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);
225 static void bnxt_free_pf_info(struct bnxt *bp)
231 static void bnxt_free_link_info(struct bnxt *bp)
233 rte_free(bp->link_info);
234 bp->link_info = NULL;
237 static void bnxt_free_leds_info(struct bnxt *bp)
246 static void bnxt_free_flow_stats_info(struct bnxt *bp)
248 rte_free(bp->flow_stat);
249 bp->flow_stat = NULL;
252 static void bnxt_free_cos_queues(struct bnxt *bp)
254 rte_free(bp->rx_cos_queue);
255 bp->rx_cos_queue = NULL;
256 rte_free(bp->tx_cos_queue);
257 bp->tx_cos_queue = NULL;
260 static void bnxt_free_mem(struct bnxt *bp, bool reconfig)
262 bnxt_free_filter_mem(bp);
263 bnxt_free_vnic_attributes(bp);
264 bnxt_free_vnic_mem(bp);
266 /* tx/rx rings are configured as part of *_queue_setup callbacks.
267 * If the number of rings change across fw update,
268 * we don't have much choice except to warn the user.
272 bnxt_free_tx_rings(bp);
273 bnxt_free_rx_rings(bp);
275 bnxt_free_async_cp_ring(bp);
276 bnxt_free_rxtx_nq_ring(bp);
278 rte_free(bp->grp_info);
282 static int bnxt_alloc_parent_info(struct bnxt *bp)
284 bp->parent = rte_zmalloc("bnxt_parent_info",
285 sizeof(struct bnxt_parent_info), 0);
286 if (bp->parent == NULL)
292 static int bnxt_alloc_pf_info(struct bnxt *bp)
294 bp->pf = rte_zmalloc("bnxt_pf_info", sizeof(struct bnxt_pf_info), 0);
301 static int bnxt_alloc_link_info(struct bnxt *bp)
304 rte_zmalloc("bnxt_link_info", sizeof(struct bnxt_link_info), 0);
305 if (bp->link_info == NULL)
311 static int bnxt_alloc_leds_info(struct bnxt *bp)
316 bp->leds = rte_zmalloc("bnxt_leds",
317 BNXT_MAX_LED * sizeof(struct bnxt_led_info),
319 if (bp->leds == NULL)
325 static int bnxt_alloc_cos_queues(struct bnxt *bp)
328 rte_zmalloc("bnxt_rx_cosq",
329 BNXT_COS_QUEUE_COUNT *
330 sizeof(struct bnxt_cos_queue_info),
332 if (bp->rx_cos_queue == NULL)
336 rte_zmalloc("bnxt_tx_cosq",
337 BNXT_COS_QUEUE_COUNT *
338 sizeof(struct bnxt_cos_queue_info),
340 if (bp->tx_cos_queue == NULL)
346 static int bnxt_alloc_flow_stats_info(struct bnxt *bp)
348 bp->flow_stat = rte_zmalloc("bnxt_flow_xstat",
349 sizeof(struct bnxt_flow_stat_info), 0);
350 if (bp->flow_stat == NULL)
356 static int bnxt_alloc_mem(struct bnxt *bp, bool reconfig)
360 rc = bnxt_alloc_ring_grps(bp);
364 rc = bnxt_alloc_async_ring_struct(bp);
368 rc = bnxt_alloc_vnic_mem(bp);
372 rc = bnxt_alloc_vnic_attributes(bp);
376 rc = bnxt_alloc_filter_mem(bp);
380 rc = bnxt_alloc_async_cp_ring(bp);
384 rc = bnxt_alloc_rxtx_nq_ring(bp);
388 if (BNXT_FLOW_XSTATS_EN(bp)) {
389 rc = bnxt_alloc_flow_stats_info(bp);
397 bnxt_free_mem(bp, reconfig);
401 static int bnxt_setup_one_vnic(struct bnxt *bp, uint16_t vnic_id)
403 struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
404 struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
405 uint64_t rx_offloads = dev_conf->rxmode.offloads;
406 struct bnxt_rx_queue *rxq;
410 rc = bnxt_vnic_grp_alloc(bp, vnic);
414 PMD_DRV_LOG(DEBUG, "vnic[%d] = %p vnic->fw_grp_ids = %p\n",
415 vnic_id, vnic, vnic->fw_grp_ids);
417 rc = bnxt_hwrm_vnic_alloc(bp, vnic);
421 /* Alloc RSS context only if RSS mode is enabled */
422 if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS) {
423 int j, nr_ctxs = bnxt_rss_ctxts(bp);
425 /* RSS table size in Thor is 512.
426 * Cap max Rx rings to same value
428 if (bp->rx_nr_rings > BNXT_RSS_TBL_SIZE_P5) {
429 PMD_DRV_LOG(ERR, "RxQ cnt %d > reta_size %d\n",
430 bp->rx_nr_rings, BNXT_RSS_TBL_SIZE_P5);
435 for (j = 0; j < nr_ctxs; j++) {
436 rc = bnxt_hwrm_vnic_ctx_alloc(bp, vnic, j);
442 "HWRM vnic %d ctx %d alloc failure rc: %x\n",
446 vnic->num_lb_ctxts = nr_ctxs;
450 * Firmware sets pf pair in default vnic cfg. If the VLAN strip
451 * setting is not available at this time, it will not be
452 * configured correctly in the CFA.
454 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
455 vnic->vlan_strip = true;
457 vnic->vlan_strip = false;
459 rc = bnxt_hwrm_vnic_cfg(bp, vnic);
463 rc = bnxt_set_hwrm_vnic_filters(bp, vnic);
467 for (j = 0; j < bp->rx_num_qs_per_vnic; j++) {
468 rxq = bp->eth_dev->data->rx_queues[j];
471 "rxq[%d]->vnic=%p vnic->fw_grp_ids=%p\n",
472 j, rxq->vnic, rxq->vnic->fw_grp_ids);
474 if (BNXT_HAS_RING_GRPS(bp) && rxq->rx_deferred_start)
475 rxq->vnic->fw_grp_ids[j] = INVALID_HW_RING_ID;
477 vnic->rx_queue_cnt++;
480 PMD_DRV_LOG(DEBUG, "vnic->rx_queue_cnt = %d\n", vnic->rx_queue_cnt);
482 rc = bnxt_vnic_rss_configure(bp, vnic);
486 bnxt_hwrm_vnic_plcmode_cfg(bp, vnic);
488 rc = bnxt_hwrm_vnic_tpa_cfg(bp, vnic,
489 (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) ?
496 PMD_DRV_LOG(ERR, "HWRM vnic %d cfg failure rc: %x\n",
501 static int bnxt_register_fc_ctx_mem(struct bnxt *bp)
505 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->rx_fc_in_tbl.dma,
506 &bp->flow_stat->rx_fc_in_tbl.ctx_id);
511 "rx_fc_in_tbl.va = %p rx_fc_in_tbl.dma = %p"
512 " rx_fc_in_tbl.ctx_id = %d\n",
513 bp->flow_stat->rx_fc_in_tbl.va,
514 (void *)((uintptr_t)bp->flow_stat->rx_fc_in_tbl.dma),
515 bp->flow_stat->rx_fc_in_tbl.ctx_id);
517 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->rx_fc_out_tbl.dma,
518 &bp->flow_stat->rx_fc_out_tbl.ctx_id);
523 "rx_fc_out_tbl.va = %p rx_fc_out_tbl.dma = %p"
524 " rx_fc_out_tbl.ctx_id = %d\n",
525 bp->flow_stat->rx_fc_out_tbl.va,
526 (void *)((uintptr_t)bp->flow_stat->rx_fc_out_tbl.dma),
527 bp->flow_stat->rx_fc_out_tbl.ctx_id);
529 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->tx_fc_in_tbl.dma,
530 &bp->flow_stat->tx_fc_in_tbl.ctx_id);
535 "tx_fc_in_tbl.va = %p tx_fc_in_tbl.dma = %p"
536 " tx_fc_in_tbl.ctx_id = %d\n",
537 bp->flow_stat->tx_fc_in_tbl.va,
538 (void *)((uintptr_t)bp->flow_stat->tx_fc_in_tbl.dma),
539 bp->flow_stat->tx_fc_in_tbl.ctx_id);
541 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->tx_fc_out_tbl.dma,
542 &bp->flow_stat->tx_fc_out_tbl.ctx_id);
547 "tx_fc_out_tbl.va = %p tx_fc_out_tbl.dma = %p"
548 " tx_fc_out_tbl.ctx_id = %d\n",
549 bp->flow_stat->tx_fc_out_tbl.va,
550 (void *)((uintptr_t)bp->flow_stat->tx_fc_out_tbl.dma),
551 bp->flow_stat->tx_fc_out_tbl.ctx_id);
553 memset(bp->flow_stat->rx_fc_out_tbl.va,
555 bp->flow_stat->rx_fc_out_tbl.size);
556 rc = bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_RX,
557 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
558 bp->flow_stat->rx_fc_out_tbl.ctx_id,
559 bp->flow_stat->max_fc,
564 memset(bp->flow_stat->tx_fc_out_tbl.va,
566 bp->flow_stat->tx_fc_out_tbl.size);
567 rc = bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_TX,
568 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
569 bp->flow_stat->tx_fc_out_tbl.ctx_id,
570 bp->flow_stat->max_fc,
576 static int bnxt_alloc_ctx_mem_buf(char *type, size_t size,
577 struct bnxt_ctx_mem_buf_info *ctx)
582 ctx->va = rte_zmalloc(type, size, 0);
585 rte_mem_lock_page(ctx->va);
587 ctx->dma = rte_mem_virt2iova(ctx->va);
588 if (ctx->dma == RTE_BAD_IOVA)
594 static int bnxt_init_fc_ctx_mem(struct bnxt *bp)
596 struct rte_pci_device *pdev = bp->pdev;
597 char type[RTE_MEMZONE_NAMESIZE];
601 max_fc = bp->flow_stat->max_fc;
603 sprintf(type, "bnxt_rx_fc_in_" PCI_PRI_FMT, pdev->addr.domain,
604 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
605 /* 4 bytes for each counter-id */
606 rc = bnxt_alloc_ctx_mem_buf(type,
608 &bp->flow_stat->rx_fc_in_tbl);
612 sprintf(type, "bnxt_rx_fc_out_" PCI_PRI_FMT, pdev->addr.domain,
613 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
614 /* 16 bytes for each counter - 8 bytes pkt_count, 8 bytes byte_count */
615 rc = bnxt_alloc_ctx_mem_buf(type,
617 &bp->flow_stat->rx_fc_out_tbl);
621 sprintf(type, "bnxt_tx_fc_in_" PCI_PRI_FMT, pdev->addr.domain,
622 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
623 /* 4 bytes for each counter-id */
624 rc = bnxt_alloc_ctx_mem_buf(type,
626 &bp->flow_stat->tx_fc_in_tbl);
630 sprintf(type, "bnxt_tx_fc_out_" PCI_PRI_FMT, pdev->addr.domain,
631 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
632 /* 16 bytes for each counter - 8 bytes pkt_count, 8 bytes byte_count */
633 rc = bnxt_alloc_ctx_mem_buf(type,
635 &bp->flow_stat->tx_fc_out_tbl);
639 rc = bnxt_register_fc_ctx_mem(bp);
644 static int bnxt_init_ctx_mem(struct bnxt *bp)
648 if (!(bp->fw_cap & BNXT_FW_CAP_ADV_FLOW_COUNTERS) ||
649 !(BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) ||
650 !BNXT_FLOW_XSTATS_EN(bp))
653 rc = bnxt_hwrm_cfa_counter_qcaps(bp, &bp->flow_stat->max_fc);
657 rc = bnxt_init_fc_ctx_mem(bp);
662 static int bnxt_update_phy_setting(struct bnxt *bp)
664 struct rte_eth_link new;
667 rc = bnxt_get_hwrm_link_config(bp, &new);
669 PMD_DRV_LOG(ERR, "Failed to get link settings\n");
674 * On BCM957508-N2100 adapters, FW will not allow any user other
675 * than BMC to shutdown the port. bnxt_get_hwrm_link_config() call
676 * always returns link up. Force phy update always in that case.
678 if (!new.link_status || IS_BNXT_DEV_957508_N2100(bp)) {
679 rc = bnxt_set_hwrm_link_config(bp, true);
681 PMD_DRV_LOG(ERR, "Failed to update PHY settings\n");
689 static void bnxt_free_prev_ring_stats(struct bnxt *bp)
691 rte_free(bp->prev_rx_ring_stats);
692 rte_free(bp->prev_tx_ring_stats);
694 bp->prev_rx_ring_stats = NULL;
695 bp->prev_tx_ring_stats = NULL;
698 static int bnxt_alloc_prev_ring_stats(struct bnxt *bp)
700 bp->prev_rx_ring_stats = rte_zmalloc("bnxt_prev_rx_ring_stats",
701 sizeof(struct bnxt_ring_stats) *
704 if (bp->prev_rx_ring_stats == NULL)
707 bp->prev_tx_ring_stats = rte_zmalloc("bnxt_prev_tx_ring_stats",
708 sizeof(struct bnxt_ring_stats) *
711 if (bp->prev_tx_ring_stats == NULL)
717 bnxt_free_prev_ring_stats(bp);
721 static int bnxt_start_nic(struct bnxt *bp)
723 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(bp->eth_dev);
724 struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
725 uint32_t intr_vector = 0;
726 uint32_t queue_id, base = BNXT_MISC_VEC_ID;
727 uint32_t vec = BNXT_MISC_VEC_ID;
731 if (bp->eth_dev->data->mtu > RTE_ETHER_MTU) {
732 bp->eth_dev->data->dev_conf.rxmode.offloads |=
733 DEV_RX_OFFLOAD_JUMBO_FRAME;
734 bp->flags |= BNXT_FLAG_JUMBO;
736 bp->eth_dev->data->dev_conf.rxmode.offloads &=
737 ~DEV_RX_OFFLOAD_JUMBO_FRAME;
738 bp->flags &= ~BNXT_FLAG_JUMBO;
741 /* THOR does not support ring groups.
742 * But we will use the array to save RSS context IDs.
744 if (BNXT_CHIP_P5(bp))
745 bp->max_ring_grps = BNXT_MAX_RSS_CTXTS_P5;
747 rc = bnxt_alloc_all_hwrm_stat_ctxs(bp);
749 PMD_DRV_LOG(ERR, "HWRM stat ctx alloc failure rc: %x\n", rc);
753 rc = bnxt_alloc_hwrm_rings(bp);
755 PMD_DRV_LOG(ERR, "HWRM ring alloc failure rc: %x\n", rc);
759 rc = bnxt_alloc_all_hwrm_ring_grps(bp);
761 PMD_DRV_LOG(ERR, "HWRM ring grp alloc failure: %x\n", rc);
765 if (!(bp->vnic_cap_flags & BNXT_VNIC_CAP_COS_CLASSIFY))
768 for (j = 0, i = 0; i < BNXT_COS_QUEUE_COUNT; i++) {
769 if (bp->rx_cos_queue[i].id != 0xff) {
770 struct bnxt_vnic_info *vnic = &bp->vnic_info[j++];
774 "Num pools more than FW profile\n");
778 vnic->cos_queue_id = bp->rx_cos_queue[i].id;
784 rc = bnxt_mq_rx_configure(bp);
786 PMD_DRV_LOG(ERR, "MQ mode configure failure rc: %x\n", rc);
791 rc = bnxt_setup_one_vnic(bp, 0);
794 /* VNIC configuration */
795 if (BNXT_RFS_NEEDS_VNIC(bp)) {
796 for (i = 1; i < bp->nr_vnics; i++) {
797 rc = bnxt_setup_one_vnic(bp, i);
803 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, &bp->vnic_info[0], 0, NULL);
806 "HWRM cfa l2 rx mask failure rc: %x\n", rc);
810 /* check and configure queue intr-vector mapping */
811 if ((rte_intr_cap_multiple(intr_handle) ||
812 !RTE_ETH_DEV_SRIOV(bp->eth_dev).active) &&
813 bp->eth_dev->data->dev_conf.intr_conf.rxq != 0) {
814 intr_vector = bp->eth_dev->data->nb_rx_queues;
815 PMD_DRV_LOG(DEBUG, "intr_vector = %d\n", intr_vector);
816 if (intr_vector > bp->rx_cp_nr_rings) {
817 PMD_DRV_LOG(ERR, "At most %d intr queues supported",
821 rc = rte_intr_efd_enable(intr_handle, intr_vector);
826 if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
827 intr_handle->intr_vec =
828 rte_zmalloc("intr_vec",
829 bp->eth_dev->data->nb_rx_queues *
831 if (intr_handle->intr_vec == NULL) {
832 PMD_DRV_LOG(ERR, "Failed to allocate %d rx_queues"
833 " intr_vec", bp->eth_dev->data->nb_rx_queues);
837 PMD_DRV_LOG(DEBUG, "intr_handle->intr_vec = %p "
838 "intr_handle->nb_efd = %d intr_handle->max_intr = %d\n",
839 intr_handle->intr_vec, intr_handle->nb_efd,
840 intr_handle->max_intr);
841 for (queue_id = 0; queue_id < bp->eth_dev->data->nb_rx_queues;
843 intr_handle->intr_vec[queue_id] =
844 vec + BNXT_RX_VEC_START;
845 if (vec < base + intr_handle->nb_efd - 1)
850 /* enable uio/vfio intr/eventfd mapping */
851 rc = rte_intr_enable(intr_handle);
852 #ifndef RTE_EXEC_ENV_FREEBSD
853 /* In FreeBSD OS, nic_uio driver does not support interrupts */
858 rc = bnxt_update_phy_setting(bp);
862 bp->mark_table = rte_zmalloc("bnxt_mark_table", BNXT_MARK_TABLE_SZ, 0);
864 PMD_DRV_LOG(ERR, "Allocation of mark table failed\n");
869 /* Some of the error status returned by FW may not be from errno.h */
876 static int bnxt_shutdown_nic(struct bnxt *bp)
878 bnxt_free_all_hwrm_resources(bp);
879 bnxt_free_all_filters(bp);
880 bnxt_free_all_vnics(bp);
885 * Device configuration and status function
888 uint32_t bnxt_get_speed_capabilities(struct bnxt *bp)
890 uint32_t link_speed = 0;
891 uint32_t speed_capa = 0;
893 if (bp->link_info == NULL)
896 link_speed = bp->link_info->support_speeds;
898 /* If PAM4 is configured, use PAM4 supported speed */
899 if (link_speed == 0 && bp->link_info->support_pam4_speeds > 0)
900 link_speed = bp->link_info->support_pam4_speeds;
902 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_100MB)
903 speed_capa |= ETH_LINK_SPEED_100M;
904 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_100MBHD)
905 speed_capa |= ETH_LINK_SPEED_100M_HD;
906 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_1GB)
907 speed_capa |= ETH_LINK_SPEED_1G;
908 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_2_5GB)
909 speed_capa |= ETH_LINK_SPEED_2_5G;
910 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_10GB)
911 speed_capa |= ETH_LINK_SPEED_10G;
912 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_20GB)
913 speed_capa |= ETH_LINK_SPEED_20G;
914 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_25GB)
915 speed_capa |= ETH_LINK_SPEED_25G;
916 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_40GB)
917 speed_capa |= ETH_LINK_SPEED_40G;
918 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_50GB)
919 speed_capa |= ETH_LINK_SPEED_50G;
920 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_100GB)
921 speed_capa |= ETH_LINK_SPEED_100G;
922 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_50G)
923 speed_capa |= ETH_LINK_SPEED_50G;
924 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_100G)
925 speed_capa |= ETH_LINK_SPEED_100G;
926 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_200G)
927 speed_capa |= ETH_LINK_SPEED_200G;
929 if (bp->link_info->auto_mode ==
930 HWRM_PORT_PHY_QCFG_OUTPUT_AUTO_MODE_NONE)
931 speed_capa |= ETH_LINK_SPEED_FIXED;
936 static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
937 struct rte_eth_dev_info *dev_info)
939 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(eth_dev->device);
940 struct bnxt *bp = eth_dev->data->dev_private;
941 uint16_t max_vnics, i, j, vpool, vrxq;
942 unsigned int max_rx_rings;
945 rc = is_bnxt_in_error(bp);
950 dev_info->max_mac_addrs = bp->max_l2_ctx;
951 dev_info->max_hash_mac_addrs = 0;
953 /* PF/VF specifics */
955 dev_info->max_vfs = pdev->max_vfs;
957 max_rx_rings = bnxt_max_rings(bp);
958 /* For the sake of symmetry, max_rx_queues = max_tx_queues */
959 dev_info->max_rx_queues = max_rx_rings;
960 dev_info->max_tx_queues = max_rx_rings;
961 dev_info->reta_size = bnxt_rss_hash_tbl_size(bp);
962 dev_info->hash_key_size = 40;
963 max_vnics = bp->max_vnics;
966 dev_info->min_mtu = RTE_ETHER_MIN_MTU;
967 dev_info->max_mtu = BNXT_MAX_MTU;
969 /* Fast path specifics */
970 dev_info->min_rx_bufsize = 1;
971 dev_info->max_rx_pktlen = BNXT_MAX_PKT_LEN;
973 dev_info->rx_offload_capa = BNXT_DEV_RX_OFFLOAD_SUPPORT;
974 if (bp->flags & BNXT_FLAG_PTP_SUPPORTED)
975 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TIMESTAMP;
976 dev_info->tx_queue_offload_capa = DEV_TX_OFFLOAD_MBUF_FAST_FREE;
977 dev_info->tx_offload_capa = BNXT_DEV_TX_OFFLOAD_SUPPORT |
978 dev_info->tx_queue_offload_capa;
979 dev_info->flow_type_rss_offloads = BNXT_ETH_RSS_SUPPORT;
981 dev_info->speed_capa = bnxt_get_speed_capabilities(bp);
984 dev_info->default_rxconf = (struct rte_eth_rxconf) {
990 .rx_free_thresh = 32,
991 .rx_drop_en = BNXT_DEFAULT_RX_DROP_EN,
994 dev_info->default_txconf = (struct rte_eth_txconf) {
1000 .tx_free_thresh = 32,
1003 eth_dev->data->dev_conf.intr_conf.lsc = 1;
1005 eth_dev->data->dev_conf.intr_conf.rxq = 1;
1006 dev_info->rx_desc_lim.nb_min = BNXT_MIN_RING_DESC;
1007 dev_info->rx_desc_lim.nb_max = BNXT_MAX_RX_RING_DESC;
1008 dev_info->tx_desc_lim.nb_min = BNXT_MIN_RING_DESC;
1009 dev_info->tx_desc_lim.nb_max = BNXT_MAX_TX_RING_DESC;
1011 if (BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) {
1012 dev_info->switch_info.name = eth_dev->device->name;
1013 dev_info->switch_info.domain_id = bp->switch_domain_id;
1014 dev_info->switch_info.port_id =
1015 BNXT_PF(bp) ? BNXT_SWITCH_PORT_ID_PF :
1016 BNXT_SWITCH_PORT_ID_TRUSTED_VF;
1022 * TODO: default_rxconf, default_txconf, rx_desc_lim, and tx_desc_lim
1023 * need further investigation.
