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_FLOW_XSTAT "flow-xstat"
91 #define BNXT_DEVARG_MAX_NUM_KFLOWS "max-num-kflows"
92 #define BNXT_DEVARG_REPRESENTOR "representor"
93 #define BNXT_DEVARG_REP_BASED_PF "rep-based-pf"
94 #define BNXT_DEVARG_REP_IS_PF "rep-is-pf"
95 #define BNXT_DEVARG_REP_Q_R2F "rep-q-r2f"
96 #define BNXT_DEVARG_REP_Q_F2R "rep-q-f2r"
97 #define BNXT_DEVARG_REP_FC_R2F "rep-fc-r2f"
98 #define BNXT_DEVARG_REP_FC_F2R "rep-fc-f2r"
99 #define BNXT_DEVARG_APP_ID "app-id"
101 static const char *const bnxt_dev_args[] = {
102 BNXT_DEVARG_REPRESENTOR,
103 BNXT_DEVARG_FLOW_XSTAT,
104 BNXT_DEVARG_MAX_NUM_KFLOWS,
105 BNXT_DEVARG_REP_BASED_PF,
106 BNXT_DEVARG_REP_IS_PF,
107 BNXT_DEVARG_REP_Q_R2F,
108 BNXT_DEVARG_REP_Q_F2R,
109 BNXT_DEVARG_REP_FC_R2F,
110 BNXT_DEVARG_REP_FC_F2R,
116 * app-id = an non-negative 8-bit number
118 #define BNXT_DEVARG_APP_ID_INVALID(val) ((val) > 255)
121 * flow_xstat == false to disable the feature
122 * flow_xstat == true to enable the feature
124 #define BNXT_DEVARG_FLOW_XSTAT_INVALID(flow_xstat) ((flow_xstat) > 1)
127 * rep_is_pf == false to indicate VF representor
128 * rep_is_pf == true to indicate PF representor
130 #define BNXT_DEVARG_REP_IS_PF_INVALID(rep_is_pf) ((rep_is_pf) > 1)
133 * rep_based_pf == Physical index of the PF
135 #define BNXT_DEVARG_REP_BASED_PF_INVALID(rep_based_pf) ((rep_based_pf) > 15)
137 * rep_q_r2f == Logical COS Queue index for the rep to endpoint direction
139 #define BNXT_DEVARG_REP_Q_R2F_INVALID(rep_q_r2f) ((rep_q_r2f) > 3)
142 * rep_q_f2r == Logical COS Queue index for the endpoint to rep direction
144 #define BNXT_DEVARG_REP_Q_F2R_INVALID(rep_q_f2r) ((rep_q_f2r) > 3)
147 * rep_fc_r2f == Flow control for the representor to endpoint direction
149 #define BNXT_DEVARG_REP_FC_R2F_INVALID(rep_fc_r2f) ((rep_fc_r2f) > 1)
152 * rep_fc_f2r == Flow control for the endpoint to representor direction
154 #define BNXT_DEVARG_REP_FC_F2R_INVALID(rep_fc_f2r) ((rep_fc_f2r) > 1)
156 int bnxt_cfa_code_dynfield_offset = -1;
159 * max_num_kflows must be >= 32
160 * and must be a power-of-2 supported value
161 * return: 1 -> invalid
164 static int bnxt_devarg_max_num_kflow_invalid(uint16_t max_num_kflows)
166 if (max_num_kflows < 32 || !rte_is_power_of_2(max_num_kflows))
171 static int bnxt_vlan_offload_set_op(struct rte_eth_dev *dev, int mask);
172 static int bnxt_dev_uninit(struct rte_eth_dev *eth_dev);
173 static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev);
174 static int bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev);
175 static void bnxt_cancel_fw_health_check(struct bnxt *bp);
176 static int bnxt_restore_vlan_filters(struct bnxt *bp);
177 static void bnxt_dev_recover(void *arg);
178 static void bnxt_free_error_recovery_info(struct bnxt *bp);
179 static void bnxt_free_rep_info(struct bnxt *bp);
181 int is_bnxt_in_error(struct bnxt *bp)
183 if (bp->flags & BNXT_FLAG_FATAL_ERROR)
185 if (bp->flags & BNXT_FLAG_FW_RESET)
191 /***********************/
194 * High level utility functions
197 static uint16_t bnxt_rss_ctxts(const struct bnxt *bp)
199 unsigned int num_rss_rings = RTE_MIN(bp->rx_nr_rings,
200 BNXT_RSS_TBL_SIZE_P5);
202 if (!BNXT_CHIP_P5(bp))
205 return RTE_ALIGN_MUL_CEIL(num_rss_rings,
206 BNXT_RSS_ENTRIES_PER_CTX_P5) /
207 BNXT_RSS_ENTRIES_PER_CTX_P5;
210 uint16_t bnxt_rss_hash_tbl_size(const struct bnxt *bp)
212 if (!BNXT_CHIP_P5(bp))
213 return HW_HASH_INDEX_SIZE;
215 return bnxt_rss_ctxts(bp) * BNXT_RSS_ENTRIES_PER_CTX_P5;
218 static void bnxt_free_parent_info(struct bnxt *bp)
220 rte_free(bp->parent);
224 static void bnxt_free_pf_info(struct bnxt *bp)
230 static void bnxt_free_link_info(struct bnxt *bp)
232 rte_free(bp->link_info);
233 bp->link_info = NULL;
236 static void bnxt_free_leds_info(struct bnxt *bp)
245 static void bnxt_free_flow_stats_info(struct bnxt *bp)
247 rte_free(bp->flow_stat);
248 bp->flow_stat = NULL;
251 static void bnxt_free_cos_queues(struct bnxt *bp)
253 rte_free(bp->rx_cos_queue);
254 bp->rx_cos_queue = NULL;
255 rte_free(bp->tx_cos_queue);
256 bp->tx_cos_queue = NULL;
259 static void bnxt_free_mem(struct bnxt *bp, bool reconfig)
261 bnxt_free_filter_mem(bp);
262 bnxt_free_vnic_attributes(bp);
263 bnxt_free_vnic_mem(bp);
265 /* tx/rx rings are configured as part of *_queue_setup callbacks.
266 * If the number of rings change across fw update,
267 * we don't have much choice except to warn the user.
271 bnxt_free_tx_rings(bp);
272 bnxt_free_rx_rings(bp);
274 bnxt_free_async_cp_ring(bp);
275 bnxt_free_rxtx_nq_ring(bp);
277 rte_free(bp->grp_info);
281 static int bnxt_alloc_parent_info(struct bnxt *bp)
283 bp->parent = rte_zmalloc("bnxt_parent_info",
284 sizeof(struct bnxt_parent_info), 0);
285 if (bp->parent == NULL)
291 static int bnxt_alloc_pf_info(struct bnxt *bp)
293 bp->pf = rte_zmalloc("bnxt_pf_info", sizeof(struct bnxt_pf_info), 0);
300 static int bnxt_alloc_link_info(struct bnxt *bp)
303 rte_zmalloc("bnxt_link_info", sizeof(struct bnxt_link_info), 0);
304 if (bp->link_info == NULL)
310 static int bnxt_alloc_leds_info(struct bnxt *bp)
315 bp->leds = rte_zmalloc("bnxt_leds",
316 BNXT_MAX_LED * sizeof(struct bnxt_led_info),
318 if (bp->leds == NULL)
324 static int bnxt_alloc_cos_queues(struct bnxt *bp)
327 rte_zmalloc("bnxt_rx_cosq",
328 BNXT_COS_QUEUE_COUNT *
329 sizeof(struct bnxt_cos_queue_info),
331 if (bp->rx_cos_queue == NULL)
335 rte_zmalloc("bnxt_tx_cosq",
336 BNXT_COS_QUEUE_COUNT *
337 sizeof(struct bnxt_cos_queue_info),
339 if (bp->tx_cos_queue == NULL)
345 static int bnxt_alloc_flow_stats_info(struct bnxt *bp)
347 bp->flow_stat = rte_zmalloc("bnxt_flow_xstat",
348 sizeof(struct bnxt_flow_stat_info), 0);
349 if (bp->flow_stat == NULL)
355 static int bnxt_alloc_mem(struct bnxt *bp, bool reconfig)
359 rc = bnxt_alloc_ring_grps(bp);
363 rc = bnxt_alloc_async_ring_struct(bp);
367 rc = bnxt_alloc_vnic_mem(bp);
371 rc = bnxt_alloc_vnic_attributes(bp);
375 rc = bnxt_alloc_filter_mem(bp);
379 rc = bnxt_alloc_async_cp_ring(bp);
383 rc = bnxt_alloc_rxtx_nq_ring(bp);
387 if (BNXT_FLOW_XSTATS_EN(bp)) {
388 rc = bnxt_alloc_flow_stats_info(bp);
396 bnxt_free_mem(bp, reconfig);
400 static int bnxt_setup_one_vnic(struct bnxt *bp, uint16_t vnic_id)
402 struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
403 struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
404 uint64_t rx_offloads = dev_conf->rxmode.offloads;
405 struct bnxt_rx_queue *rxq;
409 rc = bnxt_vnic_grp_alloc(bp, vnic);
413 PMD_DRV_LOG(DEBUG, "vnic[%d] = %p vnic->fw_grp_ids = %p\n",
414 vnic_id, vnic, vnic->fw_grp_ids);
416 rc = bnxt_hwrm_vnic_alloc(bp, vnic);
420 /* Alloc RSS context only if RSS mode is enabled */
421 if (dev_conf->rxmode.mq_mode & RTE_ETH_MQ_RX_RSS) {
422 int j, nr_ctxs = bnxt_rss_ctxts(bp);
424 /* RSS table size in Thor is 512.
425 * Cap max Rx rings to same value
427 if (bp->rx_nr_rings > BNXT_RSS_TBL_SIZE_P5) {
428 PMD_DRV_LOG(ERR, "RxQ cnt %d > reta_size %d\n",
429 bp->rx_nr_rings, BNXT_RSS_TBL_SIZE_P5);
434 for (j = 0; j < nr_ctxs; j++) {
435 rc = bnxt_hwrm_vnic_ctx_alloc(bp, vnic, j);
441 "HWRM vnic %d ctx %d alloc failure rc: %x\n",
445 vnic->num_lb_ctxts = nr_ctxs;
449 * Firmware sets pf pair in default vnic cfg. If the VLAN strip
450 * setting is not available at this time, it will not be
451 * configured correctly in the CFA.
453 if (rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP)
454 vnic->vlan_strip = true;
456 vnic->vlan_strip = false;
458 rc = bnxt_hwrm_vnic_cfg(bp, vnic);
462 rc = bnxt_set_hwrm_vnic_filters(bp, vnic);
466 for (j = 0; j < bp->rx_num_qs_per_vnic; j++) {
467 rxq = bp->eth_dev->data->rx_queues[j];
470 "rxq[%d]->vnic=%p vnic->fw_grp_ids=%p\n",
471 j, rxq->vnic, rxq->vnic->fw_grp_ids);
473 if (BNXT_HAS_RING_GRPS(bp) && rxq->rx_deferred_start)
474 rxq->vnic->fw_grp_ids[j] = INVALID_HW_RING_ID;
476 vnic->rx_queue_cnt++;
479 PMD_DRV_LOG(DEBUG, "vnic->rx_queue_cnt = %d\n", vnic->rx_queue_cnt);
481 rc = bnxt_vnic_rss_configure(bp, vnic);
485 bnxt_hwrm_vnic_plcmode_cfg(bp, vnic);
487 rc = bnxt_hwrm_vnic_tpa_cfg(bp, vnic,
488 (rx_offloads & RTE_ETH_RX_OFFLOAD_TCP_LRO) ?
