1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2014-2021 Broadcom
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>
19 #include "bnxt_filter.h"
20 #include "bnxt_hwrm.h"
22 #include "bnxt_reps.h"
23 #include "bnxt_ring.h"
26 #include "bnxt_stats.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"
36 #define DRV_MODULE_NAME "bnxt"
37 static const char bnxt_version[] =
38 "Broadcom NetXtreme driver " DRV_MODULE_NAME;
41 * The set of PCI devices this driver supports
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 */ },
90 #define BNXT_DEVARG_TRUFLOW "host-based-truflow"
91 #define BNXT_DEVARG_FLOW_XSTAT "flow-xstat"
92 #define BNXT_DEVARG_MAX_NUM_KFLOWS "max-num-kflows"
93 #define BNXT_DEVARG_REPRESENTOR "representor"
94 #define BNXT_DEVARG_REP_BASED_PF "rep-based-pf"
95 #define BNXT_DEVARG_REP_IS_PF "rep-is-pf"
96 #define BNXT_DEVARG_REP_Q_R2F "rep-q-r2f"
97 #define BNXT_DEVARG_REP_Q_F2R "rep-q-f2r"
98 #define BNXT_DEVARG_REP_FC_R2F "rep-fc-r2f"
99 #define BNXT_DEVARG_REP_FC_F2R "rep-fc-f2r"
101 static const char *const bnxt_dev_args[] = {
102 BNXT_DEVARG_REPRESENTOR,
104 BNXT_DEVARG_FLOW_XSTAT,
105 BNXT_DEVARG_MAX_NUM_KFLOWS,
106 BNXT_DEVARG_REP_BASED_PF,
107 BNXT_DEVARG_REP_IS_PF,
108 BNXT_DEVARG_REP_Q_R2F,
109 BNXT_DEVARG_REP_Q_F2R,
110 BNXT_DEVARG_REP_FC_R2F,
111 BNXT_DEVARG_REP_FC_F2R,
116 * truflow == false to disable the feature
117 * truflow == true to enable the feature
119 #define BNXT_DEVARG_TRUFLOW_INVALID(truflow) ((truflow) > 1)
122 * flow_xstat == false to disable the feature
123 * flow_xstat == true to enable the feature
125 #define BNXT_DEVARG_FLOW_XSTAT_INVALID(flow_xstat) ((flow_xstat) > 1)
128 * rep_is_pf == false to indicate VF representor
129 * rep_is_pf == true to indicate PF representor
131 #define BNXT_DEVARG_REP_IS_PF_INVALID(rep_is_pf) ((rep_is_pf) > 1)
134 * rep_based_pf == Physical index of the PF
136 #define BNXT_DEVARG_REP_BASED_PF_INVALID(rep_based_pf) ((rep_based_pf) > 15)
138 * rep_q_r2f == Logical COS Queue index for the rep to endpoint direction
140 #define BNXT_DEVARG_REP_Q_R2F_INVALID(rep_q_r2f) ((rep_q_r2f) > 3)
143 * rep_q_f2r == Logical COS Queue index for the endpoint to rep direction
145 #define BNXT_DEVARG_REP_Q_F2R_INVALID(rep_q_f2r) ((rep_q_f2r) > 3)
148 * rep_fc_r2f == Flow control for the representor to endpoint direction
150 #define BNXT_DEVARG_REP_FC_R2F_INVALID(rep_fc_r2f) ((rep_fc_r2f) > 1)
153 * rep_fc_f2r == Flow control for the endpoint to representor direction
155 #define BNXT_DEVARG_REP_FC_F2R_INVALID(rep_fc_f2r) ((rep_fc_f2r) > 1)
157 int bnxt_cfa_code_dynfield_offset = -1;
160 * max_num_kflows must be >= 32
161 * and must be a power-of-2 supported value
162 * return: 1 -> invalid
165 static int bnxt_devarg_max_num_kflow_invalid(uint16_t max_num_kflows)
167 if (max_num_kflows < 32 || !rte_is_power_of_2(max_num_kflows))
172 static int bnxt_vlan_offload_set_op(struct rte_eth_dev *dev, int mask);
173 static int bnxt_dev_uninit(struct rte_eth_dev *eth_dev);
174 static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev);
175 static int bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev);
176 static void bnxt_cancel_fw_health_check(struct bnxt *bp);
177 static int bnxt_restore_vlan_filters(struct bnxt *bp);
178 static void bnxt_dev_recover(void *arg);
179 static void bnxt_free_error_recovery_info(struct bnxt *bp);
180 static void bnxt_free_rep_info(struct bnxt *bp);
182 int is_bnxt_in_error(struct bnxt *bp)
184 if (bp->flags & BNXT_FLAG_FATAL_ERROR)
186 if (bp->flags & BNXT_FLAG_FW_RESET)
192 /***********************/
195 * High level utility functions
198 static uint16_t bnxt_rss_ctxts(const struct bnxt *bp)
200 unsigned int num_rss_rings = RTE_MIN(bp->rx_nr_rings,
201 BNXT_RSS_TBL_SIZE_P5);
203 if (!BNXT_CHIP_P5(bp))
206 return RTE_ALIGN_MUL_CEIL(num_rss_rings,
207 BNXT_RSS_ENTRIES_PER_CTX_P5) /
208 BNXT_RSS_ENTRIES_PER_CTX_P5;
211 uint16_t bnxt_rss_hash_tbl_size(const struct bnxt *bp)
213 if (!BNXT_CHIP_P5(bp))
214 return HW_HASH_INDEX_SIZE;
216 return bnxt_rss_ctxts(bp) * BNXT_RSS_ENTRIES_PER_CTX_P5;
219 static void bnxt_free_parent_info(struct bnxt *bp)
221 rte_free(bp->parent);
224 static void bnxt_free_pf_info(struct bnxt *bp)
229 static void bnxt_free_link_info(struct bnxt *bp)
231 rte_free(bp->link_info);
234 static void bnxt_free_leds_info(struct bnxt *bp)
243 static void bnxt_free_flow_stats_info(struct bnxt *bp)
245 rte_free(bp->flow_stat);
246 bp->flow_stat = NULL;
249 static void bnxt_free_cos_queues(struct bnxt *bp)
251 rte_free(bp->rx_cos_queue);
252 rte_free(bp->tx_cos_queue);
255 static void bnxt_free_mem(struct bnxt *bp, bool reconfig)
257 bnxt_free_filter_mem(bp);
258 bnxt_free_vnic_attributes(bp);
259 bnxt_free_vnic_mem(bp);
261 /* tx/rx rings are configured as part of *_queue_setup callbacks.
262 * If the number of rings change across fw update,
263 * we don't have much choice except to warn the user.
267 bnxt_free_tx_rings(bp);
268 bnxt_free_rx_rings(bp);
270 bnxt_free_async_cp_ring(bp);
271 bnxt_free_rxtx_nq_ring(bp);
273 rte_free(bp->grp_info);
277 static int bnxt_alloc_parent_info(struct bnxt *bp)
279 bp->parent = rte_zmalloc("bnxt_parent_info",
280 sizeof(struct bnxt_parent_info), 0);
281 if (bp->parent == NULL)
287 static int bnxt_alloc_pf_info(struct bnxt *bp)
289 bp->pf = rte_zmalloc("bnxt_pf_info", sizeof(struct bnxt_pf_info), 0);
296 static int bnxt_alloc_link_info(struct bnxt *bp)
299 rte_zmalloc("bnxt_link_info", sizeof(struct bnxt_link_info), 0);
300 if (bp->link_info == NULL)
306 static int bnxt_alloc_leds_info(struct bnxt *bp)
311 bp->leds = rte_zmalloc("bnxt_leds",
312 BNXT_MAX_LED * sizeof(struct bnxt_led_info),
314 if (bp->leds == NULL)
320 static int bnxt_alloc_cos_queues(struct bnxt *bp)
323 rte_zmalloc("bnxt_rx_cosq",
324 BNXT_COS_QUEUE_COUNT *
325 sizeof(struct bnxt_cos_queue_info),
327 if (bp->rx_cos_queue == NULL)
331 rte_zmalloc("bnxt_tx_cosq",
332 BNXT_COS_QUEUE_COUNT *
333 sizeof(struct bnxt_cos_queue_info),
335 if (bp->tx_cos_queue == NULL)
341 static int bnxt_alloc_flow_stats_info(struct bnxt *bp)
343 bp->flow_stat = rte_zmalloc("bnxt_flow_xstat",
344 sizeof(struct bnxt_flow_stat_info), 0);
345 if (bp->flow_stat == NULL)
351 static int bnxt_alloc_mem(struct bnxt *bp, bool reconfig)
355 rc = bnxt_alloc_ring_grps(bp);
359 rc = bnxt_alloc_async_ring_struct(bp);
363 rc = bnxt_alloc_vnic_mem(bp);
367 rc = bnxt_alloc_vnic_attributes(bp);
371 rc = bnxt_alloc_filter_mem(bp);
375 rc = bnxt_alloc_async_cp_ring(bp);
379 rc = bnxt_alloc_rxtx_nq_ring(bp);
383 if (BNXT_FLOW_XSTATS_EN(bp)) {
384 rc = bnxt_alloc_flow_stats_info(bp);
392 bnxt_free_mem(bp, reconfig);
396 static int bnxt_setup_one_vnic(struct bnxt *bp, uint16_t vnic_id)
398 struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
399 struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
400 uint64_t rx_offloads = dev_conf->rxmode.offloads;
401 struct bnxt_rx_queue *rxq;
405 rc = bnxt_vnic_grp_alloc(bp, vnic);
409 PMD_DRV_LOG(DEBUG, "vnic[%d] = %p vnic->fw_grp_ids = %p\n",
410 vnic_id, vnic, vnic->fw_grp_ids);
412 rc = bnxt_hwrm_vnic_alloc(bp, vnic);
416 /* Alloc RSS context only if RSS mode is enabled */
417 if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS) {
418 int j, nr_ctxs = bnxt_rss_ctxts(bp);
420 if (bp->rx_nr_rings > BNXT_RSS_TBL_SIZE_P5) {
421 PMD_DRV_LOG(ERR, "RxQ cnt %d > reta_size %d\n",
422 bp->rx_nr_rings, BNXT_RSS_TBL_SIZE_P5);
424 "Only queues 0-%d will be in RSS table\n",
425 BNXT_RSS_TBL_SIZE_P5 - 1);
429 for (j = 0; j < nr_ctxs; j++) {
430 rc = bnxt_hwrm_vnic_ctx_alloc(bp, vnic, j);
436 "HWRM vnic %d ctx %d alloc failure rc: %x\n",
440 vnic->num_lb_ctxts = nr_ctxs;
444 * Firmware sets pf pair in default vnic cfg. If the VLAN strip
445 * setting is not available at this time, it will not be
446 * configured correctly in the CFA.
448 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
449 vnic->vlan_strip = true;
451 vnic->vlan_strip = false;
453 rc = bnxt_hwrm_vnic_cfg(bp, vnic);
457 rc = bnxt_set_hwrm_vnic_filters(bp, vnic);
461 for (j = 0; j < bp->rx_num_qs_per_vnic; j++) {
462 rxq = bp->eth_dev->data->rx_queues[j];
465 "rxq[%d]->vnic=%p vnic->fw_grp_ids=%p\n",
466 j, rxq->vnic, rxq->vnic->fw_grp_ids);
468 if (BNXT_HAS_RING_GRPS(bp) && rxq->rx_deferred_start)
469 rxq->vnic->fw_grp_ids[j] = INVALID_HW_RING_ID;
471 vnic->rx_queue_cnt++;
474 PMD_DRV_LOG(DEBUG, "vnic->rx_queue_cnt = %d\n", vnic->rx_queue_cnt);
476 rc = bnxt_vnic_rss_configure(bp, vnic);
480 bnxt_hwrm_vnic_plcmode_cfg(bp, vnic);
482 if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
483 bnxt_hwrm_vnic_tpa_cfg(bp, vnic, 1);
485 bnxt_hwrm_vnic_tpa_cfg(bp, vnic, 0);
489 PMD_DRV_LOG(ERR, "HWRM vnic %d cfg failure rc: %x\n",
494 static int bnxt_register_fc_ctx_mem(struct bnxt *bp)
498 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->rx_fc_in_tbl.dma,
499 &bp->flow_stat->rx_fc_in_tbl.ctx_id);
504 "rx_fc_in_tbl.va = %p rx_fc_in_tbl.dma = %p"
505 " rx_fc_in_tbl.ctx_id = %d\n",
506 bp->flow_stat->rx_fc_in_tbl.va,
507 (void *)((uintptr_t)bp->flow_stat->rx_fc_in_tbl.dma),
508 bp->flow_stat->rx_fc_in_tbl.ctx_id);
510 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->rx_fc_out_tbl.dma,
511 &bp->flow_stat->rx_fc_out_tbl.ctx_id);
516 "rx_fc_out_tbl.va = %p rx_fc_out_tbl.dma = %p"
517 " rx_fc_out_tbl.ctx_id = %d\n",
518 bp->flow_stat->rx_fc_out_tbl.va,
519 (void *)((uintptr_t)bp->flow_stat->rx_fc_out_tbl.dma),
520 bp->flow_stat->rx_fc_out_tbl.ctx_id);
522 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->tx_fc_in_tbl.dma,
523 &bp->flow_stat->tx_fc_in_tbl.ctx_id);
528 "tx_fc_in_tbl.va = %p tx_fc_in_tbl.dma = %p"
529 " tx_fc_in_tbl.ctx_id = %d\n",
530 bp->flow_stat->tx_fc_in_tbl.va,
531 (void *)((uintptr_t)bp->flow_stat->tx_fc_in_tbl.dma),
532 bp->flow_stat->tx_fc_in_tbl.ctx_id);
534 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->tx_fc_out_tbl.dma,
535 &bp->flow_stat->tx_fc_out_tbl.ctx_id);
540 "tx_fc_out_tbl.va = %p tx_fc_out_tbl.dma = %p"
541 " tx_fc_out_tbl.ctx_id = %d\n",
542 bp->flow_stat->tx_fc_out_tbl.va,
543 (void *)((uintptr_t)bp->flow_stat->tx_fc_out_tbl.dma),
544 bp->flow_stat->tx_fc_out_tbl.ctx_id);
546 memset(bp->flow_stat->rx_fc_out_tbl.va,
548 bp->flow_stat->rx_fc_out_tbl.size);
549 rc = bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_RX,
550 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
551 bp->flow_stat->rx_fc_out_tbl.ctx_id,
552 bp->flow_stat->max_fc,
557 memset(bp->flow_stat->tx_fc_out_tbl.va,
559 bp->flow_stat->tx_fc_out_tbl.size);
560 rc = bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_TX,
561 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
562 bp->flow_stat->tx_fc_out_tbl.ctx_id,
563 bp->flow_stat->max_fc,
569 static int bnxt_alloc_ctx_mem_buf(char *type, size_t size,
570 struct bnxt_ctx_mem_buf_info *ctx)
575 ctx->va = rte_zmalloc(type, size, 0);
578 rte_mem_lock_page(ctx->va);
580 ctx->dma = rte_mem_virt2iova(ctx->va);
581 if (ctx->dma == RTE_BAD_IOVA)
587 static int bnxt_init_fc_ctx_mem(struct bnxt *bp)
589 struct rte_pci_device *pdev = bp->pdev;
590 char type[RTE_MEMZONE_NAMESIZE];
594 max_fc = bp->flow_stat->max_fc;
596 sprintf(type, "bnxt_rx_fc_in_" PCI_PRI_FMT, pdev->addr.domain,
597 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
598 /* 4 bytes for each counter-id */
599 rc = bnxt_alloc_ctx_mem_buf(type,
601 &bp->flow_stat->rx_fc_in_tbl);
605 sprintf(type, "bnxt_rx_fc_out_" PCI_PRI_FMT, pdev->addr.domain,
606 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
607 /* 16 bytes for each counter - 8 bytes pkt_count, 8 bytes byte_count */
608 rc = bnxt_alloc_ctx_mem_buf(type,
610 &bp->flow_stat->rx_fc_out_tbl);
614 sprintf(type, "bnxt_tx_fc_in_" PCI_PRI_FMT, pdev->addr.domain,
615 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
616 /* 4 bytes for each counter-id */
617 rc = bnxt_alloc_ctx_mem_buf(type,
619 &bp->flow_stat->tx_fc_in_tbl);
623 sprintf(type, "bnxt_tx_fc_out_" PCI_PRI_FMT, pdev->addr.domain,
624 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
625 /* 16 bytes for each counter - 8 bytes pkt_count, 8 bytes byte_count */
626 rc = bnxt_alloc_ctx_mem_buf(type,
628 &bp->flow_stat->tx_fc_out_tbl);
632 rc = bnxt_register_fc_ctx_mem(bp);
637 static int bnxt_init_ctx_mem(struct bnxt *bp)
641 if (!(bp->fw_cap & BNXT_FW_CAP_ADV_FLOW_COUNTERS) ||
642 !(BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) ||
643 !BNXT_FLOW_XSTATS_EN(bp))
646 rc = bnxt_hwrm_cfa_counter_qcaps(bp, &bp->flow_stat->max_fc);
650 rc = bnxt_init_fc_ctx_mem(bp);
655 static int bnxt_update_phy_setting(struct bnxt *bp)
657 struct rte_eth_link new;
660 rc = bnxt_get_hwrm_link_config(bp, &new);
662 PMD_DRV_LOG(ERR, "Failed to get link settings\n");
667 * On BCM957508-N2100 adapters, FW will not allow any user other
668 * than BMC to shutdown the port. bnxt_get_hwrm_link_config() call
669 * always returns link up. Force phy update always in that case.
671 if (!new.link_status || IS_BNXT_DEV_957508_N2100(bp)) {
672 rc = bnxt_set_hwrm_link_config(bp, true);
674 PMD_DRV_LOG(ERR, "Failed to update PHY settings\n");
682 static int bnxt_start_nic(struct bnxt *bp)
684 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(bp->eth_dev);
685 struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
686 uint32_t intr_vector = 0;
687 uint32_t queue_id, base = BNXT_MISC_VEC_ID;
688 uint32_t vec = BNXT_MISC_VEC_ID;
692 if (bp->eth_dev->data->mtu > RTE_ETHER_MTU) {
693 bp->eth_dev->data->dev_conf.rxmode.offloads |=
694 DEV_RX_OFFLOAD_JUMBO_FRAME;
695 bp->flags |= BNXT_FLAG_JUMBO;
697 bp->eth_dev->data->dev_conf.rxmode.offloads &=
698 ~DEV_RX_OFFLOAD_JUMBO_FRAME;
699 bp->flags &= ~BNXT_FLAG_JUMBO;
702 /* THOR does not support ring groups.
703 * But we will use the array to save RSS context IDs.