1026 /* VMDq resources */
1027 vpool = 64; /* ETH_64_POOLS */
1028 vrxq = 128; /* ETH_VMDQ_DCB_NUM_QUEUES */
1029 for (i = 0; i < 4; vpool >>= 1, i++) {
1030 if (max_vnics > vpool) {
1031 for (j = 0; j < 5; vrxq >>= 1, j++) {
1032 if (dev_info->max_rx_queues > vrxq) {
1038 /* Not enough resources to support VMDq */
1042 /* Not enough resources to support VMDq */
1046 dev_info->max_vmdq_pools = vpool;
1047 dev_info->vmdq_queue_num = vrxq;
1049 dev_info->vmdq_pool_base = 0;
1050 dev_info->vmdq_queue_base = 0;
1055 /* Configure the device based on the configuration provided */
1056 static int bnxt_dev_configure_op(struct rte_eth_dev *eth_dev)
1058 struct bnxt *bp = eth_dev->data->dev_private;
1059 uint64_t rx_offloads = eth_dev->data->dev_conf.rxmode.offloads;
1062 bp->rx_queues = (void *)eth_dev->data->rx_queues;
1063 bp->tx_queues = (void *)eth_dev->data->tx_queues;
1064 bp->tx_nr_rings = eth_dev->data->nb_tx_queues;
1065 bp->rx_nr_rings = eth_dev->data->nb_rx_queues;
1067 rc = is_bnxt_in_error(bp);
1071 if (BNXT_VF(bp) && (bp->flags & BNXT_FLAG_NEW_RM)) {
1072 rc = bnxt_hwrm_check_vf_rings(bp);
1074 PMD_DRV_LOG(ERR, "HWRM insufficient resources\n");
1078 /* If a resource has already been allocated - in this case
1079 * it is the async completion ring, free it. Reallocate it after
1080 * resource reservation. This will ensure the resource counts
1081 * are calculated correctly.
1084 pthread_mutex_lock(&bp->def_cp_lock);
1086 if (!BNXT_HAS_NQ(bp) && bp->async_cp_ring) {
1087 bnxt_disable_int(bp);
1088 bnxt_free_cp_ring(bp, bp->async_cp_ring);
1091 rc = bnxt_hwrm_func_reserve_vf_resc(bp, false);
1093 PMD_DRV_LOG(ERR, "HWRM resource alloc fail:%x\n", rc);
1094 pthread_mutex_unlock(&bp->def_cp_lock);
1098 if (!BNXT_HAS_NQ(bp) && bp->async_cp_ring) {
1099 rc = bnxt_alloc_async_cp_ring(bp);
1101 pthread_mutex_unlock(&bp->def_cp_lock);
1104 bnxt_enable_int(bp);
1107 pthread_mutex_unlock(&bp->def_cp_lock);
1110 /* Inherit new configurations */
1111 if (eth_dev->data->nb_rx_queues > bp->max_rx_rings ||
1112 eth_dev->data->nb_tx_queues > bp->max_tx_rings ||
1113 eth_dev->data->nb_rx_queues + eth_dev->data->nb_tx_queues
1114 + BNXT_NUM_ASYNC_CPR(bp) > bp->max_cp_rings ||
1115 eth_dev->data->nb_rx_queues + eth_dev->data->nb_tx_queues >
1117 goto resource_error;
1119 if (BNXT_HAS_RING_GRPS(bp) &&
1120 (uint32_t)(eth_dev->data->nb_rx_queues) > bp->max_ring_grps)
1121 goto resource_error;
1123 if (!(eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) &&
1124 bp->max_vnics < eth_dev->data->nb_rx_queues)
1125 goto resource_error;
1127 bp->rx_cp_nr_rings = bp->rx_nr_rings;
1128 bp->tx_cp_nr_rings = bp->tx_nr_rings;
1130 if (eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
1131 rx_offloads |= DEV_RX_OFFLOAD_RSS_HASH;
1132 eth_dev->data->dev_conf.rxmode.offloads = rx_offloads;
1134 if (rx_offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) {
1135 eth_dev->data->mtu =
1136 eth_dev->data->dev_conf.rxmode.max_rx_pkt_len -
1137 RTE_ETHER_HDR_LEN - RTE_ETHER_CRC_LEN - VLAN_TAG_SIZE *
1139 bnxt_mtu_set_op(eth_dev, eth_dev->data->mtu);
1145 "Insufficient resources to support requested config\n");
1147 "Num Queues Requested: Tx %d, Rx %d\n",
1148 eth_dev->data->nb_tx_queues,
1149 eth_dev->data->nb_rx_queues);
1151 "MAX: TxQ %d, RxQ %d, CQ %d Stat %d, Grp %d, Vnic %d\n",
1152 bp->max_tx_rings, bp->max_rx_rings, bp->max_cp_rings,
1153 bp->max_stat_ctx, bp->max_ring_grps, bp->max_vnics);
1157 void bnxt_print_link_info(struct rte_eth_dev *eth_dev)
1159 struct rte_eth_link *link = ð_dev->data->dev_link;
1161 if (link->link_status)
1162 PMD_DRV_LOG(INFO, "Port %d Link Up - speed %u Mbps - %s\n",
1163 eth_dev->data->port_id,
1164 (uint32_t)link->link_speed,
1165 (link->link_duplex == ETH_LINK_FULL_DUPLEX) ?
1166 ("full-duplex") : ("half-duplex\n"));
1168 PMD_DRV_LOG(INFO, "Port %d Link Down\n",
1169 eth_dev->data->port_id);
1173 * Determine whether the current configuration requires support for scattered
1174 * receive; return 1 if scattered receive is required and 0 if not.
1176 static int bnxt_scattered_rx(struct rte_eth_dev *eth_dev)
1181 if (eth_dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_SCATTER)
1184 if (eth_dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_TCP_LRO)
1187 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
1188 struct bnxt_rx_queue *rxq = eth_dev->data->rx_queues[i];
1190 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
1191 RTE_PKTMBUF_HEADROOM);
1192 if (eth_dev->data->dev_conf.rxmode.max_rx_pkt_len > buf_size)
1198 static eth_rx_burst_t
1199 bnxt_receive_function(struct rte_eth_dev *eth_dev)
1201 struct bnxt *bp = eth_dev->data->dev_private;
1203 /* Disable vector mode RX for Stingray2 for now */
1204 if (BNXT_CHIP_SR2(bp)) {
1205 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1206 return bnxt_recv_pkts;
1209 #if (defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)) && \
1210 !defined(RTE_LIBRTE_IEEE1588)
1212 /* Vector mode receive cannot be enabled if scattered rx is in use. */
1213 if (eth_dev->data->scattered_rx)
1217 * Vector mode receive cannot be enabled if Truflow is enabled or if
1218 * asynchronous completions and receive completions can be placed in
1219 * the same completion ring.
1221 if (BNXT_TRUFLOW_EN(bp) || !BNXT_NUM_ASYNC_CPR(bp))
1225 * Vector mode receive cannot be enabled if any receive offloads outside
1226 * a limited subset have been enabled.
1228 if (eth_dev->data->dev_conf.rxmode.offloads &
1229 ~(DEV_RX_OFFLOAD_VLAN_STRIP |
1230 DEV_RX_OFFLOAD_KEEP_CRC |
1231 DEV_RX_OFFLOAD_JUMBO_FRAME |
1232 DEV_RX_OFFLOAD_IPV4_CKSUM |
1233 DEV_RX_OFFLOAD_UDP_CKSUM |
1234 DEV_RX_OFFLOAD_TCP_CKSUM |
1235 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1236 DEV_RX_OFFLOAD_OUTER_UDP_CKSUM |
1237 DEV_RX_OFFLOAD_RSS_HASH |
1238 DEV_RX_OFFLOAD_VLAN_FILTER))
1241 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
1242 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
1243 rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1) {
1245 "Using AVX2 vector mode receive for port %d\n",
1246 eth_dev->data->port_id);
1247 bp->flags |= BNXT_FLAG_RX_VECTOR_PKT_MODE;
1248 return bnxt_recv_pkts_vec_avx2;
1251 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_128) {
1253 "Using SSE vector mode receive for port %d\n",
1254 eth_dev->data->port_id);
1255 bp->flags |= BNXT_FLAG_RX_VECTOR_PKT_MODE;
1256 return bnxt_recv_pkts_vec;
1260 PMD_DRV_LOG(INFO, "Vector mode receive disabled for port %d\n",
1261 eth_dev->data->port_id);
1263 "Port %d scatter: %d rx offload: %" PRIX64 "\n",
1264 eth_dev->data->port_id,
1265 eth_dev->data->scattered_rx,
1266 eth_dev->data->dev_conf.rxmode.offloads);
1268 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1269 return bnxt_recv_pkts;
1272 static eth_tx_burst_t
1273 bnxt_transmit_function(struct rte_eth_dev *eth_dev)
1275 struct bnxt *bp = eth_dev->data->dev_private;
1277 /* Disable vector mode TX for Stingray2 for now */
1278 if (BNXT_CHIP_SR2(bp))
1279 return bnxt_xmit_pkts;
1281 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64) && \
1282 !defined(RTE_LIBRTE_IEEE1588)
1283 uint64_t offloads = eth_dev->data->dev_conf.txmode.offloads;
1286 * Vector mode transmit can be enabled only if not using scatter rx
1289 if (eth_dev->data->scattered_rx ||
1290 (offloads & ~DEV_TX_OFFLOAD_MBUF_FAST_FREE) ||
1291 BNXT_TRUFLOW_EN(bp))
1294 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
1295 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
1296 rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1) {
1298 "Using AVX2 vector mode transmit for port %d\n",
1299 eth_dev->data->port_id);
1300 return bnxt_xmit_pkts_vec_avx2;
1303 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_128) {
1305 "Using SSE vector mode transmit for port %d\n",
1306 eth_dev->data->port_id);
1307 return bnxt_xmit_pkts_vec;
1311 PMD_DRV_LOG(INFO, "Vector mode transmit disabled for port %d\n",
1312 eth_dev->data->port_id);
1314 "Port %d scatter: %d tx offload: %" PRIX64 "\n",
1315 eth_dev->data->port_id,
1316 eth_dev->data->scattered_rx,
1319 return bnxt_xmit_pkts;
1322 static int bnxt_handle_if_change_status(struct bnxt *bp)
1326 /* Since fw has undergone a reset and lost all contexts,
1327 * set fatal flag to not issue hwrm during cleanup
1329 bp->flags |= BNXT_FLAG_FATAL_ERROR;
1330 bnxt_uninit_resources(bp, true);
1332 /* clear fatal flag so that re-init happens */
1333 bp->flags &= ~BNXT_FLAG_FATAL_ERROR;
1334 rc = bnxt_init_resources(bp, true);
1336 bp->flags &= ~BNXT_FLAG_IF_CHANGE_HOT_FW_RESET_DONE;
1341 static int bnxt_dev_set_link_up_op(struct rte_eth_dev *eth_dev)
1343 struct bnxt *bp = eth_dev->data->dev_private;
1346 if (!BNXT_SINGLE_PF(bp))
1349 if (!bp->link_info->link_up)
1350 rc = bnxt_set_hwrm_link_config(bp, true);
1352 eth_dev->data->dev_link.link_status = 1;
1354 bnxt_print_link_info(eth_dev);
1358 static int bnxt_dev_set_link_down_op(struct rte_eth_dev *eth_dev)
1360 struct bnxt *bp = eth_dev->data->dev_private;
1362 if (!BNXT_SINGLE_PF(bp))
1365 eth_dev->data->dev_link.link_status = 0;
1366 bnxt_set_hwrm_link_config(bp, false);
1367 bp->link_info->link_up = 0;
1372 static void bnxt_free_switch_domain(struct bnxt *bp)
1376 if (!(BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)))
1379 rc = rte_eth_switch_domain_free(bp->switch_domain_id);
1381 PMD_DRV_LOG(ERR, "free switch domain:%d fail: %d\n",
1382 bp->switch_domain_id, rc);
1385 static void bnxt_ptp_get_current_time(void *arg)
1387 struct bnxt *bp = arg;
1388 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
1391 rc = is_bnxt_in_error(bp);
1398 bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
1399 &ptp->current_time);
1401 rc = rte_eal_alarm_set(US_PER_S, bnxt_ptp_get_current_time, (void *)bp);
1403 PMD_DRV_LOG(ERR, "Failed to re-schedule PTP alarm\n");
1404 bp->flags2 &= ~BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1408 static int bnxt_schedule_ptp_alarm(struct bnxt *bp)
1410 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
1413 if (bp->flags2 & BNXT_FLAGS2_PTP_ALARM_SCHEDULED)
1416 bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
1417 &ptp->current_time);
1419 rc = rte_eal_alarm_set(US_PER_S, bnxt_ptp_get_current_time, (void *)bp);
1423 static void bnxt_cancel_ptp_alarm(struct bnxt *bp)
1425 if (bp->flags2 & BNXT_FLAGS2_PTP_ALARM_SCHEDULED) {
1426 rte_eal_alarm_cancel(bnxt_ptp_get_current_time, (void *)bp);
1427 bp->flags2 &= ~BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1431 static void bnxt_ptp_stop(struct bnxt *bp)
1433 bnxt_cancel_ptp_alarm(bp);
1434 bp->flags2 &= ~BNXT_FLAGS2_PTP_TIMESYNC_ENABLED;
1437 static int bnxt_ptp_start(struct bnxt *bp)
1441 rc = bnxt_schedule_ptp_alarm(bp);
1443 PMD_DRV_LOG(ERR, "Failed to schedule PTP alarm\n");
1445 bp->flags2 |= BNXT_FLAGS2_PTP_TIMESYNC_ENABLED;
1446 bp->flags2 |= BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1452 static int bnxt_dev_stop(struct rte_eth_dev *eth_dev)
1454 struct bnxt *bp = eth_dev->data->dev_private;
1455 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1456 struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1457 struct rte_eth_link link;
1460 eth_dev->data->dev_started = 0;
1461 eth_dev->data->scattered_rx = 0;
1463 /* Prevent crashes when queues are still in use */
1464 eth_dev->rx_pkt_burst = &bnxt_dummy_recv_pkts;
1465 eth_dev->tx_pkt_burst = &bnxt_dummy_xmit_pkts;
1467 bnxt_disable_int(bp);
1469 /* disable uio/vfio intr/eventfd mapping */
1470 rte_intr_disable(intr_handle);
1472 /* Stop the child representors for this device */
1473 ret = bnxt_rep_stop_all(bp);
1477 /* delete the bnxt ULP port details */
1478 bnxt_ulp_port_deinit(bp);
1480 bnxt_cancel_fw_health_check(bp);
1482 if (BNXT_P5_PTP_TIMESYNC_ENABLED(bp))
1483 bnxt_cancel_ptp_alarm(bp);
1485 /* Do not bring link down during reset recovery */
1486 if (!is_bnxt_in_error(bp)) {
1487 bnxt_dev_set_link_down_op(eth_dev);
1488 /* Wait for link to be reset */
1489 if (BNXT_SINGLE_PF(bp))
1491 /* clear the recorded link status */
1492 memset(&link, 0, sizeof(link));
1493 rte_eth_linkstatus_set(eth_dev, &link);
1496 /* Clean queue intr-vector mapping */
1497 rte_intr_efd_disable(intr_handle);
1498 if (intr_handle->intr_vec != NULL) {
1499 rte_free(intr_handle->intr_vec);
1500 intr_handle->intr_vec = NULL;
1503 bnxt_hwrm_port_clr_stats(bp);
1504 bnxt_free_tx_mbufs(bp);
1505 bnxt_free_rx_mbufs(bp);
1506 /* Process any remaining notifications in default completion queue */
1507 bnxt_int_handler(eth_dev);
1508 bnxt_shutdown_nic(bp);
1509 bnxt_hwrm_if_change(bp, false);
1511 bnxt_free_prev_ring_stats(bp);
1512 rte_free(bp->mark_table);
1513 bp->mark_table = NULL;
1515 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1516 bp->rx_cosq_cnt = 0;
1517 /* All filters are deleted on a port stop. */
1518 if (BNXT_FLOW_XSTATS_EN(bp))
1519 bp->flow_stat->flow_count = 0;
1524 /* Unload the driver, release resources */
1525 static int bnxt_dev_stop_op(struct rte_eth_dev *eth_dev)
1527 struct bnxt *bp = eth_dev->data->dev_private;
1529 pthread_mutex_lock(&bp->err_recovery_lock);
1530 if (bp->flags & BNXT_FLAG_FW_RESET) {
1532 "Adapter recovering from error..Please retry\n");
1533 pthread_mutex_unlock(&bp->err_recovery_lock);
1536 pthread_mutex_unlock(&bp->err_recovery_lock);
1538 return bnxt_dev_stop(eth_dev);
1541 static int bnxt_dev_start_op(struct rte_eth_dev *eth_dev)
1543 struct bnxt *bp = eth_dev->data->dev_private;
1544 uint64_t rx_offloads = eth_dev->data->dev_conf.rxmode.offloads;
1546 int rc, retry_cnt = BNXT_IF_CHANGE_RETRY_COUNT;
1548 if (!eth_dev->data->nb_tx_queues || !eth_dev->data->nb_rx_queues) {
1549 PMD_DRV_LOG(ERR, "Queues are not configured yet!\n");
1553 if (bp->rx_cp_nr_rings > RTE_ETHDEV_QUEUE_STAT_CNTRS)
1555 "RxQ cnt %d > RTE_ETHDEV_QUEUE_STAT_CNTRS %d\n",
1556 bp->rx_cp_nr_rings, RTE_ETHDEV_QUEUE_STAT_CNTRS);
1559 rc = bnxt_hwrm_if_change(bp, true);
1560 if (rc == 0 || rc != -EAGAIN)
1563 rte_delay_ms(BNXT_IF_CHANGE_RETRY_INTERVAL);
1564 } while (retry_cnt--);
1569 if (bp->flags & BNXT_FLAG_IF_CHANGE_HOT_FW_RESET_DONE) {
1570 rc = bnxt_handle_if_change_status(bp);
1575 bnxt_enable_int(bp);
1577 eth_dev->data->scattered_rx = bnxt_scattered_rx(eth_dev);
1579 rc = bnxt_start_nic(bp);
1583 rc = bnxt_alloc_prev_ring_stats(bp);
1587 eth_dev->data->dev_started = 1;
1589 bnxt_link_update_op(eth_dev, 1);
1591 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1592 vlan_mask |= ETH_VLAN_FILTER_MASK;
1593 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
1594 vlan_mask |= ETH_VLAN_STRIP_MASK;
1595 rc = bnxt_vlan_offload_set_op(eth_dev, vlan_mask);
1599 /* Initialize bnxt ULP port details */
1600 rc = bnxt_ulp_port_init(bp);
1604 eth_dev->rx_pkt_burst = bnxt_receive_function(eth_dev);
1605 eth_dev->tx_pkt_burst = bnxt_transmit_function(eth_dev);
1607 bnxt_schedule_fw_health_check(bp);
1609 if (BNXT_P5_PTP_TIMESYNC_ENABLED(bp))
1610 bnxt_schedule_ptp_alarm(bp);
1615 bnxt_dev_stop(eth_dev);
1620 bnxt_uninit_locks(struct bnxt *bp)
1622 pthread_mutex_destroy(&bp->flow_lock);
1623 pthread_mutex_destroy(&bp->def_cp_lock);
1624 pthread_mutex_destroy(&bp->health_check_lock);
1625 pthread_mutex_destroy(&bp->err_recovery_lock);
1627 pthread_mutex_destroy(&bp->rep_info->vfr_lock);
1628 pthread_mutex_destroy(&bp->rep_info->vfr_start_lock);
1632 static void bnxt_drv_uninit(struct bnxt *bp)
1634 bnxt_free_leds_info(bp);
1635 bnxt_free_cos_queues(bp);
1636 bnxt_free_link_info(bp);
1637 bnxt_free_parent_info(bp);
1638 bnxt_uninit_locks(bp);
1640 rte_memzone_free((const struct rte_memzone *)bp->tx_mem_zone);
1641 bp->tx_mem_zone = NULL;
1642 rte_memzone_free((const struct rte_memzone *)bp->rx_mem_zone);
1643 bp->rx_mem_zone = NULL;
1645 bnxt_free_vf_info(bp);
1646 bnxt_free_pf_info(bp);
1648 rte_free(bp->grp_info);
1649 bp->grp_info = NULL;
1652 static int bnxt_dev_close_op(struct rte_eth_dev *eth_dev)
1654 struct bnxt *bp = eth_dev->data->dev_private;
1657 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1660 pthread_mutex_lock(&bp->err_recovery_lock);
1661 if (bp->flags & BNXT_FLAG_FW_RESET) {
1663 "Adapter recovering from error...Please retry\n");
1664 pthread_mutex_unlock(&bp->err_recovery_lock);
1667 pthread_mutex_unlock(&bp->err_recovery_lock);
1669 /* cancel the recovery handler before remove dev */
1670 rte_eal_alarm_cancel(bnxt_dev_reset_and_resume, (void *)bp);
1671 rte_eal_alarm_cancel(bnxt_dev_recover, (void *)bp);
1672 bnxt_cancel_fc_thread(bp);
1674 if (eth_dev->data->dev_started)
1675 ret = bnxt_dev_stop(eth_dev);
1677 bnxt_uninit_resources(bp, false);
1679 bnxt_drv_uninit(bp);
1684 static void bnxt_mac_addr_remove_op(struct rte_eth_dev *eth_dev,
1687 struct bnxt *bp = eth_dev->data->dev_private;
1688 uint64_t pool_mask = eth_dev->data->mac_pool_sel[index];
1689 struct bnxt_vnic_info *vnic;
1690 struct bnxt_filter_info *filter, *temp_filter;
1693 if (is_bnxt_in_error(bp))
1697 * Loop through all VNICs from the specified filter flow pools to
1698 * remove the corresponding MAC addr filter
1700 for (i = 0; i < bp->nr_vnics; i++) {
1701 if (!(pool_mask & (1ULL << i)))
1704 vnic = &bp->vnic_info[i];
1705 filter = STAILQ_FIRST(&vnic->filter);
1707 temp_filter = STAILQ_NEXT(filter, next);
1708 if (filter->mac_index == index) {
1709 STAILQ_REMOVE(&vnic->filter, filter,
1710 bnxt_filter_info, next);
1711 bnxt_hwrm_clear_l2_filter(bp, filter);
1712 bnxt_free_filter(bp, filter);
1714 filter = temp_filter;
1719 static int bnxt_add_mac_filter(struct bnxt *bp, struct bnxt_vnic_info *vnic,
1720 struct rte_ether_addr *mac_addr, uint32_t index,
1723 struct bnxt_filter_info *filter;
1726 /* Attach requested MAC address to the new l2_filter */
1727 STAILQ_FOREACH(filter, &vnic->filter, next) {
1728 if (filter->mac_index == index) {
1730 "MAC addr already existed for pool %d\n",
1736 filter = bnxt_alloc_filter(bp);
1738 PMD_DRV_LOG(ERR, "L2 filter alloc failed\n");
1742 /* bnxt_alloc_filter copies default MAC to filter->l2_addr. So,
1743 * if the MAC that's been programmed now is a different one, then,
1744 * copy that addr to filter->l2_addr
1747 memcpy(filter->l2_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1748 filter->flags |= HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
1750 rc = bnxt_hwrm_set_l2_filter(bp, vnic->fw_vnic_id, filter);
1752 filter->mac_index = index;
1753 if (filter->mac_index == 0)
1754 STAILQ_INSERT_HEAD(&vnic->filter, filter, next);
1756 STAILQ_INSERT_TAIL(&vnic->filter, filter, next);
1758 bnxt_free_filter(bp, filter);
1764 static int bnxt_mac_addr_add_op(struct rte_eth_dev *eth_dev,
1765 struct rte_ether_addr *mac_addr,
1766 uint32_t index, uint32_t pool)
1768 struct bnxt *bp = eth_dev->data->dev_private;
1769 struct bnxt_vnic_info *vnic = &bp->vnic_info[pool];
1772 rc = is_bnxt_in_error(bp);
1776 if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp)) {
1777 PMD_DRV_LOG(ERR, "Cannot add MAC address to a VF interface\n");
1782 PMD_DRV_LOG(ERR, "VNIC not found for pool %d!\n", pool);
1786 /* Filter settings will get applied when port is started */
1787 if (!eth_dev->data->dev_started)
1790 rc = bnxt_add_mac_filter(bp, vnic, mac_addr, index, pool);
1795 int bnxt_link_update_op(struct rte_eth_dev *eth_dev, int wait_to_complete)
1798 struct bnxt *bp = eth_dev->data->dev_private;
1799 struct rte_eth_link new;
1800 int cnt = wait_to_complete ? BNXT_MAX_LINK_WAIT_CNT :
1801 BNXT_MIN_LINK_WAIT_CNT;
1803 rc = is_bnxt_in_error(bp);
1807 memset(&new, 0, sizeof(new));
1809 if (bp->link_info == NULL)
1813 /* Retrieve link info from hardware */
1814 rc = bnxt_get_hwrm_link_config(bp, &new);
1816 new.link_speed = ETH_LINK_SPEED_100M;
1817 new.link_duplex = ETH_LINK_FULL_DUPLEX;
1819 "Failed to retrieve link rc = 0x%x!\n", rc);
1823 if (!wait_to_complete || new.link_status)
1826 rte_delay_ms(BNXT_LINK_WAIT_INTERVAL);
1829 /* Only single function PF can bring phy down.
