re_init_hwrm_hdr(struct bnxt_re_dev *rdev, struct input *hdr,
330 				  u16 opcd, u16 crid, u16 trid)
331 {
332 	hdr->req_type = cpu_to_le16(opcd);
333 	hdr->cmpl_ring = cpu_to_le16(crid);
334 	hdr->target_id = cpu_to_le16(trid);
335 }
336 
337 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
338 				int msg_len, void *resp, int resp_max_len,
339 				int timeout)
340 {
341 	fw_msg->msg = msg;
342 	fw_msg->msg_len = msg_len;
343 	fw_msg->resp = resp;
344 	fw_msg->resp_max_len = resp_max_len;
345 	fw_msg->timeout = timeout;
346 }
347 
348 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev, u16 fw_ring_id)
349 {
350 	struct bnxt_en_dev *en_dev = rdev->en_dev;
351 	struct hwrm_ring_free_input req = {0};
352 	struct hwrm_ring_free_output resp;
353 	struct bnxt_fw_msg fw_msg;
354 	int rc = -EINVAL;
355 
356 	if (!en_dev)
357 		return rc;
358 
359 	memset(&fw_msg, 0, sizeof(fw_msg));
360 
361 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1);
362 	req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
363 	req.ring_id = cpu_to_le16(fw_ring_id);
364 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
365 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
366 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
367 	if (rc)
368 		dev_err(rdev_to_dev(rdev),
369 			"Failed to free HW ring:%d :%#x", req.ring_id, rc);
370 	return rc;
371 }
372 
373 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev, dma_addr_t *dma_arr,
374 				  int pages, int type, u32 ring_mask,
375 				  u32 map_index, u16 *fw_ring_id)
376 {
377 	struct bnxt_en_dev *en_dev = rdev->en_dev;
378 	struct hwrm_ring_alloc_input req = {0};
379 	struct hwrm_ring_alloc_output resp;
380 	struct bnxt_fw_msg fw_msg;
381 	int rc = -EINVAL;
382 
383 	if (!en_dev)
384 		return rc;
385 
386 	memset(&fw_msg, 0, sizeof(fw_msg));
387 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_ALLOC, -1, -1);
388 	req.enables = 0;
389 	req.page_tbl_addr =  cpu_to_le64(dma_arr[0]);
390 	if (pages > 1) {
391 		/* Page size is in log2 units */
392 		req.page_size = BNXT_PAGE_SHIFT;
393 		req.page_tbl_depth = 1;
394 	}
395 	req.fbo = 0;
396 	/* Association of ring index with doorbell index and MSIX number */
397 	req.logical_id = cpu_to_le16(map_index);
398 	req.length = cpu_to_le32(ring_mask + 1);
399 	req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
400 	req.int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
401 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
402 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
403 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
404 	if (!rc)
405 		*fw_ring_id = le16_to_cpu(resp.ring_id);
406 
407 	return rc;
408 }
409 
410 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
411 				      u32 fw_stats_ctx_id)
412 {
413 	struct bnxt_en_dev *en_dev = rdev->en_dev;
414 	struct hwrm_stat_ctx_free_input req = {0};
415 	struct bnxt_fw_msg fw_msg;
416 	int rc = -EINVAL;
417 
418 	if (!en_dev)
419 		return rc;
420 
421 	memset(&fw_msg, 0, sizeof(fw_msg));
422 
423 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_FREE, -1, -1);
424 	req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
425 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&req,
426 			    sizeof(req), DFLT_HWRM_CMD_TIMEOUT);
427 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
428 	if (rc)
429 		dev_err(rdev_to_dev(rdev),
430 			"Failed to free HW stats context %#x", rc);
431 
432 	return rc;
433 }
434 
435 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
436 				       dma_addr_t dma_map,
437 				       u32 *fw_stats_ctx_id)
438 {
439 	struct hwrm_stat_ctx_alloc_output resp = {0};
440 	struct hwrm_stat_ctx_alloc_input req = {0};
441 	struct bnxt_en_dev *en_dev = rdev->en_dev;
442 	struct bnxt_fw_msg fw_msg;
443 	int rc = -EINVAL;
444 
445 	*fw_stats_ctx_id = INVALID_STATS_CTX_ID;
446 
447 	if (!en_dev)
448 		return rc;
449 
450 	memset(&fw_msg, 0, sizeof(fw_msg));
451 
452 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1);
453 	req.update_period_ms = cpu_to_le32(1000);
454 	req.stats_dma_addr = cpu_to_le64(dma_map);
455 	req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
456 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
457 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
458 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
459 	if (!rc)
460 		*fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
461 
462 	return rc;
463 }
464 
465 /* Device */
466 
467 static bool is_bnxt_re_dev(struct net_device *netdev)
468 {
469 	struct ethtool_drvinfo drvinfo;
470 
471 	if (netdev->ethtool_ops && netdev->ethtool_ops->get_drvinfo) {
472 		memset(&drvinfo, 0, sizeof(drvinfo));
473 		netdev->ethtool_ops->get_drvinfo(netdev, &drvinfo);
474 
475 		if (strcmp(drvinfo.driver, "bnxt_en"))
476 			return false;
477 		return true;
478 	}
479 	return false;
480 }
481 
482 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
483 {
484 	struct bnxt_re_dev *rdev;
485 
486 	rcu_read_lock();
487 	list_for_each_entry_rcu(rdev, &bnxt_re_dev_list, list) {
488 		if (rdev->netdev == netdev) {
489 			rcu_read_unlock();
490 			return rdev;
491 		}
492 	}
493 	rcu_read_unlock();
494 	return NULL;
495 }
496 
