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_ACCUM_STATS "accum-stats"
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"
100 #define BNXT_DEVARG_APP_ID "app-id"
102 static const char *const bnxt_dev_args[] = {
103 BNXT_DEVARG_REPRESENTOR,
104 BNXT_DEVARG_ACCUM_STATS,
105 BNXT_DEVARG_FLOW_XSTAT,
106 BNXT_DEVARG_MAX_NUM_KFLOWS,
107 BNXT_DEVARG_REP_BASED_PF,
108 BNXT_DEVARG_REP_IS_PF,
109 BNXT_DEVARG_REP_Q_R2F,
110 BNXT_DEVARG_REP_Q_F2R,
111 BNXT_DEVARG_REP_FC_R2F,
112 BNXT_DEVARG_REP_FC_F2R,
118 * accum-stats == false to disable flow counter accumulation
119 * accum-stats == true to enable flow counter accumulation
121 #define BNXT_DEVARG_ACCUM_STATS_INVALID(accum_stats) ((accum_stats) > 1)
124 * app-id = an non-negative 8-bit number
126 #define BNXT_DEVARG_APP_ID_INVALID(val) ((val) > 255)
129 * flow_xstat == false to disable the feature
130 * flow_xstat == true to enable the feature
132 #define BNXT_DEVARG_FLOW_XSTAT_INVALID(flow_xstat) ((flow_xstat) > 1)
135 * rep_is_pf == false to indicate VF representor
136 * rep_is_pf == true to indicate PF representor
138 #define BNXT_DEVARG_REP_IS_PF_INVALID(rep_is_pf) ((rep_is_pf) > 1)
141 * rep_based_pf == Physical index of the PF
143 #define BNXT_DEVARG_REP_BASED_PF_INVALID(rep_based_pf) ((rep_based_pf) > 15)
145 * rep_q_r2f == Logical COS Queue index for the rep to endpoint direction
147 #define BNXT_DEVARG_REP_Q_R2F_INVALID(rep_q_r2f) ((rep_q_r2f) > 3)
150 * rep_q_f2r == Logical COS Queue index for the endpoint to rep direction
152 #define BNXT_DEVARG_REP_Q_F2R_INVALID(rep_q_f2r) ((rep_q_f2r) > 3)
155 * rep_fc_r2f == Flow control for the representor to endpoint direction
157 #define BNXT_DEVARG_REP_FC_R2F_INVALID(rep_fc_r2f) ((rep_fc_r2f) > 1)
160 * rep_fc_f2r == Flow control for the endpoint to representor direction
162 #define BNXT_DEVARG_REP_FC_F2R_INVALID(rep_fc_f2r) ((rep_fc_f2r) > 1)
164 int bnxt_cfa_code_dynfield_offset = -1;
167 * max_num_kflows must be >= 32
168 * and must be a power-of-2 supported value
169 * return: 1 -> invalid
172 static int bnxt_devarg_max_num_kflow_invalid(uint16_t max_num_kflows)
174 if (max_num_kflows < 32 || !rte_is_power_of_2(max_num_kflows))
179 static int bnxt_vlan_offload_set_op(struct rte_eth_dev *dev, int mask);
180 static int bnxt_dev_uninit(struct rte_eth_dev *eth_dev);
181 static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev);
182 static int bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev);
183 static void bnxt_cancel_fw_health_check(struct bnxt *bp);
184 static int bnxt_restore_vlan_filters(struct bnxt *bp);
185 static void bnxt_dev_recover(void *arg);
186 static void bnxt_free_error_recovery_info(struct bnxt *bp);
187 static void bnxt_free_rep_info(struct bnxt *bp);
189 int is_bnxt_in_error(struct bnxt *bp)
191 if (bp->flags & BNXT_FLAG_FATAL_ERROR)
193 if (bp->flags & BNXT_FLAG_FW_RESET)
199 /***********************/
202 * High level utility functions
205 static uint16_t bnxt_rss_ctxts(const struct bnxt *bp)
207 unsigned int num_rss_rings = RTE_MIN(bp->rx_nr_rings,
208 BNXT_RSS_TBL_SIZE_P5);
210 if (!BNXT_CHIP_P5(bp))
213 return RTE_ALIGN_MUL_CEIL(num_rss_rings,
214 BNXT_RSS_ENTRIES_PER_CTX_P5) /
215 BNXT_RSS_ENTRIES_PER_CTX_P5;
218 uint16_t bnxt_rss_hash_tbl_size(const struct bnxt *bp)
220 if (!BNXT_CHIP_P5(bp))
221 return HW_HASH_INDEX_SIZE;
223 return bnxt_rss_ctxts(bp) * BNXT_RSS_ENTRIES_PER_CTX_P5;
226 static void bnxt_free_parent_info(struct bnxt *bp)
228 rte_free(bp->parent);
232 static void bnxt_free_pf_info(struct bnxt *bp)
238 static void bnxt_free_link_info(struct bnxt *bp)
240 rte_free(bp->link_info);
241 bp->link_info = NULL;
244 static void bnxt_free_leds_info(struct bnxt *bp)
253 static void bnxt_free_flow_stats_info(struct bnxt *bp)
255 rte_free(bp->flow_stat);
256 bp->flow_stat = NULL;
259 static void bnxt_free_cos_queues(struct bnxt *bp)
261 rte_free(bp->rx_cos_queue);
262 bp->rx_cos_queue = NULL;
263 rte_free(bp->tx_cos_queue);
264 bp->tx_cos_queue = NULL;
267 static void bnxt_free_mem(struct bnxt *bp, bool reconfig)
269 bnxt_free_filter_mem(bp);
270 bnxt_free_vnic_attributes(bp);
271 bnxt_free_vnic_mem(bp);
273 /* tx/rx rings are configured as part of *_queue_setup callbacks.
274 * If the number of rings change across fw update,
275 * we don't have much choice except to warn the user.
279 bnxt_free_tx_rings(bp);
280 bnxt_free_rx_rings(bp);
282 bnxt_free_async_cp_ring(bp);
283 bnxt_free_rxtx_nq_ring(bp);
285 rte_free(bp->grp_info);
289 static int bnxt_alloc_parent_info(struct bnxt *bp)
291 bp->parent = rte_zmalloc("bnxt_parent_info",
292 sizeof(struct bnxt_parent_info), 0);
293 if (bp->parent == NULL)
299 static int bnxt_alloc_pf_info(struct bnxt *bp)
301 bp->pf = rte_zmalloc("bnxt_pf_info", sizeof(struct bnxt_pf_info), 0);
308 static int bnxt_alloc_link_info(struct bnxt *bp)
311 rte_zmalloc("bnxt_link_info", sizeof(struct bnxt_link_info), 0);
312 if (bp->link_info == NULL)
318 static int bnxt_alloc_leds_info(struct bnxt *bp)
323 bp->leds = rte_zmalloc("bnxt_leds",
324 BNXT_MAX_LED * sizeof(struct bnxt_led_info),
326 if (bp->leds == NULL)
332 static int bnxt_alloc_cos_queues(struct bnxt *bp)
335 rte_zmalloc("bnxt_rx_cosq",
336 BNXT_COS_QUEUE_COUNT *
337 sizeof(struct bnxt_cos_queue_info),
339 if (bp->rx_cos_queue == NULL)
343 rte_zmalloc("bnxt_tx_cosq",
344 BNXT_COS_QUEUE_COUNT *
345 sizeof(struct bnxt_cos_queue_info),
347 if (bp->tx_cos_queue == NULL)
353 static int bnxt_alloc_flow_stats_info(struct bnxt *bp)
355 bp->flow_stat = rte_zmalloc("bnxt_flow_xstat",
356 sizeof(struct bnxt_flow_stat_info), 0);
357 if (bp->flow_stat == NULL)
363 static int bnxt_alloc_mem(struct bnxt *bp, bool reconfig)
367 rc = bnxt_alloc_ring_grps(bp);
371 rc = bnxt_alloc_async_ring_struct(bp);
375 rc = bnxt_alloc_vnic_mem(bp);
379 rc = bnxt_alloc_vnic_attributes(bp);
383 rc = bnxt_alloc_filter_mem(bp);
387 rc = bnxt_alloc_async_cp_ring(bp);
391 rc = bnxt_alloc_rxtx_nq_ring(bp);
395 if (BNXT_FLOW_XSTATS_EN(bp)) {
396 rc = bnxt_alloc_flow_stats_info(bp);
404 bnxt_free_mem(bp, reconfig);
408 static int bnxt_setup_one_vnic(struct bnxt *bp, uint16_t vnic_id)
410 struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
411 struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
412 uint64_t rx_offloads = dev_conf->rxmode.offloads;
413 struct bnxt_rx_queue *rxq;
417 rc = bnxt_vnic_grp_alloc(bp, vnic);
421 PMD_DRV_LOG(DEBUG, "vnic[%d] = %p vnic->fw_grp_ids = %p\n",
422 vnic_id, vnic, vnic->fw_grp_ids);
424 rc = bnxt_hwrm_vnic_alloc(bp, vnic);
428 /* Alloc RSS context only if RSS mode is enabled */
429 if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS) {
430 int j, nr_ctxs = bnxt_rss_ctxts(bp);
432 /* RSS table size in Thor is 512.
433 * Cap max Rx rings to same value
435 if (bp->rx_nr_rings > BNXT_RSS_TBL_SIZE_P5) {
436 PMD_DRV_LOG(ERR, "RxQ cnt %d > reta_size %d\n",
437 bp->rx_nr_rings, BNXT_RSS_TBL_SIZE_P5);
442 for (j = 0; j < nr_ctxs; j++) {
443 rc = bnxt_hwrm_vnic_ctx_alloc(bp, vnic, j);
449 "HWRM vnic %d ctx %d alloc failure rc: %x\n",
453 vnic->num_lb_ctxts = nr_ctxs;
457 * Firmware sets pf pair in default vnic cfg. If the VLAN strip
458 * setting is not available at this time, it will not be
459 * configured correctly in the CFA.
461 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
462 vnic->vlan_strip = true;
464 vnic->vlan_strip = false;
466 rc = bnxt_hwrm_vnic_cfg(bp, vnic);
470 rc = bnxt_set_hwrm_vnic_filters(bp, vnic);
474 for (j = 0; j < bp->rx_num_qs_per_vnic; j++) {
475 rxq = bp->eth_dev->data->rx_queues[j];
478 "rxq[%d]->vnic=%p vnic->fw_grp_ids=%p\n",
479 j, rxq->vnic, rxq->vnic->fw_grp_ids);
481 if (BNXT_HAS_RING_GRPS(bp) && rxq->rx_deferred_start)
482 rxq->vnic->fw_grp_ids[j] = INVALID_HW_RING_ID;
484 vnic->rx_queue_cnt++;
487 PMD_DRV_LOG(DEBUG, "vnic->rx_queue_cnt = %d\n", vnic->rx_queue_cnt);
489 rc = bnxt_vnic_rss_configure(bp, vnic);
493 bnxt_hwrm_vnic_plcmode_cfg(bp, vnic);
495 rc = bnxt_hwrm_vnic_tpa_cfg(bp, vnic,
496 (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) ?
503 PMD_DRV_LOG(ERR, "HWRM vnic %d cfg failure rc: %x\n",
508 static int bnxt_register_fc_ctx_mem(struct bnxt *bp)
512 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->rx_fc_in_tbl.dma,
513 &bp->flow_stat->rx_fc_in_tbl.ctx_id);
518 "rx_fc_in_tbl.va = %p rx_fc_in_tbl.dma = %p"
519 " rx_fc_in_tbl.ctx_id = %d\n",
520 bp->flow_stat->rx_fc_in_tbl.va,
521 (void *)((uintptr_t)bp->flow_stat->rx_fc_in_tbl.dma),
522 bp->flow_stat->rx_fc_in_tbl.ctx_id);
524 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->rx_fc_out_tbl.dma,
525 &bp->flow_stat->rx_fc_out_tbl.ctx_id);
530 "rx_fc_out_tbl.va = %p rx_fc_out_tbl.dma = %p"
531 " rx_fc_out_tbl.ctx_id = %d\n",
532 bp->flow_stat->rx_fc_out_tbl.va,
533 (void *)((uintptr_t)bp->flow_stat->rx_fc_out_tbl.dma),
534 bp->flow_stat->rx_fc_out_tbl.ctx_id);
536 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->tx_fc_in_tbl.dma,
537 &bp->flow_stat->tx_fc_in_tbl.ctx_id);
542 "tx_fc_in_tbl.va = %p tx_fc_in_tbl.dma = %p"
543 " tx_fc_in_tbl.ctx_id = %d\n",
544 bp->flow_stat->tx_fc_in_tbl.va,
545 (void *)((uintptr_t)bp->flow_stat->tx_fc_in_tbl.dma),
546 bp->flow_stat->tx_fc_in_tbl.ctx_id);
548 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->tx_fc_out_tbl.dma,
549 &bp->flow_stat->tx_fc_out_tbl.ctx_id);
554 "tx_fc_out_tbl.va = %p tx_fc_out_tbl.dma = %p"
555 " tx_fc_out_tbl.ctx_id = %d\n",
556 bp->flow_stat->tx_fc_out_tbl.va,
557 (void *)((uintptr_t)bp->flow_stat->tx_fc_out_tbl.dma),
558 bp->flow_stat->tx_fc_out_tbl.ctx_id);
560 memset(bp->flow_stat->rx_fc_out_tbl.va,
562 bp->flow_stat->rx_fc_out_tbl.size);
563 rc = bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_RX,
564 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
565 bp->flow_stat->rx_fc_out_tbl.ctx_id,
566 bp->flow_stat->max_fc,
571 memset(bp->flow_stat->tx_fc_out_tbl.va,
573 bp->flow_stat->tx_fc_out_tbl.size);
574 rc = bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_TX,
575 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
576 bp->flow_stat->tx_fc_out_tbl.ctx_id,
577 bp->flow_stat->max_fc,
583 static int bnxt_alloc_ctx_mem_buf(char *type, size_t size,
584 struct bnxt_ctx_mem_buf_info *ctx)
589 ctx->va = rte_zmalloc(type, size, 0);
592 rte_mem_lock_page(ctx->va);
594 ctx->dma = rte_mem_virt2iova(ctx->va);
595 if (ctx->dma == RTE_BAD_IOVA)
601 static int bnxt_init_fc_ctx_mem(struct bnxt *bp)
603 struct rte_pci_device *pdev = bp->pdev;
604 char type[RTE_MEMZONE_NAMESIZE];
608 max_fc = bp->flow_stat->max_fc;
610 sprintf(type, "bnxt_rx_fc_in_" PCI_PRI_FMT, pdev->addr.domain,
611 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
612 /* 4 bytes for each counter-id */
613 rc = bnxt_alloc_ctx_mem_buf(type,
615 &bp->flow_stat->rx_fc_in_tbl);
619 sprintf(type, "bnxt_rx_fc_out_" PCI_PRI_FMT, pdev->addr.domain,
620 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
621 /* 16 bytes for each counter - 8 bytes pkt_count, 8 bytes byte_count */
622 rc = bnxt_alloc_ctx_mem_buf(type,
624 &bp->flow_stat->rx_fc_out_tbl);
628 sprintf(type, "bnxt_tx_fc_in_" PCI_PRI_FMT, pdev->addr.domain,
629 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
630 /* 4 bytes for each counter-id */
631 rc = bnxt_alloc_ctx_mem_buf(type,
633 &bp->flow_stat->tx_fc_in_tbl);
637 sprintf(type, "bnxt_tx_fc_out_" PCI_PRI_FMT, pdev->addr.domain,
638 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
639 /* 16 bytes for each counter - 8 bytes pkt_count, 8 bytes byte_count */
640 rc = bnxt_alloc_ctx_mem_buf(type,
642 &bp->flow_stat->tx_fc_out_tbl);
646 rc = bnxt_register_fc_ctx_mem(bp);
651 static int bnxt_init_ctx_mem(struct bnxt *bp)
655 if (!(bp->fw_cap & BNXT_FW_CAP_ADV_FLOW_COUNTERS) ||
656 !(BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) ||
657 !BNXT_FLOW_XSTATS_EN(bp))
660 rc = bnxt_hwrm_cfa_counter_qcaps(bp, &bp->flow_stat->max_fc);
664 rc = bnxt_init_fc_ctx_mem(bp);
669 static int bnxt_update_phy_setting(struct bnxt *bp)
671 struct rte_eth_link new;
674 rc = bnxt_get_hwrm_link_config(bp, &new);
676 PMD_DRV_LOG(ERR, "Failed to get link settings\n");
681 * On BCM957508-N2100 adapters, FW will not allow any user other
682 * than BMC to shutdown the port. bnxt_get_hwrm_link_config() call
683 * always returns link up. Force phy update always in that case.
685 if (!new.link_status || IS_BNXT_DEV_957508_N2100(bp)) {
686 rc = bnxt_set_hwrm_link_config(bp, true);
688 PMD_DRV_LOG(ERR, "Failed to update PHY settings\n");
696 static void bnxt_free_prev_ring_stats(struct bnxt *bp)
698 rte_free(bp->prev_rx_ring_stats);
699 rte_free(bp->prev_tx_ring_stats);
701 bp->prev_rx_ring_stats = NULL;
702 bp->prev_tx_ring_stats = NULL;
705 static int bnxt_alloc_prev_ring_stats(struct bnxt *bp)
707 bp->prev_rx_ring_stats = rte_zmalloc("bnxt_prev_rx_ring_stats",
708 sizeof(struct bnxt_ring_stats) *
711 if (bp->prev_rx_ring_stats == NULL)
714 bp->prev_tx_ring_stats = rte_zmalloc("bnxt_prev_tx_ring_stats",
715 sizeof(struct bnxt_ring_stats) *
718 if (bp->prev_tx_ring_stats == NULL)
724 bnxt_free_prev_ring_stats(bp);
728 static int bnxt_start_nic(struct bnxt *bp)
730 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(bp->eth_dev);
731 struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
732 uint32_t intr_vector = 0;
733 uint32_t queue_id, base = BNXT_MISC_VEC_ID;
734 uint32_t vec = BNXT_MISC_VEC_ID;
738 if (bp->eth_dev->data->mtu > RTE_ETHER_MTU) {
739 bp->eth_dev->data->dev_conf.rxmode.offloads |=
740 DEV_RX_OFFLOAD_JUMBO_FRAME;
741 bp->flags |= BNXT_FLAG_JUMBO;
743 bp->eth_dev->data->dev_conf.rxmode.offloads &=
744 ~DEV_RX_OFFLOAD_JUMBO_FRAME;
745 bp->flags &= ~BNXT_FLAG_JUMBO;
748 /* THOR does not support ring groups.
749 * But we will use the array to save RSS context IDs.
751 if (BNXT_CHIP_P5(bp))
752 bp->max_ring_grps = BNXT_MAX_RSS_CTXTS_P5;
754 rc = bnxt_alloc_all_hwrm_stat_ctxs(bp);
756 PMD_DRV_LOG(ERR, "HWRM stat ctx alloc failure rc: %x\n", rc);
760 rc = bnxt_alloc_hwrm_rings(bp);
762 PMD_DRV_LOG(ERR, "HWRM ring alloc failure rc: %x\n", rc);
766 rc = bnxt_alloc_all_hwrm_ring_grps(bp);
768 PMD_DRV_LOG(ERR, "HWRM ring grp alloc failure: %x\n", rc);
772 if (!(bp->vnic_cap_flags & BNXT_VNIC_CAP_COS_CLASSIFY))
775 for (j = 0, i = 0; i < BNXT_COS_QUEUE_COUNT; i++) {
776 if (bp->rx_cos_queue[i].id != 0xff) {
777 struct bnxt_vnic_info *vnic = &bp->vnic_info[j++];
781 "Num pools more than FW profile\n");
785 vnic->cos_queue_id = bp->rx_cos_queue[i].id;
791 rc = bnxt_mq_rx_configure(bp);
793 PMD_DRV_LOG(ERR, "MQ mode configure failure rc: %x\n", rc);
798 rc = bnxt_setup_one_vnic(bp, 0);
801 /* VNIC configuration */
802 if (BNXT_RFS_NEEDS_VNIC(bp)) {
803 for (i = 1; i < bp->nr_vnics; i++) {
804 rc = bnxt_setup_one_vnic(bp, i);
810 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, &bp->vnic_info[0], 0, NULL);
813 "HWRM cfa l2 rx mask failure rc: %x\n", rc);
817 /* check and configure queue intr-vector mapping */
818 if ((rte_intr_cap_multiple(intr_handle) ||
819 !RTE_ETH_DEV_SRIOV(bp->eth_dev).active) &&
820 bp->eth_dev->data->dev_conf.intr_conf.rxq != 0) {
821 intr_vector = bp->eth_dev->data->nb_rx_queues;
822 PMD_DRV_LOG(DEBUG, "intr_vector = %d\n", intr_vector);
823 if (intr_vector > bp->rx_cp_nr_rings) {
824 PMD_DRV_LOG(ERR, "At most %d intr queues supported",
828 rc = rte_intr_efd_enable(intr_handle, intr_vector);
833 if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
834 intr_handle->intr_vec =
835 rte_zmalloc("intr_vec",
836 bp->eth_dev->data->nb_rx_queues *
838 if (intr_handle->intr_vec == NULL) {
839 PMD_DRV_LOG(ERR, "Failed to allocate %d rx_queues"
840 " intr_vec", bp->eth_dev->data->nb_rx_queues);
844 PMD_DRV_LOG(DEBUG, "intr_handle->intr_vec = %p "
845 "intr_handle->nb_efd = %d intr_handle->max_intr = %d\n",
846 intr_handle->intr_vec, intr_handle->nb_efd,
847 intr_handle->max_intr);
848 for (queue_id = 0; queue_id < bp->eth_dev->data->nb_rx_queues;
850 intr_handle->intr_vec[queue_id] =
851 vec + BNXT_RX_VEC_START;
852 if (vec < base + intr_handle->nb_efd - 1)
857 /* enable uio/vfio intr/eventfd mapping */
858 rc = rte_intr_enable(intr_handle);
859 #ifndef RTE_EXEC_ENV_FREEBSD
860 /* In FreeBSD OS, nic_uio driver does not support interrupts */
865 rc = bnxt_update_phy_setting(bp);
869 bp->mark_table = rte_zmalloc("bnxt_mark_table", BNXT_MARK_TABLE_SZ, 0);
871 PMD_DRV_LOG(ERR, "Allocation of mark table failed\n");
876 /* Some of the error status returned by FW may not be from errno.h */
883 static int bnxt_shutdown_nic(struct bnxt *bp)
885 bnxt_free_all_hwrm_resources(bp);
886 bnxt_free_all_filters(bp);
887 bnxt_free_all_vnics(bp);
892 * Device configuration and status function
895 uint32_t bnxt_get_speed_capabilities(struct bnxt *bp)
897 uint32_t link_speed = 0;
898 uint32_t speed_capa = 0;
900 if (bp->link_info == NULL)
903 link_speed = bp->link_info->support_speeds;
905 /* If PAM4 is configured, use PAM4 supported speed */
906 if (link_speed == 0 && bp->link_info->support_pam4_speeds > 0)
907 link_speed = bp->link_info->support_pam4_speeds;
909 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_100MB)
910 speed_capa |= ETH_LINK_SPEED_100M;
911 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_100MBHD)
912 speed_capa |= ETH_LINK_SPEED_100M_HD;
913 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_1GB)
914 speed_capa |= ETH_LINK_SPEED_1G;
915 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_2_5GB)
916 speed_capa |= ETH_LINK_SPEED_2_5G;
917 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_10GB)
918 speed_capa |= ETH_LINK_SPEED_10G;
919 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_20GB)
920 speed_capa |= ETH_LINK_SPEED_20G;
921 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_25GB)
922 speed_capa |= ETH_LINK_SPEED_25G;
923 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_40GB)
924 speed_capa |= ETH_LINK_SPEED_40G;
925 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_50GB)
926 speed_capa |= ETH_LINK_SPEED_50G;
927 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_100GB)
928 speed_capa |= ETH_LINK_SPEED_100G;
929 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_50G)
930 speed_capa |= ETH_LINK_SPEED_50G;
931 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_100G)
932 speed_capa |= ETH_LINK_SPEED_100G;
933 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_200G)
934 speed_capa |= ETH_LINK_SPEED_200G;
936 if (bp->link_info->auto_mode ==
937 HWRM_PORT_PHY_QCFG_OUTPUT_AUTO_MODE_NONE)
938 speed_capa |= ETH_LINK_SPEED_FIXED;
943 static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
944 struct rte_eth_dev_info *dev_info)
946 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(eth_dev->device);
947 struct bnxt *bp = eth_dev->data->dev_private;
948 uint16_t max_vnics, i, j, vpool, vrxq;
949 unsigned int max_rx_rings;
952 rc = is_bnxt_in_error(bp);
957 dev_info->max_mac_addrs = bp->max_l2_ctx;
958 dev_info->max_hash_mac_addrs = 0;
960 /* PF/VF specifics */
962 dev_info->max_vfs = pdev->max_vfs;
964 max_rx_rings = bnxt_max_rings(bp);
965 /* For the sake of symmetry, max_rx_queues = max_tx_queues */
966 dev_info->max_rx_queues = max_rx_rings;
967 dev_info->max_tx_queues = max_rx_rings;
968 dev_info->reta_size = bnxt_rss_hash_tbl_size(bp);
969 dev_info->hash_key_size = 40;
970 max_vnics = bp->max_vnics;
973 dev_info->min_mtu = RTE_ETHER_MIN_MTU;
974 dev_info->max_mtu = BNXT_MAX_MTU;
976 /* Fast path specifics */
977 dev_info->min_rx_bufsize = 1;
978 dev_info->max_rx_pktlen = BNXT_MAX_PKT_LEN;
980 dev_info->rx_offload_capa = BNXT_DEV_RX_OFFLOAD_SUPPORT;
981 if (bp->flags & BNXT_FLAG_PTP_SUPPORTED)
982 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TIMESTAMP;
983 dev_info->tx_queue_offload_capa = DEV_TX_OFFLOAD_MBUF_FAST_FREE;
984 dev_info->tx_offload_capa = BNXT_DEV_TX_OFFLOAD_SUPPORT |
985 dev_info->tx_queue_offload_capa;
986 dev_info->flow_type_rss_offloads = BNXT_ETH_RSS_SUPPORT;
988 dev_info->speed_capa = bnxt_get_speed_capabilities(bp);
990 dev_info->default_rxconf = (struct rte_eth_rxconf) {
996 .rx_free_thresh = 32,
997 .rx_drop_en = BNXT_DEFAULT_RX_DROP_EN,
1000 dev_info->default_txconf = (struct rte_eth_txconf) {
1006 .tx_free_thresh = 32,
1009 eth_dev->data->dev_conf.intr_conf.lsc = 1;
1011 eth_dev->data->dev_conf.intr_conf.rxq = 1;
1012 dev_info->rx_desc_lim.nb_min = BNXT_MIN_RING_DESC;
1013 dev_info->rx_desc_lim.nb_max = BNXT_MAX_RX_RING_DESC;
1014 dev_info->tx_desc_lim.nb_min = BNXT_MIN_RING_DESC;
1015 dev_info->tx_desc_lim.nb_max = BNXT_MAX_TX_RING_DESC;
1017 if (BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) {
1018 dev_info->switch_info.name = eth_dev->device->name;
1019 dev_info->switch_info.domain_id = bp->switch_domain_id;
1020 dev_info->switch_info.port_id =
1021 BNXT_PF(bp) ? BNXT_SWITCH_PORT_ID_PF :
1022 BNXT_SWITCH_PORT_ID_TRUSTED_VF;
1026 * TODO: default_rxconf, default_txconf, rx_desc_lim, and tx_desc_lim
1027 * need further investigation.