1830 * When port is stopped, report link down for VF/MH/NPAR functions.
1832 if (!BNXT_SINGLE_PF(bp) && !eth_dev->data->dev_started)
1833 memset(&new, 0, sizeof(new));
1836 /* Timed out or success */
1837 if (new.link_status != eth_dev->data->dev_link.link_status ||
1838 new.link_speed != eth_dev->data->dev_link.link_speed) {
1839 rte_eth_linkstatus_set(eth_dev, &new);
1841 rte_eth_dev_callback_process(eth_dev,
1842 RTE_ETH_EVENT_INTR_LSC,
1845 bnxt_print_link_info(eth_dev);
1851 static int bnxt_promiscuous_enable_op(struct rte_eth_dev *eth_dev)
1853 struct bnxt *bp = eth_dev->data->dev_private;
1854 struct bnxt_vnic_info *vnic;
1858 rc = is_bnxt_in_error(bp);
1862 /* Filter settings will get applied when port is started */
1863 if (!eth_dev->data->dev_started)
1866 if (bp->vnic_info == NULL)
1869 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1871 old_flags = vnic->flags;
1872 vnic->flags |= BNXT_VNIC_INFO_PROMISC;
1873 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1875 vnic->flags = old_flags;
1880 static int bnxt_promiscuous_disable_op(struct rte_eth_dev *eth_dev)
1882 struct bnxt *bp = eth_dev->data->dev_private;
1883 struct bnxt_vnic_info *vnic;
1887 rc = is_bnxt_in_error(bp);
1891 /* Filter settings will get applied when port is started */
1892 if (!eth_dev->data->dev_started)
1895 if (bp->vnic_info == NULL)
1898 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1900 old_flags = vnic->flags;
1901 vnic->flags &= ~BNXT_VNIC_INFO_PROMISC;
1902 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1904 vnic->flags = old_flags;
1909 static int bnxt_allmulticast_enable_op(struct rte_eth_dev *eth_dev)
1911 struct bnxt *bp = eth_dev->data->dev_private;
1912 struct bnxt_vnic_info *vnic;
1916 rc = is_bnxt_in_error(bp);
1920 /* Filter settings will get applied when port is started */
1921 if (!eth_dev->data->dev_started)
1924 if (bp->vnic_info == NULL)
1927 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1929 old_flags = vnic->flags;
1930 vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
1931 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1933 vnic->flags = old_flags;
1938 static int bnxt_allmulticast_disable_op(struct rte_eth_dev *eth_dev)
1940 struct bnxt *bp = eth_dev->data->dev_private;
1941 struct bnxt_vnic_info *vnic;
1945 rc = is_bnxt_in_error(bp);
1949 /* Filter settings will get applied when port is started */
1950 if (!eth_dev->data->dev_started)
1953 if (bp->vnic_info == NULL)
1956 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1958 old_flags = vnic->flags;
1959 vnic->flags &= ~BNXT_VNIC_INFO_ALLMULTI;
1960 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1962 vnic->flags = old_flags;
1967 /* Return bnxt_rx_queue pointer corresponding to a given rxq. */
1968 static struct bnxt_rx_queue *bnxt_qid_to_rxq(struct bnxt *bp, uint16_t qid)
1970 if (qid >= bp->rx_nr_rings)
1973 return bp->eth_dev->data->rx_queues[qid];
1976 /* Return rxq corresponding to a given rss table ring/group ID. */
1977 static uint16_t bnxt_rss_to_qid(struct bnxt *bp, uint16_t fwr)
1979 struct bnxt_rx_queue *rxq;
1982 if (!BNXT_HAS_RING_GRPS(bp)) {
1983 for (i = 0; i < bp->rx_nr_rings; i++) {
1984 rxq = bp->eth_dev->data->rx_queues[i];
1985 if (rxq->rx_ring->rx_ring_struct->fw_ring_id == fwr)
1989 for (i = 0; i < bp->rx_nr_rings; i++) {
1990 if (bp->grp_info[i].fw_grp_id == fwr)
1995 return INVALID_HW_RING_ID;
1998 static int bnxt_reta_update_op(struct rte_eth_dev *eth_dev,
1999 struct rte_eth_rss_reta_entry64 *reta_conf,
2002 struct bnxt *bp = eth_dev->data->dev_private;
2003 struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
2004 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2005 uint16_t tbl_size = bnxt_rss_hash_tbl_size(bp);
2009 rc = is_bnxt_in_error(bp);
2013 if (!vnic->rss_table)
2016 if (!(dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
2019 if (reta_size != tbl_size) {
2020 PMD_DRV_LOG(ERR, "The configured hash table lookup size "
2021 "(%d) must equal the size supported by the hardware "
2022 "(%d)\n", reta_size, tbl_size);
2026 for (i = 0; i < reta_size; i++) {
2027 struct bnxt_rx_queue *rxq;
2029 idx = i / RTE_RETA_GROUP_SIZE;
2030 sft = i % RTE_RETA_GROUP_SIZE;
2032 if (!(reta_conf[idx].mask & (1ULL << sft)))
2035 rxq = bnxt_qid_to_rxq(bp, reta_conf[idx].reta[sft]);
2037 PMD_DRV_LOG(ERR, "Invalid ring in reta_conf.\n");
2041 if (BNXT_CHIP_P5(bp)) {
2042 vnic->rss_table[i * 2] =
2043 rxq->rx_ring->rx_ring_struct->fw_ring_id;
2044 vnic->rss_table[i * 2 + 1] =
2045 rxq->cp_ring->cp_ring_struct->fw_ring_id;
2047 vnic->rss_table[i] =
2048 vnic->fw_grp_ids[reta_conf[idx].reta[sft]];
2052 rc = bnxt_hwrm_vnic_rss_cfg(bp, vnic);
2056 static int bnxt_reta_query_op(struct rte_eth_dev *eth_dev,
2057 struct rte_eth_rss_reta_entry64 *reta_conf,
2060 struct bnxt *bp = eth_dev->data->dev_private;
2061 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2062 uint16_t tbl_size = bnxt_rss_hash_tbl_size(bp);
2063 uint16_t idx, sft, i;
2066 rc = is_bnxt_in_error(bp);
2070 /* Retrieve from the default VNIC */
2073 if (!vnic->rss_table)
2076 if (reta_size != tbl_size) {
2077 PMD_DRV_LOG(ERR, "The configured hash table lookup size "
2078 "(%d) must equal the size supported by the hardware "
2079 "(%d)\n", reta_size, tbl_size);
2083 for (idx = 0, i = 0; i < reta_size; i++) {
2084 idx = i / RTE_RETA_GROUP_SIZE;
2085 sft = i % RTE_RETA_GROUP_SIZE;
2087 if (reta_conf[idx].mask & (1ULL << sft)) {
2090 if (BNXT_CHIP_P5(bp))
2091 qid = bnxt_rss_to_qid(bp,
2092 vnic->rss_table[i * 2]);
2094 qid = bnxt_rss_to_qid(bp, vnic->rss_table[i]);
2096 if (qid == INVALID_HW_RING_ID) {
2097 PMD_DRV_LOG(ERR, "Inv. entry in rss table.\n");
2100 reta_conf[idx].reta[sft] = qid;
2107 static int bnxt_rss_hash_update_op(struct rte_eth_dev *eth_dev,
2108 struct rte_eth_rss_conf *rss_conf)
2110 struct bnxt *bp = eth_dev->data->dev_private;
2111 struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
2112 struct bnxt_vnic_info *vnic;
2115 rc = is_bnxt_in_error(bp);
2120 * If RSS enablement were different than dev_configure,
2121 * then return -EINVAL
2123 if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
2124 if (!rss_conf->rss_hf)
2125 PMD_DRV_LOG(ERR, "Hash type NONE\n");
2127 if (rss_conf->rss_hf & BNXT_ETH_RSS_SUPPORT)
2131 bp->flags |= BNXT_FLAG_UPDATE_HASH;
2132 memcpy(ð_dev->data->dev_conf.rx_adv_conf.rss_conf,
2136 /* Update the default RSS VNIC(s) */
2137 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2138 vnic->hash_type = bnxt_rte_to_hwrm_hash_types(rss_conf->rss_hf);
2140 bnxt_rte_to_hwrm_hash_level(bp, rss_conf->rss_hf,
2141 ETH_RSS_LEVEL(rss_conf->rss_hf));
2144 * If hashkey is not specified, use the previously configured
2147 if (!rss_conf->rss_key)
2150 if (rss_conf->rss_key_len != HW_HASH_KEY_SIZE) {
2152 "Invalid hashkey length, should be 16 bytes\n");
2155 memcpy(vnic->rss_hash_key, rss_conf->rss_key, rss_conf->rss_key_len);
2158 rc = bnxt_hwrm_vnic_rss_cfg(bp, vnic);
2162 static int bnxt_rss_hash_conf_get_op(struct rte_eth_dev *eth_dev,
2163 struct rte_eth_rss_conf *rss_conf)
2165 struct bnxt *bp = eth_dev->data->dev_private;
2166 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2168 uint32_t hash_types;
2170 rc = is_bnxt_in_error(bp);
2174 /* RSS configuration is the same for all VNICs */
2175 if (vnic && vnic->rss_hash_key) {
2176 if (rss_conf->rss_key) {
2177 len = rss_conf->rss_key_len <= HW_HASH_KEY_SIZE ?
2178 rss_conf->rss_key_len : HW_HASH_KEY_SIZE;
2179 memcpy(rss_conf->rss_key, vnic->rss_hash_key, len);
2182 hash_types = vnic->hash_type;
2183 rss_conf->rss_hf = 0;
2184 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV4) {
2185 rss_conf->rss_hf |= ETH_RSS_IPV4;
2186 hash_types &= ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV4;
2188 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV4) {
2189 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
2191 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV4;
2193 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV4) {
2194 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
2196 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV4;
2198 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV6) {
2199 rss_conf->rss_hf |= ETH_RSS_IPV6;
2200 hash_types &= ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV6;
2202 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV6) {
2203 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
2205 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV6;
2207 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV6) {
2208 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
2210 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV6;
2214 bnxt_hwrm_to_rte_rss_level(bp, vnic->hash_mode);
2218 "Unknown RSS config from firmware (%08x), RSS disabled",
2223 rss_conf->rss_hf = 0;
2228 static int bnxt_flow_ctrl_get_op(struct rte_eth_dev *dev,
2229 struct rte_eth_fc_conf *fc_conf)
2231 struct bnxt *bp = dev->data->dev_private;
2232 struct rte_eth_link link_info;
2235 rc = is_bnxt_in_error(bp);
2239 rc = bnxt_get_hwrm_link_config(bp, &link_info);
2243 memset(fc_conf, 0, sizeof(*fc_conf));
2244 if (bp->link_info->auto_pause)
2245 fc_conf->autoneg = 1;
2246 switch (bp->link_info->pause) {
2248 fc_conf->mode = RTE_FC_NONE;
2250 case HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_TX:
2251 fc_conf->mode = RTE_FC_TX_PAUSE;
2253 case HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_RX:
2254 fc_conf->mode = RTE_FC_RX_PAUSE;
2256 case (HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_TX |
2257 HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_RX):
2258 fc_conf->mode = RTE_FC_FULL;
2264 static int bnxt_flow_ctrl_set_op(struct rte_eth_dev *dev,
2265 struct rte_eth_fc_conf *fc_conf)
2267 struct bnxt *bp = dev->data->dev_private;
2270 rc = is_bnxt_in_error(bp);
2274 if (!BNXT_SINGLE_PF(bp)) {
2276 "Flow Control Settings cannot be modified on VF or on shared PF\n");
2280 switch (fc_conf->mode) {
2282 bp->link_info->auto_pause = 0;
2283 bp->link_info->force_pause = 0;
2285 case RTE_FC_RX_PAUSE:
2286 if (fc_conf->autoneg) {
2287 bp->link_info->auto_pause =
2288 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_RX;
2289 bp->link_info->force_pause = 0;
2291 bp->link_info->auto_pause = 0;
2292 bp->link_info->force_pause =
2293 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_RX;
2296 case RTE_FC_TX_PAUSE:
2297 if (fc_conf->autoneg) {
2298 bp->link_info->auto_pause =
2299 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_TX;
2300 bp->link_info->force_pause = 0;
2302 bp->link_info->auto_pause = 0;
2303 bp->link_info->force_pause =
2304 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_TX;
2308 if (fc_conf->autoneg) {
2309 bp->link_info->auto_pause =
2310 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_TX |
2311 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_RX;
2312 bp->link_info->force_pause = 0;
2314 bp->link_info->auto_pause = 0;
2315 bp->link_info->force_pause =
2316 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_TX |
2317 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_RX;
2321 return bnxt_set_hwrm_link_config(bp, true);
2324 /* Add UDP tunneling port */
2326 bnxt_udp_tunnel_port_add_op(struct rte_eth_dev *eth_dev,
2327 struct rte_eth_udp_tunnel *udp_tunnel)
2329 struct bnxt *bp = eth_dev->data->dev_private;
2330 uint16_t tunnel_type = 0;
2333 rc = is_bnxt_in_error(bp);
2337 switch (udp_tunnel->prot_type) {
2338 case RTE_TUNNEL_TYPE_VXLAN:
2339 if (bp->vxlan_port_cnt) {
2340 PMD_DRV_LOG(ERR, "Tunnel Port %d already programmed\n",
2341 udp_tunnel->udp_port);
2342 if (bp->vxlan_port != udp_tunnel->udp_port) {
2343 PMD_DRV_LOG(ERR, "Only one port allowed\n");
2346 bp->vxlan_port_cnt++;
2350 HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_VXLAN;
2351 bp->vxlan_port_cnt++;
2353 case RTE_TUNNEL_TYPE_GENEVE:
2354 if (bp->geneve_port_cnt) {
2355 PMD_DRV_LOG(ERR, "Tunnel Port %d already programmed\n",
2356 udp_tunnel->udp_port);
2357 if (bp->geneve_port != udp_tunnel->udp_port) {
2358 PMD_DRV_LOG(ERR, "Only one port allowed\n");
2361 bp->geneve_port_cnt++;
2365 HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_GENEVE;
2366 bp->geneve_port_cnt++;
2369 PMD_DRV_LOG(ERR, "Tunnel type is not supported\n");
2372 rc = bnxt_hwrm_tunnel_dst_port_alloc(bp, udp_tunnel->udp_port,
2378 bnxt_udp_tunnel_port_del_op(struct rte_eth_dev *eth_dev,
2379 struct rte_eth_udp_tunnel *udp_tunnel)
2381 struct bnxt *bp = eth_dev->data->dev_private;
2382 uint16_t tunnel_type = 0;
2386 rc = is_bnxt_in_error(bp);
2390 switch (udp_tunnel->prot_type) {
2391 case RTE_TUNNEL_TYPE_VXLAN:
2392 if (!bp->vxlan_port_cnt) {
2393 PMD_DRV_LOG(ERR, "No Tunnel port configured yet\n");
2396 if (bp->vxlan_port != udp_tunnel->udp_port) {
2397 PMD_DRV_LOG(ERR, "Req Port: %d. Configured port: %d\n",
2398 udp_tunnel->udp_port, bp->vxlan_port);
2401 if (--bp->vxlan_port_cnt)
2405 HWRM_TUNNEL_DST_PORT_FREE_INPUT_TUNNEL_TYPE_VXLAN;
2406 port = bp->vxlan_fw_dst_port_id;
2408 case RTE_TUNNEL_TYPE_GENEVE:
2409 if (!bp->geneve_port_cnt) {
2410 PMD_DRV_LOG(ERR, "No Tunnel port configured yet\n");
2413 if (bp->geneve_port != udp_tunnel->udp_port) {
2414 PMD_DRV_LOG(ERR, "Req Port: %d. Configured port: %d\n",
2415 udp_tunnel->udp_port, bp->geneve_port);
2418 if (--bp->geneve_port_cnt)
2422 HWRM_TUNNEL_DST_PORT_FREE_INPUT_TUNNEL_TYPE_GENEVE;
2423 port = bp->geneve_fw_dst_port_id;
2426 PMD_DRV_LOG(ERR, "Tunnel type is not supported\n");
2430 rc = bnxt_hwrm_tunnel_dst_port_free(bp, port, tunnel_type);
2434 static int bnxt_del_vlan_filter(struct bnxt *bp, uint16_t vlan_id)
2436 struct bnxt_filter_info *filter;
2437 struct bnxt_vnic_info *vnic;
2439 uint32_t chk = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN;
2441 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2442 filter = STAILQ_FIRST(&vnic->filter);
2444 /* Search for this matching MAC+VLAN filter */
2445 if (bnxt_vlan_filter_exists(bp, filter, chk, vlan_id)) {
2446 /* Delete the filter */
2447 rc = bnxt_hwrm_clear_l2_filter(bp, filter);
2450 STAILQ_REMOVE(&vnic->filter, filter,
2451 bnxt_filter_info, next);
2452 bnxt_free_filter(bp, filter);
2454 "Deleted vlan filter for %d\n",
2458 filter = STAILQ_NEXT(filter, next);
2463 static int bnxt_add_vlan_filter(struct bnxt *bp, uint16_t vlan_id)
2465 struct bnxt_filter_info *filter;
2466 struct bnxt_vnic_info *vnic;
2468 uint32_t en = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN |
2469 HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN_MASK;
2470 uint32_t chk = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN;
2472 /* Implementation notes on the use of VNIC in this command:
2474 * By default, these filters belong to default vnic for the function.
2475 * Once these filters are set up, only destination VNIC can be modified.
2476 * If the destination VNIC is not specified in this command,
2477 * then the HWRM shall only create an l2 context id.
2480 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2481 filter = STAILQ_FIRST(&vnic->filter);
2482 /* Check if the VLAN has already been added */
2484 if (bnxt_vlan_filter_exists(bp, filter, chk, vlan_id))
2487 filter = STAILQ_NEXT(filter, next);
2490 /* No match found. Alloc a fresh filter and issue the L2_FILTER_ALLOC
2491 * command to create MAC+VLAN filter with the right flags, enables set.