497 static void bnxt_re_dev_unprobe(struct net_device *netdev,
498 				struct bnxt_en_dev *en_dev)
499 {
500 	dev_put(netdev);
501 	module_put(en_dev->pdev->driver->driver.owner);
502 }
503 
504 static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev)
505 {
506 	struct bnxt *bp = netdev_priv(netdev);
507 	struct bnxt_en_dev *en_dev;
508 	struct pci_dev *pdev;
509 
510 	/* Call bnxt_en's RoCE probe via indirect API */
511 	if (!bp->ulp_probe)
512 		return ERR_PTR(-EINVAL);
513 
514 	en_dev = bp->ulp_probe(netdev);
515 	if (IS_ERR(en_dev))
516 		return en_dev;
517 
518 	pdev = en_dev->pdev;
519 	if (!pdev)
520 		return ERR_PTR(-EINVAL);
521 
522 	if (!(en_dev->flags & BNXT_EN_FLAG_ROCE_CAP)) {
523 		dev_info(&pdev->dev,
524 			"%s: probe error: RoCE is not supported on this device",
525 			ROCE_DRV_MODULE_NAME);
526 		return ERR_PTR(-ENODEV);
527 	}
528 
529 	/* Bump net device reference count */
530 	if (!try_module_get(pdev->driver->driver.owner))
531 		return ERR_PTR(-ENODEV);
532 
533 	dev_hold(netdev);
534 
535 	return en_dev;
536 }
537 
538 static void bnxt_re_unregister_ib(struct bnxt_re_dev *rdev)
539 {
540 	ib_unregister_device(&rdev->ibdev);
541 }
542 
543 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
544 {
545 	struct ib_device *ibdev = &rdev->ibdev;
546 
547 	/* ib device init */
548 	ibdev->owner = THIS_MODULE;
549 	ibdev->node_type = RDMA_NODE_IB_CA;
550 	strlcpy(ibdev->name, "bnxt_re%d", IB_DEVICE_NAME_MAX);
551 	strlcpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
552 		strlen(BNXT_RE_DESC) + 5);
553 	ibdev->phys_port_cnt = 1;
554 
555 	bnxt_qplib_get_guid(rdev->netdev->dev_addr, (u8 *)&ibdev->node_guid);
556 
557 	ibdev->num_comp_vectors	= 1;
558 	ibdev->dev.parent = &rdev->en_dev->pdev->dev;
559 	ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
560 
561 	/* User space */
562 	ibdev->uverbs_abi_ver = BNXT_RE_ABI_VERSION;
563 	ibdev->uverbs_cmd_mask =
564 			(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
565 			(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
566 			(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
567 			(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
568 			(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
569 			(1ull << IB_USER_VERBS_CMD_REG_MR)		|
570 			(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
571 			(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
572 			(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
573 			(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
574 			(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
575 			(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
576 			(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
577 			(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
578 			(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
579 			(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
580 			(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
581 			(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
582 			(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
583 			(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
584 			(1ull << IB_USER_VERBS_CMD_CREATE_AH)		|
585 			(1ull << IB_USER_VERBS_CMD_MODIFY_AH)		|
586 			(1ull << IB_USER_VERBS_CMD_QUERY_AH)		|
587 			(1ull << IB_USER_VERBS_CMD_DESTROY_AH);
588 	/* POLL_CQ and REQ_NOTIFY_CQ is directly handled in libbnxt_re */
589 
590 	/* Kernel verbs */
591 	ibdev->query_device		= bnxt_re_query_device;
592 	ibdev->modify_device		= bnxt_re_modify_device;
593 
594 	ibdev->query_port		= bnxt_re_query_port;
595 	ibdev->get_port_immutable	= bnxt_re_get_port_immutable;
596 	ibdev->get_dev_fw_str           = bnxt_re_query_fw_str;
597 	ibdev->query_pkey		= bnxt_re_query_pkey;
598 	ibdev->get_netdev		= bnxt_re_get_netdev;
599 	ibdev->add_gid			= bnxt_re_add_gid;
600 	ibdev->del_gid			= bnxt_re_del_gid;
601 	ibdev->get_link_layer		= bnxt_re_get_link_layer;
602 
603 	ibdev->alloc_pd			= bnxt_re_alloc_pd;
604 	ibdev->dealloc_pd		= bnxt_re_dealloc_pd;
605 
606 	ibdev->create_ah		= bnxt_re_create_ah;
607 	ibdev->modify_ah		= bnxt_re_modify_ah;
608 	ibdev->query_ah			= bnxt_re_query_ah;
609 	ibdev->destroy_ah		= bnxt_re_destroy_ah;
610 
611 	ibdev->create_srq		= bnxt_re_create_srq;
612 	ibdev->modify_srq		= bnxt_re_modify_srq;
613 	ibdev->query_srq		= bnxt_re_query_srq;
614 	ibdev->destroy_srq		= bnxt_re_destroy_srq;
615 	ibdev->post_srq_recv		= bnxt_re_post_srq_recv;
616 
617 	ibdev->create_qp		= bnxt_re_create_qp;
618 	ibdev->modify_qp		= bnxt_re_modify_qp;
619 	ibdev->query_qp			= bnxt_re_query_qp;
620 	ibdev->destroy_qp		= bnxt_re_destroy_qp;
621 
622 	ibdev->post_send		= bnxt_re_post_send;
623 	ibdev->post_recv		= bnxt_re_post_recv;
624 
625 	ibdev->create_cq		= bnxt_re_create_cq;
626 	ibdev->destroy_cq		= bnxt_re_destroy_cq;
627 	ibdev->poll_cq			= bnxt_re_poll_cq;
628 	ibdev->req_notify_cq		= bnxt_re_req_notify_cq;
629 
630 	ibdev->get_dma_mr		= bnxt_re_get_dma_mr;
631 	ibdev->dereg_mr			= bnxt_re_dereg_mr;
632 	ibdev->alloc_mr			= bnxt_re_alloc_mr;
633 	ibdev->map_mr_sg		= bnxt_re_map_mr_sg;
634 
635 	ibdev->reg_user_mr		= bnxt_re_reg_user_mr;
636 	ibdev->alloc_ucontext		= bnxt_re_alloc_ucontext;