1030 /* VMDq resources */
1031 vpool = 64; /* ETH_64_POOLS */
1032 vrxq = 128; /* ETH_VMDQ_DCB_NUM_QUEUES */
1033 for (i = 0; i < 4; vpool >>= 1, i++) {
1034 if (max_vnics > vpool) {
1035 for (j = 0; j < 5; vrxq >>= 1, j++) {
1036 if (dev_info->max_rx_queues > vrxq) {
1042 /* Not enough resources to support VMDq */
1046 /* Not enough resources to support VMDq */
1050 dev_info->max_vmdq_pools = vpool;
1051 dev_info->vmdq_queue_num = vrxq;
1053 dev_info->vmdq_pool_base = 0;
1054 dev_info->vmdq_queue_base = 0;
1059 /* Configure the device based on the configuration provided */
1060 static int bnxt_dev_configure_op(struct rte_eth_dev *eth_dev)
1062 struct bnxt *bp = eth_dev->data->dev_private;
1063 uint64_t rx_offloads = eth_dev->data->dev_conf.rxmode.offloads;
1066 bp->rx_queues = (void *)eth_dev->data->rx_queues;
1067 bp->tx_queues = (void *)eth_dev->data->tx_queues;
1068 bp->tx_nr_rings = eth_dev->data->nb_tx_queues;
1069 bp->rx_nr_rings = eth_dev->data->nb_rx_queues;
1071 rc = is_bnxt_in_error(bp);
1075 if (BNXT_VF(bp) && (bp->flags & BNXT_FLAG_NEW_RM)) {
1076 rc = bnxt_hwrm_check_vf_rings(bp);
1078 PMD_DRV_LOG(ERR, "HWRM insufficient resources\n");
1082 /* If a resource has already been allocated - in this case
1083 * it is the async completion ring, free it. Reallocate it after
1084 * resource reservation. This will ensure the resource counts
1085 * are calculated correctly.
1088 pthread_mutex_lock(&bp->def_cp_lock);
1090 if (!BNXT_HAS_NQ(bp) && bp->async_cp_ring) {
1091 bnxt_disable_int(bp);
1092 bnxt_free_cp_ring(bp, bp->async_cp_ring);
1095 rc = bnxt_hwrm_func_reserve_vf_resc(bp, false);
1097 PMD_DRV_LOG(ERR, "HWRM resource alloc fail:%x\n", rc);
1098 pthread_mutex_unlock(&bp->def_cp_lock);
1102 if (!BNXT_HAS_NQ(bp) && bp->async_cp_ring) {
1103 rc = bnxt_alloc_async_cp_ring(bp);
1105 pthread_mutex_unlock(&bp->def_cp_lock);
1108 bnxt_enable_int(bp);
1111 pthread_mutex_unlock(&bp->def_cp_lock);
1114 /* Inherit new configurations */
1115 if (eth_dev->data->nb_rx_queues > bp->max_rx_rings ||
1116 eth_dev->data->nb_tx_queues > bp->max_tx_rings ||
1117 eth_dev->data->nb_rx_queues + eth_dev->data->nb_tx_queues
1118 + BNXT_NUM_ASYNC_CPR(bp) > bp->max_cp_rings ||
1119 eth_dev->data->nb_rx_queues + eth_dev->data->nb_tx_queues >
1121 goto resource_error;
1123 if (BNXT_HAS_RING_GRPS(bp) &&
1124 (uint32_t)(eth_dev->data->nb_rx_queues) > bp->max_ring_grps)
1125 goto resource_error;
1127 if (!(eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) &&
1128 bp->max_vnics < eth_dev->data->nb_rx_queues)
1129 goto resource_error;
1131 bp->rx_cp_nr_rings = bp->rx_nr_rings;
1132 bp->tx_cp_nr_rings = bp->tx_nr_rings;
1134 if (eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
1135 rx_offloads |= DEV_RX_OFFLOAD_RSS_HASH;
1136 eth_dev->data->dev_conf.rxmode.offloads = rx_offloads;
1138 if (rx_offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) {
1139 eth_dev->data->mtu =
1140 eth_dev->data->dev_conf.rxmode.max_rx_pkt_len -
1141 RTE_ETHER_HDR_LEN - RTE_ETHER_CRC_LEN - VLAN_TAG_SIZE *
1143 bnxt_mtu_set_op(eth_dev, eth_dev->data->mtu);
1149 "Insufficient resources to support requested config\n");
1151 "Num Queues Requested: Tx %d, Rx %d\n",
1152 eth_dev->data->nb_tx_queues,
1153 eth_dev->data->nb_rx_queues);
1155 "MAX: TxQ %d, RxQ %d, CQ %d Stat %d, Grp %d, Vnic %d\n",
1156 bp->max_tx_rings, bp->max_rx_rings, bp->max_cp_rings,
1157 bp->max_stat_ctx, bp->max_ring_grps, bp->max_vnics);
1161 void bnxt_print_link_info(struct rte_eth_dev *eth_dev)
1163 struct rte_eth_link *link = ð_dev->data->dev_link;
1165 if (link->link_status)
1166 PMD_DRV_LOG(INFO, "Port %d Link Up - speed %u Mbps - %s\n",
1167 eth_dev->data->port_id,
1168 (uint32_t)link->link_speed,
1169 (link->link_duplex == ETH_LINK_FULL_DUPLEX) ?
1170 ("full-duplex") : ("half-duplex\n"));
1172 PMD_DRV_LOG(INFO, "Port %d Link Down\n",
1173 eth_dev->data->port_id);
1177 * Determine whether the current configuration requires support for scattered
1178 * receive; return 1 if scattered receive is required and 0 if not.
1180 static int bnxt_scattered_rx(struct rte_eth_dev *eth_dev)
1185 if (eth_dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_SCATTER)
1188 if (eth_dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_TCP_LRO)
1191 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
1192 struct bnxt_rx_queue *rxq = eth_dev->data->rx_queues[i];
1194 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
1195 RTE_PKTMBUF_HEADROOM);
1196 if (eth_dev->data->dev_conf.rxmode.max_rx_pkt_len > buf_size)
1202 static eth_rx_burst_t
1203 bnxt_receive_function(struct rte_eth_dev *eth_dev)
1205 struct bnxt *bp = eth_dev->data->dev_private;
1207 /* Disable vector mode RX for Stingray2 for now */
1208 if (BNXT_CHIP_SR2(bp)) {
1209 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1210 return bnxt_recv_pkts;
1213 #if (defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)) && \
1214 !defined(RTE_LIBRTE_IEEE1588)
1216 /* Vector mode receive cannot be enabled if scattered rx is in use. */
1217 if (eth_dev->data->scattered_rx)
1221 * Vector mode receive cannot be enabled if Truflow is enabled or if
1222 * asynchronous completions and receive completions can be placed in
1223 * the same completion ring.
1225 if (BNXT_TRUFLOW_EN(bp) || !BNXT_NUM_ASYNC_CPR(bp))
1229 * Vector mode receive cannot be enabled if any receive offloads outside
1230 * a limited subset have been enabled.
1232 if (eth_dev->data->dev_conf.rxmode.offloads &
1233 ~(DEV_RX_OFFLOAD_VLAN_STRIP |
1234 DEV_RX_OFFLOAD_KEEP_CRC |
1235 DEV_RX_OFFLOAD_JUMBO_FRAME |
1236 DEV_RX_OFFLOAD_IPV4_CKSUM |
1237 DEV_RX_OFFLOAD_UDP_CKSUM |
1238 DEV_RX_OFFLOAD_TCP_CKSUM |
1239 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1240 DEV_RX_OFFLOAD_OUTER_UDP_CKSUM |
1241 DEV_RX_OFFLOAD_RSS_HASH |
1242 DEV_RX_OFFLOAD_VLAN_FILTER))
1245 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
1246 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
1247 rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1) {
1249 "Using AVX2 vector mode receive for port %d\n",
1250 eth_dev->data->port_id);
1251 bp->flags |= BNXT_FLAG_RX_VECTOR_PKT_MODE;
1252 return bnxt_recv_pkts_vec_avx2;
1255 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_128) {
1257 "Using SSE vector mode receive for port %d\n",
1258 eth_dev->data->port_id);
1259 bp->flags |= BNXT_FLAG_RX_VECTOR_PKT_MODE;
1260 return bnxt_recv_pkts_vec;
1264 PMD_DRV_LOG(INFO, "Vector mode receive disabled for port %d\n",
1265 eth_dev->data->port_id);
1267 "Port %d scatter: %d rx offload: %" PRIX64 "\n",
1268 eth_dev->data->port_id,
1269 eth_dev->data->scattered_rx,
1270 eth_dev->data->dev_conf.rxmode.offloads);
1272 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1273 return bnxt_recv_pkts;
1276 static eth_tx_burst_t
1277 bnxt_transmit_function(struct rte_eth_dev *eth_dev)
1279 struct bnxt *bp = eth_dev->data->dev_private;
1281 /* Disable vector mode TX for Stingray2 for now */
1282 if (BNXT_CHIP_SR2(bp))
1283 return bnxt_xmit_pkts;
1285 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64) && \
1286 !defined(RTE_LIBRTE_IEEE1588)
1287 uint64_t offloads = eth_dev->data->dev_conf.txmode.offloads;
1290 * Vector mode transmit can be enabled only if not using scatter rx
1293 if (eth_dev->data->scattered_rx ||
1294 (offloads & ~DEV_TX_OFFLOAD_MBUF_FAST_FREE) ||
1295 BNXT_TRUFLOW_EN(bp))
1298 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
1299 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
1300 rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1) {
1302 "Using AVX2 vector mode transmit for port %d\n",
1303 eth_dev->data->port_id);
1304 return bnxt_xmit_pkts_vec_avx2;
1307 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_128) {
1309 "Using SSE vector mode transmit for port %d\n",
1310 eth_dev->data->port_id);
1311 return bnxt_xmit_pkts_vec;
1315 PMD_DRV_LOG(INFO, "Vector mode transmit disabled for port %d\n",
1316 eth_dev->data->port_id);
1318 "Port %d scatter: %d tx offload: %" PRIX64 "\n",
1319 eth_dev->data->port_id,
1320 eth_dev->data->scattered_rx,
1323 return bnxt_xmit_pkts;
1326 static int bnxt_handle_if_change_status(struct bnxt *bp)
1330 /* Since fw has undergone a reset and lost all contexts,
1331 * set fatal flag to not issue hwrm during cleanup
1333 bp->flags |= BNXT_FLAG_FATAL_ERROR;
1334 bnxt_uninit_resources(bp, true);
1336 /* clear fatal flag so that re-init happens */
1337 bp->flags &= ~BNXT_FLAG_FATAL_ERROR;
1338 rc = bnxt_init_resources(bp, true);
1340 bp->flags &= ~BNXT_FLAG_IF_CHANGE_HOT_FW_RESET_DONE;
1345 static int bnxt_dev_set_link_up_op(struct rte_eth_dev *eth_dev)
1347 struct bnxt *bp = eth_dev->data->dev_private;
1350 if (!BNXT_SINGLE_PF(bp))
1353 if (!bp->link_info->link_up)
1354 rc = bnxt_set_hwrm_link_config(bp, true);
1356 eth_dev->data->dev_link.link_status = 1;
1358 bnxt_print_link_info(eth_dev);
1362 static int bnxt_dev_set_link_down_op(struct rte_eth_dev *eth_dev)
1364 struct bnxt *bp = eth_dev->data->dev_private;
1366 if (!BNXT_SINGLE_PF(bp))
1369 eth_dev->data->dev_link.link_status = 0;
1370 bnxt_set_hwrm_link_config(bp, false);
1371 bp->link_info->link_up = 0;
1376 static void bnxt_free_switch_domain(struct bnxt *bp)
1380 if (!(BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)))
1383 rc = rte_eth_switch_domain_free(bp->switch_domain_id);
1385 PMD_DRV_LOG(ERR, "free switch domain:%d fail: %d\n",
1386 bp->switch_domain_id, rc);
1389 static void bnxt_ptp_get_current_time(void *arg)
1391 struct bnxt *bp = arg;
1392 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
1395 rc = is_bnxt_in_error(bp);
1402 bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
1403 &ptp->current_time);
1405 rc = rte_eal_alarm_set(US_PER_S, bnxt_ptp_get_current_time, (void *)bp);
1407 PMD_DRV_LOG(ERR, "Failed to re-schedule PTP alarm\n");
1408 bp->flags2 &= ~BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1412 static int bnxt_schedule_ptp_alarm(struct bnxt *bp)
1414 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
1417 if (bp->flags2 & BNXT_FLAGS2_PTP_ALARM_SCHEDULED)
1420 bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
1421 &ptp->current_time);
1423 rc = rte_eal_alarm_set(US_PER_S, bnxt_ptp_get_current_time, (void *)bp);
1427 static void bnxt_cancel_ptp_alarm(struct bnxt *bp)
1429 if (bp->flags2 & BNXT_FLAGS2_PTP_ALARM_SCHEDULED) {
1430 rte_eal_alarm_cancel(bnxt_ptp_get_current_time, (void *)bp);
1431 bp->flags2 &= ~BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1435 static void bnxt_ptp_stop(struct bnxt *bp)
1437 bnxt_cancel_ptp_alarm(bp);
1438 bp->flags2 &= ~BNXT_FLAGS2_PTP_TIMESYNC_ENABLED;
1441 static int bnxt_ptp_start(struct bnxt *bp)
1445 rc = bnxt_schedule_ptp_alarm(bp);
1447 PMD_DRV_LOG(ERR, "Failed to schedule PTP alarm\n");
1449 bp->flags2 |= BNXT_FLAGS2_PTP_TIMESYNC_ENABLED;
1450 bp->flags2 |= BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1456 static int bnxt_dev_stop(struct rte_eth_dev *eth_dev)
1458 struct bnxt *bp = eth_dev->data->dev_private;
1459 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1460 struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1461 struct rte_eth_link link;
1464 eth_dev->data->dev_started = 0;
1465 eth_dev->data->scattered_rx = 0;
1467 /* Prevent crashes when queues are still in use */
1468 eth_dev->rx_pkt_burst = &bnxt_dummy_recv_pkts;
1469 eth_dev->tx_pkt_burst = &bnxt_dummy_xmit_pkts;
1471 bnxt_disable_int(bp);
1473 /* disable uio/vfio intr/eventfd mapping */
1474 rte_intr_disable(intr_handle);
1476 /* Stop the child representors for this device */
1477 ret = bnxt_rep_stop_all(bp);
1481 /* delete the bnxt ULP port details */
1482 bnxt_ulp_port_deinit(bp);
1484 bnxt_cancel_fw_health_check(bp);
1486 if (BNXT_P5_PTP_TIMESYNC_ENABLED(bp))
1487 bnxt_cancel_ptp_alarm(bp);
1489 /* Do not bring link down during reset recovery */
1490 if (!is_bnxt_in_error(bp)) {
1491 bnxt_dev_set_link_down_op(eth_dev);
1492 /* Wait for link to be reset */
1493 if (BNXT_SINGLE_PF(bp))
1495 /* clear the recorded link status */
1496 memset(&link, 0, sizeof(link));
1497 rte_eth_linkstatus_set(eth_dev, &link);
1500 /* Clean queue intr-vector mapping */
1501 rte_intr_efd_disable(intr_handle);
1502 if (intr_handle->intr_vec != NULL) {
1503 rte_free(intr_handle->intr_vec);
1504 intr_handle->intr_vec = NULL;
1507 bnxt_hwrm_port_clr_stats(bp);
1508 bnxt_free_tx_mbufs(bp);
1509 bnxt_free_rx_mbufs(bp);
1510 /* Process any remaining notifications in default completion queue */
1511 bnxt_int_handler(eth_dev);
1512 bnxt_shutdown_nic(bp);
1513 bnxt_hwrm_if_change(bp, false);
1515 bnxt_free_prev_ring_stats(bp);
1516 rte_free(bp->mark_table);
1517 bp->mark_table = NULL;
1519 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1520 bp->rx_cosq_cnt = 0;
1521 /* All filters are deleted on a port stop. */
1522 if (BNXT_FLOW_XSTATS_EN(bp))
1523 bp->flow_stat->flow_count = 0;
1528 /* Unload the driver, release resources */
1529 static int bnxt_dev_stop_op(struct rte_eth_dev *eth_dev)
1531 struct bnxt *bp = eth_dev->data->dev_private;
1533 pthread_mutex_lock(&bp->err_recovery_lock);
1534 if (bp->flags & BNXT_FLAG_FW_RESET) {
1536 "Adapter recovering from error..Please retry\n");
1537 pthread_mutex_unlock(&bp->err_recovery_lock);
1540 pthread_mutex_unlock(&bp->err_recovery_lock);
1542 return bnxt_dev_stop(eth_dev);
1545 static int bnxt_dev_start_op(struct rte_eth_dev *eth_dev)
1547 struct bnxt *bp = eth_dev->data->dev_private;
1548 uint64_t rx_offloads = eth_dev->data->dev_conf.rxmode.offloads;
1550 int rc, retry_cnt = BNXT_IF_CHANGE_RETRY_COUNT;
1552 if (!eth_dev->data->nb_tx_queues || !eth_dev->data->nb_rx_queues) {
1553 PMD_DRV_LOG(ERR, "Queues are not configured yet!\n");
1557 if (bp->rx_cp_nr_rings > RTE_ETHDEV_QUEUE_STAT_CNTRS)
1559 "RxQ cnt %d > RTE_ETHDEV_QUEUE_STAT_CNTRS %d\n",
1560 bp->rx_cp_nr_rings, RTE_ETHDEV_QUEUE_STAT_CNTRS);
1563 rc = bnxt_hwrm_if_change(bp, true);
1564 if (rc == 0 || rc != -EAGAIN)
1567 rte_delay_ms(BNXT_IF_CHANGE_RETRY_INTERVAL);
1568 } while (retry_cnt--);
1573 if (bp->flags & BNXT_FLAG_IF_CHANGE_HOT_FW_RESET_DONE) {
1574 rc = bnxt_handle_if_change_status(bp);
1579 bnxt_enable_int(bp);
1581 eth_dev->data->scattered_rx = bnxt_scattered_rx(eth_dev);
1583 rc = bnxt_start_nic(bp);
1587 rc = bnxt_alloc_prev_ring_stats(bp);
1591 eth_dev->data->dev_started = 1;
1593 bnxt_link_update_op(eth_dev, 1);
1595 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1596 vlan_mask |= ETH_VLAN_FILTER_MASK;
1597 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
1598 vlan_mask |= ETH_VLAN_STRIP_MASK;
1599 rc = bnxt_vlan_offload_set_op(eth_dev, vlan_mask);
1603 /* Initialize bnxt ULP port details */
1604 rc = bnxt_ulp_port_init(bp);
1608 eth_dev->rx_pkt_burst = bnxt_receive_function(eth_dev);
1609 eth_dev->tx_pkt_burst = bnxt_transmit_function(eth_dev);
1611 bnxt_schedule_fw_health_check(bp);
1613 if (BNXT_P5_PTP_TIMESYNC_ENABLED(bp))
1614 bnxt_schedule_ptp_alarm(bp);
1619 bnxt_dev_stop(eth_dev);
1624 bnxt_uninit_locks(struct bnxt *bp)
1626 pthread_mutex_destroy(&bp->flow_lock);
1627 pthread_mutex_destroy(&bp->def_cp_lock);
1628 pthread_mutex_destroy(&bp->health_check_lock);
1629 pthread_mutex_destroy(&bp->err_recovery_lock);
1631 pthread_mutex_destroy(&bp->rep_info->vfr_lock);
1632 pthread_mutex_destroy(&bp->rep_info->vfr_start_lock);
1636 static void bnxt_drv_uninit(struct bnxt *bp)
1638 bnxt_free_leds_info(bp);
1639 bnxt_free_cos_queues(bp);
1640 bnxt_free_link_info(bp);
1641 bnxt_free_parent_info(bp);
1642 bnxt_uninit_locks(bp);
1644 rte_memzone_free((const struct rte_memzone *)bp->tx_mem_zone);
1645 bp->tx_mem_zone = NULL;
1646 rte_memzone_free((const struct rte_memzone *)bp->rx_mem_zone);
1647 bp->rx_mem_zone = NULL;
1649 bnxt_free_vf_info(bp);
1650 bnxt_free_pf_info(bp);
1652 rte_free(bp->grp_info);
1653 bp->grp_info = NULL;
1656 static int bnxt_dev_close_op(struct rte_eth_dev *eth_dev)
1658 struct bnxt *bp = eth_dev->data->dev_private;
1661 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1664 pthread_mutex_lock(&bp->err_recovery_lock);
1665 if (bp->flags & BNXT_FLAG_FW_RESET) {
1667 "Adapter recovering from error...Please retry\n");
1668 pthread_mutex_unlock(&bp->err_recovery_lock);
1671 pthread_mutex_unlock(&bp->err_recovery_lock);
1673 /* cancel the recovery handler before remove dev */
1674 rte_eal_alarm_cancel(bnxt_dev_reset_and_resume, (void *)bp);
1675 rte_eal_alarm_cancel(bnxt_dev_recover, (void *)bp);
1676 bnxt_cancel_fc_thread(bp);
1678 if (eth_dev->data->dev_started)
1679 ret = bnxt_dev_stop(eth_dev);
1681 bnxt_uninit_resources(bp, false);
1683 bnxt_drv_uninit(bp);
1688 static void bnxt_mac_addr_remove_op(struct rte_eth_dev *eth_dev,
1691 struct bnxt *bp = eth_dev->data->dev_private;
1692 uint64_t pool_mask = eth_dev->data->mac_pool_sel[index];
1693 struct bnxt_vnic_info *vnic;
1694 struct bnxt_filter_info *filter, *temp_filter;
1697 if (is_bnxt_in_error(bp))
1701 * Loop through all VNICs from the specified filter flow pools to
1702 * remove the corresponding MAC addr filter
1704 for (i = 0; i < bp->nr_vnics; i++) {
1705 if (!(pool_mask & (1ULL << i)))
1708 vnic = &bp->vnic_info[i];
1709 filter = STAILQ_FIRST(&vnic->filter);
1711 temp_filter = STAILQ_NEXT(filter, next);
1712 if (filter->mac_index == index) {
1713 STAILQ_REMOVE(&vnic->filter, filter,
1714 bnxt_filter_info, next);
1715 bnxt_hwrm_clear_l2_filter(bp, filter);
1716 bnxt_free_filter(bp, filter);
1718 filter = temp_filter;
1723 static int bnxt_add_mac_filter(struct bnxt *bp, struct bnxt_vnic_info *vnic,
1724 struct rte_ether_addr *mac_addr, uint32_t index,
1727 struct bnxt_filter_info *filter;
1730 /* Attach requested MAC address to the new l2_filter */
1731 STAILQ_FOREACH(filter, &vnic->filter, next) {
1732 if (filter->mac_index == index) {
1734 "MAC addr already existed for pool %d\n",
1740 filter = bnxt_alloc_filter(bp);
1742 PMD_DRV_LOG(ERR, "L2 filter alloc failed\n");
1746 /* bnxt_alloc_filter copies default MAC to filter->l2_addr. So,
1747 * if the MAC that's been programmed now is a different one, then,
1748 * copy that addr to filter->l2_addr
1751 memcpy(filter->l2_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1752 filter->flags |= HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
1754 rc = bnxt_hwrm_set_l2_filter(bp, vnic->fw_vnic_id, filter);
1756 filter->mac_index = index;
1757 if (filter->mac_index == 0)
1758 STAILQ_INSERT_HEAD(&vnic->filter, filter, next);
1760 STAILQ_INSERT_TAIL(&vnic->filter, filter, next);
1762 bnxt_free_filter(bp, filter);
1768 static int bnxt_mac_addr_add_op(struct rte_eth_dev *eth_dev,
1769 struct rte_ether_addr *mac_addr,
1770 uint32_t index, uint32_t pool)
1772 struct bnxt *bp = eth_dev->data->dev_private;
1773 struct bnxt_vnic_info *vnic = &bp->vnic_info[pool];
1776 rc = is_bnxt_in_error(bp);
1780 if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp)) {
1781 PMD_DRV_LOG(ERR, "Cannot add MAC address to a VF interface\n");
1786 PMD_DRV_LOG(ERR, "VNIC not found for pool %d!\n", pool);
1790 /* Filter settings will get applied when port is started */
1791 if (!eth_dev->data->dev_started)
1794 rc = bnxt_add_mac_filter(bp, vnic, mac_addr, index, pool);
1799 int bnxt_link_update_op(struct rte_eth_dev *eth_dev, int wait_to_complete)
1802 struct bnxt *bp = eth_dev->data->dev_private;
1803 struct rte_eth_link new;
1804 int cnt = wait_to_complete ? BNXT_MAX_LINK_WAIT_CNT :
1805 BNXT_MIN_LINK_WAIT_CNT;
1807 rc = is_bnxt_in_error(bp);
1811 memset(&new, 0, sizeof(new));
1813 if (bp->link_info == NULL)
1817 /* Retrieve link info from hardware */
1818 rc = bnxt_get_hwrm_link_config(bp, &new);
1820 new.link_speed = ETH_LINK_SPEED_100M;
1821 new.link_duplex = ETH_LINK_FULL_DUPLEX;
1823 "Failed to retrieve link rc = 0x%x!\n", rc);
1827 if (!wait_to_complete || new.link_status)
1830 rte_delay_ms(BNXT_LINK_WAIT_INTERVAL);
1833 /* Only single function PF can bring phy down.
