xref: /dpdk/drivers/net/bnxt/bnxt_ethdev.c (revision 97b914f4e715565d53d38ac6e04815b9be5e58a9)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2014-2021 Broadcom
3  * All rights reserved.
4  */
5 
6 #include <inttypes.h>
7 #include <stdbool.h>
8 
9 #include <rte_dev.h>
10 #include <ethdev_driver.h>
11 #include <ethdev_pci.h>
12 #include <rte_malloc.h>
13 #include <rte_cycles.h>
14 #include <rte_alarm.h>
15 #include <rte_kvargs.h>
16 #include <rte_vect.h>
17 
18 #include "bnxt.h"
19 #include "bnxt_filter.h"
20 #include "bnxt_hwrm.h"
21 #include "bnxt_irq.h"
22 #include "bnxt_reps.h"
23 #include "bnxt_ring.h"
24 #include "bnxt_rxq.h"
25 #include "bnxt_rxr.h"
26 #include "bnxt_stats.h"
27 #include "bnxt_txq.h"
28 #include "bnxt_txr.h"
29 #include "bnxt_vnic.h"
30 #include "hsi_struct_def_dpdk.h"
31 #include "bnxt_nvm_defs.h"
32 #include "bnxt_tf_common.h"
33 #include "ulp_flow_db.h"
34 #include "rte_pmd_bnxt.h"
35 
36 #define DRV_MODULE_NAME		"bnxt"
37 static const char bnxt_version[] =
38 	"Broadcom NetXtreme driver " DRV_MODULE_NAME;
39 
40 /*
41  * The set of PCI devices this driver supports
42  */
43 static const struct rte_pci_id bnxt_pci_id_map[] = {
44 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM,
45 			 BROADCOM_DEV_ID_STRATUS_NIC_VF1) },
46 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM,
47 			 BROADCOM_DEV_ID_STRATUS_NIC_VF2) },
48 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_STRATUS_NIC) },
49 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57414_VF) },
50 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57304_VF) },
51 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_NS2) },
52 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57406_VF) },
53 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57407_MF) },
54 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_5741X_VF) },
55 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_5731X_VF) },
56 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57417_MF) },
57 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57412) },
58 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57414) },
59 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57416_RJ45) },
60 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57417_RJ45) },
61 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57412_MF) },
62 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57317_RJ45) },
63 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57417_SFP) },
64 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57416_SFP) },
65 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57317_SFP) },
66 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57414_MF) },
67 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57416_MF) },
68 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_58802) },
69 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_58804) },
70 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_58808) },
71 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_58802_VF) },
72 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57508) },
73 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57504) },
74 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57502) },
75 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57500_VF1) },
76 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57500_VF2) },
77 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57508_MF1) },
78 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57504_MF1) },
79 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57502_MF1) },
80 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57508_MF2) },
81 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57504_MF2) },
82 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57502_MF2) },
83 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_58812) },
84 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_58814) },
85 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_58818) },
86 	{ RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_58818_VF) },
87 	{ .vendor_id = 0, /* sentinel */ },
88 };
89 
90 #define BNXT_DEVARG_FLOW_XSTAT	"flow-xstat"
91 #define BNXT_DEVARG_MAX_NUM_KFLOWS  "max-num-kflows"
92 #define BNXT_DEVARG_REPRESENTOR	"representor"
93 #define BNXT_DEVARG_REP_BASED_PF  "rep-based-pf"
94 #define BNXT_DEVARG_REP_IS_PF  "rep-is-pf"
95 #define BNXT_DEVARG_REP_Q_R2F  "rep-q-r2f"
96 #define BNXT_DEVARG_REP_Q_F2R  "rep-q-f2r"
97 #define BNXT_DEVARG_REP_FC_R2F  "rep-fc-r2f"
98 #define BNXT_DEVARG_REP_FC_F2R  "rep-fc-f2r"
99 #define BNXT_DEVARG_APP_ID	"app-id"
100 
101 static const char *const bnxt_dev_args[] = {
102 	BNXT_DEVARG_REPRESENTOR,
103 	BNXT_DEVARG_FLOW_XSTAT,
104 	BNXT_DEVARG_MAX_NUM_KFLOWS,
105 	BNXT_DEVARG_REP_BASED_PF,
106 	BNXT_DEVARG_REP_IS_PF,
107 	BNXT_DEVARG_REP_Q_R2F,
108 	BNXT_DEVARG_REP_Q_F2R,
109 	BNXT_DEVARG_REP_FC_R2F,
110 	BNXT_DEVARG_REP_FC_F2R,
111 	BNXT_DEVARG_APP_ID,
112 	NULL
113 };
114 
115 /*
116  * app-id = an non-negative 8-bit number
117  */
118 #define BNXT_DEVARG_APP_ID_INVALID(val)			((val) > 255)
119 
120 /*
121  * flow_xstat == false to disable the feature
122  * flow_xstat == true to enable the feature
123  */
124 #define	BNXT_DEVARG_FLOW_XSTAT_INVALID(flow_xstat)	((flow_xstat) > 1)
125 
126 /*
127  * rep_is_pf == false to indicate VF representor
128  * rep_is_pf == true to indicate PF representor
129  */
130 #define	BNXT_DEVARG_REP_IS_PF_INVALID(rep_is_pf)	((rep_is_pf) > 1)
131 
132 /*
133  * rep_based_pf == Physical index of the PF
134  */
135 #define	BNXT_DEVARG_REP_BASED_PF_INVALID(rep_based_pf)	((rep_based_pf) > 15)
136 /*
137  * rep_q_r2f == Logical COS Queue index for the rep to endpoint direction
138  */
139 #define	BNXT_DEVARG_REP_Q_R2F_INVALID(rep_q_r2f)	((rep_q_r2f) > 3)
140 
141 /*
142  * rep_q_f2r == Logical COS Queue index for the endpoint to rep direction
143  */
144 #define	BNXT_DEVARG_REP_Q_F2R_INVALID(rep_q_f2r)	((rep_q_f2r) > 3)
145 
146 /*
147  * rep_fc_r2f == Flow control for the representor to endpoint direction
148  */
149 #define BNXT_DEVARG_REP_FC_R2F_INVALID(rep_fc_r2f)	((rep_fc_r2f) > 1)
150 
151 /*
152  * rep_fc_f2r == Flow control for the endpoint to representor direction
153  */
154 #define BNXT_DEVARG_REP_FC_F2R_INVALID(rep_fc_f2r)	((rep_fc_f2r) > 1)
155 
156 int bnxt_cfa_code_dynfield_offset = -1;
157 
158 /*
159  * max_num_kflows must be >= 32
160  * and must be a power-of-2 supported value
161  * return: 1 -> invalid
162  *         0 -> valid
163  */
164 static int bnxt_devarg_max_num_kflow_invalid(uint16_t max_num_kflows)
165 {
166 	if (max_num_kflows < 32 || !rte_is_power_of_2(max_num_kflows))
167 		return 1;
168 	return 0;
169 }
170 
171 static int bnxt_vlan_offload_set_op(struct rte_eth_dev *dev, int mask);
172 static int bnxt_dev_uninit(struct rte_eth_dev *eth_dev);
173 static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev);
174 static int bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev);
175 static void bnxt_cancel_fw_health_check(struct bnxt *bp);
176 static int bnxt_restore_vlan_filters(struct bnxt *bp);
177 static void bnxt_dev_recover(void *arg);
178 static void bnxt_free_error_recovery_info(struct bnxt *bp);
179 static void bnxt_free_rep_info(struct bnxt *bp);
180 static int bnxt_check_fw_ready(struct bnxt *bp);
181 
182 int is_bnxt_in_error(struct bnxt *bp)
183 {
184 	if (bp->flags & BNXT_FLAG_FATAL_ERROR)
185 		return -EIO;
186 	if (bp->flags & BNXT_FLAG_FW_RESET)
187 		return -EBUSY;
188 
189 	return 0;
190 }
191 
192 /***********************/
193 
194 /*
195  * High level utility functions
196  */
197 
198 static uint16_t bnxt_rss_ctxts(const struct bnxt *bp)
199 {
200 	unsigned int num_rss_rings = RTE_MIN(bp->rx_nr_rings,
201 					     BNXT_RSS_TBL_SIZE_P5);
202 
203 	if (!BNXT_CHIP_P5(bp))
204 		return 1;
205 
206 	return RTE_ALIGN_MUL_CEIL(num_rss_rings,
207 				  BNXT_RSS_ENTRIES_PER_CTX_P5) /
208 				  BNXT_RSS_ENTRIES_PER_CTX_P5;
209 }
210 
211 uint16_t bnxt_rss_hash_tbl_size(const struct bnxt *bp)
212 {
213 	if (!BNXT_CHIP_P5(bp))
214 		return HW_HASH_INDEX_SIZE;
215 
216 	return bnxt_rss_ctxts(bp) * BNXT_RSS_ENTRIES_PER_CTX_P5;
217 }
218 
219 static void bnxt_free_parent_info(struct bnxt *bp)
220 {
221 	rte_free(bp->parent);
222 	bp->parent = NULL;
223 }
224 
225 static void bnxt_free_pf_info(struct bnxt *bp)
226 {
227 	rte_free(bp->pf);
228 	bp->pf = NULL;
229 }
230 
231 static void bnxt_free_link_info(struct bnxt *bp)
232 {
233 	rte_free(bp->link_info);
234 	bp->link_info = NULL;
235 }
236 
237 static void bnxt_free_leds_info(struct bnxt *bp)
238 {
239 	if (BNXT_VF(bp))
240 		return;
241 
242 	rte_free(bp->leds);
243 	bp->leds = NULL;
244 }
245 
246 static void bnxt_free_flow_stats_info(struct bnxt *bp)
247 {
248 	rte_free(bp->flow_stat);
249 	bp->flow_stat = NULL;
250 }
251 
252 static void bnxt_free_cos_queues(struct bnxt *bp)
253 {
254 	rte_free(bp->rx_cos_queue);
255 	bp->rx_cos_queue = NULL;
256 	rte_free(bp->tx_cos_queue);
257 	bp->tx_cos_queue = NULL;
258 }
259 
260 static void bnxt_free_mem(struct bnxt *bp, bool reconfig)
261 {
262 	bnxt_free_filter_mem(bp);
263 	bnxt_free_vnic_attributes(bp);
264 	bnxt_free_vnic_mem(bp);
265 
266 	/* tx/rx rings are configured as part of *_queue_setup callbacks.
267 	 * If the number of rings change across fw update,
268 	 * we don't have much choice except to warn the user.
269 	 */
270 	if (!reconfig) {
271 		bnxt_free_stats(bp);
272 		bnxt_free_tx_rings(bp);
273 		bnxt_free_rx_rings(bp);
274 	}
275 	bnxt_free_async_cp_ring(bp);
276 	bnxt_free_rxtx_nq_ring(bp);
277 
278 	rte_free(bp->grp_info);
279 	bp->grp_info = NULL;
280 }
281 
282 static int bnxt_alloc_parent_info(struct bnxt *bp)
283 {
284 	bp->parent = rte_zmalloc("bnxt_parent_info",
285 				 sizeof(struct bnxt_parent_info), 0);
286 	if (bp->parent == NULL)
287 		return -ENOMEM;
288 
289 	return 0;
290 }
291 
292 static int bnxt_alloc_pf_info(struct bnxt *bp)
293 {
294 	bp->pf = rte_zmalloc("bnxt_pf_info", sizeof(struct bnxt_pf_info), 0);
295 	if (bp->pf == NULL)
296 		return -ENOMEM;
297 
298 	return 0;
299 }
300 
301 static int bnxt_alloc_link_info(struct bnxt *bp)
302 {
303 	bp->link_info =
304 		rte_zmalloc("bnxt_link_info", sizeof(struct bnxt_link_info), 0);
305 	if (bp->link_info == NULL)
306 		return -ENOMEM;
307 
308 	return 0;
309 }
310 
311 static int bnxt_alloc_leds_info(struct bnxt *bp)
312 {
313 	if (BNXT_VF(bp))
314 		return 0;
315 
316 	bp->leds = rte_zmalloc("bnxt_leds",
317 			       BNXT_MAX_LED * sizeof(struct bnxt_led_info),
318 			       0);
319 	if (bp->leds == NULL)
320 		return -ENOMEM;
321 
322 	return 0;
323 }
324 
325 static int bnxt_alloc_cos_queues(struct bnxt *bp)
326 {
327 	bp->rx_cos_queue =
328 		rte_zmalloc("bnxt_rx_cosq",
329 			    BNXT_COS_QUEUE_COUNT *
330 			    sizeof(struct bnxt_cos_queue_info),
331 			    0);
332 	if (bp->rx_cos_queue == NULL)
333 		return -ENOMEM;
334 
335 	bp->tx_cos_queue =
336 		rte_zmalloc("bnxt_tx_cosq",
337 			    BNXT_COS_QUEUE_COUNT *
338 			    sizeof(struct bnxt_cos_queue_info),
339 			    0);
340 	if (bp->tx_cos_queue == NULL)
341 		return -ENOMEM;
342 
343 	return 0;
344 }
345 
346 static int bnxt_alloc_flow_stats_info(struct bnxt *bp)
347 {
348 	bp->flow_stat = rte_zmalloc("bnxt_flow_xstat",
349 				    sizeof(struct bnxt_flow_stat_info), 0);
350 	if (bp->flow_stat == NULL)
351 		return -ENOMEM;
352 
353 	return 0;
354 }
355 
356 static int bnxt_alloc_mem(struct bnxt *bp, bool reconfig)
357 {
358 	int rc;
359 
360 	rc = bnxt_alloc_ring_grps(bp);
361 	if (rc)
362 		goto alloc_mem_err;
363 
364 	rc = bnxt_alloc_async_ring_struct(bp);
365 	if (rc)
366 		goto alloc_mem_err;
367 
368 	rc = bnxt_alloc_vnic_mem(bp);
369 	if (rc)
370 		goto alloc_mem_err;
371 
372 	rc = bnxt_alloc_vnic_attributes(bp, reconfig);
373 	if (rc)
374 		goto alloc_mem_err;
375 
376 	rc = bnxt_alloc_filter_mem(bp);
377 	if (rc)
378 		goto alloc_mem_err;
379 
380 	rc = bnxt_alloc_async_cp_ring(bp);
381 	if (rc)
382 		goto alloc_mem_err;
383 
384 	rc = bnxt_alloc_rxtx_nq_ring(bp);
385 	if (rc)
386 		goto alloc_mem_err;
387 
388 	if (BNXT_FLOW_XSTATS_EN(bp)) {
389 		rc = bnxt_alloc_flow_stats_info(bp);
390 		if (rc)
391 			goto alloc_mem_err;
392 	}
393 
394 	return 0;
395 
396 alloc_mem_err:
397 	bnxt_free_mem(bp, reconfig);
398 	return rc;
399 }
400 
401 static int bnxt_setup_one_vnic(struct bnxt *bp, uint16_t vnic_id)
402 {
403 	struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
404 	struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
405 	uint64_t rx_offloads = dev_conf->rxmode.offloads;
406 	struct bnxt_rx_queue *rxq;
407 	unsigned int j;
408 	int rc;
409 
410 	rc = bnxt_vnic_grp_alloc(bp, vnic);
411 	if (rc)
412 		goto err_out;
413 
414 	PMD_DRV_LOG(DEBUG, "vnic[%d] = %p vnic->fw_grp_ids = %p\n",
415 		    vnic_id, vnic, vnic->fw_grp_ids);
416 
417 	rc = bnxt_hwrm_vnic_alloc(bp, vnic);
418 	if (rc)
419 		goto err_out;
420 
421 	/* Alloc RSS context only if RSS mode is enabled */
422 	if (dev_conf->rxmode.mq_mode & RTE_ETH_MQ_RX_RSS) {
423 		int j, nr_ctxs = bnxt_rss_ctxts(bp);
424 
425 		/* RSS table size in Thor is 512.
426 		 * Cap max Rx rings to same value
427 		 */
428 		if (bp->rx_nr_rings > BNXT_RSS_TBL_SIZE_P5) {
429 			PMD_DRV_LOG(ERR, "RxQ cnt %d > reta_size %d\n",
430 				    bp->rx_nr_rings, BNXT_RSS_TBL_SIZE_P5);
431 			goto err_out;
432 		}
433 
434 		rc = 0;
435 		for (j = 0; j < nr_ctxs; j++) {
436 			rc = bnxt_hwrm_vnic_ctx_alloc(bp, vnic, j);
437 			if (rc)
438 				break;
439 		}
440 		if (rc) {
441 			PMD_DRV_LOG(ERR,
442 				    "HWRM vnic %d ctx %d alloc failure rc: %x\n",
443 				    vnic_id, j, rc);
444 			goto err_out;
445 		}
446 		vnic->num_lb_ctxts = nr_ctxs;
447 	}
448 
449 	/*
450 	 * Firmware sets pf pair in default vnic cfg. If the VLAN strip
451 	 * setting is not available at this time, it will not be
452 	 * configured correctly in the CFA.
453 	 */
454 	if (rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP)
455 		vnic->vlan_strip = true;
456 	else
457 		vnic->vlan_strip = false;
458 
459 	rc = bnxt_hwrm_vnic_cfg(bp, vnic);
460 	if (rc)
461 		goto err_out;
462 
463 	rc = bnxt_set_hwrm_vnic_filters(bp, vnic);
464 	if (rc)
465 		goto err_out;
466 
467 	for (j = 0; j < bp->rx_num_qs_per_vnic; j++) {
468 		rxq = bp->eth_dev->data->rx_queues[j];
469 
470 		PMD_DRV_LOG(DEBUG,
471 			    "rxq[%d]->vnic=%p vnic->fw_grp_ids=%p\n",
472 			    j, rxq->vnic, rxq->vnic->fw_grp_ids);
473 
474 		if (BNXT_HAS_RING_GRPS(bp) && rxq->rx_deferred_start)
475 			rxq->vnic->fw_grp_ids[j] = INVALID_HW_RING_ID;
476 		else
477 			vnic->rx_queue_cnt++;
478 	}
479 
480 	PMD_DRV_LOG(DEBUG, "vnic->rx_queue_cnt = %d\n", vnic->rx_queue_cnt);
481 
482 	rc = bnxt_vnic_rss_configure(bp, vnic);
483 	if (rc)
484 		goto err_out;
485 
486 	bnxt_hwrm_vnic_plcmode_cfg(bp, vnic);
487 
488 	rc = bnxt_hwrm_vnic_tpa_cfg(bp, vnic,
489 				    (rx_offloads & RTE_ETH_RX_OFFLOAD_TCP_LRO) ?
490 				    true : false);
491 	if (rc)
492 		goto err_out;
493 
494 	return 0;
495 err_out:
496 	PMD_DRV_LOG(ERR, "HWRM vnic %d cfg failure rc: %x\n",
497 		    vnic_id, rc);
498 	return rc;
499 }
500 
501 static int bnxt_register_fc_ctx_mem(struct bnxt *bp)
502 {
503 	int rc = 0;
504 
505 	rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->rx_fc_in_tbl.dma,
506 				&bp->flow_stat->rx_fc_in_tbl.ctx_id);
507 	if (rc)
508 		return rc;
509 
510 	PMD_DRV_LOG(DEBUG,
511 		    "rx_fc_in_tbl.va = %p rx_fc_in_tbl.dma = %p"
512 		    " rx_fc_in_tbl.ctx_id = %d\n",
513 		    bp->flow_stat->rx_fc_in_tbl.va,
514 		    (void *)((uintptr_t)bp->flow_stat->rx_fc_in_tbl.dma),
515 		    bp->flow_stat->rx_fc_in_tbl.ctx_id);
516 
517 	rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->rx_fc_out_tbl.dma,
518 				&bp->flow_stat->rx_fc_out_tbl.ctx_id);
519 	if (rc)
520 		return rc;
521 
522 	PMD_DRV_LOG(DEBUG,
523 		    "rx_fc_out_tbl.va = %p rx_fc_out_tbl.dma = %p"
524 		    " rx_fc_out_tbl.ctx_id = %d\n",
525 		    bp->flow_stat->rx_fc_out_tbl.va,
526 		    (void *)((uintptr_t)bp->flow_stat->rx_fc_out_tbl.dma),
527 		    bp->flow_stat->rx_fc_out_tbl.ctx_id);
528 
529 	rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->tx_fc_in_tbl.dma,
530 				&bp->flow_stat->tx_fc_in_tbl.ctx_id);
531 	if (rc)
532 		return rc;
533 
534 	PMD_DRV_LOG(DEBUG,
535 		    "tx_fc_in_tbl.va = %p tx_fc_in_tbl.dma = %p"
536 		    " tx_fc_in_tbl.ctx_id = %d\n",
537 		    bp->flow_stat->tx_fc_in_tbl.va,
538 		    (void *)((uintptr_t)bp->flow_stat->tx_fc_in_tbl.dma),
539 		    bp->flow_stat->tx_fc_in_tbl.ctx_id);
540 
541 	rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->tx_fc_out_tbl.dma,
542 				&bp->flow_stat->tx_fc_out_tbl.ctx_id);
543 	if (rc)
544 		return rc;
545 
546 	PMD_DRV_LOG(DEBUG,
547 		    "tx_fc_out_tbl.va = %p tx_fc_out_tbl.dma = %p"
548 		    " tx_fc_out_tbl.ctx_id = %d\n",
549 		    bp->flow_stat->tx_fc_out_tbl.va,
550 		    (void *)((uintptr_t)bp->flow_stat->tx_fc_out_tbl.dma),
551 		    bp->flow_stat->tx_fc_out_tbl.ctx_id);
552 
553 	memset(bp->flow_stat->rx_fc_out_tbl.va,
554 	       0,
555 	       bp->flow_stat->rx_fc_out_tbl.size);
556 	rc = bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_RX,
557 				       CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
558 				       bp->flow_stat->rx_fc_out_tbl.ctx_id,
559 				       bp->flow_stat->max_fc,
560 				       true);
561 	if (rc)
562 		return rc;
563 
564 	memset(bp->flow_stat->tx_fc_out_tbl.va,
565 	       0,
566 	       bp->flow_stat->tx_fc_out_tbl.size);
567 	rc = bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_TX,
568 				       CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
569 				       bp->flow_stat->tx_fc_out_tbl.ctx_id,
570 				       bp->flow_stat->max_fc,
571 				       true);
572 
573 	return rc;
574 }
575 
576 static int bnxt_alloc_ctx_mem_buf(struct bnxt *bp, char *type, size_t size,
577 				  struct bnxt_ctx_mem_buf_info *ctx)
578 {
579 	if (!ctx)
580 		return -EINVAL;
581 
582 	ctx->va = rte_zmalloc_socket(type, size, 0,
583 				     bp->eth_dev->device->numa_node);
584 	if (ctx->va == NULL)
585 		return -ENOMEM;
586 	rte_mem_lock_page(ctx->va);
587 	ctx->size = size;
588 	ctx->dma = rte_mem_virt2iova(ctx->va);
589 	if (ctx->dma == RTE_BAD_IOVA)
590 		return -ENOMEM;
591 
592 	return 0;
593 }
594 
595 static int bnxt_init_fc_ctx_mem(struct bnxt *bp)
596 {
597 	struct rte_pci_device *pdev = bp->pdev;
598 	char type[RTE_MEMZONE_NAMESIZE];
599 	uint16_t max_fc;
600 	int rc = 0;
601 
602 	max_fc = bp->flow_stat->max_fc;
603 
604 	sprintf(type, "bnxt_rx_fc_in_" PCI_PRI_FMT, pdev->addr.domain,
605 		pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
606 	/* 4 bytes for each counter-id */
607 	rc = bnxt_alloc_ctx_mem_buf(bp, type,
608 				    max_fc * 4,
609 				    &bp->flow_stat->rx_fc_in_tbl);
610 	if (rc)
611 		return rc;
612 
613 	sprintf(type, "bnxt_rx_fc_out_" PCI_PRI_FMT, pdev->addr.domain,
614 		pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
615 	/* 16 bytes for each counter - 8 bytes pkt_count, 8 bytes byte_count */
616 	rc = bnxt_alloc_ctx_mem_buf(bp, type,
617 				    max_fc * 16,
618 				    &bp->flow_stat->rx_fc_out_tbl);
619 	if (rc)
620 		return rc;
621 
622 	sprintf(type, "bnxt_tx_fc_in_" PCI_PRI_FMT, pdev->addr.domain,
623 		pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
624 	/* 4 bytes for each counter-id */
625 	rc = bnxt_alloc_ctx_mem_buf(bp, type,
626 				    max_fc * 4,
627 				    &bp->flow_stat->tx_fc_in_tbl);
628 	if (rc)
629 		return rc;
630 
631 	sprintf(type, "bnxt_tx_fc_out_" PCI_PRI_FMT, pdev->addr.domain,
632 		pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
633 	/* 16 bytes for each counter - 8 bytes pkt_count, 8 bytes byte_count */
634 	rc = bnxt_alloc_ctx_mem_buf(bp, type,
635 				    max_fc * 16,
636 				    &bp->flow_stat->tx_fc_out_tbl);
637 	if (rc)
638 		return rc;
639 
640 	rc = bnxt_register_fc_ctx_mem(bp);
641 
642 	return rc;
643 }
644 
645 static int bnxt_init_ctx_mem(struct bnxt *bp)
646 {
647 	int rc = 0;
648 
649 	if (!(bp->fw_cap & BNXT_FW_CAP_ADV_FLOW_COUNTERS) ||
650 	    !(BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) ||
651 	    !BNXT_FLOW_XSTATS_EN(bp))
652 		return 0;
653 
654 	rc = bnxt_hwrm_cfa_counter_qcaps(bp, &bp->flow_stat->max_fc);
655 	if (rc)
656 		return rc;
657 
658 	rc = bnxt_init_fc_ctx_mem(bp);
659 
660 	return rc;
661 }
662 
663 static inline bool bnxt_force_link_config(struct bnxt *bp)
664 {
665 	uint16_t subsystem_device_id = bp->pdev->id.subsystem_device_id;
666 
667 	switch (subsystem_device_id) {
668 	case BROADCOM_DEV_957508_N2100:
669 	case BROADCOM_DEV_957414_N225:
670 		return true;
671 	default:
672 		return false;
673 	}
674 }
675 
676 static int bnxt_update_phy_setting(struct bnxt *bp)
677 {
678 	struct rte_eth_link new;
679 	int rc;
680 
681 	rc = bnxt_get_hwrm_link_config(bp, &new);
682 	if (rc) {
683 		PMD_DRV_LOG(ERR, "Failed to get link settings\n");
684 		return rc;
685 	}
686 
687 	/*
688 	 * Device is not obliged link down in certain scenarios, even
689 	 * when forced. When FW does not allow any user other than BMC
690 	 * to shutdown the port, bnxt_get_hwrm_link_config() call always
691 	 * returns link up. Force phy update always in that case.
692 	 */
693 	if (!new.link_status || bnxt_force_link_config(bp)) {
694 		rc = bnxt_set_hwrm_link_config(bp, true);
695 		if (rc) {
696 			PMD_DRV_LOG(ERR, "Failed to update PHY settings\n");
697 			return rc;
698 		}
699 	}
700 
701 	return rc;
702 }
703 
704 static void bnxt_free_prev_ring_stats(struct bnxt *bp)
705 {
706 	rte_free(bp->prev_rx_ring_stats);
707 	rte_free(bp->prev_tx_ring_stats);
708 
709 	bp->prev_rx_ring_stats = NULL;
710 	bp->prev_tx_ring_stats = NULL;
711 }
712 
713 static int bnxt_alloc_prev_ring_stats(struct bnxt *bp)
714 {
715 	bp->prev_rx_ring_stats =  rte_zmalloc("bnxt_prev_rx_ring_stats",
716 					      sizeof(struct bnxt_ring_stats) *
717 					      bp->rx_cp_nr_rings,
718 					      0);
719 	if (bp->prev_rx_ring_stats == NULL)
720 		return -ENOMEM;
721 
722 	bp->prev_tx_ring_stats = rte_zmalloc("bnxt_prev_tx_ring_stats",
723 					     sizeof(struct bnxt_ring_stats) *
724 					     bp->tx_cp_nr_rings,
725 					     0);
726 	if (bp->prev_tx_ring_stats == NULL)
727 		goto error;
728 
729 	return 0;
730 
731 error:
732 	bnxt_free_prev_ring_stats(bp);
733 	return -ENOMEM;
734 }
735 
736 static int bnxt_start_nic(struct bnxt *bp)
737 {
738 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(bp->eth_dev);
739 	struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
740 	uint32_t intr_vector = 0;
741 	uint32_t queue_id, base = BNXT_MISC_VEC_ID;
742 	uint32_t vec = BNXT_MISC_VEC_ID;
743 	unsigned int i, j;
744 	int rc;
745 
746 	if (bp->eth_dev->data->mtu > RTE_ETHER_MTU)
747 		bp->flags |= BNXT_FLAG_JUMBO;
748 	else
749 		bp->flags &= ~BNXT_FLAG_JUMBO;
750 
751 	/* THOR does not support ring groups.
752 	 * But we will use the array to save RSS context IDs.