495 PMD_DRV_LOG(ERR, "HWRM vnic %d cfg failure rc: %x\n",
500 static int bnxt_register_fc_ctx_mem(struct bnxt *bp)
504 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->rx_fc_in_tbl.dma,
505 &bp->flow_stat->rx_fc_in_tbl.ctx_id);
510 "rx_fc_in_tbl.va = %p rx_fc_in_tbl.dma = %p"
511 " rx_fc_in_tbl.ctx_id = %d\n",
512 bp->flow_stat->rx_fc_in_tbl.va,
513 (void *)((uintptr_t)bp->flow_stat->rx_fc_in_tbl.dma),
514 bp->flow_stat->rx_fc_in_tbl.ctx_id);
516 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->rx_fc_out_tbl.dma,
517 &bp->flow_stat->rx_fc_out_tbl.ctx_id);
522 "rx_fc_out_tbl.va = %p rx_fc_out_tbl.dma = %p"
523 " rx_fc_out_tbl.ctx_id = %d\n",
524 bp->flow_stat->rx_fc_out_tbl.va,
525 (void *)((uintptr_t)bp->flow_stat->rx_fc_out_tbl.dma),
526 bp->flow_stat->rx_fc_out_tbl.ctx_id);
528 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->tx_fc_in_tbl.dma,
529 &bp->flow_stat->tx_fc_in_tbl.ctx_id);
534 "tx_fc_in_tbl.va = %p tx_fc_in_tbl.dma = %p"
535 " tx_fc_in_tbl.ctx_id = %d\n",
536 bp->flow_stat->tx_fc_in_tbl.va,
537 (void *)((uintptr_t)bp->flow_stat->tx_fc_in_tbl.dma),
538 bp->flow_stat->tx_fc_in_tbl.ctx_id);
540 rc = bnxt_hwrm_ctx_rgtr(bp, bp->flow_stat->tx_fc_out_tbl.dma,
541 &bp->flow_stat->tx_fc_out_tbl.ctx_id);
546 "tx_fc_out_tbl.va = %p tx_fc_out_tbl.dma = %p"
547 " tx_fc_out_tbl.ctx_id = %d\n",
548 bp->flow_stat->tx_fc_out_tbl.va,
549 (void *)((uintptr_t)bp->flow_stat->tx_fc_out_tbl.dma),
550 bp->flow_stat->tx_fc_out_tbl.ctx_id);
552 memset(bp->flow_stat->rx_fc_out_tbl.va,
554 bp->flow_stat->rx_fc_out_tbl.size);
555 rc = bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_RX,
556 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
557 bp->flow_stat->rx_fc_out_tbl.ctx_id,
558 bp->flow_stat->max_fc,
563 memset(bp->flow_stat->tx_fc_out_tbl.va,
565 bp->flow_stat->tx_fc_out_tbl.size);
566 rc = bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_TX,
567 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
568 bp->flow_stat->tx_fc_out_tbl.ctx_id,
569 bp->flow_stat->max_fc,
575 static int bnxt_alloc_ctx_mem_buf(struct bnxt *bp, char *type, size_t size,
576 struct bnxt_ctx_mem_buf_info *ctx)
581 ctx->va = rte_zmalloc_socket(type, size, 0,
582 bp->eth_dev->device->numa_node);
585 rte_mem_lock_page(ctx->va);
587 ctx->dma = rte_mem_virt2iova(ctx->va);
588 if (ctx->dma == RTE_BAD_IOVA)
594 static int bnxt_init_fc_ctx_mem(struct bnxt *bp)
596 struct rte_pci_device *pdev = bp->pdev;
597 char type[RTE_MEMZONE_NAMESIZE];
601 max_fc = bp->flow_stat->max_fc;
603 sprintf(type, "bnxt_rx_fc_in_" PCI_PRI_FMT, pdev->addr.domain,
604 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
605 /* 4 bytes for each counter-id */
606 rc = bnxt_alloc_ctx_mem_buf(bp, type,
608 &bp->flow_stat->rx_fc_in_tbl);
612 sprintf(type, "bnxt_rx_fc_out_" PCI_PRI_FMT, pdev->addr.domain,
613 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
614 /* 16 bytes for each counter - 8 bytes pkt_count, 8 bytes byte_count */
615 rc = bnxt_alloc_ctx_mem_buf(bp, type,
617 &bp->flow_stat->rx_fc_out_tbl);
621 sprintf(type, "bnxt_tx_fc_in_" PCI_PRI_FMT, pdev->addr.domain,
622 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
623 /* 4 bytes for each counter-id */
624 rc = bnxt_alloc_ctx_mem_buf(bp, type,
626 &bp->flow_stat->tx_fc_in_tbl);
630 sprintf(type, "bnxt_tx_fc_out_" PCI_PRI_FMT, pdev->addr.domain,
631 pdev->addr.bus, pdev->addr.devid, pdev->addr.function);
632 /* 16 bytes for each counter - 8 bytes pkt_count, 8 bytes byte_count */
633 rc = bnxt_alloc_ctx_mem_buf(bp, type,
635 &bp->flow_stat->tx_fc_out_tbl);
639 rc = bnxt_register_fc_ctx_mem(bp);
644 static int bnxt_init_ctx_mem(struct bnxt *bp)
648 if (!(bp->fw_cap & BNXT_FW_CAP_ADV_FLOW_COUNTERS) ||
649 !(BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) ||
650 !BNXT_FLOW_XSTATS_EN(bp))
653 rc = bnxt_hwrm_cfa_counter_qcaps(bp, &bp->flow_stat->max_fc);
657 rc = bnxt_init_fc_ctx_mem(bp);
662 static int bnxt_update_phy_setting(struct bnxt *bp)
664 struct rte_eth_link new;
667 rc = bnxt_get_hwrm_link_config(bp, &new);
669 PMD_DRV_LOG(ERR, "Failed to get link settings\n");
674 * On BCM957508-N2100 adapters, FW will not allow any user other
675 * than BMC to shutdown the port. bnxt_get_hwrm_link_config() call
676 * always returns link up. Force phy update always in that case.
678 if (!new.link_status || IS_BNXT_DEV_957508_N2100(bp)) {
679 rc = bnxt_set_hwrm_link_config(bp, true);
681 PMD_DRV_LOG(ERR, "Failed to update PHY settings\n");
689 static void bnxt_free_prev_ring_stats(struct bnxt *bp)
691 rte_free(bp->prev_rx_ring_stats);
692 rte_free(bp->prev_tx_ring_stats);
694 bp->prev_rx_ring_stats = NULL;
695 bp->prev_tx_ring_stats = NULL;
698 static int bnxt_alloc_prev_ring_stats(struct bnxt *bp)
700 bp->prev_rx_ring_stats = rte_zmalloc("bnxt_prev_rx_ring_stats",
701 sizeof(struct bnxt_ring_stats) *
704 if (bp->prev_rx_ring_stats == NULL)
707 bp->prev_tx_ring_stats = rte_zmalloc("bnxt_prev_tx_ring_stats",
708 sizeof(struct bnxt_ring_stats) *
711 if (bp->prev_tx_ring_stats == NULL)
717 bnxt_free_prev_ring_stats(bp);
721 static int bnxt_start_nic(struct bnxt *bp)
723 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(bp->eth_dev);
724 struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
725 uint32_t intr_vector = 0;
726 uint32_t queue_id, base = BNXT_MISC_VEC_ID;
727 uint32_t vec = BNXT_MISC_VEC_ID;
731 if (bp->eth_dev->data->mtu > RTE_ETHER_MTU)
732 bp->flags |= BNXT_FLAG_JUMBO;
734 bp->flags &= ~BNXT_FLAG_JUMBO;
736 /* THOR does not support ring groups.
737 * But we will use the array to save RSS context IDs.
739 if (BNXT_CHIP_P5(bp))
740 bp->max_ring_grps = BNXT_MAX_RSS_CTXTS_P5;
742 rc = bnxt_alloc_hwrm_rings(bp);
744 PMD_DRV_LOG(ERR, "HWRM ring alloc failure rc: %x\n", rc);
748 rc = bnxt_alloc_all_hwrm_ring_grps(bp);
750 PMD_DRV_LOG(ERR, "HWRM ring grp alloc failure: %x\n", rc);
754 if (!(bp->vnic_cap_flags & BNXT_VNIC_CAP_COS_CLASSIFY))
757 for (j = 0, i = 0; i < BNXT_COS_QUEUE_COUNT; i++) {
758 if (bp->rx_cos_queue[i].id != 0xff) {
759 struct bnxt_vnic_info *vnic = &bp->vnic_info[j++];
763 "Num pools more than FW profile\n");
767 vnic->cos_queue_id = bp->rx_cos_queue[i].id;
773 rc = bnxt_mq_rx_configure(bp);
775 PMD_DRV_LOG(ERR, "MQ mode configure failure rc: %x\n", rc);
779 for (j = 0; j < bp->rx_nr_rings; j++) {
780 struct bnxt_rx_queue *rxq = bp->rx_queues[j];
782 if (!rxq->rx_deferred_start) {
783 bp->eth_dev->data->rx_queue_state[j] =
784 RTE_ETH_QUEUE_STATE_STARTED;
785 rxq->rx_started = true;
790 rc = bnxt_setup_one_vnic(bp, 0);
793 /* VNIC configuration */
794 if (BNXT_RFS_NEEDS_VNIC(bp)) {
795 for (i = 1; i < bp->nr_vnics; i++) {
796 rc = bnxt_setup_one_vnic(bp, i);
802 for (j = 0; j < bp->tx_nr_rings; j++) {
803 struct bnxt_tx_queue *txq = bp->tx_queues[j];
805 if (!txq->tx_deferred_start) {
806 bp->eth_dev->data->tx_queue_state[j] =
807 RTE_ETH_QUEUE_STATE_STARTED;
808 txq->tx_started = true;
812 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, &bp->vnic_info[0], 0, NULL);
815 "HWRM cfa l2 rx mask failure rc: %x\n", rc);
819 /* check and configure queue intr-vector mapping */
820 if ((rte_intr_cap_multiple(intr_handle) ||
821 !RTE_ETH_DEV_SRIOV(bp->eth_dev).active) &&
822 bp->eth_dev->data->dev_conf.intr_conf.rxq != 0) {
823 intr_vector = bp->eth_dev->data->nb_rx_queues;
824 PMD_DRV_LOG(DEBUG, "intr_vector = %d\n", intr_vector);
825 if (intr_vector > bp->rx_cp_nr_rings) {
826 PMD_DRV_LOG(ERR, "At most %d intr queues supported",
830 rc = rte_intr_efd_enable(intr_handle, intr_vector);
835 if (rte_intr_dp_is_en(intr_handle)) {
836 if (rte_intr_vec_list_alloc(intr_handle, "intr_vec",
837 bp->eth_dev->data->nb_rx_queues)) {
838 PMD_DRV_LOG(ERR, "Failed to allocate %d rx_queues"
839 " intr_vec", bp->eth_dev->data->nb_rx_queues);
843 PMD_DRV_LOG(DEBUG, "intr_handle->nb_efd = %d "
844 "intr_handle->max_intr = %d\n",
845 rte_intr_nb_efd_get(intr_handle),
846 rte_intr_max_intr_get(intr_handle));
847 for (queue_id = 0; queue_id < bp->eth_dev->data->nb_rx_queues;
849 rte_intr_vec_list_index_set(intr_handle,
850 queue_id, vec + BNXT_RX_VEC_START);
851 if (vec < base + rte_intr_nb_efd_get(intr_handle)
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 |= RTE_ETH_LINK_SPEED_100M;
911 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_100MBHD)
912 speed_capa |= RTE_ETH_LINK_SPEED_100M_HD;
913 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_1GB)
914 speed_capa |= RTE_ETH_LINK_SPEED_1G;
915 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_2_5GB)
916 speed_capa |= RTE_ETH_LINK_SPEED_2_5G;
917 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_10GB)
918 speed_capa |= RTE_ETH_LINK_SPEED_10G;
919 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_20GB)
920 speed_capa |= RTE_ETH_LINK_SPEED_20G;
921 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_25GB)
922 speed_capa |= RTE_ETH_LINK_SPEED_25G;
923 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_40GB)
924 speed_capa |= RTE_ETH_LINK_SPEED_40G;
925 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_50GB)
926 speed_capa |= RTE_ETH_LINK_SPEED_50G;
927 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_100GB)
928 speed_capa |= RTE_ETH_LINK_SPEED_100G;
929 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_50G)
930 speed_capa |= RTE_ETH_LINK_SPEED_50G;
931 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_100G)
932 speed_capa |= RTE_ETH_LINK_SPEED_100G;
933 if (link_speed & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_PAM4_SPEEDS_200G)
934 speed_capa |= RTE_ETH_LINK_SPEED_200G;
936 if (bp->link_info->auto_mode ==
937 HWRM_PORT_PHY_QCFG_OUTPUT_AUTO_MODE_NONE)
938 speed_capa |= RTE_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 = HW_HASH_KEY_SIZE;
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 |= RTE_ETH_RX_OFFLOAD_TIMESTAMP;
983 if (bp->vnic_cap_flags & BNXT_VNIC_CAP_VLAN_RX_STRIP)
984 dev_info->rx_offload_capa |= RTE_ETH_RX_OFFLOAD_VLAN_STRIP;
985 dev_info->tx_queue_offload_capa = RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
986 dev_info->tx_offload_capa = BNXT_DEV_TX_OFFLOAD_SUPPORT |
987 dev_info->tx_queue_offload_capa;
988 if (bp->fw_cap & BNXT_FW_CAP_VLAN_TX_INSERT)
989 dev_info->tx_offload_capa |= RTE_ETH_TX_OFFLOAD_VLAN_INSERT;
990 dev_info->flow_type_rss_offloads = BNXT_ETH_RSS_SUPPORT;
992 dev_info->speed_capa = bnxt_get_speed_capabilities(bp);
993 dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
994 RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
995 dev_info->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
997 dev_info->default_rxconf = (struct rte_eth_rxconf) {
1003 .rx_free_thresh = 32,
1004 .rx_drop_en = BNXT_DEFAULT_RX_DROP_EN,
1007 dev_info->default_txconf = (struct rte_eth_txconf) {
1013 .tx_free_thresh = 32,
1016 eth_dev->data->dev_conf.intr_conf.lsc = 1;
1018 dev_info->rx_desc_lim.nb_min = BNXT_MIN_RING_DESC;
1019 dev_info->rx_desc_lim.nb_max = BNXT_MAX_RX_RING_DESC;
1020 dev_info->tx_desc_lim.nb_min = BNXT_MIN_RING_DESC;
1021 dev_info->tx_desc_lim.nb_max = BNXT_MAX_TX_RING_DESC;
1023 if (BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) {
1024 dev_info->switch_info.name = eth_dev->device->name;
1025 dev_info->switch_info.domain_id = bp->switch_domain_id;
1026 dev_info->switch_info.port_id =
1027 BNXT_PF(bp) ? BNXT_SWITCH_PORT_ID_PF :
1028 BNXT_SWITCH_PORT_ID_TRUSTED_VF;
1032 * TODO: default_rxconf, default_txconf, rx_desc_lim, and tx_desc_lim
1033 * need further investigation.