2493 filter = bnxt_alloc_filter(bp);
2496 "MAC/VLAN filter alloc failed\n");
2499 /* MAC + VLAN ID filter */
2500 /* If l2_ivlan == 0 and l2_ivlan_mask != 0, only
2501 * untagged packets are received
2503 * If l2_ivlan != 0 and l2_ivlan_mask != 0, untagged
2504 * packets and only the programmed vlan's packets are received
2506 filter->l2_ivlan = vlan_id;
2507 filter->l2_ivlan_mask = 0x0FFF;
2508 filter->enables |= en;
2509 filter->flags |= HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
2511 rc = bnxt_hwrm_set_l2_filter(bp, vnic->fw_vnic_id, filter);
2513 /* Free the newly allocated filter as we were
2514 * not able to create the filter in hardware.
2516 bnxt_free_filter(bp, filter);
2520 filter->mac_index = 0;
2521 /* Add this new filter to the list */
2523 STAILQ_INSERT_HEAD(&vnic->filter, filter, next);
2525 STAILQ_INSERT_TAIL(&vnic->filter, filter, next);
2528 "Added Vlan filter for %d\n", vlan_id);
2532 static int bnxt_vlan_filter_set_op(struct rte_eth_dev *eth_dev,
2533 uint16_t vlan_id, int on)
2535 struct bnxt *bp = eth_dev->data->dev_private;
2538 rc = is_bnxt_in_error(bp);
2542 if (!eth_dev->data->dev_started) {
2543 PMD_DRV_LOG(ERR, "port must be started before setting vlan\n");
2547 /* These operations apply to ALL existing MAC/VLAN filters */
2549 return bnxt_add_vlan_filter(bp, vlan_id);
2551 return bnxt_del_vlan_filter(bp, vlan_id);
2554 static int bnxt_del_dflt_mac_filter(struct bnxt *bp,
2555 struct bnxt_vnic_info *vnic)
2557 struct bnxt_filter_info *filter;
2560 filter = STAILQ_FIRST(&vnic->filter);
2562 if (filter->mac_index == 0 &&
2563 !memcmp(filter->l2_addr, bp->mac_addr,
2564 RTE_ETHER_ADDR_LEN)) {
2565 rc = bnxt_hwrm_clear_l2_filter(bp, filter);
2567 STAILQ_REMOVE(&vnic->filter, filter,
2568 bnxt_filter_info, next);
2569 bnxt_free_filter(bp, filter);
2573 filter = STAILQ_NEXT(filter, next);
2579 bnxt_config_vlan_hw_filter(struct bnxt *bp, uint64_t rx_offloads)
2581 struct bnxt_vnic_info *vnic;
2585 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2586 if (!(rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)) {
2587 /* Remove any VLAN filters programmed */
2588 for (i = 0; i < RTE_ETHER_MAX_VLAN_ID; i++)
2589 bnxt_del_vlan_filter(bp, i);
2591 rc = bnxt_add_mac_filter(bp, vnic, NULL, 0, 0);
2595 /* Default filter will allow packets that match the
2596 * dest mac. So, it has to be deleted, otherwise, we
2597 * will endup receiving vlan packets for which the
2598 * filter is not programmed, when hw-vlan-filter
2599 * configuration is ON
2601 bnxt_del_dflt_mac_filter(bp, vnic);
2602 /* This filter will allow only untagged packets */
2603 bnxt_add_vlan_filter(bp, 0);
2605 PMD_DRV_LOG(DEBUG, "VLAN Filtering: %d\n",
2606 !!(rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER));
2611 static int bnxt_free_one_vnic(struct bnxt *bp, uint16_t vnic_id)
2613 struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
2617 /* Destroy vnic filters and vnic */
2618 if (bp->eth_dev->data->dev_conf.rxmode.offloads &
2619 DEV_RX_OFFLOAD_VLAN_FILTER) {
2620 for (i = 0; i < RTE_ETHER_MAX_VLAN_ID; i++)
2621 bnxt_del_vlan_filter(bp, i);
2623 bnxt_del_dflt_mac_filter(bp, vnic);
2625 rc = bnxt_hwrm_vnic_ctx_free(bp, vnic);
2629 rc = bnxt_hwrm_vnic_free(bp, vnic);
2633 rte_free(vnic->fw_grp_ids);
2634 vnic->fw_grp_ids = NULL;
2636 vnic->rx_queue_cnt = 0;
2642 bnxt_config_vlan_hw_stripping(struct bnxt *bp, uint64_t rx_offloads)
2644 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2647 /* Destroy, recreate and reconfigure the default vnic */
2648 rc = bnxt_free_one_vnic(bp, 0);
2652 /* default vnic 0 */
2653 rc = bnxt_setup_one_vnic(bp, 0);
2657 if (bp->eth_dev->data->dev_conf.rxmode.offloads &
2658 DEV_RX_OFFLOAD_VLAN_FILTER) {
2659 rc = bnxt_add_vlan_filter(bp, 0);
2662 rc = bnxt_restore_vlan_filters(bp);
2666 rc = bnxt_add_mac_filter(bp, vnic, NULL, 0, 0);
2671 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
2675 PMD_DRV_LOG(DEBUG, "VLAN Strip Offload: %d\n",
2676 !!(rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP));
2682 bnxt_vlan_offload_set_op(struct rte_eth_dev *dev, int mask)
2684 uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
2685 struct bnxt *bp = dev->data->dev_private;
2688 rc = is_bnxt_in_error(bp);
2692 /* Filter settings will get applied when port is started */
2693 if (!dev->data->dev_started)
2696 if (mask & ETH_VLAN_FILTER_MASK) {
2697 /* Enable or disable VLAN filtering */
2698 rc = bnxt_config_vlan_hw_filter(bp, rx_offloads);
2703 if (mask & ETH_VLAN_STRIP_MASK) {
2704 /* Enable or disable VLAN stripping */
2705 rc = bnxt_config_vlan_hw_stripping(bp, rx_offloads);
2710 if (mask & ETH_VLAN_EXTEND_MASK) {
2711 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)
2712 PMD_DRV_LOG(DEBUG, "Extend VLAN supported\n");
2714 PMD_DRV_LOG(INFO, "Extend VLAN unsupported\n");
2721 bnxt_vlan_tpid_set_op(struct rte_eth_dev *dev, enum rte_vlan_type vlan_type,
2724 struct bnxt *bp = dev->data->dev_private;
2725 int qinq = dev->data->dev_conf.rxmode.offloads &
2726 DEV_RX_OFFLOAD_VLAN_EXTEND;
2728 if (vlan_type != ETH_VLAN_TYPE_INNER &&
2729 vlan_type != ETH_VLAN_TYPE_OUTER) {
2731 "Unsupported vlan type.");
2736 "QinQ not enabled. Needs to be ON as we can "
2737 "accelerate only outer vlan\n");
2741 if (vlan_type == ETH_VLAN_TYPE_OUTER) {
2743 case RTE_ETHER_TYPE_QINQ:
2745 TX_BD_LONG_CFA_META_VLAN_TPID_TPID88A8;
2747 case RTE_ETHER_TYPE_VLAN:
2749 TX_BD_LONG_CFA_META_VLAN_TPID_TPID8100;
2751 case RTE_ETHER_TYPE_QINQ1:
2753 TX_BD_LONG_CFA_META_VLAN_TPID_TPID9100;
2755 case RTE_ETHER_TYPE_QINQ2:
2757 TX_BD_LONG_CFA_META_VLAN_TPID_TPID9200;
2759 case RTE_ETHER_TYPE_QINQ3:
2761 TX_BD_LONG_CFA_META_VLAN_TPID_TPID9300;
2764 PMD_DRV_LOG(ERR, "Invalid TPID: %x\n", tpid);
2767 bp->outer_tpid_bd |= tpid;
2768 PMD_DRV_LOG(INFO, "outer_tpid_bd = %x\n", bp->outer_tpid_bd);
2769 } else if (vlan_type == ETH_VLAN_TYPE_INNER) {
2771 "Can accelerate only outer vlan in QinQ\n");
2779 bnxt_set_default_mac_addr_op(struct rte_eth_dev *dev,
2780 struct rte_ether_addr *addr)
2782 struct bnxt *bp = dev->data->dev_private;
2783 /* Default Filter is tied to VNIC 0 */
2784 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2787 rc = is_bnxt_in_error(bp);
2791 if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp))
2794 if (rte_is_zero_ether_addr(addr))
2797 /* Filter settings will get applied when port is started */
2798 if (!dev->data->dev_started)
2801 /* Check if the requested MAC is already added */
2802 if (memcmp(addr, bp->mac_addr, RTE_ETHER_ADDR_LEN) == 0)
2805 /* Destroy filter and re-create it */
2806 bnxt_del_dflt_mac_filter(bp, vnic);
2808 memcpy(bp->mac_addr, addr, RTE_ETHER_ADDR_LEN);
2809 if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_VLAN_FILTER) {
2810 /* This filter will allow only untagged packets */
2811 rc = bnxt_add_vlan_filter(bp, 0);
2813 rc = bnxt_add_mac_filter(bp, vnic, addr, 0, 0);
2816 PMD_DRV_LOG(DEBUG, "Set MAC addr\n");
2821 bnxt_dev_set_mc_addr_list_op(struct rte_eth_dev *eth_dev,
2822 struct rte_ether_addr *mc_addr_set,
2823 uint32_t nb_mc_addr)
2825 struct bnxt *bp = eth_dev->data->dev_private;
2826 char *mc_addr_list = (char *)mc_addr_set;
2827 struct bnxt_vnic_info *vnic;
2828 uint32_t off = 0, i = 0;
2831 rc = is_bnxt_in_error(bp);
2835 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2837 if (nb_mc_addr > BNXT_MAX_MC_ADDRS) {
2838 vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
2842 /* TODO Check for Duplicate mcast addresses */
2843 vnic->flags &= ~BNXT_VNIC_INFO_ALLMULTI;
2844 for (i = 0; i < nb_mc_addr; i++) {
2845 memcpy(vnic->mc_list + off, &mc_addr_list[i],
2846 RTE_ETHER_ADDR_LEN);
2847 off += RTE_ETHER_ADDR_LEN;
2850 vnic->mc_addr_cnt = i;
2851 if (vnic->mc_addr_cnt)
2852 vnic->flags |= BNXT_VNIC_INFO_MCAST;
2854 vnic->flags &= ~BNXT_VNIC_INFO_MCAST;
2857 return bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
2861 bnxt_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
2863 struct bnxt *bp = dev->data->dev_private;
2864 uint8_t fw_major = (bp->fw_ver >> 24) & 0xff;
2865 uint8_t fw_minor = (bp->fw_ver >> 16) & 0xff;
2866 uint8_t fw_updt = (bp->fw_ver >> 8) & 0xff;
2867 uint8_t fw_rsvd = bp->fw_ver & 0xff;
2870 ret = snprintf(fw_version, fw_size, "%d.%d.%d.%d",
2871 fw_major, fw_minor, fw_updt, fw_rsvd);
2875 ret += 1; /* add the size of '\0' */
2876 if (fw_size < (size_t)ret)
2883 bnxt_rxq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
2884 struct rte_eth_rxq_info *qinfo)
2886 struct bnxt *bp = dev->data->dev_private;
2887 struct bnxt_rx_queue *rxq;
2889 if (is_bnxt_in_error(bp))
2892 rxq = dev->data->rx_queues[queue_id];
2894 qinfo->mp = rxq->mb_pool;
2895 qinfo->scattered_rx = dev->data->scattered_rx;
2896 qinfo->nb_desc = rxq->nb_rx_desc;
2898 qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
2899 qinfo->conf.rx_drop_en = rxq->drop_en;
2900 qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
2901 qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
2905 bnxt_txq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
2906 struct rte_eth_txq_info *qinfo)
2908 struct bnxt *bp = dev->data->dev_private;
2909 struct bnxt_tx_queue *txq;
2911 if (is_bnxt_in_error(bp))
2914 txq = dev->data->tx_queues[queue_id];
2916 qinfo->nb_desc = txq->nb_tx_desc;
2918 qinfo->conf.tx_thresh.pthresh = txq->pthresh;
2919 qinfo->conf.tx_thresh.hthresh = txq->hthresh;
2920 qinfo->conf.tx_thresh.wthresh = txq->wthresh;
2922 qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
2923 qinfo->conf.tx_rs_thresh = 0;
2924 qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
2925 qinfo->conf.offloads = txq->offloads;
2928 static const struct {
2929 eth_rx_burst_t pkt_burst;
2931 } bnxt_rx_burst_info[] = {
2932 {bnxt_recv_pkts, "Scalar"},
2933 #if defined(RTE_ARCH_X86)
2934 {bnxt_recv_pkts_vec, "Vector SSE"},
2936 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
2937 {bnxt_recv_pkts_vec_avx2, "Vector AVX2"},
2939 #if defined(RTE_ARCH_ARM64)
2940 {bnxt_recv_pkts_vec, "Vector Neon"},
2945 bnxt_rx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
2946 struct rte_eth_burst_mode *mode)
2948 eth_rx_burst_t pkt_burst = dev->rx_pkt_burst;
2951 for (i = 0; i < RTE_DIM(bnxt_rx_burst_info); i++) {
2952 if (pkt_burst == bnxt_rx_burst_info[i].pkt_burst) {
2953 snprintf(mode->info, sizeof(mode->info), "%s",
2954 bnxt_rx_burst_info[i].info);
2962 static const struct {
2963 eth_tx_burst_t pkt_burst;
2965 } bnxt_tx_burst_info[] = {
2966 {bnxt_xmit_pkts, "Scalar"},
2967 #if defined(RTE_ARCH_X86)
2968 {bnxt_xmit_pkts_vec, "Vector SSE"},
2970 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
2971 {bnxt_xmit_pkts_vec_avx2, "Vector AVX2"},
2973 #if defined(RTE_ARCH_ARM64)
2974 {bnxt_xmit_pkts_vec, "Vector Neon"},
2979 bnxt_tx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
2980 struct rte_eth_burst_mode *mode)
2982 eth_tx_burst_t pkt_burst = dev->tx_pkt_burst;
2985 for (i = 0; i < RTE_DIM(bnxt_tx_burst_info); i++) {
2986 if (pkt_burst == bnxt_tx_burst_info[i].pkt_burst) {
2987 snprintf(mode->info, sizeof(mode->info), "%s",
2988 bnxt_tx_burst_info[i].info);
2996 int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
2998 struct bnxt *bp = eth_dev->data->dev_private;
2999 uint32_t new_pkt_size;
3003 rc = is_bnxt_in_error(bp);
3007 /* Exit if receive queues are not configured yet */
3008 if (!eth_dev->data->nb_rx_queues)
3011 new_pkt_size = new_mtu + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN +
3012 VLAN_TAG_SIZE * BNXT_NUM_VLANS;
3015 * Disallow any MTU change that would require scattered receive support
3016 * if it is not already enabled.
3018 if (eth_dev->data->dev_started &&
3019 !eth_dev->data->scattered_rx &&
3021 eth_dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM)) {
3023 "MTU change would require scattered rx support. ");
3024 PMD_DRV_LOG(ERR, "Stop port before changing MTU.\n");
3028 if (new_mtu > RTE_ETHER_MTU) {
3029 bp->flags |= BNXT_FLAG_JUMBO;
3030 bp->eth_dev->data->dev_conf.rxmode.offloads |=
3031 DEV_RX_OFFLOAD_JUMBO_FRAME;
3033 bp->eth_dev->data->dev_conf.rxmode.offloads &=
3034 ~DEV_RX_OFFLOAD_JUMBO_FRAME;
3035 bp->flags &= ~BNXT_FLAG_JUMBO;
3038 /* Is there a change in mtu setting? */
3039 if (eth_dev->data->dev_conf.rxmode.max_rx_pkt_len == new_pkt_size)
3042 for (i = 0; i < bp->nr_vnics; i++) {
3043 struct bnxt_vnic_info *vnic = &bp->vnic_info[i];
3046 vnic->mru = BNXT_VNIC_MRU(new_mtu);
3047 rc = bnxt_hwrm_vnic_cfg(bp, vnic);
3051 size = rte_pktmbuf_data_room_size(bp->rx_queues[0]->mb_pool);
3052 size -= RTE_PKTMBUF_HEADROOM;
3054 if (size < new_mtu) {
3055 rc = bnxt_hwrm_vnic_plcmode_cfg(bp, vnic);
3062 eth_dev->data->dev_conf.rxmode.max_rx_pkt_len = new_pkt_size;
3064 PMD_DRV_LOG(INFO, "New MTU is %d\n", new_mtu);
3070 bnxt_vlan_pvid_set_op(struct rte_eth_dev *dev, uint16_t pvid, int on)
3072 struct bnxt *bp = dev->data->dev_private;
3073 uint16_t vlan = bp->vlan;
3076 rc = is_bnxt_in_error(bp);
3080 if (!BNXT_SINGLE_PF(bp)) {
3081 PMD_DRV_LOG(ERR, "PVID cannot be modified on VF or on shared PF\n");
3084 bp->vlan = on ? pvid : 0;
3086 rc = bnxt_hwrm_set_default_vlan(bp, 0, 0);
3093 bnxt_dev_led_on_op(struct rte_eth_dev *dev)
3095 struct bnxt *bp = dev->data->dev_private;
3098 rc = is_bnxt_in_error(bp);
3102 return bnxt_hwrm_port_led_cfg(bp, true);
3106 bnxt_dev_led_off_op(struct rte_eth_dev *dev)
3108 struct bnxt *bp = dev->data->dev_private;
3111 rc = is_bnxt_in_error(bp);
3115 return bnxt_hwrm_port_led_cfg(bp, false);
3119 bnxt_rx_queue_count_op(struct rte_eth_dev *dev, uint16_t rx_queue_id)
3121 struct bnxt *bp = (struct bnxt *)dev->data->dev_private;
3122 struct bnxt_cp_ring_info *cpr;
3123 uint32_t desc = 0, raw_cons;
3124 struct bnxt_rx_queue *rxq;
3125 struct rx_pkt_cmpl *rxcmp;
3128 rc = is_bnxt_in_error(bp);
3132 rxq = dev->data->rx_queues[rx_queue_id];
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 agg_cnt = BNXT_RX_L2_AGG_BUFS(rxcmp);
3151 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3155 case CMPL_BASE_TYPE_RX_TPA_END:
3156 if (BNXT_CHIP_P5(rxq->bp)) {
3157 struct rx_tpa_v2_end_cmpl_hi *p5_tpa_end;
3159 p5_tpa_end = (void *)rxcmp;
3160 agg_cnt = BNXT_TPA_END_AGG_BUFS_TH(p5_tpa_end);
3162 struct rx_tpa_end_cmpl *tpa_end;
3164 tpa_end = (void *)rxcmp;
3165 agg_cnt = BNXT_TPA_END_AGG_BUFS(tpa_end);
3168 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3173 raw_cons += CMP_LEN(cmpl_type);
3181 bnxt_rx_descriptor_status_op(void *rx_queue, uint16_t offset)
3183 struct bnxt_rx_queue *rxq = rx_queue;
3184 struct bnxt_cp_ring_info *cpr;
3185 struct bnxt_rx_ring_info *rxr;
3186 uint32_t desc, raw_cons;
3187 struct bnxt *bp = rxq->bp;
3188 struct rx_pkt_cmpl *rxcmp;
3191 rc = is_bnxt_in_error(bp);
3195 if (offset >= rxq->nb_rx_desc)
3202 * For the vector receive case, the completion at the requested
3203 * offset can be indexed directly.
3205 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
3206 if (bp->flags & BNXT_FLAG_RX_VECTOR_PKT_MODE) {
3207 struct rx_pkt_cmpl *rxcmp;
3210 /* Check status of completion descriptor. */
3211 raw_cons = cpr->cp_raw_cons +
3212 offset * CMP_LEN(CMPL_BASE_TYPE_RX_L2);
3213 cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3214 rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3216 if (CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct))
3217 return RTE_ETH_RX_DESC_DONE;
3219 /* Check whether rx desc has an mbuf attached. */
3220 cons = RING_CMP(rxr->rx_ring_struct, raw_cons / 2);
3221 if (cons >= rxq->rxrearm_start &&
3222 cons < rxq->rxrearm_start + rxq->rxrearm_nb) {
3223 return RTE_ETH_RX_DESC_UNAVAIL;
3226 return RTE_ETH_RX_DESC_AVAIL;
3231 * For the non-vector receive case, scan the completion ring to
3232 * locate the completion descriptor for the requested offset.
3234 raw_cons = cpr->cp_raw_cons;
3237 uint32_t agg_cnt, cons, cmpl_type;
3239 cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3240 rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3242 if (!CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct))
3245 cmpl_type = CMP_TYPE(rxcmp);
3247 switch (cmpl_type) {
3248 case CMPL_BASE_TYPE_RX_L2:
3249 case CMPL_BASE_TYPE_RX_L2_V2:
3250 if (desc == offset) {
3251 cons = rxcmp->opaque;
3252 if (rxr->rx_buf_ring[cons])
3253 return RTE_ETH_RX_DESC_DONE;
3255 return RTE_ETH_RX_DESC_UNAVAIL;
3257 agg_cnt = BNXT_RX_L2_AGG_BUFS(rxcmp);
3258 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3262 case CMPL_BASE_TYPE_RX_TPA_END:
3264 return RTE_ETH_RX_DESC_DONE;
3266 if (BNXT_CHIP_P5(rxq->bp)) {
3267 struct rx_tpa_v2_end_cmpl_hi *p5_tpa_end;
3269 p5_tpa_end = (void *)rxcmp;
3270 agg_cnt = BNXT_TPA_END_AGG_BUFS_TH(p5_tpa_end);
3272 struct rx_tpa_end_cmpl *tpa_end;
3274 tpa_end = (void *)rxcmp;
3275 agg_cnt = BNXT_TPA_END_AGG_BUFS(tpa_end);
3278 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3283 raw_cons += CMP_LEN(cmpl_type);
3287 return RTE_ETH_RX_DESC_AVAIL;
3291 bnxt_tx_descriptor_status_op(void *tx_queue, uint16_t offset)
3293 struct bnxt_tx_queue *txq = (struct bnxt_tx_queue *)tx_queue;
3294 struct bnxt_tx_ring_info *txr;
3295 struct bnxt_cp_ring_info *cpr;
3296 struct rte_mbuf **tx_buf;
3297 struct tx_pkt_cmpl *txcmp;
3298 uint32_t cons, cp_cons;
3304 rc = is_bnxt_in_error(txq->bp);
3311 if (offset >= txq->nb_tx_desc)
3314 cons = RING_CMP(cpr->cp_ring_struct, offset);
3315 txcmp = (struct tx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3316 cp_cons = cpr->cp_raw_cons;
3318 if (cons > cp_cons) {
3319 if (CMPL_VALID(txcmp, cpr->valid))
3320 return RTE_ETH_TX_DESC_UNAVAIL;
3322 if (CMPL_VALID(txcmp, !cpr->valid))
3323 return RTE_ETH_TX_DESC_UNAVAIL;
3325 tx_buf = &txr->tx_buf_ring[cons];
3326 if (*tx_buf == NULL)
3327 return RTE_ETH_TX_DESC_DONE;
3329 return RTE_ETH_TX_DESC_FULL;
3333 bnxt_flow_ops_get_op(struct rte_eth_dev *dev,
3334 const struct rte_flow_ops **ops)
3336 struct bnxt *bp = dev->data->dev_private;
3342 if (BNXT_ETH_DEV_IS_REPRESENTOR(dev)) {
3343 struct bnxt_representor *vfr = dev->data->dev_private;
3344 bp = vfr->parent_dev->data->dev_private;
3345 /* parent is deleted while children are still valid */
3347 PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR Error\n",
3348 dev->data->port_id);
3353 ret = is_bnxt_in_error(bp);
3357 /* PMD supports thread-safe flow operations. rte_flow API
3358 * functions can avoid mutex for multi-thread safety.