637 	ibdev->dealloc_ucontext		= bnxt_re_dealloc_ucontext;
638 	ibdev->mmap			= bnxt_re_mmap;
639 	ibdev->get_hw_stats             = bnxt_re_ib_get_hw_stats;
640 	ibdev->alloc_hw_stats           = bnxt_re_ib_alloc_hw_stats;
641 
642 	ibdev->driver_id = RDMA_DRIVER_BNXT_RE;
643 	return ib_register_device(ibdev, NULL);
644 }
645 
646 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
647 			char *buf)
648 {
649 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
650 
651 	return scnprintf(buf, PAGE_SIZE, "0x%x\n", rdev->en_dev->pdev->vendor);
652 }
653 
654 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
655 			char *buf)
656 {
657 	struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
658 
659 	return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->ibdev.node_desc);
660 }
661 
662 static DEVICE_ATTR(hw_rev, 0444, show_rev, NULL);
663 static DEVICE_ATTR(hca_type, 0444, show_hca, NULL);
664 
665 static struct device_attribute *bnxt_re_attributes[] = {
666 	&dev_attr_hw_rev,
667 	&dev_attr_hca_type
668 };
669 
670 static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev)
671 {
672 	dev_put(rdev->netdev);
673 	rdev->netdev = NULL;
674 
675 	mutex_lock(&bnxt_re_dev_lock);
676 	list_del_rcu(&rdev->list);
677 	mutex_unlock(&bnxt_re_dev_lock);
678 
679 	synchronize_rcu();
680 
681 	ib_dealloc_device(&rdev->ibdev);
682 	/* rdev is gone */
683 }
684 
685 static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev,
686 					   struct bnxt_en_dev *en_dev)
687 {
688 	struct bnxt_re_dev *rdev;
689 
690 	/* Allocate bnxt_re_dev instance here */
691 	rdev = (struct bnxt_re_dev *)ib_alloc_device(sizeof(*rdev));
692 	if (!rdev) {
693 		dev_err(NULL, "%s: bnxt_re_dev allocation failure!",
694 			ROCE_DRV_MODULE_NAME);
695 		return NULL;
696 	}
697 	/* Default values */
698 	rdev->netdev = netdev;
699 	dev_hold(rdev->netdev);
700 	rdev->en_dev = en_dev;
701 	rdev->id = rdev->en_dev->pdev->devfn;
702 	INIT_LIST_HEAD(&rdev->qp_list);
703 	mutex_init(&rdev->qp_lock);
704 	atomic_set(&rdev->qp_count, 0);
705 	atomic_set(&rdev->cq_count, 0);
706 	atomic_set(&rdev->srq_count, 0);
707 	atomic_set(&rdev->mr_count, 0);
708 	atomic_set(&rdev->mw_count, 0);
709 	rdev->cosq[0] = 0xFFFF;
710 	rdev->cosq[1] = 0xFFFF;
711 
712 	mutex_lock(&bnxt_re_dev_lock);
713 	list_add_tail_rcu(&rdev->list, &bnxt_re_dev_list);
714 	mutex_unlock(&bnxt_re_dev_lock);
715 	return rdev;
716 }
717 
718 static int bnxt_re_handle_unaffi_async_event(struct creq_func_event
719 					     *unaffi_async)
720 {
721 	switch (unaffi_async->event) {
722 	case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
723 		break;
724 	case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
725 		break;
726 	case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
727 		break;
728 	case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
729 		break;
730 	case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
731 		break;
732 	case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
733 		break;
734 	case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
735 		break;
736 	case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
737 		break;
738 	case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
739 		break;
740 	case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
741 		break;
742 	case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
743 		break;
744 	default:
745 		return -EINVAL;
746 	}
747 	return 0;
748 }
749 
750 static int bnxt_re_handle_qp_async_event(struct creq_qp_event *qp_event,
751 					 struct bnxt_re_qp *qp)
752 {
753 	struct ib_event event;
754 	unsigned int flags;
755 
756 	if (qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
757 		flags = bnxt_re_lock_cqs(qp);
758 		bnxt_qplib_add_flush_qp(&qp->qplib_qp);
759 		bnxt_re_unlock_cqs(qp, flags);
760 	}
761 
762 	memset(&event, 0, sizeof(event));
763 	if (qp->qplib_qp.srq) {
764 		event.device = &qp->rdev->ibdev;
765 		event.element.qp = &qp->ib_qp;
766 		event.event = IB_EVENT_QP_LAST_WQE_REACHED;
767 	}
768 
769 	if (event.device && qp->ib_qp.event_handler)
770 		qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
771 
772 	return 0;
773 }
774 
775 static int bnxt_re_handle_affi_async_event(struct creq_qp_event *affi_async,
776 					   void *obj)
777 {
778 	int rc = 0;
779 	u8 event;
780 
781 	if (!obj)
782 		return rc; /* QP was already dead, still return success */
783 
784 	event = affi_async->event;
785 	if (event == CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION) {
786 		struct bnxt_qplib_qp *lib_qp = obj;
787 		struct bnxt_re_qp *qp = container_of(lib_qp, struct bnxt_re_qp,
788 						     qplib_qp);
789 		rc = bnxt_re_handle_qp_async_event(affi_async, qp);
790 	}
791 	return rc;
792 }
793 
794 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
795 			       void *aeqe, void *obj)
796 {
797 	struct creq_qp_event *affi_async;
798 	struct creq_func_event *unaffi_async;
799 	u8 type;
800 	int rc;
801 
802 	type = ((struct creq_base *)aeqe)->type;
803 	if (type == CREQ_BASE_TYPE_FUNC_EVENT) {
804 		unaffi_async = aeqe;
805 		rc = bnxt_re_handle_unaffi_async_event(unaffi_async);
806 	} else {
807 		affi_async = aeqe;
808 		rc = bnxt_re_handle_affi_async_event(affi_async, obj);
809 	}