1834 * When port is stopped, report link down for VF/MH/NPAR functions.
1836 if (!BNXT_SINGLE_PF(bp) && !eth_dev->data->dev_started)
1837 memset(&new, 0, sizeof(new));
1840 /* Timed out or success */
1841 if (new.link_status != eth_dev->data->dev_link.link_status ||
1842 new.link_speed != eth_dev->data->dev_link.link_speed) {
1843 rte_eth_linkstatus_set(eth_dev, &new);
1845 rte_eth_dev_callback_process(eth_dev,
1846 RTE_ETH_EVENT_INTR_LSC,
1849 bnxt_print_link_info(eth_dev);
1855 static int bnxt_promiscuous_enable_op(struct rte_eth_dev *eth_dev)
1857 struct bnxt *bp = eth_dev->data->dev_private;
1858 struct bnxt_vnic_info *vnic;
1862 rc = is_bnxt_in_error(bp);
1866 /* Filter settings will get applied when port is started */
1867 if (!eth_dev->data->dev_started)
1870 if (bp->vnic_info == NULL)
1873 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1875 old_flags = vnic->flags;
1876 vnic->flags |= BNXT_VNIC_INFO_PROMISC;
1877 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1879 vnic->flags = old_flags;
1884 static int bnxt_promiscuous_disable_op(struct rte_eth_dev *eth_dev)
1886 struct bnxt *bp = eth_dev->data->dev_private;
1887 struct bnxt_vnic_info *vnic;
1891 rc = is_bnxt_in_error(bp);
1895 /* Filter settings will get applied when port is started */
1896 if (!eth_dev->data->dev_started)
1899 if (bp->vnic_info == NULL)
1902 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1904 old_flags = vnic->flags;
1905 vnic->flags &= ~BNXT_VNIC_INFO_PROMISC;
1906 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1908 vnic->flags = old_flags;
1913 static int bnxt_allmulticast_enable_op(struct rte_eth_dev *eth_dev)
1915 struct bnxt *bp = eth_dev->data->dev_private;
1916 struct bnxt_vnic_info *vnic;
1920 rc = is_bnxt_in_error(bp);
1924 /* Filter settings will get applied when port is started */
1925 if (!eth_dev->data->dev_started)
1928 if (bp->vnic_info == NULL)
1931 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1933 old_flags = vnic->flags;
1934 vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
1935 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1937 vnic->flags = old_flags;
1942 static int bnxt_allmulticast_disable_op(struct rte_eth_dev *eth_dev)
1944 struct bnxt *bp = eth_dev->data->dev_private;
1945 struct bnxt_vnic_info *vnic;
1949 rc = is_bnxt_in_error(bp);
1953 /* Filter settings will get applied when port is started */
1954 if (!eth_dev->data->dev_started)
1957 if (bp->vnic_info == NULL)
1960 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1962 old_flags = vnic->flags;
1963 vnic->flags &= ~BNXT_VNIC_INFO_ALLMULTI;
1964 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1966 vnic->flags = old_flags;
1971 /* Return bnxt_rx_queue pointer corresponding to a given rxq. */
1972 static struct bnxt_rx_queue *bnxt_qid_to_rxq(struct bnxt *bp, uint16_t qid)
1974 if (qid >= bp->rx_nr_rings)
1977 return bp->eth_dev->data->rx_queues[qid];
1980 /* Return rxq corresponding to a given rss table ring/group ID. */
1981 static uint16_t bnxt_rss_to_qid(struct bnxt *bp, uint16_t fwr)
1983 struct bnxt_rx_queue *rxq;
1986 if (!BNXT_HAS_RING_GRPS(bp)) {
1987 for (i = 0; i < bp->rx_nr_rings; i++) {
1988 rxq = bp->eth_dev->data->rx_queues[i];
1989 if (rxq->rx_ring->rx_ring_struct->fw_ring_id == fwr)
1993 for (i = 0; i < bp->rx_nr_rings; i++) {
1994 if (bp->grp_info[i].fw_grp_id == fwr)
1999 return INVALID_HW_RING_ID;
2002 static int bnxt_reta_update_op(struct rte_eth_dev *eth_dev,
2003 struct rte_eth_rss_reta_entry64 *reta_conf,
2006 struct bnxt *bp = eth_dev->data->dev_private;
2007 struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
2008 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2009 uint16_t tbl_size = bnxt_rss_hash_tbl_size(bp);
2013 rc = is_bnxt_in_error(bp);
2017 if (!vnic->rss_table)
2020 if (!(dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG))
2023 if (reta_size != tbl_size) {
2024 PMD_DRV_LOG(ERR, "The configured hash table lookup size "
2025 "(%d) must equal the size supported by the hardware "
2026 "(%d)\n", reta_size, tbl_size);
2030 for (i = 0; i < reta_size; i++) {
2031 struct bnxt_rx_queue *rxq;
2033 idx = i / RTE_RETA_GROUP_SIZE;
2034 sft = i % RTE_RETA_GROUP_SIZE;
2036 if (!(reta_conf[idx].mask & (1ULL << sft)))
2039 rxq = bnxt_qid_to_rxq(bp, reta_conf[idx].reta[sft]);
2041 PMD_DRV_LOG(ERR, "Invalid ring in reta_conf.\n");
2045 if (BNXT_CHIP_P5(bp)) {
2046 vnic->rss_table[i * 2] =
2047 rxq->rx_ring->rx_ring_struct->fw_ring_id;
2048 vnic->rss_table[i * 2 + 1] =
2049 rxq->cp_ring->cp_ring_struct->fw_ring_id;
2051 vnic->rss_table[i] =
2052 vnic->fw_grp_ids[reta_conf[idx].reta[sft]];
2056 rc = bnxt_hwrm_vnic_rss_cfg(bp, vnic);
2060 static int bnxt_reta_query_op(struct rte_eth_dev *eth_dev,
2061 struct rte_eth_rss_reta_entry64 *reta_conf,
2064 struct bnxt *bp = eth_dev->data->dev_private;
2065 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2066 uint16_t tbl_size = bnxt_rss_hash_tbl_size(bp);
2067 uint16_t idx, sft, i;
2070 rc = is_bnxt_in_error(bp);
2074 /* Retrieve from the default VNIC */
2077 if (!vnic->rss_table)
2080 if (reta_size != tbl_size) {
2081 PMD_DRV_LOG(ERR, "The configured hash table lookup size "
2082 "(%d) must equal the size supported by the hardware "
2083 "(%d)\n", reta_size, tbl_size);
2087 for (idx = 0, i = 0; i < reta_size; i++) {
2088 idx = i / RTE_RETA_GROUP_SIZE;
2089 sft = i % RTE_RETA_GROUP_SIZE;
2091 if (reta_conf[idx].mask & (1ULL << sft)) {
2094 if (BNXT_CHIP_P5(bp))
2095 qid = bnxt_rss_to_qid(bp,
2096 vnic->rss_table[i * 2]);
2098 qid = bnxt_rss_to_qid(bp, vnic->rss_table[i]);
2100 if (qid == INVALID_HW_RING_ID) {
2101 PMD_DRV_LOG(ERR, "Inv. entry in rss table.\n");
2104 reta_conf[idx].reta[sft] = qid;
2111 static int bnxt_rss_hash_update_op(struct rte_eth_dev *eth_dev,
2112 struct rte_eth_rss_conf *rss_conf)
2114 struct bnxt *bp = eth_dev->data->dev_private;
2115 struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
2116 struct bnxt_vnic_info *vnic;
2119 rc = is_bnxt_in_error(bp);
2124 * If RSS enablement were different than dev_configure,
2125 * then return -EINVAL
2127 if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
2128 if (!rss_conf->rss_hf)
2129 PMD_DRV_LOG(ERR, "Hash type NONE\n");
2131 if (rss_conf->rss_hf & BNXT_ETH_RSS_SUPPORT)
2135 bp->flags |= BNXT_FLAG_UPDATE_HASH;
2136 memcpy(ð_dev->data->dev_conf.rx_adv_conf.rss_conf,
2140 /* Update the default RSS VNIC(s) */
2141 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2142 vnic->hash_type = bnxt_rte_to_hwrm_hash_types(rss_conf->rss_hf);
2144 bnxt_rte_to_hwrm_hash_level(bp, rss_conf->rss_hf,
2145 ETH_RSS_LEVEL(rss_conf->rss_hf));
2148 * If hashkey is not specified, use the previously configured
2151 if (!rss_conf->rss_key)
2154 if (rss_conf->rss_key_len != HW_HASH_KEY_SIZE) {
2156 "Invalid hashkey length, should be 16 bytes\n");
2159 memcpy(vnic->rss_hash_key, rss_conf->rss_key, rss_conf->rss_key_len);
2162 rc = bnxt_hwrm_vnic_rss_cfg(bp, vnic);
2166 static int bnxt_rss_hash_conf_get_op(struct rte_eth_dev *eth_dev,
2167 struct rte_eth_rss_conf *rss_conf)
2169 struct bnxt *bp = eth_dev->data->dev_private;
2170 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2172 uint32_t hash_types;
2174 rc = is_bnxt_in_error(bp);
2178 /* RSS configuration is the same for all VNICs */
2179 if (vnic && vnic->rss_hash_key) {
2180 if (rss_conf->rss_key) {
2181 len = rss_conf->rss_key_len <= HW_HASH_KEY_SIZE ?
2182 rss_conf->rss_key_len : HW_HASH_KEY_SIZE;
2183 memcpy(rss_conf->rss_key, vnic->rss_hash_key, len);
2186 hash_types = vnic->hash_type;
2187 rss_conf->rss_hf = 0;
2188 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV4) {
2189 rss_conf->rss_hf |= ETH_RSS_IPV4;
2190 hash_types &= ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV4;
2192 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV4) {
2193 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
2195 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV4;
2197 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV4) {
2198 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
2200 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV4;
2202 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV6) {
2203 rss_conf->rss_hf |= ETH_RSS_IPV6;
2204 hash_types &= ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV6;
2206 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV6) {
2207 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
2209 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV6;
2211 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV6) {
2212 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
2214 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV6;
2218 bnxt_hwrm_to_rte_rss_level(bp, vnic->hash_mode);
2222 "Unknown RSS config from firmware (%08x), RSS disabled",
2227 rss_conf->rss_hf = 0;
2232 static int bnxt_flow_ctrl_get_op(struct rte_eth_dev *dev,
2233 struct rte_eth_fc_conf *fc_conf)
2235 struct bnxt *bp = dev->data->dev_private;
2236 struct rte_eth_link link_info;
2239 rc = is_bnxt_in_error(bp);
2243 rc = bnxt_get_hwrm_link_config(bp, &link_info);
2247 memset(fc_conf, 0, sizeof(*fc_conf));
2248 if (bp->link_info->auto_pause)
2249 fc_conf->autoneg = 1;
2250 switch (bp->link_info->pause) {
2252 fc_conf->mode = RTE_FC_NONE;
2254 case HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_TX:
2255 fc_conf->mode = RTE_FC_TX_PAUSE;
2257 case HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_RX:
2258 fc_conf->mode = RTE_FC_RX_PAUSE;
2260 case (HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_TX |
2261 HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_RX):
2262 fc_conf->mode = RTE_FC_FULL;
2268 static int bnxt_flow_ctrl_set_op(struct rte_eth_dev *dev,
2269 struct rte_eth_fc_conf *fc_conf)
2271 struct bnxt *bp = dev->data->dev_private;
2274 rc = is_bnxt_in_error(bp);
2278 if (!BNXT_SINGLE_PF(bp)) {
2280 "Flow Control Settings cannot be modified on VF or on shared PF\n");
2284 switch (fc_conf->mode) {
2286 bp->link_info->auto_pause = 0;
2287 bp->link_info->force_pause = 0;
2289 case RTE_FC_RX_PAUSE:
2290 if (fc_conf->autoneg) {
2291 bp->link_info->auto_pause =
2292 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_RX;
2293 bp->link_info->force_pause = 0;
2295 bp->link_info->auto_pause = 0;
2296 bp->link_info->force_pause =
2297 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_RX;
2300 case RTE_FC_TX_PAUSE:
2301 if (fc_conf->autoneg) {
2302 bp->link_info->auto_pause =
2303 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_TX;
2304 bp->link_info->force_pause = 0;
2306 bp->link_info->auto_pause = 0;
2307 bp->link_info->force_pause =
2308 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_TX;
2312 if (fc_conf->autoneg) {
2313 bp->link_info->auto_pause =
2314 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_TX |
2315 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_RX;
2316 bp->link_info->force_pause = 0;
2318 bp->link_info->auto_pause = 0;
2319 bp->link_info->force_pause =
2320 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_TX |
2321 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_RX;
2325 return bnxt_set_hwrm_link_config(bp, true);
2328 /* Add UDP tunneling port */
2330 bnxt_udp_tunnel_port_add_op(struct rte_eth_dev *eth_dev,
2331 struct rte_eth_udp_tunnel *udp_tunnel)
2333 struct bnxt *bp = eth_dev->data->dev_private;
2334 uint16_t tunnel_type = 0;
2337 rc = is_bnxt_in_error(bp);
2341 switch (udp_tunnel->prot_type) {
2342 case RTE_TUNNEL_TYPE_VXLAN:
2343 if (bp->vxlan_port_cnt) {
2344 PMD_DRV_LOG(ERR, "Tunnel Port %d already programmed\n",
2345 udp_tunnel->udp_port);
2346 if (bp->vxlan_port != udp_tunnel->udp_port) {
2347 PMD_DRV_LOG(ERR, "Only one port allowed\n");
2350 bp->vxlan_port_cnt++;
2354 HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_VXLAN;
2355 bp->vxlan_port_cnt++;
2357 case RTE_TUNNEL_TYPE_GENEVE:
2358 if (bp->geneve_port_cnt) {
2359 PMD_DRV_LOG(ERR, "Tunnel Port %d already programmed\n",
2360 udp_tunnel->udp_port);
2361 if (bp->geneve_port != udp_tunnel->udp_port) {
2362 PMD_DRV_LOG(ERR, "Only one port allowed\n");
2365 bp->geneve_port_cnt++;
2369 HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_GENEVE;
2370 bp->geneve_port_cnt++;
2373 PMD_DRV_LOG(ERR, "Tunnel type is not supported\n");
2376 rc = bnxt_hwrm_tunnel_dst_port_alloc(bp, udp_tunnel->udp_port,
2382 bnxt_udp_tunnel_port_del_op(struct rte_eth_dev *eth_dev,
2383 struct rte_eth_udp_tunnel *udp_tunnel)
2385 struct bnxt *bp = eth_dev->data->dev_private;
2386 uint16_t tunnel_type = 0;
2390 rc = is_bnxt_in_error(bp);
2394 switch (udp_tunnel->prot_type) {
2395 case RTE_TUNNEL_TYPE_VXLAN:
2396 if (!bp->vxlan_port_cnt) {
2397 PMD_DRV_LOG(ERR, "No Tunnel port configured yet\n");
2400 if (bp->vxlan_port != udp_tunnel->udp_port) {
2401 PMD_DRV_LOG(ERR, "Req Port: %d. Configured port: %d\n",
2402 udp_tunnel->udp_port, bp->vxlan_port);
2405 if (--bp->vxlan_port_cnt)
2409 HWRM_TUNNEL_DST_PORT_FREE_INPUT_TUNNEL_TYPE_VXLAN;
2410 port = bp->vxlan_fw_dst_port_id;
2412 case RTE_TUNNEL_TYPE_GENEVE:
2413 if (!bp->geneve_port_cnt) {
2414 PMD_DRV_LOG(ERR, "No Tunnel port configured yet\n");
2417 if (bp->geneve_port != udp_tunnel->udp_port) {
2418 PMD_DRV_LOG(ERR, "Req Port: %d. Configured port: %d\n",
2419 udp_tunnel->udp_port, bp->geneve_port);
2422 if (--bp->geneve_port_cnt)
2426 HWRM_TUNNEL_DST_PORT_FREE_INPUT_TUNNEL_TYPE_GENEVE;
2427 port = bp->geneve_fw_dst_port_id;
2430 PMD_DRV_LOG(ERR, "Tunnel type is not supported\n");
2434 rc = bnxt_hwrm_tunnel_dst_port_free(bp, port, tunnel_type);
2438 static int bnxt_del_vlan_filter(struct bnxt *bp, uint16_t vlan_id)
2440 struct bnxt_filter_info *filter;
2441 struct bnxt_vnic_info *vnic;
2443 uint32_t chk = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN;
2445 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2446 filter = STAILQ_FIRST(&vnic->filter);
2448 /* Search for this matching MAC+VLAN filter */
2449 if (bnxt_vlan_filter_exists(bp, filter, chk, vlan_id)) {
2450 /* Delete the filter */
2451 rc = bnxt_hwrm_clear_l2_filter(bp, filter);
2454 STAILQ_REMOVE(&vnic->filter, filter,
2455 bnxt_filter_info, next);
2456 bnxt_free_filter(bp, filter);
2458 "Deleted vlan filter for %d\n",
2462 filter = STAILQ_NEXT(filter, next);
2467 static int bnxt_add_vlan_filter(struct bnxt *bp, uint16_t vlan_id)
2469 struct bnxt_filter_info *filter;
2470 struct bnxt_vnic_info *vnic;
2472 uint32_t en = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN |
2473 HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN_MASK;
2474 uint32_t chk = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN;
2476 /* Implementation notes on the use of VNIC in this command:
2478 * By default, these filters belong to default vnic for the function.