1036 /* VMDq resources */
1037 vpool = 64; /* RTE_ETH_64_POOLS */
1038 vrxq = 128; /* RTE_ETH_VMDQ_DCB_NUM_QUEUES */
1039 for (i = 0; i < 4; vpool >>= 1, i++) {
1040 if (max_vnics > vpool) {
1041 for (j = 0; j < 5; vrxq >>= 1, j++) {
1042 if (dev_info->max_rx_queues > vrxq) {
1048 /* Not enough resources to support VMDq */
1052 /* Not enough resources to support VMDq */
1056 dev_info->max_vmdq_pools = vpool;
1057 dev_info->vmdq_queue_num = vrxq;
1059 dev_info->vmdq_pool_base = 0;
1060 dev_info->vmdq_queue_base = 0;
1065 /* Configure the device based on the configuration provided */
1066 static int bnxt_dev_configure_op(struct rte_eth_dev *eth_dev)
1068 struct bnxt *bp = eth_dev->data->dev_private;
1069 uint64_t rx_offloads = eth_dev->data->dev_conf.rxmode.offloads;
1072 bp->rx_queues = (void *)eth_dev->data->rx_queues;
1073 bp->tx_queues = (void *)eth_dev->data->tx_queues;
1074 bp->tx_nr_rings = eth_dev->data->nb_tx_queues;
1075 bp->rx_nr_rings = eth_dev->data->nb_rx_queues;
1077 rc = is_bnxt_in_error(bp);
1081 if (BNXT_VF(bp) && (bp->flags & BNXT_FLAG_NEW_RM)) {
1082 rc = bnxt_hwrm_check_vf_rings(bp);
1084 PMD_DRV_LOG(ERR, "HWRM insufficient resources\n");
1088 /* If a resource has already been allocated - in this case
1089 * it is the async completion ring, free it. Reallocate it after
1090 * resource reservation. This will ensure the resource counts
1091 * are calculated correctly.
1094 pthread_mutex_lock(&bp->def_cp_lock);
1096 if (!BNXT_HAS_NQ(bp) && bp->async_cp_ring) {
1097 bnxt_disable_int(bp);
1098 bnxt_free_cp_ring(bp, bp->async_cp_ring);
1101 rc = bnxt_hwrm_func_reserve_vf_resc(bp, false);
1103 PMD_DRV_LOG(ERR, "HWRM resource alloc fail:%x\n", rc);
1104 pthread_mutex_unlock(&bp->def_cp_lock);
1108 if (!BNXT_HAS_NQ(bp) && bp->async_cp_ring) {
1109 rc = bnxt_alloc_async_cp_ring(bp);
1111 pthread_mutex_unlock(&bp->def_cp_lock);
1114 bnxt_enable_int(bp);
1117 pthread_mutex_unlock(&bp->def_cp_lock);
1120 /* Inherit new configurations */
1121 if (eth_dev->data->nb_rx_queues > bp->max_rx_rings ||
1122 eth_dev->data->nb_tx_queues > bp->max_tx_rings ||
1123 eth_dev->data->nb_rx_queues + eth_dev->data->nb_tx_queues
1124 + BNXT_NUM_ASYNC_CPR(bp) > bp->max_cp_rings ||
1125 eth_dev->data->nb_rx_queues + eth_dev->data->nb_tx_queues >
1127 goto resource_error;
1129 if (BNXT_HAS_RING_GRPS(bp) &&
1130 (uint32_t)(eth_dev->data->nb_rx_queues) > bp->max_ring_grps)
1131 goto resource_error;
1133 if (!(eth_dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS) &&
1134 bp->max_vnics < eth_dev->data->nb_rx_queues)
1135 goto resource_error;
1137 bp->rx_cp_nr_rings = bp->rx_nr_rings;
1138 bp->tx_cp_nr_rings = bp->tx_nr_rings;
1140 if (eth_dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG)
1141 rx_offloads |= RTE_ETH_RX_OFFLOAD_RSS_HASH;
1142 eth_dev->data->dev_conf.rxmode.offloads = rx_offloads;
1144 bnxt_mtu_set_op(eth_dev, eth_dev->data->mtu);
1150 "Insufficient resources to support requested config\n");
1152 "Num Queues Requested: Tx %d, Rx %d\n",
1153 eth_dev->data->nb_tx_queues,
1154 eth_dev->data->nb_rx_queues);
1156 "MAX: TxQ %d, RxQ %d, CQ %d Stat %d, Grp %d, Vnic %d\n",
1157 bp->max_tx_rings, bp->max_rx_rings, bp->max_cp_rings,
1158 bp->max_stat_ctx, bp->max_ring_grps, bp->max_vnics);
1162 void bnxt_print_link_info(struct rte_eth_dev *eth_dev)
1164 struct rte_eth_link *link = ð_dev->data->dev_link;
1166 if (link->link_status)
1167 PMD_DRV_LOG(DEBUG, "Port %d Link Up - speed %u Mbps - %s\n",
1168 eth_dev->data->port_id,
1169 (uint32_t)link->link_speed,
1170 (link->link_duplex == RTE_ETH_LINK_FULL_DUPLEX) ?
1171 ("full-duplex") : ("half-duplex\n"));
1173 PMD_DRV_LOG(INFO, "Port %d Link Down\n",
1174 eth_dev->data->port_id);
1178 * Determine whether the current configuration requires support for scattered
1179 * receive; return 1 if scattered receive is required and 0 if not.
1181 static int bnxt_scattered_rx(struct rte_eth_dev *eth_dev)
1183 uint32_t overhead = BNXT_MAX_PKT_LEN - BNXT_MAX_MTU;
1187 if (eth_dev->data->dev_conf.rxmode.offloads & RTE_ETH_RX_OFFLOAD_SCATTER)
1190 if (eth_dev->data->dev_conf.rxmode.offloads & RTE_ETH_RX_OFFLOAD_TCP_LRO)
1193 for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
1194 struct bnxt_rx_queue *rxq = eth_dev->data->rx_queues[i];
1196 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
1197 RTE_PKTMBUF_HEADROOM);
1198 if (eth_dev->data->mtu + overhead > buf_size)
1204 static eth_rx_burst_t
1205 bnxt_receive_function(struct rte_eth_dev *eth_dev)
1207 struct bnxt *bp = eth_dev->data->dev_private;
1209 /* Disable vector mode RX for Stingray2 for now */
1210 if (BNXT_CHIP_SR2(bp)) {
1211 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1212 return bnxt_recv_pkts;
1215 #if (defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)) && \
1216 !defined(RTE_LIBRTE_IEEE1588)
1218 /* Vector mode receive cannot be enabled if scattered rx is in use. */
1219 if (eth_dev->data->scattered_rx)
1223 * Vector mode receive cannot be enabled if Truflow is enabled or if
1224 * asynchronous completions and receive completions can be placed in
1225 * the same completion ring.
1227 if (BNXT_TRUFLOW_EN(bp) || !BNXT_NUM_ASYNC_CPR(bp))
1231 * Vector mode receive cannot be enabled if any receive offloads outside
1232 * a limited subset have been enabled.