705 if (BNXT_CHIP_P5(bp))
706 bp->max_ring_grps = BNXT_MAX_RSS_CTXTS_P5;
708 rc = bnxt_alloc_all_hwrm_stat_ctxs(bp);
710 PMD_DRV_LOG(ERR, "HWRM stat ctx alloc failure rc: %x\n", rc);
714 rc = bnxt_alloc_hwrm_rings(bp);
716 PMD_DRV_LOG(ERR, "HWRM ring alloc failure rc: %x\n", rc);
720 rc = bnxt_alloc_all_hwrm_ring_grps(bp);
722 PMD_DRV_LOG(ERR, "HWRM ring grp alloc failure: %x\n", rc);
726 if (!(bp->vnic_cap_flags & BNXT_VNIC_CAP_COS_CLASSIFY))
729 for (j = 0, i = 0; i < BNXT_COS_QUEUE_COUNT; i++) {
730 if (bp->rx_cos_queue[i].id != 0xff) {
731 struct bnxt_vnic_info *vnic = &bp->vnic_info[j++];
735 "Num pools more than FW profile\n");
739 vnic->cos_queue_id = bp->rx_cos_queue[i].id;
745 rc = bnxt_mq_rx_configure(bp);
747 PMD_DRV_LOG(ERR, "MQ mode configure failure rc: %x\n", rc);
752 rc = bnxt_setup_one_vnic(bp, 0);
755 /* VNIC configuration */
756 if (BNXT_RFS_NEEDS_VNIC(bp)) {
757 for (i = 1; i < bp->nr_vnics; i++) {
758 rc = bnxt_setup_one_vnic(bp, i);
764 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, &bp->vnic_info[0], 0, NULL);
767 "HWRM cfa l2 rx mask failure rc: %x\n", rc);
771 /* check and configure queue intr-vector mapping */
772 if ((rte_intr_cap_multiple(intr_handle) ||
773 !RTE_ETH_DEV_SRIOV(bp->eth_dev).active) &&
774 bp->eth_dev->data->dev_conf.intr_conf.rxq != 0) {
775 intr_vector = bp->eth_dev->data->nb_rx_queues;
776 PMD_DRV_LOG(DEBUG, "intr_vector = %d\n", intr_vector);
777 if (intr_vector > bp->rx_cp_nr_rings) {
778 PMD_DRV_LOG(ERR, "At most %d intr queues supported",
782 rc = rte_intr_efd_enable(intr_handle, intr_vector);
787 if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
788 intr_handle->intr_vec =
789 rte_zmalloc("intr_vec",
790 bp->eth_dev->data->nb_rx_queues *
792 if (intr_handle->intr_vec == NULL) {
793 PMD_DRV_LOG(ERR, "Failed to allocate %d rx_queues"
794 " intr_vec", bp->eth_dev->data->nb_rx_queues);
798 PMD_DRV_LOG(DEBUG, "intr_handle->intr_vec = %p "
799 "intr_handle->nb_efd = %d intr_handle->max_intr = %d\n",
800 intr_handle->intr_vec, intr_handle->nb_efd,
801 intr_handle->max_intr);
802 for (queue_id = 0; queue_id < bp->eth_dev->data->nb_rx_queues;
804 intr_handle->intr_vec[queue_id] =
805 vec + BNXT_RX_VEC_START;
806 if (vec < base + intr_handle->nb_efd - 1)
811 /* enable uio/vfio intr/eventfd mapping */
812 rc = rte_intr_enable(intr_handle);
813 #ifndef RTE_EXEC_ENV_FREEBSD
814 /* In FreeBSD OS, nic_uio driver does not support interrupts */
819 rc = bnxt_update_phy_setting(bp);
823 bp->mark_table = rte_zmalloc("bnxt_mark_table", BNXT_MARK_TABLE_SZ, 0);
825 PMD_DRV_LOG(ERR, "Allocation of mark table failed\n");
830 /* Some of the error status returned by FW may not be from errno.h */
837 static int bnxt_shutdown_nic(struct bnxt *bp)
839 bnxt_free_all_hwrm_resources(bp);
840 bnxt_free_all_filters(bp);
841 bnxt_free_all_vnics(bp);
846 * Device configuration and status function
849 uint32_t bnxt_get_speed_capabilities(struct bnxt *bp)
851 uint32_t link_speed = bp->link_info->support_speeds;
852 uint32_t speed_capa = 0;
854 /* If PAM4 is configured, use PAM4 supported speed */
855 if (link_speed == 0 && bp->link_info->support_pam4_speeds > 0)
856 link_speed = bp->link_info->support_pam4_speeds;
858 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_100MB)
859 speed_capa |= ETH_LINK_SPEED_100M;
860 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_100MBHD)
861 speed_capa |= ETH_LINK_SPEED_100M_HD;
862 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_1GB)
863 speed_capa |= ETH_LINK_SPEED_1G;
864 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_2_5GB)
865 speed_capa |= ETH_LINK_SPEED_2_5G;
866 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_10GB)
867 speed_capa |= ETH_LINK_SPEED_10G;
868 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_20GB)
869 speed_capa |= ETH_LINK_SPEED_20G;
870 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_25GB)
871 speed_capa |= ETH_LINK_SPEED_25G;
872 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_40GB)
873 speed_capa |= ETH_LINK_SPEED_40G;
874 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_50GB)
875 speed_capa |= ETH_LINK_SPEED_50G;
876 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_100GB)
877 speed_capa |= ETH_LINK_SPEED_100G;
878 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_50G)
879 speed_capa |= ETH_LINK_SPEED_50G;
880 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_100G)
881 speed_capa |= ETH_LINK_SPEED_100G;
882 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_200G)
883 speed_capa |= ETH_LINK_SPEED_200G;
885 if (bp->link_info->auto_mode ==
886 HWRM_PORT_PHY_QCFG_OUTPUT_AUTO_MODE_NONE)
887 speed_capa |= ETH_LINK_SPEED_FIXED;
892 static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
893 struct rte_eth_dev_info *dev_info)
895 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(eth_dev->device);
896 struct bnxt *bp = eth_dev->data->dev_private;
897 uint16_t max_vnics, i, j, vpool, vrxq;
898 unsigned int max_rx_rings;
901 rc = is_bnxt_in_error(bp);
906 dev_info->max_mac_addrs = bp->max_l2_ctx;
907 dev_info->max_hash_mac_addrs = 0;
909 /* PF/VF specifics */
911 dev_info->max_vfs = pdev->max_vfs;
913 max_rx_rings = bnxt_max_rings(bp);
914 /* For the sake of symmetry, max_rx_queues = max_tx_queues */
915 dev_info->max_rx_queues = max_rx_rings;
916 dev_info->max_tx_queues = max_rx_rings;
917 dev_info->reta_size = bnxt_rss_hash_tbl_size(bp);
918 dev_info->hash_key_size = 40;
919 max_vnics = bp->max_vnics;
922 dev_info->min_mtu = RTE_ETHER_MIN_MTU;
923 dev_info->max_mtu = BNXT_MAX_MTU;
925 /* Fast path specifics */
926 dev_info->min_rx_bufsize = 1;
927 dev_info->max_rx_pktlen = BNXT_MAX_PKT_LEN;
929 dev_info->rx_offload_capa = BNXT_DEV_RX_OFFLOAD_SUPPORT;
930 if (bp->flags & BNXT_FLAG_PTP_SUPPORTED)
931 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TIMESTAMP;
932 dev_info->tx_queue_offload_capa = DEV_TX_OFFLOAD_MBUF_FAST_FREE;
933 dev_info->tx_offload_capa = BNXT_DEV_TX_OFFLOAD_SUPPORT |
934 dev_info->tx_queue_offload_capa;
935 dev_info->flow_type_rss_offloads = BNXT_ETH_RSS_SUPPORT;
937 dev_info->speed_capa = bnxt_get_speed_capabilities(bp);
940 dev_info->default_rxconf = (struct rte_eth_rxconf) {
946 .rx_free_thresh = 32,
947 .rx_drop_en = BNXT_DEFAULT_RX_DROP_EN,
950 dev_info->default_txconf = (struct rte_eth_txconf) {
956 .tx_free_thresh = 32,
959 eth_dev->data->dev_conf.intr_conf.lsc = 1;
961 eth_dev->data->dev_conf.intr_conf.rxq = 1;
962 dev_info->rx_desc_lim.nb_min = BNXT_MIN_RING_DESC;
963 dev_info->rx_desc_lim.nb_max = BNXT_MAX_RX_RING_DESC;
964 dev_info->tx_desc_lim.nb_min = BNXT_MIN_RING_DESC;
965 dev_info->tx_desc_lim.nb_max = BNXT_MAX_TX_RING_DESC;
967 if (BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) {
968 dev_info->switch_info.name = eth_dev->device->name;
969 dev_info->switch_info.domain_id = bp->switch_domain_id;
970 dev_info->switch_info.port_id =
971 BNXT_PF(bp) ? BNXT_SWITCH_PORT_ID_PF :
972 BNXT_SWITCH_PORT_ID_TRUSTED_VF;
978 * TODO: default_rxconf, default_txconf, rx_desc_lim, and tx_desc_lim
979 * need further investigation.
983 vpool = 64; /* ETH_64_POOLS */
984 vrxq = 128; /* ETH_VMDQ_DCB_NUM_QUEUES */
985 for (i = 0; i < 4; vpool >>= 1, i++) {
986 if (max_vnics > vpool) {
987 for (j = 0; j < 5; vrxq >>= 1, j++) {
988 if (dev_info->max_rx_queues > vrxq) {
994 /* Not enough resources to support VMDq */
998 /* Not enough resources to support VMDq */
1002 dev_info->max_vmdq_pools = vpool;
1003 dev_info->vmdq_queue_num = vrxq;
1005 dev_info->vmdq_pool_base = 0;
1006 dev_info->vmdq_queue_base = 0;
1011 /* Configure the device based on the configuration provided */
1012 static int bnxt_dev_configure_op(struct rte_eth_dev *eth_dev)
1014 struct bnxt *bp = eth_dev->data->dev_private;
1015 uint64_t rx_offloads = eth_dev->data->dev_conf.rxmode.offloads;
1018 bp->rx_queues = (void *)eth_dev->data->rx_queues;
1019 bp->tx_queues = (void *)eth_dev->data->tx_queues;
1020 bp->tx_nr_rings = eth_dev->data->nb_tx_queues;
1021 bp->rx_nr_rings = eth_dev->data->nb_rx_queues;
1023 rc = is_bnxt_in_error(bp);
1027 if (BNXT_VF(bp) && (bp->flags & BNXT_FLAG_NEW_RM)) {
1028 rc = bnxt_hwrm_check_vf_rings(bp);
1030 PMD_DRV_LOG(ERR, "HWRM insufficient resources\n");
1034 /* If a resource has already been allocated - in this case
1035 * it is the async completion ring, free it. Reallocate it after
1036 * resource reservation. This will ensure the resource counts
1037 * are calculated correctly.
1040 pthread_mutex_lock(&bp->def_cp_lock);
1042 if (!BNXT_HAS_NQ(bp) && bp->async_cp_ring) {
1043 bnxt_disable_int(bp);
1044 bnxt_free_cp_ring(bp, bp->async_cp_ring);
1047 rc = bnxt_hwrm_func_reserve_vf_resc(bp, false);
1049 PMD_DRV_LOG(ERR, "HWRM resource alloc fail:%x\n", rc);
1050 pthread_mutex_unlock(&bp->def_cp_lock);
1054 if (!BNXT_HAS_NQ(bp) && bp->async_cp_ring) {
1055 rc = bnxt_alloc_async_cp_ring(bp);
1057 pthread_mutex_unlock(&bp->def_cp_lock);
1060 bnxt_enable_int(bp);
1063 pthread_mutex_unlock(&bp->def_cp_lock);
1066 /* Inherit new configurations */
1067 if (eth_dev->data->nb_rx_queues > bp->max_rx_rings ||
1068 eth_dev->data->nb_tx_queues > bp->max_tx_rings ||
1069 eth_dev->data->nb_rx_queues + eth_dev->data->nb_tx_queues
1070 + BNXT_NUM_ASYNC_CPR(bp) > bp->max_cp_rings ||
1071 eth_dev->data->nb_rx_queues + eth_dev->data->nb_tx_queues >
1073 goto resource_error;
1075 if (BNXT_HAS_RING_GRPS(bp) &&
1076 (uint32_t)(eth_dev->data->nb_rx_queues) > bp->max_ring_grps)
1077 goto resource_error;
1079 if (!(eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) &&
1080 bp->max_vnics < eth_dev->data->nb_rx_queues)
1081 goto resource_error;
1083 bp->rx_cp_nr_rings = bp->rx_nr_rings;
1084 bp->tx_cp_nr_rings = bp->tx_nr_rings;
1086 if (eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
1087 rx_offloads |= DEV_RX_OFFLOAD_RSS_HASH;
1088 eth_dev->data->dev_conf.rxmode.offloads = rx_offloads;
1090 if (rx_offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) {
1091 eth_dev->data->mtu =
1092 eth_dev->data->dev_conf.rxmode.max_rx_pkt_len -
1093 RTE_ETHER_HDR_LEN - RTE_ETHER_CRC_LEN - VLAN_TAG_SIZE *
1095 bnxt_mtu_set_op(eth_dev, eth_dev->data->mtu);
1101 "Insufficient resources to support requested config\n");
1103 "Num Queues Requested: Tx %d, Rx %d\n",
1104 eth_dev->data->nb_tx_queues,
1105 eth_dev->data->nb_rx_queues);
1107 "MAX: TxQ %d, RxQ %d, CQ %d Stat %d, Grp %d, Vnic %d\n",
1108 bp->max_tx_rings, bp->max_rx_rings, bp->max_cp_rings,
1109 bp->max_stat_ctx, bp->max_ring_grps, bp->max_vnics);
1113 void bnxt_print_link_info(struct rte_eth_dev *eth_dev)
1115 struct rte_eth_link *link = ð_dev->data->dev_link;
1117 if (link->link_status)
1118 PMD_DRV_LOG(INFO, "Port %d Link Up - speed %u Mbps - %s\n",
1119 eth_dev->data->port_id,
1120 (uint32_t)link->link_speed,
1121 (link->link_duplex == ETH_LINK_FULL_DUPLEX) ?
1122 ("full-duplex") : ("half-duplex\n"));
1124 PMD_DRV_LOG(INFO, "Port %d Link Down\n",
1125 eth_dev->data->port_id);
1129 * Determine whether the current configuration requires support for scattered
1130 * receive; return 1 if scattered receive is required and 0 if not.
1132 static int bnxt_scattered_rx(struct rte_eth_dev *eth_dev)
1137 if (eth_dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_SCATTER)
1140 if (eth_dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_TCP_LRO)
1143 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
1144 struct bnxt_rx_queue *rxq = eth_dev->data->rx_queues[i];
1146 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
1147 RTE_PKTMBUF_HEADROOM);
1148 if (eth_dev->data->dev_conf.rxmode.max_rx_pkt_len > buf_size)
1154 static eth_rx_burst_t
1155 bnxt_receive_function(struct rte_eth_dev *eth_dev)
1157 struct bnxt *bp = eth_dev->data->dev_private;
1159 /* Disable vector mode RX for Stingray2 for now */
1160 if (BNXT_CHIP_SR2(bp)) {
1161 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1162 return bnxt_recv_pkts;
1165 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
1166 #ifndef RTE_LIBRTE_IEEE1588
1168 * Vector mode receive can be enabled only if scatter rx is not
1169 * in use and rx offloads are limited to VLAN stripping and
1172 if (!eth_dev->data->scattered_rx &&
1173 !(eth_dev->data->dev_conf.rxmode.offloads &
1174 ~(DEV_RX_OFFLOAD_VLAN_STRIP |
1175 DEV_RX_OFFLOAD_KEEP_CRC |
1176 DEV_RX_OFFLOAD_JUMBO_FRAME |
1177 DEV_RX_OFFLOAD_IPV4_CKSUM |
1178 DEV_RX_OFFLOAD_UDP_CKSUM |
1179 DEV_RX_OFFLOAD_TCP_CKSUM |
1180 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1181 DEV_RX_OFFLOAD_OUTER_UDP_CKSUM |
1182 DEV_RX_OFFLOAD_RSS_HASH |
1183 DEV_RX_OFFLOAD_VLAN_FILTER)) &&
1184 !BNXT_TRUFLOW_EN(bp) && BNXT_NUM_ASYNC_CPR(bp) &&
1185 rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_128) {
1186 PMD_DRV_LOG(INFO, "Using vector mode receive for port %d\n",
1187 eth_dev->data->port_id);
1188 bp->flags |= BNXT_FLAG_RX_VECTOR_PKT_MODE;
1189 return bnxt_recv_pkts_vec;
1191 PMD_DRV_LOG(INFO, "Vector mode receive disabled for port %d\n",
1192 eth_dev->data->port_id);
1194 "Port %d scatter: %d rx offload: %" PRIX64 "\n",
1195 eth_dev->data->port_id,
1196 eth_dev->data->scattered_rx,
1197 eth_dev->data->dev_conf.rxmode.offloads);
1200 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1201 return bnxt_recv_pkts;
1204 static eth_tx_burst_t
1205 bnxt_transmit_function(struct rte_eth_dev *eth_dev)
1207 struct bnxt *bp = eth_dev->data->dev_private;
1209 /* Disable vector mode TX for Stingray2 for now */
1210 if (BNXT_CHIP_SR2(bp))
1211 return bnxt_xmit_pkts;
1213 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
1214 #ifndef RTE_LIBRTE_IEEE1588
1215 uint64_t offloads = eth_dev->data->dev_conf.txmode.offloads;
1218 * Vector mode transmit can be enabled only if not using scatter rx
1221 if (!eth_dev->data->scattered_rx &&
1222 !(offloads & ~DEV_TX_OFFLOAD_MBUF_FAST_FREE) &&
1223 !BNXT_TRUFLOW_EN(bp) &&
1224 rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_128) {
1225 PMD_DRV_LOG(INFO, "Using vector mode transmit for port %d\n",
1226 eth_dev->data->port_id);
1227 return bnxt_xmit_pkts_vec;
1229 PMD_DRV_LOG(INFO, "Vector mode transmit disabled for port %d\n",
1230 eth_dev->data->port_id);
1232 "Port %d scatter: %d tx offload: %" PRIX64 "\n",
1233 eth_dev->data->port_id,
1234 eth_dev->data->scattered_rx,
1238 return bnxt_xmit_pkts;
1241 static int bnxt_handle_if_change_status(struct bnxt *bp)
1245 /* Since fw has undergone a reset and lost all contexts,
1246 * set fatal flag to not issue hwrm during cleanup
1248 bp->flags |= BNXT_FLAG_FATAL_ERROR;
1249 bnxt_uninit_resources(bp, true);
1251 /* clear fatal flag so that re-init happens */
1252 bp->flags &= ~BNXT_FLAG_FATAL_ERROR;
1253 rc = bnxt_init_resources(bp, true);
1255 bp->flags &= ~BNXT_FLAG_IF_CHANGE_HOT_FW_RESET_DONE;
1260 static int bnxt_dev_set_link_up_op(struct rte_eth_dev *eth_dev)
1262 struct bnxt *bp = eth_dev->data->dev_private;
1265 if (!BNXT_SINGLE_PF(bp))
1268 if (!bp->link_info->link_up)
1269 rc = bnxt_set_hwrm_link_config(bp, true);
1271 eth_dev->data->dev_link.link_status = 1;
1273 bnxt_print_link_info(eth_dev);
1277 static int bnxt_dev_set_link_down_op(struct rte_eth_dev *eth_dev)
1279 struct bnxt *bp = eth_dev->data->dev_private;
1281 if (!BNXT_SINGLE_PF(bp))
1284 eth_dev->data->dev_link.link_status = 0;
1285 bnxt_set_hwrm_link_config(bp, false);
1286 bp->link_info->link_up = 0;
1291 static void bnxt_free_switch_domain(struct bnxt *bp)
1295 if (bp->switch_domain_id) {
1296 rc = rte_eth_switch_domain_free(bp->switch_domain_id);
1298 PMD_DRV_LOG(ERR, "free switch domain:%d fail: %d\n",
1299 bp->switch_domain_id, rc);
1303 static void bnxt_ptp_get_current_time(void *arg)
1305 struct bnxt *bp = arg;
1306 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
1309 rc = is_bnxt_in_error(bp);
1316 bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
1317 &ptp->current_time);
1319 rc = rte_eal_alarm_set(US_PER_S, bnxt_ptp_get_current_time, (void *)bp);
1321 PMD_DRV_LOG(ERR, "Failed to re-schedule PTP alarm\n");
1322 bp->flags2 &= ~BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1326 static int bnxt_schedule_ptp_alarm(struct bnxt *bp)
1328 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
1331 if (bp->flags2 & BNXT_FLAGS2_PTP_ALARM_SCHEDULED)
1334 bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
1335 &ptp->current_time);
1337 rc = rte_eal_alarm_set(US_PER_S, bnxt_ptp_get_current_time, (void *)bp);
1341 static void bnxt_cancel_ptp_alarm(struct bnxt *bp)
1343 if (bp->flags2 & BNXT_FLAGS2_PTP_ALARM_SCHEDULED) {
1344 rte_eal_alarm_cancel(bnxt_ptp_get_current_time, (void *)bp);
1345 bp->flags2 &= ~BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1349 static void bnxt_ptp_stop(struct bnxt *bp)
1351 bnxt_cancel_ptp_alarm(bp);
1352 bp->flags2 &= ~BNXT_FLAGS2_PTP_TIMESYNC_ENABLED;
1355 static int bnxt_ptp_start(struct bnxt *bp)
1359 rc = bnxt_schedule_ptp_alarm(bp);
1361 PMD_DRV_LOG(ERR, "Failed to schedule PTP alarm\n");
1363 bp->flags2 |= BNXT_FLAGS2_PTP_TIMESYNC_ENABLED;
1364 bp->flags2 |= BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1370 static int bnxt_dev_stop(struct rte_eth_dev *eth_dev)
1372 struct bnxt *bp = eth_dev->data->dev_private;
1373 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1374 struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1375 struct rte_eth_link link;
1378 eth_dev->data->dev_started = 0;
1379 eth_dev->data->scattered_rx = 0;
1381 /* Prevent crashes when queues are still in use */
1382 eth_dev->rx_pkt_burst = &bnxt_dummy_recv_pkts;
1383 eth_dev->tx_pkt_burst = &bnxt_dummy_xmit_pkts;
1385 bnxt_disable_int(bp);
1387 /* disable uio/vfio intr/eventfd mapping */
1388 rte_intr_disable(intr_handle);
1390 /* Stop the child representors for this device */
1391 ret = bnxt_rep_stop_all(bp);
1395 /* delete the bnxt ULP port details */
1396 bnxt_ulp_port_deinit(bp);
1398 bnxt_cancel_fw_health_check(bp);
1400 if (BNXT_P5_PTP_TIMESYNC_ENABLED(bp))
1401 bnxt_cancel_ptp_alarm(bp);
1403 /* Do not bring link down during reset recovery */
1404 if (!is_bnxt_in_error(bp)) {
1405 bnxt_dev_set_link_down_op(eth_dev);
1406 /* Wait for link to be reset */
1407 if (BNXT_SINGLE_PF(bp))
1409 /* clear the recorded link status */
1410 memset(&link, 0, sizeof(link));
1411 rte_eth_linkstatus_set(eth_dev, &link);
1414 /* Clean queue intr-vector mapping */
1415 rte_intr_efd_disable(intr_handle);
1416 if (intr_handle->intr_vec != NULL) {
1417 rte_free(intr_handle->intr_vec);
1418 intr_handle->intr_vec = NULL;
1421 bnxt_hwrm_port_clr_stats(bp);
1422 bnxt_free_tx_mbufs(bp);
1423 bnxt_free_rx_mbufs(bp);
1424 /* Process any remaining notifications in default completion queue */
1425 bnxt_int_handler(eth_dev);
1426 bnxt_shutdown_nic(bp);
1427 bnxt_hwrm_if_change(bp, false);
1429 rte_free(bp->mark_table);
1430 bp->mark_table = NULL;
1432 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1433 bp->rx_cosq_cnt = 0;
1434 /* All filters are deleted on a port stop. */
1435 if (BNXT_FLOW_XSTATS_EN(bp))
1436 bp->flow_stat->flow_count = 0;
1441 /* Unload the driver, release resources */
1442 static int bnxt_dev_stop_op(struct rte_eth_dev *eth_dev)
1444 struct bnxt *bp = eth_dev->data->dev_private;
1446 pthread_mutex_lock(&bp->err_recovery_lock);
1447 if (bp->flags & BNXT_FLAG_FW_RESET) {
1449 "Adapter recovering from error..Please retry\n");
1450 pthread_mutex_unlock(&bp->err_recovery_lock);
1453 pthread_mutex_unlock(&bp->err_recovery_lock);
1455 return bnxt_dev_stop(eth_dev);
1458 static int bnxt_dev_start_op(struct rte_eth_dev *eth_dev)
1460 struct bnxt *bp = eth_dev->data->dev_private;
1461 uint64_t rx_offloads = eth_dev->data->dev_conf.rxmode.offloads;
1463 int rc, retry_cnt = BNXT_IF_CHANGE_RETRY_COUNT;
1465 if (!eth_dev->data->nb_tx_queues || !eth_dev->data->nb_rx_queues) {
1466 PMD_DRV_LOG(ERR, "Queues are not configured yet!\n");
1470 if (bp->rx_cp_nr_rings > RTE_ETHDEV_QUEUE_STAT_CNTRS)
1472 "RxQ cnt %d > RTE_ETHDEV_QUEUE_STAT_CNTRS %d\n",
1473 bp->rx_cp_nr_rings, RTE_ETHDEV_QUEUE_STAT_CNTRS);
1476 rc = bnxt_hwrm_if_change(bp, true);
1477 if (rc == 0 || rc != -EAGAIN)
1480 rte_delay_ms(BNXT_IF_CHANGE_RETRY_INTERVAL);
1481 } while (retry_cnt--);
1486 if (bp->flags & BNXT_FLAG_IF_CHANGE_HOT_FW_RESET_DONE) {
1487 rc = bnxt_handle_if_change_status(bp);
1492 bnxt_enable_int(bp);
1494 eth_dev->data->scattered_rx = bnxt_scattered_rx(eth_dev);
1496 rc = bnxt_start_nic(bp);