2479 * Once these filters are set up, only destination VNIC can be modified.
2480 * If the destination VNIC is not specified in this command,
2481 * then the HWRM shall only create an l2 context id.
2484 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2485 filter = STAILQ_FIRST(&vnic->filter);
2486 /* Check if the VLAN has already been added */
2488 if (bnxt_vlan_filter_exists(bp, filter, chk, vlan_id))
2491 filter = STAILQ_NEXT(filter, next);
2494 /* No match found. Alloc a fresh filter and issue the L2_FILTER_ALLOC
2495 * command to create MAC+VLAN filter with the right flags, enables set.
2497 filter = bnxt_alloc_filter(bp);
2500 "MAC/VLAN filter alloc failed\n");
2503 /* MAC + VLAN ID filter */
2504 /* If l2_ivlan == 0 and l2_ivlan_mask != 0, only
2505 * untagged packets are received
2507 * If l2_ivlan != 0 and l2_ivlan_mask != 0, untagged
2508 * packets and only the programmed vlan's packets are received
2510 filter->l2_ivlan = vlan_id;
2511 filter->l2_ivlan_mask = 0x0FFF;
2512 filter->enables |= en;
2513 filter->flags |= HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
2515 rc = bnxt_hwrm_set_l2_filter(bp, vnic->fw_vnic_id, filter);
2517 /* Free the newly allocated filter as we were
2518 * not able to create the filter in hardware.
2520 bnxt_free_filter(bp, filter);
2524 filter->mac_index = 0;
2525 /* Add this new filter to the list */
2527 STAILQ_INSERT_HEAD(&vnic->filter, filter, next);
2529 STAILQ_INSERT_TAIL(&vnic->filter, filter, next);
2532 "Added Vlan filter for %d\n", vlan_id);
2536 static int bnxt_vlan_filter_set_op(struct rte_eth_dev *eth_dev,
2537 uint16_t vlan_id, int on)
2539 struct bnxt *bp = eth_dev->data->dev_private;
2542 rc = is_bnxt_in_error(bp);
2546 if (!eth_dev->data->dev_started) {
2547 PMD_DRV_LOG(ERR, "port must be started before setting vlan\n");
2551 /* These operations apply to ALL existing MAC/VLAN filters */
2553 return bnxt_add_vlan_filter(bp, vlan_id);
2555 return bnxt_del_vlan_filter(bp, vlan_id);
2558 static int bnxt_del_dflt_mac_filter(struct bnxt *bp,
2559 struct bnxt_vnic_info *vnic)
2561 struct bnxt_filter_info *filter;
2564 filter = STAILQ_FIRST(&vnic->filter);
2566 if (filter->mac_index == 0 &&
2567 !memcmp(filter->l2_addr, bp->mac_addr,
2568 RTE_ETHER_ADDR_LEN)) {
2569 rc = bnxt_hwrm_clear_l2_filter(bp, filter);
2571 STAILQ_REMOVE(&vnic->filter, filter,
2572 bnxt_filter_info, next);
2573 bnxt_free_filter(bp, filter);
2577 filter = STAILQ_NEXT(filter, next);
2583 bnxt_config_vlan_hw_filter(struct bnxt *bp, uint64_t rx_offloads)
2585 struct bnxt_vnic_info *vnic;
2589 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2590 if (!(rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)) {
2591 /* Remove any VLAN filters programmed */
2592 for (i = 0; i < RTE_ETHER_MAX_VLAN_ID; i++)
2593 bnxt_del_vlan_filter(bp, i);
2595 rc = bnxt_add_mac_filter(bp, vnic, NULL, 0, 0);
2599 /* Default filter will allow packets that match the
2600 * dest mac. So, it has to be deleted, otherwise, we
2601 * will endup receiving vlan packets for which the
2602 * filter is not programmed, when hw-vlan-filter
2603 * configuration is ON
2605 bnxt_del_dflt_mac_filter(bp, vnic);
2606 /* This filter will allow only untagged packets */
2607 bnxt_add_vlan_filter(bp, 0);
2609 PMD_DRV_LOG(DEBUG, "VLAN Filtering: %d\n",
2610 !!(rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER));
2615 static int bnxt_free_one_vnic(struct bnxt *bp, uint16_t vnic_id)
2617 struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
2621 /* Destroy vnic filters and vnic */
2622 if (bp->eth_dev->data->dev_conf.rxmode.offloads &
2623 DEV_RX_OFFLOAD_VLAN_FILTER) {
2624 for (i = 0; i < RTE_ETHER_MAX_VLAN_ID; i++)
2625 bnxt_del_vlan_filter(bp, i);
2627 bnxt_del_dflt_mac_filter(bp, vnic);
2629 rc = bnxt_hwrm_vnic_ctx_free(bp, vnic);
2633 rc = bnxt_hwrm_vnic_free(bp, vnic);
2637 rte_free(vnic->fw_grp_ids);
2638 vnic->fw_grp_ids = NULL;
2640 vnic->rx_queue_cnt = 0;
2646 bnxt_config_vlan_hw_stripping(struct bnxt *bp, uint64_t rx_offloads)
2648 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2651 /* Destroy, recreate and reconfigure the default vnic */
2652 rc = bnxt_free_one_vnic(bp, 0);
2656 /* default vnic 0 */
2657 rc = bnxt_setup_one_vnic(bp, 0);
2661 if (bp->eth_dev->data->dev_conf.rxmode.offloads &
2662 DEV_RX_OFFLOAD_VLAN_FILTER) {
2663 rc = bnxt_add_vlan_filter(bp, 0);
2666 rc = bnxt_restore_vlan_filters(bp);
2670 rc = bnxt_add_mac_filter(bp, vnic, NULL, 0, 0);
2675 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
2679 PMD_DRV_LOG(DEBUG, "VLAN Strip Offload: %d\n",
2680 !!(rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP));
2686 bnxt_vlan_offload_set_op(struct rte_eth_dev *dev, int mask)
2688 uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
2689 struct bnxt *bp = dev->data->dev_private;
2692 rc = is_bnxt_in_error(bp);
2696 /* Filter settings will get applied when port is started */
2697 if (!dev->data->dev_started)
2700 if (mask & ETH_VLAN_FILTER_MASK) {
2701 /* Enable or disable VLAN filtering */
2702 rc = bnxt_config_vlan_hw_filter(bp, rx_offloads);
2707 if (mask & ETH_VLAN_STRIP_MASK) {
2708 /* Enable or disable VLAN stripping */
2709 rc = bnxt_config_vlan_hw_stripping(bp, rx_offloads);
2714 if (mask & ETH_VLAN_EXTEND_MASK) {
2715 if (rx_offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)
2716 PMD_DRV_LOG(DEBUG, "Extend VLAN supported\n");
2718 PMD_DRV_LOG(INFO, "Extend VLAN unsupported\n");
2725 bnxt_vlan_tpid_set_op(struct rte_eth_dev *dev, enum rte_vlan_type vlan_type,
2728 struct bnxt *bp = dev->data->dev_private;
2729 int qinq = dev->data->dev_conf.rxmode.offloads &
2730 DEV_RX_OFFLOAD_VLAN_EXTEND;
2732 if (vlan_type != ETH_VLAN_TYPE_INNER &&
2733 vlan_type != ETH_VLAN_TYPE_OUTER) {
2735 "Unsupported vlan type.");
2740 "QinQ not enabled. Needs to be ON as we can "
2741 "accelerate only outer vlan\n");
2745 if (vlan_type == ETH_VLAN_TYPE_OUTER) {
2747 case RTE_ETHER_TYPE_QINQ:
2749 TX_BD_LONG_CFA_META_VLAN_TPID_TPID88A8;
2751 case RTE_ETHER_TYPE_VLAN:
2753 TX_BD_LONG_CFA_META_VLAN_TPID_TPID8100;
2755 case RTE_ETHER_TYPE_QINQ1:
2757 TX_BD_LONG_CFA_META_VLAN_TPID_TPID9100;
2759 case RTE_ETHER_TYPE_QINQ2:
2761 TX_BD_LONG_CFA_META_VLAN_TPID_TPID9200;
2763 case RTE_ETHER_TYPE_QINQ3:
2765 TX_BD_LONG_CFA_META_VLAN_TPID_TPID9300;
2768 PMD_DRV_LOG(ERR, "Invalid TPID: %x\n", tpid);
2771 bp->outer_tpid_bd |= tpid;
2772 PMD_DRV_LOG(INFO, "outer_tpid_bd = %x\n", bp->outer_tpid_bd);
2773 } else if (vlan_type == ETH_VLAN_TYPE_INNER) {
2775 "Can accelerate only outer vlan in QinQ\n");
2783 bnxt_set_default_mac_addr_op(struct rte_eth_dev *dev,
2784 struct rte_ether_addr *addr)
2786 struct bnxt *bp = dev->data->dev_private;
2787 /* Default Filter is tied to VNIC 0 */
2788 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2791 rc = is_bnxt_in_error(bp);
2795 if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp))
2798 if (rte_is_zero_ether_addr(addr))
2801 /* Filter settings will get applied when port is started */
2802 if (!dev->data->dev_started)
2805 /* Check if the requested MAC is already added */
2806 if (memcmp(addr, bp->mac_addr, RTE_ETHER_ADDR_LEN) == 0)
2809 /* Destroy filter and re-create it */
2810 bnxt_del_dflt_mac_filter(bp, vnic);
2812 memcpy(bp->mac_addr, addr, RTE_ETHER_ADDR_LEN);
2813 if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_VLAN_FILTER) {
2814 /* This filter will allow only untagged packets */
2815 rc = bnxt_add_vlan_filter(bp, 0);
2817 rc = bnxt_add_mac_filter(bp, vnic, addr, 0, 0);
2820 PMD_DRV_LOG(DEBUG, "Set MAC addr\n");
2825 bnxt_dev_set_mc_addr_list_op(struct rte_eth_dev *eth_dev,
2826 struct rte_ether_addr *mc_addr_set,
2827 uint32_t nb_mc_addr)
2829 struct bnxt *bp = eth_dev->data->dev_private;
2830 char *mc_addr_list = (char *)mc_addr_set;
2831 struct bnxt_vnic_info *vnic;
2832 uint32_t off = 0, i = 0;
2835 rc = is_bnxt_in_error(bp);
2839 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2841 if (nb_mc_addr > BNXT_MAX_MC_ADDRS) {
2842 vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
2846 /* TODO Check for Duplicate mcast addresses */
2847 vnic->flags &= ~BNXT_VNIC_INFO_ALLMULTI;
2848 for (i = 0; i < nb_mc_addr; i++) {
2849 memcpy(vnic->mc_list + off, &mc_addr_list[i],
2850 RTE_ETHER_ADDR_LEN);
2851 off += RTE_ETHER_ADDR_LEN;
2854 vnic->mc_addr_cnt = i;
2855 if (vnic->mc_addr_cnt)
2856 vnic->flags |= BNXT_VNIC_INFO_MCAST;
2858 vnic->flags &= ~BNXT_VNIC_INFO_MCAST;
2861 return bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
2865 bnxt_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
2867 struct bnxt *bp = dev->data->dev_private;
2868 uint8_t fw_major = (bp->fw_ver >> 24) & 0xff;
2869 uint8_t fw_minor = (bp->fw_ver >> 16) & 0xff;
2870 uint8_t fw_updt = (bp->fw_ver >> 8) & 0xff;
2871 uint8_t fw_rsvd = bp->fw_ver & 0xff;
2874 ret = snprintf(fw_version, fw_size, "%d.%d.%d.%d",
2875 fw_major, fw_minor, fw_updt, fw_rsvd);
2879 ret += 1; /* add the size of '\0' */
2880 if (fw_size < (size_t)ret)
2887 bnxt_rxq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
2888 struct rte_eth_rxq_info *qinfo)
2890 struct bnxt *bp = dev->data->dev_private;
2891 struct bnxt_rx_queue *rxq;
2893 if (is_bnxt_in_error(bp))
2896 rxq = dev->data->rx_queues[queue_id];
2898 qinfo->mp = rxq->mb_pool;
2899 qinfo->scattered_rx = dev->data->scattered_rx;
2900 qinfo->nb_desc = rxq->nb_rx_desc;
2902 qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
2903 qinfo->conf.rx_drop_en = rxq->drop_en;
2904 qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
2905 qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
2909 bnxt_txq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
2910 struct rte_eth_txq_info *qinfo)
2912 struct bnxt *bp = dev->data->dev_private;
2913 struct bnxt_tx_queue *txq;
2915 if (is_bnxt_in_error(bp))
2918 txq = dev->data->tx_queues[queue_id];
2920 qinfo->nb_desc = txq->nb_tx_desc;
2922 qinfo->conf.tx_thresh.pthresh = txq->pthresh;
2923 qinfo->conf.tx_thresh.hthresh = txq->hthresh;
2924 qinfo->conf.tx_thresh.wthresh = txq->wthresh;
2926 qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
2927 qinfo->conf.tx_rs_thresh = 0;
2928 qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
2929 qinfo->conf.offloads = txq->offloads;
2932 static const struct {
2933 eth_rx_burst_t pkt_burst;
2935 } bnxt_rx_burst_info[] = {
2936 {bnxt_recv_pkts, "Scalar"},
2937 #if defined(RTE_ARCH_X86)
2938 {bnxt_recv_pkts_vec, "Vector SSE"},
2940 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
2941 {bnxt_recv_pkts_vec_avx2, "Vector AVX2"},
2943 #if defined(RTE_ARCH_ARM64)
2944 {bnxt_recv_pkts_vec, "Vector Neon"},
2949 bnxt_rx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
2950 struct rte_eth_burst_mode *mode)
2952 eth_rx_burst_t pkt_burst = dev->rx_pkt_burst;
2955 for (i = 0; i < RTE_DIM(bnxt_rx_burst_info); i++) {
2956 if (pkt_burst == bnxt_rx_burst_info[i].pkt_burst) {
2957 snprintf(mode->info, sizeof(mode->info), "%s",
2958 bnxt_rx_burst_info[i].info);
2966 static const struct {
2967 eth_tx_burst_t pkt_burst;
2969 } bnxt_tx_burst_info[] = {
2970 {bnxt_xmit_pkts, "Scalar"},
2971 #if defined(RTE_ARCH_X86)
2972 {bnxt_xmit_pkts_vec, "Vector SSE"},
2974 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
2975 {bnxt_xmit_pkts_vec_avx2, "Vector AVX2"},
2977 #if defined(RTE_ARCH_ARM64)
2978 {bnxt_xmit_pkts_vec, "Vector Neon"},
2983 bnxt_tx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
2984 struct rte_eth_burst_mode *mode)
2986 eth_tx_burst_t pkt_burst = dev->tx_pkt_burst;
2989 for (i = 0; i < RTE_DIM(bnxt_tx_burst_info); i++) {
2990 if (pkt_burst == bnxt_tx_burst_info[i].pkt_burst) {
2991 snprintf(mode->info, sizeof(mode->info), "%s",
2992 bnxt_tx_burst_info[i].info);
3000 int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
3002 struct bnxt *bp = eth_dev->data->dev_private;
3003 uint32_t new_pkt_size;
3007 rc = is_bnxt_in_error(bp);
3011 /* Exit if receive queues are not configured yet */
3012 if (!eth_dev->data->nb_rx_queues)
3015 new_pkt_size = new_mtu + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN +
3016 VLAN_TAG_SIZE * BNXT_NUM_VLANS;
3019 * Disallow any MTU change that would require scattered receive support
3020 * if it is not already enabled.
3022 if (eth_dev->data->dev_started &&
3023 !eth_dev->data->scattered_rx &&
3025 eth_dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM)) {
3027 "MTU change would require scattered rx support. ");
3028 PMD_DRV_LOG(ERR, "Stop port before changing MTU.\n");
3032 if (new_mtu > RTE_ETHER_MTU) {
3033 bp->flags |= BNXT_FLAG_JUMBO;
3034 bp->eth_dev->data->dev_conf.rxmode.offloads |=
3035 DEV_RX_OFFLOAD_JUMBO_FRAME;
3037 bp->eth_dev->data->dev_conf.rxmode.offloads &=
3038 ~DEV_RX_OFFLOAD_JUMBO_FRAME;
3039 bp->flags &= ~BNXT_FLAG_JUMBO;
3042 /* Is there a change in mtu setting? */
3043 if (eth_dev->data->dev_conf.rxmode.max_rx_pkt_len == new_pkt_size)
3046 for (i = 0; i < bp->nr_vnics; i++) {
3047 struct bnxt_vnic_info *vnic = &bp->vnic_info[i];
3050 vnic->mru = BNXT_VNIC_MRU(new_mtu);
3051 rc = bnxt_hwrm_vnic_cfg(bp, vnic);
3055 size = rte_pktmbuf_data_room_size(bp->rx_queues[0]->mb_pool);
3056 size -= RTE_PKTMBUF_HEADROOM;
3058 if (size < new_mtu) {
3059 rc = bnxt_hwrm_vnic_plcmode_cfg(bp, vnic);
3066 eth_dev->data->dev_conf.rxmode.max_rx_pkt_len = new_pkt_size;
3068 PMD_DRV_LOG(INFO, "New MTU is %d\n", new_mtu);
3074 bnxt_vlan_pvid_set_op(struct rte_eth_dev *dev, uint16_t pvid, int on)
3076 struct bnxt *bp = dev->data->dev_private;
3077 uint16_t vlan = bp->vlan;
3080 rc = is_bnxt_in_error(bp);
3084 if (!BNXT_SINGLE_PF(bp)) {
3085 PMD_DRV_LOG(ERR, "PVID cannot be modified on VF or on shared PF\n");
3088 bp->vlan = on ? pvid : 0;
3090 rc = bnxt_hwrm_set_default_vlan(bp, 0, 0);
3097 bnxt_dev_led_on_op(struct rte_eth_dev *dev)
3099 struct bnxt *bp = dev->data->dev_private;
3102 rc = is_bnxt_in_error(bp);
3106 return bnxt_hwrm_port_led_cfg(bp, true);
3110 bnxt_dev_led_off_op(struct rte_eth_dev *dev)
3112 struct bnxt *bp = dev->data->dev_private;
3115 rc = is_bnxt_in_error(bp);
3119 return bnxt_hwrm_port_led_cfg(bp, false);
3123 bnxt_rx_queue_count_op(struct rte_eth_dev *dev, uint16_t rx_queue_id)
3125 struct bnxt *bp = (struct bnxt *)dev->data->dev_private;
3126 struct bnxt_cp_ring_info *cpr;
3127 uint32_t desc = 0, raw_cons;
3128 struct bnxt_rx_queue *rxq;
3129 struct rx_pkt_cmpl *rxcmp;
3132 rc = is_bnxt_in_error(bp);
3136 rxq = dev->data->rx_queues[rx_queue_id];
3138 raw_cons = cpr->cp_raw_cons;
3141 uint32_t agg_cnt, cons, cmpl_type;
3143 cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3144 rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3146 if (!CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct))
3149 cmpl_type = CMP_TYPE(rxcmp);
3151 switch (cmpl_type) {
3152 case CMPL_BASE_TYPE_RX_L2:
3153 case CMPL_BASE_TYPE_RX_L2_V2:
3154 agg_cnt = BNXT_RX_L2_AGG_BUFS(rxcmp);
3155 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3159 case CMPL_BASE_TYPE_RX_TPA_END:
3160 if (BNXT_CHIP_P5(rxq->bp)) {
3161 struct rx_tpa_v2_end_cmpl_hi *p5_tpa_end;
3163 p5_tpa_end = (void *)rxcmp;
3164 agg_cnt = BNXT_TPA_END_AGG_BUFS_TH(p5_tpa_end);
3166 struct rx_tpa_end_cmpl *tpa_end;
3168 tpa_end = (void *)rxcmp;
3169 agg_cnt = BNXT_TPA_END_AGG_BUFS(tpa_end);
3172 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3177 raw_cons += CMP_LEN(cmpl_type);
3185 bnxt_rx_descriptor_status_op(void *rx_queue, uint16_t offset)
3187 struct bnxt_rx_queue *rxq = rx_queue;
3188 struct bnxt_cp_ring_info *cpr;
3189 struct bnxt_rx_ring_info *rxr;
3190 uint32_t desc, raw_cons;
3191 struct bnxt *bp = rxq->bp;
3192 struct rx_pkt_cmpl *rxcmp;
3195 rc = is_bnxt_in_error(bp);
3199 if (offset >= rxq->nb_rx_desc)
3206 * For the vector receive case, the completion at the requested
3207 * offset can be indexed directly.
3209 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
3210 if (bp->flags & BNXT_FLAG_RX_VECTOR_PKT_MODE) {
3211 struct rx_pkt_cmpl *rxcmp;
3214 /* Check status of completion descriptor. */
3215 raw_cons = cpr->cp_raw_cons +
3216 offset * CMP_LEN(CMPL_BASE_TYPE_RX_L2);
3217 cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3218 rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3220 if (CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct))
3221 return RTE_ETH_RX_DESC_DONE;
3223 /* Check whether rx desc has an mbuf attached. */
3224 cons = RING_CMP(rxr->rx_ring_struct, raw_cons / 2);
3225 if (cons >= rxq->rxrearm_start &&
3226 cons < rxq->rxrearm_start + rxq->rxrearm_nb) {
3227 return RTE_ETH_RX_DESC_UNAVAIL;
3230 return RTE_ETH_RX_DESC_AVAIL;
3235 * For the non-vector receive case, scan the completion ring to
3236 * locate the completion descriptor for the requested offset.