1234 if (eth_dev->data->dev_conf.rxmode.offloads &
1235 ~(RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
1236 RTE_ETH_RX_OFFLOAD_KEEP_CRC |
1237 RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
1238 RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
1239 RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
1240 RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
1241 RTE_ETH_RX_OFFLOAD_OUTER_UDP_CKSUM |
1242 RTE_ETH_RX_OFFLOAD_RSS_HASH |
1243 RTE_ETH_RX_OFFLOAD_VLAN_FILTER))
1246 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
1247 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
1248 rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1) {
1250 "Using AVX2 vector mode receive for port %d\n",
1251 eth_dev->data->port_id);
1252 bp->flags |= BNXT_FLAG_RX_VECTOR_PKT_MODE;
1253 return bnxt_recv_pkts_vec_avx2;
1256 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_128) {
1258 "Using SSE vector mode receive for port %d\n",
1259 eth_dev->data->port_id);
1260 bp->flags |= BNXT_FLAG_RX_VECTOR_PKT_MODE;
1261 return bnxt_recv_pkts_vec;
1265 PMD_DRV_LOG(INFO, "Vector mode receive disabled for port %d\n",
1266 eth_dev->data->port_id);
1268 "Port %d scatter: %d rx offload: %" PRIX64 "\n",
1269 eth_dev->data->port_id,
1270 eth_dev->data->scattered_rx,
1271 eth_dev->data->dev_conf.rxmode.offloads);
1273 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1274 return bnxt_recv_pkts;
1277 static eth_tx_burst_t
1278 bnxt_transmit_function(struct rte_eth_dev *eth_dev)
1280 struct bnxt *bp = eth_dev->data->dev_private;
1282 /* Disable vector mode TX for Stingray2 for now */
1283 if (BNXT_CHIP_SR2(bp))
1284 return bnxt_xmit_pkts;
1286 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64) && \
1287 !defined(RTE_LIBRTE_IEEE1588)
1288 uint64_t offloads = eth_dev->data->dev_conf.txmode.offloads;
1291 * Vector mode transmit can be enabled only if not using scatter rx
1294 if (eth_dev->data->scattered_rx ||
1295 (offloads & ~RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE) ||
1296 BNXT_TRUFLOW_EN(bp))
1299 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
1300 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
1301 rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1) {
1303 "Using AVX2 vector mode transmit for port %d\n",
1304 eth_dev->data->port_id);
1305 return bnxt_xmit_pkts_vec_avx2;
1308 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_128) {
1310 "Using SSE vector mode transmit for port %d\n",
1311 eth_dev->data->port_id);
1312 return bnxt_xmit_pkts_vec;
1316 PMD_DRV_LOG(INFO, "Vector mode transmit disabled for port %d\n",
1317 eth_dev->data->port_id);
1319 "Port %d scatter: %d tx offload: %" PRIX64 "\n",
1320 eth_dev->data->port_id,
1321 eth_dev->data->scattered_rx,
1324 return bnxt_xmit_pkts;
1327 static int bnxt_handle_if_change_status(struct bnxt *bp)
1331 /* Since fw has undergone a reset and lost all contexts,
1332 * set fatal flag to not issue hwrm during cleanup
1334 bp->flags |= BNXT_FLAG_FATAL_ERROR;
1335 bnxt_uninit_resources(bp, true);
1337 /* clear fatal flag so that re-init happens */
1338 bp->flags &= ~BNXT_FLAG_FATAL_ERROR;
1339 rc = bnxt_init_resources(bp, true);
1341 bp->flags &= ~BNXT_FLAG_IF_CHANGE_HOT_FW_RESET_DONE;
1346 static int bnxt_dev_set_link_up_op(struct rte_eth_dev *eth_dev)
1348 struct bnxt *bp = eth_dev->data->dev_private;
1351 if (!BNXT_SINGLE_PF(bp))
1354 if (!bp->link_info->link_up)
1355 rc = bnxt_set_hwrm_link_config(bp, true);
1357 eth_dev->data->dev_link.link_status = 1;
1359 bnxt_print_link_info(eth_dev);
1363 static int bnxt_dev_set_link_down_op(struct rte_eth_dev *eth_dev)
1365 struct bnxt *bp = eth_dev->data->dev_private;
1367 if (!BNXT_SINGLE_PF(bp))
1370 eth_dev->data->dev_link.link_status = 0;
1371 bnxt_set_hwrm_link_config(bp, false);
1372 bp->link_info->link_up = 0;
1377 static void bnxt_free_switch_domain(struct bnxt *bp)
1381 if (!(BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)))
1384 rc = rte_eth_switch_domain_free(bp->switch_domain_id);
1386 PMD_DRV_LOG(ERR, "free switch domain:%d fail: %d\n",
1387 bp->switch_domain_id, rc);
1390 static void bnxt_ptp_get_current_time(void *arg)
1392 struct bnxt *bp = arg;
1393 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
1396 rc = is_bnxt_in_error(bp);
1403 bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
1404 &ptp->current_time);
1406 rc = rte_eal_alarm_set(US_PER_S, bnxt_ptp_get_current_time, (void *)bp);
1408 PMD_DRV_LOG(ERR, "Failed to re-schedule PTP alarm\n");
1409 bp->flags2 &= ~BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1413 static int bnxt_schedule_ptp_alarm(struct bnxt *bp)
1415 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
1418 if (bp->flags2 & BNXT_FLAGS2_PTP_ALARM_SCHEDULED)
1421 bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
1422 &ptp->current_time);
1424 rc = rte_eal_alarm_set(US_PER_S, bnxt_ptp_get_current_time, (void *)bp);
1428 static void bnxt_cancel_ptp_alarm(struct bnxt *bp)
1430 if (bp->flags2 & BNXT_FLAGS2_PTP_ALARM_SCHEDULED) {
1431 rte_eal_alarm_cancel(bnxt_ptp_get_current_time, (void *)bp);
1432 bp->flags2 &= ~BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1436 static void bnxt_ptp_stop(struct bnxt *bp)
1438 bnxt_cancel_ptp_alarm(bp);
1439 bp->flags2 &= ~BNXT_FLAGS2_PTP_TIMESYNC_ENABLED;
1442 static int bnxt_ptp_start(struct bnxt *bp)
1446 rc = bnxt_schedule_ptp_alarm(bp);
1448 PMD_DRV_LOG(ERR, "Failed to schedule PTP alarm\n");
1450 bp->flags2 |= BNXT_FLAGS2_PTP_TIMESYNC_ENABLED;
1451 bp->flags2 |= BNXT_FLAGS2_PTP_ALARM_SCHEDULED;
1457 static int bnxt_dev_stop(struct rte_eth_dev *eth_dev)
1459 struct bnxt *bp = eth_dev->data->dev_private;
1460 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1461 struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
1462 struct rte_eth_link link;
1465 eth_dev->data->dev_started = 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 rte_intr_vec_list_free(intr_handle);
1504 bnxt_hwrm_port_clr_stats(bp);
1505 bnxt_free_tx_mbufs(bp);
1506 bnxt_free_rx_mbufs(bp);
1507 /* Process any remaining notifications in default completion queue */
1508 bnxt_int_handler(eth_dev);
1509 bnxt_shutdown_nic(bp);
1510 bnxt_hwrm_if_change(bp, false);
1512 bnxt_free_prev_ring_stats(bp);
1513 rte_free(bp->mark_table);
1514 bp->mark_table = NULL;
1516 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1517 bp->rx_cosq_cnt = 0;
1518 /* All filters are deleted on a port stop. */
1519 if (BNXT_FLOW_XSTATS_EN(bp))
1520 bp->flow_stat->flow_count = 0;
1522 eth_dev->data->scattered_rx = 0;
1527 /* Unload the driver, release resources */
1528 static int bnxt_dev_stop_op(struct rte_eth_dev *eth_dev)
1530 struct bnxt *bp = eth_dev->data->dev_private;
1532 pthread_mutex_lock(&bp->err_recovery_lock);
1533 if (bp->flags & BNXT_FLAG_FW_RESET) {
1535 "Adapter recovering from error..Please retry\n");
1536 pthread_mutex_unlock(&bp->err_recovery_lock);
1539 pthread_mutex_unlock(&bp->err_recovery_lock);
1541 return bnxt_dev_stop(eth_dev);
1544 static int bnxt_dev_start_op(struct rte_eth_dev *eth_dev)
1546 struct bnxt *bp = eth_dev->data->dev_private;
1547 uint64_t rx_offloads = eth_dev->data->dev_conf.rxmode.offloads;
1549 int rc, retry_cnt = BNXT_IF_CHANGE_RETRY_COUNT;
1551 if (!eth_dev->data->nb_tx_queues || !eth_dev->data->nb_rx_queues) {
1552 PMD_DRV_LOG(ERR, "Queues are not configured yet!\n");
1556 if (bp->rx_cp_nr_rings > RTE_ETHDEV_QUEUE_STAT_CNTRS)
1558 "RxQ cnt %d > RTE_ETHDEV_QUEUE_STAT_CNTRS %d\n",
1559 bp->rx_cp_nr_rings, RTE_ETHDEV_QUEUE_STAT_CNTRS);
1562 rc = bnxt_hwrm_if_change(bp, true);
1563 if (rc == 0 || rc != -EAGAIN)
1566 rte_delay_ms(BNXT_IF_CHANGE_RETRY_INTERVAL);
1567 } while (retry_cnt--);
1572 if (bp->flags & BNXT_FLAG_IF_CHANGE_HOT_FW_RESET_DONE) {
1573 rc = bnxt_handle_if_change_status(bp);
1578 bnxt_enable_int(bp);
1580 eth_dev->data->scattered_rx = bnxt_scattered_rx(eth_dev);
1582 rc = bnxt_start_nic(bp);
1586 rc = bnxt_alloc_prev_ring_stats(bp);
1590 eth_dev->data->dev_started = 1;
1592 bnxt_link_update_op(eth_dev, 1);
1594 if (rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER)
1595 vlan_mask |= RTE_ETH_VLAN_FILTER_MASK;
1596 if (rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP)
1597 vlan_mask |= RTE_ETH_VLAN_STRIP_MASK;
1598 rc = bnxt_vlan_offload_set_op(eth_dev, vlan_mask);
1602 /* Initialize bnxt ULP port details */
1603 rc = bnxt_ulp_port_init(bp);
1607 eth_dev->rx_pkt_burst = bnxt_receive_function(eth_dev);
1608 eth_dev->tx_pkt_burst = bnxt_transmit_function(eth_dev);
1610 bnxt_schedule_fw_health_check(bp);
1612 if (BNXT_P5_PTP_TIMESYNC_ENABLED(bp))
1613 bnxt_schedule_ptp_alarm(bp);
1618 bnxt_dev_stop(eth_dev);
1623 bnxt_uninit_locks(struct bnxt *bp)
1625 pthread_mutex_destroy(&bp->flow_lock);
1626 pthread_mutex_destroy(&bp->def_cp_lock);
1627 pthread_mutex_destroy(&bp->health_check_lock);
1628 pthread_mutex_destroy(&bp->err_recovery_lock);
1630 pthread_mutex_destroy(&bp->rep_info->vfr_lock);
1631 pthread_mutex_destroy(&bp->rep_info->vfr_start_lock);
1635 static void bnxt_drv_uninit(struct bnxt *bp)
1637 bnxt_free_leds_info(bp);
1638 bnxt_free_cos_queues(bp);
1639 bnxt_free_link_info(bp);
1640 bnxt_free_parent_info(bp);
1641 bnxt_uninit_locks(bp);
1643 rte_memzone_free((const struct rte_memzone *)bp->tx_mem_zone);
1644 bp->tx_mem_zone = NULL;
1645 rte_memzone_free((const struct rte_memzone *)bp->rx_mem_zone);
1646 bp->rx_mem_zone = NULL;
1648 bnxt_free_vf_info(bp);
1649 bnxt_free_pf_info(bp);
1651 rte_free(bp->grp_info);
1652 bp->grp_info = NULL;
1655 static int bnxt_dev_close_op(struct rte_eth_dev *eth_dev)
1657 struct bnxt *bp = eth_dev->data->dev_private;
1660 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1663 pthread_mutex_lock(&bp->err_recovery_lock);
1664 if (bp->flags & BNXT_FLAG_FW_RESET) {
1666 "Adapter recovering from error...Please retry\n");
1667 pthread_mutex_unlock(&bp->err_recovery_lock);
1670 pthread_mutex_unlock(&bp->err_recovery_lock);
1672 /* cancel the recovery handler before remove dev */
1673 rte_eal_alarm_cancel(bnxt_dev_reset_and_resume, (void *)bp);
1674 rte_eal_alarm_cancel(bnxt_dev_recover, (void *)bp);
1675 bnxt_cancel_fc_thread(bp);
1677 if (eth_dev->data->dev_started)
1678 ret = bnxt_dev_stop(eth_dev);
1680 bnxt_uninit_resources(bp, false);
1682 bnxt_drv_uninit(bp);
1687 static void bnxt_mac_addr_remove_op(struct rte_eth_dev *eth_dev,
1690 struct bnxt *bp = eth_dev->data->dev_private;
1691 uint64_t pool_mask = eth_dev->data->mac_pool_sel[index];
1692 struct bnxt_vnic_info *vnic;
1693 struct bnxt_filter_info *filter, *temp_filter;
1696 if (is_bnxt_in_error(bp))
1700 * Loop through all VNICs from the specified filter flow pools to
1701 * remove the corresponding MAC addr filter
1703 for (i = 0; i < bp->nr_vnics; i++) {
1704 if (!(pool_mask & (1ULL << i)))
1707 vnic = &bp->vnic_info[i];
1708 filter = STAILQ_FIRST(&vnic->filter);
1710 temp_filter = STAILQ_NEXT(filter, next);
1711 if (filter->mac_index == index) {
1712 STAILQ_REMOVE(&vnic->filter, filter,
1713 bnxt_filter_info, next);
1714 bnxt_hwrm_clear_l2_filter(bp, filter);
1715 bnxt_free_filter(bp, filter);
1717 filter = temp_filter;
1722 static int bnxt_add_mac_filter(struct bnxt *bp, struct bnxt_vnic_info *vnic,
1723 struct rte_ether_addr *mac_addr, uint32_t index,
1726 struct bnxt_filter_info *filter;
1729 /* Attach requested MAC address to the new l2_filter */
1730 STAILQ_FOREACH(filter, &vnic->filter, next) {
1731 if (filter->mac_index == index) {
1733 "MAC addr already existed for pool %d\n",
1739 filter = bnxt_alloc_filter(bp);
1741 PMD_DRV_LOG(ERR, "L2 filter alloc failed\n");
1745 /* bnxt_alloc_filter copies default MAC to filter->l2_addr. So,
1746 * if the MAC that's been programmed now is a different one, then,
1747 * copy that addr to filter->l2_addr
1750 memcpy(filter->l2_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1751 filter->flags |= HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
1753 rc = bnxt_hwrm_set_l2_filter(bp, vnic->fw_vnic_id, filter);
1755 filter->mac_index = index;
1756 if (filter->mac_index == 0)
1757 STAILQ_INSERT_HEAD(&vnic->filter, filter, next);
1759 STAILQ_INSERT_TAIL(&vnic->filter, filter, next);
1761 bnxt_free_filter(bp, filter);
1767 static int bnxt_mac_addr_add_op(struct rte_eth_dev *eth_dev,
1768 struct rte_ether_addr *mac_addr,
1769 uint32_t index, uint32_t pool)
1771 struct bnxt *bp = eth_dev->data->dev_private;
1772 struct bnxt_vnic_info *vnic = &bp->vnic_info[pool];
1775 rc = is_bnxt_in_error(bp);
1779 if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp)) {
1780 PMD_DRV_LOG(ERR, "Cannot add MAC address to a VF interface\n");
1785 PMD_DRV_LOG(ERR, "VNIC not found for pool %d!\n", pool);
1789 /* Filter settings will get applied when port is started */
1790 if (!eth_dev->data->dev_started)
1793 rc = bnxt_add_mac_filter(bp, vnic, mac_addr, index, pool);
1798 int bnxt_link_update_op(struct rte_eth_dev *eth_dev, int wait_to_complete)
1801 struct bnxt *bp = eth_dev->data->dev_private;
1802 struct rte_eth_link new;
1803 int cnt = wait_to_complete ? BNXT_MAX_LINK_WAIT_CNT :
1804 BNXT_MIN_LINK_WAIT_CNT;
1806 rc = is_bnxt_in_error(bp);
1810 memset(&new, 0, sizeof(new));
1812 if (bp->link_info == NULL)
1816 /* Retrieve link info from hardware */
1817 rc = bnxt_get_hwrm_link_config(bp, &new);
1819 new.link_speed = RTE_ETH_LINK_SPEED_100M;
1820 new.link_duplex = RTE_ETH_LINK_FULL_DUPLEX;
1822 "Failed to retrieve link rc = 0x%x!\n", rc);
1826 if (!wait_to_complete || new.link_status)
1829 rte_delay_ms(BNXT_LINK_WAIT_INTERVAL);
1832 /* Only single function PF can bring phy down.