753 	 */
754 	if (BNXT_CHIP_P5(bp))
755 		bp->max_ring_grps = BNXT_MAX_RSS_CTXTS_P5;
756 
757 	rc = bnxt_alloc_hwrm_rings(bp);
758 	if (rc) {
759 		PMD_DRV_LOG(ERR, "HWRM ring alloc failure rc: %x\n", rc);
760 		goto err_out;
761 	}
762 
763 	rc = bnxt_alloc_all_hwrm_ring_grps(bp);
764 	if (rc) {
765 		PMD_DRV_LOG(ERR, "HWRM ring grp alloc failure: %x\n", rc);
766 		goto err_out;
767 	}
768 
769 	if (!(bp->vnic_cap_flags & BNXT_VNIC_CAP_COS_CLASSIFY))
770 		goto skip_cosq_cfg;
771 
772 	for (j = 0, i = 0; i < BNXT_COS_QUEUE_COUNT; i++) {
773 		if (bp->rx_cos_queue[i].id != 0xff) {
774 			struct bnxt_vnic_info *vnic = &bp->vnic_info[j++];
775 
776 			if (!vnic) {
777 				PMD_DRV_LOG(ERR,
778 					    "Num pools more than FW profile\n");
779 				rc = -EINVAL;
780 				goto err_out;
781 			}
782 			vnic->cos_queue_id = bp->rx_cos_queue[i].id;
783 			bp->rx_cosq_cnt++;
784 		}
785 	}
786 
787 skip_cosq_cfg:
788 	rc = bnxt_mq_rx_configure(bp);
789 	if (rc) {
790 		PMD_DRV_LOG(ERR, "MQ mode configure failure rc: %x\n", rc);
791 		goto err_out;
792 	}
793 
794 	for (j = 0; j < bp->rx_nr_rings; j++) {
795 		struct bnxt_rx_queue *rxq = bp->rx_queues[j];
796 
797 		if (!rxq->rx_deferred_start) {
798 			bp->eth_dev->data->rx_queue_state[j] =
799 				RTE_ETH_QUEUE_STATE_STARTED;
800 			rxq->rx_started = true;
801 		}
802 	}
803 
804 	/* VNIC configuration */
805 	for (i = 0; i < bp->nr_vnics; i++) {
806 		rc = bnxt_setup_one_vnic(bp, i);
807 		if (rc)
808 			goto err_out;
809 	}
810 
811 	for (j = 0; j < bp->tx_nr_rings; j++) {
812 		struct bnxt_tx_queue *txq = bp->tx_queues[j];
813 
814 		if (!txq->tx_deferred_start) {
815 			bp->eth_dev->data->tx_queue_state[j] =
816 				RTE_ETH_QUEUE_STATE_STARTED;
817 			txq->tx_started = true;
818 		}
819 	}
820 
821 	rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, &bp->vnic_info[0], 0, NULL);
822 	if (rc) {
823 		PMD_DRV_LOG(ERR,
824 			"HWRM cfa l2 rx mask failure rc: %x\n", rc);
825 		goto err_out;
826 	}
827 
828 	/* check and configure queue intr-vector mapping */
829 	if ((rte_intr_cap_multiple(intr_handle) ||
830 	     !RTE_ETH_DEV_SRIOV(bp->eth_dev).active) &&
831 	    bp->eth_dev->data->dev_conf.intr_conf.rxq != 0) {
832 		intr_vector = bp->eth_dev->data->nb_rx_queues;
833 		PMD_DRV_LOG(DEBUG, "intr_vector = %d\n", intr_vector);
834 		if (intr_vector > bp->rx_cp_nr_rings) {
835 			PMD_DRV_LOG(ERR, "At most %d intr queues supported",
836 					bp->rx_cp_nr_rings);
837 			return -ENOTSUP;
838 		}
839 		rc = rte_intr_efd_enable(intr_handle, intr_vector);
840 		if (rc)
841 			return rc;
842 	}
843 
844 	if (rte_intr_dp_is_en(intr_handle)) {
845 		if (rte_intr_vec_list_alloc(intr_handle, "intr_vec",
846 					bp->eth_dev->data->nb_rx_queues)) {
847 			PMD_DRV_LOG(ERR, "Failed to allocate %d rx_queues"
848 				" intr_vec", bp->eth_dev->data->nb_rx_queues);
849 			rc = -ENOMEM;
850 			goto err_out;
851 		}
852 		PMD_DRV_LOG(DEBUG, "intr_handle->nb_efd = %d "
853 			    "intr_handle->max_intr = %d\n",
854 			    rte_intr_nb_efd_get(intr_handle),
855 			    rte_intr_max_intr_get(intr_handle));
856 		for (queue_id = 0; queue_id < bp->eth_dev->data->nb_rx_queues;
857 		     queue_id++) {
858 			rte_intr_vec_list_index_set(intr_handle,
859 					queue_id, vec + BNXT_RX_VEC_START);
860 			if (vec < base + rte_intr_nb_efd_get(intr_handle)
861 			    - 1)
862 				vec++;
863 		}
864 	}
865 
866 	/* enable uio/vfio intr/eventfd mapping */
867 	rc = rte_intr_enable(intr_handle);
868 #ifndef RTE_EXEC_ENV_FREEBSD
869 	/* In FreeBSD OS, nic_uio driver does not support interrupts */
870 	if (rc)
871 		goto err_out;
872 #endif
873 
874 	rc = bnxt_update_phy_setting(bp);
875 	if (rc)
876 		goto err_out;
877 
878 	bp->mark_table = rte_zmalloc("bnxt_mark_table", BNXT_MARK_TABLE_SZ, 0);
879 	if (!bp->mark_table)
880 		PMD_DRV_LOG(ERR, "Allocation of mark table failed\n");
881 
882 	return 0;
883 
884 err_out:
885 	/* Some of the error status returned by FW may not be from errno.h */
886 	if (rc > 0)
887 		rc = -EIO;
888 
889 	return rc;
890 }
891 
892 static int bnxt_shutdown_nic(struct bnxt *bp)
893 {
894 	bnxt_free_all_hwrm_resources(bp);
895 	bnxt_free_all_filters(bp);
896 	bnxt_free_all_vnics(bp);
897 	return 0;
898 }
899 
900 /*
901  * Device configuration and status function
902  */
903 
904 uint32_t bnxt_get_speed_capabilities(struct bnxt *bp)
905 {
906 	uint32_t link_speed = 0;
907 	uint32_t speed_capa = 0;
908 
909 	if (bp->link_info == NULL)
910 		return 0;
911 
912 	link_speed = bp->link_info->support_speeds;
913 
914 	/* If PAM4 is configured, use PAM4 supported speed */
915 	if (link_speed == 0 && bp->link_info->support_pam4_speeds > 0)
916 		link_speed = bp->link_info->support_pam4_speeds;
917 
918 	if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_100MB)
919 		speed_capa |= RTE_ETH_LINK_SPEED_100M;
920 	if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_100MBHD)
921 		speed_capa |= RTE_ETH_LINK_SPEED_100M_HD;
922 	if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_1GB)
923 		speed_capa |= RTE_ETH_LINK_SPEED_1G;
924 	if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_2_5GB)
925 		speed_capa |= RTE_ETH_LINK_SPEED_2_5G;
926 	if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_10GB)
927 		speed_capa |= RTE_ETH_LINK_SPEED_10G;
928 	if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_20GB)
929 		speed_capa |= RTE_ETH_LINK_SPEED_20G;
930 	if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_25GB)
931 		speed_capa |= RTE_ETH_LINK_SPEED_25G;
932 	if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_40GB)
933 		speed_capa |= RTE_ETH_LINK_SPEED_40G;
934 	if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_50GB)
935 		speed_capa |= RTE_ETH_LINK_SPEED_50G;
936 	if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_100GB)
937 		speed_capa |= RTE_ETH_LINK_SPEED_100G;
938 	if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_50G)
939 		speed_capa |= RTE_ETH_LINK_SPEED_50G;
940 	if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_100G)
941 		speed_capa |= RTE_ETH_LINK_SPEED_100G;
942 	if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_200G)
943 		speed_capa |= RTE_ETH_LINK_SPEED_200G;
944 
945 	if (bp->link_info->auto_mode ==
946 	    HWRM_PORT_PHY_QCFG_OUTPUT_AUTO_MODE_NONE)
947 		speed_capa |= RTE_ETH_LINK_SPEED_FIXED;
948 
949 	return speed_capa;
950 }
951 
952 static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
953 				struct rte_eth_dev_info *dev_info)
954 {
955 	struct rte_pci_device *pdev = RTE_DEV_TO_PCI(eth_dev->device);
956 	struct bnxt *bp = eth_dev->data->dev_private;
957 	uint16_t max_vnics, i, j, vpool, vrxq;
958 	unsigned int max_rx_rings;
959 	int rc;
960 
961 	rc = is_bnxt_in_error(bp);
962 	if (rc)
963 		return rc;
964 
965 	/* MAC Specifics */
966 	dev_info->max_mac_addrs = RTE_MIN(bp->max_l2_ctx, RTE_ETH_NUM_RECEIVE_MAC_ADDR);
967 	dev_info->max_hash_mac_addrs = 0;
968 
969 	/* PF/VF specifics */
970 	if (BNXT_PF(bp))
971 		dev_info->max_vfs = pdev->max_vfs;
972 
973 	max_rx_rings = bnxt_max_rings(bp);
974 	/* For the sake of symmetry, max_rx_queues = max_tx_queues */
975 	dev_info->max_rx_queues = max_rx_rings;
976 	dev_info->max_tx_queues = max_rx_rings;
977 	dev_info->reta_size = bnxt_rss_hash_tbl_size(bp);
978 	dev_info->hash_key_size = HW_HASH_KEY_SIZE;
979 	max_vnics = bp->max_vnics;
980 
981 	/* MTU specifics */
982 	dev_info->min_mtu = RTE_ETHER_MIN_MTU;
983 	dev_info->max_mtu = BNXT_MAX_MTU;
984 
985 	/* Fast path specifics */
986 	dev_info->min_rx_bufsize = 1;
987 	dev_info->max_rx_pktlen = BNXT_MAX_PKT_LEN;
988 
989 	dev_info->rx_offload_capa = bnxt_get_rx_port_offloads(bp);
990 	dev_info->tx_queue_offload_capa = RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
991 	dev_info->tx_offload_capa = bnxt_get_tx_port_offloads(bp) |
992 				    dev_info->tx_queue_offload_capa;
993 	dev_info->flow_type_rss_offloads = BNXT_ETH_RSS_SUPPORT;
994 
995 	dev_info->speed_capa = bnxt_get_speed_capabilities(bp);
996 	dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
997 			     RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
998 	dev_info->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
999 
1000 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
1001 		.rx_thresh = {
1002 			.pthresh = 8,
1003 			.hthresh = 8,
1004 			.wthresh = 0,
1005 		},
1006 		.rx_free_thresh = 32,
1007 		.rx_drop_en = BNXT_DEFAULT_RX_DROP_EN,
1008 	};
1009 
1010 	dev_info->default_txconf = (struct rte_eth_txconf) {
1011 		.tx_thresh = {
1012 			.pthresh = 32,
1013 			.hthresh = 0,
1014 			.wthresh = 0,
1015 		},
1016 		.tx_free_thresh = 32,
1017 		.tx_rs_thresh = 32,
1018 	};
1019 	eth_dev->data->dev_conf.intr_conf.lsc = 1;
1020 
1021 	dev_info->rx_desc_lim.nb_min = BNXT_MIN_RING_DESC;
1022 	dev_info->rx_desc_lim.nb_max = BNXT_MAX_RX_RING_DESC;
1023 	dev_info->tx_desc_lim.nb_min = BNXT_MIN_RING_DESC;
1024 	dev_info->tx_desc_lim.nb_max = BNXT_MAX_TX_RING_DESC;
1025 
1026 	if (BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) {
1027 		dev_info->switch_info.name = eth_dev->device->name;
1028 		dev_info->switch_info.domain_id = bp->switch_domain_id;
1029 		dev_info->switch_info.port_id =
1030 				BNXT_PF(bp) ? BNXT_SWITCH_PORT_ID_PF :
1031 				    BNXT_SWITCH_PORT_ID_TRUSTED_VF;
1032 	}
1033 
1034 	/*
1035 	 * TODO: default_rxconf, default_txconf, rx_desc_lim, and tx_desc_lim
1036 	 *       need further investigation.
1037 	 */
1038 
1039 	/* VMDq resources */
1040 	vpool = 64; /* RTE_ETH_64_POOLS */
1041 	vrxq = 128; /* RTE_ETH_VMDQ_DCB_NUM_QUEUES */
1042 	for (i = 0; i < 4; vpool >>= 1, i++) {
1043 		if (max_vnics > vpool) {
1044 			for (j = 0; j < 5; vrxq >>= 1, j++) {
1045 				if (dev_info->max_rx_queues > vrxq) {
1046 					if (vpool > vrxq)
1047 						vpool = vrxq;
1048 					goto found;
1049 				}
1050 			}
1051 			/* Not enough resources to support VMDq */
1052 			break;
1053 		}
1054 	}
1055 	/* Not enough resources to support VMDq */
1056 	vpool = 0;
1057 	vrxq = 0;
1058 found:
1059 	dev_info->max_vmdq_pools = vpool;
1060 	dev_info->vmdq_queue_num = vrxq;
1061 
1062 	dev_info->vmdq_pool_base = 0;
1063 	dev_info->vmdq_queue_base = 0;
1064 
1065 	return 0;
1066 }
1067 
1068 /* Configure the device based on the configuration provided */
1069 static int bnxt_dev_configure_op(struct rte_eth_dev *eth_dev)
1070 {
1071 	struct bnxt *bp = eth_dev->data->dev_private;
1072 	uint64_t rx_offloads = eth_dev->data->dev_conf.rxmode.offloads;
1073 	struct rte_eth_rss_conf *rss_conf = &eth_dev->data->dev_conf.rx_adv_conf.rss_conf;
1074 	int rc;
1075 
1076 	bp->rx_queues = (void *)eth_dev->data->rx_queues;
1077 	bp->tx_queues = (void *)eth_dev->data->tx_queues;
1078 	bp->tx_nr_rings = eth_dev->data->nb_tx_queues;
1079 	bp->rx_nr_rings = eth_dev->data->nb_rx_queues;
1080 
1081 	rc = is_bnxt_in_error(bp);
1082 	if (rc)
1083 		return rc;
1084 
1085 	if (BNXT_VF(bp) && (bp->flags & BNXT_FLAG_NEW_RM)) {
1086 		rc = bnxt_hwrm_check_vf_rings(bp);
1087 		if (rc) {
1088 			PMD_DRV_LOG(ERR, "HWRM insufficient resources\n");
1089 			return -ENOSPC;
1090 		}
1091 
1092 		/* If a resource has already been allocated - in this case
1093 		 * it is the async completion ring, free it. Reallocate it after
1094 		 * resource reservation. This will ensure the resource counts
1095 		 * are calculated correctly.
1096 		 */
1097 
1098 		pthread_mutex_lock(&bp->def_cp_lock);
1099 
1100 		if (!BNXT_HAS_NQ(bp) && bp->async_cp_ring) {
1101 			bnxt_disable_int(bp);
1102 			bnxt_free_cp_ring(bp, bp->async_cp_ring);
1103 		}
1104 
1105 		rc = bnxt_hwrm_func_reserve_vf_resc(bp, false);
1106 		if (rc) {
1107 			PMD_DRV_LOG(ERR, "HWRM resource alloc fail:%x\n", rc);
1108 			pthread_mutex_unlock(&bp->def_cp_lock);
1109 			return -ENOSPC;
1110 		}
1111 
1112 		if (!BNXT_HAS_NQ(bp) && bp->async_cp_ring) {
1113 			rc = bnxt_alloc_async_cp_ring(bp);
1114 			if (rc) {
1115 				pthread_mutex_unlock(&bp->def_cp_lock);
1116 				return rc;
1117 			}
1118 			bnxt_enable_int(bp);
1119 		}
1120 
1121 		pthread_mutex_unlock(&bp->def_cp_lock);
1122 	}
1123 
1124 	/* Inherit new configurations */
1125 	if (eth_dev->data->nb_rx_queues > bp->max_rx_rings ||
1126 	    eth_dev->data->nb_tx_queues > bp->max_tx_rings ||
1127 	    eth_dev->data->nb_rx_queues + eth_dev->data->nb_tx_queues
1128 		+ BNXT_NUM_ASYNC_CPR(bp) > bp->max_cp_rings ||
1129 	    eth_dev->data->nb_rx_queues + eth_dev->data->nb_tx_queues >
1130 	    bp->max_stat_ctx)
1131 		goto resource_error;
1132 
1133 	if (BNXT_HAS_RING_GRPS(bp) &&
1134 	    (uint32_t)(eth_dev->data->nb_rx_queues) > bp->max_ring_grps)
1135 		goto resource_error;
1136 
1137 	if (!(eth_dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS) &&
1138 	    bp->max_vnics < eth_dev->data->nb_rx_queues)
1139 		goto resource_error;
1140 
1141 	bp->rx_cp_nr_rings = bp->rx_nr_rings;
1142 	bp->tx_cp_nr_rings = bp->tx_nr_rings;
1143 
1144 	if (eth_dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG)
1145 		rx_offloads |= RTE_ETH_RX_OFFLOAD_RSS_HASH;
1146 	eth_dev->data->dev_conf.rxmode.offloads = rx_offloads;
1147 
1148 	/* application provides the hash key to program */
1149 	if (rss_conf->rss_key != NULL) {
1150 		if (rss_conf->rss_key_len != HW_HASH_KEY_SIZE)
1151 			PMD_DRV_LOG(WARNING, "port %u RSS key len must be %d bytes long",
1152 				    eth_dev->data->port_id, HW_HASH_KEY_SIZE);
1153 		else
1154 			memcpy(bp->rss_conf.rss_key, rss_conf->rss_key, HW_HASH_KEY_SIZE);
1155 	}
1156 	bp->rss_conf.rss_key_len = HW_HASH_KEY_SIZE;
1157 	bp->rss_conf.rss_hf = rss_conf->rss_hf;
1158 
1159 	bnxt_mtu_set_op(eth_dev, eth_dev->data->mtu);
1160 
1161 	return 0;
1162 
1163 resource_error:
1164 	PMD_DRV_LOG(ERR,
1165 		    "Insufficient resources to support requested config\n");
1166 	PMD_DRV_LOG(ERR,
1167 		    "Num Queues Requested: Tx %d, Rx %d\n",
1168 		    eth_dev->data->nb_tx_queues,
1169 		    eth_dev->data->nb_rx_queues);
1170 	PMD_DRV_LOG(ERR,
1171 		    "MAX: TxQ %d, RxQ %d, CQ %d Stat %d, Grp %d, Vnic %d\n",
1172 		    bp->max_tx_rings, bp->max_rx_rings, bp->max_cp_rings,
1173 		    bp->max_stat_ctx, bp->max_ring_grps, bp->max_vnics);
1174 	return -ENOSPC;
1175 }
1176 
1177 void bnxt_print_link_info(struct rte_eth_dev *eth_dev)
1178 {
1179 	struct rte_eth_link *link = &eth_dev->data->dev_link;
1180 
1181 	if (link->link_status)
1182 		PMD_DRV_LOG(DEBUG, "Port %d Link Up - speed %u Mbps - %s\n",
1183 			eth_dev->data->port_id,
1184 			(uint32_t)link->link_speed,
1185 			(link->link_duplex == RTE_ETH_LINK_FULL_DUPLEX) ?
1186 			("full-duplex") : ("half-duplex\n"));
1187 	else
1188 		PMD_DRV_LOG(INFO, "Port %d Link Down\n",
1189 			eth_dev->data->port_id);
1190 }
1191 
1192 /*
1193  * Determine whether the current configuration requires support for scattered
1194  * receive; return 1 if scattered receive is required and 0 if not.
1195  */
1196 static int bnxt_scattered_rx(struct rte_eth_dev *eth_dev)
1197 {
1198 	uint32_t overhead = BNXT_MAX_PKT_LEN - BNXT_MAX_MTU;
1199 	uint16_t buf_size;
1200 	int i;
1201 
1202 	if (eth_dev->data->dev_conf.rxmode.offloads & RTE_ETH_RX_OFFLOAD_SCATTER)
1203 		return 1;
1204 
1205 	if (eth_dev->data->dev_conf.rxmode.offloads & RTE_ETH_RX_OFFLOAD_TCP_LRO)
1206 		return 1;
1207 
1208 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
1209 		struct bnxt_rx_queue *rxq = eth_dev->data->rx_queues[i];
1210 
1211 		buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
1212 				      RTE_PKTMBUF_HEADROOM);
1213 		if (eth_dev->data->mtu + overhead > buf_size)
1214 			return 1;
1215 	}
1216 	return 0;
1217 }
1218 
1219 static eth_rx_burst_t
1220 bnxt_receive_function(struct rte_eth_dev *eth_dev)
1221 {
1222 	struct bnxt *bp = eth_dev->data->dev_private;
1223 
1224 	/* Disable vector mode RX for Stingray2 for now */
1225 	if (BNXT_CHIP_SR2(bp)) {
1226 		bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1227 		return bnxt_recv_pkts;
1228 	}
1229 
1230 #if (defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)) && \
1231 	!defined(RTE_LIBRTE_IEEE1588)
1232 
1233 	/* Vector mode receive cannot be enabled if scattered rx is in use. */
1234 	if (eth_dev->data->scattered_rx)
1235 		goto use_scalar_rx;
1236 
1237 	/*
1238 	 * Vector mode receive cannot be enabled if Truflow is enabled or if
1239 	 * asynchronous completions and receive completions can be placed in
1240 	 * the same completion ring.
1241 	 */
1242 	if (BNXT_TRUFLOW_EN(bp) || !BNXT_NUM_ASYNC_CPR(bp))
1243 		goto use_scalar_rx;
1244 
1245 	/*
1246 	 * Vector mode receive cannot be enabled if any receive offloads outside
1247 	 * a limited subset have been enabled.
1248 	 */
1249 	if (eth_dev->data->dev_conf.rxmode.offloads &
1250 		~(RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
1251 		  RTE_ETH_RX_OFFLOAD_KEEP_CRC |
1252 		  RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
1253 		  RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
1254 		  RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
1255 		  RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1256 		  RTE_ETH_RX_OFFLOAD_OUTER_UDP_CKSUM |
1257 		  RTE_ETH_RX_OFFLOAD_RSS_HASH |
1258 		  RTE_ETH_RX_OFFLOAD_VLAN_FILTER))
1259 		goto use_scalar_rx;
1260 
1261 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
1262 	if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
1263 	    rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1) {
1264 		PMD_DRV_LOG(INFO,
1265 			    "Using AVX2 vector mode receive for port %d\n",
1266 			    eth_dev->data->port_id);
1267 		bp->flags |= BNXT_FLAG_RX_VECTOR_PKT_MODE;
1268 		return bnxt_recv_pkts_vec_avx2;
1269 	}
1270  #endif
1271 	if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_128) {
1272 		PMD_DRV_LOG(INFO,
1273 			    "Using SSE vector mode receive for port %d\n",
1274 			    eth_dev->data->port_id);
1275 		bp->flags |= BNXT_FLAG_RX_VECTOR_PKT_MODE;
1276 		return bnxt_recv_pkts_vec;
1277 	}
1278 
1279 use_scalar_rx:
1280 	PMD_DRV_LOG(INFO, "Vector mode receive disabled for port %d\n",
1281 		    eth_dev->data->port_id);
1282 	PMD_DRV_LOG(INFO,
1283 		    "Port %d scatter: %d rx offload: %" PRIX64 "\n",
1284 		    eth_dev->data->port_id,
1285 		    eth_dev->data->scattered_rx,
1286 		    eth_dev->data->dev_conf.rxmode.offloads);
1287 #endif
1288 	bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1289 	return bnxt_recv_pkts;
1290 }
1291 
1292 static eth_tx_burst_t
1293 bnxt_transmit_function(struct rte_eth_dev *eth_dev)
1294 {
1295 	struct bnxt *bp = eth_dev->data->dev_private;
1296 
1297 	/* Disable vector mode TX for Stingray2 for now */
1298 	if (BNXT_CHIP_SR2(bp))
1299 		return bnxt_xmit_pkts;
1300 
1301 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64) && \
1302 	!defined(RTE_LIBRTE_IEEE1588)
1303 	uint64_t offloads = eth_dev->data->dev_conf.txmode.offloads;
1304 
1305 	/*
1306 	 * Vector mode transmit can be enabled only if not using scatter rx
1307 	 * or tx offloads.
1308 	 */
1309 	if (eth_dev->data->scattered_rx ||
1310 	    (offloads & ~RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE) ||
1311 	    BNXT_TRUFLOW_EN(bp))
1312 		goto use_scalar_tx;
1313 
1314 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
1315 	if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
1316 	    rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1) {
1317 		PMD_DRV_LOG(INFO,
1318 			    "Using AVX2 vector mode transmit for port %d\n",
1319 			    eth_dev->data->port_id);
1320 		return bnxt_xmit_pkts_vec_avx2;
1321 	}
1322 #endif
1323 	if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_128) {
1324 		PMD_DRV_LOG(INFO,
1325 			    "Using SSE vector mode transmit for port %d\n",
1326 			    eth_dev->data->port_id);
1327 		return bnxt_xmit_pkts_vec;
1328 	}
1329 
1330 use_scalar_tx:
1331 	PMD_DRV_LOG(INFO, "Vector mode transmit disabled for port %d\n",
1332 		    eth_dev->data->port_id);
1333 	PMD_DRV_LOG(INFO,
1334 		    "Port %d scatter: %d tx offload: %" PRIX64 "\n",
1335 		    eth_dev->data->port_id,
1336 		    eth_dev->data->scattered_rx,
1337 		    offloads);
1338 #endif
1339 	return bnxt_xmit_pkts;
1340 }
1341 
1342 static int bnxt_handle_if_change_status(struct bnxt *bp)
1343 {
1344 	int rc;
1345 
1346 	/* Since fw has undergone a reset and lost all contexts,
1347 	 * set fatal flag to not issue hwrm during cleanup
1348 	 */
1349 	bp->flags |= BNXT_FLAG_FATAL_ERROR;
1350 	bnxt_uninit_resources(bp, true);
1351 
1352 	/* clear fatal flag so that re-init happens */
1353 	bp->flags &= ~BNXT_FLAG_FATAL_ERROR;
1354 
1355 	rc = bnxt_check_fw_ready(bp);
1356 	if (rc)
1357 		return rc;
1358 
1359 	rc = bnxt_init_resources(bp, true);
1360 
1361 	bp->flags &= ~BNXT_FLAG_IF_CHANGE_HOT_FW_RESET_DONE;
1362 
1363 	return rc;
1364 }
1365 
1366 static int bnxt_dev_set_link_up_op(struct rte_eth_dev *eth_dev)
1367 {
1368 	struct bnxt *bp = eth_dev->data->dev_private;
1369 	int rc = 0;
1370 
1371 	if (!BNXT_SINGLE_PF(bp))
1372 		return -ENOTSUP;
1373 
1374 	if (!bp->link_info->link_up)
1375 		rc = bnxt_set_hwrm_link_config(bp, true);
1376 	if (!rc)
1377 		eth_dev->data->dev_link.link_status = 1;
1378 
1379 	bnxt_print_link_info(eth_dev);
1380 	return rc;
1381 }
1382 
1383 static int bnxt_dev_set_link_down_op(struct rte_eth_dev *eth_dev)
1384 {
1385 	struct bnxt *bp = eth_dev->data->dev_private;
1386 
1387 	if (!BNXT_SINGLE_PF(bp))
1388 		return -ENOTSUP;
1389 
1390 	eth_dev->data->dev_link.link_status = 0;
1391 	bnxt_set_hwrm_link_config(bp, false);
1392 	bp->link_info->link_up = 0;
1393 
1394 	return 0;
1395 }
1396 
1397 static void bnxt_free_switch_domain(struct bnxt *bp)
1398 {
1399 	int rc = 0;
1400 
1401 	if (!(BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)))
1402 		return;
1403 
1404 	rc = rte_eth_switch_domain_free(bp->switch_domain_id);
1405 	if (rc)
1406 		PMD_DRV_LOG(ERR, "free switch domain:%d fail: %d\n",
1407 			    bp->switch_domain_id, rc);
1408 }
1409 
1410 static void bnxt_ptp_get_current_time(void *arg)
1411 {
1412 	struct bnxt *bp = arg;
1413 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
1414 	int rc;
1415 
1416 	rc = is_bnxt_in_error(bp);
1417 	if (rc)
1418 		return;
1419 
1420 	if (!ptp)
1421 		return;
1422 
1423 	bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
1424 				&ptp->current_time);
1425 
1426 	rc = rte_eal_alarm_set(US_PER_S, bnxt_ptp_get_current_time, (void *)bp);
1427 	if (rc != 0) {
1428 		PMD_DRV_LOG(ERR, "Failed to re-schedule PTP alarm\n");
1429 		bp->flags2 &= ~BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1430 	}
1431 }
1432 
1433 static int bnxt_schedule_ptp_alarm(struct bnxt *bp)
1434 {
1435 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
1436 	int rc;
1437 
1438 	if (bp->flags2 & BNXT_FLAGS2_PTP_ALARM_SCHEDULED)
1439 		return 0;
1440 
1441 	bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
1442 				&ptp->current_time);
1443 
1444 	rc = rte_eal_alarm_set(US_PER_S, bnxt_ptp_get_current_time, (void *)bp);
1445 	return rc;
1446 }
1447 
1448 static void bnxt_cancel_ptp_alarm(struct bnxt *bp)
1449 {
1450 	if (bp->flags2 & BNXT_FLAGS2_PTP_ALARM_SCHEDULED) {
1451 		rte_eal_alarm_cancel(bnxt_ptp_get_current_time, (void *)bp);
1452 		bp->flags2 &= ~BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1453 	}
1454 }
1455 
1456 static void bnxt_ptp_stop(struct bnxt *bp)
1457 {
1458 	bnxt_cancel_ptp_alarm(bp);
1459 	bp->flags2 &= ~BNXT_FLAGS2_PTP_TIMESYNC_ENABLED;
1460 }
1461 
1462 static int bnxt_ptp_start(struct bnxt *bp)
1463 {
1464 	int rc;
1465 
1466 	rc = bnxt_schedule_ptp_alarm(bp);
1467 	if (rc != 0) {
1468 		PMD_DRV_LOG(ERR, "Failed to schedule PTP alarm\n");
1469 	} else {
1470 		bp->flags2 |= BNXT_FLAGS2_PTP_TIMESYNC_ENABLED;
1471 		bp->flags2 |= BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1472 	}
1473 
1474 	return rc;
1475 }
1476 
1477 static int bnxt_dev_stop(struct rte_eth_dev *eth_dev)
1478 {
1479 	struct bnxt *bp = eth_dev->data->dev_private;
1480 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1481 	struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
1482 	struct rte_eth_link link;
1483 	int ret;
1484 
1485 	eth_dev->data->dev_started = 0;
1486 
1487 	/* Prevent crashes when queues are still in use */
1488 	bnxt_stop_rxtx(eth_dev);
1489 
1490 	bnxt_disable_int(bp);
1491 
1492 	/* disable uio/vfio intr/eventfd mapping */
1493 	rte_intr_disable(intr_handle);
1494 
1495 	/* Stop the child representors for this device */
1496 	ret = bnxt_rep_stop_all(bp);
1497 	if (ret != 0)
1498 		return ret;
1499 
1500 	/* delete the bnxt ULP port details */
1501 	bnxt_ulp_port_deinit(bp);
1502 
1503 	bnxt_cancel_fw_health_check(bp);
1504 
1505 	if (BNXT_P5_PTP_TIMESYNC_ENABLED(bp))
1506 		bnxt_cancel_ptp_alarm(bp);
1507 
1508 	/* Do not bring link down during reset recovery */
1509 	if (!is_bnxt_in_error(bp)) {
1510 		bnxt_dev_set_link_down_op(eth_dev);
1511 		/* Wait for link to be reset */
1512 		if (BNXT_SINGLE_PF(bp))
1513 			rte_delay_ms(500);
1514 		/* clear the recorded link status */
1515 		memset(&link, 0, sizeof(link));
1516 		rte_eth_linkstatus_set(eth_dev, &link);
1517 	}
1518 
1519 	/* Clean queue intr-vector mapping */
1520 	rte_intr_efd_disable(intr_handle);
1521 	rte_intr_vec_list_free(intr_handle);
1522 
1523 	bnxt_hwrm_port_clr_stats(bp);
1524 	bnxt_free_tx_mbufs(bp);
1525 	bnxt_free_rx_mbufs(bp);
1526 	/* Process any remaining notifications in default completion queue */
1527 	bnxt_int_handler(eth_dev);
1528 	bnxt_shutdown_nic(bp);
1529 	bnxt_hwrm_if_change(bp, false);
1530 
1531 	bnxt_free_prev_ring_stats(bp);
1532 	rte_free(bp->mark_table);
1533 	bp->mark_table = NULL;
1534 
1535 	bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1536 	bp->rx_cosq_cnt = 0;
1537 	/* All filters are deleted on a port stop. */
1538 	if (BNXT_FLOW_XSTATS_EN(bp))
1539 		bp->flow_stat->flow_count = 0;
1540 
1541 	eth_dev->data->scattered_rx = 0;
1542 
1543 	return 0;
1544 }
1545 
1546 /* Unload the driver, release resources */
1547 int bnxt_dev_stop_op(struct rte_eth_dev *eth_dev)
1548 {
1549 	struct bnxt *bp = eth_dev->data->dev_private;
1550 
1551 	pthread_mutex_lock(&bp->err_recovery_lock);
1552 	if (bp->flags & BNXT_FLAG_FW_RESET) {
1553 		PMD_DRV_LOG(ERR,
1554 			    "Adapter recovering from error..Please retry\n");
1555 		pthread_mutex_unlock(&bp->err_recovery_lock);
1556 		return -EAGAIN;
1557 	}
1558 	pthread_mutex_unlock(&bp->err_recovery_lock);
1559 
1560 	return bnxt_dev_stop(eth_dev);
1561 }
1562 
1563 int bnxt_dev_start_op(struct rte_eth_dev *eth_dev)
1564 {
1565 	struct bnxt *bp = eth_dev->data->dev_private;
1566 	uint64_t rx_offloads = eth_dev->data->dev_conf.rxmode.offloads;
1567 	int vlan_mask = 0;
1568 	int rc, retry_cnt = BNXT_IF_CHANGE_RETRY_COUNT;
1569 
1570 	if (!eth_dev->data->nb_tx_queues || !eth_dev->data->nb_rx_queues) {
1571 		PMD_DRV_LOG(ERR, "Queues are not configured yet!\n");
1572 		return -EINVAL;
1573 	}
1574 
1575 	if (bp->rx_cp_nr_rings > RTE_ETHDEV_QUEUE_STAT_CNTRS)
1576 		PMD_DRV_LOG(ERR,
1577 			    "RxQ cnt %d > RTE_ETHDEV_QUEUE_STAT_CNTRS %d\n",
1578 			    bp->rx_cp_nr_rings, RTE_ETHDEV_QUEUE_STAT_CNTRS);
1579 
1580 	do {
1581 		rc = bnxt_hwrm_if_change(bp, true);
1582 		if (rc == 0 || rc != -EAGAIN)
1583 			break;
1584 
1585 		rte_delay_ms(BNXT_IF_CHANGE_RETRY_INTERVAL);
1586 	} while (retry_cnt--);
1587 
1588 	if (rc)
1589 		return rc;
1590 
1591 	if (bp->flags & BNXT_FLAG_IF_CHANGE_HOT_FW_RESET_DONE) {
1592 		rc = bnxt_handle_if_change_status(bp);
1593 		if (rc)
1594 			return rc;
1595 	}
1596 
1597 	bnxt_enable_int(bp);
1598 
1599 	eth_dev->data->scattered_rx = bnxt_scattered_rx(eth_dev);
1600 
1601 	rc = bnxt_start_nic(bp);
1602 	if (rc)
1603 		goto error;
1604 
1605 	rc = bnxt_alloc_prev_ring_stats(bp);
1606 	if (rc)
1607 		goto error;
1608 
1609 	eth_dev->data->dev_started = 1;
1610 
1611 	bnxt_link_update_op(eth_dev, 0);
1612 
1613 	if (rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER)
1614 		vlan_mask |= RTE_ETH_VLAN_FILTER_MASK;
1615 	if (rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP)
1616 		vlan_mask |= RTE_ETH_VLAN_STRIP_MASK;
1617 	rc = bnxt_vlan_offload_set_op(eth_dev, vlan_mask);
1618 	if (rc)
1619 		goto error;
1620 
1621 	/* Initialize bnxt ULP port details */
1622 	rc = bnxt_ulp_port_init(bp);
1623 	if (rc)
1624 		goto error;
1625 
1626 	eth_dev->rx_pkt_burst = bnxt_receive_function(eth_dev);
1627 	eth_dev->tx_pkt_burst = bnxt_transmit_function(eth_dev);
1628 
1629 	bnxt_schedule_fw_health_check(bp);
1630 
1631 	if (BNXT_P5_PTP_TIMESYNC_ENABLED(bp))
1632 		bnxt_schedule_ptp_alarm(bp);
1633 
1634 	return 0;
1635 
1636 error:
1637 	bnxt_dev_stop(eth_dev);
1638 	return rc;
1639 }
1640 
1641 static void
1642 bnxt_uninit_locks(struct bnxt *bp)
1643 {
1644 	pthread_mutex_destroy(&bp->flow_lock);
1645 	pthread_mutex_destroy(&bp->def_cp_lock);
1646 	pthread_mutex_destroy(&bp->health_check_lock);
1647 	pthread_mutex_destroy(&bp->err_recovery_lock);
1648 	if (bp->rep_info) {
1649 		pthread_mutex_destroy(&bp->rep_info->vfr_lock);
1650 		pthread_mutex_destroy(&bp->rep_info->vfr_start_lock);
1651 	}
1652 }
1653 
1654 static void bnxt_drv_uninit(struct bnxt *bp)
1655 {
1656 	bnxt_free_leds_info(bp);
1657 	bnxt_free_cos_queues(bp);
1658 	bnxt_free_link_info(bp);
1659 	bnxt_free_parent_info(bp);
1660 	bnxt_uninit_locks(bp);
1661 
1662 	rte_memzone_free((const struct rte_memzone *)bp->tx_mem_zone);
1663 	bp->tx_mem_zone = NULL;
1664 	rte_memzone_free((const struct rte_memzone *)bp->rx_mem_zone);
1665 	bp->rx_mem_zone = NULL;
1666 
1667 	bnxt_free_vf_info(bp);
1668 	bnxt_free_pf_info(bp);
1669 
1670 	rte_free(bp->grp_info);
1671 	bp->grp_info = NULL;
1672 }
1673 
1674 static int bnxt_dev_close_op(struct rte_eth_dev *eth_dev)
1675 {
1676 	struct bnxt *bp = eth_dev->data->dev_private;
1677 	int ret = 0;
1678 
1679 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1680 		return 0;
1681 
1682 	pthread_mutex_lock(&bp->err_recovery_lock);
1683 	if (bp->flags & BNXT_FLAG_FW_RESET) {
1684 		PMD_DRV_LOG(ERR,
1685 			    "Adapter recovering from error...Please retry\n");
1686 		pthread_mutex_unlock(&bp->err_recovery_lock);
1687 		return -EAGAIN;
1688 	}
1689 	pthread_mutex_unlock(&bp->err_recovery_lock);
1690 
1691 	/* cancel the recovery handler before remove dev */
1692 	rte_eal_alarm_cancel(bnxt_dev_reset_and_resume, (void *)bp);
1693 	rte_eal_alarm_cancel(bnxt_dev_recover, (void *)bp);
1694 	bnxt_cancel_fc_thread(bp);
1695 	rte_eal_alarm_cancel(bnxt_handle_vf_cfg_change, (void *)bp);
1696 
1697 	if (eth_dev->data->dev_started)
1698 		ret = bnxt_dev_stop(eth_dev);
1699 
1700 	bnxt_uninit_resources(bp, false);
1701 
1702 	bnxt_drv_uninit(bp);
1703 
1704 	return ret;
1705 }
1706 
1707 static void bnxt_mac_addr_remove_op(struct rte_eth_dev *eth_dev,
1708 				    uint32_t index)
1709 {
1710 	struct bnxt *bp = eth_dev->data->dev_private;
1711 	uint64_t pool_mask = eth_dev->data->mac_pool_sel[index];
1712 	struct bnxt_vnic_info *vnic;
1713 	struct bnxt_filter_info *filter, *temp_filter;
1714 	uint32_t i;
1715 
1716 	if (is_bnxt_in_error(bp))
1717 		return;
1718 
1719 	/*
1720 	 * Loop through all VNICs from the specified filter flow pools to
1721 	 * remove the corresponding MAC addr filter
1722 	 */
1723 	for (i = 0; i < bp->nr_vnics; i++) {
1724 		if (!(pool_mask & (1ULL << i)))
1725 			continue;
1726 
1727 		vnic = &bp->vnic_info[i];
1728 		filter = STAILQ_FIRST(&vnic->filter);
1729 		while (filter) {
1730 			temp_filter = STAILQ_NEXT(filter, next);
1731 			if (filter->mac_index == index) {
1732 				STAILQ_REMOVE(&vnic->filter, filter,
1733 						bnxt_filter_info, next);
1734 				bnxt_hwrm_clear_l2_filter(bp, filter);
1735 				bnxt_free_filter(bp, filter);
1736 			}
1737 			filter = temp_filter;
1738 		}
1739 	}
1740 }
1741 
1742 static int bnxt_add_mac_filter(struct bnxt *bp, struct bnxt_vnic_info *vnic,
1743 			       struct rte_ether_addr *mac_addr, uint32_t index,
1744 			       uint32_t pool)
1745 {
1746 	struct bnxt_filter_info *filter;
1747 	int rc = 0;
1748 
1749 	/* Attach requested MAC address to the new l2_filter */
1750 	STAILQ_FOREACH(filter, &vnic->filter, next) {
1751 		if (filter->mac_index == index) {
1752 			PMD_DRV_LOG(DEBUG,
1753 				    "MAC addr already existed for pool %d\n",
1754 				    pool);
1755 			return 0;
1756 		}
1757 	}
1758 
1759 	filter = bnxt_alloc_filter(bp);
1760 	if (!filter) {
1761 		PMD_DRV_LOG(ERR, "L2 filter alloc failed\n");
1762 		return -ENODEV;
1763 	}
1764 
1765 	/* bnxt_alloc_filter copies default MAC to filter->l2_addr. So,
1766 	 * if the MAC that's been programmed now is a different one, then,
1767 	 * copy that addr to filter->l2_addr
1768 	 */
1769 	if (mac_addr)
1770 		memcpy(filter->l2_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1771 	filter->flags |= HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
1772 
1773 	rc = bnxt_hwrm_set_l2_filter(bp, vnic->fw_vnic_id, filter);
1774 	if (!rc) {
1775 		filter->mac_index = index;
1776 		if (filter->mac_index == 0)
1777 			STAILQ_INSERT_HEAD(&vnic->filter, filter, next);
1778 		else
1779 			STAILQ_INSERT_TAIL(&vnic->filter, filter, next);
1780 	} else {
1781 		bnxt_free_filter(bp, filter);
1782 	}
1783 
1784 	return rc;
1785 }
1786 
1787 static int bnxt_mac_addr_add_op(struct rte_eth_dev *eth_dev,
1788 				struct rte_ether_addr *mac_addr,
1789 				uint32_t index, uint32_t pool)
1790 {
1791 	struct bnxt *bp = eth_dev->data->dev_private;
1792 	struct bnxt_vnic_info *vnic = &bp->vnic_info[pool];
1793 	int rc = 0;
1794 
1795 	rc = is_bnxt_in_error(bp);
1796 	if (rc)
1797 		return rc;
1798 
1799 	if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp)) {
1800 		PMD_DRV_LOG(ERR, "Cannot add MAC address to a VF interface\n");
1801 		return -ENOTSUP;
1802 	}
1803 
1804 	if (!vnic) {
1805 		PMD_DRV_LOG(ERR, "VNIC not found for pool %d!\n", pool);
1806 		return -EINVAL;
1807 	}
1808 
1809 	/* Filter settings will get applied when port is started */
1810 	if (!eth_dev->data->dev_started)
1811 		return 0;
1812 
1813 	rc = bnxt_add_mac_filter(bp, vnic, mac_addr, index, pool);
1814 
1815 	return rc;
1816 }
1817 
1818 int bnxt_link_update_op(struct rte_eth_dev *eth_dev, int wait_to_complete)
1819 {
1820 	int rc = 0;
1821 	struct bnxt *bp = eth_dev->data->dev_private;
1822 	struct rte_eth_link new;
1823 	int cnt = wait_to_complete ? BNXT_MAX_LINK_WAIT_CNT :
1824 			BNXT_MIN_LINK_WAIT_CNT;
1825 
1826 	rc = is_bnxt_in_error(bp);
1827 	if (rc)
1828 		return rc;
1829 
1830 	memset(&new, 0, sizeof(new));
1831 
1832 	if (bp->link_info == NULL)
1833 		goto out;
1834 
1835 	/* Only single function PF can bring the phy down.