1500 eth_dev->data->dev_started = 1;
1502 bnxt_link_update_op(eth_dev, 1);
1504 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1505 vlan_mask |= ETH_VLAN_FILTER_MASK;
1506 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
1507 vlan_mask |= ETH_VLAN_STRIP_MASK;
1508 rc = bnxt_vlan_offload_set_op(eth_dev, vlan_mask);
1512 /* Initialize bnxt ULP port details */
1513 rc = bnxt_ulp_port_init(bp);
1517 eth_dev->rx_pkt_burst = bnxt_receive_function(eth_dev);
1518 eth_dev->tx_pkt_burst = bnxt_transmit_function(eth_dev);
1520 bnxt_schedule_fw_health_check(bp);
1522 if (BNXT_P5_PTP_TIMESYNC_ENABLED(bp))
1523 bnxt_schedule_ptp_alarm(bp);
1528 bnxt_dev_stop(eth_dev);
1533 bnxt_uninit_locks(struct bnxt *bp)
1535 pthread_mutex_destroy(&bp->flow_lock);
1536 pthread_mutex_destroy(&bp->def_cp_lock);
1537 pthread_mutex_destroy(&bp->health_check_lock);
1538 pthread_mutex_destroy(&bp->err_recovery_lock);
1540 pthread_mutex_destroy(&bp->rep_info->vfr_lock);
1541 pthread_mutex_destroy(&bp->rep_info->vfr_start_lock);
1545 static void bnxt_drv_uninit(struct bnxt *bp)
1547 bnxt_free_switch_domain(bp);
1548 bnxt_free_leds_info(bp);
1549 bnxt_free_cos_queues(bp);
1550 bnxt_free_link_info(bp);
1551 bnxt_free_pf_info(bp);
1552 bnxt_free_parent_info(bp);
1553 bnxt_uninit_locks(bp);
1555 rte_memzone_free((const struct rte_memzone *)bp->tx_mem_zone);
1556 bp->tx_mem_zone = NULL;
1557 rte_memzone_free((const struct rte_memzone *)bp->rx_mem_zone);
1558 bp->rx_mem_zone = NULL;
1560 bnxt_free_vf_info(bp);
1562 rte_free(bp->grp_info);
1563 bp->grp_info = NULL;
1566 static int bnxt_dev_close_op(struct rte_eth_dev *eth_dev)
1568 struct bnxt *bp = eth_dev->data->dev_private;
1571 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1574 pthread_mutex_lock(&bp->err_recovery_lock);
1575 if (bp->flags & BNXT_FLAG_FW_RESET) {
1577 "Adapter recovering from error...Please retry\n");
1578 pthread_mutex_unlock(&bp->err_recovery_lock);
1581 pthread_mutex_unlock(&bp->err_recovery_lock);
1583 /* cancel the recovery handler before remove dev */
1584 rte_eal_alarm_cancel(bnxt_dev_reset_and_resume, (void *)bp);
1585 rte_eal_alarm_cancel(bnxt_dev_recover, (void *)bp);
1586 bnxt_cancel_fc_thread(bp);
1588 if (eth_dev->data->dev_started)
1589 ret = bnxt_dev_stop(eth_dev);
1591 bnxt_uninit_resources(bp, false);
1593 bnxt_drv_uninit(bp);
1598 static void bnxt_mac_addr_remove_op(struct rte_eth_dev *eth_dev,
1601 struct bnxt *bp = eth_dev->data->dev_private;
1602 uint64_t pool_mask = eth_dev->data->mac_pool_sel[index];
1603 struct bnxt_vnic_info *vnic;
1604 struct bnxt_filter_info *filter, *temp_filter;
1607 if (is_bnxt_in_error(bp))
1611 * Loop through all VNICs from the specified filter flow pools to
1612 * remove the corresponding MAC addr filter
1614 for (i = 0; i < bp->nr_vnics; i++) {
1615 if (!(pool_mask & (1ULL << i)))
1618 vnic = &bp->vnic_info[i];
1619 filter = STAILQ_FIRST(&vnic->filter);
1621 temp_filter = STAILQ_NEXT(filter, next);
1622 if (filter->mac_index == index) {
1623 STAILQ_REMOVE(&vnic->filter, filter,
1624 bnxt_filter_info, next);
1625 bnxt_hwrm_clear_l2_filter(bp, filter);
1626 bnxt_free_filter(bp, filter);
1628 filter = temp_filter;
1633 static int bnxt_add_mac_filter(struct bnxt *bp, struct bnxt_vnic_info *vnic,
1634 struct rte_ether_addr *mac_addr, uint32_t index,
1637 struct bnxt_filter_info *filter;
1640 /* Attach requested MAC address to the new l2_filter */
1641 STAILQ_FOREACH(filter, &vnic->filter, next) {
1642 if (filter->mac_index == index) {
1644 "MAC addr already existed for pool %d\n",
1650 filter = bnxt_alloc_filter(bp);
1652 PMD_DRV_LOG(ERR, "L2 filter alloc failed\n");
1656 /* bnxt_alloc_filter copies default MAC to filter->l2_addr. So,
1657 * if the MAC that's been programmed now is a different one, then,
1658 * copy that addr to filter->l2_addr
1661 memcpy(filter->l2_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1662 filter->flags |= HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
1664 rc = bnxt_hwrm_set_l2_filter(bp, vnic->fw_vnic_id, filter);
1666 filter->mac_index = index;
1667 if (filter->mac_index == 0)
1668 STAILQ_INSERT_HEAD(&vnic->filter, filter, next);
1670 STAILQ_INSERT_TAIL(&vnic->filter, filter, next);
1672 bnxt_free_filter(bp, filter);
1678 static int bnxt_mac_addr_add_op(struct rte_eth_dev *eth_dev,
1679 struct rte_ether_addr *mac_addr,
1680 uint32_t index, uint32_t pool)
1682 struct bnxt *bp = eth_dev->data->dev_private;
1683 struct bnxt_vnic_info *vnic = &bp->vnic_info[pool];
1686 rc = is_bnxt_in_error(bp);
1690 if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp)) {
1691 PMD_DRV_LOG(ERR, "Cannot add MAC address to a VF interface\n");
1696 PMD_DRV_LOG(ERR, "VNIC not found for pool %d!\n", pool);
1700 /* Filter settings will get applied when port is started */
1701 if (!eth_dev->data->dev_started)
1704 rc = bnxt_add_mac_filter(bp, vnic, mac_addr, index, pool);
1709 int bnxt_link_update_op(struct rte_eth_dev *eth_dev, int wait_to_complete)
1712 struct bnxt *bp = eth_dev->data->dev_private;
1713 struct rte_eth_link new;
1714 int cnt = wait_to_complete ? BNXT_MAX_LINK_WAIT_CNT :
1715 BNXT_MIN_LINK_WAIT_CNT;
1717 rc = is_bnxt_in_error(bp);
1721 memset(&new, 0, sizeof(new));
1723 /* Retrieve link info from hardware */
1724 rc = bnxt_get_hwrm_link_config(bp, &new);
1726 new.link_speed = ETH_LINK_SPEED_100M;
1727 new.link_duplex = ETH_LINK_FULL_DUPLEX;
1729 "Failed to retrieve link rc = 0x%x!\n", rc);
1733 if (!wait_to_complete || new.link_status)
1736 rte_delay_ms(BNXT_LINK_WAIT_INTERVAL);
1739 /* Only single function PF can bring phy down.
1740 * When port is stopped, report link down for VF/MH/NPAR functions.
1742 if (!BNXT_SINGLE_PF(bp) && !eth_dev->data->dev_started)
1743 memset(&new, 0, sizeof(new));
1746 /* Timed out or success */
1747 if (new.link_status != eth_dev->data->dev_link.link_status ||
1748 new.link_speed != eth_dev->data->dev_link.link_speed) {
1749 rte_eth_linkstatus_set(eth_dev, &new);
1751 rte_eth_dev_callback_process(eth_dev,
1752 RTE_ETH_EVENT_INTR_LSC,
1755 bnxt_print_link_info(eth_dev);
1761 static int bnxt_promiscuous_enable_op(struct rte_eth_dev *eth_dev)
1763 struct bnxt *bp = eth_dev->data->dev_private;
1764 struct bnxt_vnic_info *vnic;
1768 rc = is_bnxt_in_error(bp);
1772 /* Filter settings will get applied when port is started */
1773 if (!eth_dev->data->dev_started)
1776 if (bp->vnic_info == NULL)
1779 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1781 old_flags = vnic->flags;
1782 vnic->flags |= BNXT_VNIC_INFO_PROMISC;
1783 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1785 vnic->flags = old_flags;
1790 static int bnxt_promiscuous_disable_op(struct rte_eth_dev *eth_dev)
1792 struct bnxt *bp = eth_dev->data->dev_private;
1793 struct bnxt_vnic_info *vnic;
1797 rc = is_bnxt_in_error(bp);
1801 /* Filter settings will get applied when port is started */
1802 if (!eth_dev->data->dev_started)
1805 if (bp->vnic_info == NULL)
1808 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1810 old_flags = vnic->flags;
1811 vnic->flags &= ~BNXT_VNIC_INFO_PROMISC;
1812 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1814 vnic->flags = old_flags;
1819 static int bnxt_allmulticast_enable_op(struct rte_eth_dev *eth_dev)
1821 struct bnxt *bp = eth_dev->data->dev_private;
1822 struct bnxt_vnic_info *vnic;
1826 rc = is_bnxt_in_error(bp);
1830 /* Filter settings will get applied when port is started */
1831 if (!eth_dev->data->dev_started)
1834 if (bp->vnic_info == NULL)
1837 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1839 old_flags = vnic->flags;
1840 vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
1841 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1843 vnic->flags = old_flags;
1848 static int bnxt_allmulticast_disable_op(struct rte_eth_dev *eth_dev)
1850 struct bnxt *bp = eth_dev->data->dev_private;
1851 struct bnxt_vnic_info *vnic;
1855 rc = is_bnxt_in_error(bp);
1859 /* Filter settings will get applied when port is started */
1860 if (!eth_dev->data->dev_started)
1863 if (bp->vnic_info == NULL)
1866 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1868 old_flags = vnic->flags;
1869 vnic->flags &= ~BNXT_VNIC_INFO_ALLMULTI;
1870 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1872 vnic->flags = old_flags;
1877 /* Return bnxt_rx_queue pointer corresponding to a given rxq. */
1878 static struct bnxt_rx_queue *bnxt_qid_to_rxq(struct bnxt *bp, uint16_t qid)
1880 if (qid >= bp->rx_nr_rings)
1883 return bp->eth_dev->data->rx_queues[qid];
1886 /* Return rxq corresponding to a given rss table ring/group ID. */
1887 static uint16_t bnxt_rss_to_qid(struct bnxt *bp, uint16_t fwr)
1889 struct bnxt_rx_queue *rxq;
1892 if (!BNXT_HAS_RING_GRPS(bp)) {
1893 for (i = 0; i < bp->rx_nr_rings; i++) {
1894 rxq = bp->eth_dev->data->rx_queues[i];
1895 if (rxq->rx_ring->rx_ring_struct->fw_ring_id == fwr)
1899 for (i = 0; i < bp->rx_nr_rings; i++) {
1900 if (bp->grp_info[i].fw_grp_id == fwr)
1905 return INVALID_HW_RING_ID;
1908 static int bnxt_reta_update_op(struct rte_eth_dev *eth_dev,
1909 struct rte_eth_rss_reta_entry64 *reta_conf,
1912 struct bnxt *bp = eth_dev->data->dev_private;
1913 struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
1914 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
1915 uint16_t tbl_size = bnxt_rss_hash_tbl_size(bp);
1919 rc = is_bnxt_in_error(bp);
1923 if (!vnic->rss_table)
1926 if (!(dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
1929 if (reta_size != tbl_size) {
1930 PMD_DRV_LOG(ERR, "The configured hash table lookup size "
1931 "(%d) must equal the size supported by the hardware "
1932 "(%d)\n", reta_size, tbl_size);
1936 for (i = 0; i < reta_size; i++) {
1937 struct bnxt_rx_queue *rxq;
1939 idx = i / RTE_RETA_GROUP_SIZE;
1940 sft = i % RTE_RETA_GROUP_SIZE;
1942 if (!(reta_conf[idx].mask & (1ULL << sft)))
1945 rxq = bnxt_qid_to_rxq(bp, reta_conf[idx].reta[sft]);
1947 PMD_DRV_LOG(ERR, "Invalid ring in reta_conf.\n");
1951 if (BNXT_CHIP_P5(bp)) {
1952 vnic->rss_table[i * 2] =
1953 rxq->rx_ring->rx_ring_struct->fw_ring_id;
1954 vnic->rss_table[i * 2 + 1] =
1955 rxq->cp_ring->cp_ring_struct->fw_ring_id;
1957 vnic->rss_table[i] =
1958 vnic->fw_grp_ids[reta_conf[idx].reta[sft]];
1962 rc = bnxt_hwrm_vnic_rss_cfg(bp, vnic);
1966 static int bnxt_reta_query_op(struct rte_eth_dev *eth_dev,
1967 struct rte_eth_rss_reta_entry64 *reta_conf,
1970 struct bnxt *bp = eth_dev->data->dev_private;
1971 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
1972 uint16_t tbl_size = bnxt_rss_hash_tbl_size(bp);
1973 uint16_t idx, sft, i;
1976 rc = is_bnxt_in_error(bp);
1980 /* Retrieve from the default VNIC */
1983 if (!vnic->rss_table)
1986 if (reta_size != tbl_size) {
1987 PMD_DRV_LOG(ERR, "The configured hash table lookup size "
1988 "(%d) must equal the size supported by the hardware "
1989 "(%d)\n", reta_size, tbl_size);
1993 for (idx = 0, i = 0; i < reta_size; i++) {
1994 idx = i / RTE_RETA_GROUP_SIZE;
1995 sft = i % RTE_RETA_GROUP_SIZE;
1997 if (reta_conf[idx].mask & (1ULL << sft)) {
2000 if (BNXT_CHIP_P5(bp))
2001 qid = bnxt_rss_to_qid(bp,
2002 vnic->rss_table[i * 2]);
2004 qid = bnxt_rss_to_qid(bp, vnic->rss_table[i]);
2006 if (qid == INVALID_HW_RING_ID) {
2007 PMD_DRV_LOG(ERR, "Inv. entry in rss table.\n");
2010 reta_conf[idx].reta[sft] = qid;
2017 static int bnxt_rss_hash_update_op(struct rte_eth_dev *eth_dev,
2018 struct rte_eth_rss_conf *rss_conf)
2020 struct bnxt *bp = eth_dev->data->dev_private;
2021 struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
2022 struct bnxt_vnic_info *vnic;
2025 rc = is_bnxt_in_error(bp);
2030 * If RSS enablement were different than dev_configure,
2031 * then return -EINVAL
2033 if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
2034 if (!rss_conf->rss_hf)
2035 PMD_DRV_LOG(ERR, "Hash type NONE\n");
2037 if (rss_conf->rss_hf & BNXT_ETH_RSS_SUPPORT)
2041 bp->flags |= BNXT_FLAG_UPDATE_HASH;
2042 memcpy(ð_dev->data->dev_conf.rx_adv_conf.rss_conf,
2046 /* Update the default RSS VNIC(s) */
2047 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2048 vnic->hash_type = bnxt_rte_to_hwrm_hash_types(rss_conf->rss_hf);
2050 bnxt_rte_to_hwrm_hash_level(bp, rss_conf->rss_hf,
2051 ETH_RSS_LEVEL(rss_conf->rss_hf));
2054 * If hashkey is not specified, use the previously configured
2057 if (!rss_conf->rss_key)
2060 if (rss_conf->rss_key_len != HW_HASH_KEY_SIZE) {
2062 "Invalid hashkey length, should be 16 bytes\n");
2065 memcpy(vnic->rss_hash_key, rss_conf->rss_key, rss_conf->rss_key_len);
2068 rc = bnxt_hwrm_vnic_rss_cfg(bp, vnic);
2072 static int bnxt_rss_hash_conf_get_op(struct rte_eth_dev *eth_dev,
2073 struct rte_eth_rss_conf *rss_conf)
2075 struct bnxt *bp = eth_dev->data->dev_private;
2076 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2078 uint32_t hash_types;
2080 rc = is_bnxt_in_error(bp);
2084 /* RSS configuration is the same for all VNICs */
2085 if (vnic && vnic->rss_hash_key) {
2086 if (rss_conf->rss_key) {
2087 len = rss_conf->rss_key_len <= HW_HASH_KEY_SIZE ?
2088 rss_conf->rss_key_len : HW_HASH_KEY_SIZE;
2089 memcpy(rss_conf->rss_key, vnic->rss_hash_key, len);
2092 hash_types = vnic->hash_type;
2093 rss_conf->rss_hf = 0;
2094 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV4) {
2095 rss_conf->rss_hf |= ETH_RSS_IPV4;
2096 hash_types &= ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV4;
2098 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV4) {
2099 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
2101 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV4;
2103 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV4) {
2104 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
2106 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV4;
2108 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV6) {
2109 rss_conf->rss_hf |= ETH_RSS_IPV6;
2110 hash_types &= ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV6;
2112 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV6) {
2113 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
2115 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV6;
2117 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV6) {
2118 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
2120 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV6;
2124 bnxt_hwrm_to_rte_rss_level(bp, vnic->hash_mode);
2128 "Unknown RSS config from firmware (%08x), RSS disabled",
2133 rss_conf->rss_hf = 0;
2138 static int bnxt_flow_ctrl_get_op(struct rte_eth_dev *dev,
2139 struct rte_eth_fc_conf *fc_conf)
2141 struct bnxt *bp = dev->data->dev_private;
2142 struct rte_eth_link link_info;
2145 rc = is_bnxt_in_error(bp);
2149 rc = bnxt_get_hwrm_link_config(bp, &link_info);
2153 memset(fc_conf, 0, sizeof(*fc_conf));
2154 if (bp->link_info->auto_pause)
2155 fc_conf->autoneg = 1;
2156 switch (bp->link_info->pause) {
2158 fc_conf->mode = RTE_FC_NONE;
2160 case HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_TX:
2161 fc_conf->mode = RTE_FC_TX_PAUSE;
2163 case HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_RX:
2164 fc_conf->mode = RTE_FC_RX_PAUSE;
2166 case (HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_TX |
2167 HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_RX):
2168 fc_conf->mode = RTE_FC_FULL;
2174 static int bnxt_flow_ctrl_set_op(struct rte_eth_dev *dev,
2175 struct rte_eth_fc_conf *fc_conf)
2177 struct bnxt *bp = dev->data->dev_private;
2180 rc = is_bnxt_in_error(bp);
2184 if (!BNXT_SINGLE_PF(bp) || BNXT_VF(bp)) {
2185 PMD_DRV_LOG(ERR, "Flow Control Settings cannot be modified\n");
2189 switch (fc_conf->mode) {
2191 bp->link_info->auto_pause = 0;
2192 bp->link_info->force_pause = 0;
2194 case RTE_FC_RX_PAUSE:
2195 if (fc_conf->autoneg) {
2196 bp->link_info->auto_pause =
2197 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_RX;
2198 bp->link_info->force_pause = 0;
2200 bp->link_info->auto_pause = 0;
2201 bp->link_info->force_pause =
2202 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_RX;
2205 case RTE_FC_TX_PAUSE:
2206 if (fc_conf->autoneg) {
2207 bp->link_info->auto_pause =
2208 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_TX;
2209 bp->link_info->force_pause = 0;
2211 bp->link_info->auto_pause = 0;
2212 bp->link_info->force_pause =
2213 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_TX;
2217 if (fc_conf->autoneg) {
2218 bp->link_info->auto_pause =
2219 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_TX |
2220 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_RX;
2221 bp->link_info->force_pause = 0;
2223 bp->link_info->auto_pause = 0;
2224 bp->link_info->force_pause =
2225 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_TX |
2226 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_RX;
2230 return bnxt_set_hwrm_link_config(bp, true);
2233 /* Add UDP tunneling port */
2235 bnxt_udp_tunnel_port_add_op(struct rte_eth_dev *eth_dev,
2236 struct rte_eth_udp_tunnel *udp_tunnel)
2238 struct bnxt *bp = eth_dev->data->dev_private;
2239 uint16_t tunnel_type = 0;
2242 rc = is_bnxt_in_error(bp);
2246 switch (udp_tunnel->prot_type) {
2247 case RTE_TUNNEL_TYPE_VXLAN:
2248 if (bp->vxlan_port_cnt) {
2249 PMD_DRV_LOG(ERR, "Tunnel Port %d already programmed\n",
2250 udp_tunnel->udp_port);
2251 if (bp->vxlan_port != udp_tunnel->udp_port) {
2252 PMD_DRV_LOG(ERR, "Only one port allowed\n");
2255 bp->vxlan_port_cnt++;
2259 HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_VXLAN;
2260 bp->vxlan_port_cnt++;
2262 case RTE_TUNNEL_TYPE_GENEVE:
2263 if (bp->geneve_port_cnt) {
2264 PMD_DRV_LOG(ERR, "Tunnel Port %d already programmed\n",
2265 udp_tunnel->udp_port);
2266 if (bp->geneve_port != udp_tunnel->udp_port) {
2267 PMD_DRV_LOG(ERR, "Only one port allowed\n");
2270 bp->geneve_port_cnt++;
2274 HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_GENEVE;
2275 bp->geneve_port_cnt++;
2278 PMD_DRV_LOG(ERR, "Tunnel type is not supported\n");
2281 rc = bnxt_hwrm_tunnel_dst_port_alloc(bp, udp_tunnel->udp_port,
2287 bnxt_udp_tunnel_port_del_op(struct rte_eth_dev *eth_dev,
2288 struct rte_eth_udp_tunnel *udp_tunnel)
2290 struct bnxt *bp = eth_dev->data->dev_private;
2291 uint16_t tunnel_type = 0;
2295 rc = is_bnxt_in_error(bp);
2299 switch (udp_tunnel->prot_type) {
2300 case RTE_TUNNEL_TYPE_VXLAN:
2301 if (!bp->vxlan_port_cnt) {
2302 PMD_DRV_LOG(ERR, "No Tunnel port configured yet\n");
2305 if (bp->vxlan_port != udp_tunnel->udp_port) {
2306 PMD_DRV_LOG(ERR, "Req Port: %d. Configured port: %d\n",
2307 udp_tunnel->udp_port, bp->vxlan_port);
2310 if (--bp->vxlan_port_cnt)
2314 HWRM_TUNNEL_DST_PORT_FREE_INPUT_TUNNEL_TYPE_VXLAN;
2315 port = bp->vxlan_fw_dst_port_id;
2317 case RTE_TUNNEL_TYPE_GENEVE:
2318 if (!bp->geneve_port_cnt) {
2319 PMD_DRV_LOG(ERR, "No Tunnel port configured yet\n");
2322 if (bp->geneve_port != udp_tunnel->udp_port) {
2323 PMD_DRV_LOG(ERR, "Req Port: %d. Configured port: %d\n",
2324 udp_tunnel->udp_port, bp->geneve_port);
2327 if (--bp->geneve_port_cnt)
2331 HWRM_TUNNEL_DST_PORT_FREE_INPUT_TUNNEL_TYPE_GENEVE;
2332 port = bp->geneve_fw_dst_port_id;
2335 PMD_DRV_LOG(ERR, "Tunnel type is not supported\n");
2339 rc = bnxt_hwrm_tunnel_dst_port_free(bp, port, tunnel_type);
2343 static int bnxt_del_vlan_filter(struct bnxt *bp, uint16_t vlan_id)
2345 struct bnxt_filter_info *filter;
2346 struct bnxt_vnic_info *vnic;
2348 uint32_t chk = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN;
2350 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2351 filter = STAILQ_FIRST(&vnic->filter);
2353 /* Search for this matching MAC+VLAN filter */
2354 if (bnxt_vlan_filter_exists(bp, filter, chk, vlan_id)) {
2355 /* Delete the filter */
2356 rc = bnxt_hwrm_clear_l2_filter(bp, filter);
2359 STAILQ_REMOVE(&vnic->filter, filter,
2360 bnxt_filter_info, next);
2361 bnxt_free_filter(bp, filter);
2363 "Deleted vlan filter for %d\n",
2367 filter = STAILQ_NEXT(filter, next);
2372 static int bnxt_add_vlan_filter(struct bnxt *bp, uint16_t vlan_id)
2374 struct bnxt_filter_info *filter;
2375 struct bnxt_vnic_info *vnic;
2377 uint32_t en = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN |
2378 HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN_MASK;
2379 uint32_t chk = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN;
2381 /* Implementation notes on the use of VNIC in this command:
2383 * By default, these filters belong to default vnic for the function.