3360 dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
3362 if (BNXT_TRUFLOW_EN(bp))
3363 *ops = &bnxt_ulp_rte_flow_ops;
3365 *ops = &bnxt_flow_ops;
3370 static const uint32_t *
3371 bnxt_dev_supported_ptypes_get_op(struct rte_eth_dev *dev)
3373 static const uint32_t ptypes[] = {
3374 RTE_PTYPE_L2_ETHER_VLAN,
3375 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
3376 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
3380 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
3381 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
3382 RTE_PTYPE_INNER_L4_ICMP,
3383 RTE_PTYPE_INNER_L4_TCP,
3384 RTE_PTYPE_INNER_L4_UDP,
3388 if (!dev->rx_pkt_burst)
3394 static int bnxt_map_regs(struct bnxt *bp, uint32_t *reg_arr, int count,
3397 uint32_t reg_base = *reg_arr & 0xfffff000;
3401 for (i = 0; i < count; i++) {
3402 if ((reg_arr[i] & 0xfffff000) != reg_base)
3405 win_off = BNXT_GRCPF_REG_WINDOW_BASE_OUT + (reg_win - 1) * 4;
3406 rte_write32(reg_base, (uint8_t *)bp->bar0 + win_off);
3410 static int bnxt_map_ptp_regs(struct bnxt *bp)
3412 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3416 reg_arr = ptp->rx_regs;
3417 rc = bnxt_map_regs(bp, reg_arr, BNXT_PTP_RX_REGS, 5);
3421 reg_arr = ptp->tx_regs;
3422 rc = bnxt_map_regs(bp, reg_arr, BNXT_PTP_TX_REGS, 6);
3426 for (i = 0; i < BNXT_PTP_RX_REGS; i++)
3427 ptp->rx_mapped_regs[i] = 0x5000 + (ptp->rx_regs[i] & 0xfff);
3429 for (i = 0; i < BNXT_PTP_TX_REGS; i++)
3430 ptp->tx_mapped_regs[i] = 0x6000 + (ptp->tx_regs[i] & 0xfff);
3435 static void bnxt_unmap_ptp_regs(struct bnxt *bp)
3437 rte_write32(0, (uint8_t *)bp->bar0 +
3438 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 16);
3439 rte_write32(0, (uint8_t *)bp->bar0 +
3440 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 20);
3443 static uint64_t bnxt_cc_read(struct bnxt *bp)
3447 ns = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3448 BNXT_GRCPF_REG_SYNC_TIME));
3449 ns |= (uint64_t)(rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3450 BNXT_GRCPF_REG_SYNC_TIME + 4))) << 32;
3454 static int bnxt_get_tx_ts(struct bnxt *bp, uint64_t *ts)
3456 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3459 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3460 ptp->tx_mapped_regs[BNXT_PTP_TX_FIFO]));
3461 if (fifo & BNXT_PTP_TX_FIFO_EMPTY)
3464 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3465 ptp->tx_mapped_regs[BNXT_PTP_TX_FIFO]));
3466 *ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3467 ptp->tx_mapped_regs[BNXT_PTP_TX_TS_L]));
3468 *ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3469 ptp->tx_mapped_regs[BNXT_PTP_TX_TS_H])) << 32;
3470 rte_read32((uint8_t *)bp->bar0 + ptp->tx_mapped_regs[BNXT_PTP_TX_SEQ]);
3475 static int bnxt_clr_rx_ts(struct bnxt *bp, uint64_t *last_ts)
3477 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3478 struct bnxt_pf_info *pf = bp->pf;
3483 if (!ptp || (bp->flags & BNXT_FLAG_CHIP_P5))
3486 port_id = pf->port_id;
3487 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3488 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3489 while ((fifo & BNXT_PTP_RX_FIFO_PENDING) && (i < BNXT_PTP_RX_PND_CNT)) {
3490 rte_write32(1 << port_id, (uint8_t *)bp->bar0 +
3491 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO_ADV]);
3492 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3493 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3494 *last_ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3495 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_L]));
3496 *last_ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3497 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_H])) << 32;
3501 if (i >= BNXT_PTP_RX_PND_CNT)
3507 static int bnxt_get_rx_ts(struct bnxt *bp, uint64_t *ts)
3509 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3510 struct bnxt_pf_info *pf = bp->pf;
3514 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3515 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3516 if (!(fifo & BNXT_PTP_RX_FIFO_PENDING))
3519 port_id = pf->port_id;
3520 rte_write32(1 << port_id, (uint8_t *)bp->bar0 +
3521 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO_ADV]);
3523 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3524 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3525 if (fifo & BNXT_PTP_RX_FIFO_PENDING)
3526 return bnxt_clr_rx_ts(bp, ts);
3528 *ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3529 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_L]));
3530 *ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3531 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_H])) << 32;
3537 bnxt_timesync_write_time(struct rte_eth_dev *dev, const struct timespec *ts)
3540 struct bnxt *bp = dev->data->dev_private;
3541 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3546 ns = rte_timespec_to_ns(ts);
3547 /* Set the timecounters to a new value. */
3549 ptp->tx_tstamp_tc.nsec = ns;
3550 ptp->rx_tstamp_tc.nsec = ns;
3556 bnxt_timesync_read_time(struct rte_eth_dev *dev, struct timespec *ts)
3558 struct bnxt *bp = dev->data->dev_private;
3559 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3560 uint64_t ns, systime_cycles = 0;
3566 if (BNXT_CHIP_P5(bp))
3567 rc = bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
3570 systime_cycles = bnxt_cc_read(bp);
3572 ns = rte_timecounter_update(&ptp->tc, systime_cycles);
3573 *ts = rte_ns_to_timespec(ns);
3578 bnxt_timesync_enable(struct rte_eth_dev *dev)
3580 struct bnxt *bp = dev->data->dev_private;
3581 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3589 ptp->tx_tstamp_en = 1;
3590 ptp->rxctl = BNXT_PTP_MSG_EVENTS;
3592 rc = bnxt_hwrm_ptp_cfg(bp);
3596 memset(&ptp->tc, 0, sizeof(struct rte_timecounter));
3597 memset(&ptp->rx_tstamp_tc, 0, sizeof(struct rte_timecounter));
3598 memset(&ptp->tx_tstamp_tc, 0, sizeof(struct rte_timecounter));
3600 ptp->tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3601 ptp->tc.cc_shift = shift;
3602 ptp->tc.nsec_mask = (1ULL << shift) - 1;
3604 ptp->rx_tstamp_tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3605 ptp->rx_tstamp_tc.cc_shift = shift;
3606 ptp->rx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
3608 ptp->tx_tstamp_tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3609 ptp->tx_tstamp_tc.cc_shift = shift;
3610 ptp->tx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
3612 if (!BNXT_CHIP_P5(bp))
3613 bnxt_map_ptp_regs(bp);
3615 rc = bnxt_ptp_start(bp);
3621 bnxt_timesync_disable(struct rte_eth_dev *dev)
3623 struct bnxt *bp = dev->data->dev_private;
3624 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3630 ptp->tx_tstamp_en = 0;
3633 bnxt_hwrm_ptp_cfg(bp);
3635 if (!BNXT_CHIP_P5(bp))
3636 bnxt_unmap_ptp_regs(bp);
3644 bnxt_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
3645 struct timespec *timestamp,
3646 uint32_t flags __rte_unused)
3648 struct bnxt *bp = dev->data->dev_private;
3649 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3650 uint64_t rx_tstamp_cycles = 0;
3656 if (BNXT_CHIP_P5(bp))
3657 rx_tstamp_cycles = ptp->rx_timestamp;
3659 bnxt_get_rx_ts(bp, &rx_tstamp_cycles);
3661 ns = rte_timecounter_update(&ptp->rx_tstamp_tc, rx_tstamp_cycles);
3662 *timestamp = rte_ns_to_timespec(ns);
3667 bnxt_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
3668 struct timespec *timestamp)
3670 struct bnxt *bp = dev->data->dev_private;
3671 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3672 uint64_t tx_tstamp_cycles = 0;
3679 if (BNXT_CHIP_P5(bp))
3680 rc = bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_PATH_TX,
3683 rc = bnxt_get_tx_ts(bp, &tx_tstamp_cycles);
3685 ns = rte_timecounter_update(&ptp->tx_tstamp_tc, tx_tstamp_cycles);
3686 *timestamp = rte_ns_to_timespec(ns);
3692 bnxt_timesync_adjust_time(struct rte_eth_dev *dev, int64_t delta)
3694 struct bnxt *bp = dev->data->dev_private;
3695 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3700 ptp->tc.nsec += delta;
3701 ptp->tx_tstamp_tc.nsec += delta;
3702 ptp->rx_tstamp_tc.nsec += delta;
3708 bnxt_get_eeprom_length_op(struct rte_eth_dev *dev)
3710 struct bnxt *bp = dev->data->dev_private;
3712 uint32_t dir_entries;
3713 uint32_t entry_length;
3715 rc = is_bnxt_in_error(bp);
3719 PMD_DRV_LOG(INFO, PCI_PRI_FMT "\n",
3720 bp->pdev->addr.domain, bp->pdev->addr.bus,
3721 bp->pdev->addr.devid, bp->pdev->addr.function);
3723 rc = bnxt_hwrm_nvm_get_dir_info(bp, &dir_entries, &entry_length);
3727 return dir_entries * entry_length;
3731 bnxt_get_eeprom_op(struct rte_eth_dev *dev,
3732 struct rte_dev_eeprom_info *in_eeprom)
3734 struct bnxt *bp = dev->data->dev_private;
3739 rc = is_bnxt_in_error(bp);
3743 PMD_DRV_LOG(INFO, PCI_PRI_FMT " in_eeprom->offset = %d len = %d\n",
3744 bp->pdev->addr.domain, bp->pdev->addr.bus,
3745 bp->pdev->addr.devid, bp->pdev->addr.function,
3746 in_eeprom->offset, in_eeprom->length);
3748 if (in_eeprom->offset == 0) /* special offset value to get directory */
3749 return bnxt_get_nvram_directory(bp, in_eeprom->length,
3752 index = in_eeprom->offset >> 24;
3753 offset = in_eeprom->offset & 0xffffff;
3756 return bnxt_hwrm_get_nvram_item(bp, index - 1, offset,
3757 in_eeprom->length, in_eeprom->data);
3762 static bool bnxt_dir_type_is_ape_bin_format(uint16_t dir_type)
3765 case BNX_DIR_TYPE_CHIMP_PATCH:
3766 case BNX_DIR_TYPE_BOOTCODE:
3767 case BNX_DIR_TYPE_BOOTCODE_2:
3768 case BNX_DIR_TYPE_APE_FW:
3769 case BNX_DIR_TYPE_APE_PATCH:
3770 case BNX_DIR_TYPE_KONG_FW:
3771 case BNX_DIR_TYPE_KONG_PATCH:
3772 case BNX_DIR_TYPE_BONO_FW:
3773 case BNX_DIR_TYPE_BONO_PATCH:
3781 static bool bnxt_dir_type_is_other_exec_format(uint16_t dir_type)
3784 case BNX_DIR_TYPE_AVS:
3785 case BNX_DIR_TYPE_EXP_ROM_MBA:
3786 case BNX_DIR_TYPE_PCIE:
3787 case BNX_DIR_TYPE_TSCF_UCODE:
3788 case BNX_DIR_TYPE_EXT_PHY:
3789 case BNX_DIR_TYPE_CCM:
3790 case BNX_DIR_TYPE_ISCSI_BOOT:
3791 case BNX_DIR_TYPE_ISCSI_BOOT_IPV6:
3792 case BNX_DIR_TYPE_ISCSI_BOOT_IPV4N6:
3800 static bool bnxt_dir_type_is_executable(uint16_t dir_type)
3802 return bnxt_dir_type_is_ape_bin_format(dir_type) ||
3803 bnxt_dir_type_is_other_exec_format(dir_type);
3807 bnxt_set_eeprom_op(struct rte_eth_dev *dev,
3808 struct rte_dev_eeprom_info *in_eeprom)
3810 struct bnxt *bp = dev->data->dev_private;
3811 uint8_t index, dir_op;
3812 uint16_t type, ext, ordinal, attr;
3815 rc = is_bnxt_in_error(bp);
3819 PMD_DRV_LOG(INFO, PCI_PRI_FMT " in_eeprom->offset = %d len = %d\n",
3820 bp->pdev->addr.domain, bp->pdev->addr.bus,
3821 bp->pdev->addr.devid, bp->pdev->addr.function,
3822 in_eeprom->offset, in_eeprom->length);
3825 PMD_DRV_LOG(ERR, "NVM write not supported from a VF\n");
3829 type = in_eeprom->magic >> 16;
3831 if (type == 0xffff) { /* special value for directory operations */
3832 index = in_eeprom->magic & 0xff;
3833 dir_op = in_eeprom->magic >> 8;
3837 case 0x0e: /* erase */
3838 if (in_eeprom->offset != ~in_eeprom->magic)
3840 return bnxt_hwrm_erase_nvram_directory(bp, index - 1);
3846 /* Create or re-write an NVM item: */
3847 if (bnxt_dir_type_is_executable(type) == true)
3849 ext = in_eeprom->magic & 0xffff;
3850 ordinal = in_eeprom->offset >> 16;
3851 attr = in_eeprom->offset & 0xffff;
3853 return bnxt_hwrm_flash_nvram(bp, type, ordinal, ext, attr,
3854 in_eeprom->data, in_eeprom->length);
3861 static const struct eth_dev_ops bnxt_dev_ops = {
3862 .dev_infos_get = bnxt_dev_info_get_op,
3863 .dev_close = bnxt_dev_close_op,
3864 .dev_configure = bnxt_dev_configure_op,
3865 .dev_start = bnxt_dev_start_op,
3866 .dev_stop = bnxt_dev_stop_op,
3867 .dev_set_link_up = bnxt_dev_set_link_up_op,
3868 .dev_set_link_down = bnxt_dev_set_link_down_op,
3869 .stats_get = bnxt_stats_get_op,
3870 .stats_reset = bnxt_stats_reset_op,
3871 .rx_queue_setup = bnxt_rx_queue_setup_op,
3872 .rx_queue_release = bnxt_rx_queue_release_op,
3873 .tx_queue_setup = bnxt_tx_queue_setup_op,
3874 .tx_queue_release = bnxt_tx_queue_release_op,
3875 .rx_queue_intr_enable = bnxt_rx_queue_intr_enable_op,
3876 .rx_queue_intr_disable = bnxt_rx_queue_intr_disable_op,
3877 .reta_update = bnxt_reta_update_op,
3878 .reta_query = bnxt_reta_query_op,
3879 .rss_hash_update = bnxt_rss_hash_update_op,
3880 .rss_hash_conf_get = bnxt_rss_hash_conf_get_op,
3881 .link_update = bnxt_link_update_op,
3882 .promiscuous_enable = bnxt_promiscuous_enable_op,
3883 .promiscuous_disable = bnxt_promiscuous_disable_op,
3884 .allmulticast_enable = bnxt_allmulticast_enable_op,
3885 .allmulticast_disable = bnxt_allmulticast_disable_op,
3886 .mac_addr_add = bnxt_mac_addr_add_op,
3887 .mac_addr_remove = bnxt_mac_addr_remove_op,
3888 .flow_ctrl_get = bnxt_flow_ctrl_get_op,
3889 .flow_ctrl_set = bnxt_flow_ctrl_set_op,
3890 .udp_tunnel_port_add = bnxt_udp_tunnel_port_add_op,
3891 .udp_tunnel_port_del = bnxt_udp_tunnel_port_del_op,
3892 .vlan_filter_set = bnxt_vlan_filter_set_op,
3893 .vlan_offload_set = bnxt_vlan_offload_set_op,
3894 .vlan_tpid_set = bnxt_vlan_tpid_set_op,
3895 .vlan_pvid_set = bnxt_vlan_pvid_set_op,
3896 .mtu_set = bnxt_mtu_set_op,
3897 .mac_addr_set = bnxt_set_default_mac_addr_op,
3898 .xstats_get = bnxt_dev_xstats_get_op,
3899 .xstats_get_names = bnxt_dev_xstats_get_names_op,
3900 .xstats_reset = bnxt_dev_xstats_reset_op,
3901 .fw_version_get = bnxt_fw_version_get,
3902 .set_mc_addr_list = bnxt_dev_set_mc_addr_list_op,
3903 .rxq_info_get = bnxt_rxq_info_get_op,
3904 .txq_info_get = bnxt_txq_info_get_op,
3905 .rx_burst_mode_get = bnxt_rx_burst_mode_get,
3906 .tx_burst_mode_get = bnxt_tx_burst_mode_get,
3907 .dev_led_on = bnxt_dev_led_on_op,
3908 .dev_led_off = bnxt_dev_led_off_op,
3909 .rx_queue_start = bnxt_rx_queue_start,
3910 .rx_queue_stop = bnxt_rx_queue_stop,
3911 .tx_queue_start = bnxt_tx_queue_start,
3912 .tx_queue_stop = bnxt_tx_queue_stop,
3913 .flow_ops_get = bnxt_flow_ops_get_op,
3914 .dev_supported_ptypes_get = bnxt_dev_supported_ptypes_get_op,
3915 .get_eeprom_length = bnxt_get_eeprom_length_op,
3916 .get_eeprom = bnxt_get_eeprom_op,
3917 .set_eeprom = bnxt_set_eeprom_op,
3918 .timesync_enable = bnxt_timesync_enable,
3919 .timesync_disable = bnxt_timesync_disable,
3920 .timesync_read_time = bnxt_timesync_read_time,
3921 .timesync_write_time = bnxt_timesync_write_time,
3922 .timesync_adjust_time = bnxt_timesync_adjust_time,
3923 .timesync_read_rx_timestamp = bnxt_timesync_read_rx_timestamp,
3924 .timesync_read_tx_timestamp = bnxt_timesync_read_tx_timestamp,
3927 static uint32_t bnxt_map_reset_regs(struct bnxt *bp, uint32_t reg)
3931 /* Only pre-map the reset GRC registers using window 3 */
3932 rte_write32(reg & 0xfffff000, (uint8_t *)bp->bar0 +
3933 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 8);
3935 offset = BNXT_GRCP_WINDOW_3_BASE + (reg & 0xffc);
3940 int bnxt_map_fw_health_status_regs(struct bnxt *bp)
3942 struct bnxt_error_recovery_info *info = bp->recovery_info;
3943 uint32_t reg_base = 0xffffffff;
3946 /* Only pre-map the monitoring GRC registers using window 2 */
3947 for (i = 0; i < BNXT_FW_STATUS_REG_CNT; i++) {
3948 uint32_t reg = info->status_regs[i];
3950 if (BNXT_FW_STATUS_REG_TYPE(reg) != BNXT_FW_STATUS_REG_TYPE_GRC)
3953 if (reg_base == 0xffffffff)
3954 reg_base = reg & 0xfffff000;
3955 if ((reg & 0xfffff000) != reg_base)
3958 /* Use mask 0xffc as the Lower 2 bits indicates
3959 * address space location
3961 info->mapped_status_regs[i] = BNXT_GRCP_WINDOW_2_BASE +
3965 if (reg_base == 0xffffffff)
3968 rte_write32(reg_base, (uint8_t *)bp->bar0 +
3969 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
3974 static void bnxt_write_fw_reset_reg(struct bnxt *bp, uint32_t index)
3976 struct bnxt_error_recovery_info *info = bp->recovery_info;
3977 uint32_t delay = info->delay_after_reset[index];
3978 uint32_t val = info->reset_reg_val[index];
3979 uint32_t reg = info->reset_reg[index];
3980 uint32_t type, offset;
3983 type = BNXT_FW_STATUS_REG_TYPE(reg);
3984 offset = BNXT_FW_STATUS_REG_OFF(reg);
3987 case BNXT_FW_STATUS_REG_TYPE_CFG:
3988 ret = rte_pci_write_config(bp->pdev, &val, sizeof(val), offset);
3990 PMD_DRV_LOG(ERR, "Failed to write %#x at PCI offset %#x",
3995 case BNXT_FW_STATUS_REG_TYPE_GRC:
3996 offset = bnxt_map_reset_regs(bp, offset);
3997 rte_write32(val, (uint8_t *)bp->bar0 + offset);
3999 case BNXT_FW_STATUS_REG_TYPE_BAR0:
4000 rte_write32(val, (uint8_t *)bp->bar0 + offset);
4003 /* wait on a specific interval of time until core reset is complete */
4005 rte_delay_ms(delay);
4008 static void bnxt_dev_cleanup(struct bnxt *bp)
4010 bp->eth_dev->data->dev_link.link_status = 0;
4011 bp->link_info->link_up = 0;
4012 if (bp->eth_dev->data->dev_started)
4013 bnxt_dev_stop(bp->eth_dev);
4015 bnxt_uninit_resources(bp, true);
4019 bnxt_check_fw_reset_done(struct bnxt *bp)
4021 int timeout = bp->fw_reset_max_msecs;
4026 rc = rte_pci_read_config(bp->pdev, &val, sizeof(val), PCI_SUBSYSTEM_ID_OFFSET);
4028 PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x", PCI_SUBSYSTEM_ID_OFFSET);
4034 } while (timeout--);
4036 if (val == 0xffff) {
4037 PMD_DRV_LOG(ERR, "Firmware reset aborted, PCI config space invalid\n");
4044 static int bnxt_restore_vlan_filters(struct bnxt *bp)
4046 struct rte_eth_dev *dev = bp->eth_dev;
4047 struct rte_vlan_filter_conf *vfc;
4051 for (vlan_id = 1; vlan_id <= RTE_ETHER_MAX_VLAN_ID; vlan_id++) {
4052 vfc = &dev->data->vlan_filter_conf;
4053 vidx = vlan_id / 64;
4054 vbit = vlan_id % 64;
4056 /* Each bit corresponds to a VLAN id */
4057 if (vfc->ids[vidx] & (UINT64_C(1) << vbit)) {
4058 rc = bnxt_add_vlan_filter(bp, vlan_id);
4067 static int bnxt_restore_mac_filters(struct bnxt *bp)
4069 struct rte_eth_dev *dev = bp->eth_dev;
4070 struct rte_eth_dev_info dev_info;
4071 struct rte_ether_addr *addr;
4077 if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp))
4080 rc = bnxt_dev_info_get_op(dev, &dev_info);
4084 /* replay MAC address configuration */
4085 for (i = 1; i < dev_info.max_mac_addrs; i++) {
4086 addr = &dev->data->mac_addrs[i];
4088 /* skip zero address */
4089 if (rte_is_zero_ether_addr(addr))
4093 pool_mask = dev->data->mac_pool_sel[i];
4096 if (pool_mask & 1ULL) {
4097 rc = bnxt_mac_addr_add_op(dev, addr, i, pool);
4103 } while (pool_mask);
4109 static int bnxt_restore_filters(struct bnxt *bp)
4111 struct rte_eth_dev *dev = bp->eth_dev;
4114 if (dev->data->all_multicast) {
4115 ret = bnxt_allmulticast_enable_op(dev);
4119 if (dev->data->promiscuous) {
4120 ret = bnxt_promiscuous_enable_op(dev);
4125 ret = bnxt_restore_mac_filters(bp);
4129 ret = bnxt_restore_vlan_filters(bp);
4130 /* TODO restore other filters as well */
4134 static int bnxt_check_fw_ready(struct bnxt *bp)
4136 int timeout = bp->fw_reset_max_msecs;
4140 rc = bnxt_hwrm_poll_ver_get(bp);
4143 rte_delay_ms(BNXT_FW_READY_WAIT_INTERVAL);
4144 timeout -= BNXT_FW_READY_WAIT_INTERVAL;
4145 } while (rc && timeout > 0);
4148 PMD_DRV_LOG(ERR, "FW is not Ready after reset\n");
4153 static void bnxt_dev_recover(void *arg)
4155 struct bnxt *bp = arg;
4158 pthread_mutex_lock(&bp->err_recovery_lock);
4160 if (!bp->fw_reset_min_msecs) {
4161 rc = bnxt_check_fw_reset_done(bp);
4166 /* Clear Error flag so that device re-init should happen */
4167 bp->flags &= ~BNXT_FLAG_FATAL_ERROR;
4169 rc = bnxt_check_fw_ready(bp);
4173 rc = bnxt_init_resources(bp, true);
4176 "Failed to initialize resources after reset\n");
4179 /* clear reset flag as the device is initialized now */
4180 bp->flags &= ~BNXT_FLAG_FW_RESET;
4182 rc = bnxt_dev_start_op(bp->eth_dev);
4184 PMD_DRV_LOG(ERR, "Failed to start port after reset\n");
4188 rc = bnxt_restore_filters(bp);
4192 PMD_DRV_LOG(INFO, "Recovered from FW reset\n");
4193 pthread_mutex_unlock(&bp->err_recovery_lock);
4197 bnxt_dev_stop(bp->eth_dev);
4199 bp->flags |= BNXT_FLAG_FATAL_ERROR;
4200 bnxt_uninit_resources(bp, false);
4201 pthread_mutex_unlock(&bp->err_recovery_lock);
4202 PMD_DRV_LOG(ERR, "Failed to recover from FW reset\n");
4205 void bnxt_dev_reset_and_resume(void *arg)
4207 struct bnxt *bp = arg;
4208 uint32_t us = US_PER_MS * bp->fw_reset_min_msecs;
4212 bnxt_dev_cleanup(bp);
4214 bnxt_wait_for_device_shutdown(bp);
4216 /* During some fatal firmware error conditions, the PCI config space
4217 * register 0x2e which normally contains the subsystem ID will become
4218 * 0xffff. This register will revert back to the normal value after
4219 * the chip has completed core reset. If we detect this condition,
4220 * we can poll this config register immediately for the value to revert.