810 
811 	return rc;
812 }
813 
814 static int bnxt_re_srqn_handler(struct bnxt_qplib_nq *nq,
815 				struct bnxt_qplib_srq *handle, u8 event)
816 {
817 	struct bnxt_re_srq *srq = container_of(handle, struct bnxt_re_srq,
818 					       qplib_srq);
819 	struct ib_event ib_event;
820 	int rc = 0;
821 
822 	if (!srq) {
823 		dev_err(NULL, "%s: SRQ is NULL, SRQN not handled",
824 			ROCE_DRV_MODULE_NAME);
825 		rc = -EINVAL;
826 		goto done;
827 	}
828 	ib_event.device = &srq->rdev->ibdev;
829 	ib_event.element.srq = &srq->ib_srq;
830 	if (event == NQ_SRQ_EVENT_EVENT_SRQ_THRESHOLD_EVENT)
831 		ib_event.event = IB_EVENT_SRQ_LIMIT_REACHED;
832 	else
833 		ib_event.event = IB_EVENT_SRQ_ERR;
834 
835 	if (srq->ib_srq.event_handler) {
836 		/* Lock event_handler? */
837 		(*srq->ib_srq.event_handler)(&ib_event,
838 					     srq->ib_srq.srq_context);
839 	}
840 done:
841 	return rc;
842 }
843 
844 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
845 			       struct bnxt_qplib_cq *handle)
846 {
847 	struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
848 					     qplib_cq);
849 
850 	if (!cq) {
851 		dev_err(NULL, "%s: CQ is NULL, CQN not handled",
852 			ROCE_DRV_MODULE_NAME);
853 		return -EINVAL;
854 	}
855 	if (cq->ib_cq.comp_handler) {
856 		/* Lock comp_handler? */
857 		(*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
858 	}
859 
860 	return 0;
861 }
862 
863 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
864 {
865 	int i;
866 
867 	if (rdev->nq[0].hwq.max_elements) {
868 		for (i = 1; i < rdev->num_msix; i++)
869 			bnxt_qplib_disable_nq(&rdev->nq[i - 1]);
870 	}
871 
872 	if (rdev->qplib_res.rcfw)
873 		bnxt_qplib_cleanup_res(&rdev->qplib_res);
874 }
875 
876 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
877 {
878 	int rc = 0, i;
879 
880 	bnxt_qplib_init_res(&rdev->qplib_res);
881 
882 	for (i = 1; i < rdev->num_msix ; i++) {
883 		rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
884 					  i - 1, rdev->msix_entries[i].vector,
885 					  rdev->msix_entries[i].db_offset,
886 					  &bnxt_re_cqn_handler,
887 					  &bnxt_re_srqn_handler);
888 
889 		if (rc) {
890 			dev_err(rdev_to_dev(rdev),
891 				"Failed to enable NQ with rc = 0x%x", rc);
892 			goto fail;
893 		}
894 	}
895 	return 0;
896 fail:
897 	return rc;
898 }
899 
900 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev)
901 {
902 	int i;
903 
904 	for (i = 0; i < rdev->num_msix - 1; i++) {
905 		bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id);
906 		bnxt_qplib_free_nq(&rdev->nq[i]);
907 	}
908 }
909 
910 static void bnxt_re_free_res(struct bnxt_re_dev *rdev)
911 {
912 	bnxt_re_free_nq_res(rdev);
913 
914 	if (rdev->qplib_res.dpi_tbl.max) {
915 		bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
916 				       &rdev->qplib_res.dpi_tbl,
917 				       &rdev->dpi_privileged);
918 	}
919 	if (rdev->qplib_res.rcfw) {
920 		bnxt_qplib_free_res(&rdev->qplib_res);
921 		rdev->qplib_res.rcfw = NULL;
922 	}
923 }
924 
925 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
926 {
927 	int rc = 0, i;
928 
929 	/* Configure and allocate resources for qplib */
930 	rdev->qplib_res.rcfw = &rdev->rcfw;
931 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
932 				     rdev->is_virtfn);
933 	if (rc)
934 		goto fail;
935 
936 	rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
937 				  rdev->netdev, &rdev->dev_attr);
938 	if (rc)
939 		goto fail;
940 
941 	rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
942 				  &rdev->dpi_privileged,
943 				  rdev);
944 	if (rc)
945 		goto dealloc_res;
946 
947 	for (i = 0; i < rdev->num_msix - 1; i++) {
948 		rdev->nq[i].hwq.max_elements = BNXT_RE_MAX_CQ_COUNT +
949 			BNXT_RE_MAX_SRQC_COUNT + 2;
950 		rc = bnxt_qplib_alloc_nq(rdev->en_dev->pdev, &rdev->nq[i]);
951 		if (rc) {
952 			dev_err(rdev_to_dev(rdev), "Alloc Failed NQ%d rc:%#x",
953 				i, rc);
954 			goto dealloc_dpi;
955 		}
956 		rc = bnxt_re_net_ring_alloc
957 			(rdev, rdev->nq[i].hwq.pbl[PBL_LVL_0].pg_map_arr,
958 			 rdev->nq[i].hwq.pbl[rdev->nq[i].hwq.level].pg_count,
959 			 HWRM_RING_ALLOC_CMPL,
960 			 BNXT_QPLIB_NQE_MAX_CNT - 1,
961 			 rdev->msix_entries[i + 1].ring_idx,
962 			 &rdev->nq[i].ring_id);
963 		if (rc) {
964 			dev_err(rdev_to_dev(rdev),
965 				"Failed to allocate NQ fw id with rc = 0x%x",
966 				rc);
967 			goto free_nq;
968 		}
969 	}
970 	return 0;
971 free_nq:
972 	for (i = 0; i < rdev->num_msix - 1; i++)
973 		bnxt_qplib_free_nq(&rdev->nq[i]);
974 dealloc_dpi:
975 	bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
976 			       &rdev->qplib_res.dpi_tbl,
977 			       &rdev->dpi_privileged);
978 dealloc_res:
979 	bnxt_qplib_free_res(&rdev->qplib_res);
980 
981 fail:
982 	rdev->qplib_res.rcfw = NULL;
983 	return rc;
984 }
985 
986 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
987 				   u8 port_num, enum ib_event_type event)
988 {
989 	struct ib_event ib_event;
990 
991 	ib_event.device = ibdev;
992 	if (qp)
993 		ib_event.element.qp = qp;
994 	else
995 		ib_event.element.port_num = port_num;
996 	ib_event.event = event;
997 	ib_dispatch_event(&ib_event);
998 }
999 
1000 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN      0x02