2384 * Once these filters are set up, only destination VNIC can be modified.
2385 * If the destination VNIC is not specified in this command,
2386 * then the HWRM shall only create an l2 context id.
2389 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2390 filter = STAILQ_FIRST(&vnic->filter);
2391 /* Check if the VLAN has already been added */
2393 if (bnxt_vlan_filter_exists(bp, filter, chk, vlan_id))
2396 filter = STAILQ_NEXT(filter, next);
2399 /* No match found. Alloc a fresh filter and issue the L2_FILTER_ALLOC
2400 * command to create MAC+VLAN filter with the right flags, enables set.
2402 filter = bnxt_alloc_filter(bp);
2405 "MAC/VLAN filter alloc failed\n");
2408 /* MAC + VLAN ID filter */
2409 /* If l2_ivlan == 0 and l2_ivlan_mask != 0, only
2410 * untagged packets are received
2412 * If l2_ivlan != 0 and l2_ivlan_mask != 0, untagged
2413 * packets and only the programmed vlan's packets are received
2415 filter->l2_ivlan = vlan_id;
2416 filter->l2_ivlan_mask = 0x0FFF;
2417 filter->enables |= en;
2418 filter->flags |= HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
2420 rc = bnxt_hwrm_set_l2_filter(bp, vnic->fw_vnic_id, filter);
2422 /* Free the newly allocated filter as we were
2423 * not able to create the filter in hardware.
2425 bnxt_free_filter(bp, filter);
2429 filter->mac_index = 0;
2430 /* Add this new filter to the list */
2432 STAILQ_INSERT_HEAD(&vnic->filter, filter, next);
2434 STAILQ_INSERT_TAIL(&vnic->filter, filter, next);
2437 "Added Vlan filter for %d\n", vlan_id);
2441 static int bnxt_vlan_filter_set_op(struct rte_eth_dev *eth_dev,
2442 uint16_t vlan_id, int on)
2444 struct bnxt *bp = eth_dev->data->dev_private;
2447 rc = is_bnxt_in_error(bp);
2451 if (!eth_dev->data->dev_started) {
2452 PMD_DRV_LOG(ERR, "port must be started before setting vlan\n");
2456 /* These operations apply to ALL existing MAC/VLAN filters */
2458 return bnxt_add_vlan_filter(bp, vlan_id);
2460 return bnxt_del_vlan_filter(bp, vlan_id);
2463 static int bnxt_del_dflt_mac_filter(struct bnxt *bp,
2464 struct bnxt_vnic_info *vnic)
2466 struct bnxt_filter_info *filter;
2469 filter = STAILQ_FIRST(&vnic->filter);
2471 if (filter->mac_index == 0 &&
2472 !memcmp(filter->l2_addr, bp->mac_addr,
2473 RTE_ETHER_ADDR_LEN)) {
2474 rc = bnxt_hwrm_clear_l2_filter(bp, filter);
2476 STAILQ_REMOVE(&vnic->filter, filter,
2477 bnxt_filter_info, next);
2478 bnxt_free_filter(bp, filter);
2482 filter = STAILQ_NEXT(filter, next);
2488 bnxt_config_vlan_hw_filter(struct bnxt *bp, uint64_t rx_offloads)
2490 struct bnxt_vnic_info *vnic;
2494 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2495 if (!(rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)) {
2496 /* Remove any VLAN filters programmed */
2497 for (i = 0; i < RTE_ETHER_MAX_VLAN_ID; i++)
2498 bnxt_del_vlan_filter(bp, i);
2500 rc = bnxt_add_mac_filter(bp, vnic, NULL, 0, 0);
2504 /* Default filter will allow packets that match the
2505 * dest mac. So, it has to be deleted, otherwise, we
2506 * will endup receiving vlan packets for which the
2507 * filter is not programmed, when hw-vlan-filter
2508 * configuration is ON
2510 bnxt_del_dflt_mac_filter(bp, vnic);
2511 /* This filter will allow only untagged packets */
2512 bnxt_add_vlan_filter(bp, 0);
2514 PMD_DRV_LOG(DEBUG, "VLAN Filtering: %d\n",
2515 !!(rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER));
2520 static int bnxt_free_one_vnic(struct bnxt *bp, uint16_t vnic_id)
2522 struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
2526 /* Destroy vnic filters and vnic */
2527 if (bp->eth_dev->data->dev_conf.rxmode.offloads &
2528 DEV_RX_OFFLOAD_VLAN_FILTER) {
2529 for (i = 0; i < RTE_ETHER_MAX_VLAN_ID; i++)
2530 bnxt_del_vlan_filter(bp, i);
2532 bnxt_del_dflt_mac_filter(bp, vnic);
2534 rc = bnxt_hwrm_vnic_ctx_free(bp, vnic);
2538 rc = bnxt_hwrm_vnic_free(bp, vnic);
2542 rte_free(vnic->fw_grp_ids);
2543 vnic->fw_grp_ids = NULL;
2545 vnic->rx_queue_cnt = 0;
2551 bnxt_config_vlan_hw_stripping(struct bnxt *bp, uint64_t rx_offloads)
2553 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2556 /* Destroy, recreate and reconfigure the default vnic */
2557 rc = bnxt_free_one_vnic(bp, 0);
2561 /* default vnic 0 */
2562 rc = bnxt_setup_one_vnic(bp, 0);
2566 if (bp->eth_dev->data->dev_conf.rxmode.offloads &
2567 DEV_RX_OFFLOAD_VLAN_FILTER) {
2568 rc = bnxt_add_vlan_filter(bp, 0);
2571 rc = bnxt_restore_vlan_filters(bp);
2575 rc = bnxt_add_mac_filter(bp, vnic, NULL, 0, 0);
2580 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
2584 PMD_DRV_LOG(DEBUG, "VLAN Strip Offload: %d\n",
2585 !!(rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP));
2591 bnxt_vlan_offload_set_op(struct rte_eth_dev *dev, int mask)
2593 uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
2594 struct bnxt *bp = dev->data->dev_private;
2597 rc = is_bnxt_in_error(bp);
2601 /* Filter settings will get applied when port is started */
2602 if (!dev->data->dev_started)
2605 if (mask & ETH_VLAN_FILTER_MASK) {
2606 /* Enable or disable VLAN filtering */
2607 rc = bnxt_config_vlan_hw_filter(bp, rx_offloads);
2612 if (mask & ETH_VLAN_STRIP_MASK) {
2613 /* Enable or disable VLAN stripping */
2614 rc = bnxt_config_vlan_hw_stripping(bp, rx_offloads);
2619 if (mask & ETH_VLAN_EXTEND_MASK) {
2620 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)
2621 PMD_DRV_LOG(DEBUG, "Extend VLAN supported\n");
2623 PMD_DRV_LOG(INFO, "Extend VLAN unsupported\n");
2630 bnxt_vlan_tpid_set_op(struct rte_eth_dev *dev, enum rte_vlan_type vlan_type,
2633 struct bnxt *bp = dev->data->dev_private;
2634 int qinq = dev->data->dev_conf.rxmode.offloads &
2635 DEV_RX_OFFLOAD_VLAN_EXTEND;
2637 if (vlan_type != ETH_VLAN_TYPE_INNER &&
2638 vlan_type != ETH_VLAN_TYPE_OUTER) {
2640 "Unsupported vlan type.");
2645 "QinQ not enabled. Needs to be ON as we can "
2646 "accelerate only outer vlan\n");
2650 if (vlan_type == ETH_VLAN_TYPE_OUTER) {
2652 case RTE_ETHER_TYPE_QINQ:
2654 TX_BD_LONG_CFA_META_VLAN_TPID_TPID88A8;
2656 case RTE_ETHER_TYPE_VLAN:
2658 TX_BD_LONG_CFA_META_VLAN_TPID_TPID8100;
2660 case RTE_ETHER_TYPE_QINQ1:
2662 TX_BD_LONG_CFA_META_VLAN_TPID_TPID9100;
2664 case RTE_ETHER_TYPE_QINQ2:
2666 TX_BD_LONG_CFA_META_VLAN_TPID_TPID9200;
2668 case RTE_ETHER_TYPE_QINQ3:
2670 TX_BD_LONG_CFA_META_VLAN_TPID_TPID9300;
2673 PMD_DRV_LOG(ERR, "Invalid TPID: %x\n", tpid);
2676 bp->outer_tpid_bd |= tpid;
2677 PMD_DRV_LOG(INFO, "outer_tpid_bd = %x\n", bp->outer_tpid_bd);
2678 } else if (vlan_type == ETH_VLAN_TYPE_INNER) {
2680 "Can accelerate only outer vlan in QinQ\n");
2688 bnxt_set_default_mac_addr_op(struct rte_eth_dev *dev,
2689 struct rte_ether_addr *addr)
2691 struct bnxt *bp = dev->data->dev_private;
2692 /* Default Filter is tied to VNIC 0 */
2693 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2696 rc = is_bnxt_in_error(bp);
2700 if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp))
2703 if (rte_is_zero_ether_addr(addr))
2706 /* Filter settings will get applied when port is started */
2707 if (!dev->data->dev_started)
2710 /* Check if the requested MAC is already added */
2711 if (memcmp(addr, bp->mac_addr, RTE_ETHER_ADDR_LEN) == 0)
2714 /* Destroy filter and re-create it */
2715 bnxt_del_dflt_mac_filter(bp, vnic);
2717 memcpy(bp->mac_addr, addr, RTE_ETHER_ADDR_LEN);
2718 if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_VLAN_FILTER) {
2719 /* This filter will allow only untagged packets */
2720 rc = bnxt_add_vlan_filter(bp, 0);
2722 rc = bnxt_add_mac_filter(bp, vnic, addr, 0, 0);
2725 PMD_DRV_LOG(DEBUG, "Set MAC addr\n");
2730 bnxt_dev_set_mc_addr_list_op(struct rte_eth_dev *eth_dev,
2731 struct rte_ether_addr *mc_addr_set,
2732 uint32_t nb_mc_addr)
2734 struct bnxt *bp = eth_dev->data->dev_private;
2735 char *mc_addr_list = (char *)mc_addr_set;
2736 struct bnxt_vnic_info *vnic;
2737 uint32_t off = 0, i = 0;
2740 rc = is_bnxt_in_error(bp);
2744 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2746 if (nb_mc_addr > BNXT_MAX_MC_ADDRS) {
2747 vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
2751 /* TODO Check for Duplicate mcast addresses */
2752 vnic->flags &= ~BNXT_VNIC_INFO_ALLMULTI;
2753 for (i = 0; i < nb_mc_addr; i++) {
2754 memcpy(vnic->mc_list + off, &mc_addr_list[i],
2755 RTE_ETHER_ADDR_LEN);
2756 off += RTE_ETHER_ADDR_LEN;
2759 vnic->mc_addr_cnt = i;
2760 if (vnic->mc_addr_cnt)
2761 vnic->flags |= BNXT_VNIC_INFO_MCAST;
2763 vnic->flags &= ~BNXT_VNIC_INFO_MCAST;
2766 return bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
2770 bnxt_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
2772 struct bnxt *bp = dev->data->dev_private;
2773 uint8_t fw_major = (bp->fw_ver >> 24) & 0xff;
2774 uint8_t fw_minor = (bp->fw_ver >> 16) & 0xff;
2775 uint8_t fw_updt = (bp->fw_ver >> 8) & 0xff;
2776 uint8_t fw_rsvd = bp->fw_ver & 0xff;
2779 ret = snprintf(fw_version, fw_size, "%d.%d.%d.%d",
2780 fw_major, fw_minor, fw_updt, fw_rsvd);
2782 ret += 1; /* add the size of '\0' */
2783 if (fw_size < (uint32_t)ret)
2790 bnxt_rxq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
2791 struct rte_eth_rxq_info *qinfo)
2793 struct bnxt *bp = dev->data->dev_private;
2794 struct bnxt_rx_queue *rxq;
2796 if (is_bnxt_in_error(bp))
2799 rxq = dev->data->rx_queues[queue_id];
2801 qinfo->mp = rxq->mb_pool;
2802 qinfo->scattered_rx = dev->data->scattered_rx;
2803 qinfo->nb_desc = rxq->nb_rx_desc;
2805 qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
2806 qinfo->conf.rx_drop_en = rxq->drop_en;
2807 qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
2808 qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
2812 bnxt_txq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
2813 struct rte_eth_txq_info *qinfo)
2815 struct bnxt *bp = dev->data->dev_private;
2816 struct bnxt_tx_queue *txq;
2818 if (is_bnxt_in_error(bp))
2821 txq = dev->data->tx_queues[queue_id];
2823 qinfo->nb_desc = txq->nb_tx_desc;
2825 qinfo->conf.tx_thresh.pthresh = txq->pthresh;
2826 qinfo->conf.tx_thresh.hthresh = txq->hthresh;
2827 qinfo->conf.tx_thresh.wthresh = txq->wthresh;
2829 qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
2830 qinfo->conf.tx_rs_thresh = 0;
2831 qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
2832 qinfo->conf.offloads = txq->offloads;
2835 static const struct {
2836 eth_rx_burst_t pkt_burst;
2838 } bnxt_rx_burst_info[] = {
2839 {bnxt_recv_pkts, "Scalar"},
2840 #if defined(RTE_ARCH_X86)
2841 {bnxt_recv_pkts_vec, "Vector SSE"},
2842 #elif defined(RTE_ARCH_ARM64)
2843 {bnxt_recv_pkts_vec, "Vector Neon"},
2848 bnxt_rx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
2849 struct rte_eth_burst_mode *mode)
2851 eth_rx_burst_t pkt_burst = dev->rx_pkt_burst;
2854 for (i = 0; i < RTE_DIM(bnxt_rx_burst_info); i++) {
2855 if (pkt_burst == bnxt_rx_burst_info[i].pkt_burst) {
2856 snprintf(mode->info, sizeof(mode->info), "%s",
2857 bnxt_rx_burst_info[i].info);
2865 static const struct {
2866 eth_tx_burst_t pkt_burst;
2868 } bnxt_tx_burst_info[] = {
2869 {bnxt_xmit_pkts, "Scalar"},
2870 #if defined(RTE_ARCH_X86)
2871 {bnxt_xmit_pkts_vec, "Vector SSE"},
2872 #elif defined(RTE_ARCH_ARM64)
2873 {bnxt_xmit_pkts_vec, "Vector Neon"},
2878 bnxt_tx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
2879 struct rte_eth_burst_mode *mode)
2881 eth_tx_burst_t pkt_burst = dev->tx_pkt_burst;
2884 for (i = 0; i < RTE_DIM(bnxt_tx_burst_info); i++) {
2885 if (pkt_burst == bnxt_tx_burst_info[i].pkt_burst) {
2886 snprintf(mode->info, sizeof(mode->info), "%s",
2887 bnxt_tx_burst_info[i].info);
2895 int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
2897 struct bnxt *bp = eth_dev->data->dev_private;
2898 uint32_t new_pkt_size;
2902 rc = is_bnxt_in_error(bp);
2906 /* Exit if receive queues are not configured yet */
2907 if (!eth_dev->data->nb_rx_queues)
2910 new_pkt_size = new_mtu + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN +
2911 VLAN_TAG_SIZE * BNXT_NUM_VLANS;
2914 * Disallow any MTU change that would require scattered receive support
2915 * if it is not already enabled.
2917 if (eth_dev->data->dev_started &&
2918 !eth_dev->data->scattered_rx &&
2920 eth_dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM)) {
2922 "MTU change would require scattered rx support. ");
2923 PMD_DRV_LOG(ERR, "Stop port before changing MTU.\n");
2927 if (new_mtu > RTE_ETHER_MTU) {
2928 bp->flags |= BNXT_FLAG_JUMBO;
2929 bp->eth_dev->data->dev_conf.rxmode.offloads |=
2930 DEV_RX_OFFLOAD_JUMBO_FRAME;
2932 bp->eth_dev->data->dev_conf.rxmode.offloads &=
2933 ~DEV_RX_OFFLOAD_JUMBO_FRAME;
2934 bp->flags &= ~BNXT_FLAG_JUMBO;
2937 /* Is there a change in mtu setting? */
2938 if (eth_dev->data->dev_conf.rxmode.max_rx_pkt_len == new_pkt_size)
2941 for (i = 0; i < bp->nr_vnics; i++) {
2942 struct bnxt_vnic_info *vnic = &bp->vnic_info[i];
2945 vnic->mru = BNXT_VNIC_MRU(new_mtu);
2946 rc = bnxt_hwrm_vnic_cfg(bp, vnic);
2950 size = rte_pktmbuf_data_room_size(bp->rx_queues[0]->mb_pool);
2951 size -= RTE_PKTMBUF_HEADROOM;
2953 if (size < new_mtu) {
2954 rc = bnxt_hwrm_vnic_plcmode_cfg(bp, vnic);
2961 eth_dev->data->dev_conf.rxmode.max_rx_pkt_len = new_pkt_size;
2963 PMD_DRV_LOG(INFO, "New MTU is %d\n", new_mtu);
2969 bnxt_vlan_pvid_set_op(struct rte_eth_dev *dev, uint16_t pvid, int on)
2971 struct bnxt *bp = dev->data->dev_private;
2972 uint16_t vlan = bp->vlan;
2975 rc = is_bnxt_in_error(bp);
2979 if (!BNXT_SINGLE_PF(bp) || BNXT_VF(bp)) {
2981 "PVID cannot be modified for this function\n");
2984 bp->vlan = on ? pvid : 0;
2986 rc = bnxt_hwrm_set_default_vlan(bp, 0, 0);
2993 bnxt_dev_led_on_op(struct rte_eth_dev *dev)
2995 struct bnxt *bp = dev->data->dev_private;
2998 rc = is_bnxt_in_error(bp);
3002 return bnxt_hwrm_port_led_cfg(bp, true);
3006 bnxt_dev_led_off_op(struct rte_eth_dev *dev)
3008 struct bnxt *bp = dev->data->dev_private;
3011 rc = is_bnxt_in_error(bp);
3015 return bnxt_hwrm_port_led_cfg(bp, false);
3019 bnxt_rx_queue_count_op(struct rte_eth_dev *dev, uint16_t rx_queue_id)
3021 struct bnxt *bp = (struct bnxt *)dev->data->dev_private;
3022 struct bnxt_cp_ring_info *cpr;
3023 uint32_t desc = 0, raw_cons;
3024 struct bnxt_rx_queue *rxq;
3025 struct rx_pkt_cmpl *rxcmp;
3028 rc = is_bnxt_in_error(bp);
3032 rxq = dev->data->rx_queues[rx_queue_id];
3034 raw_cons = cpr->cp_raw_cons;
3037 uint32_t agg_cnt, cons, cmpl_type;
3039 cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3040 rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3042 if (!CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct))
3045 cmpl_type = CMP_TYPE(rxcmp);
3047 switch (cmpl_type) {
3048 case CMPL_BASE_TYPE_RX_L2:
3049 case CMPL_BASE_TYPE_RX_L2_V2:
3050 agg_cnt = BNXT_RX_L2_AGG_BUFS(rxcmp);
3051 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3055 case CMPL_BASE_TYPE_RX_TPA_END:
3056 if (BNXT_CHIP_P5(rxq->bp)) {
3057 struct rx_tpa_v2_end_cmpl_hi *p5_tpa_end;
3059 p5_tpa_end = (void *)rxcmp;
3060 agg_cnt = BNXT_TPA_END_AGG_BUFS_TH(p5_tpa_end);
3062 struct rx_tpa_end_cmpl *tpa_end;
3064 tpa_end = (void *)rxcmp;
3065 agg_cnt = BNXT_TPA_END_AGG_BUFS(tpa_end);
3068 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3073 raw_cons += CMP_LEN(cmpl_type);
3081 bnxt_rx_descriptor_status_op(void *rx_queue, uint16_t offset)
3083 struct bnxt_rx_queue *rxq = rx_queue;
3084 struct bnxt_cp_ring_info *cpr;
3085 struct bnxt_rx_ring_info *rxr;
3086 uint32_t desc, raw_cons;
3087 struct bnxt *bp = rxq->bp;
3088 struct rx_pkt_cmpl *rxcmp;
3091 rc = is_bnxt_in_error(bp);
3095 if (offset >= rxq->nb_rx_desc)
3102 * For the vector receive case, the completion at the requested
3103 * offset can be indexed directly.
3105 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
3106 if (bp->flags & BNXT_FLAG_RX_VECTOR_PKT_MODE) {
3107 struct rx_pkt_cmpl *rxcmp;
3110 /* Check status of completion descriptor. */
3111 raw_cons = cpr->cp_raw_cons +
3112 offset * CMP_LEN(CMPL_BASE_TYPE_RX_L2);
3113 cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3114 rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3116 if (CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct))
3117 return RTE_ETH_RX_DESC_DONE;
3119 /* Check whether rx desc has an mbuf attached. */
3120 cons = RING_CMP(rxr->rx_ring_struct, raw_cons / 2);
3121 if (cons >= rxq->rxrearm_start &&
3122 cons < rxq->rxrearm_start + rxq->rxrearm_nb) {
3123 return RTE_ETH_RX_DESC_UNAVAIL;
3126 return RTE_ETH_RX_DESC_AVAIL;
3131 * For the non-vector receive case, scan the completion ring to
3132 * locate the completion descriptor for the requested offset.