4222 if (bp->flags & BNXT_FLAG_FATAL_ERROR) {
4223 rc = rte_pci_read_config(bp->pdev, &val, sizeof(val), PCI_SUBSYSTEM_ID_OFFSET);
4225 PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x", PCI_SUBSYSTEM_ID_OFFSET);
4228 if (val == 0xffff) {
4229 bp->fw_reset_min_msecs = 0;
4234 rc = rte_eal_alarm_set(us, bnxt_dev_recover, (void *)bp);
4236 PMD_DRV_LOG(ERR, "Error setting recovery alarm");
4239 uint32_t bnxt_read_fw_status_reg(struct bnxt *bp, uint32_t index)
4241 struct bnxt_error_recovery_info *info = bp->recovery_info;
4242 uint32_t reg = info->status_regs[index];
4243 uint32_t type, offset, val = 0;
4246 type = BNXT_FW_STATUS_REG_TYPE(reg);
4247 offset = BNXT_FW_STATUS_REG_OFF(reg);
4250 case BNXT_FW_STATUS_REG_TYPE_CFG:
4251 ret = rte_pci_read_config(bp->pdev, &val, sizeof(val), offset);
4253 PMD_DRV_LOG(ERR, "Failed to read PCI offset %#x",
4256 case BNXT_FW_STATUS_REG_TYPE_GRC:
4257 offset = info->mapped_status_regs[index];
4259 case BNXT_FW_STATUS_REG_TYPE_BAR0:
4260 val = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
4268 static int bnxt_fw_reset_all(struct bnxt *bp)
4270 struct bnxt_error_recovery_info *info = bp->recovery_info;
4274 if (info->flags & BNXT_FLAG_ERROR_RECOVERY_HOST) {
4275 /* Reset through master function driver */
4276 for (i = 0; i < info->reg_array_cnt; i++)
4277 bnxt_write_fw_reset_reg(bp, i);
4278 /* Wait for time specified by FW after triggering reset */
4279 rte_delay_ms(info->master_func_wait_period_after_reset);
4280 } else if (info->flags & BNXT_FLAG_ERROR_RECOVERY_CO_CPU) {
4281 /* Reset with the help of Kong processor */
4282 rc = bnxt_hwrm_fw_reset(bp);
4284 PMD_DRV_LOG(ERR, "Failed to reset FW\n");
4290 static void bnxt_fw_reset_cb(void *arg)
4292 struct bnxt *bp = arg;
4293 struct bnxt_error_recovery_info *info = bp->recovery_info;
4296 /* Only Master function can do FW reset */
4297 if (bnxt_is_master_func(bp) &&
4298 bnxt_is_recovery_enabled(bp)) {
4299 rc = bnxt_fw_reset_all(bp);
4301 PMD_DRV_LOG(ERR, "Adapter recovery failed\n");
4306 /* if recovery method is ERROR_RECOVERY_CO_CPU, KONG will send
4307 * EXCEPTION_FATAL_ASYNC event to all the functions
4308 * (including MASTER FUNC). After receiving this Async, all the active
4309 * drivers should treat this case as FW initiated recovery
4311 if (info->flags & BNXT_FLAG_ERROR_RECOVERY_HOST) {
4312 bp->fw_reset_min_msecs = BNXT_MIN_FW_READY_TIMEOUT;
4313 bp->fw_reset_max_msecs = BNXT_MAX_FW_RESET_TIMEOUT;
4315 /* To recover from error */
4316 rte_eal_alarm_set(US_PER_MS, bnxt_dev_reset_and_resume,
4321 /* Driver should poll FW heartbeat, reset_counter with the frequency
4322 * advertised by FW in HWRM_ERROR_RECOVERY_QCFG.
4323 * When the driver detects heartbeat stop or change in reset_counter,
4324 * it has to trigger a reset to recover from the error condition.
4325 * A “master PF” is the function who will have the privilege to
4326 * initiate the chimp reset. The master PF will be elected by the
4327 * firmware and will be notified through async message.
4329 static void bnxt_check_fw_health(void *arg)
4331 struct bnxt *bp = arg;
4332 struct bnxt_error_recovery_info *info = bp->recovery_info;
4333 uint32_t val = 0, wait_msec;
4335 if (!info || !bnxt_is_recovery_enabled(bp) ||
4336 is_bnxt_in_error(bp))
4339 val = bnxt_read_fw_status_reg(bp, BNXT_FW_HEARTBEAT_CNT_REG);
4340 if (val == info->last_heart_beat)
4343 info->last_heart_beat = val;
4345 val = bnxt_read_fw_status_reg(bp, BNXT_FW_RECOVERY_CNT_REG);
4346 if (val != info->last_reset_counter)
4349 info->last_reset_counter = val;
4351 rte_eal_alarm_set(US_PER_MS * info->driver_polling_freq,
4352 bnxt_check_fw_health, (void *)bp);
4356 /* Stop DMA to/from device */
4357 bp->flags |= BNXT_FLAG_FATAL_ERROR;
4358 bp->flags |= BNXT_FLAG_FW_RESET;
4362 PMD_DRV_LOG(ERR, "Detected FW dead condition\n");
4364 if (bnxt_is_master_func(bp))
4365 wait_msec = info->master_func_wait_period;
4367 wait_msec = info->normal_func_wait_period;
4369 rte_eal_alarm_set(US_PER_MS * wait_msec,
4370 bnxt_fw_reset_cb, (void *)bp);
4373 void bnxt_schedule_fw_health_check(struct bnxt *bp)
4375 uint32_t polling_freq;
4377 pthread_mutex_lock(&bp->health_check_lock);
4379 if (!bnxt_is_recovery_enabled(bp))
4382 if (bp->flags & BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED)
4385 polling_freq = bp->recovery_info->driver_polling_freq;
4387 rte_eal_alarm_set(US_PER_MS * polling_freq,
4388 bnxt_check_fw_health, (void *)bp);
4389 bp->flags |= BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED;
4392 pthread_mutex_unlock(&bp->health_check_lock);
4395 static void bnxt_cancel_fw_health_check(struct bnxt *bp)
4397 rte_eal_alarm_cancel(bnxt_check_fw_health, (void *)bp);
4398 bp->flags &= ~BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED;
4401 static bool bnxt_vf_pciid(uint16_t device_id)
4403 switch (device_id) {
4404 case BROADCOM_DEV_ID_57304_VF:
4405 case BROADCOM_DEV_ID_57406_VF:
4406 case BROADCOM_DEV_ID_5731X_VF:
4407 case BROADCOM_DEV_ID_5741X_VF:
4408 case BROADCOM_DEV_ID_57414_VF:
4409 case BROADCOM_DEV_ID_STRATUS_NIC_VF1:
4410 case BROADCOM_DEV_ID_STRATUS_NIC_VF2:
4411 case BROADCOM_DEV_ID_58802_VF:
4412 case BROADCOM_DEV_ID_57500_VF1:
4413 case BROADCOM_DEV_ID_57500_VF2:
4414 case BROADCOM_DEV_ID_58818_VF:
4422 /* Phase 5 device */
4423 static bool bnxt_p5_device(uint16_t device_id)
4425 switch (device_id) {
4426 case BROADCOM_DEV_ID_57508:
4427 case BROADCOM_DEV_ID_57504:
4428 case BROADCOM_DEV_ID_57502:
4429 case BROADCOM_DEV_ID_57508_MF1:
4430 case BROADCOM_DEV_ID_57504_MF1:
4431 case BROADCOM_DEV_ID_57502_MF1:
4432 case BROADCOM_DEV_ID_57508_MF2:
4433 case BROADCOM_DEV_ID_57504_MF2:
4434 case BROADCOM_DEV_ID_57502_MF2:
4435 case BROADCOM_DEV_ID_57500_VF1:
4436 case BROADCOM_DEV_ID_57500_VF2:
4437 case BROADCOM_DEV_ID_58812:
4438 case BROADCOM_DEV_ID_58814:
4439 case BROADCOM_DEV_ID_58818:
4440 case BROADCOM_DEV_ID_58818_VF:
4448 bool bnxt_stratus_device(struct bnxt *bp)
4450 uint16_t device_id = bp->pdev->id.device_id;
4452 switch (device_id) {
4453 case BROADCOM_DEV_ID_STRATUS_NIC:
4454 case BROADCOM_DEV_ID_STRATUS_NIC_VF1:
4455 case BROADCOM_DEV_ID_STRATUS_NIC_VF2:
4463 static int bnxt_map_pci_bars(struct rte_eth_dev *eth_dev)
4465 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
4466 struct bnxt *bp = eth_dev->data->dev_private;
4468 /* enable device (incl. PCI PM wakeup), and bus-mastering */
4469 bp->bar0 = (void *)pci_dev->mem_resource[0].addr;
4470 bp->doorbell_base = (void *)pci_dev->mem_resource[2].addr;
4471 if (!bp->bar0 || !bp->doorbell_base) {
4472 PMD_DRV_LOG(ERR, "Unable to access Hardware\n");
4476 bp->eth_dev = eth_dev;
4482 static int bnxt_alloc_ctx_mem_blk(struct bnxt *bp,
4483 struct bnxt_ctx_pg_info *ctx_pg,
4488 struct bnxt_ring_mem_info *rmem = &ctx_pg->ring_mem;
4489 const struct rte_memzone *mz = NULL;
4490 char mz_name[RTE_MEMZONE_NAMESIZE];
4491 rte_iova_t mz_phys_addr;
4492 uint64_t valid_bits = 0;
4499 rmem->nr_pages = RTE_ALIGN_MUL_CEIL(mem_size, BNXT_PAGE_SIZE) /
4501 rmem->page_size = BNXT_PAGE_SIZE;
4502 rmem->pg_arr = ctx_pg->ctx_pg_arr;
4503 rmem->dma_arr = ctx_pg->ctx_dma_arr;
4504 rmem->flags = BNXT_RMEM_VALID_PTE_FLAG;
4506 valid_bits = PTU_PTE_VALID;
4508 if (rmem->nr_pages > 1) {
4509 snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4510 "bnxt_ctx_pg_tbl%s_%x_%d",
4511 suffix, idx, bp->eth_dev->data->port_id);
4512 mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4513 mz = rte_memzone_lookup(mz_name);
4515 mz = rte_memzone_reserve_aligned(mz_name,
4519 RTE_MEMZONE_SIZE_HINT_ONLY |
4520 RTE_MEMZONE_IOVA_CONTIG,
4526 memset(mz->addr, 0, mz->len);
4527 mz_phys_addr = mz->iova;
4529 rmem->pg_tbl = mz->addr;
4530 rmem->pg_tbl_map = mz_phys_addr;
4531 rmem->pg_tbl_mz = mz;
4534 snprintf(mz_name, RTE_MEMZONE_NAMESIZE, "bnxt_ctx_%s_%x_%d",
4535 suffix, idx, bp->eth_dev->data->port_id);
4536 mz = rte_memzone_lookup(mz_name);
4538 mz = rte_memzone_reserve_aligned(mz_name,
4542 RTE_MEMZONE_SIZE_HINT_ONLY |
4543 RTE_MEMZONE_IOVA_CONTIG,
4549 memset(mz->addr, 0, mz->len);
4550 mz_phys_addr = mz->iova;
4552 for (sz = 0, i = 0; sz < mem_size; sz += BNXT_PAGE_SIZE, i++) {
4553 rmem->pg_arr[i] = ((char *)mz->addr) + sz;
4554 rmem->dma_arr[i] = mz_phys_addr + sz;
4556 if (rmem->nr_pages > 1) {
4557 if (i == rmem->nr_pages - 2 &&
4558 (rmem->flags & BNXT_RMEM_RING_PTE_FLAG))
4559 valid_bits |= PTU_PTE_NEXT_TO_LAST;
4560 else if (i == rmem->nr_pages - 1 &&
4561 (rmem->flags & BNXT_RMEM_RING_PTE_FLAG))
4562 valid_bits |= PTU_PTE_LAST;
4564 rmem->pg_tbl[i] = rte_cpu_to_le_64(rmem->dma_arr[i] |
4570 if (rmem->vmem_size)
4571 rmem->vmem = (void **)mz->addr;
4572 rmem->dma_arr[0] = mz_phys_addr;
4576 static void bnxt_free_ctx_mem(struct bnxt *bp)
4580 if (!bp->ctx || !(bp->ctx->flags & BNXT_CTX_FLAG_INITED))
4583 bp->ctx->flags &= ~BNXT_CTX_FLAG_INITED;
4584 rte_memzone_free(bp->ctx->qp_mem.ring_mem.mz);
4585 rte_memzone_free(bp->ctx->srq_mem.ring_mem.mz);
4586 rte_memzone_free(bp->ctx->cq_mem.ring_mem.mz);
4587 rte_memzone_free(bp->ctx->vnic_mem.ring_mem.mz);
4588 rte_memzone_free(bp->ctx->stat_mem.ring_mem.mz);
4589 rte_memzone_free(bp->ctx->qp_mem.ring_mem.pg_tbl_mz);
4590 rte_memzone_free(bp->ctx->srq_mem.ring_mem.pg_tbl_mz);
4591 rte_memzone_free(bp->ctx->cq_mem.ring_mem.pg_tbl_mz);
4592 rte_memzone_free(bp->ctx->vnic_mem.ring_mem.pg_tbl_mz);
4593 rte_memzone_free(bp->ctx->stat_mem.ring_mem.pg_tbl_mz);
4595 for (i = 0; i < bp->ctx->tqm_fp_rings_count + 1; i++) {
4596 if (bp->ctx->tqm_mem[i])
4597 rte_memzone_free(bp->ctx->tqm_mem[i]->ring_mem.mz);
4604 #define bnxt_roundup(x, y) ((((x) + ((y) - 1)) / (y)) * (y))
4606 #define min_t(type, x, y) ({ \
4607 type __min1 = (x); \
4608 type __min2 = (y); \
4609 __min1 < __min2 ? __min1 : __min2; })
4611 #define max_t(type, x, y) ({ \
4612 type __max1 = (x); \
4613 type __max2 = (y); \
4614 __max1 > __max2 ? __max1 : __max2; })
4616 #define clamp_t(type, _x, min, max) min_t(type, max_t(type, _x, min), max)
4618 int bnxt_alloc_ctx_mem(struct bnxt *bp)
4620 struct bnxt_ctx_pg_info *ctx_pg;
4621 struct bnxt_ctx_mem_info *ctx;
4622 uint32_t mem_size, ena, entries;
4623 uint32_t entries_sp, min;
4626 rc = bnxt_hwrm_func_backing_store_qcaps(bp);
4628 PMD_DRV_LOG(ERR, "Query context mem capability failed\n");
4632 if (!ctx || (ctx->flags & BNXT_CTX_FLAG_INITED))
4635 ctx_pg = &ctx->qp_mem;
4636 ctx_pg->entries = ctx->qp_min_qp1_entries + ctx->qp_max_l2_entries;
4637 if (ctx->qp_entry_size) {
4638 mem_size = ctx->qp_entry_size * ctx_pg->entries;
4639 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "qp_mem", 0);
4644 ctx_pg = &ctx->srq_mem;
4645 ctx_pg->entries = ctx->srq_max_l2_entries;
4646 if (ctx->srq_entry_size) {
4647 mem_size = ctx->srq_entry_size * ctx_pg->entries;
4648 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "srq_mem", 0);
4653 ctx_pg = &ctx->cq_mem;
4654 ctx_pg->entries = ctx->cq_max_l2_entries;
4655 if (ctx->cq_entry_size) {
4656 mem_size = ctx->cq_entry_size * ctx_pg->entries;
4657 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "cq_mem", 0);
4662 ctx_pg = &ctx->vnic_mem;
4663 ctx_pg->entries = ctx->vnic_max_vnic_entries +
4664 ctx->vnic_max_ring_table_entries;
4665 if (ctx->vnic_entry_size) {
4666 mem_size = ctx->vnic_entry_size * ctx_pg->entries;
4667 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "vnic_mem", 0);
4672 ctx_pg = &ctx->stat_mem;
4673 ctx_pg->entries = ctx->stat_max_entries;
4674 if (ctx->stat_entry_size) {
4675 mem_size = ctx->stat_entry_size * ctx_pg->entries;
4676 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "stat_mem", 0);
4681 min = ctx->tqm_min_entries_per_ring;
4683 entries_sp = ctx->qp_max_l2_entries +
4684 ctx->vnic_max_vnic_entries +
4685 2 * ctx->qp_min_qp1_entries + min;
4686 entries_sp = bnxt_roundup(entries_sp, ctx->tqm_entries_multiple);
4688 entries = ctx->qp_max_l2_entries + ctx->qp_min_qp1_entries;
4689 entries = bnxt_roundup(entries, ctx->tqm_entries_multiple);
4690 entries = clamp_t(uint32_t, entries, min,
4691 ctx->tqm_max_entries_per_ring);
4692 for (i = 0, ena = 0; i < ctx->tqm_fp_rings_count + 1; i++) {
4693 /* i=0 is for TQM_SP. i=1 to i=8 applies to RING0 to RING7.
4694 * i > 8 is other ext rings.