1833 * When port is stopped, report link down for VF/MH/NPAR functions.
1835 if (!BNXT_SINGLE_PF(bp) && !eth_dev->data->dev_started)
1836 memset(&new, 0, sizeof(new));
1839 /* Timed out or success */
1840 if (new.link_status != eth_dev->data->dev_link.link_status ||
1841 new.link_speed != eth_dev->data->dev_link.link_speed) {
1842 rte_eth_linkstatus_set(eth_dev, &new);
1843 bnxt_print_link_info(eth_dev);
1849 static int bnxt_promiscuous_enable_op(struct rte_eth_dev *eth_dev)
1851 struct bnxt *bp = eth_dev->data->dev_private;
1852 struct bnxt_vnic_info *vnic;
1856 rc = is_bnxt_in_error(bp);
1860 /* Filter settings will get applied when port is started */
1861 if (!eth_dev->data->dev_started)
1864 if (bp->vnic_info == NULL)
1867 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1869 old_flags = vnic->flags;
1870 vnic->flags |= BNXT_VNIC_INFO_PROMISC;
1871 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1873 vnic->flags = old_flags;
1878 static int bnxt_promiscuous_disable_op(struct rte_eth_dev *eth_dev)
1880 struct bnxt *bp = eth_dev->data->dev_private;
1881 struct bnxt_vnic_info *vnic;
1885 rc = is_bnxt_in_error(bp);
1889 /* Filter settings will get applied when port is started */
1890 if (!eth_dev->data->dev_started)
1893 if (bp->vnic_info == NULL)
1896 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1898 old_flags = vnic->flags;
1899 vnic->flags &= ~BNXT_VNIC_INFO_PROMISC;
1900 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1902 vnic->flags = old_flags;
1907 static int bnxt_allmulticast_enable_op(struct rte_eth_dev *eth_dev)
1909 struct bnxt *bp = eth_dev->data->dev_private;
1910 struct bnxt_vnic_info *vnic;
1914 rc = is_bnxt_in_error(bp);
1918 /* Filter settings will get applied when port is started */
1919 if (!eth_dev->data->dev_started)
1922 if (bp->vnic_info == NULL)
1925 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1927 old_flags = vnic->flags;
1928 vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
1929 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1931 vnic->flags = old_flags;
1936 static int bnxt_allmulticast_disable_op(struct rte_eth_dev *eth_dev)
1938 struct bnxt *bp = eth_dev->data->dev_private;
1939 struct bnxt_vnic_info *vnic;
1943 rc = is_bnxt_in_error(bp);
1947 /* Filter settings will get applied when port is started */
1948 if (!eth_dev->data->dev_started)
1951 if (bp->vnic_info == NULL)
1954 vnic = BNXT_GET_DEFAULT_VNIC(bp);
1956 old_flags = vnic->flags;
1957 vnic->flags &= ~BNXT_VNIC_INFO_ALLMULTI;
1958 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
1960 vnic->flags = old_flags;
1965 /* Return bnxt_rx_queue pointer corresponding to a given rxq. */
1966 static struct bnxt_rx_queue *bnxt_qid_to_rxq(struct bnxt *bp, uint16_t qid)
1968 if (qid >= bp->rx_nr_rings)
1971 return bp->eth_dev->data->rx_queues[qid];
1974 /* Return rxq corresponding to a given rss table ring/group ID. */
1975 static uint16_t bnxt_rss_to_qid(struct bnxt *bp, uint16_t fwr)
1977 struct bnxt_rx_queue *rxq;
1980 if (!BNXT_HAS_RING_GRPS(bp)) {
1981 for (i = 0; i < bp->rx_nr_rings; i++) {
1982 rxq = bp->eth_dev->data->rx_queues[i];
1983 if (rxq->rx_ring->rx_ring_struct->fw_ring_id == fwr)
1987 for (i = 0; i < bp->rx_nr_rings; i++) {
1988 if (bp->grp_info[i].fw_grp_id == fwr)
1993 return INVALID_HW_RING_ID;
1996 static int bnxt_reta_update_op(struct rte_eth_dev *eth_dev,
1997 struct rte_eth_rss_reta_entry64 *reta_conf,
2000 struct bnxt *bp = eth_dev->data->dev_private;
2001 struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
2002 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2003 uint16_t tbl_size = bnxt_rss_hash_tbl_size(bp);
2007 rc = is_bnxt_in_error(bp);
2011 if (!vnic->rss_table)
2014 if (!(dev_conf->rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG))
2017 if (reta_size != tbl_size) {
2018 PMD_DRV_LOG(ERR, "The configured hash table lookup size "
2019 "(%d) must equal the size supported by the hardware "
2020 "(%d)\n", reta_size, tbl_size);
2024 for (i = 0; i < reta_size; i++) {
2025 struct bnxt_rx_queue *rxq;
2027 idx = i / RTE_ETH_RETA_GROUP_SIZE;
2028 sft = i % RTE_ETH_RETA_GROUP_SIZE;
2030 if (!(reta_conf[idx].mask & (1ULL << sft)))
2033 rxq = bnxt_qid_to_rxq(bp, reta_conf[idx].reta[sft]);
2035 PMD_DRV_LOG(ERR, "Invalid ring in reta_conf.\n");
2039 if (BNXT_CHIP_P5(bp)) {
2040 vnic->rss_table[i * 2] =
2041 rxq->rx_ring->rx_ring_struct->fw_ring_id;
2042 vnic->rss_table[i * 2 + 1] =
2043 rxq->cp_ring->cp_ring_struct->fw_ring_id;
2045 vnic->rss_table[i] =
2046 vnic->fw_grp_ids[reta_conf[idx].reta[sft]];
2050 rc = bnxt_hwrm_vnic_rss_cfg(bp, vnic);
2054 static int bnxt_reta_query_op(struct rte_eth_dev *eth_dev,
2055 struct rte_eth_rss_reta_entry64 *reta_conf,
2058 struct bnxt *bp = eth_dev->data->dev_private;
2059 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2060 uint16_t tbl_size = bnxt_rss_hash_tbl_size(bp);
2061 uint16_t idx, sft, i;
2064 rc = is_bnxt_in_error(bp);
2070 if (!vnic->rss_table)
2073 if (reta_size != tbl_size) {
2074 PMD_DRV_LOG(ERR, "The configured hash table lookup size "
2075 "(%d) must equal the size supported by the hardware "
2076 "(%d)\n", reta_size, tbl_size);
2080 for (idx = 0, i = 0; i < reta_size; i++) {
2081 idx = i / RTE_ETH_RETA_GROUP_SIZE;
2082 sft = i % RTE_ETH_RETA_GROUP_SIZE;
2084 if (reta_conf[idx].mask & (1ULL << sft)) {
2087 if (BNXT_CHIP_P5(bp))
2088 qid = bnxt_rss_to_qid(bp,
2089 vnic->rss_table[i * 2]);
2091 qid = bnxt_rss_to_qid(bp, vnic->rss_table[i]);
2093 if (qid == INVALID_HW_RING_ID) {
2094 PMD_DRV_LOG(ERR, "Inv. entry in rss table.\n");
2097 reta_conf[idx].reta[sft] = qid;
2104 static int bnxt_rss_hash_update_op(struct rte_eth_dev *eth_dev,
2105 struct rte_eth_rss_conf *rss_conf)
2107 struct bnxt *bp = eth_dev->data->dev_private;
2108 struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
2109 struct bnxt_vnic_info *vnic;
2112 rc = is_bnxt_in_error(bp);
2117 * If RSS enablement were different than dev_configure,
2118 * then return -EINVAL
2120 if (dev_conf->rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG) {
2121 if (!rss_conf->rss_hf)
2122 PMD_DRV_LOG(ERR, "Hash type NONE\n");
2124 if (rss_conf->rss_hf & BNXT_ETH_RSS_SUPPORT)
2128 bp->flags |= BNXT_FLAG_UPDATE_HASH;
2129 memcpy(ð_dev->data->dev_conf.rx_adv_conf.rss_conf,
2133 /* Update the default RSS VNIC(s) */
2134 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2135 vnic->hash_type = bnxt_rte_to_hwrm_hash_types(rss_conf->rss_hf);
2137 bnxt_rte_to_hwrm_hash_level(bp, rss_conf->rss_hf,
2138 RTE_ETH_RSS_LEVEL(rss_conf->rss_hf));
2141 * If hashkey is not specified, use the previously configured
2144 if (!rss_conf->rss_key)
2147 if (rss_conf->rss_key_len != HW_HASH_KEY_SIZE) {
2149 "Invalid hashkey length, should be %d bytes\n",
2153 memcpy(vnic->rss_hash_key, rss_conf->rss_key, rss_conf->rss_key_len);
2156 rc = bnxt_hwrm_vnic_rss_cfg(bp, vnic);
2160 static int bnxt_rss_hash_conf_get_op(struct rte_eth_dev *eth_dev,
2161 struct rte_eth_rss_conf *rss_conf)
2163 struct bnxt *bp = eth_dev->data->dev_private;
2164 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2166 uint32_t hash_types;
2168 rc = is_bnxt_in_error(bp);
2172 /* RSS configuration is the same for all VNICs */
2173 if (vnic && vnic->rss_hash_key) {
2174 if (rss_conf->rss_key) {
2175 len = rss_conf->rss_key_len <= HW_HASH_KEY_SIZE ?