1001 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
1002 				      u64 *cid_map)
1003 {
1004 	struct hwrm_queue_pri2cos_qcfg_input req = {0};
1005 	struct bnxt *bp = netdev_priv(rdev->netdev);
1006 	struct hwrm_queue_pri2cos_qcfg_output resp;
1007 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1008 	struct bnxt_fw_msg fw_msg;
1009 	u32 flags = 0;
1010 	u8 *qcfgmap, *tmp_map;
1011 	int rc = 0, i;
1012 
1013 	if (!cid_map)
1014 		return -EINVAL;
1015 
1016 	memset(&fw_msg, 0, sizeof(fw_msg));
1017 	bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1018 			      HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
1019 	flags |= (dir & 0x01);
1020 	flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
1021 	req.flags = cpu_to_le32(flags);
1022 	req.port_id = bp->pf.port_id;
1023 
1024 	bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1025 			    sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1026 	rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1027 	if (rc)
1028 		return rc;
1029 
1030 	if (resp.queue_cfg_info) {
1031 		dev_warn(rdev_to_dev(rdev),
1032 			 "Asymmetric cos queue configuration detected");
1033 		dev_warn(rdev_to_dev(rdev),
1034 			 " on device, QoS may not be fully functional\n");
1035 	}
1036 	qcfgmap = &resp.pri0_cos_queue_id;
1037 	tmp_map = (u8 *)cid_map;
1038 	for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
1039 		tmp_map[i] = qcfgmap[i];
1040 
1041 	return rc;
1042 }
1043 
1044 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
1045 					struct bnxt_re_qp *qp)
1046 {
1047 	return (qp->ib_qp.qp_type == IB_QPT_GSI) || (qp == rdev->qp1_sqp);
1048 }
1049 
1050 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
1051 {
1052 	int mask = IB_QP_STATE;
1053 	struct ib_qp_attr qp_attr;
1054 	struct bnxt_re_qp *qp;
1055 
1056 	qp_attr.qp_state = IB_QPS_ERR;
1057 	mutex_lock(&rdev->qp_lock);
1058 	list_for_each_entry(qp, &rdev->qp_list, list) {
1059 		/* Modify the state of all QPs except QP1/Shadow QP */
1060 		if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
1061 			if (qp->qplib_qp.state !=
1062 			    CMDQ_MODIFY_QP_NEW_STATE_RESET &&
1063 			    qp->qplib_qp.state !=
1064 			    CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1065 				bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
1066 						       1, IB_EVENT_QP_FATAL);
1067 				bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
1068 						  NULL);
1069 			}
1070 		}
1071 	}
1072 	mutex_unlock(&rdev->qp_lock);
1073 }
1074 
1075 static int bnxt_re_update_gid(struct bnxt_re_dev *rdev)
1076 {
1077 	struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
1078 	struct bnxt_qplib_gid gid;
1079 	u16 gid_idx, index;
1080 	int rc = 0;
1081 
1082 	if (!test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1083 		return 0;
1084 
1085 	if (!sgid_tbl) {
1086 		dev_err(rdev_to_dev(rdev), "QPLIB: SGID table not allocated");
1087 		return -EINVAL;
1088 	}
1089 
1090 	for (index = 0; index < sgid_tbl->active; index++) {
1091 		gid_idx = sgid_tbl->hw_id[index];
1092 
1093 		if (!memcmp(&sgid_tbl->tbl[index], &bnxt_qplib_gid_zero,
1094 			    sizeof(bnxt_qplib_gid_zero)))
1095 			continue;
1096 		/* need to modify the VLAN enable setting of non VLAN GID only
1097 		 * as setting is done for VLAN GID while adding GID
1098 		 */
1099 		if (sgid_tbl->vlan[index])
1100 			continue;
1101 
1102 		memcpy(&gid, &sgid_tbl->tbl[index], sizeof(gid));
1103 
1104 		rc = bnxt_qplib_update_sgid(sgid_tbl, &gid, gid_idx,
1105 					    rdev->qplib_res.netdev->dev_addr);
1106 	}
1107 
1108 	return rc;
1109 }
1110 
1111 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
1112 {
1113 	u32 prio_map = 0, tmp_map = 0;
1114 	struct net_device *netdev;
1115 	struct dcb_app app;
1116 
1117 	netdev = rdev->netdev;
1118 
1119 	memset(&app, 0, sizeof(app));
1120 	app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
1121 	app.protocol = ETH_P_IBOE;
1122 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1123 	prio_map = tmp_map;
1124 
1125 	app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
1126 	app.protocol = ROCE_V2_UDP_DPORT;
1127 	tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1128 	prio_map |= tmp_map;
1129 
1130 	return prio_map;
1131 }
1132 
1133 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
1134 {
1135 	u16 prio;
1136 	u8 id;
1137 
1138 	for (prio = 0, id = 0; prio < 8; prio++) {
1139 		if (prio_map & (1 << prio)) {
1140 			cosq[id] = cid_map[prio];
1141 			id++;
1142 			if (id == 2) /* Max 2 tcs supported */
1143 				break;
1144 		}
1145 	}
1146 }
1147 
1148 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
1149 {
1150 	u8 prio_map = 0;
1151 	u64 cid_map;
1152 	int rc;
1153 
1154 	/* Get priority for roce */
1155 	prio_map = bnxt_re_get_priority_mask(rdev);
1156 
1157 	if (prio_map == rdev->cur_prio_map)
1158 		return 0;
1159 	rdev->cur_prio_map = prio_map;
1160 	/* Get cosq id for this priority */
1161 	rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
1162 	if (rc) {
1163 		dev_warn(rdev_to_dev(rdev), "no cos for p_mask %x\n", prio_map);
1164 		return rc;
1165 	}
1166 	/* Parse CoS IDs for app priority */
1167 	bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
1168 
1169 	/* Config BONO. */