4696 ctx_pg = ctx->tqm_mem[i];
4697 ctx_pg->entries = i ? entries : entries_sp;
4698 if (ctx->tqm_entry_size) {
4699 mem_size = ctx->tqm_entry_size * ctx_pg->entries;
4700 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size,
4705 if (i < BNXT_MAX_TQM_LEGACY_RINGS)
4706 ena |= HWRM_FUNC_BACKING_STORE_CFG_INPUT_ENABLES_TQM_SP << i;
4708 ena |= HWRM_FUNC_BACKING_STORE_CFG_INPUT_ENABLES_TQM_RING8;
4711 ena |= FUNC_BACKING_STORE_CFG_INPUT_DFLT_ENABLES;
4712 rc = bnxt_hwrm_func_backing_store_cfg(bp, ena);
4715 "Failed to configure context mem: rc = %d\n", rc);
4717 ctx->flags |= BNXT_CTX_FLAG_INITED;
4722 static int bnxt_alloc_stats_mem(struct bnxt *bp)
4724 struct rte_pci_device *pci_dev = bp->pdev;
4725 char mz_name[RTE_MEMZONE_NAMESIZE];
4726 const struct rte_memzone *mz = NULL;
4727 uint32_t total_alloc_len;
4728 rte_iova_t mz_phys_addr;
4730 if (pci_dev->id.device_id == BROADCOM_DEV_ID_NS2)
4733 snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4734 "bnxt_" PCI_PRI_FMT "-%s", pci_dev->addr.domain,
4735 pci_dev->addr.bus, pci_dev->addr.devid,
4736 pci_dev->addr.function, "rx_port_stats");
4737 mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4738 mz = rte_memzone_lookup(mz_name);
4740 RTE_CACHE_LINE_ROUNDUP(sizeof(struct rx_port_stats) +
4741 sizeof(struct rx_port_stats_ext) + 512);
4743 mz = rte_memzone_reserve(mz_name, total_alloc_len,
4746 RTE_MEMZONE_SIZE_HINT_ONLY |
4747 RTE_MEMZONE_IOVA_CONTIG);
4751 memset(mz->addr, 0, mz->len);
4752 mz_phys_addr = mz->iova;
4754 bp->rx_mem_zone = (const void *)mz;
4755 bp->hw_rx_port_stats = mz->addr;
4756 bp->hw_rx_port_stats_map = mz_phys_addr;
4758 snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4759 "bnxt_" PCI_PRI_FMT "-%s", pci_dev->addr.domain,
4760 pci_dev->addr.bus, pci_dev->addr.devid,
4761 pci_dev->addr.function, "tx_port_stats");
4762 mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4763 mz = rte_memzone_lookup(mz_name);
4765 RTE_CACHE_LINE_ROUNDUP(sizeof(struct tx_port_stats) +
4766 sizeof(struct tx_port_stats_ext) + 512);
4768 mz = rte_memzone_reserve(mz_name,
4772 RTE_MEMZONE_SIZE_HINT_ONLY |
4773 RTE_MEMZONE_IOVA_CONTIG);
4777 memset(mz->addr, 0, mz->len);
4778 mz_phys_addr = mz->iova;
4780 bp->tx_mem_zone = (const void *)mz;
4781 bp->hw_tx_port_stats = mz->addr;
4782 bp->hw_tx_port_stats_map = mz_phys_addr;
4783 bp->flags |= BNXT_FLAG_PORT_STATS;
4785 /* Display extended statistics if FW supports it */
4786 if (bp->hwrm_spec_code < HWRM_SPEC_CODE_1_8_4 ||
4787 bp->hwrm_spec_code == HWRM_SPEC_CODE_1_9_0 ||
4788 !(bp->flags & BNXT_FLAG_EXT_STATS_SUPPORTED))
4791 bp->hw_rx_port_stats_ext = (void *)
4792 ((uint8_t *)bp->hw_rx_port_stats +
4793 sizeof(struct rx_port_stats));
4794 bp->hw_rx_port_stats_ext_map = bp->hw_rx_port_stats_map +
4795 sizeof(struct rx_port_stats);
4796 bp->flags |= BNXT_FLAG_EXT_RX_PORT_STATS;
4798 if (bp->hwrm_spec_code < HWRM_SPEC_CODE_1_9_2 ||
4799 bp->flags & BNXT_FLAG_EXT_STATS_SUPPORTED) {
4800 bp->hw_tx_port_stats_ext = (void *)
4801 ((uint8_t *)bp->hw_tx_port_stats +
4802 sizeof(struct tx_port_stats));
4803 bp->hw_tx_port_stats_ext_map =
4804 bp->hw_tx_port_stats_map +
4805 sizeof(struct tx_port_stats);
4806 bp->flags |= BNXT_FLAG_EXT_TX_PORT_STATS;
4812 static int bnxt_setup_mac_addr(struct rte_eth_dev *eth_dev)
4814 struct bnxt *bp = eth_dev->data->dev_private;
4817 eth_dev->data->mac_addrs = rte_zmalloc("bnxt_mac_addr_tbl",
4818 RTE_ETHER_ADDR_LEN *
4821 if (eth_dev->data->mac_addrs == NULL) {
4822 PMD_DRV_LOG(ERR, "Failed to alloc MAC addr tbl\n");
4826 if (!BNXT_HAS_DFLT_MAC_SET(bp)) {
4830 /* Generate a random MAC address, if none was assigned by PF */
4831 PMD_DRV_LOG(INFO, "VF MAC address not assigned by Host PF\n");
4832 bnxt_eth_hw_addr_random(bp->mac_addr);
4834 "Assign random MAC:%02X:%02X:%02X:%02X:%02X:%02X\n",
4835 bp->mac_addr[0], bp->mac_addr[1], bp->mac_addr[2],
4836 bp->mac_addr[3], bp->mac_addr[4], bp->mac_addr[5]);
4838 rc = bnxt_hwrm_set_mac(bp);
4843 /* Copy the permanent MAC from the FUNC_QCAPS response */
4844 memcpy(ð_dev->data->mac_addrs[0], bp->mac_addr, RTE_ETHER_ADDR_LEN);
4849 static int bnxt_restore_dflt_mac(struct bnxt *bp)
4853 /* MAC is already configured in FW */
4854 if (BNXT_HAS_DFLT_MAC_SET(bp))
4857 /* Restore the old MAC configured */
4858 rc = bnxt_hwrm_set_mac(bp);
4860 PMD_DRV_LOG(ERR, "Failed to restore MAC address\n");
4865 static void bnxt_config_vf_req_fwd(struct bnxt *bp)
4870 memset(bp->pf->vf_req_fwd, 0, sizeof(bp->pf->vf_req_fwd));
4872 if (!(bp->fw_cap & BNXT_FW_CAP_LINK_ADMIN))
4873 BNXT_HWRM_CMD_TO_FORWARD(HWRM_PORT_PHY_QCFG);
4874 BNXT_HWRM_CMD_TO_FORWARD(HWRM_FUNC_CFG);
4875 BNXT_HWRM_CMD_TO_FORWARD(HWRM_FUNC_VF_CFG);
4876 BNXT_HWRM_CMD_TO_FORWARD(HWRM_CFA_L2_FILTER_ALLOC);
4877 BNXT_HWRM_CMD_TO_FORWARD(HWRM_OEM_CMD);
4881 bnxt_get_svif(uint16_t port_id, bool func_svif,
4882 enum bnxt_ulp_intf_type type)
4884 struct rte_eth_dev *eth_dev;
4887 eth_dev = &rte_eth_devices[port_id];
4888 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
4889 struct bnxt_representor *vfr = eth_dev->data->dev_private;
4893 if (type == BNXT_ULP_INTF_TYPE_VF_REP)
4896 eth_dev = vfr->parent_dev;
4899 bp = eth_dev->data->dev_private;
4901 return func_svif ? bp->func_svif : bp->port_svif;
4905 bnxt_get_vnic_id(uint16_t port, enum bnxt_ulp_intf_type type)
4907 struct rte_eth_dev *eth_dev;
4908 struct bnxt_vnic_info *vnic;
4911 eth_dev = &rte_eth_devices[port];
4912 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
4913 struct bnxt_representor *vfr = eth_dev->data->dev_private;
4917 if (type == BNXT_ULP_INTF_TYPE_VF_REP)
4918 return vfr->dflt_vnic_id;
4920 eth_dev = vfr->parent_dev;
4923 bp = eth_dev->data->dev_private;
4925 vnic = BNXT_GET_DEFAULT_VNIC(bp);
4927 return vnic->fw_vnic_id;
4931 bnxt_get_fw_func_id(uint16_t port, enum bnxt_ulp_intf_type type)
4933 struct rte_eth_dev *eth_dev;
4936 eth_dev = &rte_eth_devices[port];
4937 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
4938 struct bnxt_representor *vfr = eth_dev->data->dev_private;
4942 if (type == BNXT_ULP_INTF_TYPE_VF_REP)
4945 eth_dev = vfr->parent_dev;
4948 bp = eth_dev->data->dev_private;
4953 enum bnxt_ulp_intf_type
4954 bnxt_get_interface_type(uint16_t port)
4956 struct rte_eth_dev *eth_dev;
4959 eth_dev = &rte_eth_devices[port];
4960 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev))
4961 return BNXT_ULP_INTF_TYPE_VF_REP;
4963 bp = eth_dev->data->dev_private;
4965 return BNXT_ULP_INTF_TYPE_PF;
4966 else if (BNXT_VF_IS_TRUSTED(bp))
4967 return BNXT_ULP_INTF_TYPE_TRUSTED_VF;
4968 else if (BNXT_VF(bp))
4969 return BNXT_ULP_INTF_TYPE_VF;
4971 return BNXT_ULP_INTF_TYPE_INVALID;
4975 bnxt_get_phy_port_id(uint16_t port_id)
4977 struct bnxt_representor *vfr;
4978 struct rte_eth_dev *eth_dev;
4981 eth_dev = &rte_eth_devices[port_id];
4982 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
4983 vfr = eth_dev->data->dev_private;
4987 eth_dev = vfr->parent_dev;
4990 bp = eth_dev->data->dev_private;
4992 return BNXT_PF(bp) ? bp->pf->port_id : bp->parent->port_id;
4996 bnxt_get_parif(uint16_t port_id, enum bnxt_ulp_intf_type type)
4998 struct rte_eth_dev *eth_dev;
5001 eth_dev = &rte_eth_devices[port_id];
5002 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
5003 struct bnxt_representor *vfr = eth_dev->data->dev_private;
5007 if (type == BNXT_ULP_INTF_TYPE_VF_REP)
5008 return vfr->fw_fid - 1;
5010 eth_dev = vfr->parent_dev;
5013 bp = eth_dev->data->dev_private;
5015 return BNXT_PF(bp) ? bp->fw_fid - 1 : bp->parent->fid - 1;
5019 bnxt_get_vport(uint16_t port_id)
5021 return (1 << bnxt_get_phy_port_id(port_id));
5024 static void bnxt_alloc_error_recovery_info(struct bnxt *bp)
5026 struct bnxt_error_recovery_info *info = bp->recovery_info;
5029 if (!(bp->fw_cap & BNXT_FW_CAP_HCOMM_FW_STATUS))
5030 memset(info, 0, sizeof(*info));
5034 if (!(bp->fw_cap & BNXT_FW_CAP_ERROR_RECOVERY))
5037 info = rte_zmalloc("bnxt_hwrm_error_recovery_qcfg",
5040 bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5042 bp->recovery_info = info;
5045 static void bnxt_check_fw_status(struct bnxt *bp)
5049 if (!(bp->recovery_info &&
5050 (bp->fw_cap & BNXT_FW_CAP_HCOMM_FW_STATUS)))
5053 fw_status = bnxt_read_fw_status_reg(bp, BNXT_FW_STATUS_REG);
5054 if (fw_status != BNXT_FW_STATUS_HEALTHY)
5055 PMD_DRV_LOG(ERR, "Firmware not responding, status: %#x\n",
5059 static int bnxt_map_hcomm_fw_status_reg(struct bnxt *bp)
5061 struct bnxt_error_recovery_info *info = bp->recovery_info;
5062 uint32_t status_loc;
5065 rte_write32(HCOMM_STATUS_STRUCT_LOC, (uint8_t *)bp->bar0 +
5066 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
5067 sig_ver = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
5068 BNXT_GRCP_WINDOW_2_BASE +
5069 offsetof(struct hcomm_status,
5071 /* If the signature is absent, then FW does not support this feature */
5072 if ((sig_ver & HCOMM_STATUS_SIGNATURE_MASK) !=
5073 HCOMM_STATUS_SIGNATURE_VAL)
5077 info = rte_zmalloc("bnxt_hwrm_error_recovery_qcfg",
5081 bp->recovery_info = info;
5083 memset(info, 0, sizeof(*info));
5086 status_loc = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
5087 BNXT_GRCP_WINDOW_2_BASE +
5088 offsetof(struct hcomm_status,
5091 /* Only pre-map the FW health status GRC register */
5092 if (BNXT_FW_STATUS_REG_TYPE(status_loc) != BNXT_FW_STATUS_REG_TYPE_GRC)
5095 info->status_regs[BNXT_FW_STATUS_REG] = status_loc;
5096 info->mapped_status_regs[BNXT_FW_STATUS_REG] =
5097 BNXT_GRCP_WINDOW_2_BASE + (status_loc & BNXT_GRCP_OFFSET_MASK);
5099 rte_write32((status_loc & BNXT_GRCP_BASE_MASK), (uint8_t *)bp->bar0 +
5100 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
5102 bp->fw_cap |= BNXT_FW_CAP_HCOMM_FW_STATUS;
5107 /* This function gets the FW version along with the
5108 * capabilities(MAX and current) of the function, vnic,
5109 * error recovery, phy and other chip related info
5111 static int bnxt_get_config(struct bnxt *bp)
5118 rc = bnxt_map_hcomm_fw_status_reg(bp);
5122 rc = bnxt_hwrm_ver_get(bp, DFLT_HWRM_CMD_TIMEOUT);
5124 bnxt_check_fw_status(bp);
5128 rc = bnxt_hwrm_func_reset(bp);
5132 rc = bnxt_hwrm_vnic_qcaps(bp);
5136 rc = bnxt_hwrm_queue_qportcfg(bp);
5140 /* Get the MAX capabilities for this function.
5141 * This function also allocates context memory for TQM rings and
5142 * informs the firmware about this allocated backing store memory.
5144 rc = bnxt_hwrm_func_qcaps(bp);
5148 rc = bnxt_hwrm_func_qcfg(bp, &mtu);
5152 rc = bnxt_hwrm_cfa_adv_flow_mgmt_qcaps(bp);
5156 bnxt_hwrm_port_mac_qcfg(bp);
5158 bnxt_hwrm_parent_pf_qcfg(bp);
5160 bnxt_hwrm_port_phy_qcaps(bp);
5162 bnxt_alloc_error_recovery_info(bp);
5163 /* Get the adapter error recovery support info */
5164 rc = bnxt_hwrm_error_recovery_qcfg(bp);
5166 bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5168 bnxt_hwrm_port_led_qcaps(bp);
5174 bnxt_init_locks(struct bnxt *bp)
5178 err = pthread_mutex_init(&bp->flow_lock, NULL);
5180 PMD_DRV_LOG(ERR, "Unable to initialize flow_lock\n");
5184 err = pthread_mutex_init(&bp->def_cp_lock, NULL);
5186 PMD_DRV_LOG(ERR, "Unable to initialize def_cp_lock\n");
5190 err = pthread_mutex_init(&bp->health_check_lock, NULL);
5192 PMD_DRV_LOG(ERR, "Unable to initialize health_check_lock\n");
5196 err = pthread_mutex_init(&bp->err_recovery_lock, NULL);
5198 PMD_DRV_LOG(ERR, "Unable to initialize err_recovery_lock\n");
5203 static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev)
5207 rc = bnxt_get_config(bp);
5211 if (!reconfig_dev) {
5212 rc = bnxt_setup_mac_addr(bp->eth_dev);
5216 rc = bnxt_restore_dflt_mac(bp);
5221 bnxt_config_vf_req_fwd(bp);
5223 rc = bnxt_hwrm_func_driver_register(bp);
5225 PMD_DRV_LOG(ERR, "Failed to register driver");
5230 if (bp->pdev->max_vfs) {
5231 rc = bnxt_hwrm_allocate_vfs(bp, bp->pdev->max_vfs);
5233 PMD_DRV_LOG(ERR, "Failed to allocate VFs\n");
5237 rc = bnxt_hwrm_allocate_pf_only(bp);
5240 "Failed to allocate PF resources");
5246 rc = bnxt_alloc_mem(bp, reconfig_dev);
5250 rc = bnxt_setup_int(bp);
5254 rc = bnxt_request_int(bp);
5258 rc = bnxt_init_ctx_mem(bp);
5260 PMD_DRV_LOG(ERR, "Failed to init adv_flow_counters\n");
5268 bnxt_parse_devarg_truflow(__rte_unused const char *key,
5269 const char *value, void *opaque_arg)
5271 struct bnxt *bp = opaque_arg;
5272 unsigned long truflow;
5275 if (!value || !opaque_arg) {
5277 "Invalid parameter passed to truflow devargs.\n");
5281 truflow = strtoul(value, &end, 10);
5282 if (end == NULL || *end != '\0' ||
5283 (truflow == ULONG_MAX && errno == ERANGE)) {
5285 "Invalid parameter passed to truflow devargs.\n");
5289 if (BNXT_DEVARG_TRUFLOW_INVALID(truflow)) {
5291 "Invalid value passed to truflow devargs.\n");
5296 bp->flags |= BNXT_FLAG_TRUFLOW_EN;
5297 PMD_DRV_LOG(INFO, "Host-based truflow feature enabled.\n");
5299 bp->flags &= ~BNXT_FLAG_TRUFLOW_EN;
5300 PMD_DRV_LOG(INFO, "Host-based truflow feature disabled.\n");
5307 bnxt_parse_devarg_flow_xstat(__rte_unused const char *key,
5308 const char *value, void *opaque_arg)
5310 struct bnxt *bp = opaque_arg;
5311 unsigned long flow_xstat;
5314 if (!value || !opaque_arg) {
5316 "Invalid parameter passed to flow_xstat devarg.\n");
5320 flow_xstat = strtoul(value, &end, 10);
5321 if (end == NULL || *end != '\0' ||
5322 (flow_xstat == ULONG_MAX && errno == ERANGE)) {
5324 "Invalid parameter passed to flow_xstat devarg.\n");
5328 if (BNXT_DEVARG_FLOW_XSTAT_INVALID(flow_xstat)) {
5330 "Invalid value passed to flow_xstat devarg.\n");
5334 bp->flags |= BNXT_FLAG_FLOW_XSTATS_EN;
5335 if (BNXT_FLOW_XSTATS_EN(bp))
5336 PMD_DRV_LOG(INFO, "flow_xstat feature enabled.\n");
5342 bnxt_parse_devarg_max_num_kflows(__rte_unused const char *key,
5343 const char *value, void *opaque_arg)
5345 struct bnxt *bp = opaque_arg;
5346 unsigned long max_num_kflows;
5349 if (!value || !opaque_arg) {
5351 "Invalid parameter passed to max_num_kflows devarg.\n");
5355 max_num_kflows = strtoul(value, &end, 10);
5356 if (end == NULL || *end != '\0' ||
5357 (max_num_kflows == ULONG_MAX && errno == ERANGE)) {
5359 "Invalid parameter passed to max_num_kflows devarg.\n");
5363 if (bnxt_devarg_max_num_kflow_invalid(max_num_kflows)) {
5365 "Invalid value passed to max_num_kflows devarg.\n");
5369 bp->max_num_kflows = max_num_kflows;
5370 if (bp->max_num_kflows)
5371 PMD_DRV_LOG(INFO, "max_num_kflows set as %ldK.\n",
5378 bnxt_parse_devarg_rep_is_pf(__rte_unused const char *key,
5379 const char *value, void *opaque_arg)
5381 struct bnxt_representor *vfr_bp = opaque_arg;
5382 unsigned long rep_is_pf;
5385 if (!value || !opaque_arg) {
5387 "Invalid parameter passed to rep_is_pf devargs.\n");
5391 rep_is_pf = strtoul(value, &end, 10);
5392 if (end == NULL || *end != '\0' ||
5393 (rep_is_pf == ULONG_MAX && errno == ERANGE)) {
5395 "Invalid parameter passed to rep_is_pf devargs.\n");
5399 if (BNXT_DEVARG_REP_IS_PF_INVALID(rep_is_pf)) {
5401 "Invalid value passed to rep_is_pf devargs.\n");
5405 vfr_bp->flags |= rep_is_pf;
5406 if (BNXT_REP_PF(vfr_bp))
5407 PMD_DRV_LOG(INFO, "PF representor\n");
5409 PMD_DRV_LOG(INFO, "VF representor\n");
5415 bnxt_parse_devarg_rep_based_pf(__rte_unused const char *key,
5416 const char *value, void *opaque_arg)
5418 struct bnxt_representor *vfr_bp = opaque_arg;
5419 unsigned long rep_based_pf;
5422 if (!value || !opaque_arg) {
5424 "Invalid parameter passed to rep_based_pf "
5429 rep_based_pf = strtoul(value, &end, 10);
5430 if (end == NULL || *end != '\0' ||
5431 (rep_based_pf == ULONG_MAX && errno == ERANGE)) {
5433 "Invalid parameter passed to rep_based_pf "
5438 if (BNXT_DEVARG_REP_BASED_PF_INVALID(rep_based_pf)) {
5440 "Invalid value passed to rep_based_pf devargs.\n");
5444 vfr_bp->rep_based_pf = rep_based_pf;
5445 vfr_bp->flags |= BNXT_REP_BASED_PF_VALID;
5447 PMD_DRV_LOG(INFO, "rep-based-pf = %d\n", vfr_bp->rep_based_pf);
5453 bnxt_parse_devarg_rep_q_r2f(__rte_unused const char *key,
5454 const char *value, void *opaque_arg)
5456 struct bnxt_representor *vfr_bp = opaque_arg;
5457 unsigned long rep_q_r2f;
5460 if (!value || !opaque_arg) {
5462 "Invalid parameter passed to rep_q_r2f "
5467 rep_q_r2f = strtoul(value, &end, 10);
5468 if (end == NULL || *end != '\0' ||
5469 (rep_q_r2f == ULONG_MAX && errno == ERANGE)) {
5471 "Invalid parameter passed to rep_q_r2f "
5476 if (BNXT_DEVARG_REP_Q_R2F_INVALID(rep_q_r2f)) {
5478 "Invalid value passed to rep_q_r2f devargs.\n");
5482 vfr_bp->rep_q_r2f = rep_q_r2f;
5483 vfr_bp->flags |= BNXT_REP_Q_R2F_VALID;
5484 PMD_DRV_LOG(INFO, "rep-q-r2f = %d\n", vfr_bp->rep_q_r2f);
5490 bnxt_parse_devarg_rep_q_f2r(__rte_unused const char *key,
5491 const char *value, void *opaque_arg)
5493 struct bnxt_representor *vfr_bp = opaque_arg;
5494 unsigned long rep_q_f2r;
5497 if (!value || !opaque_arg) {
5499 "Invalid parameter passed to rep_q_f2r "
5504 rep_q_f2r = strtoul(value, &end, 10);
5505 if (end == NULL || *end != '\0' ||
5506 (rep_q_f2r == ULONG_MAX && errno == ERANGE)) {
5508 "Invalid parameter passed to rep_q_f2r "
5513 if (BNXT_DEVARG_REP_Q_F2R_INVALID(rep_q_f2r)) {
5515 "Invalid value passed to rep_q_f2r devargs.\n");
5519 vfr_bp->rep_q_f2r = rep_q_f2r;
5520 vfr_bp->flags |= BNXT_REP_Q_F2R_VALID;
5521 PMD_DRV_LOG(INFO, "rep-q-f2r = %d\n", vfr_bp->rep_q_f2r);
5527 bnxt_parse_devarg_rep_fc_r2f(__rte_unused const char *key,
5528 const char *value, void *opaque_arg)
5530 struct bnxt_representor *vfr_bp = opaque_arg;
5531 unsigned long rep_fc_r2f;
5534 if (!value || !opaque_arg) {
5536 "Invalid parameter passed to rep_fc_r2f "
5541 rep_fc_r2f = strtoul(value, &end, 10);
5542 if (end == NULL || *end != '\0' ||
5543 (rep_fc_r2f == ULONG_MAX && errno == ERANGE)) {
5545 "Invalid parameter passed to rep_fc_r2f "
5550 if (BNXT_DEVARG_REP_FC_R2F_INVALID(rep_fc_r2f)) {
5552 "Invalid value passed to rep_fc_r2f devargs.\n");
5556 vfr_bp->flags |= BNXT_REP_FC_R2F_VALID;
5557 vfr_bp->rep_fc_r2f = rep_fc_r2f;
5558 PMD_DRV_LOG(INFO, "rep-fc-r2f = %lu\n", rep_fc_r2f);
5564 bnxt_parse_devarg_rep_fc_f2r(__rte_unused const char *key,
5565 const char *value, void *opaque_arg)
5567 struct bnxt_representor *vfr_bp = opaque_arg;
5568 unsigned long rep_fc_f2r;
5571 if (!value || !opaque_arg) {
5573 "Invalid parameter passed to rep_fc_f2r "
5578 rep_fc_f2r = strtoul(value, &end, 10);
5579 if (end == NULL || *end != '\0' ||
5580 (rep_fc_f2r == ULONG_MAX && errno == ERANGE)) {
5582 "Invalid parameter passed to rep_fc_f2r "
5587 if (BNXT_DEVARG_REP_FC_F2R_INVALID(rep_fc_f2r)) {
5589 "Invalid value passed to rep_fc_f2r devargs.\n");
5593 vfr_bp->flags |= BNXT_REP_FC_F2R_VALID;
5594 vfr_bp->rep_fc_f2r = rep_fc_f2r;
5595 PMD_DRV_LOG(INFO, "rep-fc-f2r = %lu\n", rep_fc_f2r);
5601 bnxt_parse_dev_args(struct bnxt *bp, struct rte_devargs *devargs)
5603 struct rte_kvargs *kvlist;
5606 if (devargs == NULL)
5609 kvlist = rte_kvargs_parse(devargs->args, bnxt_dev_args);
5614 * Handler for "truflow" devarg.