1836 	 * In certain scenarios, device is not obliged link down even when forced.
1837 	 * When port is stopped, report link down in those cases.
1838 	 */
1839 	if (!eth_dev->data->dev_started &&
1840 	    (!BNXT_SINGLE_PF(bp) || bnxt_force_link_config(bp)))
1841 		goto out;
1842 
1843 	do {
1844 		/* Retrieve link info from hardware */
1845 		rc = bnxt_get_hwrm_link_config(bp, &new);
1846 		if (rc) {
1847 			new.link_speed = RTE_ETH_LINK_SPEED_100M;
1848 			new.link_duplex = RTE_ETH_LINK_FULL_DUPLEX;
1849 			PMD_DRV_LOG(ERR,
1850 				"Failed to retrieve link rc = 0x%x!\n", rc);
1851 			goto out;
1852 		}
1853 
1854 		if (!wait_to_complete || new.link_status)
1855 			break;
1856 
1857 		rte_delay_ms(BNXT_LINK_WAIT_INTERVAL);
1858 	} while (cnt--);
1859 
1860 out:
1861 	/* Timed out or success */
1862 	if (new.link_status != eth_dev->data->dev_link.link_status ||
1863 	    new.link_speed != eth_dev->data->dev_link.link_speed) {
1864 		rte_eth_linkstatus_set(eth_dev, &new);
1865 		bnxt_print_link_info(eth_dev);
1866 	}
1867 
1868 	return rc;
1869 }
1870 
1871 static int bnxt_promiscuous_enable_op(struct rte_eth_dev *eth_dev)
1872 {
1873 	struct bnxt *bp = eth_dev->data->dev_private;
1874 	struct bnxt_vnic_info *vnic;
1875 	uint32_t old_flags;
1876 	int rc;
1877 
1878 	rc = is_bnxt_in_error(bp);
1879 	if (rc)
1880 		return rc;
1881 
1882 	/* Filter settings will get applied when port is started */
1883 	if (!eth_dev->data->dev_started)
1884 		return 0;
1885 
1886 	if (bp->vnic_info == NULL)
1887 		return 0;
1888 
1889 	vnic = BNXT_GET_DEFAULT_VNIC(bp);
1890 
1891 	old_flags = vnic->flags;
1892 	vnic->flags |= BNXT_VNIC_INFO_PROMISC;
1893 	rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1894 	if (rc != 0)
1895 		vnic->flags = old_flags;
1896 
1897 	return rc;
1898 }
1899 
1900 static int bnxt_promiscuous_disable_op(struct rte_eth_dev *eth_dev)
1901 {
1902 	struct bnxt *bp = eth_dev->data->dev_private;
1903 	struct bnxt_vnic_info *vnic;
1904 	uint32_t old_flags;
1905 	int rc;
1906 
1907 	rc = is_bnxt_in_error(bp);
1908 	if (rc)
1909 		return rc;
1910 
1911 	/* Filter settings will get applied when port is started */
1912 	if (!eth_dev->data->dev_started)
1913 		return 0;
1914 
1915 	if (bp->vnic_info == NULL)
1916 		return 0;
1917 
1918 	vnic = BNXT_GET_DEFAULT_VNIC(bp);
1919 
1920 	old_flags = vnic->flags;
1921 	vnic->flags &= ~BNXT_VNIC_INFO_PROMISC;
1922 	rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1923 	if (rc != 0)
1924 		vnic->flags = old_flags;
1925 
1926 	return rc;
1927 }
1928 
1929 static int bnxt_allmulticast_enable_op(struct rte_eth_dev *eth_dev)
1930 {
1931 	struct bnxt *bp = eth_dev->data->dev_private;
1932 	struct bnxt_vnic_info *vnic;
1933 	uint32_t old_flags;
1934 	int rc;
1935 
1936 	rc = is_bnxt_in_error(bp);
1937 	if (rc)
1938 		return rc;
1939 
1940 	/* Filter settings will get applied when port is started */
1941 	if (!eth_dev->data->dev_started)
1942 		return 0;
1943 
1944 	if (bp->vnic_info == NULL)
1945 		return 0;
1946 
1947 	vnic = BNXT_GET_DEFAULT_VNIC(bp);
1948 
1949 	old_flags = vnic->flags;
1950 	vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
1951 	rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1952 	if (rc != 0)
1953 		vnic->flags = old_flags;
1954 
1955 	return rc;
1956 }
1957 
1958 static int bnxt_allmulticast_disable_op(struct rte_eth_dev *eth_dev)
1959 {
1960 	struct bnxt *bp = eth_dev->data->dev_private;
1961 	struct bnxt_vnic_info *vnic;
1962 	uint32_t old_flags;
1963 	int rc;
1964 
1965 	rc = is_bnxt_in_error(bp);
1966 	if (rc)
1967 		return rc;
1968 
1969 	/* Filter settings will get applied when port is started */
1970 	if (!eth_dev->data->dev_started)
1971 		return 0;
1972 
1973 	if (bp->vnic_info == NULL)
1974 		return 0;
1975 
1976 	vnic = BNXT_GET_DEFAULT_VNIC(bp);
1977 
1978 	old_flags = vnic->flags;
1979 	vnic->flags &= ~BNXT_VNIC_INFO_ALLMULTI;
1980 	rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1981 	if (rc != 0)
1982 		vnic->flags = old_flags;
1983 
1984 	return rc;
1985 }
1986 
1987 /* Return bnxt_rx_queue pointer corresponding to a given rxq. */
1988 static struct bnxt_rx_queue *bnxt_qid_to_rxq(struct bnxt *bp, uint16_t qid)
1989 {
1990 	if (qid >= bp->rx_nr_rings)
1991 		return NULL;
1992 
1993 	return bp->eth_dev->data->rx_queues[qid];
1994 }
1995 
1996 /* Return rxq corresponding to a given rss table ring/group ID. */
1997 static uint16_t bnxt_rss_to_qid(struct bnxt *bp, uint16_t fwr)
1998 {
1999 	struct bnxt_rx_queue *rxq;
2000 	unsigned int i;
2001 
2002 	if (!BNXT_HAS_RING_GRPS(bp)) {
2003 		for (i = 0; i < bp->rx_nr_rings; i++) {
2004 			rxq = bp->eth_dev->data->rx_queues[i];
2005 			if (rxq->rx_ring->rx_ring_struct->fw_ring_id == fwr)
2006 				return rxq->index;
2007 		}
2008 	} else {
2009 		for (i = 0; i < bp->rx_nr_rings; i++) {
2010 			if (bp->grp_info[i].fw_grp_id == fwr)
2011 				return i;
2012 		}
2013 	}
2014 
2015 	return INVALID_HW_RING_ID;
2016 }
2017 
2018 static int bnxt_reta_update_op(struct rte_eth_dev *eth_dev,
2019 			    struct rte_eth_rss_reta_entry64 *reta_conf,
2020 			    uint16_t reta_size)
2021 {
2022 	struct bnxt *bp = eth_dev->data->dev_private;
2023 	struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
2024 	struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2025 	uint16_t tbl_size = bnxt_rss_hash_tbl_size(bp);
2026 	uint16_t idx, sft;
2027 	int i, rc;
2028 
2029 	rc = is_bnxt_in_error(bp);
2030 	if (rc)
2031 		return rc;
2032 
2033 	if (!vnic->rss_table)
2034 		return -EINVAL;
2035 
2036 	if (!(dev_conf->rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
2037 		return -EINVAL;
2038 
2039 	if (reta_size != tbl_size) {
2040 		PMD_DRV_LOG(ERR, "The configured hash table lookup size "
2041 			"(%d) must equal the size supported by the hardware "
2042 			"(%d)\n", reta_size, tbl_size);
2043 		return -EINVAL;
2044 	}
2045 
2046 	for (i = 0; i < reta_size; i++) {
2047 		struct bnxt_rx_queue *rxq;
2048 
2049 		idx = i / RTE_ETH_RETA_GROUP_SIZE;
2050 		sft = i % RTE_ETH_RETA_GROUP_SIZE;
2051 
2052 		if (!(reta_conf[idx].mask & (1ULL << sft)))
2053 			continue;
2054 
2055 		rxq = bnxt_qid_to_rxq(bp, reta_conf[idx].reta[sft]);
2056 		if (!rxq) {
2057 			PMD_DRV_LOG(ERR, "Invalid ring in reta_conf.\n");
2058 			return -EINVAL;
2059 		}
2060 
2061 		if (BNXT_CHIP_P5(bp)) {
2062 			vnic->rss_table[i * 2] =
2063 				rxq->rx_ring->rx_ring_struct->fw_ring_id;
2064 			vnic->rss_table[i * 2 + 1] =
2065 				rxq->cp_ring->cp_ring_struct->fw_ring_id;
2066 		} else {
2067 			vnic->rss_table[i] =
2068 			    vnic->fw_grp_ids[reta_conf[idx].reta[sft]];
2069 		}
2070 	}
2071 
2072 	rc = bnxt_hwrm_vnic_rss_cfg(bp, vnic);
2073 	return rc;
2074 }
2075 
2076 static int bnxt_reta_query_op(struct rte_eth_dev *eth_dev,
2077 			      struct rte_eth_rss_reta_entry64 *reta_conf,
2078 			      uint16_t reta_size)
2079 {
2080 	struct bnxt *bp = eth_dev->data->dev_private;
2081 	struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2082 	uint16_t tbl_size = bnxt_rss_hash_tbl_size(bp);
2083 	uint16_t idx, sft, i;
2084 	int rc;
2085 
2086 	rc = is_bnxt_in_error(bp);
2087 	if (rc)
2088 		return rc;
2089 
2090 	if (!vnic)
2091 		return -EINVAL;
2092 	if (!vnic->rss_table)
2093 		return -EINVAL;
2094 
2095 	if (reta_size != tbl_size) {
2096 		PMD_DRV_LOG(ERR, "The configured hash table lookup size "
2097 			"(%d) must equal the size supported by the hardware "
2098 			"(%d)\n", reta_size, tbl_size);
2099 		return -EINVAL;
2100 	}
2101 
2102 	for (idx = 0, i = 0; i < reta_size; i++) {
2103 		idx = i / RTE_ETH_RETA_GROUP_SIZE;
2104 		sft = i % RTE_ETH_RETA_GROUP_SIZE;
2105 
2106 		if (reta_conf[idx].mask & (1ULL << sft)) {
2107 			uint16_t qid;
2108 
2109 			if (BNXT_CHIP_P5(bp))
2110 				qid = bnxt_rss_to_qid(bp,
2111 						      vnic->rss_table[i * 2]);
2112 			else
2113 				qid = bnxt_rss_to_qid(bp, vnic->rss_table[i]);
2114 
2115 			if (qid == INVALID_HW_RING_ID) {
2116 				PMD_DRV_LOG(ERR, "Inv. entry in rss table.\n");
2117 				return -EINVAL;
2118 			}
2119 			reta_conf[idx].reta[sft] = qid;
2120 		}
2121 	}
2122 
2123 	return 0;
2124 }
2125 
2126 static int bnxt_rss_hash_update_op(struct rte_eth_dev *eth_dev,
2127 				   struct rte_eth_rss_conf *rss_conf)
2128 {
2129 	struct bnxt *bp = eth_dev->data->dev_private;
2130 	struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
2131 	struct bnxt_vnic_info *vnic;
2132 	int rc;
2133 
2134 	rc = is_bnxt_in_error(bp);
2135 	if (rc)
2136 		return rc;
2137 
2138 	/*
2139 	 * If RSS enablement were different than dev_configure,
2140 	 * then return -EINVAL
2141 	 */
2142 	if (dev_conf->rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG) {
2143 		if (!rss_conf->rss_hf)
2144 			PMD_DRV_LOG(ERR, "Hash type NONE\n");
2145 	} else {
2146 		if (rss_conf->rss_hf & BNXT_ETH_RSS_SUPPORT)
2147 			return -EINVAL;
2148 	}
2149 
2150 	/* Update the default RSS VNIC(s) */
2151 	vnic = BNXT_GET_DEFAULT_VNIC(bp);
2152 	vnic->hash_type = bnxt_rte_to_hwrm_hash_types(rss_conf->rss_hf);
2153 	vnic->hash_mode =
2154 		bnxt_rte_to_hwrm_hash_level(bp, rss_conf->rss_hf,
2155 					    RTE_ETH_RSS_LEVEL(rss_conf->rss_hf));
2156 
2157 	/* Cache the hash function */
2158 	bp->rss_conf.rss_hf = rss_conf->rss_hf;
2159 
2160 	/*
2161 	 * If hashkey is not specified, use the previously configured
2162 	 * hashkey
2163 	 */
2164 	if (!rss_conf->rss_key)
2165 		goto rss_config;
2166 
2167 	if (rss_conf->rss_key_len != HW_HASH_KEY_SIZE) {
2168 		PMD_DRV_LOG(ERR,
2169 			    "Invalid hashkey length, should be %d bytes\n",
2170 			    HW_HASH_KEY_SIZE);
2171 		return -EINVAL;
2172 	}
2173 	memcpy(vnic->rss_hash_key, rss_conf->rss_key, rss_conf->rss_key_len);
2174 
2175 	/* Cache the hash key */
2176 	memcpy(bp->rss_conf.rss_key, rss_conf->rss_key, HW_HASH_KEY_SIZE);
2177 
2178 rss_config:
2179 	rc = bnxt_hwrm_vnic_rss_cfg(bp, vnic);
2180 	return rc;
2181 }
2182 
2183 static int bnxt_rss_hash_conf_get_op(struct rte_eth_dev *eth_dev,
2184 				     struct rte_eth_rss_conf *rss_conf)
2185 {
2186 	struct bnxt *bp = eth_dev->data->dev_private;
2187 	struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2188 	int len, rc;
2189 	uint32_t hash_types;
2190 
2191 	rc = is_bnxt_in_error(bp);
2192 	if (rc)
2193 		return rc;
2194 
2195 	/* RSS configuration is the same for all VNICs */
2196 	if (vnic && vnic->rss_hash_key) {
2197 		if (rss_conf->rss_key) {
2198 			len = rss_conf->rss_key_len <= HW_HASH_KEY_SIZE ?
2199 			      rss_conf->rss_key_len : HW_HASH_KEY_SIZE;
2200 			memcpy(rss_conf->rss_key, vnic->rss_hash_key, len);
2201 		}
2202 
2203 		hash_types = vnic->hash_type;
2204 		rss_conf->rss_hf = 0;
2205 		if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV4) {
2206 			rss_conf->rss_hf |= RTE_ETH_RSS_IPV4;
2207 			hash_types &= ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV4;
2208 		}
2209 		if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV4) {
2210 			rss_conf->rss_hf |= RTE_ETH_RSS_NONFRAG_IPV4_TCP;
2211 			hash_types &=
2212 				~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV4;
2213 		}
2214 		if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV4) {
2215 			rss_conf->rss_hf |= RTE_ETH_RSS_NONFRAG_IPV4_UDP;
2216 			hash_types &=
2217 				~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV4;
2218 		}
2219 		if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV6) {
2220 			rss_conf->rss_hf |= RTE_ETH_RSS_IPV6;
2221 			hash_types &= ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV6;
2222 		}
2223 		if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV6) {
2224 			rss_conf->rss_hf |= RTE_ETH_RSS_NONFRAG_IPV6_TCP;
2225 			hash_types &=
2226 				~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV6;
2227 		}
2228 		if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV6) {
2229 			rss_conf->rss_hf |= RTE_ETH_RSS_NONFRAG_IPV6_UDP;
2230 			hash_types &=
2231 				~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV6;
2232 		}
2233 
2234 		rss_conf->rss_hf |=
2235 			bnxt_hwrm_to_rte_rss_level(bp, vnic->hash_mode);
2236 
2237 		if (hash_types) {
2238 			PMD_DRV_LOG(ERR,
2239 				"Unknown RSS config from firmware (%08x), RSS disabled",
2240 				vnic->hash_type);
2241 			return -ENOTSUP;
2242 		}
2243 	} else {
2244 		rss_conf->rss_hf = 0;
2245 	}
2246 	return 0;
2247 }
2248 
2249 static int bnxt_flow_ctrl_get_op(struct rte_eth_dev *dev,
2250 			       struct rte_eth_fc_conf *fc_conf)
2251 {
2252 	struct bnxt *bp = dev->data->dev_private;
2253 	struct rte_eth_link link_info;
2254 	int rc;
2255 
2256 	rc = is_bnxt_in_error(bp);
2257 	if (rc)
2258 		return rc;
2259 
2260 	rc = bnxt_get_hwrm_link_config(bp, &link_info);
2261 	if (rc)
2262 		return rc;
2263 
2264 	memset(fc_conf, 0, sizeof(*fc_conf));
2265 	if (bp->link_info->auto_pause)
2266 		fc_conf->autoneg = 1;
2267 	switch (bp->link_info->pause) {
2268 	case 0:
2269 		fc_conf->mode = RTE_ETH_FC_NONE;
2270 		break;
2271 	case HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_TX:
2272 		fc_conf->mode = RTE_ETH_FC_TX_PAUSE;
2273 		break;
2274 	case HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_RX:
2275 		fc_conf->mode = RTE_ETH_FC_RX_PAUSE;
2276 		break;
2277 	case (HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_TX |
2278 			HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_RX):
2279 		fc_conf->mode = RTE_ETH_FC_FULL;
2280 		break;
2281 	}
2282 	return 0;
2283 }
2284 
2285 static int bnxt_flow_ctrl_set_op(struct rte_eth_dev *dev,
2286 			       struct rte_eth_fc_conf *fc_conf)
2287 {
2288 	struct bnxt *bp = dev->data->dev_private;
2289 	int rc;
2290 
2291 	rc = is_bnxt_in_error(bp);
2292 	if (rc)
2293 		return rc;
2294 
2295 	if (!BNXT_SINGLE_PF(bp)) {
2296 		PMD_DRV_LOG(ERR,
2297 			    "Flow Control Settings cannot be modified on VF or on shared PF\n");
2298 		return -ENOTSUP;
2299 	}
2300 
2301 	switch (fc_conf->mode) {
2302 	case RTE_ETH_FC_NONE:
2303 		bp->link_info->auto_pause = 0;
2304 		bp->link_info->force_pause = 0;
2305 		break;
2306 	case RTE_ETH_FC_RX_PAUSE:
2307 		if (fc_conf->autoneg) {
2308 			bp->link_info->auto_pause =
2309 					HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_RX;
2310 			bp->link_info->force_pause = 0;
2311 		} else {
2312 			bp->link_info->auto_pause = 0;
2313 			bp->link_info->force_pause =
2314 					HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_RX;
2315 		}
2316 		break;
2317 	case RTE_ETH_FC_TX_PAUSE:
2318 		if (fc_conf->autoneg) {
2319 			bp->link_info->auto_pause =
2320 					HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_TX;
2321 			bp->link_info->force_pause = 0;
2322 		} else {
2323 			bp->link_info->auto_pause = 0;
2324 			bp->link_info->force_pause =
2325 					HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_TX;
2326 		}
2327 		break;
2328 	case RTE_ETH_FC_FULL:
2329 		if (fc_conf->autoneg) {
2330 			bp->link_info->auto_pause =
2331 					HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_TX |
2332 					HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_RX;
2333 			bp->link_info->force_pause = 0;
2334 		} else {
2335 			bp->link_info->auto_pause = 0;
2336 			bp->link_info->force_pause =
2337 					HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_TX |
2338 					HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_RX;
2339 		}
2340 		break;
2341 	}
2342 	return bnxt_set_hwrm_link_config(bp, true);
2343 }
2344 
2345 /* Add UDP tunneling port */
2346 static int
2347 bnxt_udp_tunnel_port_add_op(struct rte_eth_dev *eth_dev,
2348 			 struct rte_eth_udp_tunnel *udp_tunnel)
2349 {
2350 	struct bnxt *bp = eth_dev->data->dev_private;
2351 	uint16_t tunnel_type = 0;
2352 	int rc = 0;
2353 
2354 	rc = is_bnxt_in_error(bp);
2355 	if (rc)
2356 		return rc;
2357 
2358 	switch (udp_tunnel->prot_type) {
2359 	case RTE_ETH_TUNNEL_TYPE_VXLAN:
2360 		if (bp->vxlan_port_cnt) {
2361 			PMD_DRV_LOG(ERR, "Tunnel Port %d already programmed\n",
2362 				udp_tunnel->udp_port);
2363 			if (bp->vxlan_port != udp_tunnel->udp_port) {
2364 				PMD_DRV_LOG(ERR, "Only one port allowed\n");
2365 				return -ENOSPC;
2366 			}
2367 			bp->vxlan_port_cnt++;
2368 			return 0;
2369 		}
2370 		tunnel_type =
2371 			HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_VXLAN;
2372 		break;
2373 	case RTE_ETH_TUNNEL_TYPE_GENEVE:
2374 		if (bp->geneve_port_cnt) {
2375 			PMD_DRV_LOG(ERR, "Tunnel Port %d already programmed\n",
2376 				udp_tunnel->udp_port);
2377 			if (bp->geneve_port != udp_tunnel->udp_port) {
2378 				PMD_DRV_LOG(ERR, "Only one port allowed\n");
2379 				return -ENOSPC;
2380 			}
2381 			bp->geneve_port_cnt++;
2382 			return 0;
2383 		}
2384 		tunnel_type =
2385 			HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_GENEVE;
2386 		break;
2387 	default:
2388 		PMD_DRV_LOG(ERR, "Tunnel type is not supported\n");
2389 		return -ENOTSUP;
2390 	}
2391 	rc = bnxt_hwrm_tunnel_dst_port_alloc(bp, udp_tunnel->udp_port,
2392 					     tunnel_type);
2393 
2394 	if (rc != 0)
2395 		return rc;
2396 
2397 	if (tunnel_type ==
2398 	    HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_VXLAN)
2399 		bp->vxlan_port_cnt++;
2400 
2401 	if (tunnel_type ==
2402 	    HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_GENEVE)
2403 		bp->geneve_port_cnt++;
2404 
2405 	return rc;
2406 }
2407 
2408 static int
2409 bnxt_udp_tunnel_port_del_op(struct rte_eth_dev *eth_dev,
2410 			 struct rte_eth_udp_tunnel *udp_tunnel)
2411 {
2412 	struct bnxt *bp = eth_dev->data->dev_private;
2413 	uint16_t tunnel_type = 0;
2414 	uint16_t port = 0;
2415 	int rc = 0;
2416 
2417 	rc = is_bnxt_in_error(bp);
2418 	if (rc)
2419 		return rc;
2420 
2421 	switch (udp_tunnel->prot_type) {
2422 	case RTE_ETH_TUNNEL_TYPE_VXLAN:
2423 		if (!bp->vxlan_port_cnt) {
2424 			PMD_DRV_LOG(ERR, "No Tunnel port configured yet\n");
2425 			return -EINVAL;
2426 		}
2427 		if (bp->vxlan_port != udp_tunnel->udp_port) {
2428 			PMD_DRV_LOG(ERR, "Req Port: %d. Configured port: %d\n",
2429 				udp_tunnel->udp_port, bp->vxlan_port);
2430 			return -EINVAL;
2431 		}
2432 		if (--bp->vxlan_port_cnt)
2433 			return 0;
2434 
2435 		tunnel_type =
2436 			HWRM_TUNNEL_DST_PORT_FREE_INPUT_TUNNEL_TYPE_VXLAN;
2437 		port = bp->vxlan_fw_dst_port_id;
2438 		break;
2439 	case RTE_ETH_TUNNEL_TYPE_GENEVE:
2440 		if (!bp->geneve_port_cnt) {
2441 			PMD_DRV_LOG(ERR, "No Tunnel port configured yet\n");
2442 			return -EINVAL;
2443 		}
2444 		if (bp->geneve_port != udp_tunnel->udp_port) {
2445 			PMD_DRV_LOG(ERR, "Req Port: %d. Configured port: %d\n",
2446 				udp_tunnel->udp_port, bp->geneve_port);
2447 			return -EINVAL;
2448 		}
2449 		if (--bp->geneve_port_cnt)
2450 			return 0;
2451 
2452 		tunnel_type =
2453 			HWRM_TUNNEL_DST_PORT_FREE_INPUT_TUNNEL_TYPE_GENEVE;
2454 		port = bp->geneve_fw_dst_port_id;
2455 		break;
2456 	default:
2457 		PMD_DRV_LOG(ERR, "Tunnel type is not supported\n");
2458 		return -ENOTSUP;
2459 	}
2460 
2461 	rc = bnxt_hwrm_tunnel_dst_port_free(bp, port, tunnel_type);
2462 	return rc;
2463 }
2464 
2465 static int bnxt_del_vlan_filter(struct bnxt *bp, uint16_t vlan_id)
2466 {
2467 	struct bnxt_filter_info *filter;
2468 	struct bnxt_vnic_info *vnic;
2469 	int rc = 0;
2470 	uint32_t chk = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN;
2471 
2472 	vnic = BNXT_GET_DEFAULT_VNIC(bp);
2473 	filter = STAILQ_FIRST(&vnic->filter);
2474 	while (filter) {
2475 		/* Search for this matching MAC+VLAN filter */
2476 		if (bnxt_vlan_filter_exists(bp, filter, chk, vlan_id)) {
2477 			/* Delete the filter */
2478 			rc = bnxt_hwrm_clear_l2_filter(bp, filter);
2479 			if (rc)
2480 				return rc;
2481 			STAILQ_REMOVE(&vnic->filter, filter,
2482 				      bnxt_filter_info, next);
2483 			bnxt_free_filter(bp, filter);
2484 			PMD_DRV_LOG(INFO,
2485 				    "Deleted vlan filter for %d\n",
2486 				    vlan_id);
2487 			return 0;
2488 		}
2489 		filter = STAILQ_NEXT(filter, next);
2490 	}
2491 	return -ENOENT;
2492 }
2493 
2494 static int bnxt_add_vlan_filter(struct bnxt *bp, uint16_t vlan_id)
2495 {
2496 	struct bnxt_filter_info *filter;
2497 	struct bnxt_vnic_info *vnic;
2498 	int rc = 0;
2499 	uint32_t en = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN |
2500 		HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN_MASK;
2501 	uint32_t chk = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN;
2502 
2503 	/* Implementation notes on the use of VNIC in this command:
2504 	 *
2505 	 * By default, these filters belong to default vnic for the function.