1170 	rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
1171 	if (rc) {
1172 		dev_warn(rdev_to_dev(rdev), "no tc for cos{%x, %x}\n",
1173 			 rdev->cosq[0], rdev->cosq[1]);
1174 		return rc;
1175 	}
1176 
1177 	/* Actual priorities are not programmed as they are already
1178 	 * done by L2 driver; just enable or disable priority vlan tagging
1179 	 */
1180 	if ((prio_map == 0 && rdev->qplib_res.prio) ||
1181 	    (prio_map != 0 && !rdev->qplib_res.prio)) {
1182 		rdev->qplib_res.prio = prio_map ? true : false;
1183 
1184 		bnxt_re_update_gid(rdev);
1185 	}
1186 
1187 	return 0;
1188 }
1189 
1190 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev)
1191 {
1192 	int i, rc;
1193 
1194 	if (test_and_clear_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags)) {
1195 		for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++)
1196 			device_remove_file(&rdev->ibdev.dev,
1197 					   bnxt_re_attributes[i]);
1198 		/* Cleanup ib dev */
1199 		bnxt_re_unregister_ib(rdev);
1200 	}
1201 	if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
1202 		cancel_delayed_work(&rdev->worker);
1203 
1204 	bnxt_re_cleanup_res(rdev);
1205 	bnxt_re_free_res(rdev);
1206 
1207 	if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
1208 		rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
1209 		if (rc)
1210 			dev_warn(rdev_to_dev(rdev),
1211 				 "Failed to deinitialize RCFW: %#x", rc);
1212 		bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1213 		bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1214 		bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1215 		bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id);
1216 		bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1217 	}
1218 	if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
1219 		rc = bnxt_re_free_msix(rdev);
1220 		if (rc)
1221 			dev_warn(rdev_to_dev(rdev),
1222 				 "Failed to free MSI-X vectors: %#x", rc);
1223 	}
1224 	if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
1225 		rc = bnxt_re_unregister_netdev(rdev);
1226 		if (rc)
1227 			dev_warn(rdev_to_dev(rdev),
1228 				 "Failed to unregister with netdev: %#x", rc);
1229 	}
1230 }
1231 
1232 /* worker thread for polling periodic events. Now used for QoS programming*/
1233 static void bnxt_re_worker(struct work_struct *work)
1234 {
1235 	struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
1236 						worker.work);
1237 
1238 	bnxt_re_setup_qos(rdev);
1239 	schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1240 }
1241 
1242 static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
1243 {
1244 	int i, j, rc;
1245 
1246 	bool locked;
1247 
1248 	/* Acquire rtnl lock through out this function */
1249 	rtnl_lock();
1250 	locked = true;
1251 
1252 	/* Registered a new RoCE device instance to netdev */
1253 	rc = bnxt_re_register_netdev(rdev);
1254 	if (rc) {
1255 		pr_err("Failed to register with netedev: %#x\n", rc);
1256 		return -EINVAL;
1257 	}
1258 	set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1259 
1260 	/* Check whether VF or PF */
1261 	bnxt_re_get_sriov_func_type(rdev);
1262 
1263 	rc = bnxt_re_request_msix(rdev);
1264 	if (rc) {
1265 		pr_err("Failed to get MSI-X vectors: %#x\n", rc);
1266 		rc = -EINVAL;
1267 		goto fail;
1268 	}
1269 	set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
1270 
1271 	/* Establish RCFW Communication Channel to initialize the context
1272 	 * memory for the function and all child VFs
1273 	 */
1274 	rc = bnxt_qplib_alloc_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw,
1275 					   BNXT_RE_MAX_QPC_COUNT);
1276 	if (rc) {
1277 		pr_err("Failed to allocate RCFW Channel: %#x\n", rc);
1278 		goto fail;
1279 	}
1280 	rc = bnxt_re_net_ring_alloc
1281 			(rdev, rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr,
1282 			 rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count,
1283 			 HWRM_RING_ALLOC_CMPL, BNXT_QPLIB_CREQE_MAX_CNT - 1,
1284 			 rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx,
1285 			 &rdev->rcfw.creq_ring_id);
1286 	if (rc) {
1287 		pr_err("Failed to allocate CREQ: %#x\n", rc);
1288 		goto free_rcfw;
1289 	}
1290 	rc = bnxt_qplib_enable_rcfw_channel
1291 				(rdev->en_dev->pdev, &rdev->rcfw,
1292 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].vector,
1293 				 rdev->msix_entries[BNXT_RE_AEQ_IDX].db_offset,
1294 				 rdev->is_virtfn, &bnxt_re_aeq_handler);
1295 	if (rc) {
1296 		pr_err("Failed to enable RCFW channel: %#x\n", rc);
1297 		goto free_ring;
1298 	}
1299 
1300 	rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
1301 				     rdev->is_virtfn);
1302 	if (rc)
1303 		goto disable_rcfw;
1304 	if (!rdev->is_virtfn)
1305 		bnxt_re_set_resource_limits(rdev);
1306 
1307 	rc = bnxt_qplib_alloc_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx, 0);
1308 	if (rc) {
1309 		pr_err("Failed to allocate QPLIB context: %#x\n", rc);
1310 		goto disable_rcfw;
1311 	}
1312 	rc = bnxt_re_net_stats_ctx_alloc(rdev,
1313 					 rdev->qplib_ctx.stats.dma_map,
1314 					 &rdev->qplib_ctx.stats.fw_id);
1315 	if (rc) {
1316 		pr_err("Failed to allocate stats context: %#x\n", rc);
1317 		goto free_ctx;
1318 	}
1319 
1320 	rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx,
1321 				  rdev->is_virtfn);