3134 raw_cons = cpr->cp_raw_cons;
3137 uint32_t agg_cnt, cons, cmpl_type;
3139 cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3140 rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3142 if (!CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct))
3145 cmpl_type = CMP_TYPE(rxcmp);
3147 switch (cmpl_type) {
3148 case CMPL_BASE_TYPE_RX_L2:
3149 case CMPL_BASE_TYPE_RX_L2_V2:
3150 if (desc == offset) {
3151 cons = rxcmp->opaque;
3152 if (rxr->rx_buf_ring[cons])
3153 return RTE_ETH_RX_DESC_DONE;
3155 return RTE_ETH_RX_DESC_UNAVAIL;
3157 agg_cnt = BNXT_RX_L2_AGG_BUFS(rxcmp);
3158 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3162 case CMPL_BASE_TYPE_RX_TPA_END:
3164 return RTE_ETH_RX_DESC_DONE;
3166 if (BNXT_CHIP_P5(rxq->bp)) {
3167 struct rx_tpa_v2_end_cmpl_hi *p5_tpa_end;
3169 p5_tpa_end = (void *)rxcmp;
3170 agg_cnt = BNXT_TPA_END_AGG_BUFS_TH(p5_tpa_end);
3172 struct rx_tpa_end_cmpl *tpa_end;
3174 tpa_end = (void *)rxcmp;
3175 agg_cnt = BNXT_TPA_END_AGG_BUFS(tpa_end);
3178 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3183 raw_cons += CMP_LEN(cmpl_type);
3187 return RTE_ETH_RX_DESC_AVAIL;
3191 bnxt_tx_descriptor_status_op(void *tx_queue, uint16_t offset)
3193 struct bnxt_tx_queue *txq = (struct bnxt_tx_queue *)tx_queue;
3194 struct bnxt_tx_ring_info *txr;
3195 struct bnxt_cp_ring_info *cpr;
3196 struct rte_mbuf **tx_buf;
3197 struct tx_pkt_cmpl *txcmp;
3198 uint32_t cons, cp_cons;
3204 rc = is_bnxt_in_error(txq->bp);
3211 if (offset >= txq->nb_tx_desc)
3214 cons = RING_CMP(cpr->cp_ring_struct, offset);
3215 txcmp = (struct tx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3216 cp_cons = cpr->cp_raw_cons;
3218 if (cons > cp_cons) {
3219 if (CMPL_VALID(txcmp, cpr->valid))
3220 return RTE_ETH_TX_DESC_UNAVAIL;
3222 if (CMPL_VALID(txcmp, !cpr->valid))
3223 return RTE_ETH_TX_DESC_UNAVAIL;
3225 tx_buf = &txr->tx_buf_ring[cons];
3226 if (*tx_buf == NULL)
3227 return RTE_ETH_TX_DESC_DONE;
3229 return RTE_ETH_TX_DESC_FULL;
3233 bnxt_flow_ops_get_op(struct rte_eth_dev *dev,
3234 const struct rte_flow_ops **ops)
3236 struct bnxt *bp = dev->data->dev_private;
3242 if (BNXT_ETH_DEV_IS_REPRESENTOR(dev)) {
3243 struct bnxt_representor *vfr = dev->data->dev_private;
3244 bp = vfr->parent_dev->data->dev_private;
3245 /* parent is deleted while children are still valid */
3247 PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR Error\n",
3248 dev->data->port_id);
3253 ret = is_bnxt_in_error(bp);
3257 /* PMD supports thread-safe flow operations. rte_flow API
3258 * functions can avoid mutex for multi-thread safety.
3260 dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
3262 if (BNXT_TRUFLOW_EN(bp))
3263 *ops = &bnxt_ulp_rte_flow_ops;
3265 *ops = &bnxt_flow_ops;
3270 static const uint32_t *
3271 bnxt_dev_supported_ptypes_get_op(struct rte_eth_dev *dev)
3273 static const uint32_t ptypes[] = {
3274 RTE_PTYPE_L2_ETHER_VLAN,
3275 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
3276 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
3280 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
3281 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
3282 RTE_PTYPE_INNER_L4_ICMP,
3283 RTE_PTYPE_INNER_L4_TCP,
3284 RTE_PTYPE_INNER_L4_UDP,
3288 if (!dev->rx_pkt_burst)
3294 static int bnxt_map_regs(struct bnxt *bp, uint32_t *reg_arr, int count,
3297 uint32_t reg_base = *reg_arr & 0xfffff000;
3301 for (i = 0; i < count; i++) {
3302 if ((reg_arr[i] & 0xfffff000) != reg_base)
3305 win_off = BNXT_GRCPF_REG_WINDOW_BASE_OUT + (reg_win - 1) * 4;
3306 rte_write32(reg_base, (uint8_t *)bp->bar0 + win_off);
3310 static int bnxt_map_ptp_regs(struct bnxt *bp)
3312 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3316 reg_arr = ptp->rx_regs;
3317 rc = bnxt_map_regs(bp, reg_arr, BNXT_PTP_RX_REGS, 5);
3321 reg_arr = ptp->tx_regs;
3322 rc = bnxt_map_regs(bp, reg_arr, BNXT_PTP_TX_REGS, 6);
3326 for (i = 0; i < BNXT_PTP_RX_REGS; i++)
3327 ptp->rx_mapped_regs[i] = 0x5000 + (ptp->rx_regs[i] & 0xfff);
3329 for (i = 0; i < BNXT_PTP_TX_REGS; i++)
3330 ptp->tx_mapped_regs[i] = 0x6000 + (ptp->tx_regs[i] & 0xfff);
3335 static void bnxt_unmap_ptp_regs(struct bnxt *bp)
3337 rte_write32(0, (uint8_t *)bp->bar0 +
3338 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 16);
3339 rte_write32(0, (uint8_t *)bp->bar0 +
3340 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 20);
3343 static uint64_t bnxt_cc_read(struct bnxt *bp)
3347 ns = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3348 BNXT_GRCPF_REG_SYNC_TIME));
3349 ns |= (uint64_t)(rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3350 BNXT_GRCPF_REG_SYNC_TIME + 4))) << 32;
3354 static int bnxt_get_tx_ts(struct bnxt *bp, uint64_t *ts)
3356 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3359 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3360 ptp->tx_mapped_regs[BNXT_PTP_TX_FIFO]));
3361 if (fifo & BNXT_PTP_TX_FIFO_EMPTY)
3364 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3365 ptp->tx_mapped_regs[BNXT_PTP_TX_FIFO]));
3366 *ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3367 ptp->tx_mapped_regs[BNXT_PTP_TX_TS_L]));
3368 *ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3369 ptp->tx_mapped_regs[BNXT_PTP_TX_TS_H])) << 32;
3370 rte_read32((uint8_t *)bp->bar0 + ptp->tx_mapped_regs[BNXT_PTP_TX_SEQ]);
3375 static int bnxt_get_rx_ts(struct bnxt *bp, uint64_t *ts)
3377 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3378 struct bnxt_pf_info *pf = bp->pf;
3382 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3383 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3384 if (!(fifo & BNXT_PTP_RX_FIFO_PENDING))
3387 port_id = pf->port_id;
3388 rte_write32(1 << port_id, (uint8_t *)bp->bar0 +
3389 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO_ADV]);
3391 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3392 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3393 if (fifo & BNXT_PTP_RX_FIFO_PENDING) {
3394 /* bnxt_clr_rx_ts(bp); TBD */
3398 *ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3399 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_L]));
3400 *ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3401 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_H])) << 32;
3407 bnxt_timesync_write_time(struct rte_eth_dev *dev, const struct timespec *ts)
3410 struct bnxt *bp = dev->data->dev_private;
3411 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3416 ns = rte_timespec_to_ns(ts);
3417 /* Set the timecounters to a new value. */
3419 ptp->tx_tstamp_tc.nsec = ns;
3420 ptp->rx_tstamp_tc.nsec = ns;
3426 bnxt_timesync_read_time(struct rte_eth_dev *dev, struct timespec *ts)
3428 struct bnxt *bp = dev->data->dev_private;
3429 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3430 uint64_t ns, systime_cycles = 0;
3436 if (BNXT_CHIP_P5(bp))
3437 rc = bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
3440 systime_cycles = bnxt_cc_read(bp);
3442 ns = rte_timecounter_update(&ptp->tc, systime_cycles);
3443 *ts = rte_ns_to_timespec(ns);
3448 bnxt_timesync_enable(struct rte_eth_dev *dev)
3450 struct bnxt *bp = dev->data->dev_private;
3451 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3459 ptp->tx_tstamp_en = 1;
3460 ptp->rxctl = BNXT_PTP_MSG_EVENTS;
3462 rc = bnxt_hwrm_ptp_cfg(bp);
3466 memset(&ptp->tc, 0, sizeof(struct rte_timecounter));
3467 memset(&ptp->rx_tstamp_tc, 0, sizeof(struct rte_timecounter));
3468 memset(&ptp->tx_tstamp_tc, 0, sizeof(struct rte_timecounter));
3470 ptp->tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3471 ptp->tc.cc_shift = shift;
3472 ptp->tc.nsec_mask = (1ULL << shift) - 1;
3474 ptp->rx_tstamp_tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3475 ptp->rx_tstamp_tc.cc_shift = shift;
3476 ptp->rx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
3478 ptp->tx_tstamp_tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3479 ptp->tx_tstamp_tc.cc_shift = shift;
3480 ptp->tx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
3482 if (!BNXT_CHIP_P5(bp))
3483 bnxt_map_ptp_regs(bp);
3485 rc = bnxt_ptp_start(bp);
3491 bnxt_timesync_disable(struct rte_eth_dev *dev)
3493 struct bnxt *bp = dev->data->dev_private;
3494 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3500 ptp->tx_tstamp_en = 0;
3503 bnxt_hwrm_ptp_cfg(bp);
3505 if (!BNXT_CHIP_P5(bp))
3506 bnxt_unmap_ptp_regs(bp);
3514 bnxt_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
3515 struct timespec *timestamp,
3516 uint32_t flags __rte_unused)
3518 struct bnxt *bp = dev->data->dev_private;
3519 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3520 uint64_t rx_tstamp_cycles = 0;
3526 if (BNXT_CHIP_P5(bp))
3527 rx_tstamp_cycles = ptp->rx_timestamp;
3529 bnxt_get_rx_ts(bp, &rx_tstamp_cycles);
3531 ns = rte_timecounter_update(&ptp->rx_tstamp_tc, rx_tstamp_cycles);
3532 *timestamp = rte_ns_to_timespec(ns);
3537 bnxt_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
3538 struct timespec *timestamp)
3540 struct bnxt *bp = dev->data->dev_private;
3541 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3542 uint64_t tx_tstamp_cycles = 0;
3549 if (BNXT_CHIP_P5(bp))
3550 rc = bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_PATH_TX,
3553 rc = bnxt_get_tx_ts(bp, &tx_tstamp_cycles);
3555 ns = rte_timecounter_update(&ptp->tx_tstamp_tc, tx_tstamp_cycles);
3556 *timestamp = rte_ns_to_timespec(ns);
3562 bnxt_timesync_adjust_time(struct rte_eth_dev *dev, int64_t delta)
3564 struct bnxt *bp = dev->data->dev_private;
3565 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3570 ptp->tc.nsec += delta;
3571 ptp->tx_tstamp_tc.nsec += delta;
3572 ptp->rx_tstamp_tc.nsec += delta;
3578 bnxt_get_eeprom_length_op(struct rte_eth_dev *dev)
3580 struct bnxt *bp = dev->data->dev_private;
3582 uint32_t dir_entries;
3583 uint32_t entry_length;
3585 rc = is_bnxt_in_error(bp);
3589 PMD_DRV_LOG(INFO, PCI_PRI_FMT "\n",
3590 bp->pdev->addr.domain, bp->pdev->addr.bus,
3591 bp->pdev->addr.devid, bp->pdev->addr.function);
3593 rc = bnxt_hwrm_nvm_get_dir_info(bp, &dir_entries, &entry_length);
3597 return dir_entries * entry_length;
3601 bnxt_get_eeprom_op(struct rte_eth_dev *dev,
3602 struct rte_dev_eeprom_info *in_eeprom)
3604 struct bnxt *bp = dev->data->dev_private;
3609 rc = is_bnxt_in_error(bp);
3613 PMD_DRV_LOG(INFO, PCI_PRI_FMT " in_eeprom->offset = %d len = %d\n",
3614 bp->pdev->addr.domain, bp->pdev->addr.bus,
3615 bp->pdev->addr.devid, bp->pdev->addr.function,
3616 in_eeprom->offset, in_eeprom->length);
3618 if (in_eeprom->offset == 0) /* special offset value to get directory */
3619 return bnxt_get_nvram_directory(bp, in_eeprom->length,
3622 index = in_eeprom->offset >> 24;
3623 offset = in_eeprom->offset & 0xffffff;
3626 return bnxt_hwrm_get_nvram_item(bp, index - 1, offset,
3627 in_eeprom->length, in_eeprom->data);
3632 static bool bnxt_dir_type_is_ape_bin_format(uint16_t dir_type)
3635 case BNX_DIR_TYPE_CHIMP_PATCH:
3636 case BNX_DIR_TYPE_BOOTCODE:
3637 case BNX_DIR_TYPE_BOOTCODE_2:
3638 case BNX_DIR_TYPE_APE_FW:
3639 case BNX_DIR_TYPE_APE_PATCH:
3640 case BNX_DIR_TYPE_KONG_FW:
3641 case BNX_DIR_TYPE_KONG_PATCH:
3642 case BNX_DIR_TYPE_BONO_FW:
3643 case BNX_DIR_TYPE_BONO_PATCH:
3651 static bool bnxt_dir_type_is_other_exec_format(uint16_t dir_type)
3654 case BNX_DIR_TYPE_AVS:
3655 case BNX_DIR_TYPE_EXP_ROM_MBA:
3656 case BNX_DIR_TYPE_PCIE:
3657 case BNX_DIR_TYPE_TSCF_UCODE:
3658 case BNX_DIR_TYPE_EXT_PHY:
3659 case BNX_DIR_TYPE_CCM:
3660 case BNX_DIR_TYPE_ISCSI_BOOT:
3661 case BNX_DIR_TYPE_ISCSI_BOOT_IPV6:
3662 case BNX_DIR_TYPE_ISCSI_BOOT_IPV4N6:
3670 static bool bnxt_dir_type_is_executable(uint16_t dir_type)
3672 return bnxt_dir_type_is_ape_bin_format(dir_type) ||
3673 bnxt_dir_type_is_other_exec_format(dir_type);
3677 bnxt_set_eeprom_op(struct rte_eth_dev *dev,
3678 struct rte_dev_eeprom_info *in_eeprom)
3680 struct bnxt *bp = dev->data->dev_private;
3681 uint8_t index, dir_op;
3682 uint16_t type, ext, ordinal, attr;
3685 rc = is_bnxt_in_error(bp);
3689 PMD_DRV_LOG(INFO, PCI_PRI_FMT " in_eeprom->offset = %d len = %d\n",
3690 bp->pdev->addr.domain, bp->pdev->addr.bus,
3691 bp->pdev->addr.devid, bp->pdev->addr.function,
3692 in_eeprom->offset, in_eeprom->length);
3695 PMD_DRV_LOG(ERR, "NVM write not supported from a VF\n");
3699 type = in_eeprom->magic >> 16;
3701 if (type == 0xffff) { /* special value for directory operations */
3702 index = in_eeprom->magic & 0xff;
3703 dir_op = in_eeprom->magic >> 8;
3707 case 0x0e: /* erase */
3708 if (in_eeprom->offset != ~in_eeprom->magic)
3710 return bnxt_hwrm_erase_nvram_directory(bp, index - 1);
3716 /* Create or re-write an NVM item: */
3717 if (bnxt_dir_type_is_executable(type) == true)
3719 ext = in_eeprom->magic & 0xffff;
3720 ordinal = in_eeprom->offset >> 16;
3721 attr = in_eeprom->offset & 0xffff;
3723 return bnxt_hwrm_flash_nvram(bp, type, ordinal, ext, attr,
3724 in_eeprom->data, in_eeprom->length);
3731 static const struct eth_dev_ops bnxt_dev_ops = {
3732 .dev_infos_get = bnxt_dev_info_get_op,
3733 .dev_close = bnxt_dev_close_op,
3734 .dev_configure = bnxt_dev_configure_op,
3735 .dev_start = bnxt_dev_start_op,
3736 .dev_stop = bnxt_dev_stop_op,
3737 .dev_set_link_up = bnxt_dev_set_link_up_op,
3738 .dev_set_link_down = bnxt_dev_set_link_down_op,
3739 .stats_get = bnxt_stats_get_op,
3740 .stats_reset = bnxt_stats_reset_op,
3741 .rx_queue_setup = bnxt_rx_queue_setup_op,
3742 .rx_queue_release = bnxt_rx_queue_release_op,
3743 .tx_queue_setup = bnxt_tx_queue_setup_op,
3744 .tx_queue_release = bnxt_tx_queue_release_op,
3745 .rx_queue_intr_enable = bnxt_rx_queue_intr_enable_op,
3746 .rx_queue_intr_disable = bnxt_rx_queue_intr_disable_op,
3747 .reta_update = bnxt_reta_update_op,
3748 .reta_query = bnxt_reta_query_op,
3749 .rss_hash_update = bnxt_rss_hash_update_op,
3750 .rss_hash_conf_get = bnxt_rss_hash_conf_get_op,
3751 .link_update = bnxt_link_update_op,
3752 .promiscuous_enable = bnxt_promiscuous_enable_op,
3753 .promiscuous_disable = bnxt_promiscuous_disable_op,
3754 .allmulticast_enable = bnxt_allmulticast_enable_op,
3755 .allmulticast_disable = bnxt_allmulticast_disable_op,
3756 .mac_addr_add = bnxt_mac_addr_add_op,
3757 .mac_addr_remove = bnxt_mac_addr_remove_op,
3758 .flow_ctrl_get = bnxt_flow_ctrl_get_op,
3759 .flow_ctrl_set = bnxt_flow_ctrl_set_op,
3760 .udp_tunnel_port_add = bnxt_udp_tunnel_port_add_op,
3761 .udp_tunnel_port_del = bnxt_udp_tunnel_port_del_op,
3762 .vlan_filter_set = bnxt_vlan_filter_set_op,
3763 .vlan_offload_set = bnxt_vlan_offload_set_op,
3764 .vlan_tpid_set = bnxt_vlan_tpid_set_op,
3765 .vlan_pvid_set = bnxt_vlan_pvid_set_op,
3766 .mtu_set = bnxt_mtu_set_op,
3767 .mac_addr_set = bnxt_set_default_mac_addr_op,
3768 .xstats_get = bnxt_dev_xstats_get_op,
3769 .xstats_get_names = bnxt_dev_xstats_get_names_op,
3770 .xstats_reset = bnxt_dev_xstats_reset_op,
3771 .fw_version_get = bnxt_fw_version_get,
3772 .set_mc_addr_list = bnxt_dev_set_mc_addr_list_op,
3773 .rxq_info_get = bnxt_rxq_info_get_op,
3774 .txq_info_get = bnxt_txq_info_get_op,
3775 .rx_burst_mode_get = bnxt_rx_burst_mode_get,
3776 .tx_burst_mode_get = bnxt_tx_burst_mode_get,
3777 .dev_led_on = bnxt_dev_led_on_op,
3778 .dev_led_off = bnxt_dev_led_off_op,
3779 .rx_queue_start = bnxt_rx_queue_start,
3780 .rx_queue_stop = bnxt_rx_queue_stop,
3781 .tx_queue_start = bnxt_tx_queue_start,
3782 .tx_queue_stop = bnxt_tx_queue_stop,
3783 .flow_ops_get = bnxt_flow_ops_get_op,
3784 .dev_supported_ptypes_get = bnxt_dev_supported_ptypes_get_op,
3785 .get_eeprom_length = bnxt_get_eeprom_length_op,
3786 .get_eeprom = bnxt_get_eeprom_op,
3787 .set_eeprom = bnxt_set_eeprom_op,
3788 .timesync_enable = bnxt_timesync_enable,
3789 .timesync_disable = bnxt_timesync_disable,
3790 .timesync_read_time = bnxt_timesync_read_time,
3791 .timesync_write_time = bnxt_timesync_write_time,
3792 .timesync_adjust_time = bnxt_timesync_adjust_time,
3793 .timesync_read_rx_timestamp = bnxt_timesync_read_rx_timestamp,
3794 .timesync_read_tx_timestamp = bnxt_timesync_read_tx_timestamp,
3797 static uint32_t bnxt_map_reset_regs(struct bnxt *bp, uint32_t reg)
3801 /* Only pre-map the reset GRC registers using window 3 */
3802 rte_write32(reg & 0xfffff000, (uint8_t *)bp->bar0 +
3803 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 8);
3805 offset = BNXT_GRCP_WINDOW_3_BASE + (reg & 0xffc);
3810 int bnxt_map_fw_health_status_regs(struct bnxt *bp)
3812 struct bnxt_error_recovery_info *info = bp->recovery_info;
3813 uint32_t reg_base = 0xffffffff;
3816 /* Only pre-map the monitoring GRC registers using window 2 */
3817 for (i = 0; i < BNXT_FW_STATUS_REG_CNT; i++) {
3818 uint32_t reg = info->status_regs[i];
3820 if (BNXT_FW_STATUS_REG_TYPE(reg) != BNXT_FW_STATUS_REG_TYPE_GRC)
3823 if (reg_base == 0xffffffff)
3824 reg_base = reg & 0xfffff000;
3825 if ((reg & 0xfffff000) != reg_base)
3828 /* Use mask 0xffc as the Lower 2 bits indicates
3829 * address space location
3831 info->mapped_status_regs[i] = BNXT_GRCP_WINDOW_2_BASE +
3835 if (reg_base == 0xffffffff)
3838 rte_write32(reg_base, (uint8_t *)bp->bar0 +
3839 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
3844 static void bnxt_write_fw_reset_reg(struct bnxt *bp, uint32_t index)
3846 struct bnxt_error_recovery_info *info = bp->recovery_info;
3847 uint32_t delay = info->delay_after_reset[index];
3848 uint32_t val = info->reset_reg_val[index];
3849 uint32_t reg = info->reset_reg[index];
3850 uint32_t type, offset;
3853 type = BNXT_FW_STATUS_REG_TYPE(reg);
3854 offset = BNXT_FW_STATUS_REG_OFF(reg);
3857 case BNXT_FW_STATUS_REG_TYPE_CFG:
3858 ret = rte_pci_write_config(bp->pdev, &val, sizeof(val), offset);
3860 PMD_DRV_LOG(ERR, "Failed to write %#x at PCI offset %#x",
3865 case BNXT_FW_STATUS_REG_TYPE_GRC:
3866 offset = bnxt_map_reset_regs(bp, offset);
3867 rte_write32(val, (uint8_t *)bp->bar0 + offset);
3869 case BNXT_FW_STATUS_REG_TYPE_BAR0:
3870 rte_write32(val, (uint8_t *)bp->bar0 + offset);
3873 /* wait on a specific interval of time until core reset is complete */
3875 rte_delay_ms(delay);
3878 static void bnxt_dev_cleanup(struct bnxt *bp)
3880 bp->eth_dev->data->dev_link.link_status = 0;
3881 bp->link_info->link_up = 0;
3882 if (bp->eth_dev->data->dev_started)
3883 bnxt_dev_stop(bp->eth_dev);
3885 bnxt_uninit_resources(bp, true);
3889 bnxt_check_fw_reset_done(struct bnxt *bp)
3891 int timeout = bp->fw_reset_max_msecs;
3896 rc = rte_pci_read_config(bp->pdev, &val, sizeof(val), PCI_SUBSYSTEM_ID_OFFSET);
3898 PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x", PCI_SUBSYSTEM_ID_OFFSET);
3904 } while (timeout--);
3906 if (val == 0xffff) {
3907 PMD_DRV_LOG(ERR, "Firmware reset aborted, PCI config space invalid\n");
3914 static int bnxt_restore_vlan_filters(struct bnxt *bp)
3916 struct rte_eth_dev *dev = bp->eth_dev;
3917 struct rte_vlan_filter_conf *vfc;
3921 for (vlan_id = 1; vlan_id <= RTE_ETHER_MAX_VLAN_ID; vlan_id++) {
3922 vfc = &dev->data->vlan_filter_conf;
3923 vidx = vlan_id / 64;
3924 vbit = vlan_id % 64;
3926 /* Each bit corresponds to a VLAN id */
3927 if (vfc->ids[vidx] & (UINT64_C(1) << vbit)) {
3928 rc = bnxt_add_vlan_filter(bp, vlan_id);
3937 static int bnxt_restore_mac_filters(struct bnxt *bp)
3939 struct rte_eth_dev *dev = bp->eth_dev;
3940 struct rte_eth_dev_info dev_info;
3941 struct rte_ether_addr *addr;
3947 if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp))
3950 rc = bnxt_dev_info_get_op(dev, &dev_info);
3954 /* replay MAC address configuration */
3955 for (i = 1; i < dev_info.max_mac_addrs; i++) {
3956 addr = &dev->data->mac_addrs[i];
3958 /* skip zero address */
3959 if (rte_is_zero_ether_addr(addr))
3963 pool_mask = dev->data->mac_pool_sel[i];
3966 if (pool_mask & 1ULL) {
3967 rc = bnxt_mac_addr_add_op(dev, addr, i, pool);
3973 } while (pool_mask);
3979 static int bnxt_restore_filters(struct bnxt *bp)
3981 struct rte_eth_dev *dev = bp->eth_dev;
3984 if (dev->data->all_multicast) {
3985 ret = bnxt_allmulticast_enable_op(dev);
3989 if (dev->data->promiscuous) {
3990 ret = bnxt_promiscuous_enable_op(dev);
3995 ret = bnxt_restore_mac_filters(bp);
3999 ret = bnxt_restore_vlan_filters(bp);
4000 /* TODO restore other filters as well */
4004 static int bnxt_check_fw_ready(struct bnxt *bp)
4006 int timeout = bp->fw_reset_max_msecs;
4010 rc = bnxt_hwrm_poll_ver_get(bp);
4013 rte_delay_ms(BNXT_FW_READY_WAIT_INTERVAL);
4014 timeout -= BNXT_FW_READY_WAIT_INTERVAL;
4015 } while (rc && timeout > 0);
4018 PMD_DRV_LOG(ERR, "FW is not Ready after reset\n");
4023 static void bnxt_dev_recover(void *arg)
4025 struct bnxt *bp = arg;
4028 pthread_mutex_lock(&bp->err_recovery_lock);
4030 if (!bp->fw_reset_min_msecs) {
4031 rc = bnxt_check_fw_reset_done(bp);
4036 /* Clear Error flag so that device re-init should happen */
4037 bp->flags &= ~BNXT_FLAG_FATAL_ERROR;
4039 rc = bnxt_check_fw_ready(bp);
4043 rc = bnxt_init_resources(bp, true);
4046 "Failed to initialize resources after reset\n");
4049 /* clear reset flag as the device is initialized now */
4050 bp->flags &= ~BNXT_FLAG_FW_RESET;
4052 rc = bnxt_dev_start_op(bp->eth_dev);
4054 PMD_DRV_LOG(ERR, "Failed to start port after reset\n");
4058 rc = bnxt_restore_filters(bp);
4062 PMD_DRV_LOG(INFO, "Recovered from FW reset\n");
4063 pthread_mutex_unlock(&bp->err_recovery_lock);
4067 bnxt_dev_stop(bp->eth_dev);
4069 bp->flags |= BNXT_FLAG_FATAL_ERROR;
4070 bnxt_uninit_resources(bp, false);
4071 pthread_mutex_unlock(&bp->err_recovery_lock);
4072 PMD_DRV_LOG(ERR, "Failed to recover from FW reset\n");
4075 void bnxt_dev_reset_and_resume(void *arg)
4077 struct bnxt *bp = arg;
4078 uint32_t us = US_PER_MS * bp->fw_reset_min_msecs;
4082 bnxt_dev_cleanup(bp);
4084 bnxt_wait_for_device_shutdown(bp);
4086 /* During some fatal firmware error conditions, the PCI config space
4087 * register 0x2e which normally contains the subsystem ID will become
4088 * 0xffff. This register will revert back to the normal value after
4089 * the chip has completed core reset. If we detect this condition,
4090 * we can poll this config register immediately for the value to revert.