2506 	 * Once these filters are set up, only destination VNIC can be modified.
2507 	 * If the destination VNIC is not specified in this command,
2508 	 * then the HWRM shall only create an l2 context id.
2509 	 */
2510 
2511 	vnic = BNXT_GET_DEFAULT_VNIC(bp);
2512 	filter = STAILQ_FIRST(&vnic->filter);
2513 	/* Check if the VLAN has already been added */
2514 	while (filter) {
2515 		if (bnxt_vlan_filter_exists(bp, filter, chk, vlan_id))
2516 			return -EEXIST;
2517 
2518 		filter = STAILQ_NEXT(filter, next);
2519 	}
2520 
2521 	/* No match found. Alloc a fresh filter and issue the L2_FILTER_ALLOC
2522 	 * command to create MAC+VLAN filter with the right flags, enables set.
2523 	 */
2524 	filter = bnxt_alloc_filter(bp);
2525 	if (!filter) {
2526 		PMD_DRV_LOG(ERR,
2527 			    "MAC/VLAN filter alloc failed\n");
2528 		return -ENOMEM;
2529 	}
2530 	/* MAC + VLAN ID filter */
2531 	/* If l2_ivlan == 0 and l2_ivlan_mask != 0, only
2532 	 * untagged packets are received
2533 	 *
2534 	 * If l2_ivlan != 0 and l2_ivlan_mask != 0, untagged
2535 	 * packets and only the programmed vlan's packets are received
2536 	 */
2537 	filter->l2_ivlan = vlan_id;
2538 	filter->l2_ivlan_mask = 0x0FFF;
2539 	filter->enables |= en;
2540 	filter->flags |= HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
2541 
2542 	rc = bnxt_hwrm_set_l2_filter(bp, vnic->fw_vnic_id, filter);
2543 	if (rc) {
2544 		/* Free the newly allocated filter as we were
2545 		 * not able to create the filter in hardware.
2546 		 */
2547 		bnxt_free_filter(bp, filter);
2548 		return rc;
2549 	}
2550 
2551 	filter->mac_index = 0;
2552 	/* Add this new filter to the list */
2553 	if (vlan_id == 0)
2554 		STAILQ_INSERT_HEAD(&vnic->filter, filter, next);
2555 	else
2556 		STAILQ_INSERT_TAIL(&vnic->filter, filter, next);
2557 
2558 	PMD_DRV_LOG(INFO,
2559 		    "Added Vlan filter for %d\n", vlan_id);
2560 	return rc;
2561 }
2562 
2563 static int bnxt_vlan_filter_set_op(struct rte_eth_dev *eth_dev,
2564 		uint16_t vlan_id, int on)
2565 {
2566 	struct bnxt *bp = eth_dev->data->dev_private;
2567 	int rc;
2568 
2569 	rc = is_bnxt_in_error(bp);
2570 	if (rc)
2571 		return rc;
2572 
2573 	if (!eth_dev->data->dev_started) {
2574 		PMD_DRV_LOG(ERR, "port must be started before setting vlan\n");
2575 		return -EINVAL;
2576 	}
2577 
2578 	/* These operations apply to ALL existing MAC/VLAN filters */
2579 	if (on)
2580 		return bnxt_add_vlan_filter(bp, vlan_id);
2581 	else
2582 		return bnxt_del_vlan_filter(bp, vlan_id);
2583 }
2584 
2585 static int bnxt_del_dflt_mac_filter(struct bnxt *bp,
2586 				    struct bnxt_vnic_info *vnic)
2587 {
2588 	struct bnxt_filter_info *filter;
2589 	int rc;
2590 
2591 	filter = STAILQ_FIRST(&vnic->filter);
2592 	while (filter) {
2593 		if (filter->mac_index == 0 &&
2594 		    !memcmp(filter->l2_addr, bp->mac_addr,
2595 			    RTE_ETHER_ADDR_LEN)) {
2596 			rc = bnxt_hwrm_clear_l2_filter(bp, filter);
2597 			if (!rc) {
2598 				STAILQ_REMOVE(&vnic->filter, filter,
2599 					      bnxt_filter_info, next);
2600 				bnxt_free_filter(bp, filter);
2601 			}
2602 			return rc;
2603 		}
2604 		filter = STAILQ_NEXT(filter, next);
2605 	}
2606 	return 0;
2607 }
2608 
2609 static int
2610 bnxt_config_vlan_hw_filter(struct bnxt *bp, uint64_t rx_offloads)
2611 {
2612 	struct bnxt_vnic_info *vnic;
2613 	unsigned int i;
2614 	int rc;
2615 
2616 	vnic = BNXT_GET_DEFAULT_VNIC(bp);
2617 	if (!(rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER)) {
2618 		/* Remove any VLAN filters programmed */
2619 		for (i = 0; i < RTE_ETHER_MAX_VLAN_ID; i++)
2620 			bnxt_del_vlan_filter(bp, i);
2621 
2622 		rc = bnxt_add_mac_filter(bp, vnic, NULL, 0, 0);
2623 		if (rc)
2624 			return rc;
2625 	} else {
2626 		/* Default filter will allow packets that match the
2627 		 * dest mac. So, it has to be deleted, otherwise, we
2628 		 * will endup receiving vlan packets for which the
2629 		 * filter is not programmed, when hw-vlan-filter
2630 		 * configuration is ON
2631 		 */
2632 		bnxt_del_dflt_mac_filter(bp, vnic);
2633 		/* This filter will allow only untagged packets */
2634 		bnxt_add_vlan_filter(bp, 0);
2635 	}
2636 	PMD_DRV_LOG(DEBUG, "VLAN Filtering: %d\n",
2637 		    !!(rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER));
2638 
2639 	return 0;
2640 }
2641 
2642 static int bnxt_free_one_vnic(struct bnxt *bp, uint16_t vnic_id)
2643 {
2644 	struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
2645 	unsigned int i;
2646 	int rc;
2647 
2648 	/* Destroy vnic filters and vnic */
2649 	if (bp->eth_dev->data->dev_conf.rxmode.offloads &
2650 	    RTE_ETH_RX_OFFLOAD_VLAN_FILTER) {
2651 		for (i = 0; i < RTE_ETHER_MAX_VLAN_ID; i++)
2652 			bnxt_del_vlan_filter(bp, i);
2653 	}
2654 	bnxt_del_dflt_mac_filter(bp, vnic);
2655 
2656 	rc = bnxt_hwrm_vnic_ctx_free(bp, vnic);
2657 	if (rc)
2658 		return rc;
2659 
2660 	rc = bnxt_hwrm_vnic_free(bp, vnic);
2661 	if (rc)
2662 		return rc;
2663 
2664 	rte_free(vnic->fw_grp_ids);
2665 	vnic->fw_grp_ids = NULL;
2666 
2667 	vnic->rx_queue_cnt = 0;
2668 
2669 	return 0;
2670 }
2671 
2672 static int
2673 bnxt_config_vlan_hw_stripping(struct bnxt *bp, uint64_t rx_offloads)
2674 {
2675 	struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2676 	int rc;
2677 
2678 	/* Destroy, recreate and reconfigure the default vnic */
2679 	rc = bnxt_free_one_vnic(bp, 0);
2680 	if (rc)
2681 		return rc;
2682 
2683 	/* default vnic 0 */
2684 	rc = bnxt_setup_one_vnic(bp, 0);
2685 	if (rc)
2686 		return rc;
2687 
2688 	if (bp->eth_dev->data->dev_conf.rxmode.offloads &
2689 	    RTE_ETH_RX_OFFLOAD_VLAN_FILTER) {
2690 		rc = bnxt_add_vlan_filter(bp, 0);
2691 		if (rc)
2692 			return rc;
2693 		rc = bnxt_restore_vlan_filters(bp);
2694 		if (rc)
2695 			return rc;
2696 	} else {
2697 		rc = bnxt_add_mac_filter(bp, vnic, NULL, 0, 0);
2698 		if (rc)
2699 			return rc;
2700 	}
2701 
2702 	rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
2703 	if (rc)
2704 		return rc;
2705 
2706 	PMD_DRV_LOG(DEBUG, "VLAN Strip Offload: %d\n",
2707 		    !!(rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP));
2708 
2709 	return rc;
2710 }
2711 
2712 static int
2713 bnxt_vlan_offload_set_op(struct rte_eth_dev *dev, int mask)
2714 {
2715 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
2716 	struct bnxt *bp = dev->data->dev_private;
2717 	int rc;
2718 
2719 	rc = is_bnxt_in_error(bp);
2720 	if (rc)
2721 		return rc;
2722 
2723 	/* Filter settings will get applied when port is started */
2724 	if (!dev->data->dev_started)
2725 		return 0;
2726 
2727 	if (mask & RTE_ETH_VLAN_FILTER_MASK) {
2728 		/* Enable or disable VLAN filtering */
2729 		rc = bnxt_config_vlan_hw_filter(bp, rx_offloads);
2730 		if (rc)
2731 			return rc;
2732 	}
2733 
2734 	if (mask & RTE_ETH_VLAN_STRIP_MASK) {
2735 		/* Enable or disable VLAN stripping */
2736 		rc = bnxt_config_vlan_hw_stripping(bp, rx_offloads);
2737 		if (rc)
2738 			return rc;
2739 	}
2740 
2741 	if (mask & RTE_ETH_VLAN_EXTEND_MASK) {
2742 		if (rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_EXTEND)
2743 			PMD_DRV_LOG(DEBUG, "Extend VLAN supported\n");
2744 		else
2745 			PMD_DRV_LOG(INFO, "Extend VLAN unsupported\n");
2746 	}
2747 
2748 	return 0;
2749 }
2750 
2751 static int
2752 bnxt_vlan_tpid_set_op(struct rte_eth_dev *dev, enum rte_vlan_type vlan_type,
2753 		      uint16_t tpid)
2754 {
2755 	struct bnxt *bp = dev->data->dev_private;
2756 	int qinq = dev->data->dev_conf.rxmode.offloads &
2757 		   RTE_ETH_RX_OFFLOAD_VLAN_EXTEND;
2758 
2759 	if (vlan_type != RTE_ETH_VLAN_TYPE_INNER &&
2760 	    vlan_type != RTE_ETH_VLAN_TYPE_OUTER) {
2761 		PMD_DRV_LOG(ERR,
2762 			    "Unsupported vlan type.");
2763 		return -EINVAL;
2764 	}
2765 	if (!qinq) {
2766 		PMD_DRV_LOG(ERR,
2767 			    "QinQ not enabled. Needs to be ON as we can "
2768 			    "accelerate only outer vlan\n");
2769 		return -EINVAL;
2770 	}
2771 
2772 	if (vlan_type == RTE_ETH_VLAN_TYPE_OUTER) {
2773 		switch (tpid) {
2774 		case RTE_ETHER_TYPE_QINQ:
2775 			bp->outer_tpid_bd =
2776 				TX_BD_LONG_CFA_META_VLAN_TPID_TPID88A8;
2777 				break;
2778 		case RTE_ETHER_TYPE_VLAN:
2779 			bp->outer_tpid_bd =
2780 				TX_BD_LONG_CFA_META_VLAN_TPID_TPID8100;
2781 				break;
2782 		case RTE_ETHER_TYPE_QINQ1:
2783 			bp->outer_tpid_bd =
2784 				TX_BD_LONG_CFA_META_VLAN_TPID_TPID9100;
2785 				break;
2786 		case RTE_ETHER_TYPE_QINQ2:
2787 			bp->outer_tpid_bd =
2788 				TX_BD_LONG_CFA_META_VLAN_TPID_TPID9200;
2789 				break;
2790 		case RTE_ETHER_TYPE_QINQ3:
2791 			bp->outer_tpid_bd =
2792 				 TX_BD_LONG_CFA_META_VLAN_TPID_TPID9300;
2793 				break;
2794 		default:
2795 			PMD_DRV_LOG(ERR, "Invalid TPID: %x\n", tpid);
2796 			return -EINVAL;
2797 		}
2798 		bp->outer_tpid_bd |= tpid;
2799 		PMD_DRV_LOG(INFO, "outer_tpid_bd = %x\n", bp->outer_tpid_bd);
2800 	} else if (vlan_type == RTE_ETH_VLAN_TYPE_INNER) {
2801 		PMD_DRV_LOG(ERR,
2802 			    "Can accelerate only outer vlan in QinQ\n");
2803 		return -EINVAL;
2804 	}
2805 
2806 	return 0;
2807 }
2808 
2809 static int
2810 bnxt_set_default_mac_addr_op(struct rte_eth_dev *dev,
2811 			     struct rte_ether_addr *addr)
2812 {
2813 	struct bnxt *bp = dev->data->dev_private;
2814 	/* Default Filter is tied to VNIC 0 */
2815 	struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2816 	int rc;
2817 
2818 	rc = is_bnxt_in_error(bp);
2819 	if (rc)
2820 		return rc;
2821 
2822 	if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp))
2823 		return -EPERM;
2824 
2825 	if (rte_is_zero_ether_addr(addr))
2826 		return -EINVAL;
2827 
2828 	/* Filter settings will get applied when port is started */
2829 	if (!dev->data->dev_started)
2830 		return 0;
2831 
2832 	/* Check if the requested MAC is already added */
2833 	if (memcmp(addr, bp->mac_addr, RTE_ETHER_ADDR_LEN) == 0)
2834 		return 0;
2835 
2836 	/* Destroy filter and re-create it */
2837 	bnxt_del_dflt_mac_filter(bp, vnic);
2838 
2839 	memcpy(bp->mac_addr, addr, RTE_ETHER_ADDR_LEN);
2840 	if (dev->data->dev_conf.rxmode.offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER) {
2841 		/* This filter will allow only untagged packets */
2842 		rc = bnxt_add_vlan_filter(bp, 0);
2843 	} else {
2844 		rc = bnxt_add_mac_filter(bp, vnic, addr, 0, 0);
2845 	}
2846 
2847 	PMD_DRV_LOG(DEBUG, "Set MAC addr\n");
2848 	return rc;
2849 }
2850 
2851 static int
2852 bnxt_dev_set_mc_addr_list_op(struct rte_eth_dev *eth_dev,
2853 			  struct rte_ether_addr *mc_addr_set,
2854 			  uint32_t nb_mc_addr)
2855 {
2856 	struct bnxt *bp = eth_dev->data->dev_private;
2857 	struct bnxt_vnic_info *vnic;
2858 	uint32_t i = 0;
2859 	int rc;
2860 
2861 	rc = is_bnxt_in_error(bp);
2862 	if (rc)
2863 		return rc;
2864 
2865 	vnic = BNXT_GET_DEFAULT_VNIC(bp);
2866 
2867 	bp->nb_mc_addr = nb_mc_addr;
2868 
2869 	if (nb_mc_addr > BNXT_MAX_MC_ADDRS) {
2870 		vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
2871 		goto allmulti;
2872 	}
2873 
2874 	/* TODO Check for Duplicate mcast addresses */
2875 	vnic->flags &= ~BNXT_VNIC_INFO_ALLMULTI;
2876 	for (i = 0; i < nb_mc_addr; i++)
2877 		rte_ether_addr_copy(&mc_addr_set[i], &bp->mcast_addr_list[i]);
2878 
2879 	if (bp->nb_mc_addr)
2880 		vnic->flags |= BNXT_VNIC_INFO_MCAST;
2881 	else
2882 		vnic->flags &= ~BNXT_VNIC_INFO_MCAST;
2883 
2884 allmulti:
2885 	return bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
2886 }
2887 
2888 static int
2889 bnxt_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
2890 {
2891 	struct bnxt *bp = dev->data->dev_private;
2892 	uint8_t fw_major = (bp->fw_ver >> 24) & 0xff;
2893 	uint8_t fw_minor = (bp->fw_ver >> 16) & 0xff;
2894 	uint8_t fw_updt = (bp->fw_ver >> 8) & 0xff;
2895 	uint8_t fw_rsvd = bp->fw_ver & 0xff;
2896 	int ret;
2897 
2898 	ret = snprintf(fw_version, fw_size, "%d.%d.%d.%d",
2899 			fw_major, fw_minor, fw_updt, fw_rsvd);
2900 	if (ret < 0)
2901 		return -EINVAL;
2902 
2903 	ret += 1; /* add the size of '\0' */
2904 	if (fw_size < (size_t)ret)
2905 		return ret;
2906 	else
2907 		return 0;
2908 }
2909 
2910 static void
2911 bnxt_rxq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
2912 	struct rte_eth_rxq_info *qinfo)
2913 {
2914 	struct bnxt *bp = dev->data->dev_private;
2915 	struct bnxt_rx_queue *rxq;
2916 
2917 	if (is_bnxt_in_error(bp))
2918 		return;
2919 
2920 	rxq = dev->data->rx_queues[queue_id];
2921 
2922 	qinfo->mp = rxq->mb_pool;
2923 	qinfo->scattered_rx = dev->data->scattered_rx;
2924 	qinfo->nb_desc = rxq->nb_rx_desc;
2925 
2926 	qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
2927 	qinfo->conf.rx_drop_en = rxq->drop_en;
2928 	qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
2929 	qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
2930 }
2931 
2932 static void
2933 bnxt_txq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
2934 	struct rte_eth_txq_info *qinfo)
2935 {
2936 	struct bnxt *bp = dev->data->dev_private;
2937 	struct bnxt_tx_queue *txq;
2938 
2939 	if (is_bnxt_in_error(bp))
2940 		return;
2941 
2942 	txq = dev->data->tx_queues[queue_id];
2943 
2944 	qinfo->nb_desc = txq->nb_tx_desc;
2945 
2946 	qinfo->conf.tx_thresh.pthresh = txq->pthresh;
2947 	qinfo->conf.tx_thresh.hthresh = txq->hthresh;
2948 	qinfo->conf.tx_thresh.wthresh = txq->wthresh;
2949 
2950 	qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
2951 	qinfo->conf.tx_rs_thresh = 0;
2952 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
2953 	qinfo->conf.offloads = txq->offloads;
2954 }
2955 
2956 static const struct {
2957 	eth_rx_burst_t pkt_burst;
2958 	const char *info;
2959 } bnxt_rx_burst_info[] = {
2960 	{bnxt_recv_pkts,		"Scalar"},
2961 #if defined(RTE_ARCH_X86)
2962 	{bnxt_recv_pkts_vec,		"Vector SSE"},
2963 #endif
2964 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
2965 	{bnxt_recv_pkts_vec_avx2,	"Vector AVX2"},
2966 #endif
2967 #if defined(RTE_ARCH_ARM64)
2968 	{bnxt_recv_pkts_vec,		"Vector Neon"},
2969 #endif
2970 };
2971 
2972 static int
2973 bnxt_rx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
2974 		       struct rte_eth_burst_mode *mode)
2975 {
2976 	eth_rx_burst_t pkt_burst = dev->rx_pkt_burst;
2977 	size_t i;
2978 
2979 	for (i = 0; i < RTE_DIM(bnxt_rx_burst_info); i++) {
2980 		if (pkt_burst == bnxt_rx_burst_info[i].pkt_burst) {
2981 			snprintf(mode->info, sizeof(mode->info), "%s",
2982 				 bnxt_rx_burst_info[i].info);
2983 			return 0;
2984 		}
2985 	}
2986 
2987 	return -EINVAL;
2988 }
2989 
2990 static const struct {
2991 	eth_tx_burst_t pkt_burst;
2992 	const char *info;
2993 } bnxt_tx_burst_info[] = {
2994 	{bnxt_xmit_pkts,		"Scalar"},
2995 #if defined(RTE_ARCH_X86)
2996 	{bnxt_xmit_pkts_vec,		"Vector SSE"},
2997 #endif
2998 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
2999 	{bnxt_xmit_pkts_vec_avx2,	"Vector AVX2"},
3000 #endif
3001 #if defined(RTE_ARCH_ARM64)
3002 	{bnxt_xmit_pkts_vec,		"Vector Neon"},
3003 #endif
3004 };
3005 
3006 static int
3007 bnxt_tx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
3008 		       struct rte_eth_burst_mode *mode)
3009 {
3010 	eth_tx_burst_t pkt_burst = dev->tx_pkt_burst;
3011 	size_t i;
3012 
3013 	for (i = 0; i < RTE_DIM(bnxt_tx_burst_info); i++) {
3014 		if (pkt_burst == bnxt_tx_burst_info[i].pkt_burst) {
3015 			snprintf(mode->info, sizeof(mode->info), "%s",
3016 				 bnxt_tx_burst_info[i].info);
3017 			return 0;
3018 		}
3019 	}
3020 
3021 	return -EINVAL;
3022 }
3023 
3024 int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
3025 {
3026 	uint32_t overhead = BNXT_MAX_PKT_LEN - BNXT_MAX_MTU;
3027 	struct bnxt *bp = eth_dev->data->dev_private;
3028 	uint32_t new_pkt_size;
3029 	uint32_t rc;
3030 	uint32_t i;
3031 
3032 	rc = is_bnxt_in_error(bp);
3033 	if (rc)
3034 		return rc;
3035 
3036 	/* Exit if receive queues are not configured yet */
3037 	if (!eth_dev->data->nb_rx_queues)
3038 		return rc;
3039 
3040 	new_pkt_size = new_mtu + overhead;
3041 
3042 	/*
3043 	 * Disallow any MTU change that would require scattered receive support
3044 	 * if it is not already enabled.
3045 	 */
3046 	if (eth_dev->data->dev_started &&
3047 	    !eth_dev->data->scattered_rx &&
3048 	    (new_pkt_size >
3049 	     eth_dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM)) {
3050 		PMD_DRV_LOG(ERR,
3051 			    "MTU change would require scattered rx support. ");
3052 		PMD_DRV_LOG(ERR, "Stop port before changing MTU.\n");
3053 		return -EINVAL;
3054 	}
3055 
3056 	if (new_mtu > RTE_ETHER_MTU)
3057 		bp->flags |= BNXT_FLAG_JUMBO;
3058 	else
3059 		bp->flags &= ~BNXT_FLAG_JUMBO;
3060 
3061 	/* Is there a change in mtu setting? */
3062 	if (eth_dev->data->mtu == new_mtu)
3063 		return rc;
3064 
3065 	for (i = 0; i < bp->nr_vnics; i++) {
3066 		struct bnxt_vnic_info *vnic = &bp->vnic_info[i];
3067 		uint16_t size = 0;
3068 
3069 		vnic->mru = BNXT_VNIC_MRU(new_mtu);
3070 		rc = bnxt_hwrm_vnic_cfg(bp, vnic);
3071 		if (rc)
3072 			break;
3073 
3074 		size = rte_pktmbuf_data_room_size(bp->rx_queues[0]->mb_pool);
3075 		size -= RTE_PKTMBUF_HEADROOM;
3076 
3077 		if (size < new_mtu) {
3078 			rc = bnxt_hwrm_vnic_plcmode_cfg(bp, vnic);
3079 			if (rc)
3080 				return rc;
3081 		}
3082 	}
3083 
3084 	if (bnxt_hwrm_config_host_mtu(bp))
3085 		PMD_DRV_LOG(WARNING, "Failed to configure host MTU\n");
3086 
3087 	PMD_DRV_LOG(INFO, "New MTU is %d\n", new_mtu);
3088 
3089 	return rc;
3090 }
3091 
3092 static int
3093 bnxt_vlan_pvid_set_op(struct rte_eth_dev *dev, uint16_t pvid, int on)
3094 {
3095 	struct bnxt *bp = dev->data->dev_private;
3096 	uint16_t vlan = bp->vlan;
3097 	int rc;
3098 
3099 	rc = is_bnxt_in_error(bp);
3100 	if (rc)
3101 		return rc;
3102 
3103 	if (!BNXT_SINGLE_PF(bp)) {
3104 		PMD_DRV_LOG(ERR, "PVID cannot be modified on VF or on shared PF\n");
3105 		return -ENOTSUP;
3106 	}
3107 	bp->vlan = on ? pvid : 0;
3108 
3109 	rc = bnxt_hwrm_set_default_vlan(bp, 0, 0);
3110 	if (rc)
3111 		bp->vlan = vlan;
3112 	return rc;
3113 }
3114 
3115 static int
3116 bnxt_dev_led_on_op(struct rte_eth_dev *dev)
3117 {
3118 	struct bnxt *bp = dev->data->dev_private;
3119 	int rc;
3120 
3121 	rc = is_bnxt_in_error(bp);
3122 	if (rc)
3123 		return rc;
3124 
3125 	return bnxt_hwrm_port_led_cfg(bp, true);
3126 }
3127 
3128 static int
3129 bnxt_dev_led_off_op(struct rte_eth_dev *dev)
3130 {
3131 	struct bnxt *bp = dev->data->dev_private;
3132 	int rc;
3133 
3134 	rc = is_bnxt_in_error(bp);
3135 	if (rc)
3136 		return rc;
3137 
3138 	return bnxt_hwrm_port_led_cfg(bp, false);
3139 }
3140 
3141 static uint32_t
3142 bnxt_rx_queue_count_op(void *rx_queue)
3143 {
3144 	struct bnxt *bp;
3145 	struct bnxt_cp_ring_info *cpr;
3146 	uint32_t desc = 0, raw_cons, cp_ring_size;
3147 	struct bnxt_rx_queue *rxq;
3148 	struct rx_pkt_cmpl *rxcmp;
3149 	int rc;
3150 
3151 	rxq = rx_queue;
3152 	bp = rxq->bp;
3153 
3154 	rc = is_bnxt_in_error(bp);
3155 	if (rc)
3156 		return rc;
3157 
3158 	cpr = rxq->cp_ring;
3159 	raw_cons = cpr->cp_raw_cons;
3160 	cp_ring_size = cpr->cp_ring_struct->ring_size;
3161 
3162 	while (1) {
3163 		uint32_t agg_cnt, cons, cmpl_type;
3164 
3165 		cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3166 		rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3167 
3168 		if (!bnxt_cpr_cmp_valid(rxcmp, raw_cons, cp_ring_size))
3169 			break;
3170 
3171 		cmpl_type = CMP_TYPE(rxcmp);
3172 
3173 		switch (cmpl_type) {
3174 		case CMPL_BASE_TYPE_RX_L2:
3175 		case CMPL_BASE_TYPE_RX_L2_V2:
3176 			agg_cnt = BNXT_RX_L2_AGG_BUFS(rxcmp);
3177 			raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3178 			desc++;
3179 			break;
3180 
3181 		case CMPL_BASE_TYPE_RX_TPA_END:
3182 			if (BNXT_CHIP_P5(rxq->bp)) {
3183 				struct rx_tpa_v2_end_cmpl_hi *p5_tpa_end;
3184 
3185 				p5_tpa_end = (void *)rxcmp;
3186 				agg_cnt = BNXT_TPA_END_AGG_BUFS_TH(p5_tpa_end);
3187 			} else {
3188 				struct rx_tpa_end_cmpl *tpa_end;
3189 
3190 				tpa_end = (void *)rxcmp;
3191 				agg_cnt = BNXT_TPA_END_AGG_BUFS(tpa_end);
3192 			}
3193 
3194 			raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3195 			desc++;
3196 			break;
3197 
3198 		default:
3199 			raw_cons += CMP_LEN(cmpl_type);
3200 		}
3201 	}
3202 
3203 	return desc;
3204 }
3205 
3206 static int
3207 bnxt_rx_descriptor_status_op(void *rx_queue, uint16_t offset)
3208 {
3209 	struct bnxt_rx_queue *rxq = rx_queue;
3210 	struct bnxt_cp_ring_info *cpr;
3211 	struct bnxt_rx_ring_info *rxr;
3212 	uint32_t desc, raw_cons, cp_ring_size;
3213 	struct bnxt *bp = rxq->bp;
3214 	struct rx_pkt_cmpl *rxcmp;
3215 	int rc;
3216 
3217 	rc = is_bnxt_in_error(bp);
3218 	if (rc)
3219 		return rc;
3220 
3221 	if (offset >= rxq->nb_rx_desc)
3222 		return -EINVAL;
3223 
3224 	rxr = rxq->rx_ring;
3225 	cpr = rxq->cp_ring;
3226 	cp_ring_size = cpr->cp_ring_struct->ring_size;
3227 
3228 	/*
3229 	 * For the vector receive case, the completion at the requested
3230 	 * offset can be indexed directly.
3231 	 */
3232 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
3233 	if (bp->flags & BNXT_FLAG_RX_VECTOR_PKT_MODE) {
3234 		struct rx_pkt_cmpl *rxcmp;
3235 		uint32_t cons;
3236 
3237 		/* Check status of completion descriptor. */
3238 		raw_cons = cpr->cp_raw_cons +
3239 			   offset * CMP_LEN(CMPL_BASE_TYPE_RX_L2);
3240 		cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3241 		rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3242 
3243 		if (bnxt_cpr_cmp_valid(rxcmp, raw_cons, cp_ring_size))
3244 			return RTE_ETH_RX_DESC_DONE;
3245 
3246 		/* Check whether rx desc has an mbuf attached. */
3247 		cons = RING_CMP(rxr->rx_ring_struct, raw_cons / 2);
3248 		if (cons >= rxq->rxrearm_start &&
3249 		    cons < rxq->rxrearm_start + rxq->rxrearm_nb) {
3250 			return RTE_ETH_RX_DESC_UNAVAIL;
3251 		}
3252 
3253 		return RTE_ETH_RX_DESC_AVAIL;
3254 	}
3255 #endif
3256 
3257 	/*
3258 	 * For the non-vector receive case, scan the completion ring to
3259 	 * locate the completion descriptor for the requested offset.