1322 	if (rc) {
1323 		pr_err("Failed to initialize RCFW: %#x\n", rc);
1324 		goto free_sctx;
1325 	}
1326 	set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1327 
1328 	/* Resources based on the 'new' device caps */
1329 	rc = bnxt_re_alloc_res(rdev);
1330 	if (rc) {
1331 		pr_err("Failed to allocate resources: %#x\n", rc);
1332 		goto fail;
1333 	}
1334 	rc = bnxt_re_init_res(rdev);
1335 	if (rc) {
1336 		pr_err("Failed to initialize resources: %#x\n", rc);
1337 		goto fail;
1338 	}
1339 
1340 	if (!rdev->is_virtfn) {
1341 		rc = bnxt_re_setup_qos(rdev);
1342 		if (rc)
1343 			pr_info("RoCE priority not yet configured\n");
1344 
1345 		INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1346 		set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1347 		schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1348 	}
1349 
1350 	rtnl_unlock();
1351 	locked = false;
1352 
1353 	/* Register ib dev */
1354 	rc = bnxt_re_register_ib(rdev);
1355 	if (rc) {
1356 		pr_err("Failed to register with IB: %#x\n", rc);
1357 		goto fail;
1358 	}
1359 	set_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags);
1360 	dev_info(rdev_to_dev(rdev), "Device registered successfully");
1361 	for (i = 0; i < ARRAY_SIZE(bnxt_re_attributes); i++) {
1362 		rc = device_create_file(&rdev->ibdev.dev,
1363 					bnxt_re_attributes[i]);
1364 		if (rc) {
1365 			dev_err(rdev_to_dev(rdev),
1366 				"Failed to create IB sysfs: %#x", rc);
1367 			/* Must clean up all created device files */
1368 			for (j = 0; j < i; j++)
1369 				device_remove_file(&rdev->ibdev.dev,
1370 						   bnxt_re_attributes[j]);
1371 			bnxt_re_unregister_ib(rdev);
1372 			goto fail;
1373 		}
1374 	}
1375 	ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1376 			 &rdev->active_width);
1377 	set_bit(BNXT_RE_FLAG_ISSUE_ROCE_STATS, &rdev->flags);
1378 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_PORT_ACTIVE);
1379 	bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_GID_CHANGE);
1380 
1381 	return 0;
1382 free_sctx:
1383 	bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1384 free_ctx:
1385 	bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1386 disable_rcfw:
1387 	bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1388 free_ring:
1389 	bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id);
1390 free_rcfw:
1391 	bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1392 fail:
1393 	if (!locked)
1394 		rtnl_lock();
1395 	bnxt_re_ib_unreg(rdev);
1396 	rtnl_unlock();
1397 
1398 	return rc;
1399 }
1400 
1401 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1402 {
1403 	struct bnxt_en_dev *en_dev = rdev->en_dev;
1404 	struct net_device *netdev = rdev->netdev;
1405 
1406 	bnxt_re_dev_remove(rdev);
1407 
1408 	if (netdev)
1409 		bnxt_re_dev_unprobe(netdev, en_dev);
1410 }
1411 
1412 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1413 {
1414 	struct bnxt_en_dev *en_dev;
1415 	int rc = 0;
1416 
1417 	if (!is_bnxt_re_dev(netdev))
1418 		return -ENODEV;
1419 
1420 	en_dev = bnxt_re_dev_probe(netdev);
1421 	if (IS_ERR(en_dev)) {
1422 		if (en_dev != ERR_PTR(-ENODEV))
1423 			pr_err("%s: Failed to probe\n", ROCE_DRV_MODULE_NAME);
1424 		rc = PTR_ERR(en_dev);
1425 		goto exit;
1426 	}
1427 	*rdev = bnxt_re_dev_add(netdev, en_dev);
1428 	if (!*rdev) {
1429 		rc = -ENOMEM;
1430 		bnxt_re_dev_unprobe(netdev, en_dev);
1431 		goto exit;
1432 	}
1433 exit:
1434 	return rc;
1435 }
1436 
1437 static void bnxt_re_remove_one(struct bnxt_re_dev *rdev)
1438 {
1439 	pci_dev_put(rdev->en_dev->pdev);
1440 }
1441 
1442 /* Handle all deferred netevents tasks */
1443 static void bnxt_re_task(struct work_struct *work)
1444 {
1445 	struct bnxt_re_work *re_work;
1446 	struct bnxt_re_dev *rdev;
1447 	int rc = 0;
1448 
1449 	re_work = container_of(work, struct bnxt_re_work, work);
1450 	rdev = re_work->rdev;
1451 
1452 	if (re_work->event != NETDEV_REGISTER &&
1453 	    !test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1454 		return;
1455 
1456 	switch (re_work->event) {
1457 	case NETDEV_REGISTER:
1458 		rc = bnxt_re_ib_reg(rdev);
1459 		if (rc) {
1460 			dev_err(rdev_to_dev(rdev),
1461 				"Failed to register with IB: %#x", rc);
1462 			bnxt_re_remove_one(rdev);
1463 			bnxt_re_dev_unreg(rdev);
1464 		}
1465 		break;
1466 	case NETDEV_UP:
1467 		bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1468 				       IB_EVENT_PORT_ACTIVE);
1469 		break;
1470 	case NETDEV_DOWN:
1471 		bnxt_re_dev_stop(rdev);
1472 		break;
1473 	case NETDEV_CHANGE:
1474 		if (!netif_carrier_ok(rdev->netdev))
1475 			bnxt_re_dev_stop(rdev);
1476 		else if (netif_carrier_ok(rdev->netdev))
1477 			bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1478 					       IB_EVENT_PORT_ACTIVE);
1479 		ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1480 				 &rdev->active_width);
1481 		break;
1482 	default:
1483 		break;
1484 	}
1485 	smp_mb__before_atomic();
1486 	atomic_dec(&rdev->sched_count);
1487 	kfree(re_work);
1488 }
1489 
1490 static void bnxt_re_init_one(struct bnxt_re_dev *rdev)
1491 {
1492 	pci_dev_get(rdev->en_dev->pdev);
1493 }
1494 
1495 /*
1496  * "Notifier chain callback can be invoked for the same chain from
1497  * different CPUs at the same time".