2176 rss_conf->rss_key_len : HW_HASH_KEY_SIZE;
2177 memcpy(rss_conf->rss_key, vnic->rss_hash_key, len);
2180 hash_types = vnic->hash_type;
2181 rss_conf->rss_hf = 0;
2182 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV4) {
2183 rss_conf->rss_hf |= RTE_ETH_RSS_IPV4;
2184 hash_types &= ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV4;
2186 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV4) {
2187 rss_conf->rss_hf |= RTE_ETH_RSS_NONFRAG_IPV4_TCP;
2189 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV4;
2191 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV4) {
2192 rss_conf->rss_hf |= RTE_ETH_RSS_NONFRAG_IPV4_UDP;
2194 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV4;
2196 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV6) {
2197 rss_conf->rss_hf |= RTE_ETH_RSS_IPV6;
2198 hash_types &= ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV6;
2200 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV6) {
2201 rss_conf->rss_hf |= RTE_ETH_RSS_NONFRAG_IPV6_TCP;
2203 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV6;
2205 if (hash_types & HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV6) {
2206 rss_conf->rss_hf |= RTE_ETH_RSS_NONFRAG_IPV6_UDP;
2208 ~HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV6;
2212 bnxt_hwrm_to_rte_rss_level(bp, vnic->hash_mode);
2216 "Unknown RSS config from firmware (%08x), RSS disabled",
2221 rss_conf->rss_hf = 0;
2226 static int bnxt_flow_ctrl_get_op(struct rte_eth_dev *dev,
2227 struct rte_eth_fc_conf *fc_conf)
2229 struct bnxt *bp = dev->data->dev_private;
2230 struct rte_eth_link link_info;
2233 rc = is_bnxt_in_error(bp);
2237 rc = bnxt_get_hwrm_link_config(bp, &link_info);
2241 memset(fc_conf, 0, sizeof(*fc_conf));
2242 if (bp->link_info->auto_pause)
2243 fc_conf->autoneg = 1;
2244 switch (bp->link_info->pause) {
2246 fc_conf->mode = RTE_ETH_FC_NONE;
2248 case HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_TX:
2249 fc_conf->mode = RTE_ETH_FC_TX_PAUSE;
2251 case HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_RX:
2252 fc_conf->mode = RTE_ETH_FC_RX_PAUSE;
2254 case (HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_TX |
2255 HWRM_PORT_PHY_QCFG_OUTPUT_PAUSE_RX):
2256 fc_conf->mode = RTE_ETH_FC_FULL;
2262 static int bnxt_flow_ctrl_set_op(struct rte_eth_dev *dev,
2263 struct rte_eth_fc_conf *fc_conf)
2265 struct bnxt *bp = dev->data->dev_private;
2268 rc = is_bnxt_in_error(bp);
2272 if (!BNXT_SINGLE_PF(bp)) {
2274 "Flow Control Settings cannot be modified on VF or on shared PF\n");
2278 switch (fc_conf->mode) {
2279 case RTE_ETH_FC_NONE:
2280 bp->link_info->auto_pause = 0;
2281 bp->link_info->force_pause = 0;
2283 case RTE_ETH_FC_RX_PAUSE:
2284 if (fc_conf->autoneg) {
2285 bp->link_info->auto_pause =
2286 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_RX;
2287 bp->link_info->force_pause = 0;
2289 bp->link_info->auto_pause = 0;
2290 bp->link_info->force_pause =
2291 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_RX;
2294 case RTE_ETH_FC_TX_PAUSE:
2295 if (fc_conf->autoneg) {
2296 bp->link_info->auto_pause =
2297 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_TX;
2298 bp->link_info->force_pause = 0;
2300 bp->link_info->auto_pause = 0;
2301 bp->link_info->force_pause =
2302 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_TX;
2305 case RTE_ETH_FC_FULL:
2306 if (fc_conf->autoneg) {
2307 bp->link_info->auto_pause =
2308 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_TX |
2309 HWRM_PORT_PHY_CFG_INPUT_AUTO_PAUSE_RX;
2310 bp->link_info->force_pause = 0;
2312 bp->link_info->auto_pause = 0;
2313 bp->link_info->force_pause =
2314 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_TX |
2315 HWRM_PORT_PHY_CFG_INPUT_FORCE_PAUSE_RX;
2319 return bnxt_set_hwrm_link_config(bp, true);
2322 /* Add UDP tunneling port */
2324 bnxt_udp_tunnel_port_add_op(struct rte_eth_dev *eth_dev,
2325 struct rte_eth_udp_tunnel *udp_tunnel)
2327 struct bnxt *bp = eth_dev->data->dev_private;
2328 uint16_t tunnel_type = 0;
2331 rc = is_bnxt_in_error(bp);
2335 switch (udp_tunnel->prot_type) {
2336 case RTE_ETH_TUNNEL_TYPE_VXLAN:
2337 if (bp->vxlan_port_cnt) {
2338 PMD_DRV_LOG(ERR, "Tunnel Port %d already programmed\n",
2339 udp_tunnel->udp_port);
2340 if (bp->vxlan_port != udp_tunnel->udp_port) {
2341 PMD_DRV_LOG(ERR, "Only one port allowed\n");
2344 bp->vxlan_port_cnt++;
2348 HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_VXLAN;
2350 case RTE_ETH_TUNNEL_TYPE_GENEVE:
2351 if (bp->geneve_port_cnt) {
2352 PMD_DRV_LOG(ERR, "Tunnel Port %d already programmed\n",
2353 udp_tunnel->udp_port);
2354 if (bp->geneve_port != udp_tunnel->udp_port) {
2355 PMD_DRV_LOG(ERR, "Only one port allowed\n");
2358 bp->geneve_port_cnt++;
2362 HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_GENEVE;
2365 PMD_DRV_LOG(ERR, "Tunnel type is not supported\n");
2368 rc = bnxt_hwrm_tunnel_dst_port_alloc(bp, udp_tunnel->udp_port,
2375 HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_VXLAN)
2376 bp->vxlan_port_cnt++;
2379 HWRM_TUNNEL_DST_PORT_ALLOC_INPUT_TUNNEL_TYPE_GENEVE)
2380 bp->geneve_port_cnt++;
2386 bnxt_udp_tunnel_port_del_op(struct rte_eth_dev *eth_dev,
2387 struct rte_eth_udp_tunnel *udp_tunnel)
2389 struct bnxt *bp = eth_dev->data->dev_private;
2390 uint16_t tunnel_type = 0;
2394 rc = is_bnxt_in_error(bp);
2398 switch (udp_tunnel->prot_type) {
2399 case RTE_ETH_TUNNEL_TYPE_VXLAN:
2400 if (!bp->vxlan_port_cnt) {
2401 PMD_DRV_LOG(ERR, "No Tunnel port configured yet\n");
2404 if (bp->vxlan_port != udp_tunnel->udp_port) {
2405 PMD_DRV_LOG(ERR, "Req Port: %d. Configured port: %d\n",
2406 udp_tunnel->udp_port, bp->vxlan_port);
2409 if (--bp->vxlan_port_cnt)
2413 HWRM_TUNNEL_DST_PORT_FREE_INPUT_TUNNEL_TYPE_VXLAN;
2414 port = bp->vxlan_fw_dst_port_id;
2416 case RTE_ETH_TUNNEL_TYPE_GENEVE:
2417 if (!bp->geneve_port_cnt) {
2418 PMD_DRV_LOG(ERR, "No Tunnel port configured yet\n");
2421 if (bp->geneve_port != udp_tunnel->udp_port) {
2422 PMD_DRV_LOG(ERR, "Req Port: %d. Configured port: %d\n",
2423 udp_tunnel->udp_port, bp->geneve_port);
2426 if (--bp->geneve_port_cnt)
2430 HWRM_TUNNEL_DST_PORT_FREE_INPUT_TUNNEL_TYPE_GENEVE;
2431 port = bp->geneve_fw_dst_port_id;
2434 PMD_DRV_LOG(ERR, "Tunnel type is not supported\n");
2438 rc = bnxt_hwrm_tunnel_dst_port_free(bp, port, tunnel_type);
2442 static int bnxt_del_vlan_filter(struct bnxt *bp, uint16_t vlan_id)
2444 struct bnxt_filter_info *filter;
2445 struct bnxt_vnic_info *vnic;
2447 uint32_t chk = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN;
2449 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2450 filter = STAILQ_FIRST(&vnic->filter);
2452 /* Search for this matching MAC+VLAN filter */
2453 if (bnxt_vlan_filter_exists(bp, filter, chk, vlan_id)) {
2454 /* Delete the filter */
2455 rc = bnxt_hwrm_clear_l2_filter(bp, filter);
2458 STAILQ_REMOVE(&vnic->filter, filter,
2459 bnxt_filter_info, next);
2460 bnxt_free_filter(bp, filter);
2462 "Deleted vlan filter for %d\n",
2466 filter = STAILQ_NEXT(filter, next);
2471 static int bnxt_add_vlan_filter(struct bnxt *bp, uint16_t vlan_id)
2473 struct bnxt_filter_info *filter;
2474 struct bnxt_vnic_info *vnic;
2476 uint32_t en = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN |
2477 HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN_MASK;
2478 uint32_t chk = HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_IVLAN;
2480 /* Implementation notes on the use of VNIC in this command:
2482 * By default, these filters belong to default vnic for the function.
2483 * Once these filters are set up, only destination VNIC can be modified.
2484 * If the destination VNIC is not specified in this command,
2485 * then the HWRM shall only create an l2 context id.
2488 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2489 filter = STAILQ_FIRST(&vnic->filter);
2490 /* Check if the VLAN has already been added */
2492 if (bnxt_vlan_filter_exists(bp, filter, chk, vlan_id))
2495 filter = STAILQ_NEXT(filter, next);
2498 /* No match found. Alloc a fresh filter and issue the L2_FILTER_ALLOC
2499 * command to create MAC+VLAN filter with the right flags, enables set.
2501 filter = bnxt_alloc_filter(bp);
2504 "MAC/VLAN filter alloc failed\n");
2507 /* MAC + VLAN ID filter */
2508 /* If l2_ivlan == 0 and l2_ivlan_mask != 0, only
2509 * untagged packets are received
2511 * If l2_ivlan != 0 and l2_ivlan_mask != 0, untagged
2512 * packets and only the programmed vlan's packets are received
2514 filter->l2_ivlan = vlan_id;
2515 filter->l2_ivlan_mask = 0x0FFF;
2516 filter->enables |= en;
2517 filter->flags |= HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
2519 rc = bnxt_hwrm_set_l2_filter(bp, vnic->fw_vnic_id, filter);
2521 /* Free the newly allocated filter as we were
2522 * not able to create the filter in hardware.