3260 	 */
3261 	raw_cons = cpr->cp_raw_cons;
3262 	desc = 0;
3263 	while (1) {
3264 		uint32_t agg_cnt, cons, cmpl_type;
3265 
3266 		cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3267 		rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3268 
3269 		if (!bnxt_cpr_cmp_valid(rxcmp, raw_cons, cp_ring_size))
3270 			break;
3271 
3272 		cmpl_type = CMP_TYPE(rxcmp);
3273 
3274 		switch (cmpl_type) {
3275 		case CMPL_BASE_TYPE_RX_L2:
3276 		case CMPL_BASE_TYPE_RX_L2_V2:
3277 			if (desc == offset) {
3278 				cons = rxcmp->opaque;
3279 				if (rxr->rx_buf_ring[cons])
3280 					return RTE_ETH_RX_DESC_DONE;
3281 				else
3282 					return RTE_ETH_RX_DESC_UNAVAIL;
3283 			}
3284 			agg_cnt = BNXT_RX_L2_AGG_BUFS(rxcmp);
3285 			raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3286 			desc++;
3287 			break;
3288 
3289 		case CMPL_BASE_TYPE_RX_TPA_END:
3290 			if (desc == offset)
3291 				return RTE_ETH_RX_DESC_DONE;
3292 
3293 			if (BNXT_CHIP_P5(rxq->bp)) {
3294 				struct rx_tpa_v2_end_cmpl_hi *p5_tpa_end;
3295 
3296 				p5_tpa_end = (void *)rxcmp;
3297 				agg_cnt = BNXT_TPA_END_AGG_BUFS_TH(p5_tpa_end);
3298 			} else {
3299 				struct rx_tpa_end_cmpl *tpa_end;
3300 
3301 				tpa_end = (void *)rxcmp;
3302 				agg_cnt = BNXT_TPA_END_AGG_BUFS(tpa_end);
3303 			}
3304 
3305 			raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3306 			desc++;
3307 			break;
3308 
3309 		default:
3310 			raw_cons += CMP_LEN(cmpl_type);
3311 		}
3312 	}
3313 
3314 	return RTE_ETH_RX_DESC_AVAIL;
3315 }
3316 
3317 static int
3318 bnxt_tx_descriptor_status_op(void *tx_queue, uint16_t offset)
3319 {
3320 	struct bnxt_tx_queue *txq = (struct bnxt_tx_queue *)tx_queue;
3321 	struct bnxt_cp_ring_info *cpr = txq->cp_ring;
3322 	uint32_t ring_mask, raw_cons, nb_tx_pkts = 0;
3323 	struct cmpl_base *cp_desc_ring;
3324 	int rc;
3325 
3326 	rc = is_bnxt_in_error(txq->bp);
3327 	if (rc)
3328 		return rc;
3329 
3330 	if (offset >= txq->nb_tx_desc)
3331 		return -EINVAL;
3332 
3333 	/* Return "desc done" if descriptor is available for use. */
3334 	if (bnxt_tx_bds_in_hw(txq) <= offset)
3335 		return RTE_ETH_TX_DESC_DONE;
3336 
3337 	raw_cons = cpr->cp_raw_cons;
3338 	cp_desc_ring = cpr->cp_desc_ring;
3339 	ring_mask = cpr->cp_ring_struct->ring_mask;
3340 
3341 	/* Check to see if hw has posted a completion for the descriptor. */
3342 	while (1) {
3343 		struct tx_cmpl *txcmp;
3344 		uint32_t cons;
3345 
3346 		cons = RING_CMPL(ring_mask, raw_cons);
3347 		txcmp = (struct tx_cmpl *)&cp_desc_ring[cons];
3348 
3349 		if (!bnxt_cpr_cmp_valid(txcmp, raw_cons, ring_mask + 1))
3350 			break;
3351 
3352 		if (CMP_TYPE(txcmp) == TX_CMPL_TYPE_TX_L2)
3353 			nb_tx_pkts += rte_le_to_cpu_32(txcmp->opaque);
3354 
3355 		if (nb_tx_pkts > offset)
3356 			return RTE_ETH_TX_DESC_DONE;
3357 
3358 		raw_cons = NEXT_RAW_CMP(raw_cons);
3359 	}
3360 
3361 	/* Descriptor is pending transmit, not yet completed by hardware. */
3362 	return RTE_ETH_TX_DESC_FULL;
3363 }
3364 
3365 int
3366 bnxt_flow_ops_get_op(struct rte_eth_dev *dev,
3367 		     const struct rte_flow_ops **ops)
3368 {
3369 	struct bnxt *bp = dev->data->dev_private;
3370 	int ret = 0;
3371 
3372 	if (!bp)
3373 		return -EIO;
3374 
3375 	if (BNXT_ETH_DEV_IS_REPRESENTOR(dev)) {
3376 		struct bnxt_representor *vfr = dev->data->dev_private;
3377 		bp = vfr->parent_dev->data->dev_private;
3378 		/* parent is deleted while children are still valid */
3379 		if (!bp) {
3380 			PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR Error\n",
3381 				    dev->data->port_id);
3382 			return -EIO;
3383 		}
3384 	}
3385 
3386 	ret = is_bnxt_in_error(bp);
3387 	if (ret)
3388 		return ret;
3389 
3390 	/* PMD supports thread-safe flow operations.  rte_flow API
3391 	 * functions can avoid mutex for multi-thread safety.
3392 	 */
3393 	dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
3394 
3395 	if (BNXT_TRUFLOW_EN(bp))
3396 		*ops = &bnxt_ulp_rte_flow_ops;
3397 	else
3398 		*ops = &bnxt_flow_ops;
3399 
3400 	return ret;
3401 }
3402 
3403 static const uint32_t *
3404 bnxt_dev_supported_ptypes_get_op(struct rte_eth_dev *dev)
3405 {
3406 	static const uint32_t ptypes[] = {
3407 		RTE_PTYPE_L2_ETHER_VLAN,
3408 		RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
3409 		RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
3410 		RTE_PTYPE_L4_ICMP,
3411 		RTE_PTYPE_L4_TCP,
3412 		RTE_PTYPE_L4_UDP,
3413 		RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
3414 		RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
3415 		RTE_PTYPE_INNER_L4_ICMP,
3416 		RTE_PTYPE_INNER_L4_TCP,
3417 		RTE_PTYPE_INNER_L4_UDP,
3418 		RTE_PTYPE_UNKNOWN
3419 	};
3420 
3421 	if (!dev->rx_pkt_burst)
3422 		return NULL;
3423 
3424 	return ptypes;
3425 }
3426 
3427 static int bnxt_map_regs(struct bnxt *bp, uint32_t *reg_arr, int count,
3428 			 int reg_win)
3429 {
3430 	uint32_t reg_base = *reg_arr & 0xfffff000;
3431 	uint32_t win_off;
3432 	int i;
3433 
3434 	for (i = 0; i < count; i++) {
3435 		if ((reg_arr[i] & 0xfffff000) != reg_base)
3436 			return -ERANGE;
3437 	}
3438 	win_off = BNXT_GRCPF_REG_WINDOW_BASE_OUT + (reg_win - 1) * 4;
3439 	rte_write32(reg_base, (uint8_t *)bp->bar0 + win_off);
3440 	return 0;
3441 }
3442 
3443 static int bnxt_map_ptp_regs(struct bnxt *bp)
3444 {
3445 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3446 	uint32_t *reg_arr;
3447 	int rc, i;
3448 
3449 	reg_arr = ptp->rx_regs;
3450 	rc = bnxt_map_regs(bp, reg_arr, BNXT_PTP_RX_REGS, 5);
3451 	if (rc)
3452 		return rc;
3453 
3454 	reg_arr = ptp->tx_regs;
3455 	rc = bnxt_map_regs(bp, reg_arr, BNXT_PTP_TX_REGS, 6);
3456 	if (rc)
3457 		return rc;
3458 
3459 	for (i = 0; i < BNXT_PTP_RX_REGS; i++)
3460 		ptp->rx_mapped_regs[i] = 0x5000 + (ptp->rx_regs[i] & 0xfff);
3461 
3462 	for (i = 0; i < BNXT_PTP_TX_REGS; i++)
3463 		ptp->tx_mapped_regs[i] = 0x6000 + (ptp->tx_regs[i] & 0xfff);
3464 
3465 	return 0;
3466 }
3467 
3468 static void bnxt_unmap_ptp_regs(struct bnxt *bp)
3469 {
3470 	rte_write32(0, (uint8_t *)bp->bar0 +
3471 			 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 16);
3472 	rte_write32(0, (uint8_t *)bp->bar0 +
3473 			 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 20);
3474 }
3475 
3476 static uint64_t bnxt_cc_read(struct bnxt *bp)
3477 {
3478 	uint64_t ns;
3479 
3480 	ns = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3481 			      BNXT_GRCPF_REG_SYNC_TIME));
3482 	ns |= (uint64_t)(rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3483 					  BNXT_GRCPF_REG_SYNC_TIME + 4))) << 32;
3484 	return ns;
3485 }
3486 
3487 static int bnxt_get_tx_ts(struct bnxt *bp, uint64_t *ts)
3488 {
3489 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3490 	uint32_t fifo;
3491 
3492 	fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3493 				ptp->tx_mapped_regs[BNXT_PTP_TX_FIFO]));
3494 	if (fifo & BNXT_PTP_TX_FIFO_EMPTY)
3495 		return -EAGAIN;
3496 
3497 	fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3498 				ptp->tx_mapped_regs[BNXT_PTP_TX_FIFO]));
3499 	*ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3500 				ptp->tx_mapped_regs[BNXT_PTP_TX_TS_L]));
3501 	*ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3502 				ptp->tx_mapped_regs[BNXT_PTP_TX_TS_H])) << 32;
3503 	rte_read32((uint8_t *)bp->bar0 + ptp->tx_mapped_regs[BNXT_PTP_TX_SEQ]);
3504 
3505 	return 0;
3506 }
3507 
3508 static int bnxt_clr_rx_ts(struct bnxt *bp, uint64_t *last_ts)
3509 {
3510 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3511 	struct bnxt_pf_info *pf = bp->pf;
3512 	uint16_t port_id;
3513 	int i = 0;
3514 	uint32_t fifo;
3515 
3516 	if (!ptp || (bp->flags & BNXT_FLAG_CHIP_P5))
3517 		return -EINVAL;
3518 
3519 	port_id = pf->port_id;
3520 	fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3521 				ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3522 	while ((fifo & BNXT_PTP_RX_FIFO_PENDING) && (i < BNXT_PTP_RX_PND_CNT)) {
3523 		rte_write32(1 << port_id, (uint8_t *)bp->bar0 +
3524 			    ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO_ADV]);
3525 		fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3526 					ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3527 		*last_ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3528 					ptp->rx_mapped_regs[BNXT_PTP_RX_TS_L]));
3529 		*last_ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3530 					ptp->rx_mapped_regs[BNXT_PTP_RX_TS_H])) << 32;
3531 		i++;
3532 	}
3533 
3534 	if (i >= BNXT_PTP_RX_PND_CNT)
3535 		return -EBUSY;
3536 
3537 	return 0;
3538 }
3539 
3540 static int bnxt_get_rx_ts(struct bnxt *bp, uint64_t *ts)
3541 {
3542 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3543 	struct bnxt_pf_info *pf = bp->pf;
3544 	uint16_t port_id;
3545 	uint32_t fifo;
3546 
3547 	fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3548 				ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3549 	if (!(fifo & BNXT_PTP_RX_FIFO_PENDING))
3550 		return -EAGAIN;
3551 
3552 	port_id = pf->port_id;
3553 	rte_write32(1 << port_id, (uint8_t *)bp->bar0 +
3554 	       ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO_ADV]);
3555 
3556 	fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3557 				   ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3558 	if (fifo & BNXT_PTP_RX_FIFO_PENDING)
3559 		return bnxt_clr_rx_ts(bp, ts);
3560 
3561 	*ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3562 				ptp->rx_mapped_regs[BNXT_PTP_RX_TS_L]));
3563 	*ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3564 				ptp->rx_mapped_regs[BNXT_PTP_RX_TS_H])) << 32;
3565 
3566 	return 0;
3567 }
3568 
3569 static int
3570 bnxt_timesync_write_time(struct rte_eth_dev *dev, const struct timespec *ts)
3571 {
3572 	uint64_t ns;
3573 	struct bnxt *bp = dev->data->dev_private;
3574 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3575 
3576 	if (!ptp)
3577 		return -ENOTSUP;
3578 
3579 	ns = rte_timespec_to_ns(ts);
3580 	/* Set the timecounters to a new value. */
3581 	ptp->tc.nsec = ns;
3582 	ptp->tx_tstamp_tc.nsec = ns;
3583 	ptp->rx_tstamp_tc.nsec = ns;
3584 
3585 	return 0;
3586 }
3587 
3588 static int
3589 bnxt_timesync_read_time(struct rte_eth_dev *dev, struct timespec *ts)
3590 {
3591 	struct bnxt *bp = dev->data->dev_private;
3592 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3593 	uint64_t ns, systime_cycles = 0;
3594 	int rc = 0;
3595 
3596 	if (!ptp)
3597 		return -ENOTSUP;
3598 
3599 	if (BNXT_CHIP_P5(bp))
3600 		rc = bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
3601 					     &systime_cycles);
3602 	else
3603 		systime_cycles = bnxt_cc_read(bp);
3604 
3605 	ns = rte_timecounter_update(&ptp->tc, systime_cycles);
3606 	*ts = rte_ns_to_timespec(ns);
3607 
3608 	return rc;
3609 }
3610 static int
3611 bnxt_timesync_enable(struct rte_eth_dev *dev)
3612 {
3613 	struct bnxt *bp = dev->data->dev_private;
3614 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3615 	uint32_t shift = 0;
3616 	int rc;
3617 
3618 	if (!ptp)
3619 		return -ENOTSUP;
3620 
3621 	ptp->rx_filter = 1;
3622 	ptp->tx_tstamp_en = 1;
3623 	ptp->rxctl = BNXT_PTP_MSG_EVENTS;
3624 
3625 	rc = bnxt_hwrm_ptp_cfg(bp);
3626 	if (rc)
3627 		return rc;
3628 
3629 	memset(&ptp->tc, 0, sizeof(struct rte_timecounter));
3630 	memset(&ptp->rx_tstamp_tc, 0, sizeof(struct rte_timecounter));
3631 	memset(&ptp->tx_tstamp_tc, 0, sizeof(struct rte_timecounter));
3632 
3633 	ptp->tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3634 	ptp->tc.cc_shift = shift;
3635 	ptp->tc.nsec_mask = (1ULL << shift) - 1;
3636 
3637 	ptp->rx_tstamp_tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3638 	ptp->rx_tstamp_tc.cc_shift = shift;
3639 	ptp->rx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
3640 
3641 	ptp->tx_tstamp_tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3642 	ptp->tx_tstamp_tc.cc_shift = shift;
3643 	ptp->tx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
3644 
3645 	if (!BNXT_CHIP_P5(bp))
3646 		bnxt_map_ptp_regs(bp);
3647 	else
3648 		rc = bnxt_ptp_start(bp);
3649 
3650 	return rc;
3651 }
3652 
3653 static int
3654 bnxt_timesync_disable(struct rte_eth_dev *dev)
3655 {
3656 	struct bnxt *bp = dev->data->dev_private;
3657 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3658 
3659 	if (!ptp)
3660 		return -ENOTSUP;
3661 
3662 	ptp->rx_filter = 0;
3663 	ptp->tx_tstamp_en = 0;
3664 	ptp->rxctl = 0;
3665 
3666 	bnxt_hwrm_ptp_cfg(bp);
3667 
3668 	if (!BNXT_CHIP_P5(bp))
3669 		bnxt_unmap_ptp_regs(bp);
3670 	else
3671 		bnxt_ptp_stop(bp);
3672 
3673 	return 0;
3674 }
3675 
3676 static int
3677 bnxt_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
3678 				 struct timespec *timestamp,
3679 				 uint32_t flags __rte_unused)
3680 {
3681 	struct bnxt *bp = dev->data->dev_private;
3682 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3683 	uint64_t rx_tstamp_cycles = 0;
3684 	uint64_t ns;
3685 
3686 	if (!ptp)
3687 		return -ENOTSUP;
3688 
3689 	if (BNXT_CHIP_P5(bp))
3690 		rx_tstamp_cycles = ptp->rx_timestamp;
3691 	else
3692 		bnxt_get_rx_ts(bp, &rx_tstamp_cycles);
3693 
3694 	ns = rte_timecounter_update(&ptp->rx_tstamp_tc, rx_tstamp_cycles);
3695 	*timestamp = rte_ns_to_timespec(ns);
3696 	return  0;
3697 }
3698 
3699 static int
3700 bnxt_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
3701 				 struct timespec *timestamp)
3702 {
3703 	struct bnxt *bp = dev->data->dev_private;
3704 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3705 	uint64_t tx_tstamp_cycles = 0;
3706 	uint64_t ns;
3707 	int rc = 0;
3708 
3709 	if (!ptp)
3710 		return -ENOTSUP;
3711 
3712 	if (BNXT_CHIP_P5(bp))
3713 		rc = bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_PATH_TX,
3714 					     &tx_tstamp_cycles);
3715 	else
3716 		rc = bnxt_get_tx_ts(bp, &tx_tstamp_cycles);
3717 
3718 	ns = rte_timecounter_update(&ptp->tx_tstamp_tc, tx_tstamp_cycles);
3719 	*timestamp = rte_ns_to_timespec(ns);
3720 
3721 	return rc;
3722 }
3723 
3724 static int
3725 bnxt_timesync_adjust_time(struct rte_eth_dev *dev, int64_t delta)
3726 {
3727 	struct bnxt *bp = dev->data->dev_private;
3728 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3729 
3730 	if (!ptp)
3731 		return -ENOTSUP;
3732 
3733 	ptp->tc.nsec += delta;
3734 	ptp->tx_tstamp_tc.nsec += delta;
3735 	ptp->rx_tstamp_tc.nsec += delta;
3736 
3737 	return 0;
3738 }
3739 
3740 static int
3741 bnxt_get_eeprom_length_op(struct rte_eth_dev *dev)
3742 {
3743 	struct bnxt *bp = dev->data->dev_private;
3744 	int rc;
3745 	uint32_t dir_entries;
3746 	uint32_t entry_length;
3747 
3748 	rc = is_bnxt_in_error(bp);
3749 	if (rc)
3750 		return rc;
3751 
3752 	PMD_DRV_LOG(INFO, PCI_PRI_FMT "\n",
3753 		    bp->pdev->addr.domain, bp->pdev->addr.bus,
3754 		    bp->pdev->addr.devid, bp->pdev->addr.function);
3755 
3756 	rc = bnxt_hwrm_nvm_get_dir_info(bp, &dir_entries, &entry_length);
3757 	if (rc != 0)
3758 		return rc;
3759 
3760 	return dir_entries * entry_length;
3761 }
3762 
3763 static int
3764 bnxt_get_eeprom_op(struct rte_eth_dev *dev,
3765 		struct rte_dev_eeprom_info *in_eeprom)
3766 {
3767 	struct bnxt *bp = dev->data->dev_private;
3768 	uint32_t index;
3769 	uint32_t offset;
3770 	int rc;
3771 
3772 	rc = is_bnxt_in_error(bp);
3773 	if (rc)
3774 		return rc;
3775 
3776 	PMD_DRV_LOG(INFO, PCI_PRI_FMT " in_eeprom->offset = %d len = %d\n",
3777 		    bp->pdev->addr.domain, bp->pdev->addr.bus,
3778 		    bp->pdev->addr.devid, bp->pdev->addr.function,
3779 		    in_eeprom->offset, in_eeprom->length);
3780 
3781 	if (in_eeprom->offset == 0) /* special offset value to get directory */
3782 		return bnxt_get_nvram_directory(bp, in_eeprom->length,
3783 						in_eeprom->data);
3784 
3785 	index = in_eeprom->offset >> 24;
3786 	offset = in_eeprom->offset & 0xffffff;
3787 
3788 	if (index != 0)
3789 		return bnxt_hwrm_get_nvram_item(bp, index - 1, offset,
3790 					   in_eeprom->length, in_eeprom->data);
3791 
3792 	return 0;
3793 }
3794 
3795 static bool bnxt_dir_type_is_ape_bin_format(uint16_t dir_type)
3796 {
3797 	switch (dir_type) {
3798 	case BNX_DIR_TYPE_CHIMP_PATCH:
3799 	case BNX_DIR_TYPE_BOOTCODE:
3800 	case BNX_DIR_TYPE_BOOTCODE_2:
3801 	case BNX_DIR_TYPE_APE_FW:
3802 	case BNX_DIR_TYPE_APE_PATCH:
3803 	case BNX_DIR_TYPE_KONG_FW:
3804 	case BNX_DIR_TYPE_KONG_PATCH:
3805 	case BNX_DIR_TYPE_BONO_FW:
3806 	case BNX_DIR_TYPE_BONO_PATCH:
3807 		/* FALLTHROUGH */
3808 		return true;
3809 	}
3810 
3811 	return false;
3812 }
3813 
3814 static bool bnxt_dir_type_is_other_exec_format(uint16_t dir_type)
3815 {
3816 	switch (dir_type) {
3817 	case BNX_DIR_TYPE_AVS:
3818 	case BNX_DIR_TYPE_EXP_ROM_MBA:
3819 	case BNX_DIR_TYPE_PCIE:
3820 	case BNX_DIR_TYPE_TSCF_UCODE:
3821 	case BNX_DIR_TYPE_EXT_PHY:
3822 	case BNX_DIR_TYPE_CCM:
3823 	case BNX_DIR_TYPE_ISCSI_BOOT:
3824 	case BNX_DIR_TYPE_ISCSI_BOOT_IPV6:
3825 	case BNX_DIR_TYPE_ISCSI_BOOT_IPV4N6:
3826 		/* FALLTHROUGH */
3827 		return true;
3828 	}
3829 
3830 	return false;
3831 }
3832 
3833 static bool bnxt_dir_type_is_executable(uint16_t dir_type)
3834 {
3835 	return bnxt_dir_type_is_ape_bin_format(dir_type) ||
3836 		bnxt_dir_type_is_other_exec_format(dir_type);
3837 }
3838 
3839 static int
3840 bnxt_set_eeprom_op(struct rte_eth_dev *dev,
3841 		struct rte_dev_eeprom_info *in_eeprom)
3842 {
3843 	struct bnxt *bp = dev->data->dev_private;
3844 	uint8_t index, dir_op;
3845 	uint16_t type, ext, ordinal, attr;
3846 	int rc;
3847 
3848 	rc = is_bnxt_in_error(bp);
3849 	if (rc)
3850 		return rc;
3851 
3852 	PMD_DRV_LOG(INFO, PCI_PRI_FMT " in_eeprom->offset = %d len = %d\n",
3853 		    bp->pdev->addr.domain, bp->pdev->addr.bus,
3854 		    bp->pdev->addr.devid, bp->pdev->addr.function,
3855 		    in_eeprom->offset, in_eeprom->length);
3856 
3857 	if (!BNXT_PF(bp)) {
3858 		PMD_DRV_LOG(ERR, "NVM write not supported from a VF\n");
3859 		return -EINVAL;
3860 	}
3861 
3862 	type = in_eeprom->magic >> 16;
3863 
3864 	if (type == 0xffff) { /* special value for directory operations */
3865 		index = in_eeprom->magic & 0xff;
3866 		dir_op = in_eeprom->magic >> 8;
3867 		if (index == 0)
3868 			return -EINVAL;
3869 		switch (dir_op) {
3870 		case 0x0e: /* erase */
3871 			if (in_eeprom->offset != ~in_eeprom->magic)
3872 				return -EINVAL;
3873 			return bnxt_hwrm_erase_nvram_directory(bp, index - 1);
3874 		default:
3875 			return -EINVAL;
3876 		}
3877 	}
3878 
3879 	/* Create or re-write an NVM item: */
3880 	if (bnxt_dir_type_is_executable(type) == true)
3881 		return -EOPNOTSUPP;
3882 	ext = in_eeprom->magic & 0xffff;
3883 	ordinal = in_eeprom->offset >> 16;
3884 	attr = in_eeprom->offset & 0xffff;
3885 
3886 	return bnxt_hwrm_flash_nvram(bp, type, ordinal, ext, attr,
3887 				     in_eeprom->data, in_eeprom->length);
3888 }
3889 
3890 static int bnxt_get_module_info(struct rte_eth_dev *dev,
3891 				struct rte_eth_dev_module_info *modinfo)
3892 {
3893 	uint8_t module_info[SFF_DIAG_SUPPORT_OFFSET + 1];
3894 	struct bnxt *bp = dev->data->dev_private;
3895 	int rc;
3896 
3897 	/* No point in going further if phy status indicates
3898 	 * module is not inserted or if it is powered down or
3899 	 * if it is of type 10GBase-T
3900 	 */
3901 	if (bp->link_info->module_status >
3902 	    HWRM_PORT_PHY_QCFG_OUTPUT_MODULE_STATUS_WARNINGMSG) {
3903 		PMD_DRV_LOG(NOTICE, "Port %u : Module is not inserted or is powered down\n",
3904 			    dev->data->port_id);
3905 		return -ENOTSUP;
3906 	}
3907 
3908 	/* This feature is not supported in older firmware versions */
3909 	if (bp->hwrm_spec_code < 0x10202) {
3910 		PMD_DRV_LOG(NOTICE, "Port %u : Feature is not supported in older firmware\n",
3911 			    dev->data->port_id);
3912 		return -ENOTSUP;
3913 	}
3914 
3915 	rc = bnxt_hwrm_read_sfp_module_eeprom_info(bp, I2C_DEV_ADDR_A0, 0, 0,
3916 						   SFF_DIAG_SUPPORT_OFFSET + 1,
3917 						   module_info);
3918 
3919 	if (rc)
3920 		return rc;
3921 
3922 	switch (module_info[0]) {
3923 	case SFF_MODULE_ID_SFP:
3924 		modinfo->type = RTE_ETH_MODULE_SFF_8472;
3925 		modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8472_LEN;
3926 		if (module_info[SFF_DIAG_SUPPORT_OFFSET] == 0)
3927 			modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8436_LEN;
3928 		break;
3929 	case SFF_MODULE_ID_QSFP:
3930 	case SFF_MODULE_ID_QSFP_PLUS:
3931 		modinfo->type = RTE_ETH_MODULE_SFF_8436;
3932 		modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8436_LEN;
3933 		break;
3934 	case SFF_MODULE_ID_QSFP28:
3935 		modinfo->type = RTE_ETH_MODULE_SFF_8636;
3936 		modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8636_MAX_LEN;
3937 		if (module_info[SFF8636_FLATMEM_OFFSET] & SFF8636_FLATMEM_MASK)
3938 			modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8636_LEN;
3939 		break;
3940 	default:
3941 		PMD_DRV_LOG(NOTICE, "Port %u : Unsupported module\n", dev->data->port_id);
3942 		return -ENOTSUP;
3943 	}
3944 
3945 	PMD_DRV_LOG(INFO, "Port %u : modinfo->type = %d modinfo->eeprom_len = %d\n",
3946 		    dev->data->port_id, modinfo->type, modinfo->eeprom_len);
3947 
3948 	return 0;
3949 }
3950 
3951 static int bnxt_get_module_eeprom(struct rte_eth_dev *dev,
3952 				  struct rte_dev_eeprom_info *info)
3953 {
3954 	uint8_t pg_addr[5] = { I2C_DEV_ADDR_A0, I2C_DEV_ADDR_A0 };
3955 	uint32_t offset = info->offset, length = info->length;
3956 	uint8_t module_info[SFF_DIAG_SUPPORT_OFFSET + 1];
3957 	struct bnxt *bp = dev->data->dev_private;
3958 	uint8_t *data = info->data;
3959 	uint8_t page = offset >> 7;
3960 	uint8_t max_pages = 2;
3961 	uint8_t opt_pages;
3962 	int rc;
3963 
3964 	rc = bnxt_hwrm_read_sfp_module_eeprom_info(bp, I2C_DEV_ADDR_A0, 0, 0,
3965 						   SFF_DIAG_SUPPORT_OFFSET + 1,
3966 						   module_info);
3967 	if (rc)
3968 		return rc;
3969 
3970 	switch (module_info[0]) {
3971 	case SFF_MODULE_ID_SFP:
3972 		module_info[SFF_DIAG_SUPPORT_OFFSET] = 0;
3973 		if (module_info[SFF_DIAG_SUPPORT_OFFSET]) {
3974 			pg_addr[2] = I2C_DEV_ADDR_A2;
3975 			pg_addr[3] = I2C_DEV_ADDR_A2;
3976 			max_pages = 4;
3977 		}
3978 		break;
3979 	case SFF_MODULE_ID_QSFP28:
3980 		rc = bnxt_hwrm_read_sfp_module_eeprom_info(bp, I2C_DEV_ADDR_A0, 0,
3981 							   SFF8636_OPT_PAGES_OFFSET,
3982 							   1, &opt_pages);
3983 		if (rc)
3984 			return rc;
3985 
3986 		if (opt_pages & SFF8636_PAGE1_MASK) {
3987 			pg_addr[2] = I2C_DEV_ADDR_A0;
3988 			max_pages = 3;
3989 		}
3990 		if (opt_pages & SFF8636_PAGE2_MASK) {
3991 			pg_addr[3] = I2C_DEV_ADDR_A0;
3992 			max_pages = 4;
3993 		}
3994 		if (~module_info[SFF8636_FLATMEM_OFFSET] & SFF8636_FLATMEM_MASK) {
3995 			pg_addr[4] = I2C_DEV_ADDR_A0;
3996 			max_pages = 5;
3997 		}
3998 		break;
3999 	default:
4000 		break;
4001 	}
4002 
4003 	memset(data, 0, length);
4004 
4005 	offset &= 0xff;
4006 	while (length && page < max_pages) {
4007 		uint8_t raw_page = page ? page - 1 : 0;
4008 		uint16_t chunk;
4009 
4010 		if (pg_addr[page] == I2C_DEV_ADDR_A2)
4011 			raw_page = 0;
4012 		else if (page)
4013 			offset |= 0x80;
4014 		chunk = RTE_MIN(length, 256 - offset);
4015 
4016 		if (pg_addr[page]) {
4017 			rc = bnxt_hwrm_read_sfp_module_eeprom_info(bp, pg_addr[page],
4018 								   raw_page, offset,
4019 								   chunk, data);
4020 			if (rc)
4021 				return rc;
4022 		}
4023 
4024 		data += chunk;
4025 		length -= chunk;
4026 		offset = 0;
4027 		page += 1 + (chunk > 128);
4028 	}
4029 
4030 	return length ? -EINVAL : 0;
4031 }
4032 
4033 /*
4034  * Initialization
4035  */
4036 
4037 static const struct eth_dev_ops bnxt_dev_ops = {
4038 	.dev_infos_get = bnxt_dev_info_get_op,
4039 	.dev_close = bnxt_dev_close_op,
4040 	.dev_configure = bnxt_dev_configure_op,
4041 	.dev_start = bnxt_dev_start_op,
4042 	.dev_stop = bnxt_dev_stop_op,
4043 	.dev_set_link_up = bnxt_dev_set_link_up_op,
4044 	.dev_set_link_down = bnxt_dev_set_link_down_op,
4045 	.stats_get = bnxt_stats_get_op,
4046 	.stats_reset = bnxt_stats_reset_op,
4047 	.rx_queue_setup = bnxt_rx_queue_setup_op,
4048 	.rx_queue_release = bnxt_rx_queue_release_op,
4049 	.tx_queue_setup = bnxt_tx_queue_setup_op,
4050 	.tx_queue_release = bnxt_tx_queue_release_op,
4051 	.rx_queue_intr_enable = bnxt_rx_queue_intr_enable_op,
4052 	.rx_queue_intr_disable = bnxt_rx_queue_intr_disable_op,
4053 	.reta_update = bnxt_reta_update_op,
4054 	.reta_query = bnxt_reta_query_op,
4055 	.rss_hash_update = bnxt_rss_hash_update_op,
4056 	.rss_hash_conf_get = bnxt_rss_hash_conf_get_op,
4057 	.link_update = bnxt_link_update_op,
4058 	.promiscuous_enable = bnxt_promiscuous_enable_op,
4059 	.promiscuous_disable = bnxt_promiscuous_disable_op,
4060 	.allmulticast_enable = bnxt_allmulticast_enable_op,
4061 	.allmulticast_disable = bnxt_allmulticast_disable_op,
4062 	.mac_addr_add = bnxt_mac_addr_add_op,
4063 	.mac_addr_remove = bnxt_mac_addr_remove_op,
4064 	.flow_ctrl_get = bnxt_flow_ctrl_get_op,
4065 	.flow_ctrl_set = bnxt_flow_ctrl_set_op,
4066 	.udp_tunnel_port_add  = bnxt_udp_tunnel_port_add_op,
4067 	.udp_tunnel_port_del  = bnxt_udp_tunnel_port_del_op,
4068 	.vlan_filter_set = bnxt_vlan_filter_set_op,
4069 	.vlan_offload_set = bnxt_vlan_offload_set_op,
4070 	.vlan_tpid_set = bnxt_vlan_tpid_set_op,
4071 	.vlan_pvid_set = bnxt_vlan_pvid_set_op,
4072 	.mtu_set = bnxt_mtu_set_op,
4073 	.mac_addr_set = bnxt_set_default_mac_addr_op,
4074 	.xstats_get = bnxt_dev_xstats_get_op,
4075 	.xstats_get_names = bnxt_dev_xstats_get_names_op,
4076 	.xstats_reset = bnxt_dev_xstats_reset_op,
4077 	.fw_version_get = bnxt_fw_version_get,
4078 	.set_mc_addr_list = bnxt_dev_set_mc_addr_list_op,
4079 	.rxq_info_get = bnxt_rxq_info_get_op,
4080 	.txq_info_get = bnxt_txq_info_get_op,
4081 	.rx_burst_mode_get = bnxt_rx_burst_mode_get,
4082 	.tx_burst_mode_get = bnxt_tx_burst_mode_get,
4083 	.dev_led_on = bnxt_dev_led_on_op,
4084 	.dev_led_off = bnxt_dev_led_off_op,
4085 	.rx_queue_start = bnxt_rx_queue_start,
4086 	.rx_queue_stop = bnxt_rx_queue_stop,