3238 raw_cons = cpr->cp_raw_cons;
3241 uint32_t agg_cnt, cons, cmpl_type;
3243 cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3244 rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3246 if (!CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct))
3249 cmpl_type = CMP_TYPE(rxcmp);
3251 switch (cmpl_type) {
3252 case CMPL_BASE_TYPE_RX_L2:
3253 case CMPL_BASE_TYPE_RX_L2_V2:
3254 if (desc == offset) {
3255 cons = rxcmp->opaque;
3256 if (rxr->rx_buf_ring[cons])
3257 return RTE_ETH_RX_DESC_DONE;
3259 return RTE_ETH_RX_DESC_UNAVAIL;
3261 agg_cnt = BNXT_RX_L2_AGG_BUFS(rxcmp);
3262 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3266 case CMPL_BASE_TYPE_RX_TPA_END:
3268 return RTE_ETH_RX_DESC_DONE;
3270 if (BNXT_CHIP_P5(rxq->bp)) {
3271 struct rx_tpa_v2_end_cmpl_hi *p5_tpa_end;
3273 p5_tpa_end = (void *)rxcmp;
3274 agg_cnt = BNXT_TPA_END_AGG_BUFS_TH(p5_tpa_end);
3276 struct rx_tpa_end_cmpl *tpa_end;
3278 tpa_end = (void *)rxcmp;
3279 agg_cnt = BNXT_TPA_END_AGG_BUFS(tpa_end);
3282 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3287 raw_cons += CMP_LEN(cmpl_type);
3291 return RTE_ETH_RX_DESC_AVAIL;
3295 bnxt_tx_descriptor_status_op(void *tx_queue, uint16_t offset)
3297 struct bnxt_tx_queue *txq = (struct bnxt_tx_queue *)tx_queue;
3298 struct bnxt_tx_ring_info *txr;
3299 struct bnxt_cp_ring_info *cpr;
3300 struct rte_mbuf **tx_buf;
3301 struct tx_pkt_cmpl *txcmp;
3302 uint32_t cons, cp_cons;
3308 rc = is_bnxt_in_error(txq->bp);
3315 if (offset >= txq->nb_tx_desc)
3318 cons = RING_CMP(cpr->cp_ring_struct, offset);
3319 txcmp = (struct tx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3320 cp_cons = cpr->cp_raw_cons;
3322 if (cons > cp_cons) {
3323 if (CMPL_VALID(txcmp, cpr->valid))
3324 return RTE_ETH_TX_DESC_UNAVAIL;
3326 if (CMPL_VALID(txcmp, !cpr->valid))
3327 return RTE_ETH_TX_DESC_UNAVAIL;
3329 tx_buf = &txr->tx_buf_ring[cons];
3330 if (*tx_buf == NULL)
3331 return RTE_ETH_TX_DESC_DONE;
3333 return RTE_ETH_TX_DESC_FULL;
3337 bnxt_flow_ops_get_op(struct rte_eth_dev *dev,
3338 const struct rte_flow_ops **ops)
3340 struct bnxt *bp = dev->data->dev_private;
3346 if (BNXT_ETH_DEV_IS_REPRESENTOR(dev)) {
3347 struct bnxt_representor *vfr = dev->data->dev_private;
3348 bp = vfr->parent_dev->data->dev_private;
3349 /* parent is deleted while children are still valid */
3351 PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR Error\n",
3352 dev->data->port_id);
3357 ret = is_bnxt_in_error(bp);
3361 /* PMD supports thread-safe flow operations. rte_flow API
3362 * functions can avoid mutex for multi-thread safety.
3364 dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
3366 if (BNXT_TRUFLOW_EN(bp))
3367 *ops = &bnxt_ulp_rte_flow_ops;
3369 *ops = &bnxt_flow_ops;
3374 static const uint32_t *
3375 bnxt_dev_supported_ptypes_get_op(struct rte_eth_dev *dev)
3377 static const uint32_t ptypes[] = {
3378 RTE_PTYPE_L2_ETHER_VLAN,
3379 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
3380 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
3384 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
3385 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
3386 RTE_PTYPE_INNER_L4_ICMP,
3387 RTE_PTYPE_INNER_L4_TCP,
3388 RTE_PTYPE_INNER_L4_UDP,
3392 if (!dev->rx_pkt_burst)
3398 static int bnxt_map_regs(struct bnxt *bp, uint32_t *reg_arr, int count,
3401 uint32_t reg_base = *reg_arr & 0xfffff000;
3405 for (i = 0; i < count; i++) {
3406 if ((reg_arr[i] & 0xfffff000) != reg_base)
3409 win_off = BNXT_GRCPF_REG_WINDOW_BASE_OUT + (reg_win - 1) * 4;
3410 rte_write32(reg_base, (uint8_t *)bp->bar0 + win_off);
3414 static int bnxt_map_ptp_regs(struct bnxt *bp)
3416 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3420 reg_arr = ptp->rx_regs;
3421 rc = bnxt_map_regs(bp, reg_arr, BNXT_PTP_RX_REGS, 5);
3425 reg_arr = ptp->tx_regs;
3426 rc = bnxt_map_regs(bp, reg_arr, BNXT_PTP_TX_REGS, 6);
3430 for (i = 0; i < BNXT_PTP_RX_REGS; i++)
3431 ptp->rx_mapped_regs[i] = 0x5000 + (ptp->rx_regs[i] & 0xfff);
3433 for (i = 0; i < BNXT_PTP_TX_REGS; i++)
3434 ptp->tx_mapped_regs[i] = 0x6000 + (ptp->tx_regs[i] & 0xfff);
3439 static void bnxt_unmap_ptp_regs(struct bnxt *bp)
3441 rte_write32(0, (uint8_t *)bp->bar0 +
3442 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 16);
3443 rte_write32(0, (uint8_t *)bp->bar0 +
3444 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 20);
3447 static uint64_t bnxt_cc_read(struct bnxt *bp)
3451 ns = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3452 BNXT_GRCPF_REG_SYNC_TIME));
3453 ns |= (uint64_t)(rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3454 BNXT_GRCPF_REG_SYNC_TIME + 4))) << 32;
3458 static int bnxt_get_tx_ts(struct bnxt *bp, uint64_t *ts)
3460 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3463 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3464 ptp->tx_mapped_regs[BNXT_PTP_TX_FIFO]));
3465 if (fifo & BNXT_PTP_TX_FIFO_EMPTY)
3468 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3469 ptp->tx_mapped_regs[BNXT_PTP_TX_FIFO]));
3470 *ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3471 ptp->tx_mapped_regs[BNXT_PTP_TX_TS_L]));
3472 *ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3473 ptp->tx_mapped_regs[BNXT_PTP_TX_TS_H])) << 32;
3474 rte_read32((uint8_t *)bp->bar0 + ptp->tx_mapped_regs[BNXT_PTP_TX_SEQ]);
3479 static int bnxt_clr_rx_ts(struct bnxt *bp, uint64_t *last_ts)
3481 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3482 struct bnxt_pf_info *pf = bp->pf;
3487 if (!ptp || (bp->flags & BNXT_FLAG_CHIP_P5))
3490 port_id = pf->port_id;
3491 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3492 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3493 while ((fifo & BNXT_PTP_RX_FIFO_PENDING) && (i < BNXT_PTP_RX_PND_CNT)) {
3494 rte_write32(1 << port_id, (uint8_t *)bp->bar0 +
3495 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO_ADV]);
3496 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3497 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3498 *last_ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3499 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_L]));
3500 *last_ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3501 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_H])) << 32;
3505 if (i >= BNXT_PTP_RX_PND_CNT)
3511 static int bnxt_get_rx_ts(struct bnxt *bp, uint64_t *ts)
3513 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3514 struct bnxt_pf_info *pf = bp->pf;
3518 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3519 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3520 if (!(fifo & BNXT_PTP_RX_FIFO_PENDING))
3523 port_id = pf->port_id;
3524 rte_write32(1 << port_id, (uint8_t *)bp->bar0 +
3525 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO_ADV]);
3527 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3528 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3529 if (fifo & BNXT_PTP_RX_FIFO_PENDING)
3530 return bnxt_clr_rx_ts(bp, ts);
3532 *ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3533 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_L]));
3534 *ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3535 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_H])) << 32;
3541 bnxt_timesync_write_time(struct rte_eth_dev *dev, const struct timespec *ts)
3544 struct bnxt *bp = dev->data->dev_private;
3545 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3550 ns = rte_timespec_to_ns(ts);
3551 /* Set the timecounters to a new value. */
3553 ptp->tx_tstamp_tc.nsec = ns;
3554 ptp->rx_tstamp_tc.nsec = ns;
3560 bnxt_timesync_read_time(struct rte_eth_dev *dev, struct timespec *ts)
3562 struct bnxt *bp = dev->data->dev_private;
3563 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3564 uint64_t ns, systime_cycles = 0;
3570 if (BNXT_CHIP_P5(bp))
3571 rc = bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
3574 systime_cycles = bnxt_cc_read(bp);
3576 ns = rte_timecounter_update(&ptp->tc, systime_cycles);
3577 *ts = rte_ns_to_timespec(ns);
3582 bnxt_timesync_enable(struct rte_eth_dev *dev)
3584 struct bnxt *bp = dev->data->dev_private;
3585 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3593 ptp->tx_tstamp_en = 1;
3594 ptp->rxctl = BNXT_PTP_MSG_EVENTS;
3596 rc = bnxt_hwrm_ptp_cfg(bp);
3600 memset(&ptp->tc, 0, sizeof(struct rte_timecounter));
3601 memset(&ptp->rx_tstamp_tc, 0, sizeof(struct rte_timecounter));
3602 memset(&ptp->tx_tstamp_tc, 0, sizeof(struct rte_timecounter));
3604 ptp->tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3605 ptp->tc.cc_shift = shift;
3606 ptp->tc.nsec_mask = (1ULL << shift) - 1;
3608 ptp->rx_tstamp_tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3609 ptp->rx_tstamp_tc.cc_shift = shift;
3610 ptp->rx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
3612 ptp->tx_tstamp_tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3613 ptp->tx_tstamp_tc.cc_shift = shift;
3614 ptp->tx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
3616 if (!BNXT_CHIP_P5(bp))
3617 bnxt_map_ptp_regs(bp);
3619 rc = bnxt_ptp_start(bp);
3625 bnxt_timesync_disable(struct rte_eth_dev *dev)
3627 struct bnxt *bp = dev->data->dev_private;
3628 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3634 ptp->tx_tstamp_en = 0;
3637 bnxt_hwrm_ptp_cfg(bp);
3639 if (!BNXT_CHIP_P5(bp))
3640 bnxt_unmap_ptp_regs(bp);
3648 bnxt_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
3649 struct timespec *timestamp,
3650 uint32_t flags __rte_unused)
3652 struct bnxt *bp = dev->data->dev_private;
3653 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3654 uint64_t rx_tstamp_cycles = 0;
3660 if (BNXT_CHIP_P5(bp))
3661 rx_tstamp_cycles = ptp->rx_timestamp;
3663 bnxt_get_rx_ts(bp, &rx_tstamp_cycles);
3665 ns = rte_timecounter_update(&ptp->rx_tstamp_tc, rx_tstamp_cycles);
3666 *timestamp = rte_ns_to_timespec(ns);
3671 bnxt_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
3672 struct timespec *timestamp)
3674 struct bnxt *bp = dev->data->dev_private;
3675 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3676 uint64_t tx_tstamp_cycles = 0;
3683 if (BNXT_CHIP_P5(bp))
3684 rc = bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_PATH_TX,
3687 rc = bnxt_get_tx_ts(bp, &tx_tstamp_cycles);
3689 ns = rte_timecounter_update(&ptp->tx_tstamp_tc, tx_tstamp_cycles);
3690 *timestamp = rte_ns_to_timespec(ns);
3696 bnxt_timesync_adjust_time(struct rte_eth_dev *dev, int64_t delta)
3698 struct bnxt *bp = dev->data->dev_private;
3699 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3704 ptp->tc.nsec += delta;
3705 ptp->tx_tstamp_tc.nsec += delta;
3706 ptp->rx_tstamp_tc.nsec += delta;
3712 bnxt_get_eeprom_length_op(struct rte_eth_dev *dev)
3714 struct bnxt *bp = dev->data->dev_private;
3716 uint32_t dir_entries;
3717 uint32_t entry_length;
3719 rc = is_bnxt_in_error(bp);
3723 PMD_DRV_LOG(INFO, PCI_PRI_FMT "\n",
3724 bp->pdev->addr.domain, bp->pdev->addr.bus,
3725 bp->pdev->addr.devid, bp->pdev->addr.function);
3727 rc = bnxt_hwrm_nvm_get_dir_info(bp, &dir_entries, &entry_length);
3731 return dir_entries * entry_length;
3735 bnxt_get_eeprom_op(struct rte_eth_dev *dev,
3736 struct rte_dev_eeprom_info *in_eeprom)
3738 struct bnxt *bp = dev->data->dev_private;
3743 rc = is_bnxt_in_error(bp);
3747 PMD_DRV_LOG(INFO, PCI_PRI_FMT " in_eeprom->offset = %d len = %d\n",
3748 bp->pdev->addr.domain, bp->pdev->addr.bus,
3749 bp->pdev->addr.devid, bp->pdev->addr.function,
3750 in_eeprom->offset, in_eeprom->length);
3752 if (in_eeprom->offset == 0) /* special offset value to get directory */
3753 return bnxt_get_nvram_directory(bp, in_eeprom->length,
3756 index = in_eeprom->offset >> 24;
3757 offset = in_eeprom->offset & 0xffffff;
3760 return bnxt_hwrm_get_nvram_item(bp, index - 1, offset,
3761 in_eeprom->length, in_eeprom->data);
3766 static bool bnxt_dir_type_is_ape_bin_format(uint16_t dir_type)
3769 case BNX_DIR_TYPE_CHIMP_PATCH:
3770 case BNX_DIR_TYPE_BOOTCODE:
3771 case BNX_DIR_TYPE_BOOTCODE_2:
3772 case BNX_DIR_TYPE_APE_FW:
3773 case BNX_DIR_TYPE_APE_PATCH:
3774 case BNX_DIR_TYPE_KONG_FW:
3775 case BNX_DIR_TYPE_KONG_PATCH:
3776 case BNX_DIR_TYPE_BONO_FW:
3777 case BNX_DIR_TYPE_BONO_PATCH:
3785 static bool bnxt_dir_type_is_other_exec_format(uint16_t dir_type)
3788 case BNX_DIR_TYPE_AVS:
3789 case BNX_DIR_TYPE_EXP_ROM_MBA:
3790 case BNX_DIR_TYPE_PCIE:
3791 case BNX_DIR_TYPE_TSCF_UCODE:
3792 case BNX_DIR_TYPE_EXT_PHY:
3793 case BNX_DIR_TYPE_CCM:
3794 case BNX_DIR_TYPE_ISCSI_BOOT:
3795 case BNX_DIR_TYPE_ISCSI_BOOT_IPV6:
3796 case BNX_DIR_TYPE_ISCSI_BOOT_IPV4N6:
3804 static bool bnxt_dir_type_is_executable(uint16_t dir_type)
3806 return bnxt_dir_type_is_ape_bin_format(dir_type) ||
3807 bnxt_dir_type_is_other_exec_format(dir_type);
3811 bnxt_set_eeprom_op(struct rte_eth_dev *dev,
3812 struct rte_dev_eeprom_info *in_eeprom)
3814 struct bnxt *bp = dev->data->dev_private;
3815 uint8_t index, dir_op;
3816 uint16_t type, ext, ordinal, attr;
3819 rc = is_bnxt_in_error(bp);
3823 PMD_DRV_LOG(INFO, PCI_PRI_FMT " in_eeprom->offset = %d len = %d\n",
3824 bp->pdev->addr.domain, bp->pdev->addr.bus,
3825 bp->pdev->addr.devid, bp->pdev->addr.function,
3826 in_eeprom->offset, in_eeprom->length);
3829 PMD_DRV_LOG(ERR, "NVM write not supported from a VF\n");
3833 type = in_eeprom->magic >> 16;
3835 if (type == 0xffff) { /* special value for directory operations */
3836 index = in_eeprom->magic & 0xff;
3837 dir_op = in_eeprom->magic >> 8;
3841 case 0x0e: /* erase */
3842 if (in_eeprom->offset != ~in_eeprom->magic)
3844 return bnxt_hwrm_erase_nvram_directory(bp, index - 1);
3850 /* Create or re-write an NVM item: */
3851 if (bnxt_dir_type_is_executable(type) == true)
3853 ext = in_eeprom->magic & 0xffff;
3854 ordinal = in_eeprom->offset >> 16;
3855 attr = in_eeprom->offset & 0xffff;
3857 return bnxt_hwrm_flash_nvram(bp, type, ordinal, ext, attr,
3858 in_eeprom->data, in_eeprom->length);
3865 static const struct eth_dev_ops bnxt_dev_ops = {
3866 .dev_infos_get = bnxt_dev_info_get_op,
3867 .dev_close = bnxt_dev_close_op,
3868 .dev_configure = bnxt_dev_configure_op,
3869 .dev_start = bnxt_dev_start_op,
3870 .dev_stop = bnxt_dev_stop_op,
3871 .dev_set_link_up = bnxt_dev_set_link_up_op,
3872 .dev_set_link_down = bnxt_dev_set_link_down_op,
3873 .stats_get = bnxt_stats_get_op,
3874 .stats_reset = bnxt_stats_reset_op,
3875 .rx_queue_setup = bnxt_rx_queue_setup_op,
3876 .rx_queue_release = bnxt_rx_queue_release_op,
3877 .tx_queue_setup = bnxt_tx_queue_setup_op,
3878 .tx_queue_release = bnxt_tx_queue_release_op,
3879 .rx_queue_intr_enable = bnxt_rx_queue_intr_enable_op,
3880 .rx_queue_intr_disable = bnxt_rx_queue_intr_disable_op,
3881 .reta_update = bnxt_reta_update_op,
3882 .reta_query = bnxt_reta_query_op,
3883 .rss_hash_update = bnxt_rss_hash_update_op,
3884 .rss_hash_conf_get = bnxt_rss_hash_conf_get_op,
3885 .link_update = bnxt_link_update_op,
3886 .promiscuous_enable = bnxt_promiscuous_enable_op,
3887 .promiscuous_disable = bnxt_promiscuous_disable_op,
3888 .allmulticast_enable = bnxt_allmulticast_enable_op,
3889 .allmulticast_disable = bnxt_allmulticast_disable_op,
3890 .mac_addr_add = bnxt_mac_addr_add_op,
3891 .mac_addr_remove = bnxt_mac_addr_remove_op,
3892 .flow_ctrl_get = bnxt_flow_ctrl_get_op,
3893 .flow_ctrl_set = bnxt_flow_ctrl_set_op,
3894 .udp_tunnel_port_add = bnxt_udp_tunnel_port_add_op,
3895 .udp_tunnel_port_del = bnxt_udp_tunnel_port_del_op,
3896 .vlan_filter_set = bnxt_vlan_filter_set_op,
3897 .vlan_offload_set = bnxt_vlan_offload_set_op,
3898 .vlan_tpid_set = bnxt_vlan_tpid_set_op,
3899 .vlan_pvid_set = bnxt_vlan_pvid_set_op,
3900 .mtu_set = bnxt_mtu_set_op,
3901 .mac_addr_set = bnxt_set_default_mac_addr_op,
3902 .xstats_get = bnxt_dev_xstats_get_op,
3903 .xstats_get_names = bnxt_dev_xstats_get_names_op,
3904 .xstats_reset = bnxt_dev_xstats_reset_op,
3905 .fw_version_get = bnxt_fw_version_get,
3906 .set_mc_addr_list = bnxt_dev_set_mc_addr_list_op,
3907 .rxq_info_get = bnxt_rxq_info_get_op,
3908 .txq_info_get = bnxt_txq_info_get_op,
3909 .rx_burst_mode_get = bnxt_rx_burst_mode_get,
3910 .tx_burst_mode_get = bnxt_tx_burst_mode_get,
3911 .dev_led_on = bnxt_dev_led_on_op,
3912 .dev_led_off = bnxt_dev_led_off_op,
3913 .rx_queue_start = bnxt_rx_queue_start,
3914 .rx_queue_stop = bnxt_rx_queue_stop,
3915 .tx_queue_start = bnxt_tx_queue_start,
3916 .tx_queue_stop = bnxt_tx_queue_stop,
3917 .flow_ops_get = bnxt_flow_ops_get_op,
3918 .dev_supported_ptypes_get = bnxt_dev_supported_ptypes_get_op,
3919 .get_eeprom_length = bnxt_get_eeprom_length_op,
3920 .get_eeprom = bnxt_get_eeprom_op,
3921 .set_eeprom = bnxt_set_eeprom_op,
3922 .timesync_enable = bnxt_timesync_enable,
3923 .timesync_disable = bnxt_timesync_disable,
3924 .timesync_read_time = bnxt_timesync_read_time,
3925 .timesync_write_time = bnxt_timesync_write_time,
3926 .timesync_adjust_time = bnxt_timesync_adjust_time,
3927 .timesync_read_rx_timestamp = bnxt_timesync_read_rx_timestamp,
3928 .timesync_read_tx_timestamp = bnxt_timesync_read_tx_timestamp,
3931 static uint32_t bnxt_map_reset_regs(struct bnxt *bp, uint32_t reg)
3935 /* Only pre-map the reset GRC registers using window 3 */
3936 rte_write32(reg & 0xfffff000, (uint8_t *)bp->bar0 +
3937 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 8);
3939 offset = BNXT_GRCP_WINDOW_3_BASE + (reg & 0xffc);
3944 int bnxt_map_fw_health_status_regs(struct bnxt *bp)
3946 struct bnxt_error_recovery_info *info = bp->recovery_info;
3947 uint32_t reg_base = 0xffffffff;
3950 /* Only pre-map the monitoring GRC registers using window 2 */
3951 for (i = 0; i < BNXT_FW_STATUS_REG_CNT; i++) {
3952 uint32_t reg = info->status_regs[i];
3954 if (BNXT_FW_STATUS_REG_TYPE(reg) != BNXT_FW_STATUS_REG_TYPE_GRC)
3957 if (reg_base == 0xffffffff)
3958 reg_base = reg & 0xfffff000;
3959 if ((reg & 0xfffff000) != reg_base)
3962 /* Use mask 0xffc as the Lower 2 bits indicates
3963 * address space location
3965 info->mapped_status_regs[i] = BNXT_GRCP_WINDOW_2_BASE +
3969 if (reg_base == 0xffffffff)
3972 rte_write32(reg_base, (uint8_t *)bp->bar0 +
3973 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
3978 static void bnxt_write_fw_reset_reg(struct bnxt *bp, uint32_t index)
3980 struct bnxt_error_recovery_info *info = bp->recovery_info;
3981 uint32_t delay = info->delay_after_reset[index];
3982 uint32_t val = info->reset_reg_val[index];
3983 uint32_t reg = info->reset_reg[index];
3984 uint32_t type, offset;
3987 type = BNXT_FW_STATUS_REG_TYPE(reg);
3988 offset = BNXT_FW_STATUS_REG_OFF(reg);
3991 case BNXT_FW_STATUS_REG_TYPE_CFG:
3992 ret = rte_pci_write_config(bp->pdev, &val, sizeof(val), offset);
3994 PMD_DRV_LOG(ERR, "Failed to write %#x at PCI offset %#x",
3999 case BNXT_FW_STATUS_REG_TYPE_GRC:
4000 offset = bnxt_map_reset_regs(bp, offset);
4001 rte_write32(val, (uint8_t *)bp->bar0 + offset);
4003 case BNXT_FW_STATUS_REG_TYPE_BAR0:
4004 rte_write32(val, (uint8_t *)bp->bar0 + offset);
4007 /* wait on a specific interval of time until core reset is complete */
4009 rte_delay_ms(delay);
4012 static void bnxt_dev_cleanup(struct bnxt *bp)
4014 bp->eth_dev->data->dev_link.link_status = 0;
4015 bp->link_info->link_up = 0;
4016 if (bp->eth_dev->data->dev_started)
4017 bnxt_dev_stop(bp->eth_dev);
4019 bnxt_uninit_resources(bp, true);
4023 bnxt_check_fw_reset_done(struct bnxt *bp)
4025 int timeout = bp->fw_reset_max_msecs;
4030 rc = rte_pci_read_config(bp->pdev, &val, sizeof(val), PCI_SUBSYSTEM_ID_OFFSET);
4032 PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x", PCI_SUBSYSTEM_ID_OFFSET);
4038 } while (timeout--);
4040 if (val == 0xffff) {
4041 PMD_DRV_LOG(ERR, "Firmware reset aborted, PCI config space invalid\n");
4048 static int bnxt_restore_vlan_filters(struct bnxt *bp)
4050 struct rte_eth_dev *dev = bp->eth_dev;
4051 struct rte_vlan_filter_conf *vfc;
4055 for (vlan_id = 1; vlan_id <= RTE_ETHER_MAX_VLAN_ID; vlan_id++) {
4056 vfc = &dev->data->vlan_filter_conf;
4057 vidx = vlan_id / 64;
4058 vbit = vlan_id % 64;
4060 /* Each bit corresponds to a VLAN id */
4061 if (vfc->ids[vidx] & (UINT64_C(1) << vbit)) {
4062 rc = bnxt_add_vlan_filter(bp, vlan_id);
4071 static int bnxt_restore_mac_filters(struct bnxt *bp)
4073 struct rte_eth_dev *dev = bp->eth_dev;
4074 struct rte_eth_dev_info dev_info;
4075 struct rte_ether_addr *addr;
4081 if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp))
4084 rc = bnxt_dev_info_get_op(dev, &dev_info);
4088 /* replay MAC address configuration */
4089 for (i = 1; i < dev_info.max_mac_addrs; i++) {
4090 addr = &dev->data->mac_addrs[i];
4092 /* skip zero address */
4093 if (rte_is_zero_ether_addr(addr))
4097 pool_mask = dev->data->mac_pool_sel[i];
4100 if (pool_mask & 1ULL) {
4101 rc = bnxt_mac_addr_add_op(dev, addr, i, pool);
4107 } while (pool_mask);
4113 static int bnxt_restore_filters(struct bnxt *bp)
4115 struct rte_eth_dev *dev = bp->eth_dev;
4118 if (dev->data->all_multicast) {
4119 ret = bnxt_allmulticast_enable_op(dev);
4123 if (dev->data->promiscuous) {
4124 ret = bnxt_promiscuous_enable_op(dev);
4129 ret = bnxt_restore_mac_filters(bp);
4133 ret = bnxt_restore_vlan_filters(bp);
4134 /* TODO restore other filters as well */
4138 static int bnxt_check_fw_ready(struct bnxt *bp)
4140 int timeout = bp->fw_reset_max_msecs;
4144 rc = bnxt_hwrm_poll_ver_get(bp);
4147 rte_delay_ms(BNXT_FW_READY_WAIT_INTERVAL);
4148 timeout -= BNXT_FW_READY_WAIT_INTERVAL;
4149 } while (rc && timeout > 0);
4152 PMD_DRV_LOG(ERR, "FW is not Ready after reset\n");
4157 static void bnxt_dev_recover(void *arg)
4159 struct bnxt *bp = arg;
4162 pthread_mutex_lock(&bp->err_recovery_lock);
4164 if (!bp->fw_reset_min_msecs) {
4165 rc = bnxt_check_fw_reset_done(bp);
4170 /* Clear Error flag so that device re-init should happen */
4171 bp->flags &= ~BNXT_FLAG_FATAL_ERROR;
4173 rc = bnxt_check_fw_ready(bp);
4177 rc = bnxt_init_resources(bp, true);
4180 "Failed to initialize resources after reset\n");
4183 /* clear reset flag as the device is initialized now */
4184 bp->flags &= ~BNXT_FLAG_FW_RESET;
4186 rc = bnxt_dev_start_op(bp->eth_dev);
4188 PMD_DRV_LOG(ERR, "Failed to start port after reset\n");
4192 rc = bnxt_restore_filters(bp);
4196 PMD_DRV_LOG(INFO, "Recovered from FW reset\n");
4197 pthread_mutex_unlock(&bp->err_recovery_lock);
4201 bnxt_dev_stop(bp->eth_dev);
4203 bp->flags |= BNXT_FLAG_FATAL_ERROR;
4204 bnxt_uninit_resources(bp, false);
4205 pthread_mutex_unlock(&bp->err_recovery_lock);
4206 PMD_DRV_LOG(ERR, "Failed to recover from FW reset\n");
4209 void bnxt_dev_reset_and_resume(void *arg)
4211 struct bnxt *bp = arg;
4212 uint32_t us = US_PER_MS * bp->fw_reset_min_msecs;
4216 bnxt_dev_cleanup(bp);
4218 bnxt_wait_for_device_shutdown(bp);
4220 /* During some fatal firmware error conditions, the PCI config space
4221 * register 0x2e which normally contains the subsystem ID will become
4222 * 0xffff. This register will revert back to the normal value after
4223 * the chip has completed core reset. If we detect this condition,
4224 * we can poll this config register immediately for the value to revert.