2524 bnxt_free_filter(bp, filter);
2528 filter->mac_index = 0;
2529 /* Add this new filter to the list */
2531 STAILQ_INSERT_HEAD(&vnic->filter, filter, next);
2533 STAILQ_INSERT_TAIL(&vnic->filter, filter, next);
2536 "Added Vlan filter for %d\n", vlan_id);
2540 static int bnxt_vlan_filter_set_op(struct rte_eth_dev *eth_dev,
2541 uint16_t vlan_id, int on)
2543 struct bnxt *bp = eth_dev->data->dev_private;
2546 rc = is_bnxt_in_error(bp);
2550 if (!eth_dev->data->dev_started) {
2551 PMD_DRV_LOG(ERR, "port must be started before setting vlan\n");
2555 /* These operations apply to ALL existing MAC/VLAN filters */
2557 return bnxt_add_vlan_filter(bp, vlan_id);
2559 return bnxt_del_vlan_filter(bp, vlan_id);
2562 static int bnxt_del_dflt_mac_filter(struct bnxt *bp,
2563 struct bnxt_vnic_info *vnic)
2565 struct bnxt_filter_info *filter;
2568 filter = STAILQ_FIRST(&vnic->filter);
2570 if (filter->mac_index == 0 &&
2571 !memcmp(filter->l2_addr, bp->mac_addr,
2572 RTE_ETHER_ADDR_LEN)) {
2573 rc = bnxt_hwrm_clear_l2_filter(bp, filter);
2575 STAILQ_REMOVE(&vnic->filter, filter,
2576 bnxt_filter_info, next);
2577 bnxt_free_filter(bp, filter);
2581 filter = STAILQ_NEXT(filter, next);
2587 bnxt_config_vlan_hw_filter(struct bnxt *bp, uint64_t rx_offloads)
2589 struct bnxt_vnic_info *vnic;
2593 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2594 if (!(rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER)) {
2595 /* Remove any VLAN filters programmed */
2596 for (i = 0; i < RTE_ETHER_MAX_VLAN_ID; i++)
2597 bnxt_del_vlan_filter(bp, i);
2599 rc = bnxt_add_mac_filter(bp, vnic, NULL, 0, 0);
2603 /* Default filter will allow packets that match the
2604 * dest mac. So, it has to be deleted, otherwise, we
2605 * will endup receiving vlan packets for which the
2606 * filter is not programmed, when hw-vlan-filter
2607 * configuration is ON
2609 bnxt_del_dflt_mac_filter(bp, vnic);
2610 /* This filter will allow only untagged packets */
2611 bnxt_add_vlan_filter(bp, 0);
2613 PMD_DRV_LOG(DEBUG, "VLAN Filtering: %d\n",
2614 !!(rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER));
2619 static int bnxt_free_one_vnic(struct bnxt *bp, uint16_t vnic_id)
2621 struct bnxt_vnic_info *vnic = &bp->vnic_info[vnic_id];
2625 /* Destroy vnic filters and vnic */
2626 if (bp->eth_dev->data->dev_conf.rxmode.offloads &
2627 RTE_ETH_RX_OFFLOAD_VLAN_FILTER) {
2628 for (i = 0; i < RTE_ETHER_MAX_VLAN_ID; i++)
2629 bnxt_del_vlan_filter(bp, i);
2631 bnxt_del_dflt_mac_filter(bp, vnic);
2633 rc = bnxt_hwrm_vnic_ctx_free(bp, vnic);
2637 rc = bnxt_hwrm_vnic_free(bp, vnic);
2641 rte_free(vnic->fw_grp_ids);
2642 vnic->fw_grp_ids = NULL;
2644 vnic->rx_queue_cnt = 0;
2650 bnxt_config_vlan_hw_stripping(struct bnxt *bp, uint64_t rx_offloads)
2652 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2655 /* Destroy, recreate and reconfigure the default vnic */
2656 rc = bnxt_free_one_vnic(bp, 0);
2660 /* default vnic 0 */
2661 rc = bnxt_setup_one_vnic(bp, 0);
2665 if (bp->eth_dev->data->dev_conf.rxmode.offloads &
2666 RTE_ETH_RX_OFFLOAD_VLAN_FILTER) {
2667 rc = bnxt_add_vlan_filter(bp, 0);
2670 rc = bnxt_restore_vlan_filters(bp);
2674 rc = bnxt_add_mac_filter(bp, vnic, NULL, 0, 0);
2679 rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
2683 PMD_DRV_LOG(DEBUG, "VLAN Strip Offload: %d\n",
2684 !!(rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP));
2690 bnxt_vlan_offload_set_op(struct rte_eth_dev *dev, int mask)
2692 uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
2693 struct bnxt *bp = dev->data->dev_private;
2696 rc = is_bnxt_in_error(bp);
2700 /* Filter settings will get applied when port is started */
2701 if (!dev->data->dev_started)
2704 if (mask & RTE_ETH_VLAN_FILTER_MASK) {
2705 /* Enable or disable VLAN filtering */
2706 rc = bnxt_config_vlan_hw_filter(bp, rx_offloads);
2711 if (mask & RTE_ETH_VLAN_STRIP_MASK) {
2712 /* Enable or disable VLAN stripping */
2713 rc = bnxt_config_vlan_hw_stripping(bp, rx_offloads);
2718 if (mask & RTE_ETH_VLAN_EXTEND_MASK) {
2719 if (rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_EXTEND)
2720 PMD_DRV_LOG(DEBUG, "Extend VLAN supported\n");
2722 PMD_DRV_LOG(INFO, "Extend VLAN unsupported\n");
2729 bnxt_vlan_tpid_set_op(struct rte_eth_dev *dev, enum rte_vlan_type vlan_type,
2732 struct bnxt *bp = dev->data->dev_private;
2733 int qinq = dev->data->dev_conf.rxmode.offloads &
2734 RTE_ETH_RX_OFFLOAD_VLAN_EXTEND;
2736 if (vlan_type != RTE_ETH_VLAN_TYPE_INNER &&
2737 vlan_type != RTE_ETH_VLAN_TYPE_OUTER) {
2739 "Unsupported vlan type.");
2744 "QinQ not enabled. Needs to be ON as we can "
2745 "accelerate only outer vlan\n");
2749 if (vlan_type == RTE_ETH_VLAN_TYPE_OUTER) {
2751 case RTE_ETHER_TYPE_QINQ:
2753 TX_BD_LONG_CFA_META_VLAN_TPID_TPID88A8;
2755 case RTE_ETHER_TYPE_VLAN:
2757 TX_BD_LONG_CFA_META_VLAN_TPID_TPID8100;
2759 case RTE_ETHER_TYPE_QINQ1:
2761 TX_BD_LONG_CFA_META_VLAN_TPID_TPID9100;
2763 case RTE_ETHER_TYPE_QINQ2:
2765 TX_BD_LONG_CFA_META_VLAN_TPID_TPID9200;
2767 case RTE_ETHER_TYPE_QINQ3:
2769 TX_BD_LONG_CFA_META_VLAN_TPID_TPID9300;
2772 PMD_DRV_LOG(ERR, "Invalid TPID: %x\n", tpid);
2775 bp->outer_tpid_bd |= tpid;
2776 PMD_DRV_LOG(INFO, "outer_tpid_bd = %x\n", bp->outer_tpid_bd);
2777 } else if (vlan_type == RTE_ETH_VLAN_TYPE_INNER) {
2779 "Can accelerate only outer vlan in QinQ\n");
2787 bnxt_set_default_mac_addr_op(struct rte_eth_dev *dev,
2788 struct rte_ether_addr *addr)
2790 struct bnxt *bp = dev->data->dev_private;
2791 /* Default Filter is tied to VNIC 0 */
2792 struct bnxt_vnic_info *vnic = BNXT_GET_DEFAULT_VNIC(bp);
2795 rc = is_bnxt_in_error(bp);
2799 if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp))
2802 if (rte_is_zero_ether_addr(addr))
2805 /* Filter settings will get applied when port is started */
2806 if (!dev->data->dev_started)
2809 /* Check if the requested MAC is already added */
2810 if (memcmp(addr, bp->mac_addr, RTE_ETHER_ADDR_LEN) == 0)
2813 /* Destroy filter and re-create it */
2814 bnxt_del_dflt_mac_filter(bp, vnic);
2816 memcpy(bp->mac_addr, addr, RTE_ETHER_ADDR_LEN);
2817 if (dev->data->dev_conf.rxmode.offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER) {
2818 /* This filter will allow only untagged packets */
2819 rc = bnxt_add_vlan_filter(bp, 0);
2821 rc = bnxt_add_mac_filter(bp, vnic, addr, 0, 0);
2824 PMD_DRV_LOG(DEBUG, "Set MAC addr\n");
2829 bnxt_dev_set_mc_addr_list_op(struct rte_eth_dev *eth_dev,
2830 struct rte_ether_addr *mc_addr_set,
2831 uint32_t nb_mc_addr)
2833 struct bnxt *bp = eth_dev->data->dev_private;
2834 char *mc_addr_list = (char *)mc_addr_set;
2835 struct bnxt_vnic_info *vnic;
2836 uint32_t off = 0, i = 0;
2839 rc = is_bnxt_in_error(bp);
2843 vnic = BNXT_GET_DEFAULT_VNIC(bp);
2845 if (nb_mc_addr > BNXT_MAX_MC_ADDRS) {
2846 vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
2850 /* TODO Check for Duplicate mcast addresses */
2851 vnic->flags &= ~BNXT_VNIC_INFO_ALLMULTI;
2852 for (i = 0; i < nb_mc_addr; i++) {
2853 memcpy(vnic->mc_list + off, &mc_addr_list[i],
2854 RTE_ETHER_ADDR_LEN);
2855 off += RTE_ETHER_ADDR_LEN;
2858 vnic->mc_addr_cnt = i;
2859 if (vnic->mc_addr_cnt)
2860 vnic->flags |= BNXT_VNIC_INFO_MCAST;
2862 vnic->flags &= ~BNXT_VNIC_INFO_MCAST;
2865 return bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
2869 bnxt_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
2871 struct bnxt *bp = dev->data->dev_private;
2872 uint8_t fw_major = (bp->fw_ver >> 24) & 0xff;
2873 uint8_t fw_minor = (bp->fw_ver >> 16) & 0xff;
2874 uint8_t fw_updt = (bp->fw_ver >> 8) & 0xff;
2875 uint8_t fw_rsvd = bp->fw_ver & 0xff;
2878 ret = snprintf(fw_version, fw_size, "%d.%d.%d.%d",
2879 fw_major, fw_minor, fw_updt, fw_rsvd);
2883 ret += 1; /* add the size of '\0' */
2884 if (fw_size < (size_t)ret)
2891 bnxt_rxq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
2892 struct rte_eth_rxq_info *qinfo)
2894 struct bnxt *bp = dev->data->dev_private;
2895 struct bnxt_rx_queue *rxq;
2897 if (is_bnxt_in_error(bp))
2900 rxq = dev->data->rx_queues[queue_id];
2902 qinfo->mp = rxq->mb_pool;
2903 qinfo->scattered_rx = dev->data->scattered_rx;
2904 qinfo->nb_desc = rxq->nb_rx_desc;
2906 qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
2907 qinfo->conf.rx_drop_en = rxq->drop_en;
2908 qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
2909 qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
2913 bnxt_txq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
2914 struct rte_eth_txq_info *qinfo)
2916 struct bnxt *bp = dev->data->dev_private;
2917 struct bnxt_tx_queue *txq;
2919 if (is_bnxt_in_error(bp))
2922 txq = dev->data->tx_queues[queue_id];
2924 qinfo->nb_desc = txq->nb_tx_desc;
2926 qinfo->conf.tx_thresh.pthresh = txq->pthresh;
2927 qinfo->conf.tx_thresh.hthresh = txq->hthresh;
2928 qinfo->conf.tx_thresh.wthresh = txq->wthresh;
2930 qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
2931 qinfo->conf.tx_rs_thresh = 0;
2932 qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
2933 qinfo->conf.offloads = txq->offloads;
2936 static const struct {
2937 eth_rx_burst_t pkt_burst;
2939 } bnxt_rx_burst_info[] = {
2940 {bnxt_recv_pkts, "Scalar"},
2941 #if defined(RTE_ARCH_X86)
2942 {bnxt_recv_pkts_vec, "Vector SSE"},
2944 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
2945 {bnxt_recv_pkts_vec_avx2, "Vector AVX2"},
2947 #if defined(RTE_ARCH_ARM64)
2948 {bnxt_recv_pkts_vec, "Vector Neon"},
2953 bnxt_rx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
2954 struct rte_eth_burst_mode *mode)
2956 eth_rx_burst_t pkt_burst = dev->rx_pkt_burst;
2959 for (i = 0; i < RTE_DIM(bnxt_rx_burst_info); i++) {
2960 if (pkt_burst == bnxt_rx_burst_info[i].pkt_burst) {
2961 snprintf(mode->info, sizeof(mode->info), "%s",
2962 bnxt_rx_burst_info[i].info);
2970 static const struct {
2971 eth_tx_burst_t pkt_burst;
2973 } bnxt_tx_burst_info[] = {
2974 {bnxt_xmit_pkts, "Scalar"},
2975 #if defined(RTE_ARCH_X86)
2976 {bnxt_xmit_pkts_vec, "Vector SSE"},
2978 #if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
2979 {bnxt_xmit_pkts_vec_avx2, "Vector AVX2"},
2981 #if defined(RTE_ARCH_ARM64)
2982 {bnxt_xmit_pkts_vec, "Vector Neon"},
2987 bnxt_tx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
2988 struct rte_eth_burst_mode *mode)
2990 eth_tx_burst_t pkt_burst = dev->tx_pkt_burst;
2993 for (i = 0; i < RTE_DIM(bnxt_tx_burst_info); i++) {
2994 if (pkt_burst == bnxt_tx_burst_info[i].pkt_burst) {
2995 snprintf(mode->info, sizeof(mode->info), "%s",
2996 bnxt_tx_burst_info[i].info);
3004 int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
3006 uint32_t overhead = BNXT_MAX_PKT_LEN - BNXT_MAX_MTU;
3007 struct bnxt *bp = eth_dev->data->dev_private;
3008 uint32_t new_pkt_size;
3012 rc = is_bnxt_in_error(bp);
3016 /* Exit if receive queues are not configured yet */
3017 if (!eth_dev->data->nb_rx_queues)
3020 new_pkt_size = new_mtu + overhead;
3023 * Disallow any MTU change that would require scattered receive support
3024 * if it is not already enabled.