4092 if (bp->flags & BNXT_FLAG_FATAL_ERROR) {
4093 rc = rte_pci_read_config(bp->pdev, &val, sizeof(val), PCI_SUBSYSTEM_ID_OFFSET);
4095 PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x", PCI_SUBSYSTEM_ID_OFFSET);
4098 if (val == 0xffff) {
4099 bp->fw_reset_min_msecs = 0;
4104 rc = rte_eal_alarm_set(us, bnxt_dev_recover, (void *)bp);
4106 PMD_DRV_LOG(ERR, "Error setting recovery alarm");
4109 uint32_t bnxt_read_fw_status_reg(struct bnxt *bp, uint32_t index)
4111 struct bnxt_error_recovery_info *info = bp->recovery_info;
4112 uint32_t reg = info->status_regs[index];
4113 uint32_t type, offset, val = 0;
4115 type = BNXT_FW_STATUS_REG_TYPE(reg);
4116 offset = BNXT_FW_STATUS_REG_OFF(reg);
4119 case BNXT_FW_STATUS_REG_TYPE_CFG:
4120 rte_pci_read_config(bp->pdev, &val, sizeof(val), offset);
4122 case BNXT_FW_STATUS_REG_TYPE_GRC:
4123 offset = info->mapped_status_regs[index];
4125 case BNXT_FW_STATUS_REG_TYPE_BAR0:
4126 val = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
4134 static int bnxt_fw_reset_all(struct bnxt *bp)
4136 struct bnxt_error_recovery_info *info = bp->recovery_info;
4140 if (info->flags & BNXT_FLAG_ERROR_RECOVERY_HOST) {
4141 /* Reset through master function driver */
4142 for (i = 0; i < info->reg_array_cnt; i++)
4143 bnxt_write_fw_reset_reg(bp, i);
4144 /* Wait for time specified by FW after triggering reset */
4145 rte_delay_ms(info->master_func_wait_period_after_reset);
4146 } else if (info->flags & BNXT_FLAG_ERROR_RECOVERY_CO_CPU) {
4147 /* Reset with the help of Kong processor */
4148 rc = bnxt_hwrm_fw_reset(bp);
4150 PMD_DRV_LOG(ERR, "Failed to reset FW\n");
4156 static void bnxt_fw_reset_cb(void *arg)
4158 struct bnxt *bp = arg;
4159 struct bnxt_error_recovery_info *info = bp->recovery_info;
4162 /* Only Master function can do FW reset */
4163 if (bnxt_is_master_func(bp) &&
4164 bnxt_is_recovery_enabled(bp)) {
4165 rc = bnxt_fw_reset_all(bp);
4167 PMD_DRV_LOG(ERR, "Adapter recovery failed\n");
4172 /* if recovery method is ERROR_RECOVERY_CO_CPU, KONG will send
4173 * EXCEPTION_FATAL_ASYNC event to all the functions
4174 * (including MASTER FUNC). After receiving this Async, all the active
4175 * drivers should treat this case as FW initiated recovery
4177 if (info->flags & BNXT_FLAG_ERROR_RECOVERY_HOST) {
4178 bp->fw_reset_min_msecs = BNXT_MIN_FW_READY_TIMEOUT;
4179 bp->fw_reset_max_msecs = BNXT_MAX_FW_RESET_TIMEOUT;
4181 /* To recover from error */
4182 rte_eal_alarm_set(US_PER_MS, bnxt_dev_reset_and_resume,
4187 /* Driver should poll FW heartbeat, reset_counter with the frequency
4188 * advertised by FW in HWRM_ERROR_RECOVERY_QCFG.
4189 * When the driver detects heartbeat stop or change in reset_counter,
4190 * it has to trigger a reset to recover from the error condition.
4191 * A “master PF” is the function who will have the privilege to
4192 * initiate the chimp reset. The master PF will be elected by the
4193 * firmware and will be notified through async message.
4195 static void bnxt_check_fw_health(void *arg)
4197 struct bnxt *bp = arg;
4198 struct bnxt_error_recovery_info *info = bp->recovery_info;
4199 uint32_t val = 0, wait_msec;
4201 if (!info || !bnxt_is_recovery_enabled(bp) ||
4202 is_bnxt_in_error(bp))
4205 val = bnxt_read_fw_status_reg(bp, BNXT_FW_HEARTBEAT_CNT_REG);
4206 if (val == info->last_heart_beat)
4209 info->last_heart_beat = val;
4211 val = bnxt_read_fw_status_reg(bp, BNXT_FW_RECOVERY_CNT_REG);
4212 if (val != info->last_reset_counter)
4215 info->last_reset_counter = val;
4217 rte_eal_alarm_set(US_PER_MS * info->driver_polling_freq,
4218 bnxt_check_fw_health, (void *)bp);
4222 /* Stop DMA to/from device */
4223 bp->flags |= BNXT_FLAG_FATAL_ERROR;
4224 bp->flags |= BNXT_FLAG_FW_RESET;
4226 PMD_DRV_LOG(ERR, "Detected FW dead condition\n");
4228 if (bnxt_is_master_func(bp))
4229 wait_msec = info->master_func_wait_period;
4231 wait_msec = info->normal_func_wait_period;
4233 rte_eal_alarm_set(US_PER_MS * wait_msec,
4234 bnxt_fw_reset_cb, (void *)bp);
4237 void bnxt_schedule_fw_health_check(struct bnxt *bp)
4239 uint32_t polling_freq;
4241 pthread_mutex_lock(&bp->health_check_lock);
4243 if (!bnxt_is_recovery_enabled(bp))
4246 if (bp->flags & BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED)
4249 polling_freq = bp->recovery_info->driver_polling_freq;
4251 rte_eal_alarm_set(US_PER_MS * polling_freq,
4252 bnxt_check_fw_health, (void *)bp);
4253 bp->flags |= BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED;
4256 pthread_mutex_unlock(&bp->health_check_lock);
4259 static void bnxt_cancel_fw_health_check(struct bnxt *bp)
4261 if (!bnxt_is_recovery_enabled(bp))
4264 rte_eal_alarm_cancel(bnxt_check_fw_health, (void *)bp);
4265 bp->flags &= ~BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED;
4268 static bool bnxt_vf_pciid(uint16_t device_id)
4270 switch (device_id) {
4271 case BROADCOM_DEV_ID_57304_VF:
4272 case BROADCOM_DEV_ID_57406_VF:
4273 case BROADCOM_DEV_ID_5731X_VF:
4274 case BROADCOM_DEV_ID_5741X_VF:
4275 case BROADCOM_DEV_ID_57414_VF:
4276 case BROADCOM_DEV_ID_STRATUS_NIC_VF1:
4277 case BROADCOM_DEV_ID_STRATUS_NIC_VF2:
4278 case BROADCOM_DEV_ID_58802_VF:
4279 case BROADCOM_DEV_ID_57500_VF1:
4280 case BROADCOM_DEV_ID_57500_VF2:
4281 case BROADCOM_DEV_ID_58818_VF:
4289 /* Phase 5 device */
4290 static bool bnxt_p5_device(uint16_t device_id)
4292 switch (device_id) {
4293 case BROADCOM_DEV_ID_57508:
4294 case BROADCOM_DEV_ID_57504:
4295 case BROADCOM_DEV_ID_57502:
4296 case BROADCOM_DEV_ID_57508_MF1:
4297 case BROADCOM_DEV_ID_57504_MF1:
4298 case BROADCOM_DEV_ID_57502_MF1:
4299 case BROADCOM_DEV_ID_57508_MF2:
4300 case BROADCOM_DEV_ID_57504_MF2:
4301 case BROADCOM_DEV_ID_57502_MF2:
4302 case BROADCOM_DEV_ID_57500_VF1:
4303 case BROADCOM_DEV_ID_57500_VF2:
4304 case BROADCOM_DEV_ID_58812:
4305 case BROADCOM_DEV_ID_58814:
4306 case BROADCOM_DEV_ID_58818:
4307 case BROADCOM_DEV_ID_58818_VF:
4315 bool bnxt_stratus_device(struct bnxt *bp)
4317 uint16_t device_id = bp->pdev->id.device_id;
4319 switch (device_id) {
4320 case BROADCOM_DEV_ID_STRATUS_NIC:
4321 case BROADCOM_DEV_ID_STRATUS_NIC_VF1:
4322 case BROADCOM_DEV_ID_STRATUS_NIC_VF2:
4330 static int bnxt_map_pci_bars(struct rte_eth_dev *eth_dev)
4332 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
4333 struct bnxt *bp = eth_dev->data->dev_private;
4335 /* enable device (incl. PCI PM wakeup), and bus-mastering */
4336 bp->bar0 = (void *)pci_dev->mem_resource[0].addr;
4337 bp->doorbell_base = (void *)pci_dev->mem_resource[2].addr;
4338 if (!bp->bar0 || !bp->doorbell_base) {
4339 PMD_DRV_LOG(ERR, "Unable to access Hardware\n");
4343 bp->eth_dev = eth_dev;
4349 static int bnxt_alloc_ctx_mem_blk(struct bnxt *bp,
4350 struct bnxt_ctx_pg_info *ctx_pg,
4355 struct bnxt_ring_mem_info *rmem = &ctx_pg->ring_mem;
4356 const struct rte_memzone *mz = NULL;
4357 char mz_name[RTE_MEMZONE_NAMESIZE];
4358 rte_iova_t mz_phys_addr;
4359 uint64_t valid_bits = 0;
4366 rmem->nr_pages = RTE_ALIGN_MUL_CEIL(mem_size, BNXT_PAGE_SIZE) /
4368 rmem->page_size = BNXT_PAGE_SIZE;
4369 rmem->pg_arr = ctx_pg->ctx_pg_arr;
4370 rmem->dma_arr = ctx_pg->ctx_dma_arr;
4371 rmem->flags = BNXT_RMEM_VALID_PTE_FLAG;
4373 valid_bits = PTU_PTE_VALID;
4375 if (rmem->nr_pages > 1) {
4376 snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4377 "bnxt_ctx_pg_tbl%s_%x_%d",
4378 suffix, idx, bp->eth_dev->data->port_id);
4379 mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4380 mz = rte_memzone_lookup(mz_name);
4382 mz = rte_memzone_reserve_aligned(mz_name,
4386 RTE_MEMZONE_SIZE_HINT_ONLY |
4387 RTE_MEMZONE_IOVA_CONTIG,
4393 memset(mz->addr, 0, mz->len);
4394 mz_phys_addr = mz->iova;
4396 rmem->pg_tbl = mz->addr;
4397 rmem->pg_tbl_map = mz_phys_addr;
4398 rmem->pg_tbl_mz = mz;
4401 snprintf(mz_name, RTE_MEMZONE_NAMESIZE, "bnxt_ctx_%s_%x_%d",
4402 suffix, idx, bp->eth_dev->data->port_id);
4403 mz = rte_memzone_lookup(mz_name);
4405 mz = rte_memzone_reserve_aligned(mz_name,
4409 RTE_MEMZONE_SIZE_HINT_ONLY |
4410 RTE_MEMZONE_IOVA_CONTIG,
4416 memset(mz->addr, 0, mz->len);
4417 mz_phys_addr = mz->iova;
4419 for (sz = 0, i = 0; sz < mem_size; sz += BNXT_PAGE_SIZE, i++) {
4420 rmem->pg_arr[i] = ((char *)mz->addr) + sz;
4421 rmem->dma_arr[i] = mz_phys_addr + sz;
4423 if (rmem->nr_pages > 1) {
4424 if (i == rmem->nr_pages - 2 &&
4425 (rmem->flags & BNXT_RMEM_RING_PTE_FLAG))
4426 valid_bits |= PTU_PTE_NEXT_TO_LAST;
4427 else if (i == rmem->nr_pages - 1 &&
4428 (rmem->flags & BNXT_RMEM_RING_PTE_FLAG))
4429 valid_bits |= PTU_PTE_LAST;
4431 rmem->pg_tbl[i] = rte_cpu_to_le_64(rmem->dma_arr[i] |
4437 if (rmem->vmem_size)
4438 rmem->vmem = (void **)mz->addr;
4439 rmem->dma_arr[0] = mz_phys_addr;
4443 static void bnxt_free_ctx_mem(struct bnxt *bp)
4447 if (!bp->ctx || !(bp->ctx->flags & BNXT_CTX_FLAG_INITED))
4450 bp->ctx->flags &= ~BNXT_CTX_FLAG_INITED;
4451 rte_memzone_free(bp->ctx->qp_mem.ring_mem.mz);
4452 rte_memzone_free(bp->ctx->srq_mem.ring_mem.mz);
4453 rte_memzone_free(bp->ctx->cq_mem.ring_mem.mz);
4454 rte_memzone_free(bp->ctx->vnic_mem.ring_mem.mz);
4455 rte_memzone_free(bp->ctx->stat_mem.ring_mem.mz);
4456 rte_memzone_free(bp->ctx->qp_mem.ring_mem.pg_tbl_mz);
4457 rte_memzone_free(bp->ctx->srq_mem.ring_mem.pg_tbl_mz);
4458 rte_memzone_free(bp->ctx->cq_mem.ring_mem.pg_tbl_mz);
4459 rte_memzone_free(bp->ctx->vnic_mem.ring_mem.pg_tbl_mz);
4460 rte_memzone_free(bp->ctx->stat_mem.ring_mem.pg_tbl_mz);
4462 for (i = 0; i < bp->ctx->tqm_fp_rings_count + 1; i++) {
4463 if (bp->ctx->tqm_mem[i])
4464 rte_memzone_free(bp->ctx->tqm_mem[i]->ring_mem.mz);
4471 #define bnxt_roundup(x, y) ((((x) + ((y) - 1)) / (y)) * (y))
4473 #define min_t(type, x, y) ({ \
4474 type __min1 = (x); \
4475 type __min2 = (y); \
4476 __min1 < __min2 ? __min1 : __min2; })
4478 #define max_t(type, x, y) ({ \
4479 type __max1 = (x); \
4480 type __max2 = (y); \
4481 __max1 > __max2 ? __max1 : __max2; })
4483 #define clamp_t(type, _x, min, max) min_t(type, max_t(type, _x, min), max)
4485 int bnxt_alloc_ctx_mem(struct bnxt *bp)
4487 struct bnxt_ctx_pg_info *ctx_pg;
4488 struct bnxt_ctx_mem_info *ctx;
4489 uint32_t mem_size, ena, entries;
4490 uint32_t entries_sp, min;
4493 rc = bnxt_hwrm_func_backing_store_qcaps(bp);
4495 PMD_DRV_LOG(ERR, "Query context mem capability failed\n");
4499 if (!ctx || (ctx->flags & BNXT_CTX_FLAG_INITED))
4502 ctx_pg = &ctx->qp_mem;
4503 ctx_pg->entries = ctx->qp_min_qp1_entries + ctx->qp_max_l2_entries;
4504 if (ctx->qp_entry_size) {
4505 mem_size = ctx->qp_entry_size * ctx_pg->entries;
4506 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "qp_mem", 0);
4511 ctx_pg = &ctx->srq_mem;
4512 ctx_pg->entries = ctx->srq_max_l2_entries;
4513 if (ctx->srq_entry_size) {
4514 mem_size = ctx->srq_entry_size * ctx_pg->entries;
4515 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "srq_mem", 0);
4520 ctx_pg = &ctx->cq_mem;
4521 ctx_pg->entries = ctx->cq_max_l2_entries;
4522 if (ctx->cq_entry_size) {
4523 mem_size = ctx->cq_entry_size * ctx_pg->entries;
4524 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "cq_mem", 0);
4529 ctx_pg = &ctx->vnic_mem;
4530 ctx_pg->entries = ctx->vnic_max_vnic_entries +
4531 ctx->vnic_max_ring_table_entries;
4532 if (ctx->vnic_entry_size) {
4533 mem_size = ctx->vnic_entry_size * ctx_pg->entries;
4534 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "vnic_mem", 0);
4539 ctx_pg = &ctx->stat_mem;
4540 ctx_pg->entries = ctx->stat_max_entries;
4541 if (ctx->stat_entry_size) {
4542 mem_size = ctx->stat_entry_size * ctx_pg->entries;
4543 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "stat_mem", 0);
4548 min = ctx->tqm_min_entries_per_ring;
4550 entries_sp = ctx->qp_max_l2_entries +
4551 ctx->vnic_max_vnic_entries +
4552 2 * ctx->qp_min_qp1_entries + min;
4553 entries_sp = bnxt_roundup(entries_sp, ctx->tqm_entries_multiple);
4555 entries = ctx->qp_max_l2_entries + ctx->qp_min_qp1_entries;
4556 entries = bnxt_roundup(entries, ctx->tqm_entries_multiple);
4557 entries = clamp_t(uint32_t, entries, min,
4558 ctx->tqm_max_entries_per_ring);
4559 for (i = 0, ena = 0; i < ctx->tqm_fp_rings_count + 1; i++) {
4560 /* i=0 is for TQM_SP. i=1 to i=8 applies to RING0 to RING7.