4226 if (bp->flags & BNXT_FLAG_FATAL_ERROR) {
4227 rc = rte_pci_read_config(bp->pdev, &val, sizeof(val), PCI_SUBSYSTEM_ID_OFFSET);
4229 PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x", PCI_SUBSYSTEM_ID_OFFSET);
4232 if (val == 0xffff) {
4233 bp->fw_reset_min_msecs = 0;
4238 rc = rte_eal_alarm_set(us, bnxt_dev_recover, (void *)bp);
4240 PMD_DRV_LOG(ERR, "Error setting recovery alarm");
4243 uint32_t bnxt_read_fw_status_reg(struct bnxt *bp, uint32_t index)
4245 struct bnxt_error_recovery_info *info = bp->recovery_info;
4246 uint32_t reg = info->status_regs[index];
4247 uint32_t type, offset, val = 0;
4250 type = BNXT_FW_STATUS_REG_TYPE(reg);
4251 offset = BNXT_FW_STATUS_REG_OFF(reg);
4254 case BNXT_FW_STATUS_REG_TYPE_CFG:
4255 ret = rte_pci_read_config(bp->pdev, &val, sizeof(val), offset);
4257 PMD_DRV_LOG(ERR, "Failed to read PCI offset %#x",
4260 case BNXT_FW_STATUS_REG_TYPE_GRC:
4261 offset = info->mapped_status_regs[index];
4263 case BNXT_FW_STATUS_REG_TYPE_BAR0:
4264 val = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
4272 static int bnxt_fw_reset_all(struct bnxt *bp)
4274 struct bnxt_error_recovery_info *info = bp->recovery_info;
4278 if (info->flags & BNXT_FLAG_ERROR_RECOVERY_HOST) {
4279 /* Reset through master function driver */
4280 for (i = 0; i < info->reg_array_cnt; i++)
4281 bnxt_write_fw_reset_reg(bp, i);
4282 /* Wait for time specified by FW after triggering reset */
4283 rte_delay_ms(info->master_func_wait_period_after_reset);
4284 } else if (info->flags & BNXT_FLAG_ERROR_RECOVERY_CO_CPU) {
4285 /* Reset with the help of Kong processor */
4286 rc = bnxt_hwrm_fw_reset(bp);
4288 PMD_DRV_LOG(ERR, "Failed to reset FW\n");
4294 static void bnxt_fw_reset_cb(void *arg)
4296 struct bnxt *bp = arg;
4297 struct bnxt_error_recovery_info *info = bp->recovery_info;
4300 /* Only Master function can do FW reset */
4301 if (bnxt_is_master_func(bp) &&
4302 bnxt_is_recovery_enabled(bp)) {
4303 rc = bnxt_fw_reset_all(bp);
4305 PMD_DRV_LOG(ERR, "Adapter recovery failed\n");
4310 /* if recovery method is ERROR_RECOVERY_CO_CPU, KONG will send
4311 * EXCEPTION_FATAL_ASYNC event to all the functions
4312 * (including MASTER FUNC). After receiving this Async, all the active
4313 * drivers should treat this case as FW initiated recovery
4315 if (info->flags & BNXT_FLAG_ERROR_RECOVERY_HOST) {
4316 bp->fw_reset_min_msecs = BNXT_MIN_FW_READY_TIMEOUT;
4317 bp->fw_reset_max_msecs = BNXT_MAX_FW_RESET_TIMEOUT;
4319 /* To recover from error */
4320 rte_eal_alarm_set(US_PER_MS, bnxt_dev_reset_and_resume,
4325 /* Driver should poll FW heartbeat, reset_counter with the frequency
4326 * advertised by FW in HWRM_ERROR_RECOVERY_QCFG.
4327 * When the driver detects heartbeat stop or change in reset_counter,
4328 * it has to trigger a reset to recover from the error condition.
4329 * A “master PF” is the function who will have the privilege to
4330 * initiate the chimp reset. The master PF will be elected by the
4331 * firmware and will be notified through async message.
4333 static void bnxt_check_fw_health(void *arg)
4335 struct bnxt *bp = arg;
4336 struct bnxt_error_recovery_info *info = bp->recovery_info;
4337 uint32_t val = 0, wait_msec;
4339 if (!info || !bnxt_is_recovery_enabled(bp) ||
4340 is_bnxt_in_error(bp))
4343 val = bnxt_read_fw_status_reg(bp, BNXT_FW_HEARTBEAT_CNT_REG);
4344 if (val == info->last_heart_beat)
4347 info->last_heart_beat = val;
4349 val = bnxt_read_fw_status_reg(bp, BNXT_FW_RECOVERY_CNT_REG);
4350 if (val != info->last_reset_counter)
4353 info->last_reset_counter = val;
4355 rte_eal_alarm_set(US_PER_MS * info->driver_polling_freq,
4356 bnxt_check_fw_health, (void *)bp);
4360 /* Stop DMA to/from device */
4361 bp->flags |= BNXT_FLAG_FATAL_ERROR;
4362 bp->flags |= BNXT_FLAG_FW_RESET;
4366 PMD_DRV_LOG(ERR, "Detected FW dead condition\n");
4368 if (bnxt_is_master_func(bp))
4369 wait_msec = info->master_func_wait_period;
4371 wait_msec = info->normal_func_wait_period;
4373 rte_eal_alarm_set(US_PER_MS * wait_msec,
4374 bnxt_fw_reset_cb, (void *)bp);
4377 void bnxt_schedule_fw_health_check(struct bnxt *bp)
4379 uint32_t polling_freq;
4381 pthread_mutex_lock(&bp->health_check_lock);
4383 if (!bnxt_is_recovery_enabled(bp))
4386 if (bp->flags & BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED)
4389 polling_freq = bp->recovery_info->driver_polling_freq;
4391 rte_eal_alarm_set(US_PER_MS * polling_freq,
4392 bnxt_check_fw_health, (void *)bp);
4393 bp->flags |= BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED;
4396 pthread_mutex_unlock(&bp->health_check_lock);
4399 static void bnxt_cancel_fw_health_check(struct bnxt *bp)
4401 rte_eal_alarm_cancel(bnxt_check_fw_health, (void *)bp);
4402 bp->flags &= ~BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED;
4405 static bool bnxt_vf_pciid(uint16_t device_id)
4407 switch (device_id) {
4408 case BROADCOM_DEV_ID_57304_VF:
4409 case BROADCOM_DEV_ID_57406_VF:
4410 case BROADCOM_DEV_ID_5731X_VF:
4411 case BROADCOM_DEV_ID_5741X_VF:
4412 case BROADCOM_DEV_ID_57414_VF:
4413 case BROADCOM_DEV_ID_STRATUS_NIC_VF1:
4414 case BROADCOM_DEV_ID_STRATUS_NIC_VF2:
4415 case BROADCOM_DEV_ID_58802_VF:
4416 case BROADCOM_DEV_ID_57500_VF1:
4417 case BROADCOM_DEV_ID_57500_VF2:
4418 case BROADCOM_DEV_ID_58818_VF:
4426 /* Phase 5 device */
4427 static bool bnxt_p5_device(uint16_t device_id)
4429 switch (device_id) {
4430 case BROADCOM_DEV_ID_57508:
4431 case BROADCOM_DEV_ID_57504:
4432 case BROADCOM_DEV_ID_57502:
4433 case BROADCOM_DEV_ID_57508_MF1:
4434 case BROADCOM_DEV_ID_57504_MF1:
4435 case BROADCOM_DEV_ID_57502_MF1:
4436 case BROADCOM_DEV_ID_57508_MF2:
4437 case BROADCOM_DEV_ID_57504_MF2:
4438 case BROADCOM_DEV_ID_57502_MF2:
4439 case BROADCOM_DEV_ID_57500_VF1:
4440 case BROADCOM_DEV_ID_57500_VF2:
4441 case BROADCOM_DEV_ID_58812:
4442 case BROADCOM_DEV_ID_58814:
4443 case BROADCOM_DEV_ID_58818:
4444 case BROADCOM_DEV_ID_58818_VF:
4452 bool bnxt_stratus_device(struct bnxt *bp)
4454 uint16_t device_id = bp->pdev->id.device_id;
4456 switch (device_id) {
4457 case BROADCOM_DEV_ID_STRATUS_NIC:
4458 case BROADCOM_DEV_ID_STRATUS_NIC_VF1:
4459 case BROADCOM_DEV_ID_STRATUS_NIC_VF2:
4467 static int bnxt_map_pci_bars(struct rte_eth_dev *eth_dev)
4469 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
4470 struct bnxt *bp = eth_dev->data->dev_private;
4472 /* enable device (incl. PCI PM wakeup), and bus-mastering */
4473 bp->bar0 = (void *)pci_dev->mem_resource[0].addr;
4474 bp->doorbell_base = (void *)pci_dev->mem_resource[2].addr;
4475 if (!bp->bar0 || !bp->doorbell_base) {
4476 PMD_DRV_LOG(ERR, "Unable to access Hardware\n");
4480 bp->eth_dev = eth_dev;
4486 static int bnxt_alloc_ctx_mem_blk(struct bnxt *bp,
4487 struct bnxt_ctx_pg_info *ctx_pg,
4492 struct bnxt_ring_mem_info *rmem = &ctx_pg->ring_mem;
4493 const struct rte_memzone *mz = NULL;
4494 char mz_name[RTE_MEMZONE_NAMESIZE];
4495 rte_iova_t mz_phys_addr;
4496 uint64_t valid_bits = 0;
4503 rmem->nr_pages = RTE_ALIGN_MUL_CEIL(mem_size, BNXT_PAGE_SIZE) /
4505 rmem->page_size = BNXT_PAGE_SIZE;
4506 rmem->pg_arr = ctx_pg->ctx_pg_arr;
4507 rmem->dma_arr = ctx_pg->ctx_dma_arr;
4508 rmem->flags = BNXT_RMEM_VALID_PTE_FLAG;
4510 valid_bits = PTU_PTE_VALID;
4512 if (rmem->nr_pages > 1) {
4513 snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4514 "bnxt_ctx_pg_tbl%s_%x_%d",
4515 suffix, idx, bp->eth_dev->data->port_id);
4516 mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4517 mz = rte_memzone_lookup(mz_name);
4519 mz = rte_memzone_reserve_aligned(mz_name,
4523 RTE_MEMZONE_SIZE_HINT_ONLY |
4524 RTE_MEMZONE_IOVA_CONTIG,
4530 memset(mz->addr, 0, mz->len);
4531 mz_phys_addr = mz->iova;
4533 rmem->pg_tbl = mz->addr;
4534 rmem->pg_tbl_map = mz_phys_addr;
4535 rmem->pg_tbl_mz = mz;
4538 snprintf(mz_name, RTE_MEMZONE_NAMESIZE, "bnxt_ctx_%s_%x_%d",
4539 suffix, idx, bp->eth_dev->data->port_id);
4540 mz = rte_memzone_lookup(mz_name);
4542 mz = rte_memzone_reserve_aligned(mz_name,
4546 RTE_MEMZONE_SIZE_HINT_ONLY |
4547 RTE_MEMZONE_IOVA_CONTIG,
4553 memset(mz->addr, 0, mz->len);
4554 mz_phys_addr = mz->iova;
4556 for (sz = 0, i = 0; sz < mem_size; sz += BNXT_PAGE_SIZE, i++) {
4557 rmem->pg_arr[i] = ((char *)mz->addr) + sz;
4558 rmem->dma_arr[i] = mz_phys_addr + sz;
4560 if (rmem->nr_pages > 1) {
4561 if (i == rmem->nr_pages - 2 &&
4562 (rmem->flags & BNXT_RMEM_RING_PTE_FLAG))
4563 valid_bits |= PTU_PTE_NEXT_TO_LAST;
4564 else if (i == rmem->nr_pages - 1 &&
4565 (rmem->flags & BNXT_RMEM_RING_PTE_FLAG))
4566 valid_bits |= PTU_PTE_LAST;
4568 rmem->pg_tbl[i] = rte_cpu_to_le_64(rmem->dma_arr[i] |
4574 if (rmem->vmem_size)
4575 rmem->vmem = (void **)mz->addr;
4576 rmem->dma_arr[0] = mz_phys_addr;
4580 static void bnxt_free_ctx_mem(struct bnxt *bp)
4584 if (!bp->ctx || !(bp->ctx->flags & BNXT_CTX_FLAG_INITED))
4587 bp->ctx->flags &= ~BNXT_CTX_FLAG_INITED;
4588 rte_memzone_free(bp->ctx->qp_mem.ring_mem.mz);
4589 rte_memzone_free(bp->ctx->srq_mem.ring_mem.mz);
4590 rte_memzone_free(bp->ctx->cq_mem.ring_mem.mz);
4591 rte_memzone_free(bp->ctx->vnic_mem.ring_mem.mz);
4592 rte_memzone_free(bp->ctx->stat_mem.ring_mem.mz);
4593 rte_memzone_free(bp->ctx->qp_mem.ring_mem.pg_tbl_mz);
4594 rte_memzone_free(bp->ctx->srq_mem.ring_mem.pg_tbl_mz);
4595 rte_memzone_free(bp->ctx->cq_mem.ring_mem.pg_tbl_mz);
4596 rte_memzone_free(bp->ctx->vnic_mem.ring_mem.pg_tbl_mz);
4597 rte_memzone_free(bp->ctx->stat_mem.ring_mem.pg_tbl_mz);
4599 for (i = 0; i < bp->ctx->tqm_fp_rings_count + 1; i++) {
4600 if (bp->ctx->tqm_mem[i])
4601 rte_memzone_free(bp->ctx->tqm_mem[i]->ring_mem.mz);
4608 #define bnxt_roundup(x, y) ((((x) + ((y) - 1)) / (y)) * (y))
4610 #define min_t(type, x, y) ({ \
4611 type __min1 = (x); \
4612 type __min2 = (y); \
4613 __min1 < __min2 ? __min1 : __min2; })
4615 #define max_t(type, x, y) ({ \
4616 type __max1 = (x); \
4617 type __max2 = (y); \
4618 __max1 > __max2 ? __max1 : __max2; })
4620 #define clamp_t(type, _x, min, max) min_t(type, max_t(type, _x, min), max)
4622 int bnxt_alloc_ctx_mem(struct bnxt *bp)
4624 struct bnxt_ctx_pg_info *ctx_pg;
4625 struct bnxt_ctx_mem_info *ctx;
4626 uint32_t mem_size, ena, entries;
4627 uint32_t entries_sp, min;
4630 rc = bnxt_hwrm_func_backing_store_qcaps(bp);
4632 PMD_DRV_LOG(ERR, "Query context mem capability failed\n");
4636 if (!ctx || (ctx->flags & BNXT_CTX_FLAG_INITED))
4639 ctx_pg = &ctx->qp_mem;
4640 ctx_pg->entries = ctx->qp_min_qp1_entries + ctx->qp_max_l2_entries;
4641 if (ctx->qp_entry_size) {
4642 mem_size = ctx->qp_entry_size * ctx_pg->entries;
4643 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "qp_mem", 0);
4648 ctx_pg = &ctx->srq_mem;
4649 ctx_pg->entries = ctx->srq_max_l2_entries;
4650 if (ctx->srq_entry_size) {
4651 mem_size = ctx->srq_entry_size * ctx_pg->entries;
4652 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "srq_mem", 0);
4657 ctx_pg = &ctx->cq_mem;
4658 ctx_pg->entries = ctx->cq_max_l2_entries;
4659 if (ctx->cq_entry_size) {
4660 mem_size = ctx->cq_entry_size * ctx_pg->entries;
4661 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "cq_mem", 0);
4666 ctx_pg = &ctx->vnic_mem;
4667 ctx_pg->entries = ctx->vnic_max_vnic_entries +
4668 ctx->vnic_max_ring_table_entries;
4669 if (ctx->vnic_entry_size) {
4670 mem_size = ctx->vnic_entry_size * ctx_pg->entries;
4671 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "vnic_mem", 0);
4676 ctx_pg = &ctx->stat_mem;
4677 ctx_pg->entries = ctx->stat_max_entries;
4678 if (ctx->stat_entry_size) {
4679 mem_size = ctx->stat_entry_size * ctx_pg->entries;
4680 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "stat_mem", 0);
4685 min = ctx->tqm_min_entries_per_ring;
4687 entries_sp = ctx->qp_max_l2_entries +
4688 ctx->vnic_max_vnic_entries +
4689 2 * ctx->qp_min_qp1_entries + min;
4690 entries_sp = bnxt_roundup(entries_sp, ctx->tqm_entries_multiple);
4692 entries = ctx->qp_max_l2_entries + ctx->qp_min_qp1_entries;
4693 entries = bnxt_roundup(entries, ctx->tqm_entries_multiple);
4694 entries = clamp_t(uint32_t, entries, min,
4695 ctx->tqm_max_entries_per_ring);
4696 for (i = 0, ena = 0; i < ctx->tqm_fp_rings_count + 1; i++) {
4697 /* i=0 is for TQM_SP. i=1 to i=8 applies to RING0 to RING7.
4698 * i > 8 is other ext rings.