4087 	.tx_queue_start = bnxt_tx_queue_start,
4088 	.tx_queue_stop = bnxt_tx_queue_stop,
4089 	.flow_ops_get = bnxt_flow_ops_get_op,
4090 	.dev_supported_ptypes_get = bnxt_dev_supported_ptypes_get_op,
4091 	.get_eeprom_length    = bnxt_get_eeprom_length_op,
4092 	.get_eeprom           = bnxt_get_eeprom_op,
4093 	.set_eeprom           = bnxt_set_eeprom_op,
4094 	.get_module_info = bnxt_get_module_info,
4095 	.get_module_eeprom = bnxt_get_module_eeprom,
4096 	.timesync_enable      = bnxt_timesync_enable,
4097 	.timesync_disable     = bnxt_timesync_disable,
4098 	.timesync_read_time   = bnxt_timesync_read_time,
4099 	.timesync_write_time   = bnxt_timesync_write_time,
4100 	.timesync_adjust_time = bnxt_timesync_adjust_time,
4101 	.timesync_read_rx_timestamp = bnxt_timesync_read_rx_timestamp,
4102 	.timesync_read_tx_timestamp = bnxt_timesync_read_tx_timestamp,
4103 };
4104 
4105 static uint32_t bnxt_map_reset_regs(struct bnxt *bp, uint32_t reg)
4106 {
4107 	uint32_t offset;
4108 
4109 	/* Only pre-map the reset GRC registers using window 3 */
4110 	rte_write32(reg & 0xfffff000, (uint8_t *)bp->bar0 +
4111 		    BNXT_GRCPF_REG_WINDOW_BASE_OUT + 8);
4112 
4113 	offset = BNXT_GRCP_WINDOW_3_BASE + (reg & 0xffc);
4114 
4115 	return offset;
4116 }
4117 
4118 int bnxt_map_fw_health_status_regs(struct bnxt *bp)
4119 {
4120 	struct bnxt_error_recovery_info *info = bp->recovery_info;
4121 	uint32_t reg_base = 0xffffffff;
4122 	int i;
4123 
4124 	/* Only pre-map the monitoring GRC registers using window 2 */
4125 	for (i = 0; i < BNXT_FW_STATUS_REG_CNT; i++) {
4126 		uint32_t reg = info->status_regs[i];
4127 
4128 		if (BNXT_FW_STATUS_REG_TYPE(reg) != BNXT_FW_STATUS_REG_TYPE_GRC)
4129 			continue;
4130 
4131 		if (reg_base == 0xffffffff)
4132 			reg_base = reg & 0xfffff000;
4133 		if ((reg & 0xfffff000) != reg_base)
4134 			return -ERANGE;
4135 
4136 		/* Use mask 0xffc as the Lower 2 bits indicates
4137 		 * address space location
4138 		 */
4139 		info->mapped_status_regs[i] = BNXT_GRCP_WINDOW_2_BASE +
4140 						(reg & 0xffc);
4141 	}
4142 
4143 	if (reg_base == 0xffffffff)
4144 		return 0;
4145 
4146 	rte_write32(reg_base, (uint8_t *)bp->bar0 +
4147 		    BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
4148 
4149 	return 0;
4150 }
4151 
4152 static void bnxt_write_fw_reset_reg(struct bnxt *bp, uint32_t index)
4153 {
4154 	struct bnxt_error_recovery_info *info = bp->recovery_info;
4155 	uint32_t delay = info->delay_after_reset[index];
4156 	uint32_t val = info->reset_reg_val[index];
4157 	uint32_t reg = info->reset_reg[index];
4158 	uint32_t type, offset;
4159 	int ret;
4160 
4161 	type = BNXT_FW_STATUS_REG_TYPE(reg);
4162 	offset = BNXT_FW_STATUS_REG_OFF(reg);
4163 
4164 	switch (type) {
4165 	case BNXT_FW_STATUS_REG_TYPE_CFG:
4166 		ret = rte_pci_write_config(bp->pdev, &val, sizeof(val), offset);
4167 		if (ret < 0) {
4168 			PMD_DRV_LOG(ERR, "Failed to write %#x at PCI offset %#x",
4169 				    val, offset);
4170 			return;
4171 		}
4172 		break;
4173 	case BNXT_FW_STATUS_REG_TYPE_GRC:
4174 		offset = bnxt_map_reset_regs(bp, offset);
4175 		rte_write32(val, (uint8_t *)bp->bar0 + offset);
4176 		break;
4177 	case BNXT_FW_STATUS_REG_TYPE_BAR0:
4178 		rte_write32(val, (uint8_t *)bp->bar0 + offset);
4179 		break;
4180 	}
4181 	/* wait on a specific interval of time until core reset is complete */
4182 	if (delay)
4183 		rte_delay_ms(delay);
4184 }
4185 
4186 static void bnxt_dev_cleanup(struct bnxt *bp)
4187 {
4188 	bp->eth_dev->data->dev_link.link_status = 0;
4189 	bp->link_info->link_up = 0;
4190 	if (bp->eth_dev->data->dev_started)
4191 		bnxt_dev_stop(bp->eth_dev);
4192 
4193 	bnxt_uninit_resources(bp, true);
4194 }
4195 
4196 static int
4197 bnxt_check_fw_reset_done(struct bnxt *bp)
4198 {
4199 	int timeout = bp->fw_reset_max_msecs;
4200 	uint16_t val = 0;
4201 	int rc;
4202 
4203 	do {
4204 		rc = rte_pci_read_config(bp->pdev, &val, sizeof(val), PCI_SUBSYSTEM_ID_OFFSET);
4205 		if (rc < 0) {
4206 			PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x", PCI_SUBSYSTEM_ID_OFFSET);
4207 			return rc;
4208 		}
4209 		if (val != 0xffff)
4210 			break;
4211 		rte_delay_ms(1);
4212 	} while (timeout--);
4213 
4214 	if (val == 0xffff) {
4215 		PMD_DRV_LOG(ERR, "Firmware reset aborted, PCI config space invalid\n");
4216 		return -1;
4217 	}
4218 
4219 	return 0;
4220 }
4221 
4222 static int bnxt_restore_vlan_filters(struct bnxt *bp)
4223 {
4224 	struct rte_eth_dev *dev = bp->eth_dev;
4225 	struct rte_vlan_filter_conf *vfc;
4226 	int vidx, vbit, rc;
4227 	uint16_t vlan_id;
4228 
4229 	for (vlan_id = 1; vlan_id <= RTE_ETHER_MAX_VLAN_ID; vlan_id++) {
4230 		vfc = &dev->data->vlan_filter_conf;
4231 		vidx = vlan_id / 64;
4232 		vbit = vlan_id % 64;
4233 
4234 		/* Each bit corresponds to a VLAN id */
4235 		if (vfc->ids[vidx] & (UINT64_C(1) << vbit)) {
4236 			rc = bnxt_add_vlan_filter(bp, vlan_id);
4237 			if (rc)
4238 				return rc;
4239 		}
4240 	}
4241 
4242 	return 0;
4243 }
4244 
4245 static int bnxt_restore_mac_filters(struct bnxt *bp)
4246 {
4247 	struct rte_eth_dev *dev = bp->eth_dev;
4248 	struct rte_eth_dev_info dev_info;
4249 	struct rte_ether_addr *addr;
4250 	uint64_t pool_mask;
4251 	uint32_t pool = 0;
4252 	uint32_t i;
4253 	int rc;
4254 
4255 	if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp))
4256 		return 0;
4257 
4258 	rc = bnxt_dev_info_get_op(dev, &dev_info);
4259 	if (rc)
4260 		return rc;
4261 
4262 	/* replay MAC address configuration */
4263 	for (i = 1; i < dev_info.max_mac_addrs; i++) {
4264 		addr = &dev->data->mac_addrs[i];
4265 
4266 		/* skip zero address */
4267 		if (rte_is_zero_ether_addr(addr))
4268 			continue;
4269 
4270 		pool = 0;
4271 		pool_mask = dev->data->mac_pool_sel[i];
4272 
4273 		do {
4274 			if (pool_mask & 1ULL) {
4275 				rc = bnxt_mac_addr_add_op(dev, addr, i, pool);
4276 				if (rc)
4277 					return rc;
4278 			}
4279 			pool_mask >>= 1;
4280 			pool++;
4281 		} while (pool_mask);
4282 	}
4283 
4284 	return 0;
4285 }
4286 
4287 static int bnxt_restore_mcast_mac_filters(struct bnxt *bp)
4288 {
4289 	int ret = 0;
4290 
4291 	ret = bnxt_dev_set_mc_addr_list_op(bp->eth_dev, bp->mcast_addr_list,
4292 					   bp->nb_mc_addr);
4293 	if (ret)
4294 		PMD_DRV_LOG(ERR, "Failed to restore multicast MAC addreeses\n");
4295 
4296 	return ret;
4297 }
4298 
4299 static int bnxt_restore_filters(struct bnxt *bp)
4300 {
4301 	struct rte_eth_dev *dev = bp->eth_dev;
4302 	int ret = 0;
4303 
4304 	if (dev->data->all_multicast) {
4305 		ret = bnxt_allmulticast_enable_op(dev);
4306 		if (ret)
4307 			return ret;
4308 	}
4309 	if (dev->data->promiscuous) {
4310 		ret = bnxt_promiscuous_enable_op(dev);
4311 		if (ret)
4312 			return ret;
4313 	}
4314 
4315 	ret = bnxt_restore_mac_filters(bp);
4316 	if (ret)
4317 		return ret;
4318 
4319 	/* if vlans are already programmed, this can fail with -EEXIST */
4320 	ret = bnxt_restore_vlan_filters(bp);
4321 	if (ret && ret != -EEXIST)
4322 		return ret;
4323 
4324 	ret = bnxt_restore_mcast_mac_filters(bp);
4325 	if (ret)
4326 		return ret;
4327 
4328 	return ret;
4329 }
4330 
4331 static int bnxt_check_fw_ready(struct bnxt *bp)
4332 {
4333 	int timeout = bp->fw_reset_max_msecs ? : BNXT_MAX_FW_RESET_TIMEOUT;
4334 	int rc = 0;
4335 
4336 	do {
4337 		rc = bnxt_hwrm_poll_ver_get(bp);
4338 		if (rc == 0)
4339 			break;
4340 		rte_delay_ms(BNXT_FW_READY_WAIT_INTERVAL);
4341 		timeout -= BNXT_FW_READY_WAIT_INTERVAL;
4342 	} while (rc && timeout > 0);
4343 
4344 	if (rc)
4345 		PMD_DRV_LOG(ERR, "FW is not Ready after reset\n");
4346 
4347 	return rc;
4348 }
4349 
4350 static void bnxt_dev_recover(void *arg)
4351 {
4352 	struct bnxt *bp = arg;
4353 	int rc = 0;
4354 
4355 	pthread_mutex_lock(&bp->err_recovery_lock);
4356 
4357 	if (!bp->fw_reset_min_msecs) {
4358 		rc = bnxt_check_fw_reset_done(bp);
4359 		if (rc)
4360 			goto err;
4361 	}
4362 
4363 	/* Clear Error flag so that device re-init should happen */
4364 	bp->flags &= ~BNXT_FLAG_FATAL_ERROR;
4365 	PMD_DRV_LOG(INFO, "Port: %u Starting recovery...\n",
4366 		    bp->eth_dev->data->port_id);
4367 
4368 	rc = bnxt_check_fw_ready(bp);
4369 	if (rc)
4370 		goto err;
4371 
4372 	rc = bnxt_init_resources(bp, true);
4373 	if (rc) {
4374 		PMD_DRV_LOG(ERR,
4375 			    "Failed to initialize resources after reset\n");
4376 		goto err;
4377 	}
4378 	/* clear reset flag as the device is initialized now */
4379 	bp->flags &= ~BNXT_FLAG_FW_RESET;
4380 
4381 	rc = bnxt_dev_start_op(bp->eth_dev);
4382 	if (rc) {
4383 		PMD_DRV_LOG(ERR, "Failed to start port after reset\n");
4384 		goto err_start;
4385 	}
4386 
4387 	rc = bnxt_restore_filters(bp);
4388 	if (rc)
4389 		goto err_start;
4390 
4391 	rte_eth_fp_ops[bp->eth_dev->data->port_id].rx_pkt_burst =
4392 		bp->eth_dev->rx_pkt_burst;
4393 	rte_eth_fp_ops[bp->eth_dev->data->port_id].tx_pkt_burst =
4394 		bp->eth_dev->tx_pkt_burst;
4395 	rte_mb();
4396 
4397 	PMD_DRV_LOG(INFO, "Port: %u Recovered from FW reset\n",
4398 		    bp->eth_dev->data->port_id);
4399 	pthread_mutex_unlock(&bp->err_recovery_lock);
4400 
4401 	return;
4402 err_start:
4403 	bnxt_dev_stop(bp->eth_dev);
4404 err:
4405 	bp->flags |= BNXT_FLAG_FATAL_ERROR;
4406 	bnxt_uninit_resources(bp, false);
4407 	if (bp->eth_dev->data->dev_conf.intr_conf.rmv)
4408 		rte_eth_dev_callback_process(bp->eth_dev,
4409 					     RTE_ETH_EVENT_INTR_RMV,
4410 					     NULL);
4411 	pthread_mutex_unlock(&bp->err_recovery_lock);
4412 	PMD_DRV_LOG(ERR, "Port %u: Failed to recover from FW reset\n",
4413 		    bp->eth_dev->data->port_id);
4414 }
4415 
4416 void bnxt_dev_reset_and_resume(void *arg)
4417 {
4418 	struct bnxt *bp = arg;
4419 	uint32_t us = US_PER_MS * bp->fw_reset_min_msecs;
4420 	uint16_t val = 0;
4421 	int rc;
4422 
4423 	bnxt_dev_cleanup(bp);
4424 	PMD_DRV_LOG(INFO, "Port: %u Finished bnxt_dev_cleanup\n",
4425 		    bp->eth_dev->data->port_id);
4426 
4427 	bnxt_wait_for_device_shutdown(bp);
4428 
4429 	/* During some fatal firmware error conditions, the PCI config space
4430 	 * register 0x2e which normally contains the subsystem ID will become
4431 	 * 0xffff. This register will revert back to the normal value after
4432 	 * the chip has completed core reset. If we detect this condition,
4433 	 * we can poll this config register immediately for the value to revert.
4434 	 */
4435 	if (bp->flags & BNXT_FLAG_FATAL_ERROR) {
4436 		rc = rte_pci_read_config(bp->pdev, &val, sizeof(val), PCI_SUBSYSTEM_ID_OFFSET);
4437 		if (rc < 0) {
4438 			PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x", PCI_SUBSYSTEM_ID_OFFSET);
4439 			return;
4440 		}
4441 		if (val == 0xffff) {
4442 			bp->fw_reset_min_msecs = 0;
4443 			us = 1;
4444 		}
4445 	}
4446 
4447 	rc = rte_eal_alarm_set(us, bnxt_dev_recover, (void *)bp);
4448 	if (rc)
4449 		PMD_DRV_LOG(ERR, "Port %u: Error setting recovery alarm",
4450 			    bp->eth_dev->data->port_id);
4451 }
4452 
4453 uint32_t bnxt_read_fw_status_reg(struct bnxt *bp, uint32_t index)
4454 {
4455 	struct bnxt_error_recovery_info *info = bp->recovery_info;
4456 	uint32_t reg = info->status_regs[index];
4457 	uint32_t type, offset, val = 0;
4458 	int ret = 0;
4459 
4460 	type = BNXT_FW_STATUS_REG_TYPE(reg);
4461 	offset = BNXT_FW_STATUS_REG_OFF(reg);
4462 
4463 	switch (type) {
4464 	case BNXT_FW_STATUS_REG_TYPE_CFG:
4465 		ret = rte_pci_read_config(bp->pdev, &val, sizeof(val), offset);
4466 		if (ret < 0)
4467 			PMD_DRV_LOG(ERR, "Failed to read PCI offset %#x",
4468 				    offset);
4469 		break;
4470 	case BNXT_FW_STATUS_REG_TYPE_GRC:
4471 		offset = info->mapped_status_regs[index];
4472 		/* FALLTHROUGH */
4473 	case BNXT_FW_STATUS_REG_TYPE_BAR0:
4474 		val = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
4475 				       offset));
4476 		break;
4477 	}
4478 
4479 	return val;
4480 }
4481 
4482 static int bnxt_fw_reset_all(struct bnxt *bp)
4483 {
4484 	struct bnxt_error_recovery_info *info = bp->recovery_info;
4485 	uint32_t i;
4486 	int rc = 0;
4487 
4488 	if (info->flags & BNXT_FLAG_ERROR_RECOVERY_HOST) {
4489 		/* Reset through primary function driver */
4490 		for (i = 0; i < info->reg_array_cnt; i++)
4491 			bnxt_write_fw_reset_reg(bp, i);
4492 		/* Wait for time specified by FW after triggering reset */
4493 		rte_delay_ms(info->primary_func_wait_period_after_reset);
4494 	} else if (info->flags & BNXT_FLAG_ERROR_RECOVERY_CO_CPU) {
4495 		/* Reset with the help of Kong processor */
4496 		rc = bnxt_hwrm_fw_reset(bp);
4497 		if (rc)
4498 			PMD_DRV_LOG(ERR, "Failed to reset FW\n");
4499 	}
4500 
4501 	return rc;
4502 }
4503 
4504 static void bnxt_fw_reset_cb(void *arg)
4505 {
4506 	struct bnxt *bp = arg;
4507 	struct bnxt_error_recovery_info *info = bp->recovery_info;
4508 	int rc = 0;
4509 
4510 	/* Only Primary function can do FW reset */
4511 	if (bnxt_is_primary_func(bp) &&
4512 	    bnxt_is_recovery_enabled(bp)) {
4513 		rc = bnxt_fw_reset_all(bp);
4514 		if (rc) {
4515 			PMD_DRV_LOG(ERR, "Adapter recovery failed\n");
4516 			return;
4517 		}
4518 	}
4519 
4520 	/* if recovery method is ERROR_RECOVERY_CO_CPU, KONG will send
4521 	 * EXCEPTION_FATAL_ASYNC event to all the functions
4522 	 * (including MASTER FUNC). After receiving this Async, all the active
4523 	 * drivers should treat this case as FW initiated recovery
4524 	 */
4525 	if (info->flags & BNXT_FLAG_ERROR_RECOVERY_HOST) {
4526 		bp->fw_reset_min_msecs = BNXT_MIN_FW_READY_TIMEOUT;
4527 		bp->fw_reset_max_msecs = BNXT_MAX_FW_RESET_TIMEOUT;
4528 
4529 		/* To recover from error */
4530 		rte_eal_alarm_set(US_PER_MS, bnxt_dev_reset_and_resume,
4531 				  (void *)bp);
4532 	}
4533 }
4534 
4535 /* Driver should poll FW heartbeat, reset_counter with the frequency
4536  * advertised by FW in HWRM_ERROR_RECOVERY_QCFG.
4537  * When the driver detects heartbeat stop or change in reset_counter,
4538  * it has to trigger a reset to recover from the error condition.
4539  * A “primary function” is the function who will have the privilege to
4540  * initiate the chimp reset. The primary function will be elected by the
4541  * firmware and will be notified through async message.
4542  */
4543 static void bnxt_check_fw_health(void *arg)
4544 {
4545 	struct bnxt *bp = arg;
4546 	struct bnxt_error_recovery_info *info = bp->recovery_info;
4547 	uint32_t val = 0, wait_msec;
4548 
4549 	if (!info || !bnxt_is_recovery_enabled(bp) ||
4550 	    is_bnxt_in_error(bp))
4551 		return;
4552 
4553 	val = bnxt_read_fw_status_reg(bp, BNXT_FW_HEARTBEAT_CNT_REG);
4554 	if (val == info->last_heart_beat)
4555 		goto reset;
4556 
4557 	info->last_heart_beat = val;
4558 
4559 	val = bnxt_read_fw_status_reg(bp, BNXT_FW_RECOVERY_CNT_REG);
4560 	if (val != info->last_reset_counter)
4561 		goto reset;
4562 
4563 	info->last_reset_counter = val;
4564 
4565 	rte_eal_alarm_set(US_PER_MS * info->driver_polling_freq,
4566 			  bnxt_check_fw_health, (void *)bp);
4567 
4568 	return;
4569 reset:
4570 	/* Stop DMA to/from device */
4571 	bp->flags |= BNXT_FLAG_FATAL_ERROR;
4572 	bp->flags |= BNXT_FLAG_FW_RESET;
4573 
4574 	bnxt_stop_rxtx(bp->eth_dev);
4575 
4576 	PMD_DRV_LOG(ERR, "Detected FW dead condition\n");
4577 
4578 	if (bnxt_is_primary_func(bp))
4579 		wait_msec = info->primary_func_wait_period;
4580 	else
4581 		wait_msec = info->normal_func_wait_period;
4582 
4583 	rte_eal_alarm_set(US_PER_MS * wait_msec,
4584 			  bnxt_fw_reset_cb, (void *)bp);
4585 }
4586 
4587 void bnxt_schedule_fw_health_check(struct bnxt *bp)
4588 {
4589 	uint32_t polling_freq;
4590 
4591 	pthread_mutex_lock(&bp->health_check_lock);
4592 
4593 	if (!bnxt_is_recovery_enabled(bp))
4594 		goto done;
4595 
4596 	if (bp->flags & BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED)
4597 		goto done;
4598 
4599 	polling_freq = bp->recovery_info->driver_polling_freq;
4600 
4601 	rte_eal_alarm_set(US_PER_MS * polling_freq,
4602 			  bnxt_check_fw_health, (void *)bp);
4603 	bp->flags |= BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED;
4604 
4605 done:
4606 	pthread_mutex_unlock(&bp->health_check_lock);
4607 }
4608 
4609 static void bnxt_cancel_fw_health_check(struct bnxt *bp)
4610 {
4611 	rte_eal_alarm_cancel(bnxt_check_fw_health, (void *)bp);
4612 	bp->flags &= ~BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED;
4613 }
4614 
4615 static bool bnxt_vf_pciid(uint16_t device_id)
4616 {
4617 	switch (device_id) {
4618 	case BROADCOM_DEV_ID_57304_VF:
4619 	case BROADCOM_DEV_ID_57406_VF:
4620 	case BROADCOM_DEV_ID_5731X_VF:
4621 	case BROADCOM_DEV_ID_5741X_VF:
4622 	case BROADCOM_DEV_ID_57414_VF:
4623 	case BROADCOM_DEV_ID_STRATUS_NIC_VF1:
4624 	case BROADCOM_DEV_ID_STRATUS_NIC_VF2:
4625 	case BROADCOM_DEV_ID_58802_VF:
4626 	case BROADCOM_DEV_ID_57500_VF1:
4627 	case BROADCOM_DEV_ID_57500_VF2:
4628 	case BROADCOM_DEV_ID_58818_VF:
4629 		/* FALLTHROUGH */
4630 		return true;
4631 	default:
4632 		return false;
4633 	}
4634 }
4635 
4636 /* Phase 5 device */
4637 static bool bnxt_p5_device(uint16_t device_id)
4638 {
4639 	switch (device_id) {
4640 	case BROADCOM_DEV_ID_57508:
4641 	case BROADCOM_DEV_ID_57504:
4642 	case BROADCOM_DEV_ID_57502:
4643 	case BROADCOM_DEV_ID_57508_MF1:
4644 	case BROADCOM_DEV_ID_57504_MF1:
4645 	case BROADCOM_DEV_ID_57502_MF1:
4646 	case BROADCOM_DEV_ID_57508_MF2:
4647 	case BROADCOM_DEV_ID_57504_MF2:
4648 	case BROADCOM_DEV_ID_57502_MF2:
4649 	case BROADCOM_DEV_ID_57500_VF1:
4650 	case BROADCOM_DEV_ID_57500_VF2:
4651 	case BROADCOM_DEV_ID_58812:
4652 	case BROADCOM_DEV_ID_58814:
4653 	case BROADCOM_DEV_ID_58818:
4654 	case BROADCOM_DEV_ID_58818_VF:
4655 		/* FALLTHROUGH */
4656 		return true;
4657 	default:
4658 		return false;
4659 	}
4660 }
4661 
4662 bool bnxt_stratus_device(struct bnxt *bp)
4663 {
4664 	uint16_t device_id = bp->pdev->id.device_id;
4665 
4666 	switch (device_id) {
4667 	case BROADCOM_DEV_ID_STRATUS_NIC:
4668 	case BROADCOM_DEV_ID_STRATUS_NIC_VF1:
4669 	case BROADCOM_DEV_ID_STRATUS_NIC_VF2:
4670 		/* FALLTHROUGH */
4671 		return true;
4672 	default:
4673 		return false;
4674 	}
4675 }
4676 
4677 static int bnxt_map_pci_bars(struct rte_eth_dev *eth_dev)
4678 {
4679 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
4680 	struct bnxt *bp = eth_dev->data->dev_private;
4681 
4682 	/* enable device (incl. PCI PM wakeup), and bus-mastering */
4683 	bp->bar0 = (void *)pci_dev->mem_resource[0].addr;
4684 	bp->doorbell_base = (void *)pci_dev->mem_resource[2].addr;
4685 	if (!bp->bar0 || !bp->doorbell_base) {
4686 		PMD_DRV_LOG(ERR, "Unable to access Hardware\n");
4687 		return -ENODEV;
4688 	}
4689 
4690 	bp->eth_dev = eth_dev;
4691 	bp->pdev = pci_dev;
4692 
4693 	return 0;
4694 }
4695 
4696 static int bnxt_alloc_ctx_mem_blk(struct bnxt *bp,
4697 				  struct bnxt_ctx_pg_info *ctx_pg,
4698 				  uint32_t mem_size,
4699 				  const char *suffix,
4700 				  uint16_t idx)
4701 {
4702 	struct bnxt_ring_mem_info *rmem = &ctx_pg->ring_mem;
4703 	const struct rte_memzone *mz = NULL;
4704 	char mz_name[RTE_MEMZONE_NAMESIZE];
4705 	rte_iova_t mz_phys_addr;
4706 	uint64_t valid_bits = 0;
4707 	uint32_t sz;
4708 	int i;
4709 
4710 	if (!mem_size)
4711 		return 0;
4712 
4713 	rmem->nr_pages = RTE_ALIGN_MUL_CEIL(mem_size, BNXT_PAGE_SIZE) /
4714 			 BNXT_PAGE_SIZE;
4715 	rmem->page_size = BNXT_PAGE_SIZE;
4716 	rmem->pg_arr = ctx_pg->ctx_pg_arr;
4717 	rmem->dma_arr = ctx_pg->ctx_dma_arr;
4718 	rmem->flags = BNXT_RMEM_VALID_PTE_FLAG;
4719 
4720 	valid_bits = PTU_PTE_VALID;
4721 
4722 	if (rmem->nr_pages > 1) {
4723 		snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4724 			 "bnxt_ctx_pg_tbl%s_%x_%d",
4725 			 suffix, idx, bp->eth_dev->data->port_id);
4726 		mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4727 		mz = rte_memzone_lookup(mz_name);
4728 		if (!mz) {
4729 			mz = rte_memzone_reserve_aligned(mz_name,
4730 						rmem->nr_pages * 8,
4731 						bp->eth_dev->device->numa_node,
4732 						RTE_MEMZONE_2MB |
4733 						RTE_MEMZONE_SIZE_HINT_ONLY |
4734 						RTE_MEMZONE_IOVA_CONTIG,
4735 						BNXT_PAGE_SIZE);
4736 			if (mz == NULL)
4737 				return -ENOMEM;
4738 		}
4739 
4740 		memset(mz->addr, 0, mz->len);
4741 		mz_phys_addr = mz->iova;
4742 
4743 		rmem->pg_tbl = mz->addr;
4744 		rmem->pg_tbl_map = mz_phys_addr;
4745 		rmem->pg_tbl_mz = mz;
4746 	}
4747 
4748 	snprintf(mz_name, RTE_MEMZONE_NAMESIZE, "bnxt_ctx_%s_%x_%d",
4749 		 suffix, idx, bp->eth_dev->data->port_id);
4750 	mz = rte_memzone_lookup(mz_name);
4751 	if (!mz) {
4752 		mz = rte_memzone_reserve_aligned(mz_name,
4753 						 mem_size,
4754 						 bp->eth_dev->device->numa_node,
4755 						 RTE_MEMZONE_1GB |
4756 						 RTE_MEMZONE_SIZE_HINT_ONLY |
4757 						 RTE_MEMZONE_IOVA_CONTIG,
4758 						 BNXT_PAGE_SIZE);
4759 		if (mz == NULL)
4760 			return -ENOMEM;
4761 	}
4762 
4763 	memset(mz->addr, 0, mz->len);
4764 	mz_phys_addr = mz->iova;
4765 
4766 	for (sz = 0, i = 0; sz < mem_size; sz += BNXT_PAGE_SIZE, i++) {
4767 		rmem->pg_arr[i] = ((char *)mz->addr) + sz;
4768 		rmem->dma_arr[i] = mz_phys_addr + sz;
4769 
4770 		if (rmem->nr_pages > 1) {
4771 			if (i == rmem->nr_pages - 2 &&
4772 			    (rmem->flags & BNXT_RMEM_RING_PTE_FLAG))
4773 				valid_bits |= PTU_PTE_NEXT_TO_LAST;
4774 			else if (i == rmem->nr_pages - 1 &&
4775 				 (rmem->flags & BNXT_RMEM_RING_PTE_FLAG))
4776 				valid_bits |= PTU_PTE_LAST;
4777 
4778 			rmem->pg_tbl[i] = rte_cpu_to_le_64(rmem->dma_arr[i] |
4779 							   valid_bits);
4780 		}
4781 	}
4782 
4783 	rmem->mz = mz;
4784 	if (rmem->vmem_size)
4785 		rmem->vmem = (void **)mz->addr;
4786 	rmem->dma_arr[0] = mz_phys_addr;
4787 	return 0;
4788 }
4789 
4790 static void bnxt_free_ctx_mem(struct bnxt *bp)
4791 {
4792 	int i;
4793 
4794 	if (!bp->ctx || !(bp->ctx->flags & BNXT_CTX_FLAG_INITED))
4795 		return;
4796 
4797 	bp->ctx->flags &= ~BNXT_CTX_FLAG_INITED;
4798 	rte_memzone_free(bp->ctx->qp_mem.ring_mem.mz);
4799 	rte_memzone_free(bp->ctx->srq_mem.ring_mem.mz);
4800 	rte_memzone_free(bp->ctx->cq_mem.ring_mem.mz);
4801 	rte_memzone_free(bp->ctx->vnic_mem.ring_mem.mz);
4802 	rte_memzone_free(bp->ctx->stat_mem.ring_mem.mz);
4803 	rte_memzone_free(bp->ctx->qp_mem.ring_mem.pg_tbl_mz);
4804 	rte_memzone_free(bp->ctx->srq_mem.ring_mem.pg_tbl_mz);
4805 	rte_memzone_free(bp->ctx->cq_mem.ring_mem.pg_tbl_mz);
4806 	rte_memzone_free(bp->ctx->vnic_mem.ring_mem.pg_tbl_mz);
4807 	rte_memzone_free(bp->ctx->stat_mem.ring_mem.pg_tbl_mz);
4808 
4809 	for (i = 0; i < bp->ctx->tqm_fp_rings_count + 1; i++) {
4810 		if (bp->ctx->tqm_mem[i])
4811 			rte_memzone_free(bp->ctx->tqm_mem[i]->ring_mem.mz);
4812 	}
4813 
4814 	rte_free(bp->ctx);
4815 	bp->ctx = NULL;
4816 }
4817 
4818 #define bnxt_roundup(x, y)   ((((x) + ((y) - 1)) / (y)) * (y))
4819 
4820 #define min_t(type, x, y) ({                    \
4821 	type __min1 = (x);                      \
4822 	type __min2 = (y);                      \
4823 	__min1 < __min2 ? __min1 : __min2; })
4824 
4825 #define max_t(type, x, y) ({                    \
4826 	type __max1 = (x);                      \
4827 	type __max2 = (y);                      \
4828 	__max1 > __max2 ? __max1 : __max2; })
4829 
4830 #define clamp_t(type, _x, min, max)     min_t(type, max_t(type, _x, min), max)
4831 
4832 int bnxt_alloc_ctx_mem(struct bnxt *bp)
4833 {
4834 	struct bnxt_ctx_pg_info *ctx_pg;
4835 	struct bnxt_ctx_mem_info *ctx;
4836 	uint32_t mem_size, ena, entries;
4837 	uint32_t entries_sp, min;
4838 	int i, rc;
4839 
4840 	rc = bnxt_hwrm_func_backing_store_qcaps(bp);
4841 	if (rc) {
4842 		PMD_DRV_LOG(ERR, "Query context mem capability failed\n");
4843 		return rc;
4844 	}
4845 	ctx = bp->ctx;
4846 	if (!ctx || (ctx->flags & BNXT_CTX_FLAG_INITED))
4847 		return 0;
4848 
4849 	ctx_pg = &ctx->qp_mem;
4850 	ctx_pg->entries = ctx->qp_min_qp1_entries + ctx->qp_max_l2_entries;
4851 	if (ctx->qp_entry_size) {
4852 		mem_size = ctx->qp_entry_size * ctx_pg->entries;
4853 		rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "qp_mem", 0);
4854 		if (rc)
4855 			return rc;
4856 	}
4857 
4858 	ctx_pg = &ctx->srq_mem;
4859 	ctx_pg->entries = ctx->srq_max_l2_entries;
4860 	if (ctx->srq_entry_size) {
4861 		mem_size = ctx->srq_entry_size * ctx_pg->entries;
4862 		rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "srq_mem", 0);
4863 		if (rc)
4864 			return rc;
4865 	}
4866 
4867 	ctx_pg = &ctx->cq_mem;
4868 	ctx_pg->entries = ctx->cq_max_l2_entries;
4869 	if (ctx->cq_entry_size) {
4870 		mem_size = ctx->cq_entry_size * ctx_pg->entries;
4871 		rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "cq_mem", 0);
4872 		if (rc)
4873 			return rc;
4874 	}
4875 
4876 	ctx_pg = &ctx->vnic_mem;
4877 	ctx_pg->entries = ctx->vnic_max_vnic_entries +
4878 		ctx->vnic_max_ring_table_entries;
4879 	if (ctx->vnic_entry_size) {
4880 		mem_size = ctx->vnic_entry_size * ctx_pg->entries;
4881 		rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "vnic_mem", 0);
4882 		if (rc)
4883 			return rc;
4884 	}
4885 
4886 	ctx_pg = &ctx->stat_mem;
4887 	ctx_pg->entries = ctx->stat_max_entries;
4888 	if (ctx->stat_entry_size) {
4889 		mem_size = ctx->stat_entry_size * ctx_pg->entries;
4890 		rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "stat_mem", 0);
4891 		if (rc)
4892 			return rc;
4893 	}
4894 
4895 	min = ctx->tqm_min_entries_per_ring;
4896 
4897 	entries_sp = ctx->qp_max_l2_entries +
4898 		     ctx->vnic_max_vnic_entries +
4899 		     2 * ctx->qp_min_qp1_entries + min;
4900 	entries_sp = bnxt_roundup(entries_sp, ctx->tqm_entries_multiple);
4901 
4902 	entries = ctx->qp_max_l2_entries + ctx->qp_min_qp1_entries;
4903 	entries = bnxt_roundup(entries, ctx->tqm_entries_multiple);
4904 	entries = clamp_t(uint32_t, entries, min,
4905 			  ctx->tqm_max_entries_per_ring);
4906 	for (i = 0, ena = 0; i < ctx->tqm_fp_rings_count + 1; i++) {
4907 		/* i=0 is for TQM_SP. i=1 to i=8 applies to RING0 to RING7.