5615 * Invoked as for ex: "-a 0000:00:0d.0,host-based-truflow=1"
5617 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_TRUFLOW,
5618 bnxt_parse_devarg_truflow, bp);
5623 * Handler for "flow_xstat" devarg.
5624 * Invoked as for ex: "-a 0000:00:0d.0,flow_xstat=1"
5626 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_FLOW_XSTAT,
5627 bnxt_parse_devarg_flow_xstat, bp);
5632 * Handler for "max_num_kflows" devarg.
5633 * Invoked as for ex: "-a 000:00:0d.0,max_num_kflows=32"
5635 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_MAX_NUM_KFLOWS,
5636 bnxt_parse_devarg_max_num_kflows, bp);
5641 rte_kvargs_free(kvlist);
5645 static int bnxt_alloc_switch_domain(struct bnxt *bp)
5649 if (BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) {
5650 rc = rte_eth_switch_domain_alloc(&bp->switch_domain_id);
5653 "Failed to alloc switch domain: %d\n", rc);
5656 "Switch domain allocated %d\n",
5657 bp->switch_domain_id);
5663 /* Allocate and initialize various fields in bnxt struct that
5664 * need to be allocated/destroyed only once in the lifetime of the driver
5666 static int bnxt_drv_init(struct rte_eth_dev *eth_dev)
5668 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
5669 struct bnxt *bp = eth_dev->data->dev_private;
5672 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
5674 if (bnxt_vf_pciid(pci_dev->id.device_id))
5675 bp->flags |= BNXT_FLAG_VF;
5677 if (bnxt_p5_device(pci_dev->id.device_id))
5678 bp->flags |= BNXT_FLAG_CHIP_P5;
5680 if (pci_dev->id.device_id == BROADCOM_DEV_ID_58802 ||
5681 pci_dev->id.device_id == BROADCOM_DEV_ID_58804 ||
5682 pci_dev->id.device_id == BROADCOM_DEV_ID_58808 ||
5683 pci_dev->id.device_id == BROADCOM_DEV_ID_58802_VF)
5684 bp->flags |= BNXT_FLAG_STINGRAY;
5686 if (BNXT_TRUFLOW_EN(bp)) {
5687 /* extra mbuf field is required to store CFA code from mark */
5688 static const struct rte_mbuf_dynfield bnxt_cfa_code_dynfield_desc = {
5689 .name = RTE_PMD_BNXT_CFA_CODE_DYNFIELD_NAME,
5690 .size = sizeof(bnxt_cfa_code_dynfield_t),
5691 .align = __alignof__(bnxt_cfa_code_dynfield_t),
5693 bnxt_cfa_code_dynfield_offset =
5694 rte_mbuf_dynfield_register(&bnxt_cfa_code_dynfield_desc);
5695 if (bnxt_cfa_code_dynfield_offset < 0) {
5697 "Failed to register mbuf field for TruFlow mark\n");
5702 rc = bnxt_map_pci_bars(eth_dev);
5705 "Failed to initialize board rc: %x\n", rc);
5709 rc = bnxt_alloc_pf_info(bp);
5713 rc = bnxt_alloc_link_info(bp);
5717 rc = bnxt_alloc_parent_info(bp);
5721 rc = bnxt_alloc_hwrm_resources(bp);
5724 "Failed to allocate response buffer rc: %x\n", rc);
5727 rc = bnxt_alloc_leds_info(bp);
5731 rc = bnxt_alloc_cos_queues(bp);
5735 rc = bnxt_init_locks(bp);
5739 rc = bnxt_alloc_switch_domain(bp);
5747 bnxt_dev_init(struct rte_eth_dev *eth_dev, void *params __rte_unused)
5749 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
5750 static int version_printed;
5754 if (version_printed++ == 0)
5755 PMD_DRV_LOG(INFO, "%s\n", bnxt_version);
5757 eth_dev->dev_ops = &bnxt_dev_ops;
5758 eth_dev->rx_queue_count = bnxt_rx_queue_count_op;
5759 eth_dev->rx_descriptor_status = bnxt_rx_descriptor_status_op;
5760 eth_dev->tx_descriptor_status = bnxt_tx_descriptor_status_op;
5761 eth_dev->rx_pkt_burst = &bnxt_recv_pkts;
5762 eth_dev->tx_pkt_burst = &bnxt_xmit_pkts;
5765 * For secondary processes, we don't initialise any further
5766 * as primary has already done this work.
5768 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
5771 rte_eth_copy_pci_info(eth_dev, pci_dev);
5772 eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
5774 bp = eth_dev->data->dev_private;
5776 /* Parse dev arguments passed on when starting the DPDK application. */
5777 rc = bnxt_parse_dev_args(bp, pci_dev->device.devargs);
5781 rc = bnxt_drv_init(eth_dev);
5785 rc = bnxt_init_resources(bp, false);
5789 rc = bnxt_alloc_stats_mem(bp);
5794 DRV_MODULE_NAME "found at mem %" PRIX64 ", node addr %pM\n",
5795 pci_dev->mem_resource[0].phys_addr,
5796 pci_dev->mem_resource[0].addr);
5801 bnxt_dev_uninit(eth_dev);
5806 static void bnxt_free_ctx_mem_buf(struct bnxt_ctx_mem_buf_info *ctx)
5815 ctx->dma = RTE_BAD_IOVA;
5816 ctx->ctx_id = BNXT_CTX_VAL_INVAL;
5819 static void bnxt_unregister_fc_ctx_mem(struct bnxt *bp)
5821 bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_RX,
5822 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
5823 bp->flow_stat->rx_fc_out_tbl.ctx_id,
5824 bp->flow_stat->max_fc,
5827 bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_TX,
5828 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
5829 bp->flow_stat->tx_fc_out_tbl.ctx_id,
5830 bp->flow_stat->max_fc,
5833 if (bp->flow_stat->rx_fc_in_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5834 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->rx_fc_in_tbl.ctx_id);
5835 bp->flow_stat->rx_fc_in_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5837 if (bp->flow_stat->rx_fc_out_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5838 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->rx_fc_out_tbl.ctx_id);
5839 bp->flow_stat->rx_fc_out_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5841 if (bp->flow_stat->tx_fc_in_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5842 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->tx_fc_in_tbl.ctx_id);
5843 bp->flow_stat->tx_fc_in_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5845 if (bp->flow_stat->tx_fc_out_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5846 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->tx_fc_out_tbl.ctx_id);
5847 bp->flow_stat->tx_fc_out_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5850 static void bnxt_uninit_fc_ctx_mem(struct bnxt *bp)
5852 bnxt_unregister_fc_ctx_mem(bp);
5854 bnxt_free_ctx_mem_buf(&bp->flow_stat->rx_fc_in_tbl);
5855 bnxt_free_ctx_mem_buf(&bp->flow_stat->rx_fc_out_tbl);
5856 bnxt_free_ctx_mem_buf(&bp->flow_stat->tx_fc_in_tbl);
5857 bnxt_free_ctx_mem_buf(&bp->flow_stat->tx_fc_out_tbl);
5860 static void bnxt_uninit_ctx_mem(struct bnxt *bp)
5862 if (BNXT_FLOW_XSTATS_EN(bp))
5863 bnxt_uninit_fc_ctx_mem(bp);
5867 bnxt_free_error_recovery_info(struct bnxt *bp)
5869 rte_free(bp->recovery_info);
5870 bp->recovery_info = NULL;
5871 bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5875 bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev)
5880 bnxt_free_mem(bp, reconfig_dev);
5882 bnxt_hwrm_func_buf_unrgtr(bp);
5883 if (bp->pf != NULL) {
5884 rte_free(bp->pf->vf_req_buf);
5885 bp->pf->vf_req_buf = NULL;
5888 rc = bnxt_hwrm_func_driver_unregister(bp, 0);
5889 bp->flags &= ~BNXT_FLAG_REGISTERED;
5890 bnxt_free_ctx_mem(bp);
5891 if (!reconfig_dev) {
5892 bnxt_free_hwrm_resources(bp);
5893 bnxt_free_error_recovery_info(bp);
5896 bnxt_uninit_ctx_mem(bp);
5898 bnxt_free_flow_stats_info(bp);
5899 if (bp->rep_info != NULL)
5900 bnxt_free_switch_domain(bp);
5901 bnxt_free_rep_info(bp);
5902 rte_free(bp->ptp_cfg);
5908 bnxt_dev_uninit(struct rte_eth_dev *eth_dev)
5910 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
5913 PMD_DRV_LOG(DEBUG, "Calling Device uninit\n");
5915 if (eth_dev->state != RTE_ETH_DEV_UNUSED)
5916 bnxt_dev_close_op(eth_dev);
5921 static int bnxt_pci_remove_dev_with_reps(struct rte_eth_dev *eth_dev)
5923 struct bnxt *bp = eth_dev->data->dev_private;
5924 struct rte_eth_dev *vf_rep_eth_dev;
5930 for (i = 0; i < bp->num_reps; i++) {
5931 vf_rep_eth_dev = bp->rep_info[i].vfr_eth_dev;
5932 if (!vf_rep_eth_dev)
5934 PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR pci remove\n",
5935 vf_rep_eth_dev->data->port_id);
5936 rte_eth_dev_destroy(vf_rep_eth_dev, bnxt_representor_uninit);
5938 PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci remove\n",
5939 eth_dev->data->port_id);
5940 ret = rte_eth_dev_destroy(eth_dev, bnxt_dev_uninit);
5945 static void bnxt_free_rep_info(struct bnxt *bp)
5947 rte_free(bp->rep_info);
5948 bp->rep_info = NULL;
5949 rte_free(bp->cfa_code_map);
5950 bp->cfa_code_map = NULL;
5953 static int bnxt_init_rep_info(struct bnxt *bp)
5960 bp->rep_info = rte_zmalloc("bnxt_rep_info",
5961 sizeof(bp->rep_info[0]) * BNXT_MAX_VF_REPS,
5963 if (!bp->rep_info) {
5964 PMD_DRV_LOG(ERR, "Failed to alloc memory for rep info\n");
5967 bp->cfa_code_map = rte_zmalloc("bnxt_cfa_code_map",
5968 sizeof(*bp->cfa_code_map) *
5969 BNXT_MAX_CFA_CODE, 0);
5970 if (!bp->cfa_code_map) {
5971 PMD_DRV_LOG(ERR, "Failed to alloc memory for cfa_code_map\n");
5972 bnxt_free_rep_info(bp);
5976 for (i = 0; i < BNXT_MAX_CFA_CODE; i++)
5977 bp->cfa_code_map[i] = BNXT_VF_IDX_INVALID;
5979 rc = pthread_mutex_init(&bp->rep_info->vfr_lock, NULL);
5981 PMD_DRV_LOG(ERR, "Unable to initialize vfr_lock\n");
5982 bnxt_free_rep_info(bp);
5986 rc = pthread_mutex_init(&bp->rep_info->vfr_start_lock, NULL);
5988 PMD_DRV_LOG(ERR, "Unable to initialize vfr_start_lock\n");
5989 bnxt_free_rep_info(bp);
5996 static int bnxt_rep_port_probe(struct rte_pci_device *pci_dev,
5997 struct rte_eth_devargs *eth_da,
5998 struct rte_eth_dev *backing_eth_dev,
5999 const char *dev_args)
6001 struct rte_eth_dev *vf_rep_eth_dev;
6002 char name[RTE_ETH_NAME_MAX_LEN];
6003 struct bnxt *backing_bp;
6006 struct rte_kvargs *kvlist = NULL;
6008 if (eth_da->type == RTE_ETH_REPRESENTOR_NONE)
6010 if (eth_da->type != RTE_ETH_REPRESENTOR_VF) {
6011 PMD_DRV_LOG(ERR, "unsupported representor type %d\n",
6015 num_rep = eth_da->nb_representor_ports;
6016 if (num_rep > BNXT_MAX_VF_REPS) {
6017 PMD_DRV_LOG(ERR, "nb_representor_ports = %d > %d MAX VF REPS\n",
6018 num_rep, BNXT_MAX_VF_REPS);
6022 if (num_rep >= RTE_MAX_ETHPORTS) {
6024 "nb_representor_ports = %d > %d MAX ETHPORTS\n",
6025 num_rep, RTE_MAX_ETHPORTS);
6029 backing_bp = backing_eth_dev->data->dev_private;
6031 if (!(BNXT_PF(backing_bp) || BNXT_VF_IS_TRUSTED(backing_bp))) {
6033 "Not a PF or trusted VF. No Representor support\n");
6034 /* Returning an error is not an option.
6035 * Applications are not handling this correctly
6040 if (bnxt_init_rep_info(backing_bp))
6043 for (i = 0; i < num_rep; i++) {
6044 struct bnxt_representor representor = {
6045 .vf_id = eth_da->representor_ports[i],
6046 .switch_domain_id = backing_bp->switch_domain_id,
6047 .parent_dev = backing_eth_dev
6050 if (representor.vf_id >= BNXT_MAX_VF_REPS) {
6051 PMD_DRV_LOG(ERR, "VF-Rep id %d >= %d MAX VF ID\n",
6052 representor.vf_id, BNXT_MAX_VF_REPS);
6056 /* representor port net_bdf_port */
6057 snprintf(name, sizeof(name), "net_%s_representor_%d",
6058 pci_dev->device.name, eth_da->representor_ports[i]);
6060 kvlist = rte_kvargs_parse(dev_args, bnxt_dev_args);
6063 * Handler for "rep_is_pf" devarg.
6064 * Invoked as for ex: "-a 000:00:0d.0,
6065 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6067 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_IS_PF,
6068 bnxt_parse_devarg_rep_is_pf,
6069 (void *)&representor);
6075 * Handler for "rep_based_pf" devarg.
6076 * Invoked as for ex: "-a 000:00:0d.0,
6077 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6079 ret = rte_kvargs_process(kvlist,
6080 BNXT_DEVARG_REP_BASED_PF,
6081 bnxt_parse_devarg_rep_based_pf,
6082 (void *)&representor);
6088 * Handler for "rep_based_pf" devarg.
6089 * Invoked as for ex: "-a 000:00:0d.0,
6090 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6092 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_Q_R2F,
6093 bnxt_parse_devarg_rep_q_r2f,
6094 (void *)&representor);
6100 * Handler for "rep_based_pf" devarg.
6101 * Invoked as for ex: "-a 000:00:0d.0,
6102 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6104 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_Q_F2R,
6105 bnxt_parse_devarg_rep_q_f2r,
6106 (void *)&representor);
6112 * Handler for "rep_based_pf" devarg.
6113 * Invoked as for ex: "-a 000:00:0d.0,
6114 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6116 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_FC_R2F,
6117 bnxt_parse_devarg_rep_fc_r2f,
6118 (void *)&representor);
6124 * Handler for "rep_based_pf" devarg.
6125 * Invoked as for ex: "-a 000:00:0d.0,
6126 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6128 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_FC_F2R,
6129 bnxt_parse_devarg_rep_fc_f2r,
6130 (void *)&representor);
6137 ret = rte_eth_dev_create(&pci_dev->device, name,
6138 sizeof(struct bnxt_representor),
6140 bnxt_representor_init,
6143 PMD_DRV_LOG(ERR, "failed to create bnxt vf "
6144 "representor %s.", name);
6148 vf_rep_eth_dev = rte_eth_dev_allocated(name);
6149 if (!vf_rep_eth_dev) {
6150 PMD_DRV_LOG(ERR, "Failed to find the eth_dev"
6151 " for VF-Rep: %s.", name);
6156 PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR pci probe\n",
6157 backing_eth_dev->data->port_id);
6158 backing_bp->rep_info[representor.vf_id].vfr_eth_dev =
6160 backing_bp->num_reps++;
6164 rte_kvargs_free(kvlist);
6168 /* If num_rep > 1, then rollback already created
6169 * ports, since we'll be failing the probe anyway
6172 bnxt_pci_remove_dev_with_reps(backing_eth_dev);
6174 rte_kvargs_free(kvlist);
6179 static int bnxt_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
6180 struct rte_pci_device *pci_dev)
6182 struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
6183 struct rte_eth_dev *backing_eth_dev;
6187 if (pci_dev->device.devargs) {
6188 ret = rte_eth_devargs_parse(pci_dev->device.devargs->args,
6194 num_rep = eth_da.nb_representor_ports;
6195 PMD_DRV_LOG(DEBUG, "nb_representor_ports = %d\n",
6198 /* We could come here after first level of probe is already invoked
6199 * as part of an application bringup(OVS-DPDK vswitchd), so first check
6200 * for already allocated eth_dev for the backing device (PF/Trusted VF)
6202 backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6203 if (backing_eth_dev == NULL) {
6204 ret = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
6205 sizeof(struct bnxt),
6206 eth_dev_pci_specific_init, pci_dev,
6207 bnxt_dev_init, NULL);
6209 if (ret || !num_rep)
6212 backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6214 PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci probe\n",
6215 backing_eth_dev->data->port_id);
6220 /* probe representor ports now */
6221 ret = bnxt_rep_port_probe(pci_dev, ð_da, backing_eth_dev,
6222 pci_dev->device.devargs->args);
6227 static int bnxt_pci_remove(struct rte_pci_device *pci_dev)
6229 struct rte_eth_dev *eth_dev;
6231 eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6233 return 0; /* Invoked typically only by OVS-DPDK, by the
6234 * time it comes here the eth_dev is already
6235 * deleted by rte_eth_dev_close(), so returning
6236 * +ve value will at least help in proper cleanup
6239 PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci remove\n", eth_dev->data->port_id);
6240 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
6241 if (eth_dev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
6242 return rte_eth_dev_destroy(eth_dev,
6243 bnxt_representor_uninit);
6245 return rte_eth_dev_destroy(eth_dev,
6248 return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
6252 static struct rte_pci_driver bnxt_rte_pmd = {
6253 .id_table = bnxt_pci_id_map,
6254 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
6255 RTE_PCI_DRV_PROBE_AGAIN, /* Needed in case of VF-REPs
6258 .probe = bnxt_pci_probe,
6259 .remove = bnxt_pci_remove,
6263 is_device_supported(struct rte_eth_dev *dev, struct rte_pci_driver *drv)
6265 if (strcmp(dev->device->driver->name, drv->driver.name))
6271 bool is_bnxt_supported(struct rte_eth_dev *dev)
6273 return is_device_supported(dev, &bnxt_rte_pmd);
6276 RTE_LOG_REGISTER_SUFFIX(bnxt_logtype_driver, driver, NOTICE);
6277 RTE_PMD_REGISTER_PCI(net_bnxt, bnxt_rte_pmd);
6278 RTE_PMD_REGISTER_PCI_TABLE(net_bnxt, bnxt_pci_id_map);
6279 RTE_PMD_REGISTER_KMOD_DEP(net_bnxt, "* igb_uio | uio_pci_generic | vfio-pci");