4700 ctx_pg = ctx->tqm_mem[i];
4701 ctx_pg->entries = i ? entries : entries_sp;
4702 if (ctx->tqm_entry_size) {
4703 mem_size = ctx->tqm_entry_size * ctx_pg->entries;
4704 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size,
4709 if (i < BNXT_MAX_TQM_LEGACY_RINGS)
4710 ena |= HWRM_FUNC_BACKING_STORE_CFG_INPUT_ENABLES_TQM_SP << i;
4712 ena |= HWRM_FUNC_BACKING_STORE_CFG_INPUT_ENABLES_TQM_RING8;
4715 ena |= FUNC_BACKING_STORE_CFG_INPUT_DFLT_ENABLES;
4716 rc = bnxt_hwrm_func_backing_store_cfg(bp, ena);
4719 "Failed to configure context mem: rc = %d\n", rc);
4721 ctx->flags |= BNXT_CTX_FLAG_INITED;
4726 static int bnxt_alloc_stats_mem(struct bnxt *bp)
4728 struct rte_pci_device *pci_dev = bp->pdev;
4729 char mz_name[RTE_MEMZONE_NAMESIZE];
4730 const struct rte_memzone *mz = NULL;
4731 uint32_t total_alloc_len;
4732 rte_iova_t mz_phys_addr;
4734 if (pci_dev->id.device_id == BROADCOM_DEV_ID_NS2)
4737 snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4738 "bnxt_" PCI_PRI_FMT "-%s", pci_dev->addr.domain,
4739 pci_dev->addr.bus, pci_dev->addr.devid,
4740 pci_dev->addr.function, "rx_port_stats");
4741 mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4742 mz = rte_memzone_lookup(mz_name);
4744 RTE_CACHE_LINE_ROUNDUP(sizeof(struct rx_port_stats) +
4745 sizeof(struct rx_port_stats_ext) + 512);
4747 mz = rte_memzone_reserve(mz_name, total_alloc_len,
4750 RTE_MEMZONE_SIZE_HINT_ONLY |
4751 RTE_MEMZONE_IOVA_CONTIG);
4755 memset(mz->addr, 0, mz->len);
4756 mz_phys_addr = mz->iova;
4758 bp->rx_mem_zone = (const void *)mz;
4759 bp->hw_rx_port_stats = mz->addr;
4760 bp->hw_rx_port_stats_map = mz_phys_addr;
4762 snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4763 "bnxt_" PCI_PRI_FMT "-%s", pci_dev->addr.domain,
4764 pci_dev->addr.bus, pci_dev->addr.devid,
4765 pci_dev->addr.function, "tx_port_stats");
4766 mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4767 mz = rte_memzone_lookup(mz_name);
4769 RTE_CACHE_LINE_ROUNDUP(sizeof(struct tx_port_stats) +
4770 sizeof(struct tx_port_stats_ext) + 512);
4772 mz = rte_memzone_reserve(mz_name,
4776 RTE_MEMZONE_SIZE_HINT_ONLY |
4777 RTE_MEMZONE_IOVA_CONTIG);
4781 memset(mz->addr, 0, mz->len);
4782 mz_phys_addr = mz->iova;
4784 bp->tx_mem_zone = (const void *)mz;
4785 bp->hw_tx_port_stats = mz->addr;
4786 bp->hw_tx_port_stats_map = mz_phys_addr;
4787 bp->flags |= BNXT_FLAG_PORT_STATS;
4789 /* Display extended statistics if FW supports it */
4790 if (bp->hwrm_spec_code < HWRM_SPEC_CODE_1_8_4 ||
4791 bp->hwrm_spec_code == HWRM_SPEC_CODE_1_9_0 ||
4792 !(bp->flags & BNXT_FLAG_EXT_STATS_SUPPORTED))
4795 bp->hw_rx_port_stats_ext = (void *)
4796 ((uint8_t *)bp->hw_rx_port_stats +
4797 sizeof(struct rx_port_stats));
4798 bp->hw_rx_port_stats_ext_map = bp->hw_rx_port_stats_map +
4799 sizeof(struct rx_port_stats);
4800 bp->flags |= BNXT_FLAG_EXT_RX_PORT_STATS;
4802 if (bp->hwrm_spec_code < HWRM_SPEC_CODE_1_9_2 ||
4803 bp->flags & BNXT_FLAG_EXT_STATS_SUPPORTED) {
4804 bp->hw_tx_port_stats_ext = (void *)
4805 ((uint8_t *)bp->hw_tx_port_stats +
4806 sizeof(struct tx_port_stats));
4807 bp->hw_tx_port_stats_ext_map =
4808 bp->hw_tx_port_stats_map +
4809 sizeof(struct tx_port_stats);
4810 bp->flags |= BNXT_FLAG_EXT_TX_PORT_STATS;
4816 static int bnxt_setup_mac_addr(struct rte_eth_dev *eth_dev)
4818 struct bnxt *bp = eth_dev->data->dev_private;
4821 eth_dev->data->mac_addrs = rte_zmalloc("bnxt_mac_addr_tbl",
4822 RTE_ETHER_ADDR_LEN *
4825 if (eth_dev->data->mac_addrs == NULL) {
4826 PMD_DRV_LOG(ERR, "Failed to alloc MAC addr tbl\n");
4830 if (!BNXT_HAS_DFLT_MAC_SET(bp)) {
4834 /* Generate a random MAC address, if none was assigned by PF */
4835 PMD_DRV_LOG(INFO, "VF MAC address not assigned by Host PF\n");
4836 bnxt_eth_hw_addr_random(bp->mac_addr);
4838 "Assign random MAC:%02X:%02X:%02X:%02X:%02X:%02X\n",
4839 bp->mac_addr[0], bp->mac_addr[1], bp->mac_addr[2],
4840 bp->mac_addr[3], bp->mac_addr[4], bp->mac_addr[5]);
4842 rc = bnxt_hwrm_set_mac(bp);
4847 /* Copy the permanent MAC from the FUNC_QCAPS response */
4848 memcpy(ð_dev->data->mac_addrs[0], bp->mac_addr, RTE_ETHER_ADDR_LEN);
4853 static int bnxt_restore_dflt_mac(struct bnxt *bp)
4857 /* MAC is already configured in FW */
4858 if (BNXT_HAS_DFLT_MAC_SET(bp))
4861 /* Restore the old MAC configured */
4862 rc = bnxt_hwrm_set_mac(bp);
4864 PMD_DRV_LOG(ERR, "Failed to restore MAC address\n");
4869 static void bnxt_config_vf_req_fwd(struct bnxt *bp)
4874 memset(bp->pf->vf_req_fwd, 0, sizeof(bp->pf->vf_req_fwd));
4876 if (!(bp->fw_cap & BNXT_FW_CAP_LINK_ADMIN))
4877 BNXT_HWRM_CMD_TO_FORWARD(HWRM_PORT_PHY_QCFG);
4878 BNXT_HWRM_CMD_TO_FORWARD(HWRM_FUNC_CFG);
4879 BNXT_HWRM_CMD_TO_FORWARD(HWRM_FUNC_VF_CFG);
4880 BNXT_HWRM_CMD_TO_FORWARD(HWRM_CFA_L2_FILTER_ALLOC);
4881 BNXT_HWRM_CMD_TO_FORWARD(HWRM_OEM_CMD);
4885 bnxt_get_bp(uint16_t port)
4888 struct rte_eth_dev *dev;
4890 if (!rte_eth_dev_is_valid_port(port)) {
4891 PMD_DRV_LOG(ERR, "Invalid port %d\n", port);
4895 dev = &rte_eth_devices[port];
4896 if (!is_bnxt_supported(dev)) {
4897 PMD_DRV_LOG(ERR, "Device %d not supported\n", port);
4901 bp = (struct bnxt *)dev->data->dev_private;
4902 if (!BNXT_TRUFLOW_EN(bp)) {
4903 PMD_DRV_LOG(ERR, "TRUFLOW not enabled\n");
4911 bnxt_get_svif(uint16_t port_id, bool func_svif,
4912 enum bnxt_ulp_intf_type type)
4914 struct rte_eth_dev *eth_dev;
4917 eth_dev = &rte_eth_devices[port_id];
4918 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
4919 struct bnxt_representor *vfr = eth_dev->data->dev_private;
4923 if (type == BNXT_ULP_INTF_TYPE_VF_REP)
4926 eth_dev = vfr->parent_dev;
4929 bp = eth_dev->data->dev_private;
4931 return func_svif ? bp->func_svif : bp->port_svif;
4935 bnxt_get_iface_mac(uint16_t port, enum bnxt_ulp_intf_type type,
4936 uint8_t *mac, uint8_t *parent_mac)
4938 struct rte_eth_dev *eth_dev;
4941 if (type != BNXT_ULP_INTF_TYPE_TRUSTED_VF &&
4942 type != BNXT_ULP_INTF_TYPE_PF)
4945 eth_dev = &rte_eth_devices[port];
4946 bp = eth_dev->data->dev_private;
4947 memcpy(mac, bp->mac_addr, RTE_ETHER_ADDR_LEN);
4949 if (type == BNXT_ULP_INTF_TYPE_TRUSTED_VF)
4950 memcpy(parent_mac, bp->parent->mac_addr, RTE_ETHER_ADDR_LEN);
4954 bnxt_get_parent_vnic_id(uint16_t port, enum bnxt_ulp_intf_type type)
4956 struct rte_eth_dev *eth_dev;
4959 if (type != BNXT_ULP_INTF_TYPE_TRUSTED_VF)
4962 eth_dev = &rte_eth_devices[port];
4963 bp = eth_dev->data->dev_private;
4965 return bp->parent->vnic;
4968 bnxt_get_vnic_id(uint16_t port, enum bnxt_ulp_intf_type type)
4970 struct rte_eth_dev *eth_dev;
4971 struct bnxt_vnic_info *vnic;
4974 eth_dev = &rte_eth_devices[port];
4975 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
4976 struct bnxt_representor *vfr = eth_dev->data->dev_private;
4980 if (type == BNXT_ULP_INTF_TYPE_VF_REP)
4981 return vfr->dflt_vnic_id;
4983 eth_dev = vfr->parent_dev;
4986 bp = eth_dev->data->dev_private;
4988 vnic = BNXT_GET_DEFAULT_VNIC(bp);
4990 return vnic->fw_vnic_id;
4994 bnxt_get_fw_func_id(uint16_t port, enum bnxt_ulp_intf_type type)
4996 struct rte_eth_dev *eth_dev;
4999 eth_dev = &rte_eth_devices[port];
5000 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
5001 struct bnxt_representor *vfr = eth_dev->data->dev_private;
5005 if (type == BNXT_ULP_INTF_TYPE_VF_REP)
5008 eth_dev = vfr->parent_dev;
5011 bp = eth_dev->data->dev_private;
5016 enum bnxt_ulp_intf_type
5017 bnxt_get_interface_type(uint16_t port)
5019 struct rte_eth_dev *eth_dev;
5022 eth_dev = &rte_eth_devices[port];
5023 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev))
5024 return BNXT_ULP_INTF_TYPE_VF_REP;
5026 bp = eth_dev->data->dev_private;
5028 return BNXT_ULP_INTF_TYPE_PF;
5029 else if (BNXT_VF_IS_TRUSTED(bp))
5030 return BNXT_ULP_INTF_TYPE_TRUSTED_VF;
5031 else if (BNXT_VF(bp))
5032 return BNXT_ULP_INTF_TYPE_VF;
5034 return BNXT_ULP_INTF_TYPE_INVALID;
5038 bnxt_get_phy_port_id(uint16_t port_id)
5040 struct bnxt_representor *vfr;
5041 struct rte_eth_dev *eth_dev;
5044 eth_dev = &rte_eth_devices[port_id];
5045 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
5046 vfr = eth_dev->data->dev_private;
5050 eth_dev = vfr->parent_dev;
5053 bp = eth_dev->data->dev_private;
5055 return BNXT_PF(bp) ? bp->pf->port_id : bp->parent->port_id;
5059 bnxt_get_parif(uint16_t port_id, enum bnxt_ulp_intf_type type)
5061 struct rte_eth_dev *eth_dev;
5064 eth_dev = &rte_eth_devices[port_id];
5065 if (BNXT_ETH_DEV_IS_REPRESENTOR(eth_dev)) {
5066 struct bnxt_representor *vfr = eth_dev->data->dev_private;
5070 if (type == BNXT_ULP_INTF_TYPE_VF_REP)
5071 return vfr->fw_fid - 1;
5073 eth_dev = vfr->parent_dev;
5076 bp = eth_dev->data->dev_private;
5078 return BNXT_PF(bp) ? bp->fw_fid - 1 : bp->parent->fid - 1;
5082 bnxt_get_vport(uint16_t port_id)
5084 return (1 << bnxt_get_phy_port_id(port_id));
5087 static void bnxt_alloc_error_recovery_info(struct bnxt *bp)
5089 struct bnxt_error_recovery_info *info = bp->recovery_info;
5092 if (!(bp->fw_cap & BNXT_FW_CAP_HCOMM_FW_STATUS))
5093 memset(info, 0, sizeof(*info));
5097 if (!(bp->fw_cap & BNXT_FW_CAP_ERROR_RECOVERY))
5100 info = rte_zmalloc("bnxt_hwrm_error_recovery_qcfg",
5103 bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5105 bp->recovery_info = info;
5108 static void bnxt_check_fw_status(struct bnxt *bp)
5112 if (!(bp->recovery_info &&
5113 (bp->fw_cap & BNXT_FW_CAP_HCOMM_FW_STATUS)))
5116 fw_status = bnxt_read_fw_status_reg(bp, BNXT_FW_STATUS_REG);
5117 if (fw_status != BNXT_FW_STATUS_HEALTHY)
5118 PMD_DRV_LOG(ERR, "Firmware not responding, status: %#x\n",
5122 static int bnxt_map_hcomm_fw_status_reg(struct bnxt *bp)
5124 struct bnxt_error_recovery_info *info = bp->recovery_info;
5125 uint32_t status_loc;
5128 rte_write32(HCOMM_STATUS_STRUCT_LOC, (uint8_t *)bp->bar0 +
5129 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
5130 sig_ver = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
5131 BNXT_GRCP_WINDOW_2_BASE +
5132 offsetof(struct hcomm_status,
5134 /* If the signature is absent, then FW does not support this feature */
5135 if ((sig_ver & HCOMM_STATUS_SIGNATURE_MASK) !=
5136 HCOMM_STATUS_SIGNATURE_VAL)
5140 info = rte_zmalloc("bnxt_hwrm_error_recovery_qcfg",
5144 bp->recovery_info = info;
5146 memset(info, 0, sizeof(*info));
5149 status_loc = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
5150 BNXT_GRCP_WINDOW_2_BASE +
5151 offsetof(struct hcomm_status,
5154 /* Only pre-map the FW health status GRC register */
5155 if (BNXT_FW_STATUS_REG_TYPE(status_loc) != BNXT_FW_STATUS_REG_TYPE_GRC)
5158 info->status_regs[BNXT_FW_STATUS_REG] = status_loc;
5159 info->mapped_status_regs[BNXT_FW_STATUS_REG] =
5160 BNXT_GRCP_WINDOW_2_BASE + (status_loc & BNXT_GRCP_OFFSET_MASK);
5162 rte_write32((status_loc & BNXT_GRCP_BASE_MASK), (uint8_t *)bp->bar0 +
5163 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
5165 bp->fw_cap |= BNXT_FW_CAP_HCOMM_FW_STATUS;
5170 /* This function gets the FW version along with the
5171 * capabilities(MAX and current) of the function, vnic,
5172 * error recovery, phy and other chip related info
5174 static int bnxt_get_config(struct bnxt *bp)
5181 rc = bnxt_map_hcomm_fw_status_reg(bp);
5185 rc = bnxt_hwrm_ver_get(bp, DFLT_HWRM_CMD_TIMEOUT);
5187 bnxt_check_fw_status(bp);
5191 rc = bnxt_hwrm_func_reset(bp);
5195 rc = bnxt_hwrm_vnic_qcaps(bp);
5199 rc = bnxt_hwrm_queue_qportcfg(bp);
5203 /* Get the MAX capabilities for this function.
5204 * This function also allocates context memory for TQM rings and
5205 * informs the firmware about this allocated backing store memory.
5207 rc = bnxt_hwrm_func_qcaps(bp);
5211 rc = bnxt_hwrm_func_qcfg(bp, &mtu);
5215 rc = bnxt_hwrm_cfa_adv_flow_mgmt_qcaps(bp);
5219 bnxt_hwrm_port_mac_qcfg(bp);
5221 bnxt_hwrm_parent_pf_qcfg(bp);
5223 bnxt_hwrm_port_phy_qcaps(bp);
5225 bnxt_alloc_error_recovery_info(bp);
5226 /* Get the adapter error recovery support info */
5227 rc = bnxt_hwrm_error_recovery_qcfg(bp);
5229 bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5231 bnxt_hwrm_port_led_qcaps(bp);
5237 bnxt_init_locks(struct bnxt *bp)
5241 err = pthread_mutex_init(&bp->flow_lock, NULL);
5243 PMD_DRV_LOG(ERR, "Unable to initialize flow_lock\n");
5247 err = pthread_mutex_init(&bp->def_cp_lock, NULL);
5249 PMD_DRV_LOG(ERR, "Unable to initialize def_cp_lock\n");
5253 err = pthread_mutex_init(&bp->health_check_lock, NULL);
5255 PMD_DRV_LOG(ERR, "Unable to initialize health_check_lock\n");
5259 err = pthread_mutex_init(&bp->err_recovery_lock, NULL);
5261 PMD_DRV_LOG(ERR, "Unable to initialize err_recovery_lock\n");
5266 static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev)
5270 rc = bnxt_get_config(bp);
5274 if (!reconfig_dev) {
5275 rc = bnxt_setup_mac_addr(bp->eth_dev);
5279 rc = bnxt_restore_dflt_mac(bp);
5284 bnxt_config_vf_req_fwd(bp);
5286 rc = bnxt_hwrm_func_driver_register(bp);
5288 PMD_DRV_LOG(ERR, "Failed to register driver");
5293 if (bp->pdev->max_vfs) {
5294 rc = bnxt_hwrm_allocate_vfs(bp, bp->pdev->max_vfs);
5296 PMD_DRV_LOG(ERR, "Failed to allocate VFs\n");
5300 rc = bnxt_hwrm_allocate_pf_only(bp);
5303 "Failed to allocate PF resources");
5309 rc = bnxt_alloc_mem(bp, reconfig_dev);
5313 rc = bnxt_setup_int(bp);
5317 rc = bnxt_request_int(bp);
5321 rc = bnxt_init_ctx_mem(bp);
5323 PMD_DRV_LOG(ERR, "Failed to init adv_flow_counters\n");
5331 bnxt_parse_devarg_accum_stats(__rte_unused const char *key,
5332 const char *value, void *opaque_arg)
5334 struct bnxt *bp = opaque_arg;
5335 unsigned long accum_stats;
5338 if (!value || !opaque_arg) {
5340 "Invalid parameter passed to accum-stats devargs.\n");
5344 accum_stats = strtoul(value, &end, 10);
5345 if (end == NULL || *end != '\0' ||
5346 (accum_stats == ULONG_MAX && errno == ERANGE)) {
5348 "Invalid parameter passed to accum-stats devargs.\n");
5352 if (BNXT_DEVARG_ACCUM_STATS_INVALID(accum_stats)) {
5354 "Invalid value passed to accum-stats devargs.\n");
5359 bp->flags2 |= BNXT_FLAGS2_ACCUM_STATS_EN;
5360 PMD_DRV_LOG(INFO, "Host-based accum-stats feature enabled.\n");
5362 bp->flags2 &= ~BNXT_FLAGS2_ACCUM_STATS_EN;
5363 PMD_DRV_LOG(INFO, "Host-based accum-stats feature disabled.\n");
5370 bnxt_parse_devarg_flow_xstat(__rte_unused const char *key,
5371 const char *value, void *opaque_arg)
5373 struct bnxt *bp = opaque_arg;
5374 unsigned long flow_xstat;
5377 if (!value || !opaque_arg) {
5379 "Invalid parameter passed to flow_xstat devarg.\n");
5383 flow_xstat = strtoul(value, &end, 10);
5384 if (end == NULL || *end != '\0' ||
5385 (flow_xstat == ULONG_MAX && errno == ERANGE)) {
5387 "Invalid parameter passed to flow_xstat devarg.\n");
5391 if (BNXT_DEVARG_FLOW_XSTAT_INVALID(flow_xstat)) {
5393 "Invalid value passed to flow_xstat devarg.\n");
5397 bp->flags |= BNXT_FLAG_FLOW_XSTATS_EN;
5398 if (BNXT_FLOW_XSTATS_EN(bp))
5399 PMD_DRV_LOG(INFO, "flow_xstat feature enabled.\n");
5405 bnxt_parse_devarg_max_num_kflows(__rte_unused const char *key,
5406 const char *value, void *opaque_arg)
5408 struct bnxt *bp = opaque_arg;
5409 unsigned long max_num_kflows;
5412 if (!value || !opaque_arg) {
5414 "Invalid parameter passed to max_num_kflows devarg.\n");
5418 max_num_kflows = strtoul(value, &end, 10);
5419 if (end == NULL || *end != '\0' ||
5420 (max_num_kflows == ULONG_MAX && errno == ERANGE)) {
5422 "Invalid parameter passed to max_num_kflows devarg.\n");
5426 if (bnxt_devarg_max_num_kflow_invalid(max_num_kflows)) {
5428 "Invalid value passed to max_num_kflows devarg.\n");
5432 bp->max_num_kflows = max_num_kflows;
5433 if (bp->max_num_kflows)
5434 PMD_DRV_LOG(INFO, "max_num_kflows set as %ldK.\n",
5441 bnxt_parse_devarg_app_id(__rte_unused const char *key,
5442 const char *value, void *opaque_arg)
5444 struct bnxt *bp = opaque_arg;
5445 unsigned long app_id;
5448 if (!value || !opaque_arg) {
5450 "Invalid parameter passed to app-id "
5455 app_id = strtoul(value, &end, 10);
5456 if (end == NULL || *end != '\0' ||
5457 (app_id == ULONG_MAX && errno == ERANGE)) {
5459 "Invalid parameter passed to app_id "
5464 if (BNXT_DEVARG_APP_ID_INVALID(app_id)) {
5465 PMD_DRV_LOG(ERR, "Invalid app-id(%d) devargs.\n",
5470 bp->app_id = app_id;
5471 PMD_DRV_LOG(INFO, "app-id=%d feature enabled.\n", (uint16_t)app_id);
5477 bnxt_parse_devarg_rep_is_pf(__rte_unused const char *key,
5478 const char *value, void *opaque_arg)
5480 struct bnxt_representor *vfr_bp = opaque_arg;
5481 unsigned long rep_is_pf;
5484 if (!value || !opaque_arg) {
5486 "Invalid parameter passed to rep_is_pf devargs.\n");
5490 rep_is_pf = strtoul(value, &end, 10);
5491 if (end == NULL || *end != '\0' ||
5492 (rep_is_pf == ULONG_MAX && errno == ERANGE)) {
5494 "Invalid parameter passed to rep_is_pf devargs.\n");
5498 if (BNXT_DEVARG_REP_IS_PF_INVALID(rep_is_pf)) {
5500 "Invalid value passed to rep_is_pf devargs.\n");
5504 vfr_bp->flags |= rep_is_pf;
5505 if (BNXT_REP_PF(vfr_bp))
5506 PMD_DRV_LOG(INFO, "PF representor\n");
5508 PMD_DRV_LOG(INFO, "VF representor\n");
5514 bnxt_parse_devarg_rep_based_pf(__rte_unused const char *key,
5515 const char *value, void *opaque_arg)
5517 struct bnxt_representor *vfr_bp = opaque_arg;
5518 unsigned long rep_based_pf;
5521 if (!value || !opaque_arg) {
5523 "Invalid parameter passed to rep_based_pf "
5528 rep_based_pf = strtoul(value, &end, 10);
5529 if (end == NULL || *end != '\0' ||
5530 (rep_based_pf == ULONG_MAX && errno == ERANGE)) {
5532 "Invalid parameter passed to rep_based_pf "
5537 if (BNXT_DEVARG_REP_BASED_PF_INVALID(rep_based_pf)) {
5539 "Invalid value passed to rep_based_pf devargs.\n");
5543 vfr_bp->rep_based_pf = rep_based_pf;
5544 vfr_bp->flags |= BNXT_REP_BASED_PF_VALID;
5546 PMD_DRV_LOG(INFO, "rep-based-pf = %d\n", vfr_bp->rep_based_pf);
5552 bnxt_parse_devarg_rep_q_r2f(__rte_unused const char *key,
5553 const char *value, void *opaque_arg)
5555 struct bnxt_representor *vfr_bp = opaque_arg;
5556 unsigned long rep_q_r2f;
5559 if (!value || !opaque_arg) {
5561 "Invalid parameter passed to rep_q_r2f "
5566 rep_q_r2f = strtoul(value, &end, 10);
5567 if (end == NULL || *end != '\0' ||
5568 (rep_q_r2f == ULONG_MAX && errno == ERANGE)) {
5570 "Invalid parameter passed to rep_q_r2f "
5575 if (BNXT_DEVARG_REP_Q_R2F_INVALID(rep_q_r2f)) {
5577 "Invalid value passed to rep_q_r2f devargs.\n");
5581 vfr_bp->rep_q_r2f = rep_q_r2f;
5582 vfr_bp->flags |= BNXT_REP_Q_R2F_VALID;
5583 PMD_DRV_LOG(INFO, "rep-q-r2f = %d\n", vfr_bp->rep_q_r2f);
5589 bnxt_parse_devarg_rep_q_f2r(__rte_unused const char *key,
5590 const char *value, void *opaque_arg)
5592 struct bnxt_representor *vfr_bp = opaque_arg;
5593 unsigned long rep_q_f2r;
5596 if (!value || !opaque_arg) {
5598 "Invalid parameter passed to rep_q_f2r "
5603 rep_q_f2r = strtoul(value, &end, 10);
5604 if (end == NULL || *end != '\0' ||
5605 (rep_q_f2r == ULONG_MAX && errno == ERANGE)) {
5607 "Invalid parameter passed to rep_q_f2r "
5612 if (BNXT_DEVARG_REP_Q_F2R_INVALID(rep_q_f2r)) {
5614 "Invalid value passed to rep_q_f2r devargs.\n");
5618 vfr_bp->rep_q_f2r = rep_q_f2r;
5619 vfr_bp->flags |= BNXT_REP_Q_F2R_VALID;
5620 PMD_DRV_LOG(INFO, "rep-q-f2r = %d\n", vfr_bp->rep_q_f2r);
5626 bnxt_parse_devarg_rep_fc_r2f(__rte_unused const char *key,
5627 const char *value, void *opaque_arg)
5629 struct bnxt_representor *vfr_bp = opaque_arg;
5630 unsigned long rep_fc_r2f;
5633 if (!value || !opaque_arg) {
5635 "Invalid parameter passed to rep_fc_r2f "
5640 rep_fc_r2f = strtoul(value, &end, 10);
5641 if (end == NULL || *end != '\0' ||
5642 (rep_fc_r2f == ULONG_MAX && errno == ERANGE)) {
5644 "Invalid parameter passed to rep_fc_r2f "
5649 if (BNXT_DEVARG_REP_FC_R2F_INVALID(rep_fc_r2f)) {
5651 "Invalid value passed to rep_fc_r2f devargs.\n");
5655 vfr_bp->flags |= BNXT_REP_FC_R2F_VALID;
5656 vfr_bp->rep_fc_r2f = rep_fc_r2f;
5657 PMD_DRV_LOG(INFO, "rep-fc-r2f = %lu\n", rep_fc_r2f);
5663 bnxt_parse_devarg_rep_fc_f2r(__rte_unused const char *key,
5664 const char *value, void *opaque_arg)
5666 struct bnxt_representor *vfr_bp = opaque_arg;
5667 unsigned long rep_fc_f2r;
5670 if (!value || !opaque_arg) {
5672 "Invalid parameter passed to rep_fc_f2r "
5677 rep_fc_f2r = strtoul(value, &end, 10);
5678 if (end == NULL || *end != '\0' ||
5679 (rep_fc_f2r == ULONG_MAX && errno == ERANGE)) {
5681 "Invalid parameter passed to rep_fc_f2r "
5686 if (BNXT_DEVARG_REP_FC_F2R_INVALID(rep_fc_f2r)) {
5688 "Invalid value passed to rep_fc_f2r devargs.\n");
5692 vfr_bp->flags |= BNXT_REP_FC_F2R_VALID;
5693 vfr_bp->rep_fc_f2r = rep_fc_f2r;
5694 PMD_DRV_LOG(INFO, "rep-fc-f2r = %lu\n", rep_fc_f2r);
5700 bnxt_parse_dev_args(struct bnxt *bp, struct rte_devargs *devargs)
5702 struct rte_kvargs *kvlist;
5705 if (devargs == NULL)
5708 kvlist = rte_kvargs_parse(devargs->args, bnxt_dev_args);
5713 * Handler for "flow_xstat" devarg.