4561 * i > 8 is other ext rings.
4563 ctx_pg = ctx->tqm_mem[i];
4564 ctx_pg->entries = i ? entries : entries_sp;
4565 if (ctx->tqm_entry_size) {
4566 mem_size = ctx->tqm_entry_size * ctx_pg->entries;
4567 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size,
4572 if (i < BNXT_MAX_TQM_LEGACY_RINGS)
4573 ena |= HWRM_FUNC_BACKING_STORE_CFG_INPUT_ENABLES_TQM_SP << i;
4575 ena |= HWRM_FUNC_BACKING_STORE_CFG_INPUT_ENABLES_TQM_RING8;
4578 ena |= FUNC_BACKING_STORE_CFG_INPUT_DFLT_ENABLES;
4579 rc = bnxt_hwrm_func_backing_store_cfg(bp, ena);
4582 "Failed to configure context mem: rc = %d\n", rc);
4584 ctx->flags |= BNXT_CTX_FLAG_INITED;
4589 static int bnxt_alloc_stats_mem(struct bnxt *bp)
4591 struct rte_pci_device *pci_dev = bp->pdev;
4592 char mz_name[RTE_MEMZONE_NAMESIZE];
4593 const struct rte_memzone *mz = NULL;
4594 uint32_t total_alloc_len;
4595 rte_iova_t mz_phys_addr;
4597 if (pci_dev->id.device_id == BROADCOM_DEV_ID_NS2)
4600 snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4601 "bnxt_" PCI_PRI_FMT "-%s", pci_dev->addr.domain,
4602 pci_dev->addr.bus, pci_dev->addr.devid,
4603 pci_dev->addr.function, "rx_port_stats");
4604 mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4605 mz = rte_memzone_lookup(mz_name);
4607 RTE_CACHE_LINE_ROUNDUP(sizeof(struct rx_port_stats) +
4608 sizeof(struct rx_port_stats_ext) + 512);
4610 mz = rte_memzone_reserve(mz_name, total_alloc_len,
4613 RTE_MEMZONE_SIZE_HINT_ONLY |
4614 RTE_MEMZONE_IOVA_CONTIG);
4618 memset(mz->addr, 0, mz->len);
4619 mz_phys_addr = mz->iova;
4621 bp->rx_mem_zone = (const void *)mz;
4622 bp->hw_rx_port_stats = mz->addr;
4623 bp->hw_rx_port_stats_map = mz_phys_addr;
4625 snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4626 "bnxt_" PCI_PRI_FMT "-%s", pci_dev->addr.domain,
4627 pci_dev->addr.bus, pci_dev->addr.devid,
4628 pci_dev->addr.function, "tx_port_stats");
4629 mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4630 mz = rte_memzone_lookup(mz_name);
4632 RTE_CACHE_LINE_ROUNDUP(sizeof(struct tx_port_stats) +
4633 sizeof(struct tx_port_stats_ext) + 512);
4635 mz = rte_memzone_reserve(mz_name,
4639 RTE_MEMZONE_SIZE_HINT_ONLY |
4640 RTE_MEMZONE_IOVA_CONTIG);
4644 memset(mz->addr, 0, mz->len);
4645 mz_phys_addr = mz->iova;
4647 bp->tx_mem_zone = (const void *)mz;
4648 bp->hw_tx_port_stats = mz->addr;
4649 bp->hw_tx_port_stats_map = mz_phys_addr;
4650 bp->flags |= BNXT_FLAG_PORT_STATS;
4652 /* Display extended statistics if FW supports it */
4653 if (bp->hwrm_spec_code < HWRM_SPEC_CODE_1_8_4 ||
4654 bp->hwrm_spec_code == HWRM_SPEC_CODE_1_9_0 ||
4655 !(bp->flags & BNXT_FLAG_EXT_STATS_SUPPORTED))
4658 bp->hw_rx_port_stats_ext = (void *)
4659 ((uint8_t *)bp->hw_rx_port_stats +
4660 sizeof(struct rx_port_stats));
4661 bp->hw_rx_port_stats_ext_map = bp->hw_rx_port_stats_map +
4662 sizeof(struct rx_port_stats);
4663 bp->flags |= BNXT_FLAG_EXT_RX_PORT_STATS;
4665 if (bp->hwrm_spec_code < HWRM_SPEC_CODE_1_9_2 ||
4666 bp->flags & BNXT_FLAG_EXT_STATS_SUPPORTED) {
4667 bp->hw_tx_port_stats_ext = (void *)
4668 ((uint8_t *)bp->hw_tx_port_stats +
4669 sizeof(struct tx_port_stats));
4670 bp->hw_tx_port_stats_ext_map =
4671 bp->hw_tx_port_stats_map +
4672 sizeof(struct tx_port_stats);
4673 bp->flags |= BNXT_FLAG_EXT_TX_PORT_STATS;
4679 static int bnxt_setup_mac_addr(struct rte_eth_dev *eth_dev)
4681 struct bnxt *bp = eth_dev->data->dev_private;
4684 eth_dev->data->mac_addrs = rte_zmalloc("bnxt_mac_addr_tbl",
4685 RTE_ETHER_ADDR_LEN *
4688 if (eth_dev->data->mac_addrs == NULL) {
4689 PMD_DRV_LOG(ERR, "Failed to alloc MAC addr tbl\n");
4693 if (!BNXT_HAS_DFLT_MAC_SET(bp)) {
4697 /* Generate a random MAC address, if none was assigned by PF */
4698 PMD_DRV_LOG(INFO, "VF MAC address not assigned by Host PF\n");
4699 bnxt_eth_hw_addr_random(bp->mac_addr);
4701 "Assign random MAC:%02X:%02X:%02X:%02X:%02X:%02X\n",
4702 bp->mac_addr[0], bp->mac_addr[1], bp->mac_addr[2],
4703 bp->mac_addr[3], bp->mac_addr[4], bp->mac_addr[5]);
4705 rc = bnxt_hwrm_set_mac(bp);
4710 /* Copy the permanent MAC from the FUNC_QCAPS response */
4711 memcpy(ð_dev->data->mac_addrs[0], bp->mac_addr, RTE_ETHER_ADDR_LEN);
4716 static int bnxt_restore_dflt_mac(struct bnxt *bp)
4720 /* MAC is already configured in FW */
4721 if (BNXT_HAS_DFLT_MAC_SET(bp))
4724 /* Restore the old MAC configured */
4725 rc = bnxt_hwrm_set_mac(bp);
4727 PMD_DRV_LOG(ERR, "Failed to restore MAC address\n");
4732 static void bnxt_config_vf_req_fwd(struct bnxt *bp)
4737 memset(bp->pf->vf_req_fwd, 0, sizeof(bp->pf->vf_req_fwd));
4739 if (!(bp->fw_cap & BNXT_FW_CAP_LINK_ADMIN))
4740 BNXT_HWRM_CMD_TO_FORWARD(HWRM_PORT_PHY_QCFG);
4741 BNXT_HWRM_CMD_TO_FORWARD(HWRM_FUNC_CFG);
4742 BNXT_HWRM_CMD_TO_FORWARD(HWRM_FUNC_VF_CFG);
4743 BNXT_HWRM_CMD_TO_FORWARD(HWRM_CFA_L2_FILTER_ALLOC);
4744 BNXT_HWRM_CMD_TO_FORWARD(HWRM_OEM_CMD);
4748 bnxt_get_svif(uint16_t port_id, bool func_svif,
4749 enum bnxt_ulp_intf_type type)
4751 struct rte_eth_dev *eth_dev;
4754 eth_dev = &rte_eth_devices[port_id];
4755 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
4756 struct bnxt_representor *vfr = eth_dev->data->dev_private;
4760 if (type == BNXT_ULP_INTF_TYPE_VF_REP)
4763 eth_dev = vfr->parent_dev;
4766 bp = eth_dev->data->dev_private;
4768 return func_svif ? bp->func_svif : bp->port_svif;
4772 bnxt_get_vnic_id(uint16_t port, enum bnxt_ulp_intf_type type)
4774 struct rte_eth_dev *eth_dev;
4775 struct bnxt_vnic_info *vnic;
4778 eth_dev = &rte_eth_devices[port];
4779 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
4780 struct bnxt_representor *vfr = eth_dev->data->dev_private;
4784 if (type == BNXT_ULP_INTF_TYPE_VF_REP)
4785 return vfr->dflt_vnic_id;
4787 eth_dev = vfr->parent_dev;
4790 bp = eth_dev->data->dev_private;
4792 vnic = BNXT_GET_DEFAULT_VNIC(bp);
4794 return vnic->fw_vnic_id;
4798 bnxt_get_fw_func_id(uint16_t port, enum bnxt_ulp_intf_type type)
4800 struct rte_eth_dev *eth_dev;
4803 eth_dev = &rte_eth_devices[port];
4804 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
4805 struct bnxt_representor *vfr = eth_dev->data->dev_private;
4809 if (type == BNXT_ULP_INTF_TYPE_VF_REP)
4812 eth_dev = vfr->parent_dev;
4815 bp = eth_dev->data->dev_private;
4820 enum bnxt_ulp_intf_type
4821 bnxt_get_interface_type(uint16_t port)
4823 struct rte_eth_dev *eth_dev;
4826 eth_dev = &rte_eth_devices[port];
4827 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev))
4828 return BNXT_ULP_INTF_TYPE_VF_REP;
4830 bp = eth_dev->data->dev_private;
4832 return BNXT_ULP_INTF_TYPE_PF;
4833 else if (BNXT_VF_IS_TRUSTED(bp))
4834 return BNXT_ULP_INTF_TYPE_TRUSTED_VF;
4835 else if (BNXT_VF(bp))
4836 return BNXT_ULP_INTF_TYPE_VF;
4838 return BNXT_ULP_INTF_TYPE_INVALID;
4842 bnxt_get_phy_port_id(uint16_t port_id)
4844 struct bnxt_representor *vfr;
4845 struct rte_eth_dev *eth_dev;
4848 eth_dev = &rte_eth_devices[port_id];
4849 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
4850 vfr = eth_dev->data->dev_private;
4854 eth_dev = vfr->parent_dev;
4857 bp = eth_dev->data->dev_private;
4859 return BNXT_PF(bp) ? bp->pf->port_id : bp->parent->port_id;
4863 bnxt_get_parif(uint16_t port_id, enum bnxt_ulp_intf_type type)
4865 struct rte_eth_dev *eth_dev;
4868 eth_dev = &rte_eth_devices[port_id];
4869 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
4870 struct bnxt_representor *vfr = eth_dev->data->dev_private;
4874 if (type == BNXT_ULP_INTF_TYPE_VF_REP)
4875 return vfr->fw_fid - 1;
4877 eth_dev = vfr->parent_dev;
4880 bp = eth_dev->data->dev_private;
4882 return BNXT_PF(bp) ? bp->fw_fid - 1 : bp->parent->fid - 1;
4886 bnxt_get_vport(uint16_t port_id)
4888 return (1 << bnxt_get_phy_port_id(port_id));
4891 static void bnxt_alloc_error_recovery_info(struct bnxt *bp)
4893 struct bnxt_error_recovery_info *info = bp->recovery_info;
4896 if (!(bp->fw_cap & BNXT_FW_CAP_HCOMM_FW_STATUS))
4897 memset(info, 0, sizeof(*info));
4901 if (!(bp->fw_cap & BNXT_FW_CAP_ERROR_RECOVERY))
4904 info = rte_zmalloc("bnxt_hwrm_error_recovery_qcfg",
4907 bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
4909 bp->recovery_info = info;
4912 static void bnxt_check_fw_status(struct bnxt *bp)
4916 if (!(bp->recovery_info &&
4917 (bp->fw_cap & BNXT_FW_CAP_HCOMM_FW_STATUS)))
4920 fw_status = bnxt_read_fw_status_reg(bp, BNXT_FW_STATUS_REG);
4921 if (fw_status != BNXT_FW_STATUS_HEALTHY)
4922 PMD_DRV_LOG(ERR, "Firmware not responding, status: %#x\n",
4926 static int bnxt_map_hcomm_fw_status_reg(struct bnxt *bp)
4928 struct bnxt_error_recovery_info *info = bp->recovery_info;
4929 uint32_t status_loc;
4932 rte_write32(HCOMM_STATUS_STRUCT_LOC, (uint8_t *)bp->bar0 +
4933 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
4934 sig_ver = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
4935 BNXT_GRCP_WINDOW_2_BASE +
4936 offsetof(struct hcomm_status,
4938 /* If the signature is absent, then FW does not support this feature */
4939 if ((sig_ver & HCOMM_STATUS_SIGNATURE_MASK) !=
4940 HCOMM_STATUS_SIGNATURE_VAL)
4944 info = rte_zmalloc("bnxt_hwrm_error_recovery_qcfg",
4948 bp->recovery_info = info;
4950 memset(info, 0, sizeof(*info));
4953 status_loc = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
4954 BNXT_GRCP_WINDOW_2_BASE +
4955 offsetof(struct hcomm_status,
4958 /* Only pre-map the FW health status GRC register */
4959 if (BNXT_FW_STATUS_REG_TYPE(status_loc) != BNXT_FW_STATUS_REG_TYPE_GRC)
4962 info->status_regs[BNXT_FW_STATUS_REG] = status_loc;
4963 info->mapped_status_regs[BNXT_FW_STATUS_REG] =
4964 BNXT_GRCP_WINDOW_2_BASE + (status_loc & BNXT_GRCP_OFFSET_MASK);
4966 rte_write32((status_loc & BNXT_GRCP_BASE_MASK), (uint8_t *)bp->bar0 +
4967 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
4969 bp->fw_cap |= BNXT_FW_CAP_HCOMM_FW_STATUS;
4974 /* This function gets the FW version along with the
4975 * capabilities(MAX and current) of the function, vnic,
4976 * error recovery, phy and other chip related info
4978 static int bnxt_get_config(struct bnxt *bp)
4985 rc = bnxt_map_hcomm_fw_status_reg(bp);
4989 rc = bnxt_hwrm_ver_get(bp, DFLT_HWRM_CMD_TIMEOUT);
4991 bnxt_check_fw_status(bp);
4995 rc = bnxt_hwrm_func_reset(bp);
4999 rc = bnxt_hwrm_vnic_qcaps(bp);
5003 rc = bnxt_hwrm_queue_qportcfg(bp);
5007 /* Get the MAX capabilities for this function.
5008 * This function also allocates context memory for TQM rings and
5009 * informs the firmware about this allocated backing store memory.
5011 rc = bnxt_hwrm_func_qcaps(bp);
5015 rc = bnxt_hwrm_func_qcfg(bp, &mtu);
5019 rc = bnxt_hwrm_cfa_adv_flow_mgmt_qcaps(bp);
5023 bnxt_hwrm_port_mac_qcfg(bp);
5025 bnxt_hwrm_parent_pf_qcfg(bp);
5027 bnxt_hwrm_port_phy_qcaps(bp);
5029 bnxt_alloc_error_recovery_info(bp);
5030 /* Get the adapter error recovery support info */
5031 rc = bnxt_hwrm_error_recovery_qcfg(bp);
5033 bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5035 bnxt_hwrm_port_led_qcaps(bp);
5041 bnxt_init_locks(struct bnxt *bp)
5045 err = pthread_mutex_init(&bp->flow_lock, NULL);
5047 PMD_DRV_LOG(ERR, "Unable to initialize flow_lock\n");
5051 err = pthread_mutex_init(&bp->def_cp_lock, NULL);
5053 PMD_DRV_LOG(ERR, "Unable to initialize def_cp_lock\n");
5057 err = pthread_mutex_init(&bp->health_check_lock, NULL);
5059 PMD_DRV_LOG(ERR, "Unable to initialize health_check_lock\n");
5063 err = pthread_mutex_init(&bp->err_recovery_lock, NULL);
5065 PMD_DRV_LOG(ERR, "Unable to initialize err_recovery_lock\n");
5070 static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev)
5074 rc = bnxt_get_config(bp);
5078 if (!reconfig_dev) {
5079 rc = bnxt_setup_mac_addr(bp->eth_dev);
5083 rc = bnxt_restore_dflt_mac(bp);
5088 bnxt_config_vf_req_fwd(bp);
5090 rc = bnxt_hwrm_func_driver_register(bp);
5092 PMD_DRV_LOG(ERR, "Failed to register driver");
5097 if (bp->pdev->max_vfs) {
5098 rc = bnxt_hwrm_allocate_vfs(bp, bp->pdev->max_vfs);
5100 PMD_DRV_LOG(ERR, "Failed to allocate VFs\n");
5104 rc = bnxt_hwrm_allocate_pf_only(bp);
5107 "Failed to allocate PF resources");
5113 rc = bnxt_alloc_mem(bp, reconfig_dev);
5117 rc = bnxt_setup_int(bp);
5121 rc = bnxt_request_int(bp);
5125 rc = bnxt_init_ctx_mem(bp);
5127 PMD_DRV_LOG(ERR, "Failed to init adv_flow_counters\n");
5135 bnxt_parse_devarg_truflow(__rte_unused const char *key,
5136 const char *value, void *opaque_arg)
5138 struct bnxt *bp = opaque_arg;
5139 unsigned long truflow;
5142 if (!value || !opaque_arg) {
5144 "Invalid parameter passed to truflow devargs.\n");
5148 truflow = strtoul(value, &end, 10);
5149 if (end == NULL || *end != '\0' ||
5150 (truflow == ULONG_MAX && errno == ERANGE)) {
5152 "Invalid parameter passed to truflow devargs.\n");
5156 if (BNXT_DEVARG_TRUFLOW_INVALID(truflow)) {
5158 "Invalid value passed to truflow devargs.\n");
5163 bp->flags |= BNXT_FLAG_TRUFLOW_EN;
5164 PMD_DRV_LOG(INFO, "Host-based truflow feature enabled.\n");
5166 bp->flags &= ~BNXT_FLAG_TRUFLOW_EN;
5167 PMD_DRV_LOG(INFO, "Host-based truflow feature disabled.\n");
5174 bnxt_parse_devarg_flow_xstat(__rte_unused const char *key,
5175 const char *value, void *opaque_arg)
5177 struct bnxt *bp = opaque_arg;
5178 unsigned long flow_xstat;
5181 if (!value || !opaque_arg) {
5183 "Invalid parameter passed to flow_xstat devarg.\n");
5187 flow_xstat = strtoul(value, &end, 10);
5188 if (end == NULL || *end != '\0' ||
5189 (flow_xstat == ULONG_MAX && errno == ERANGE)) {
5191 "Invalid parameter passed to flow_xstat devarg.\n");
5195 if (BNXT_DEVARG_FLOW_XSTAT_INVALID(flow_xstat)) {
5197 "Invalid value passed to flow_xstat devarg.\n");
5201 bp->flags |= BNXT_FLAG_FLOW_XSTATS_EN;
5202 if (BNXT_FLOW_XSTATS_EN(bp))
5203 PMD_DRV_LOG(INFO, "flow_xstat feature enabled.\n");
5209 bnxt_parse_devarg_max_num_kflows(__rte_unused const char *key,
5210 const char *value, void *opaque_arg)
5212 struct bnxt *bp = opaque_arg;
5213 unsigned long max_num_kflows;
5216 if (!value || !opaque_arg) {
5218 "Invalid parameter passed to max_num_kflows devarg.\n");
5222 max_num_kflows = strtoul(value, &end, 10);
5223 if (end == NULL || *end != '\0' ||
5224 (max_num_kflows == ULONG_MAX && errno == ERANGE)) {
5226 "Invalid parameter passed to max_num_kflows devarg.\n");
5230 if (bnxt_devarg_max_num_kflow_invalid(max_num_kflows)) {
5232 "Invalid value passed to max_num_kflows devarg.\n");
5236 bp->max_num_kflows = max_num_kflows;
5237 if (bp->max_num_kflows)
5238 PMD_DRV_LOG(INFO, "max_num_kflows set as %ldK.\n",
5245 bnxt_parse_devarg_rep_is_pf(__rte_unused const char *key,
5246 const char *value, void *opaque_arg)
5248 struct bnxt_representor *vfr_bp = opaque_arg;
5249 unsigned long rep_is_pf;
5252 if (!value || !opaque_arg) {
5254 "Invalid parameter passed to rep_is_pf devargs.\n");
5258 rep_is_pf = strtoul(value, &end, 10);
5259 if (end == NULL || *end != '\0' ||
5260 (rep_is_pf == ULONG_MAX && errno == ERANGE)) {
5262 "Invalid parameter passed to rep_is_pf devargs.\n");
5266 if (BNXT_DEVARG_REP_IS_PF_INVALID(rep_is_pf)) {
5268 "Invalid value passed to rep_is_pf devargs.\n");
5272 vfr_bp->flags |= rep_is_pf;
5273 if (BNXT_REP_PF(vfr_bp))
5274 PMD_DRV_LOG(INFO, "PF representor\n");
5276 PMD_DRV_LOG(INFO, "VF representor\n");
5282 bnxt_parse_devarg_rep_based_pf(__rte_unused const char *key,
5283 const char *value, void *opaque_arg)
5285 struct bnxt_representor *vfr_bp = opaque_arg;
5286 unsigned long rep_based_pf;
5289 if (!value || !opaque_arg) {
5291 "Invalid parameter passed to rep_based_pf "
5296 rep_based_pf = strtoul(value, &end, 10);
5297 if (end == NULL || *end != '\0' ||
5298 (rep_based_pf == ULONG_MAX && errno == ERANGE)) {
5300 "Invalid parameter passed to rep_based_pf "
5305 if (BNXT_DEVARG_REP_BASED_PF_INVALID(rep_based_pf)) {
5307 "Invalid value passed to rep_based_pf devargs.\n");
5311 vfr_bp->rep_based_pf = rep_based_pf;
5312 vfr_bp->flags |= BNXT_REP_BASED_PF_VALID;
5314 PMD_DRV_LOG(INFO, "rep-based-pf = %d\n", vfr_bp->rep_based_pf);
5320 bnxt_parse_devarg_rep_q_r2f(__rte_unused const char *key,
5321 const char *value, void *opaque_arg)
5323 struct bnxt_representor *vfr_bp = opaque_arg;
5324 unsigned long rep_q_r2f;
5327 if (!value || !opaque_arg) {
5329 "Invalid parameter passed to rep_q_r2f "
5334 rep_q_r2f = strtoul(value, &end, 10);
5335 if (end == NULL || *end != '\0' ||
5336 (rep_q_r2f == ULONG_MAX && errno == ERANGE)) {
5338 "Invalid parameter passed to rep_q_r2f "
5343 if (BNXT_DEVARG_REP_Q_R2F_INVALID(rep_q_r2f)) {
5345 "Invalid value passed to rep_q_r2f devargs.\n");
5349 vfr_bp->rep_q_r2f = rep_q_r2f;
5350 vfr_bp->flags |= BNXT_REP_Q_R2F_VALID;
5351 PMD_DRV_LOG(INFO, "rep-q-r2f = %d\n", vfr_bp->rep_q_r2f);
5357 bnxt_parse_devarg_rep_q_f2r(__rte_unused const char *key,
5358 const char *value, void *opaque_arg)
5360 struct bnxt_representor *vfr_bp = opaque_arg;
5361 unsigned long rep_q_f2r;
5364 if (!value || !opaque_arg) {
5366 "Invalid parameter passed to rep_q_f2r "
5371 rep_q_f2r = strtoul(value, &end, 10);
5372 if (end == NULL || *end != '\0' ||
5373 (rep_q_f2r == ULONG_MAX && errno == ERANGE)) {
5375 "Invalid parameter passed to rep_q_f2r "
5380 if (BNXT_DEVARG_REP_Q_F2R_INVALID(rep_q_f2r)) {
5382 "Invalid value passed to rep_q_f2r devargs.\n");
5386 vfr_bp->rep_q_f2r = rep_q_f2r;
5387 vfr_bp->flags |= BNXT_REP_Q_F2R_VALID;
5388 PMD_DRV_LOG(INFO, "rep-q-f2r = %d\n", vfr_bp->rep_q_f2r);
5394 bnxt_parse_devarg_rep_fc_r2f(__rte_unused const char *key,
5395 const char *value, void *opaque_arg)
5397 struct bnxt_representor *vfr_bp = opaque_arg;
5398 unsigned long rep_fc_r2f;
5401 if (!value || !opaque_arg) {
5403 "Invalid parameter passed to rep_fc_r2f "
5408 rep_fc_r2f = strtoul(value, &end, 10);
5409 if (end == NULL || *end != '\0' ||
5410 (rep_fc_r2f == ULONG_MAX && errno == ERANGE)) {
5412 "Invalid parameter passed to rep_fc_r2f "
5417 if (BNXT_DEVARG_REP_FC_R2F_INVALID(rep_fc_r2f)) {
5419 "Invalid value passed to rep_fc_r2f devargs.\n");
5423 vfr_bp->flags |= BNXT_REP_FC_R2F_VALID;
5424 vfr_bp->rep_fc_r2f = rep_fc_r2f;
5425 PMD_DRV_LOG(INFO, "rep-fc-r2f = %lu\n", rep_fc_r2f);
5431 bnxt_parse_devarg_rep_fc_f2r(__rte_unused const char *key,
5432 const char *value, void *opaque_arg)
5434 struct bnxt_representor *vfr_bp = opaque_arg;
5435 unsigned long rep_fc_f2r;
5438 if (!value || !opaque_arg) {
5440 "Invalid parameter passed to rep_fc_f2r "
5445 rep_fc_f2r = strtoul(value, &end, 10);
5446 if (end == NULL || *end != '\0' ||
5447 (rep_fc_f2r == ULONG_MAX && errno == ERANGE)) {
5449 "Invalid parameter passed to rep_fc_f2r "
5454 if (BNXT_DEVARG_REP_FC_F2R_INVALID(rep_fc_f2r)) {
5456 "Invalid value passed to rep_fc_f2r devargs.\n");
5460 vfr_bp->flags |= BNXT_REP_FC_F2R_VALID;
5461 vfr_bp->rep_fc_f2r = rep_fc_f2r;
5462 PMD_DRV_LOG(INFO, "rep-fc-f2r = %lu\n", rep_fc_f2r);
5468 bnxt_parse_dev_args(struct bnxt *bp, struct rte_devargs *devargs)
5470 struct rte_kvargs *kvlist;
5473 if (devargs == NULL)
5476 kvlist = rte_kvargs_parse(devargs->args, bnxt_dev_args);
5481 * Handler for "truflow" devarg.