4908 		 * i > 8 is other ext rings.
4909 		 */
4910 		ctx_pg = ctx->tqm_mem[i];
4911 		ctx_pg->entries = i ? entries : entries_sp;
4912 		if (ctx->tqm_entry_size) {
4913 			mem_size = ctx->tqm_entry_size * ctx_pg->entries;
4914 			rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size,
4915 						    "tqm_mem", i);
4916 			if (rc)
4917 				return rc;
4918 		}
4919 		if (i < BNXT_MAX_TQM_LEGACY_RINGS)
4920 			ena |= HWRM_FUNC_BACKING_STORE_CFG_INPUT_ENABLES_TQM_SP << i;
4921 		else
4922 			ena |= HWRM_FUNC_BACKING_STORE_CFG_INPUT_ENABLES_TQM_RING8;
4923 	}
4924 
4925 	ena |= FUNC_BACKING_STORE_CFG_INPUT_DFLT_ENABLES;
4926 	rc = bnxt_hwrm_func_backing_store_cfg(bp, ena);
4927 	if (rc)
4928 		PMD_DRV_LOG(ERR,
4929 			    "Failed to configure context mem: rc = %d\n", rc);
4930 	else
4931 		ctx->flags |= BNXT_CTX_FLAG_INITED;
4932 
4933 	return rc;
4934 }
4935 
4936 static int bnxt_alloc_stats_mem(struct bnxt *bp)
4937 {
4938 	struct rte_pci_device *pci_dev = bp->pdev;
4939 	char mz_name[RTE_MEMZONE_NAMESIZE];
4940 	const struct rte_memzone *mz = NULL;
4941 	uint32_t total_alloc_len;
4942 	rte_iova_t mz_phys_addr;
4943 
4944 	if (pci_dev->id.device_id == BROADCOM_DEV_ID_NS2)
4945 		return 0;
4946 
4947 	snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4948 		 "bnxt_" PCI_PRI_FMT "-%s", pci_dev->addr.domain,
4949 		 pci_dev->addr.bus, pci_dev->addr.devid,
4950 		 pci_dev->addr.function, "rx_port_stats");
4951 	mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4952 	mz = rte_memzone_lookup(mz_name);
4953 	total_alloc_len =
4954 		RTE_CACHE_LINE_ROUNDUP(sizeof(struct rx_port_stats) +
4955 				       sizeof(struct rx_port_stats_ext) + 512);
4956 	if (!mz) {
4957 		mz = rte_memzone_reserve(mz_name, total_alloc_len,
4958 					 SOCKET_ID_ANY,
4959 					 RTE_MEMZONE_2MB |
4960 					 RTE_MEMZONE_SIZE_HINT_ONLY |
4961 					 RTE_MEMZONE_IOVA_CONTIG);
4962 		if (mz == NULL)
4963 			return -ENOMEM;
4964 	}
4965 	memset(mz->addr, 0, mz->len);
4966 	mz_phys_addr = mz->iova;
4967 
4968 	bp->rx_mem_zone = (const void *)mz;
4969 	bp->hw_rx_port_stats = mz->addr;
4970 	bp->hw_rx_port_stats_map = mz_phys_addr;
4971 
4972 	snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4973 		 "bnxt_" PCI_PRI_FMT "-%s", pci_dev->addr.domain,
4974 		 pci_dev->addr.bus, pci_dev->addr.devid,
4975 		 pci_dev->addr.function, "tx_port_stats");
4976 	mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4977 	mz = rte_memzone_lookup(mz_name);
4978 	total_alloc_len =
4979 		RTE_CACHE_LINE_ROUNDUP(sizeof(struct tx_port_stats) +
4980 				       sizeof(struct tx_port_stats_ext) + 512);
4981 	if (!mz) {
4982 		mz = rte_memzone_reserve(mz_name,
4983 					 total_alloc_len,
4984 					 SOCKET_ID_ANY,
4985 					 RTE_MEMZONE_2MB |
4986 					 RTE_MEMZONE_SIZE_HINT_ONLY |
4987 					 RTE_MEMZONE_IOVA_CONTIG);
4988 		if (mz == NULL)
4989 			return -ENOMEM;
4990 	}
4991 	memset(mz->addr, 0, mz->len);
4992 	mz_phys_addr = mz->iova;
4993 
4994 	bp->tx_mem_zone = (const void *)mz;
4995 	bp->hw_tx_port_stats = mz->addr;
4996 	bp->hw_tx_port_stats_map = mz_phys_addr;
4997 	bp->flags |= BNXT_FLAG_PORT_STATS;
4998 
4999 	/* Display extended statistics if FW supports it */
5000 	if (bp->hwrm_spec_code < HWRM_SPEC_CODE_1_8_4 ||
5001 	    bp->hwrm_spec_code == HWRM_SPEC_CODE_1_9_0 ||
5002 	    !(bp->flags & BNXT_FLAG_EXT_STATS_SUPPORTED))
5003 		return 0;
5004 
5005 	bp->hw_rx_port_stats_ext = (void *)
5006 		((uint8_t *)bp->hw_rx_port_stats +
5007 		 sizeof(struct rx_port_stats));
5008 	bp->hw_rx_port_stats_ext_map = bp->hw_rx_port_stats_map +
5009 		sizeof(struct rx_port_stats);
5010 	bp->flags |= BNXT_FLAG_EXT_RX_PORT_STATS;
5011 
5012 	if (bp->hwrm_spec_code < HWRM_SPEC_CODE_1_9_2 ||
5013 	    bp->flags & BNXT_FLAG_EXT_STATS_SUPPORTED) {
5014 		bp->hw_tx_port_stats_ext = (void *)
5015 			((uint8_t *)bp->hw_tx_port_stats +
5016 			 sizeof(struct tx_port_stats));
5017 		bp->hw_tx_port_stats_ext_map =
5018 			bp->hw_tx_port_stats_map +
5019 			sizeof(struct tx_port_stats);
5020 		bp->flags |= BNXT_FLAG_EXT_TX_PORT_STATS;
5021 	}
5022 
5023 	return 0;
5024 }
5025 
5026 static int bnxt_setup_mac_addr(struct rte_eth_dev *eth_dev)
5027 {
5028 	struct bnxt *bp = eth_dev->data->dev_private;
5029 	size_t max_mac_addr = RTE_MIN(bp->max_l2_ctx, RTE_ETH_NUM_RECEIVE_MAC_ADDR);
5030 	int rc = 0;
5031 
5032 	if (bp->max_l2_ctx > RTE_ETH_NUM_RECEIVE_MAC_ADDR)
5033 		PMD_DRV_LOG(INFO, "Max number of MAC addrs supported is %d, but will be limited to %d\n",
5034 			    bp->max_l2_ctx, RTE_ETH_NUM_RECEIVE_MAC_ADDR);
5035 
5036 	eth_dev->data->mac_addrs = rte_zmalloc("bnxt_mac_addr_tbl",
5037 					       RTE_ETHER_ADDR_LEN * max_mac_addr,
5038 					       0);
5039 	if (eth_dev->data->mac_addrs == NULL) {
5040 		PMD_DRV_LOG(ERR, "Failed to alloc MAC addr tbl\n");
5041 		return -ENOMEM;
5042 	}
5043 
5044 	if (!BNXT_HAS_DFLT_MAC_SET(bp)) {
5045 		if (BNXT_PF(bp))
5046 			return -EINVAL;
5047 
5048 		/* Generate a random MAC address, if none was assigned by PF */
5049 		PMD_DRV_LOG(INFO, "VF MAC address not assigned by Host PF\n");
5050 		bnxt_eth_hw_addr_random(bp->mac_addr);
5051 		PMD_DRV_LOG(INFO,
5052 			    "Assign random MAC:" RTE_ETHER_ADDR_PRT_FMT "\n",
5053 			    bp->mac_addr[0], bp->mac_addr[1], bp->mac_addr[2],
5054 			    bp->mac_addr[3], bp->mac_addr[4], bp->mac_addr[5]);
5055 
5056 		rc = bnxt_hwrm_set_mac(bp);
5057 		if (rc)
5058 			return rc;
5059 	}
5060 
5061 	/* Copy the permanent MAC from the FUNC_QCAPS response */
5062 	memcpy(&eth_dev->data->mac_addrs[0], bp->mac_addr, RTE_ETHER_ADDR_LEN);
5063 
5064 	/*
5065 	 *  Allocate memory to hold multicast mac addresses added.
5066 	 *  Used to restore them during reset recovery
5067 	 */
5068 	bp->mcast_addr_list = rte_zmalloc("bnxt_mcast_addr_tbl",
5069 					  sizeof(struct rte_ether_addr) *
5070 					  BNXT_MAX_MC_ADDRS, 0);
5071 	if (bp->mcast_addr_list == NULL) {
5072 		PMD_DRV_LOG(ERR, "Failed to allocate multicast addr table\n");
5073 		return -ENOMEM;
5074 	}
5075 	bp->mc_list_dma_addr = rte_malloc_virt2iova(bp->mcast_addr_list);
5076 	if (bp->mc_list_dma_addr == RTE_BAD_IOVA) {
5077 		PMD_DRV_LOG(ERR, "Fail to map mcast_addr_list to physical memory\n");
5078 		return -ENOMEM;
5079 	}
5080 
5081 	return rc;
5082 }
5083 
5084 static int bnxt_restore_dflt_mac(struct bnxt *bp)
5085 {
5086 	int rc = 0;
5087 
5088 	/* MAC is already configured in FW */
5089 	if (BNXT_HAS_DFLT_MAC_SET(bp))
5090 		return 0;
5091 
5092 	/* Restore the old MAC configured */
5093 	rc = bnxt_hwrm_set_mac(bp);
5094 	if (rc)
5095 		PMD_DRV_LOG(ERR, "Failed to restore MAC address\n");
5096 
5097 	return rc;
5098 }
5099 
5100 static void bnxt_config_vf_req_fwd(struct bnxt *bp)
5101 {
5102 	if (!BNXT_PF(bp))
5103 		return;
5104 
5105 	memset(bp->pf->vf_req_fwd, 0, sizeof(bp->pf->vf_req_fwd));
5106 
5107 	if (!(bp->fw_cap & BNXT_FW_CAP_LINK_ADMIN))
5108 		BNXT_HWRM_CMD_TO_FORWARD(HWRM_PORT_PHY_QCFG);
5109 	BNXT_HWRM_CMD_TO_FORWARD(HWRM_FUNC_CFG);
5110 	BNXT_HWRM_CMD_TO_FORWARD(HWRM_FUNC_VF_CFG);
5111 	BNXT_HWRM_CMD_TO_FORWARD(HWRM_CFA_L2_FILTER_ALLOC);
5112 	BNXT_HWRM_CMD_TO_FORWARD(HWRM_OEM_CMD);
5113 }
5114 
5115 static void bnxt_alloc_error_recovery_info(struct bnxt *bp)
5116 {
5117 	struct bnxt_error_recovery_info *info = bp->recovery_info;
5118 
5119 	if (info) {
5120 		if (!(bp->fw_cap & BNXT_FW_CAP_HCOMM_FW_STATUS))
5121 			memset(info, 0, sizeof(*info));
5122 		return;
5123 	}
5124 
5125 	if (!(bp->fw_cap & BNXT_FW_CAP_ERROR_RECOVERY))
5126 		return;
5127 
5128 	info = rte_zmalloc("bnxt_hwrm_error_recovery_qcfg",
5129 			   sizeof(*info), 0);
5130 	if (!info)
5131 		bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5132 
5133 	bp->recovery_info = info;
5134 }
5135 
5136 static void bnxt_check_fw_status(struct bnxt *bp)
5137 {
5138 	uint32_t fw_status;
5139 
5140 	if (!(bp->recovery_info &&
5141 	      (bp->fw_cap & BNXT_FW_CAP_HCOMM_FW_STATUS)))
5142 		return;
5143 
5144 	fw_status = bnxt_read_fw_status_reg(bp, BNXT_FW_STATUS_REG);
5145 	if (fw_status != BNXT_FW_STATUS_HEALTHY)
5146 		PMD_DRV_LOG(ERR, "Firmware not responding, status: %#x\n",
5147 			    fw_status);
5148 }
5149 
5150 static int bnxt_map_hcomm_fw_status_reg(struct bnxt *bp)
5151 {
5152 	struct bnxt_error_recovery_info *info = bp->recovery_info;
5153 	uint32_t status_loc;
5154 	uint32_t sig_ver;
5155 
5156 	rte_write32(HCOMM_STATUS_STRUCT_LOC, (uint8_t *)bp->bar0 +
5157 		    BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
5158 	sig_ver = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
5159 				   BNXT_GRCP_WINDOW_2_BASE +
5160 				   offsetof(struct hcomm_status,
5161 					    sig_ver)));
5162 	/* If the signature is absent, then FW does not support this feature */
5163 	if ((sig_ver & HCOMM_STATUS_SIGNATURE_MASK) !=
5164 	    HCOMM_STATUS_SIGNATURE_VAL)
5165 		return 0;
5166 
5167 	if (!info) {
5168 		info = rte_zmalloc("bnxt_hwrm_error_recovery_qcfg",
5169 				   sizeof(*info), 0);
5170 		if (!info)
5171 			return -ENOMEM;
5172 		bp->recovery_info = info;
5173 	} else {
5174 		memset(info, 0, sizeof(*info));
5175 	}
5176 
5177 	status_loc = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
5178 				      BNXT_GRCP_WINDOW_2_BASE +
5179 				      offsetof(struct hcomm_status,
5180 					       fw_status_loc)));
5181 
5182 	/* Only pre-map the FW health status GRC register */
5183 	if (BNXT_FW_STATUS_REG_TYPE(status_loc) != BNXT_FW_STATUS_REG_TYPE_GRC)
5184 		return 0;
5185 
5186 	info->status_regs[BNXT_FW_STATUS_REG] = status_loc;
5187 	info->mapped_status_regs[BNXT_FW_STATUS_REG] =
5188 		BNXT_GRCP_WINDOW_2_BASE + (status_loc & BNXT_GRCP_OFFSET_MASK);
5189 
5190 	rte_write32((status_loc & BNXT_GRCP_BASE_MASK), (uint8_t *)bp->bar0 +
5191 		    BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
5192 
5193 	bp->fw_cap |= BNXT_FW_CAP_HCOMM_FW_STATUS;
5194 
5195 	return 0;
5196 }
5197 
5198 /* This function gets the FW version along with the
5199  * capabilities(MAX and current) of the function, vnic,
5200  * error recovery, phy and other chip related info
5201  */
5202 static int bnxt_get_config(struct bnxt *bp)
5203 {
5204 	uint16_t mtu;
5205 	int rc = 0;
5206 
5207 	bp->fw_cap = 0;
5208 
5209 	rc = bnxt_map_hcomm_fw_status_reg(bp);
5210 	if (rc)
5211 		return rc;
5212 
5213 	rc = bnxt_hwrm_ver_get(bp, DFLT_HWRM_CMD_TIMEOUT);
5214 	if (rc) {
5215 		bnxt_check_fw_status(bp);
5216 		return rc;
5217 	}
5218 
5219 	rc = bnxt_hwrm_func_reset(bp);
5220 	if (rc)
5221 		return -EIO;
5222 
5223 	rc = bnxt_hwrm_vnic_qcaps(bp);
5224 	if (rc)
5225 		return rc;
5226 
5227 	rc = bnxt_hwrm_queue_qportcfg(bp);
5228 	if (rc)
5229 		return rc;
5230 
5231 	/* Get the MAX capabilities for this function.
5232 	 * This function also allocates context memory for TQM rings and
5233 	 * informs the firmware about this allocated backing store memory.
5234 	 */
5235 	rc = bnxt_hwrm_func_qcaps(bp);
5236 	if (rc)
5237 		return rc;
5238 
5239 	rc = bnxt_hwrm_func_qcfg(bp, &mtu);
5240 	if (rc)
5241 		return rc;
5242 
5243 	bnxt_hwrm_port_mac_qcfg(bp);
5244 
5245 	bnxt_hwrm_parent_pf_qcfg(bp);
5246 
5247 	bnxt_hwrm_port_phy_qcaps(bp);
5248 
5249 	bnxt_alloc_error_recovery_info(bp);
5250 	/* Get the adapter error recovery support info */
5251 	rc = bnxt_hwrm_error_recovery_qcfg(bp);
5252 	if (rc)
5253 		bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5254 
5255 	bnxt_hwrm_port_led_qcaps(bp);
5256 
5257 	return 0;
5258 }
5259 
5260 static int
5261 bnxt_init_locks(struct bnxt *bp)
5262 {
5263 	int err;
5264 
5265 	err = pthread_mutex_init(&bp->flow_lock, NULL);
5266 	if (err) {
5267 		PMD_DRV_LOG(ERR, "Unable to initialize flow_lock\n");
5268 		return err;
5269 	}
5270 
5271 	err = pthread_mutex_init(&bp->def_cp_lock, NULL);
5272 	if (err) {
5273 		PMD_DRV_LOG(ERR, "Unable to initialize def_cp_lock\n");
5274 		return err;
5275 	}
5276 
5277 	err = pthread_mutex_init(&bp->health_check_lock, NULL);
5278 	if (err) {
5279 		PMD_DRV_LOG(ERR, "Unable to initialize health_check_lock\n");
5280 		return err;
5281 	}
5282 
5283 	err = pthread_mutex_init(&bp->err_recovery_lock, NULL);
5284 	if (err)
5285 		PMD_DRV_LOG(ERR, "Unable to initialize err_recovery_lock\n");
5286 
5287 	return err;
5288 }
5289 
5290 static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev)
5291 {
5292 	int rc = 0;
5293 
5294 	rc = bnxt_get_config(bp);
5295 	if (rc)
5296 		return rc;
5297 
5298 	if (!reconfig_dev) {
5299 		rc = bnxt_setup_mac_addr(bp->eth_dev);
5300 		if (rc)
5301 			return rc;
5302 	} else {
5303 		rc = bnxt_restore_dflt_mac(bp);
5304 		if (rc)
5305 			return rc;
5306 	}
5307 
5308 	bnxt_config_vf_req_fwd(bp);
5309 
5310 	rc = bnxt_hwrm_func_driver_register(bp);
5311 	if (rc) {
5312 		PMD_DRV_LOG(ERR, "Failed to register driver");
5313 		return -EBUSY;
5314 	}
5315 
5316 	if (BNXT_PF(bp)) {
5317 		if (bp->pdev->max_vfs) {
5318 			rc = bnxt_hwrm_allocate_vfs(bp, bp->pdev->max_vfs);
5319 			if (rc) {
5320 				PMD_DRV_LOG(ERR, "Failed to allocate VFs\n");
5321 				return rc;
5322 			}
5323 		} else {
5324 			rc = bnxt_hwrm_allocate_pf_only(bp);
5325 			if (rc) {
5326 				PMD_DRV_LOG(ERR,
5327 					    "Failed to allocate PF resources");
5328 				return rc;
5329 			}
5330 		}
5331 	}
5332 
5333 	if (!reconfig_dev) {
5334 		bp->rss_conf.rss_key = rte_zmalloc("bnxt_rss_key",
5335 						   HW_HASH_KEY_SIZE, 0);
5336 		if (bp->rss_conf.rss_key == NULL) {
5337 			PMD_DRV_LOG(ERR, "port %u cannot allocate RSS hash key memory",
5338 				    bp->eth_dev->data->port_id);
5339 			return -ENOMEM;
5340 		}
5341 	}
5342 
5343 	rc = bnxt_alloc_mem(bp, reconfig_dev);
5344 	if (rc)
5345 		return rc;
5346 
5347 	rc = bnxt_setup_int(bp);
5348 	if (rc)
5349 		return rc;
5350 
5351 	rc = bnxt_request_int(bp);
5352 	if (rc)
5353 		return rc;
5354 
5355 	rc = bnxt_init_ctx_mem(bp);
5356 	if (rc) {
5357 		PMD_DRV_LOG(ERR, "Failed to init adv_flow_counters\n");
5358 		return rc;
5359 	}
5360 
5361 	return 0;
5362 }
5363 
5364 static int
5365 bnxt_parse_devarg_flow_xstat(__rte_unused const char *key,
5366 			     const char *value, void *opaque_arg)
5367 {
5368 	struct bnxt *bp = opaque_arg;
5369 	unsigned long flow_xstat;
5370 	char *end = NULL;
5371 
5372 	if (!value || !opaque_arg) {
5373 		PMD_DRV_LOG(ERR,
5374 			    "Invalid parameter passed to flow_xstat devarg.\n");
5375 		return -EINVAL;
5376 	}
5377 
5378 	flow_xstat = strtoul(value, &end, 10);
5379 	if (end == NULL || *end != '\0' ||
5380 	    (flow_xstat == ULONG_MAX && errno == ERANGE)) {
5381 		PMD_DRV_LOG(ERR,
5382 			    "Invalid parameter passed to flow_xstat devarg.\n");
5383 		return -EINVAL;
5384 	}
5385 
5386 	if (BNXT_DEVARG_FLOW_XSTAT_INVALID(flow_xstat)) {
5387 		PMD_DRV_LOG(ERR,
5388 			    "Invalid value passed to flow_xstat devarg.\n");
5389 		return -EINVAL;
5390 	}
5391 
5392 	bp->flags |= BNXT_FLAG_FLOW_XSTATS_EN;
5393 	if (BNXT_FLOW_XSTATS_EN(bp))
5394 		PMD_DRV_LOG(INFO, "flow_xstat feature enabled.\n");
5395 
5396 	return 0;
5397 }
5398 
5399 static int
5400 bnxt_parse_devarg_max_num_kflows(__rte_unused const char *key,
5401 					const char *value, void *opaque_arg)
5402 {
5403 	struct bnxt *bp = opaque_arg;
5404 	unsigned long max_num_kflows;
5405 	char *end = NULL;
5406 
5407 	if (!value || !opaque_arg) {
5408 		PMD_DRV_LOG(ERR,
5409 			"Invalid parameter passed to max_num_kflows devarg.\n");
5410 		return -EINVAL;
5411 	}
5412 
5413 	max_num_kflows = strtoul(value, &end, 10);
5414 	if (end == NULL || *end != '\0' ||
5415 		(max_num_kflows == ULONG_MAX && errno == ERANGE)) {
5416 		PMD_DRV_LOG(ERR,
5417 			"Invalid parameter passed to max_num_kflows devarg.\n");
5418 		return -EINVAL;
5419 	}
5420 
5421 	if (bnxt_devarg_max_num_kflow_invalid(max_num_kflows)) {
5422 		PMD_DRV_LOG(ERR,
5423 			"Invalid value passed to max_num_kflows devarg.\n");
5424 		return -EINVAL;
5425 	}
5426 
5427 	bp->max_num_kflows = max_num_kflows;
5428 	if (bp->max_num_kflows)
5429 		PMD_DRV_LOG(INFO, "max_num_kflows set as %ldK.\n",
5430 				max_num_kflows);
5431 
5432 	return 0;
5433 }
5434 
5435 static int
5436 bnxt_parse_devarg_app_id(__rte_unused const char *key,
5437 				 const char *value, void *opaque_arg)
5438 {
5439 	struct bnxt *bp = opaque_arg;
5440 	unsigned long app_id;
5441 	char *end = NULL;
5442 
5443 	if (!value || !opaque_arg) {
5444 		PMD_DRV_LOG(ERR,
5445 			    "Invalid parameter passed to app-id "
5446 			    "devargs.\n");
5447 		return -EINVAL;
5448 	}
5449 
5450 	app_id = strtoul(value, &end, 10);
5451 	if (end == NULL || *end != '\0' ||
5452 	    (app_id == ULONG_MAX && errno == ERANGE)) {
5453 		PMD_DRV_LOG(ERR,
5454 			    "Invalid parameter passed to app_id "
5455 			    "devargs.\n");
5456 		return -EINVAL;
5457 	}
5458 
5459 	if (BNXT_DEVARG_APP_ID_INVALID(app_id)) {
5460 		PMD_DRV_LOG(ERR, "Invalid app-id(%d) devargs.\n",
5461 			    (uint16_t)app_id);
5462 		return -EINVAL;
5463 	}
5464 
5465 	bp->app_id = app_id;
5466 	PMD_DRV_LOG(INFO, "app-id=%d feature enabled.\n", (uint16_t)app_id);
5467 
5468 	return 0;
5469 }
5470 
5471 static int
5472 bnxt_parse_devarg_rep_is_pf(__rte_unused const char *key,
5473 			    const char *value, void *opaque_arg)
5474 {
5475 	struct bnxt_representor *vfr_bp = opaque_arg;
5476 	unsigned long rep_is_pf;
5477 	char *end = NULL;
5478 
5479 	if (!value || !opaque_arg) {
5480 		PMD_DRV_LOG(ERR,
5481 			    "Invalid parameter passed to rep_is_pf devargs.\n");
5482 		return -EINVAL;
5483 	}
5484 
5485 	rep_is_pf = strtoul(value, &end, 10);
5486 	if (end == NULL || *end != '\0' ||
5487 	    (rep_is_pf == ULONG_MAX && errno == ERANGE)) {
5488 		PMD_DRV_LOG(ERR,
5489 			    "Invalid parameter passed to rep_is_pf devargs.\n");
5490 		return -EINVAL;
5491 	}
5492 
5493 	if (BNXT_DEVARG_REP_IS_PF_INVALID(rep_is_pf)) {
5494 		PMD_DRV_LOG(ERR,
5495 			    "Invalid value passed to rep_is_pf devargs.\n");
5496 		return -EINVAL;
5497 	}
5498 
5499 	vfr_bp->flags |= rep_is_pf;
5500 	if (BNXT_REP_PF(vfr_bp))
5501 		PMD_DRV_LOG(INFO, "PF representor\n");
5502 	else
5503 		PMD_DRV_LOG(INFO, "VF representor\n");
5504 
5505 	return 0;
5506 }
5507 
5508 static int
5509 bnxt_parse_devarg_rep_based_pf(__rte_unused const char *key,
5510 			       const char *value, void *opaque_arg)
5511 {
5512 	struct bnxt_representor *vfr_bp = opaque_arg;
5513 	unsigned long rep_based_pf;
5514 	char *end = NULL;
5515 
5516 	if (!value || !opaque_arg) {
5517 		PMD_DRV_LOG(ERR,
5518 			    "Invalid parameter passed to rep_based_pf "
5519 			    "devargs.\n");
5520 		return -EINVAL;
5521 	}
5522 
5523 	rep_based_pf = strtoul(value, &end, 10);
5524 	if (end == NULL || *end != '\0' ||
5525 	    (rep_based_pf == ULONG_MAX && errno == ERANGE)) {
5526 		PMD_DRV_LOG(ERR,
5527 			    "Invalid parameter passed to rep_based_pf "
5528 			    "devargs.\n");
5529 		return -EINVAL;
5530 	}
5531 
5532 	if (BNXT_DEVARG_REP_BASED_PF_INVALID(rep_based_pf)) {
5533 		PMD_DRV_LOG(ERR,
5534 			    "Invalid value passed to rep_based_pf devargs.\n");
5535 		return -EINVAL;
5536 	}
5537 
5538 	vfr_bp->rep_based_pf = rep_based_pf;
5539 	vfr_bp->flags |= BNXT_REP_BASED_PF_VALID;
5540 
5541 	PMD_DRV_LOG(INFO, "rep-based-pf = %d\n", vfr_bp->rep_based_pf);
5542 
5543 	return 0;
5544 }
5545 
5546 static int
5547 bnxt_parse_devarg_rep_q_r2f(__rte_unused const char *key,
5548 			    const char *value, void *opaque_arg)
5549 {
5550 	struct bnxt_representor *vfr_bp = opaque_arg;
5551 	unsigned long rep_q_r2f;
5552 	char *end = NULL;
5553 
5554 	if (!value || !opaque_arg) {
5555 		PMD_DRV_LOG(ERR,
5556 			    "Invalid parameter passed to rep_q_r2f "
5557 			    "devargs.\n");
5558 		return -EINVAL;
5559 	}
5560 
5561 	rep_q_r2f = strtoul(value, &end, 10);
5562 	if (end == NULL || *end != '\0' ||
5563 	    (rep_q_r2f == ULONG_MAX && errno == ERANGE)) {
5564 		PMD_DRV_LOG(ERR,
5565 			    "Invalid parameter passed to rep_q_r2f "
5566 			    "devargs.\n");
5567 		return -EINVAL;
5568 	}
5569 
5570 	if (BNXT_DEVARG_REP_Q_R2F_INVALID(rep_q_r2f)) {
5571 		PMD_DRV_LOG(ERR,
5572 			    "Invalid value passed to rep_q_r2f devargs.\n");
5573 		return -EINVAL;
5574 	}
5575 
5576 	vfr_bp->rep_q_r2f = rep_q_r2f;
5577 	vfr_bp->flags |= BNXT_REP_Q_R2F_VALID;
5578 	PMD_DRV_LOG(INFO, "rep-q-r2f = %d\n", vfr_bp->rep_q_r2f);
5579 
5580 	return 0;
5581 }
5582 
5583 static int
5584 bnxt_parse_devarg_rep_q_f2r(__rte_unused const char *key,
5585 			    const char *value, void *opaque_arg)
5586 {
5587 	struct bnxt_representor *vfr_bp = opaque_arg;
5588 	unsigned long rep_q_f2r;
5589 	char *end = NULL;
5590 
5591 	if (!value || !opaque_arg) {
5592 		PMD_DRV_LOG(ERR,
5593 			    "Invalid parameter passed to rep_q_f2r "
5594 			    "devargs.\n");
5595 		return -EINVAL;
5596 	}
5597 
5598 	rep_q_f2r = strtoul(value, &end, 10);
5599 	if (end == NULL || *end != '\0' ||
5600 	    (rep_q_f2r == ULONG_MAX && errno == ERANGE)) {
5601 		PMD_DRV_LOG(ERR,
5602 			    "Invalid parameter passed to rep_q_f2r "
5603 			    "devargs.\n");
5604 		return -EINVAL;
5605 	}
5606 
5607 	if (BNXT_DEVARG_REP_Q_F2R_INVALID(rep_q_f2r)) {
5608 		PMD_DRV_LOG(ERR,
5609 			    "Invalid value passed to rep_q_f2r devargs.\n");
5610 		return -EINVAL;
5611 	}
5612 
5613 	vfr_bp->rep_q_f2r = rep_q_f2r;
5614 	vfr_bp->flags |= BNXT_REP_Q_F2R_VALID;
5615 	PMD_DRV_LOG(INFO, "rep-q-f2r = %d\n", vfr_bp->rep_q_f2r);
5616 
5617 	return 0;
5618 }
5619 
5620 static int
5621 bnxt_parse_devarg_rep_fc_r2f(__rte_unused const char *key,
5622 			     const char *value, void *opaque_arg)
5623 {
5624 	struct bnxt_representor *vfr_bp = opaque_arg;
5625 	unsigned long rep_fc_r2f;
5626 	char *end = NULL;
5627 
5628 	if (!value || !opaque_arg) {
5629 		PMD_DRV_LOG(ERR,
5630 			    "Invalid parameter passed to rep_fc_r2f "
5631 			    "devargs.\n");
5632 		return -EINVAL;
5633 	}
5634 
5635 	rep_fc_r2f = strtoul(value, &end, 10);
5636 	if (end == NULL || *end != '\0' ||
5637 	    (rep_fc_r2f == ULONG_MAX && errno == ERANGE)) {
5638 		PMD_DRV_LOG(ERR,
5639 			    "Invalid parameter passed to rep_fc_r2f "
5640 			    "devargs.\n");
5641 		return -EINVAL;
5642 	}
5643 
5644 	if (BNXT_DEVARG_REP_FC_R2F_INVALID(rep_fc_r2f)) {
5645 		PMD_DRV_LOG(ERR,
5646 			    "Invalid value passed to rep_fc_r2f devargs.\n");
5647 		return -EINVAL;
5648 	}
5649 
5650 	vfr_bp->flags |= BNXT_REP_FC_R2F_VALID;
5651 	vfr_bp->rep_fc_r2f = rep_fc_r2f;
5652 	PMD_DRV_LOG(INFO, "rep-fc-r2f = %lu\n", rep_fc_r2f);
5653 
5654 	return 0;
5655 }
5656 
5657 static int
5658 bnxt_parse_devarg_rep_fc_f2r(__rte_unused const char *key,
5659 			     const char *value, void *opaque_arg)
5660 {
5661 	struct bnxt_representor *vfr_bp = opaque_arg;
5662 	unsigned long rep_fc_f2r;
5663 	char *end = NULL;
5664 
5665 	if (!value || !opaque_arg) {
5666 		PMD_DRV_LOG(ERR,
5667 			    "Invalid parameter passed to rep_fc_f2r "
5668 			    "devargs.\n");
5669 		return -EINVAL;
5670 	}
5671 
5672 	rep_fc_f2r = strtoul(value, &end, 10);
5673 	if (end == NULL || *end != '\0' ||
5674 	    (rep_fc_f2r == ULONG_MAX && errno == ERANGE)) {
5675 		PMD_DRV_LOG(ERR,
5676 			    "Invalid parameter passed to rep_fc_f2r "
5677 			    "devargs.\n");
5678 		return -EINVAL;
5679 	}
5680 
5681 	if (BNXT_DEVARG_REP_FC_F2R_INVALID(rep_fc_f2r)) {
5682 		PMD_DRV_LOG(ERR,
5683 			    "Invalid value passed to rep_fc_f2r devargs.\n");
5684 		return -EINVAL;
5685 	}
5686 
5687 	vfr_bp->flags |= BNXT_REP_FC_F2R_VALID;
5688 	vfr_bp->rep_fc_f2r = rep_fc_f2r;
5689 	PMD_DRV_LOG(INFO, "rep-fc-f2r = %lu\n", rep_fc_f2r);
5690 
5691 	return 0;
5692 }
5693 
5694 static int
5695 bnxt_parse_dev_args(struct bnxt *bp, struct rte_devargs *devargs)
5696 {
5697 	struct rte_kvargs *kvlist;
5698 	int ret;
5699 
5700 	if (devargs == NULL)
5701 		return 0;
5702 
5703 	kvlist = rte_kvargs_parse(devargs->args, bnxt_dev_args);
5704 	if (kvlist == NULL)
5705 		return -EINVAL;
5706 
5707 	/*
5708 	 * Handler for "flow_xstat" devarg.