5714 * Invoked as for ex: "-a 0000:00:0d.0,flow_xstat=1"
5716 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_FLOW_XSTAT,
5717 bnxt_parse_devarg_flow_xstat, bp);
5722 * Handler for "accum-stats" devarg.
5723 * Invoked as for ex: "-a 0000:00:0d.0,accum-stats=1"
5725 rte_kvargs_process(kvlist, BNXT_DEVARG_ACCUM_STATS,
5726 bnxt_parse_devarg_accum_stats, bp);
5728 * Handler for "max_num_kflows" devarg.
5729 * Invoked as for ex: "-a 000:00:0d.0,max_num_kflows=32"
5731 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_MAX_NUM_KFLOWS,
5732 bnxt_parse_devarg_max_num_kflows, bp);
5738 * Handler for "app-id" devarg.
5739 * Invoked as for ex: "-a 000:00:0d.0,app-id=1"
5741 rte_kvargs_process(kvlist, BNXT_DEVARG_APP_ID,
5742 bnxt_parse_devarg_app_id, bp);
5744 rte_kvargs_free(kvlist);
5748 static int bnxt_alloc_switch_domain(struct bnxt *bp)
5752 if (BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) {
5753 rc = rte_eth_switch_domain_alloc(&bp->switch_domain_id);
5756 "Failed to alloc switch domain: %d\n", rc);
5759 "Switch domain allocated %d\n",
5760 bp->switch_domain_id);
5766 /* Allocate and initialize various fields in bnxt struct that
5767 * need to be allocated/destroyed only once in the lifetime of the driver
5769 static int bnxt_drv_init(struct rte_eth_dev *eth_dev)
5771 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
5772 struct bnxt *bp = eth_dev->data->dev_private;
5775 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
5777 if (bnxt_vf_pciid(pci_dev->id.device_id))
5778 bp->flags |= BNXT_FLAG_VF;
5780 if (bnxt_p5_device(pci_dev->id.device_id))
5781 bp->flags |= BNXT_FLAG_CHIP_P5;
5783 if (pci_dev->id.device_id == BROADCOM_DEV_ID_58802 ||
5784 pci_dev->id.device_id == BROADCOM_DEV_ID_58804 ||
5785 pci_dev->id.device_id == BROADCOM_DEV_ID_58808 ||
5786 pci_dev->id.device_id == BROADCOM_DEV_ID_58802_VF)
5787 bp->flags |= BNXT_FLAG_STINGRAY;
5789 if (BNXT_TRUFLOW_EN(bp)) {
5790 /* extra mbuf field is required to store CFA code from mark */
5791 static const struct rte_mbuf_dynfield bnxt_cfa_code_dynfield_desc = {
5792 .name = RTE_PMD_BNXT_CFA_CODE_DYNFIELD_NAME,
5793 .size = sizeof(bnxt_cfa_code_dynfield_t),
5794 .align = __alignof__(bnxt_cfa_code_dynfield_t),
5796 bnxt_cfa_code_dynfield_offset =
5797 rte_mbuf_dynfield_register(&bnxt_cfa_code_dynfield_desc);
5798 if (bnxt_cfa_code_dynfield_offset < 0) {
5800 "Failed to register mbuf field for TruFlow mark\n");
5805 rc = bnxt_map_pci_bars(eth_dev);
5808 "Failed to initialize board rc: %x\n", rc);
5812 rc = bnxt_alloc_pf_info(bp);
5816 rc = bnxt_alloc_link_info(bp);
5820 rc = bnxt_alloc_parent_info(bp);
5824 rc = bnxt_alloc_hwrm_resources(bp);
5827 "Failed to allocate response buffer rc: %x\n", rc);
5830 rc = bnxt_alloc_leds_info(bp);
5834 rc = bnxt_alloc_cos_queues(bp);
5838 rc = bnxt_init_locks(bp);
5842 rc = bnxt_alloc_switch_domain(bp);
5850 bnxt_dev_init(struct rte_eth_dev *eth_dev, void *params __rte_unused)
5852 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
5853 static int version_printed;
5857 if (version_printed++ == 0)
5858 PMD_DRV_LOG(INFO, "%s\n", bnxt_version);
5860 eth_dev->dev_ops = &bnxt_dev_ops;
5861 eth_dev->rx_queue_count = bnxt_rx_queue_count_op;
5862 eth_dev->rx_descriptor_status = bnxt_rx_descriptor_status_op;
5863 eth_dev->tx_descriptor_status = bnxt_tx_descriptor_status_op;
5864 eth_dev->rx_pkt_burst = &bnxt_recv_pkts;
5865 eth_dev->tx_pkt_burst = &bnxt_xmit_pkts;
5868 * For secondary processes, we don't initialise any further
5869 * as primary has already done this work.
5871 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
5874 rte_eth_copy_pci_info(eth_dev, pci_dev);
5875 eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
5877 bp = eth_dev->data->dev_private;
5879 /* Parse dev arguments passed on when starting the DPDK application. */
5880 rc = bnxt_parse_dev_args(bp, pci_dev->device.devargs);
5884 rc = bnxt_drv_init(eth_dev);
5888 rc = bnxt_init_resources(bp, false);
5892 rc = bnxt_alloc_stats_mem(bp);
5897 "Found %s device at mem %" PRIX64 ", node addr %pM\n",
5899 pci_dev->mem_resource[0].phys_addr,
5900 pci_dev->mem_resource[0].addr);
5905 bnxt_dev_uninit(eth_dev);
5910 static void bnxt_free_ctx_mem_buf(struct bnxt_ctx_mem_buf_info *ctx)
5919 ctx->dma = RTE_BAD_IOVA;
5920 ctx->ctx_id = BNXT_CTX_VAL_INVAL;
5923 static void bnxt_unregister_fc_ctx_mem(struct bnxt *bp)
5925 bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_RX,
5926 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
5927 bp->flow_stat->rx_fc_out_tbl.ctx_id,
5928 bp->flow_stat->max_fc,
5931 bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_TX,
5932 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
5933 bp->flow_stat->tx_fc_out_tbl.ctx_id,
5934 bp->flow_stat->max_fc,
5937 if (bp->flow_stat->rx_fc_in_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5938 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->rx_fc_in_tbl.ctx_id);
5939 bp->flow_stat->rx_fc_in_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5941 if (bp->flow_stat->rx_fc_out_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5942 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->rx_fc_out_tbl.ctx_id);
5943 bp->flow_stat->rx_fc_out_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5945 if (bp->flow_stat->tx_fc_in_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5946 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->tx_fc_in_tbl.ctx_id);
5947 bp->flow_stat->tx_fc_in_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5949 if (bp->flow_stat->tx_fc_out_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5950 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->tx_fc_out_tbl.ctx_id);
5951 bp->flow_stat->tx_fc_out_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5954 static void bnxt_uninit_fc_ctx_mem(struct bnxt *bp)
5956 bnxt_unregister_fc_ctx_mem(bp);
5958 bnxt_free_ctx_mem_buf(&bp->flow_stat->rx_fc_in_tbl);
5959 bnxt_free_ctx_mem_buf(&bp->flow_stat->rx_fc_out_tbl);
5960 bnxt_free_ctx_mem_buf(&bp->flow_stat->tx_fc_in_tbl);
5961 bnxt_free_ctx_mem_buf(&bp->flow_stat->tx_fc_out_tbl);
5964 static void bnxt_uninit_ctx_mem(struct bnxt *bp)
5966 if (BNXT_FLOW_XSTATS_EN(bp))
5967 bnxt_uninit_fc_ctx_mem(bp);
5971 bnxt_free_error_recovery_info(struct bnxt *bp)
5973 rte_free(bp->recovery_info);
5974 bp->recovery_info = NULL;
5975 bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5979 bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev)
5984 bnxt_free_mem(bp, reconfig_dev);
5986 bnxt_hwrm_func_buf_unrgtr(bp);
5987 if (bp->pf != NULL) {
5988 rte_free(bp->pf->vf_req_buf);
5989 bp->pf->vf_req_buf = NULL;
5992 rc = bnxt_hwrm_func_driver_unregister(bp, 0);
5993 bp->flags &= ~BNXT_FLAG_REGISTERED;
5994 bnxt_free_ctx_mem(bp);
5995 if (!reconfig_dev) {
5996 bnxt_free_hwrm_resources(bp);
5997 bnxt_free_error_recovery_info(bp);
6000 bnxt_uninit_ctx_mem(bp);
6002 bnxt_free_flow_stats_info(bp);
6003 if (bp->rep_info != NULL)
6004 bnxt_free_switch_domain(bp);
6005 bnxt_free_rep_info(bp);
6006 rte_free(bp->ptp_cfg);
6012 bnxt_dev_uninit(struct rte_eth_dev *eth_dev)
6014 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
6017 PMD_DRV_LOG(DEBUG, "Calling Device uninit\n");
6019 if (eth_dev->state != RTE_ETH_DEV_UNUSED)
6020 bnxt_dev_close_op(eth_dev);
6025 static int bnxt_pci_remove_dev_with_reps(struct rte_eth_dev *eth_dev)
6027 struct bnxt *bp = eth_dev->data->dev_private;
6028 struct rte_eth_dev *vf_rep_eth_dev;
6034 for (i = 0; i < bp->num_reps; i++) {
6035 vf_rep_eth_dev = bp->rep_info[i].vfr_eth_dev;
6036 if (!vf_rep_eth_dev)
6038 PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR pci remove\n",
6039 vf_rep_eth_dev->data->port_id);
6040 rte_eth_dev_destroy(vf_rep_eth_dev, bnxt_representor_uninit);
6042 PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci remove\n",
6043 eth_dev->data->port_id);
6044 ret = rte_eth_dev_destroy(eth_dev, bnxt_dev_uninit);
6049 static void bnxt_free_rep_info(struct bnxt *bp)
6051 rte_free(bp->rep_info);
6052 bp->rep_info = NULL;
6053 rte_free(bp->cfa_code_map);
6054 bp->cfa_code_map = NULL;
6057 static int bnxt_init_rep_info(struct bnxt *bp)
6064 bp->rep_info = rte_zmalloc("bnxt_rep_info",
6065 sizeof(bp->rep_info[0]) * BNXT_MAX_VF_REPS,
6067 if (!bp->rep_info) {
6068 PMD_DRV_LOG(ERR, "Failed to alloc memory for rep info\n");
6071 bp->cfa_code_map = rte_zmalloc("bnxt_cfa_code_map",
6072 sizeof(*bp->cfa_code_map) *
6073 BNXT_MAX_CFA_CODE, 0);
6074 if (!bp->cfa_code_map) {
6075 PMD_DRV_LOG(ERR, "Failed to alloc memory for cfa_code_map\n");
6076 bnxt_free_rep_info(bp);
6080 for (i = 0; i < BNXT_MAX_CFA_CODE; i++)
6081 bp->cfa_code_map[i] = BNXT_VF_IDX_INVALID;
6083 rc = pthread_mutex_init(&bp->rep_info->vfr_lock, NULL);
6085 PMD_DRV_LOG(ERR, "Unable to initialize vfr_lock\n");
6086 bnxt_free_rep_info(bp);
6090 rc = pthread_mutex_init(&bp->rep_info->vfr_start_lock, NULL);
6092 PMD_DRV_LOG(ERR, "Unable to initialize vfr_start_lock\n");
6093 bnxt_free_rep_info(bp);
6100 static int bnxt_rep_port_probe(struct rte_pci_device *pci_dev,
6101 struct rte_eth_devargs *eth_da,
6102 struct rte_eth_dev *backing_eth_dev,
6103 const char *dev_args)
6105 struct rte_eth_dev *vf_rep_eth_dev;
6106 char name[RTE_ETH_NAME_MAX_LEN];
6107 struct bnxt *backing_bp;
6110 struct rte_kvargs *kvlist = NULL;
6112 if (eth_da->type == RTE_ETH_REPRESENTOR_NONE)
6114 if (eth_da->type != RTE_ETH_REPRESENTOR_VF) {
6115 PMD_DRV_LOG(ERR, "unsupported representor type %d\n",
6119 num_rep = eth_da->nb_representor_ports;
6120 if (num_rep > BNXT_MAX_VF_REPS) {
6121 PMD_DRV_LOG(ERR, "nb_representor_ports = %d > %d MAX VF REPS\n",
6122 num_rep, BNXT_MAX_VF_REPS);
6126 if (num_rep >= RTE_MAX_ETHPORTS) {
6128 "nb_representor_ports = %d > %d MAX ETHPORTS\n",
6129 num_rep, RTE_MAX_ETHPORTS);
6133 backing_bp = backing_eth_dev->data->dev_private;
6135 if (!(BNXT_PF(backing_bp) || BNXT_VF_IS_TRUSTED(backing_bp))) {
6137 "Not a PF or trusted VF. No Representor support\n");
6138 /* Returning an error is not an option.
6139 * Applications are not handling this correctly
6144 if (bnxt_init_rep_info(backing_bp))
6147 for (i = 0; i < num_rep; i++) {
6148 struct bnxt_representor representor = {
6149 .vf_id = eth_da->representor_ports[i],
6150 .switch_domain_id = backing_bp->switch_domain_id,
6151 .parent_dev = backing_eth_dev
6154 if (representor.vf_id >= BNXT_MAX_VF_REPS) {
6155 PMD_DRV_LOG(ERR, "VF-Rep id %d >= %d MAX VF ID\n",
6156 representor.vf_id, BNXT_MAX_VF_REPS);
6160 /* representor port net_bdf_port */
6161 snprintf(name, sizeof(name), "net_%s_representor_%d",
6162 pci_dev->device.name, eth_da->representor_ports[i]);
6164 kvlist = rte_kvargs_parse(dev_args, bnxt_dev_args);
6167 * Handler for "rep_is_pf" devarg.
6168 * Invoked as for ex: "-a 000:00:0d.0,
6169 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6171 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_IS_PF,
6172 bnxt_parse_devarg_rep_is_pf,
6173 (void *)&representor);
6179 * Handler for "rep_based_pf" devarg.
6180 * Invoked as for ex: "-a 000:00:0d.0,
6181 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6183 ret = rte_kvargs_process(kvlist,
6184 BNXT_DEVARG_REP_BASED_PF,
6185 bnxt_parse_devarg_rep_based_pf,
6186 (void *)&representor);
6192 * Handler for "rep_based_pf" devarg.
6193 * Invoked as for ex: "-a 000:00:0d.0,
6194 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6196 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_Q_R2F,
6197 bnxt_parse_devarg_rep_q_r2f,
6198 (void *)&representor);
6204 * Handler for "rep_based_pf" devarg.
6205 * Invoked as for ex: "-a 000:00:0d.0,
6206 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6208 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_Q_F2R,
6209 bnxt_parse_devarg_rep_q_f2r,
6210 (void *)&representor);
6216 * Handler for "rep_based_pf" devarg.
6217 * Invoked as for ex: "-a 000:00:0d.0,
6218 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6220 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_FC_R2F,
6221 bnxt_parse_devarg_rep_fc_r2f,
6222 (void *)&representor);
6228 * Handler for "rep_based_pf" devarg.
6229 * Invoked as for ex: "-a 000:00:0d.0,
6230 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6232 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_FC_F2R,
6233 bnxt_parse_devarg_rep_fc_f2r,
6234 (void *)&representor);
6241 ret = rte_eth_dev_create(&pci_dev->device, name,
6242 sizeof(struct bnxt_representor),
6244 bnxt_representor_init,
6247 PMD_DRV_LOG(ERR, "failed to create bnxt vf "
6248 "representor %s.", name);
6252 vf_rep_eth_dev = rte_eth_dev_allocated(name);
6253 if (!vf_rep_eth_dev) {
6254 PMD_DRV_LOG(ERR, "Failed to find the eth_dev"
6255 " for VF-Rep: %s.", name);
6260 PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR pci probe\n",
6261 backing_eth_dev->data->port_id);
6262 backing_bp->rep_info[representor.vf_id].vfr_eth_dev =
6264 backing_bp->num_reps++;
6268 rte_kvargs_free(kvlist);
6272 /* If num_rep > 1, then rollback already created
6273 * ports, since we'll be failing the probe anyway
6276 bnxt_pci_remove_dev_with_reps(backing_eth_dev);
6278 rte_kvargs_free(kvlist);
6283 static int bnxt_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
6284 struct rte_pci_device *pci_dev)
6286 struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
6287 struct rte_eth_dev *backing_eth_dev;
6291 if (pci_dev->device.devargs) {
6292 ret = rte_eth_devargs_parse(pci_dev->device.devargs->args,
6298 num_rep = eth_da.nb_representor_ports;
6299 PMD_DRV_LOG(DEBUG, "nb_representor_ports = %d\n",
6302 /* We could come here after first level of probe is already invoked
6303 * as part of an application bringup(OVS-DPDK vswitchd), so first check
6304 * for already allocated eth_dev for the backing device (PF/Trusted VF)
6306 backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6307 if (backing_eth_dev == NULL) {
6308 ret = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
6309 sizeof(struct bnxt),
6310 eth_dev_pci_specific_init, pci_dev,
6311 bnxt_dev_init, NULL);
6313 if (ret || !num_rep)
6316 backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6318 PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci probe\n",
6319 backing_eth_dev->data->port_id);
6324 /* probe representor ports now */
6325 ret = bnxt_rep_port_probe(pci_dev, ð_da, backing_eth_dev,
6326 pci_dev->device.devargs->args);
6331 static int bnxt_pci_remove(struct rte_pci_device *pci_dev)
6333 struct rte_eth_dev *eth_dev;
6335 eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6337 return 0; /* Invoked typically only by OVS-DPDK, by the
6338 * time it comes here the eth_dev is already
6339 * deleted by rte_eth_dev_close(), so returning
6340 * +ve value will at least help in proper cleanup
6343 PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci remove\n", eth_dev->data->port_id);
6344 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
6345 if (eth_dev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
6346 return rte_eth_dev_destroy(eth_dev,
6347 bnxt_representor_uninit);
6349 return rte_eth_dev_destroy(eth_dev,
6352 return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
6356 static struct rte_pci_driver bnxt_rte_pmd = {
6357 .id_table = bnxt_pci_id_map,
6358 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
6359 RTE_PCI_DRV_PROBE_AGAIN, /* Needed in case of VF-REPs
6362 .probe = bnxt_pci_probe,
6363 .remove = bnxt_pci_remove,
6367 is_device_supported(struct rte_eth_dev *dev, struct rte_pci_driver *drv)
6369 if (strcmp(dev->device->driver->name, drv->driver.name))
6375 bool is_bnxt_supported(struct rte_eth_dev *dev)
6377 return is_device_supported(dev, &bnxt_rte_pmd);
6380 RTE_LOG_REGISTER_SUFFIX(bnxt_logtype_driver, driver, NOTICE);
6381 RTE_PMD_REGISTER_PCI(net_bnxt, bnxt_rte_pmd);
6382 RTE_PMD_REGISTER_PCI_TABLE(net_bnxt, bnxt_pci_id_map);
6383 RTE_PMD_REGISTER_KMOD_DEP(net_bnxt, "* igb_uio | uio_pci_generic | vfio-pci");