5482 * Invoked as for ex: "-a 0000:00:0d.0,host-based-truflow=1"
5484 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_TRUFLOW,
5485 bnxt_parse_devarg_truflow, bp);
5490 * Handler for "flow_xstat" devarg.
5491 * Invoked as for ex: "-a 0000:00:0d.0,flow_xstat=1"
5493 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_FLOW_XSTAT,
5494 bnxt_parse_devarg_flow_xstat, bp);
5499 * Handler for "max_num_kflows" devarg.
5500 * Invoked as for ex: "-a 000:00:0d.0,max_num_kflows=32"
5502 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_MAX_NUM_KFLOWS,
5503 bnxt_parse_devarg_max_num_kflows, bp);
5508 rte_kvargs_free(kvlist);
5512 static int bnxt_alloc_switch_domain(struct bnxt *bp)
5516 if (BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) {
5517 rc = rte_eth_switch_domain_alloc(&bp->switch_domain_id);
5520 "Failed to alloc switch domain: %d\n", rc);
5523 "Switch domain allocated %d\n",
5524 bp->switch_domain_id);
5530 /* Allocate and initialize various fields in bnxt struct that
5531 * need to be allocated/destroyed only once in the lifetime of the driver
5533 static int bnxt_drv_init(struct rte_eth_dev *eth_dev)
5535 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
5536 struct bnxt *bp = eth_dev->data->dev_private;
5539 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
5541 if (bnxt_vf_pciid(pci_dev->id.device_id))
5542 bp->flags |= BNXT_FLAG_VF;
5544 if (bnxt_p5_device(pci_dev->id.device_id))
5545 bp->flags |= BNXT_FLAG_CHIP_P5;
5547 if (pci_dev->id.device_id == BROADCOM_DEV_ID_58802 ||
5548 pci_dev->id.device_id == BROADCOM_DEV_ID_58804 ||
5549 pci_dev->id.device_id == BROADCOM_DEV_ID_58808 ||
5550 pci_dev->id.device_id == BROADCOM_DEV_ID_58802_VF)
5551 bp->flags |= BNXT_FLAG_STINGRAY;
5553 if (BNXT_TRUFLOW_EN(bp)) {
5554 /* extra mbuf field is required to store CFA code from mark */
5555 static const struct rte_mbuf_dynfield bnxt_cfa_code_dynfield_desc = {
5556 .name = RTE_PMD_BNXT_CFA_CODE_DYNFIELD_NAME,
5557 .size = sizeof(bnxt_cfa_code_dynfield_t),
5558 .align = __alignof__(bnxt_cfa_code_dynfield_t),
5560 bnxt_cfa_code_dynfield_offset =
5561 rte_mbuf_dynfield_register(&bnxt_cfa_code_dynfield_desc);
5562 if (bnxt_cfa_code_dynfield_offset < 0) {
5564 "Failed to register mbuf field for TruFlow mark\n");
5569 rc = bnxt_map_pci_bars(eth_dev);
5572 "Failed to initialize board rc: %x\n", rc);
5576 rc = bnxt_alloc_pf_info(bp);
5580 rc = bnxt_alloc_link_info(bp);
5584 rc = bnxt_alloc_parent_info(bp);
5588 rc = bnxt_alloc_hwrm_resources(bp);
5591 "Failed to allocate response buffer rc: %x\n", rc);
5594 rc = bnxt_alloc_leds_info(bp);
5598 rc = bnxt_alloc_cos_queues(bp);
5602 rc = bnxt_init_locks(bp);
5606 rc = bnxt_alloc_switch_domain(bp);
5614 bnxt_dev_init(struct rte_eth_dev *eth_dev, void *params __rte_unused)
5616 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
5617 static int version_printed;
5621 if (version_printed++ == 0)
5622 PMD_DRV_LOG(INFO, "%s\n", bnxt_version);
5624 eth_dev->dev_ops = &bnxt_dev_ops;
5625 eth_dev->rx_queue_count = bnxt_rx_queue_count_op;
5626 eth_dev->rx_descriptor_status = bnxt_rx_descriptor_status_op;
5627 eth_dev->tx_descriptor_status = bnxt_tx_descriptor_status_op;
5628 eth_dev->rx_pkt_burst = &bnxt_recv_pkts;
5629 eth_dev->tx_pkt_burst = &bnxt_xmit_pkts;
5632 * For secondary processes, we don't initialise any further
5633 * as primary has already done this work.
5635 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
5638 rte_eth_copy_pci_info(eth_dev, pci_dev);
5639 eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
5641 bp = eth_dev->data->dev_private;
5643 /* Parse dev arguments passed on when starting the DPDK application. */
5644 rc = bnxt_parse_dev_args(bp, pci_dev->device.devargs);
5648 rc = bnxt_drv_init(eth_dev);
5652 rc = bnxt_init_resources(bp, false);
5656 rc = bnxt_alloc_stats_mem(bp);
5661 DRV_MODULE_NAME "found at mem %" PRIX64 ", node addr %pM\n",
5662 pci_dev->mem_resource[0].phys_addr,
5663 pci_dev->mem_resource[0].addr);
5668 bnxt_dev_uninit(eth_dev);
5673 static void bnxt_free_ctx_mem_buf(struct bnxt_ctx_mem_buf_info *ctx)
5682 ctx->dma = RTE_BAD_IOVA;
5683 ctx->ctx_id = BNXT_CTX_VAL_INVAL;
5686 static void bnxt_unregister_fc_ctx_mem(struct bnxt *bp)
5688 bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_RX,
5689 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
5690 bp->flow_stat->rx_fc_out_tbl.ctx_id,
5691 bp->flow_stat->max_fc,
5694 bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_TX,
5695 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
5696 bp->flow_stat->tx_fc_out_tbl.ctx_id,
5697 bp->flow_stat->max_fc,
5700 if (bp->flow_stat->rx_fc_in_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5701 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->rx_fc_in_tbl.ctx_id);
5702 bp->flow_stat->rx_fc_in_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5704 if (bp->flow_stat->rx_fc_out_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5705 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->rx_fc_out_tbl.ctx_id);
5706 bp->flow_stat->rx_fc_out_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5708 if (bp->flow_stat->tx_fc_in_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5709 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->tx_fc_in_tbl.ctx_id);
5710 bp->flow_stat->tx_fc_in_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5712 if (bp->flow_stat->tx_fc_out_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5713 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->tx_fc_out_tbl.ctx_id);
5714 bp->flow_stat->tx_fc_out_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5717 static void bnxt_uninit_fc_ctx_mem(struct bnxt *bp)
5719 bnxt_unregister_fc_ctx_mem(bp);
5721 bnxt_free_ctx_mem_buf(&bp->flow_stat->rx_fc_in_tbl);
5722 bnxt_free_ctx_mem_buf(&bp->flow_stat->rx_fc_out_tbl);
5723 bnxt_free_ctx_mem_buf(&bp->flow_stat->tx_fc_in_tbl);
5724 bnxt_free_ctx_mem_buf(&bp->flow_stat->tx_fc_out_tbl);
5727 static void bnxt_uninit_ctx_mem(struct bnxt *bp)
5729 if (BNXT_FLOW_XSTATS_EN(bp))
5730 bnxt_uninit_fc_ctx_mem(bp);
5734 bnxt_free_error_recovery_info(struct bnxt *bp)
5736 rte_free(bp->recovery_info);
5737 bp->recovery_info = NULL;
5738 bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5742 bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev)
5747 bnxt_free_mem(bp, reconfig_dev);
5749 bnxt_hwrm_func_buf_unrgtr(bp);
5750 rte_free(bp->pf->vf_req_buf);
5752 rc = bnxt_hwrm_func_driver_unregister(bp, 0);
5753 bp->flags &= ~BNXT_FLAG_REGISTERED;
5754 bnxt_free_ctx_mem(bp);
5755 if (!reconfig_dev) {
5756 bnxt_free_hwrm_resources(bp);
5757 bnxt_free_error_recovery_info(bp);
5760 bnxt_uninit_ctx_mem(bp);
5762 bnxt_free_flow_stats_info(bp);
5763 bnxt_free_rep_info(bp);
5764 rte_free(bp->ptp_cfg);
5770 bnxt_dev_uninit(struct rte_eth_dev *eth_dev)
5772 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
5775 PMD_DRV_LOG(DEBUG, "Calling Device uninit\n");
5777 if (eth_dev->state != RTE_ETH_DEV_UNUSED)
5778 bnxt_dev_close_op(eth_dev);
5783 static int bnxt_pci_remove_dev_with_reps(struct rte_eth_dev *eth_dev)
5785 struct bnxt *bp = eth_dev->data->dev_private;
5786 struct rte_eth_dev *vf_rep_eth_dev;
5792 for (i = 0; i < bp->num_reps; i++) {
5793 vf_rep_eth_dev = bp->rep_info[i].vfr_eth_dev;
5794 if (!vf_rep_eth_dev)
5796 PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR pci remove\n",
5797 vf_rep_eth_dev->data->port_id);
5798 rte_eth_dev_destroy(vf_rep_eth_dev, bnxt_representor_uninit);
5800 PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci remove\n",
5801 eth_dev->data->port_id);
5802 ret = rte_eth_dev_destroy(eth_dev, bnxt_dev_uninit);
5807 static void bnxt_free_rep_info(struct bnxt *bp)
5809 rte_free(bp->rep_info);
5810 bp->rep_info = NULL;
5811 rte_free(bp->cfa_code_map);
5812 bp->cfa_code_map = NULL;
5815 static int bnxt_init_rep_info(struct bnxt *bp)
5822 bp->rep_info = rte_zmalloc("bnxt_rep_info",
5823 sizeof(bp->rep_info[0]) * BNXT_MAX_VF_REPS,
5825 if (!bp->rep_info) {
5826 PMD_DRV_LOG(ERR, "Failed to alloc memory for rep info\n");
5829 bp->cfa_code_map = rte_zmalloc("bnxt_cfa_code_map",
5830 sizeof(*bp->cfa_code_map) *
5831 BNXT_MAX_CFA_CODE, 0);
5832 if (!bp->cfa_code_map) {
5833 PMD_DRV_LOG(ERR, "Failed to alloc memory for cfa_code_map\n");
5834 bnxt_free_rep_info(bp);
5838 for (i = 0; i < BNXT_MAX_CFA_CODE; i++)
5839 bp->cfa_code_map[i] = BNXT_VF_IDX_INVALID;
5841 rc = pthread_mutex_init(&bp->rep_info->vfr_lock, NULL);
5843 PMD_DRV_LOG(ERR, "Unable to initialize vfr_lock\n");
5844 bnxt_free_rep_info(bp);
5848 rc = pthread_mutex_init(&bp->rep_info->vfr_start_lock, NULL);
5850 PMD_DRV_LOG(ERR, "Unable to initialize vfr_start_lock\n");
5851 bnxt_free_rep_info(bp);
5858 static int bnxt_rep_port_probe(struct rte_pci_device *pci_dev,
5859 struct rte_eth_devargs *eth_da,
5860 struct rte_eth_dev *backing_eth_dev,
5861 const char *dev_args)
5863 struct rte_eth_dev *vf_rep_eth_dev;
5864 char name[RTE_ETH_NAME_MAX_LEN];
5865 struct bnxt *backing_bp;
5868 struct rte_kvargs *kvlist = NULL;
5870 if (eth_da->type == RTE_ETH_REPRESENTOR_NONE)
5872 if (eth_da->type != RTE_ETH_REPRESENTOR_VF) {
5873 PMD_DRV_LOG(ERR, "unsupported representor type %d\n",
5877 num_rep = eth_da->nb_representor_ports;
5878 if (num_rep > BNXT_MAX_VF_REPS) {
5879 PMD_DRV_LOG(ERR, "nb_representor_ports = %d > %d MAX VF REPS\n",
5880 num_rep, BNXT_MAX_VF_REPS);
5884 if (num_rep >= RTE_MAX_ETHPORTS) {
5886 "nb_representor_ports = %d > %d MAX ETHPORTS\n",
5887 num_rep, RTE_MAX_ETHPORTS);
5891 backing_bp = backing_eth_dev->data->dev_private;
5893 if (!(BNXT_PF(backing_bp) || BNXT_VF_IS_TRUSTED(backing_bp))) {
5895 "Not a PF or trusted VF. No Representor support\n");
5896 /* Returning an error is not an option.
5897 * Applications are not handling this correctly
5902 if (bnxt_init_rep_info(backing_bp))
5905 for (i = 0; i < num_rep; i++) {
5906 struct bnxt_representor representor = {
5907 .vf_id = eth_da->representor_ports[i],
5908 .switch_domain_id = backing_bp->switch_domain_id,
5909 .parent_dev = backing_eth_dev
5912 if (representor.vf_id >= BNXT_MAX_VF_REPS) {
5913 PMD_DRV_LOG(ERR, "VF-Rep id %d >= %d MAX VF ID\n",
5914 representor.vf_id, BNXT_MAX_VF_REPS);
5918 /* representor port net_bdf_port */
5919 snprintf(name, sizeof(name), "net_%s_representor_%d",
5920 pci_dev->device.name, eth_da->representor_ports[i]);
5922 kvlist = rte_kvargs_parse(dev_args, bnxt_dev_args);
5925 * Handler for "rep_is_pf" devarg.
5926 * Invoked as for ex: "-a 000:00:0d.0,
5927 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
5929 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_IS_PF,
5930 bnxt_parse_devarg_rep_is_pf,
5931 (void *)&representor);
5937 * Handler for "rep_based_pf" devarg.
5938 * Invoked as for ex: "-a 000:00:0d.0,
5939 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
5941 ret = rte_kvargs_process(kvlist,
5942 BNXT_DEVARG_REP_BASED_PF,
5943 bnxt_parse_devarg_rep_based_pf,
5944 (void *)&representor);
5950 * Handler for "rep_based_pf" devarg.
5951 * Invoked as for ex: "-a 000:00:0d.0,
5952 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
5954 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_Q_R2F,
5955 bnxt_parse_devarg_rep_q_r2f,
5956 (void *)&representor);
5962 * Handler for "rep_based_pf" devarg.
5963 * Invoked as for ex: "-a 000:00:0d.0,
5964 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
5966 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_Q_F2R,
5967 bnxt_parse_devarg_rep_q_f2r,
5968 (void *)&representor);
5974 * Handler for "rep_based_pf" devarg.
5975 * Invoked as for ex: "-a 000:00:0d.0,
5976 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
5978 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_FC_R2F,
5979 bnxt_parse_devarg_rep_fc_r2f,
5980 (void *)&representor);
5986 * Handler for "rep_based_pf" devarg.
5987 * Invoked as for ex: "-a 000:00:0d.0,
5988 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
5990 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_FC_F2R,
5991 bnxt_parse_devarg_rep_fc_f2r,
5992 (void *)&representor);
5999 ret = rte_eth_dev_create(&pci_dev->device, name,
6000 sizeof(struct bnxt_representor),
6002 bnxt_representor_init,
6005 PMD_DRV_LOG(ERR, "failed to create bnxt vf "
6006 "representor %s.", name);
6010 vf_rep_eth_dev = rte_eth_dev_allocated(name);
6011 if (!vf_rep_eth_dev) {
6012 PMD_DRV_LOG(ERR, "Failed to find the eth_dev"
6013 " for VF-Rep: %s.", name);
6018 PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR pci probe\n",
6019 backing_eth_dev->data->port_id);
6020 backing_bp->rep_info[representor.vf_id].vfr_eth_dev =
6022 backing_bp->num_reps++;
6026 rte_kvargs_free(kvlist);
6030 /* If num_rep > 1, then rollback already created
6031 * ports, since we'll be failing the probe anyway
6034 bnxt_pci_remove_dev_with_reps(backing_eth_dev);
6036 rte_kvargs_free(kvlist);
6041 static int bnxt_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
6042 struct rte_pci_device *pci_dev)
6044 struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
6045 struct rte_eth_dev *backing_eth_dev;
6049 if (pci_dev->device.devargs) {
6050 ret = rte_eth_devargs_parse(pci_dev->device.devargs->args,
6056 num_rep = eth_da.nb_representor_ports;
6057 PMD_DRV_LOG(DEBUG, "nb_representor_ports = %d\n",
6060 /* We could come here after first level of probe is already invoked
6061 * as part of an application bringup(OVS-DPDK vswitchd), so first check
6062 * for already allocated eth_dev for the backing device (PF/Trusted VF)
6064 backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6065 if (backing_eth_dev == NULL) {
6066 ret = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
6067 sizeof(struct bnxt),
6068 eth_dev_pci_specific_init, pci_dev,
6069 bnxt_dev_init, NULL);
6071 if (ret || !num_rep)
6074 backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6076 PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci probe\n",
6077 backing_eth_dev->data->port_id);
6082 /* probe representor ports now */
6083 ret = bnxt_rep_port_probe(pci_dev, ð_da, backing_eth_dev,
6084 pci_dev->device.devargs->args);
6089 static int bnxt_pci_remove(struct rte_pci_device *pci_dev)
6091 struct rte_eth_dev *eth_dev;
6093 eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6095 return 0; /* Invoked typically only by OVS-DPDK, by the
6096 * time it comes here the eth_dev is already
6097 * deleted by rte_eth_dev_close(), so returning
6098 * +ve value will at least help in proper cleanup
6101 PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci remove\n", eth_dev->data->port_id);
6102 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
6103 if (eth_dev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
6104 return rte_eth_dev_destroy(eth_dev,
6105 bnxt_representor_uninit);
6107 return rte_eth_dev_destroy(eth_dev,
6110 return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
6114 static struct rte_pci_driver bnxt_rte_pmd = {
6115 .id_table = bnxt_pci_id_map,
6116 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
6117 RTE_PCI_DRV_PROBE_AGAIN, /* Needed in case of VF-REPs
6120 .probe = bnxt_pci_probe,
6121 .remove = bnxt_pci_remove,
6125 is_device_supported(struct rte_eth_dev *dev, struct rte_pci_driver *drv)
6127 if (strcmp(dev->device->driver->name, drv->driver.name))
6133 bool is_bnxt_supported(struct rte_eth_dev *dev)
6135 return is_device_supported(dev, &bnxt_rte_pmd);
6138 RTE_LOG_REGISTER(bnxt_logtype_driver, pmd.net.bnxt.driver, NOTICE);
6139 RTE_PMD_REGISTER_PCI(net_bnxt, bnxt_rte_pmd);
6140 RTE_PMD_REGISTER_PCI_TABLE(net_bnxt, bnxt_pci_id_map);
6141 RTE_PMD_REGISTER_KMOD_DEP(net_bnxt, "* igb_uio | uio_pci_generic | vfio-pci");