5709 	 * Invoked as for ex: "-a 0000:00:0d.0,flow_xstat=1"
5710 	 */
5711 	ret = rte_kvargs_process(kvlist, BNXT_DEVARG_FLOW_XSTAT,
5712 				 bnxt_parse_devarg_flow_xstat, bp);
5713 	if (ret)
5714 		goto err;
5715 
5716 	/*
5717 	 * Handler for "max_num_kflows" devarg.
5718 	 * Invoked as for ex: "-a 000:00:0d.0,max_num_kflows=32"
5719 	 */
5720 	ret = rte_kvargs_process(kvlist, BNXT_DEVARG_MAX_NUM_KFLOWS,
5721 				 bnxt_parse_devarg_max_num_kflows, bp);
5722 	if (ret)
5723 		goto err;
5724 
5725 err:
5726 	/*
5727 	 * Handler for "app-id" devarg.
5728 	 * Invoked as for ex: "-a 000:00:0d.0,app-id=1"
5729 	 */
5730 	rte_kvargs_process(kvlist, BNXT_DEVARG_APP_ID,
5731 			   bnxt_parse_devarg_app_id, bp);
5732 
5733 	rte_kvargs_free(kvlist);
5734 	return ret;
5735 }
5736 
5737 static int bnxt_alloc_switch_domain(struct bnxt *bp)
5738 {
5739 	int rc = 0;
5740 
5741 	if (BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) {
5742 		rc = rte_eth_switch_domain_alloc(&bp->switch_domain_id);
5743 		if (rc)
5744 			PMD_DRV_LOG(ERR,
5745 				    "Failed to alloc switch domain: %d\n", rc);
5746 		else
5747 			PMD_DRV_LOG(INFO,
5748 				    "Switch domain allocated %d\n",
5749 				    bp->switch_domain_id);
5750 	}
5751 
5752 	return rc;
5753 }
5754 
5755 /* Allocate and initialize various fields in bnxt struct that
5756  * need to be allocated/destroyed only once in the lifetime of the driver
5757  */
5758 static int bnxt_drv_init(struct rte_eth_dev *eth_dev)
5759 {
5760 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
5761 	struct bnxt *bp = eth_dev->data->dev_private;
5762 	int rc = 0;
5763 
5764 	bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
5765 
5766 	if (bnxt_vf_pciid(pci_dev->id.device_id))
5767 		bp->flags |= BNXT_FLAG_VF;
5768 
5769 	if (bnxt_p5_device(pci_dev->id.device_id))
5770 		bp->flags |= BNXT_FLAG_CHIP_P5;
5771 
5772 	if (pci_dev->id.device_id == BROADCOM_DEV_ID_58802 ||
5773 	    pci_dev->id.device_id == BROADCOM_DEV_ID_58804 ||
5774 	    pci_dev->id.device_id == BROADCOM_DEV_ID_58808 ||
5775 	    pci_dev->id.device_id == BROADCOM_DEV_ID_58802_VF)
5776 		bp->flags |= BNXT_FLAG_STINGRAY;
5777 
5778 	if (BNXT_TRUFLOW_EN(bp)) {
5779 		/* extra mbuf field is required to store CFA code from mark */
5780 		static const struct rte_mbuf_dynfield bnxt_cfa_code_dynfield_desc = {
5781 			.name = RTE_PMD_BNXT_CFA_CODE_DYNFIELD_NAME,
5782 			.size = sizeof(bnxt_cfa_code_dynfield_t),
5783 			.align = __alignof__(bnxt_cfa_code_dynfield_t),
5784 		};
5785 		bnxt_cfa_code_dynfield_offset =
5786 			rte_mbuf_dynfield_register(&bnxt_cfa_code_dynfield_desc);
5787 		if (bnxt_cfa_code_dynfield_offset < 0) {
5788 			PMD_DRV_LOG(ERR,
5789 			    "Failed to register mbuf field for TruFlow mark\n");
5790 			return -rte_errno;
5791 		}
5792 	}
5793 
5794 	rc = bnxt_map_pci_bars(eth_dev);
5795 	if (rc) {
5796 		PMD_DRV_LOG(ERR,
5797 			    "Failed to initialize board rc: %x\n", rc);
5798 		return rc;
5799 	}
5800 
5801 	rc = bnxt_alloc_pf_info(bp);
5802 	if (rc)
5803 		return rc;
5804 
5805 	rc = bnxt_alloc_link_info(bp);
5806 	if (rc)
5807 		return rc;
5808 
5809 	rc = bnxt_alloc_parent_info(bp);
5810 	if (rc)
5811 		return rc;
5812 
5813 	rc = bnxt_alloc_hwrm_resources(bp);
5814 	if (rc) {
5815 		PMD_DRV_LOG(ERR,
5816 			    "Failed to allocate response buffer rc: %x\n", rc);
5817 		return rc;
5818 	}
5819 	rc = bnxt_alloc_leds_info(bp);
5820 	if (rc)
5821 		return rc;
5822 
5823 	rc = bnxt_alloc_cos_queues(bp);
5824 	if (rc)
5825 		return rc;
5826 
5827 	rc = bnxt_init_locks(bp);
5828 	if (rc)
5829 		return rc;
5830 
5831 	rc = bnxt_alloc_switch_domain(bp);
5832 	if (rc)
5833 		return rc;
5834 
5835 	return rc;
5836 }
5837 
5838 static int
5839 bnxt_dev_init(struct rte_eth_dev *eth_dev, void *params __rte_unused)
5840 {
5841 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
5842 	static int version_printed;
5843 	struct bnxt *bp;
5844 	int rc;
5845 
5846 	if (version_printed++ == 0)
5847 		PMD_DRV_LOG(INFO, "%s\n", bnxt_version);
5848 
5849 	eth_dev->dev_ops = &bnxt_dev_ops;
5850 	eth_dev->rx_queue_count = bnxt_rx_queue_count_op;
5851 	eth_dev->rx_descriptor_status = bnxt_rx_descriptor_status_op;
5852 	eth_dev->tx_descriptor_status = bnxt_tx_descriptor_status_op;
5853 	eth_dev->rx_pkt_burst = &bnxt_recv_pkts;
5854 	eth_dev->tx_pkt_burst = &bnxt_xmit_pkts;
5855 
5856 	/*
5857 	 * For secondary processes, we don't initialise any further
5858 	 * as primary has already done this work.
5859 	 */
5860 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
5861 		return 0;
5862 
5863 	rte_eth_copy_pci_info(eth_dev, pci_dev);
5864 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
5865 
5866 	bp = eth_dev->data->dev_private;
5867 
5868 	/* Parse dev arguments passed on when starting the DPDK application. */
5869 	rc = bnxt_parse_dev_args(bp, pci_dev->device.devargs);
5870 	if (rc)
5871 		goto error_free;
5872 
5873 	rc = bnxt_drv_init(eth_dev);
5874 	if (rc)
5875 		goto error_free;
5876 
5877 	rc = bnxt_init_resources(bp, false);
5878 	if (rc)
5879 		goto error_free;
5880 
5881 	rc = bnxt_alloc_stats_mem(bp);
5882 	if (rc)
5883 		goto error_free;
5884 
5885 	PMD_DRV_LOG(INFO,
5886 		    "Found %s device at mem %" PRIX64 ", node addr %pM\n",
5887 		    DRV_MODULE_NAME,
5888 		    pci_dev->mem_resource[0].phys_addr,
5889 		    pci_dev->mem_resource[0].addr);
5890 
5891 	return 0;
5892 
5893 error_free:
5894 	bnxt_dev_uninit(eth_dev);
5895 	return rc;
5896 }
5897 
5898 
5899 static void bnxt_free_ctx_mem_buf(struct bnxt_ctx_mem_buf_info *ctx)
5900 {
5901 	if (!ctx)
5902 		return;
5903 
5904 	rte_free(ctx->va);
5905 
5906 	ctx->va = NULL;
5907 	ctx->dma = RTE_BAD_IOVA;
5908 	ctx->ctx_id = BNXT_CTX_VAL_INVAL;
5909 }
5910 
5911 static void bnxt_unregister_fc_ctx_mem(struct bnxt *bp)
5912 {
5913 	bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_RX,
5914 				  CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
5915 				  bp->flow_stat->rx_fc_out_tbl.ctx_id,
5916 				  bp->flow_stat->max_fc,
5917 				  false);
5918 
5919 	bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_TX,
5920 				  CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
5921 				  bp->flow_stat->tx_fc_out_tbl.ctx_id,
5922 				  bp->flow_stat->max_fc,
5923 				  false);
5924 
5925 	if (bp->flow_stat->rx_fc_in_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5926 		bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->rx_fc_in_tbl.ctx_id);
5927 	bp->flow_stat->rx_fc_in_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5928 
5929 	if (bp->flow_stat->rx_fc_out_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5930 		bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->rx_fc_out_tbl.ctx_id);
5931 	bp->flow_stat->rx_fc_out_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5932 
5933 	if (bp->flow_stat->tx_fc_in_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5934 		bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->tx_fc_in_tbl.ctx_id);
5935 	bp->flow_stat->tx_fc_in_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5936 
5937 	if (bp->flow_stat->tx_fc_out_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5938 		bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->tx_fc_out_tbl.ctx_id);
5939 	bp->flow_stat->tx_fc_out_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5940 }
5941 
5942 static void bnxt_uninit_fc_ctx_mem(struct bnxt *bp)
5943 {
5944 	bnxt_unregister_fc_ctx_mem(bp);
5945 
5946 	bnxt_free_ctx_mem_buf(&bp->flow_stat->rx_fc_in_tbl);
5947 	bnxt_free_ctx_mem_buf(&bp->flow_stat->rx_fc_out_tbl);
5948 	bnxt_free_ctx_mem_buf(&bp->flow_stat->tx_fc_in_tbl);
5949 	bnxt_free_ctx_mem_buf(&bp->flow_stat->tx_fc_out_tbl);
5950 }
5951 
5952 static void bnxt_uninit_ctx_mem(struct bnxt *bp)
5953 {
5954 	if (BNXT_FLOW_XSTATS_EN(bp))
5955 		bnxt_uninit_fc_ctx_mem(bp);
5956 }
5957 
5958 static void
5959 bnxt_free_error_recovery_info(struct bnxt *bp)
5960 {
5961 	rte_free(bp->recovery_info);
5962 	bp->recovery_info = NULL;
5963 	bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5964 }
5965 
5966 static int
5967 bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev)
5968 {
5969 	int rc;
5970 
5971 	bnxt_free_int(bp);
5972 	bnxt_free_mem(bp, reconfig_dev);
5973 
5974 	bnxt_hwrm_func_buf_unrgtr(bp);
5975 	if (bp->pf != NULL) {
5976 		rte_free(bp->pf->vf_req_buf);
5977 		bp->pf->vf_req_buf = NULL;
5978 	}
5979 
5980 	rc = bnxt_hwrm_func_driver_unregister(bp);
5981 	bp->flags &= ~BNXT_FLAG_REGISTERED;
5982 	bnxt_free_ctx_mem(bp);
5983 	if (!reconfig_dev) {
5984 		bnxt_free_hwrm_resources(bp);
5985 		bnxt_free_error_recovery_info(bp);
5986 		rte_free(bp->mcast_addr_list);
5987 		bp->mcast_addr_list = NULL;
5988 		rte_free(bp->rss_conf.rss_key);
5989 		bp->rss_conf.rss_key = NULL;
5990 	}
5991 
5992 	bnxt_uninit_ctx_mem(bp);
5993 
5994 	bnxt_free_flow_stats_info(bp);
5995 	if (bp->rep_info != NULL)
5996 		bnxt_free_switch_domain(bp);
5997 	bnxt_free_rep_info(bp);
5998 	rte_free(bp->ptp_cfg);
5999 	bp->ptp_cfg = NULL;
6000 	return rc;
6001 }
6002 
6003 static int
6004 bnxt_dev_uninit(struct rte_eth_dev *eth_dev)
6005 {
6006 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
6007 		return -EPERM;
6008 
6009 	PMD_DRV_LOG(DEBUG, "Calling Device uninit\n");
6010 
6011 	if (eth_dev->state != RTE_ETH_DEV_UNUSED)
6012 		bnxt_dev_close_op(eth_dev);
6013 
6014 	return 0;
6015 }
6016 
6017 static int bnxt_pci_remove_dev_with_reps(struct rte_eth_dev *eth_dev)
6018 {
6019 	struct bnxt *bp = eth_dev->data->dev_private;
6020 	struct rte_eth_dev *vf_rep_eth_dev;
6021 	int ret = 0, i;
6022 
6023 	if (!bp)
6024 		return -EINVAL;
6025 
6026 	for (i = 0; i < bp->num_reps; i++) {
6027 		vf_rep_eth_dev = bp->rep_info[i].vfr_eth_dev;
6028 		if (!vf_rep_eth_dev)
6029 			continue;
6030 		PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR pci remove\n",
6031 			    vf_rep_eth_dev->data->port_id);
6032 		rte_eth_dev_destroy(vf_rep_eth_dev, bnxt_representor_uninit);
6033 	}
6034 	PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci remove\n",
6035 		    eth_dev->data->port_id);
6036 	ret = rte_eth_dev_destroy(eth_dev, bnxt_dev_uninit);
6037 
6038 	return ret;
6039 }
6040 
6041 static void bnxt_free_rep_info(struct bnxt *bp)
6042 {
6043 	rte_free(bp->rep_info);
6044 	bp->rep_info = NULL;
6045 	rte_free(bp->cfa_code_map);
6046 	bp->cfa_code_map = NULL;
6047 }
6048 
6049 static int bnxt_init_rep_info(struct bnxt *bp)
6050 {
6051 	int i = 0, rc;
6052 
6053 	if (bp->rep_info)
6054 		return 0;
6055 
6056 	bp->rep_info = rte_zmalloc("bnxt_rep_info",
6057 				   sizeof(bp->rep_info[0]) * BNXT_MAX_VF_REPS(bp),
6058 				   0);
6059 	if (!bp->rep_info) {
6060 		PMD_DRV_LOG(ERR, "Failed to alloc memory for rep info\n");
6061 		return -ENOMEM;
6062 	}
6063 	bp->cfa_code_map = rte_zmalloc("bnxt_cfa_code_map",
6064 				       sizeof(*bp->cfa_code_map) *
6065 				       BNXT_MAX_CFA_CODE, 0);
6066 	if (!bp->cfa_code_map) {
6067 		PMD_DRV_LOG(ERR, "Failed to alloc memory for cfa_code_map\n");
6068 		bnxt_free_rep_info(bp);
6069 		return -ENOMEM;
6070 	}
6071 
6072 	for (i = 0; i < BNXT_MAX_CFA_CODE; i++)
6073 		bp->cfa_code_map[i] = BNXT_VF_IDX_INVALID;
6074 
6075 	rc = pthread_mutex_init(&bp->rep_info->vfr_lock, NULL);
6076 	if (rc) {
6077 		PMD_DRV_LOG(ERR, "Unable to initialize vfr_lock\n");
6078 		bnxt_free_rep_info(bp);
6079 		return rc;
6080 	}
6081 
6082 	rc = pthread_mutex_init(&bp->rep_info->vfr_start_lock, NULL);
6083 	if (rc) {
6084 		PMD_DRV_LOG(ERR, "Unable to initialize vfr_start_lock\n");
6085 		bnxt_free_rep_info(bp);
6086 		return rc;
6087 	}
6088 
6089 	return rc;
6090 }
6091 
6092 static int bnxt_rep_port_probe(struct rte_pci_device *pci_dev,
6093 			       struct rte_eth_devargs *eth_da,
6094 			       struct rte_eth_dev *backing_eth_dev,
6095 			       const char *dev_args)
6096 {
6097 	struct rte_eth_dev *vf_rep_eth_dev;
6098 	char name[RTE_ETH_NAME_MAX_LEN];
6099 	struct bnxt *backing_bp = backing_eth_dev->data->dev_private;
6100 	uint16_t max_vf_reps = BNXT_MAX_VF_REPS(backing_bp);
6101 
6102 	uint16_t num_rep;
6103 	int i, ret = 0;
6104 	struct rte_kvargs *kvlist = NULL;
6105 
6106 	if (eth_da->type == RTE_ETH_REPRESENTOR_NONE)
6107 		return 0;
6108 	if (eth_da->type != RTE_ETH_REPRESENTOR_VF) {
6109 		PMD_DRV_LOG(ERR, "unsupported representor type %d\n",
6110 			    eth_da->type);
6111 		return -ENOTSUP;
6112 	}
6113 	num_rep = eth_da->nb_representor_ports;
6114 	if (num_rep > max_vf_reps) {
6115 		PMD_DRV_LOG(ERR, "nb_representor_ports = %d > %d MAX VF REPS\n",
6116 			    num_rep, max_vf_reps);
6117 		return -EINVAL;
6118 	}
6119 
6120 	if (num_rep >= RTE_MAX_ETHPORTS) {
6121 		PMD_DRV_LOG(ERR,
6122 			    "nb_representor_ports = %d > %d MAX ETHPORTS\n",
6123 			    num_rep, RTE_MAX_ETHPORTS);
6124 		return -EINVAL;
6125 	}
6126 
6127 	if (!(BNXT_PF(backing_bp) || BNXT_VF_IS_TRUSTED(backing_bp))) {
6128 		PMD_DRV_LOG(ERR,
6129 			    "Not a PF or trusted VF. No Representor support\n");
6130 		/* Returning an error is not an option.
6131 		 * Applications are not handling this correctly
6132 		 */
6133 		return 0;
6134 	}
6135 
6136 	if (bnxt_init_rep_info(backing_bp))
6137 		return 0;
6138 
6139 	for (i = 0; i < num_rep; i++) {
6140 		struct bnxt_representor representor = {
6141 			.vf_id = eth_da->representor_ports[i],
6142 			.switch_domain_id = backing_bp->switch_domain_id,
6143 			.parent_dev = backing_eth_dev
6144 		};
6145 
6146 		if (representor.vf_id >= max_vf_reps) {
6147 			PMD_DRV_LOG(ERR, "VF-Rep id %d >= %d MAX VF ID\n",
6148 				    representor.vf_id, max_vf_reps);
6149 			continue;
6150 		}
6151 
6152 		/* representor port net_bdf_port */
6153 		snprintf(name, sizeof(name), "net_%s_representor_%d",
6154 			 pci_dev->device.name, eth_da->representor_ports[i]);
6155 
6156 		kvlist = rte_kvargs_parse(dev_args, bnxt_dev_args);
6157 		if (kvlist) {
6158 			/*
6159 			 * Handler for "rep_is_pf" devarg.
6160 			 * Invoked as for ex: "-a 000:00:0d.0,
6161 			 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6162 			 */
6163 			ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_IS_PF,
6164 						 bnxt_parse_devarg_rep_is_pf,
6165 						 (void *)&representor);
6166 			if (ret) {
6167 				ret = -EINVAL;
6168 				goto err;
6169 			}
6170 			/*
6171 			 * Handler for "rep_based_pf" devarg.
6172 			 * Invoked as for ex: "-a 000:00:0d.0,
6173 			 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6174 			 */
6175 			ret = rte_kvargs_process(kvlist,
6176 						 BNXT_DEVARG_REP_BASED_PF,
6177 						 bnxt_parse_devarg_rep_based_pf,
6178 						 (void *)&representor);
6179 			if (ret) {
6180 				ret = -EINVAL;
6181 				goto err;
6182 			}
6183 			/*
6184 			 * Handler for "rep_based_pf" devarg.
6185 			 * Invoked as for ex: "-a 000:00:0d.0,
6186 			 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6187 			 */
6188 			ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_Q_R2F,
6189 						 bnxt_parse_devarg_rep_q_r2f,
6190 						 (void *)&representor);
6191 			if (ret) {
6192 				ret = -EINVAL;
6193 				goto err;
6194 			}
6195 			/*
6196 			 * Handler for "rep_based_pf" devarg.
6197 			 * Invoked as for ex: "-a 000:00:0d.0,
6198 			 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6199 			 */
6200 			ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_Q_F2R,
6201 						 bnxt_parse_devarg_rep_q_f2r,
6202 						 (void *)&representor);
6203 			if (ret) {
6204 				ret = -EINVAL;
6205 				goto err;
6206 			}
6207 			/*
6208 			 * Handler for "rep_based_pf" devarg.
6209 			 * Invoked as for ex: "-a 000:00:0d.0,
6210 			 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6211 			 */
6212 			ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_FC_R2F,
6213 						 bnxt_parse_devarg_rep_fc_r2f,
6214 						 (void *)&representor);
6215 			if (ret) {
6216 				ret = -EINVAL;
6217 				goto err;
6218 			}
6219 			/*
6220 			 * Handler for "rep_based_pf" devarg.
6221 			 * Invoked as for ex: "-a 000:00:0d.0,
6222 			 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6223 			 */
6224 			ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_FC_F2R,
6225 						 bnxt_parse_devarg_rep_fc_f2r,
6226 						 (void *)&representor);
6227 			if (ret) {
6228 				ret = -EINVAL;
6229 				goto err;
6230 			}
6231 		}
6232 
6233 		ret = rte_eth_dev_create(&pci_dev->device, name,
6234 					 sizeof(struct bnxt_representor),
6235 					 NULL, NULL,
6236 					 bnxt_representor_init,
6237 					 &representor);
6238 		if (ret) {
6239 			PMD_DRV_LOG(ERR, "failed to create bnxt vf "
6240 				    "representor %s.", name);
6241 			goto err;
6242 		}
6243 
6244 		vf_rep_eth_dev = rte_eth_dev_allocated(name);
6245 		if (!vf_rep_eth_dev) {
6246 			PMD_DRV_LOG(ERR, "Failed to find the eth_dev"
6247 				    " for VF-Rep: %s.", name);
6248 			ret = -ENODEV;
6249 			goto err;
6250 		}
6251 
6252 		PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR pci probe\n",
6253 			    backing_eth_dev->data->port_id);
6254 		backing_bp->rep_info[representor.vf_id].vfr_eth_dev =
6255 							 vf_rep_eth_dev;
6256 		backing_bp->num_reps++;
6257 
6258 	}
6259 
6260 	rte_kvargs_free(kvlist);
6261 	return 0;
6262 
6263 err:
6264 	/* If num_rep > 1, then rollback already created
6265 	 * ports, since we'll be failing the probe anyway
6266 	 */
6267 	if (num_rep > 1)
6268 		bnxt_pci_remove_dev_with_reps(backing_eth_dev);
6269 	rte_errno = -ret;
6270 	rte_kvargs_free(kvlist);
6271 
6272 	return ret;
6273 }
6274 
6275 static int bnxt_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
6276 			  struct rte_pci_device *pci_dev)
6277 {
6278 	struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
6279 	struct rte_eth_dev *backing_eth_dev;
6280 	uint16_t num_rep;
6281 	int ret = 0;
6282 
6283 	if (pci_dev->device.devargs) {
6284 		ret = rte_eth_devargs_parse(pci_dev->device.devargs->args,
6285 					    &eth_da);
6286 		if (ret)
6287 			return ret;
6288 	}
6289 
6290 	num_rep = eth_da.nb_representor_ports;
6291 	PMD_DRV_LOG(DEBUG, "nb_representor_ports = %d\n",
6292 		    num_rep);
6293 
6294 	/* We could come here after first level of probe is already invoked
6295 	 * as part of an application bringup(OVS-DPDK vswitchd), so first check
6296 	 * for already allocated eth_dev for the backing device (PF/Trusted VF)
6297 	 */
6298 	backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6299 	if (backing_eth_dev == NULL) {
6300 		ret = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
6301 					 sizeof(struct bnxt),
6302 					 eth_dev_pci_specific_init, pci_dev,
6303 					 bnxt_dev_init, NULL);
6304 
6305 		if (ret || !num_rep)
6306 			return ret;
6307 
6308 		backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6309 	}
6310 	PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci probe\n",
6311 		    backing_eth_dev->data->port_id);
6312 
6313 	if (!num_rep)
6314 		return ret;
6315 
6316 	/* probe representor ports now */
6317 	ret = bnxt_rep_port_probe(pci_dev, &eth_da, backing_eth_dev,
6318 				  pci_dev->device.devargs->args);
6319 
6320 	return ret;
6321 }
6322 
6323 static int bnxt_pci_remove(struct rte_pci_device *pci_dev)
6324 {
6325 	struct rte_eth_dev *eth_dev;
6326 
6327 	eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6328 	if (!eth_dev)
6329 		return 0; /* Invoked typically only by OVS-DPDK, by the
6330 			   * time it comes here the eth_dev is already
6331 			   * deleted by rte_eth_dev_close(), so returning
6332 			   * +ve value will at least help in proper cleanup
6333 			   */
6334 
6335 	PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci remove\n", eth_dev->data->port_id);
6336 	if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
6337 		if (eth_dev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
6338 			return rte_eth_dev_destroy(eth_dev,
6339 						   bnxt_representor_uninit);
6340 		else
6341 			return rte_eth_dev_destroy(eth_dev,
6342 						   bnxt_dev_uninit);
6343 	} else {
6344 		return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
6345 	}
6346 }
6347 
6348 static struct rte_pci_driver bnxt_rte_pmd = {
6349 	.id_table = bnxt_pci_id_map,
6350 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
6351 			RTE_PCI_DRV_INTR_RMV |
6352 			RTE_PCI_DRV_PROBE_AGAIN, /* Needed in case of VF-REPs
6353 						  * and OVS-DPDK
6354 						  */
6355 	.probe = bnxt_pci_probe,
6356 	.remove = bnxt_pci_remove,
6357 };
6358 
6359 static bool
6360 is_device_supported(struct rte_eth_dev *dev, struct rte_pci_driver *drv)
6361 {
6362 	if (strcmp(dev->device->driver->name, drv->driver.name))
6363 		return false;
6364 
6365 	return true;
6366 }
6367 
6368 bool is_bnxt_supported(struct rte_eth_dev *dev)
6369 {
6370 	return is_device_supported(dev, &bnxt_rte_pmd);
6371 }
6372 
6373 RTE_LOG_REGISTER_SUFFIX(bnxt_logtype_driver, driver, NOTICE);
6374 RTE_PMD_REGISTER_PCI(net_bnxt, bnxt_rte_pmd);
6375 RTE_PMD_REGISTER_PCI_TABLE(net_bnxt, bnxt_pci_id_map);
6376 RTE_PMD_REGISTER_KMOD_DEP(net_bnxt, "* igb_uio | uio_pci_generic | vfio-pci");
6377