1498  *
1499  * For cases when the netdev is already present, our call to the
1500  * register_netdevice_notifier() will actually get the rtnl_lock()
1501  * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1502  * events.
1503  *
1504  * But for cases when the netdev is not already present, the notifier
1505  * chain is subjected to be invoked from different CPUs simultaneously.
1506  *
1507  * This is protected by the netdev_mutex.
1508  */
1509 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1510 				unsigned long event, void *ptr)
1511 {
1512 	struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1513 	struct bnxt_re_work *re_work;
1514 	struct bnxt_re_dev *rdev;
1515 	int rc = 0;
1516 	bool sch_work = false;
1517 
1518 	real_dev = rdma_vlan_dev_real_dev(netdev);
1519 	if (!real_dev)
1520 		real_dev = netdev;
1521 
1522 	rdev = bnxt_re_from_netdev(real_dev);
1523 	if (!rdev && event != NETDEV_REGISTER)
1524 		goto exit;
1525 	if (real_dev != netdev)
1526 		goto exit;
1527 
1528 	switch (event) {
1529 	case NETDEV_REGISTER:
1530 		if (rdev)
1531 			break;
1532 		rc = bnxt_re_dev_reg(&rdev, real_dev);
1533 		if (rc == -ENODEV)
1534 			break;
1535 		if (rc) {
1536 			pr_err("Failed to register with the device %s: %#x\n",
1537 			       real_dev->name, rc);
1538 			break;
1539 		}
1540 		bnxt_re_init_one(rdev);
1541 		sch_work = true;
1542 		break;
1543 
1544 	case NETDEV_UNREGISTER:
1545 		/* netdev notifier will call NETDEV_UNREGISTER again later since
1546 		 * we are still holding the reference to the netdev
1547 		 */
1548 		if (atomic_read(&rdev->sched_count) > 0)
1549 			goto exit;
1550 		bnxt_re_ib_unreg(rdev);
1551 		bnxt_re_remove_one(rdev);
1552 		bnxt_re_dev_unreg(rdev);
1553 		break;
1554 
1555 	default:
1556 		sch_work = true;
1557 		break;
1558 	}
1559 	if (sch_work) {
1560 		/* Allocate for the deferred task */
1561 		re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1562 		if (re_work) {
1563 			re_work->rdev = rdev;
1564 			re_work->event = event;
1565 			re_work->vlan_dev = (real_dev == netdev ?
1566 					     NULL : netdev);
1567 			INIT_WORK(&re_work->work, bnxt_re_task);
1568 			atomic_inc(&rdev->sched_count);
1569 			queue_work(bnxt_re_wq, &re_work->work);
1570 		}
1571 	}
1572 
1573 exit:
1574 	return NOTIFY_DONE;
1575 }
1576 
1577 static struct notifier_block bnxt_re_netdev_notifier = {
1578 	.notifier_call = bnxt_re_netdev_event
1579 };
1580 
1581 static int __init bnxt_re_mod_init(void)
1582 {
1583 	int rc = 0;
1584 
1585 	pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1586 
1587 	bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1588 	if (!bnxt_re_wq)
1589 		return -ENOMEM;
1590 
1591 	INIT_LIST_HEAD(&bnxt_re_dev_list);
1592 
1593 	rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1594 	if (rc) {
1595 		pr_err("%s: Cannot register to netdevice_notifier",
1596 		       ROCE_DRV_MODULE_NAME);
1597 		goto err_netdev;
1598 	}
1599 	return 0;
1600 
1601 err_netdev:
1602 	destroy_workqueue(bnxt_re_wq);
1603 
1604 	return rc;
1605 }
1606 
1607 static void __exit bnxt_re_mod_exit(void)
1608 {
1609 	struct bnxt_re_dev *rdev, *next;
1610 	LIST_HEAD(to_be_deleted);
1611 
1612 	mutex_lock(&bnxt_re_dev_lock);
1613 	/* Free all adapter allocated resources */
1614 	if (!list_empty(&bnxt_re_dev_list))
1615 		list_splice_init(&bnxt_re_dev_list, &to_be_deleted);
1616 	mutex_unlock(&bnxt_re_dev_lock);
1617        /*
1618 	* Cleanup the devices in reverse order so that the VF device
1619 	* cleanup is done before PF cleanup
1620 	*/
1621 	list_for_each_entry_safe_reverse(rdev, next, &to_be_deleted, list) {
1622 		dev_info(rdev_to_dev(rdev), "Unregistering Device");
1623 		/*
1624 		 * Flush out any scheduled tasks before destroying the
1625 		 * resources
1626 		 */
1627 		flush_workqueue(bnxt_re_wq);
1628 		bnxt_re_dev_stop(rdev);
1629 		/* Acquire the rtnl_lock as the L2 resources are freed here */
1630 		rtnl_lock();
1631 		bnxt_re_ib_unreg(rdev);
1632 		rtnl_unlock();
1633 		bnxt_re_remove_one(rdev);
1634 		bnxt_re_dev_unreg(rdev);
1635 	}
1636 	unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1637 	if (bnxt_re_wq)
1638 		destroy_workqueue(bnxt_re_wq);
1639 }
1640 
1641 module_init(bnxt_re_mod_init);
1642 module_exit(bnxt_re_mod_exit);
1643