3026 if (eth_dev->data->dev_started &&
3027 !eth_dev->data->scattered_rx &&
3029 eth_dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM)) {
3031 "MTU change would require scattered rx support. ");
3032 PMD_DRV_LOG(ERR, "Stop port before changing MTU.\n");
3036 if (new_mtu > RTE_ETHER_MTU)
3037 bp->flags |= BNXT_FLAG_JUMBO;
3039 bp->flags &= ~BNXT_FLAG_JUMBO;
3041 /* Is there a change in mtu setting? */
3042 if (eth_dev->data->mtu == new_mtu)
3045 for (i = 0; i < bp->nr_vnics; i++) {
3046 struct bnxt_vnic_info *vnic = &bp->vnic_info[i];
3049 vnic->mru = BNXT_VNIC_MRU(new_mtu);
3050 rc = bnxt_hwrm_vnic_cfg(bp, vnic);
3054 size = rte_pktmbuf_data_room_size(bp->rx_queues[0]->mb_pool);
3055 size -= RTE_PKTMBUF_HEADROOM;
3057 if (size < new_mtu) {
3058 rc = bnxt_hwrm_vnic_plcmode_cfg(bp, vnic);
3064 if (bnxt_hwrm_config_host_mtu(bp))
3065 PMD_DRV_LOG(WARNING, "Failed to configure host MTU\n");
3067 PMD_DRV_LOG(INFO, "New MTU is %d\n", new_mtu);
3073 bnxt_vlan_pvid_set_op(struct rte_eth_dev *dev, uint16_t pvid, int on)
3075 struct bnxt *bp = dev->data->dev_private;
3076 uint16_t vlan = bp->vlan;
3079 rc = is_bnxt_in_error(bp);
3083 if (!BNXT_SINGLE_PF(bp)) {
3084 PMD_DRV_LOG(ERR, "PVID cannot be modified on VF or on shared PF\n");
3087 bp->vlan = on ? pvid : 0;
3089 rc = bnxt_hwrm_set_default_vlan(bp, 0, 0);
3096 bnxt_dev_led_on_op(struct rte_eth_dev *dev)
3098 struct bnxt *bp = dev->data->dev_private;
3101 rc = is_bnxt_in_error(bp);
3105 return bnxt_hwrm_port_led_cfg(bp, true);
3109 bnxt_dev_led_off_op(struct rte_eth_dev *dev)
3111 struct bnxt *bp = dev->data->dev_private;
3114 rc = is_bnxt_in_error(bp);
3118 return bnxt_hwrm_port_led_cfg(bp, false);
3122 bnxt_rx_queue_count_op(void *rx_queue)
3125 struct bnxt_cp_ring_info *cpr;
3126 uint32_t desc = 0, raw_cons, cp_ring_size;
3127 struct bnxt_rx_queue *rxq;
3128 struct rx_pkt_cmpl *rxcmp;
3134 rc = is_bnxt_in_error(bp);
3139 raw_cons = cpr->cp_raw_cons;
3140 cp_ring_size = cpr->cp_ring_struct->ring_size;
3143 uint32_t agg_cnt, cons, cmpl_type;
3145 cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3146 rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3148 if (!bnxt_cpr_cmp_valid(rxcmp, raw_cons, cp_ring_size))
3151 cmpl_type = CMP_TYPE(rxcmp);
3153 switch (cmpl_type) {
3154 case CMPL_BASE_TYPE_RX_L2:
3155 case CMPL_BASE_TYPE_RX_L2_V2:
3156 agg_cnt = BNXT_RX_L2_AGG_BUFS(rxcmp);
3157 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3161 case CMPL_BASE_TYPE_RX_TPA_END:
3162 if (BNXT_CHIP_P5(rxq->bp)) {
3163 struct rx_tpa_v2_end_cmpl_hi *p5_tpa_end;
3165 p5_tpa_end = (void *)rxcmp;
3166 agg_cnt = BNXT_TPA_END_AGG_BUFS_TH(p5_tpa_end);
3168 struct rx_tpa_end_cmpl *tpa_end;
3170 tpa_end = (void *)rxcmp;
3171 agg_cnt = BNXT_TPA_END_AGG_BUFS(tpa_end);
3174 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3179 raw_cons += CMP_LEN(cmpl_type);
3187 bnxt_rx_descriptor_status_op(void *rx_queue, uint16_t offset)
3189 struct bnxt_rx_queue *rxq = rx_queue;
3190 struct bnxt_cp_ring_info *cpr;
3191 struct bnxt_rx_ring_info *rxr;
3192 uint32_t desc, raw_cons, cp_ring_size;
3193 struct bnxt *bp = rxq->bp;
3194 struct rx_pkt_cmpl *rxcmp;
3197 rc = is_bnxt_in_error(bp);
3201 if (offset >= rxq->nb_rx_desc)
3206 cp_ring_size = cpr->cp_ring_struct->ring_size;
3209 * For the vector receive case, the completion at the requested
3210 * offset can be indexed directly.
3212 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
3213 if (bp->flags & BNXT_FLAG_RX_VECTOR_PKT_MODE) {
3214 struct rx_pkt_cmpl *rxcmp;
3217 /* Check status of completion descriptor. */
3218 raw_cons = cpr->cp_raw_cons +
3219 offset * CMP_LEN(CMPL_BASE_TYPE_RX_L2);
3220 cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3221 rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3223 if (bnxt_cpr_cmp_valid(rxcmp, raw_cons, cp_ring_size))
3224 return RTE_ETH_RX_DESC_DONE;
3226 /* Check whether rx desc has an mbuf attached. */
3227 cons = RING_CMP(rxr->rx_ring_struct, raw_cons / 2);
3228 if (cons >= rxq->rxrearm_start &&
3229 cons < rxq->rxrearm_start + rxq->rxrearm_nb) {
3230 return RTE_ETH_RX_DESC_UNAVAIL;
3233 return RTE_ETH_RX_DESC_AVAIL;
3238 * For the non-vector receive case, scan the completion ring to
3239 * locate the completion descriptor for the requested offset.
3241 raw_cons = cpr->cp_raw_cons;
3244 uint32_t agg_cnt, cons, cmpl_type;
3246 cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
3247 rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
3249 if (!bnxt_cpr_cmp_valid(rxcmp, raw_cons, cp_ring_size))
3252 cmpl_type = CMP_TYPE(rxcmp);
3254 switch (cmpl_type) {
3255 case CMPL_BASE_TYPE_RX_L2:
3256 case CMPL_BASE_TYPE_RX_L2_V2:
3257 if (desc == offset) {
3258 cons = rxcmp->opaque;
3259 if (rxr->rx_buf_ring[cons])
3260 return RTE_ETH_RX_DESC_DONE;
3262 return RTE_ETH_RX_DESC_UNAVAIL;
3264 agg_cnt = BNXT_RX_L2_AGG_BUFS(rxcmp);
3265 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3269 case CMPL_BASE_TYPE_RX_TPA_END:
3271 return RTE_ETH_RX_DESC_DONE;
3273 if (BNXT_CHIP_P5(rxq->bp)) {
3274 struct rx_tpa_v2_end_cmpl_hi *p5_tpa_end;
3276 p5_tpa_end = (void *)rxcmp;
3277 agg_cnt = BNXT_TPA_END_AGG_BUFS_TH(p5_tpa_end);
3279 struct rx_tpa_end_cmpl *tpa_end;
3281 tpa_end = (void *)rxcmp;
3282 agg_cnt = BNXT_TPA_END_AGG_BUFS(tpa_end);
3285 raw_cons = raw_cons + CMP_LEN(cmpl_type) + agg_cnt;
3290 raw_cons += CMP_LEN(cmpl_type);
3294 return RTE_ETH_RX_DESC_AVAIL;
3298 bnxt_tx_descriptor_status_op(void *tx_queue, uint16_t offset)
3300 struct bnxt_tx_queue *txq = (struct bnxt_tx_queue *)tx_queue;
3301 struct bnxt_cp_ring_info *cpr = txq->cp_ring;
3302 uint32_t ring_mask, raw_cons, nb_tx_pkts = 0;
3303 struct cmpl_base *cp_desc_ring;
3306 rc = is_bnxt_in_error(txq->bp);
3310 if (offset >= txq->nb_tx_desc)
3313 /* Return "desc done" if descriptor is available for use. */
3314 if (bnxt_tx_bds_in_hw(txq) <= offset)
3315 return RTE_ETH_TX_DESC_DONE;
3317 raw_cons = cpr->cp_raw_cons;
3318 cp_desc_ring = cpr->cp_desc_ring;
3319 ring_mask = cpr->cp_ring_struct->ring_mask;
3321 /* Check to see if hw has posted a completion for the descriptor. */
3323 struct tx_cmpl *txcmp;
3326 cons = RING_CMPL(ring_mask, raw_cons);
3327 txcmp = (struct tx_cmpl *)&cp_desc_ring[cons];
3329 if (!bnxt_cpr_cmp_valid(txcmp, raw_cons, ring_mask + 1))
3332 if (CMP_TYPE(txcmp) == TX_CMPL_TYPE_TX_L2)
3333 nb_tx_pkts += rte_le_to_cpu_32(txcmp->opaque);
3335 if (nb_tx_pkts > offset)
3336 return RTE_ETH_TX_DESC_DONE;
3338 raw_cons = NEXT_RAW_CMP(raw_cons);
3341 /* Descriptor is pending transmit, not yet completed by hardware. */
3342 return RTE_ETH_TX_DESC_FULL;
3346 bnxt_flow_ops_get_op(struct rte_eth_dev *dev,
3347 const struct rte_flow_ops **ops)
3349 struct bnxt *bp = dev->data->dev_private;
3355 if (BNXT_ETH_DEV_IS_REPRESENTOR(dev)) {
3356 struct bnxt_representor *vfr = dev->data->dev_private;
3357 bp = vfr->parent_dev->data->dev_private;
3358 /* parent is deleted while children are still valid */
3360 PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR Error\n",
3361 dev->data->port_id);
3366 ret = is_bnxt_in_error(bp);
3370 /* PMD supports thread-safe flow operations. rte_flow API
3371 * functions can avoid mutex for multi-thread safety.
3373 dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
3375 if (BNXT_TRUFLOW_EN(bp))
3376 *ops = &bnxt_ulp_rte_flow_ops;
3378 *ops = &bnxt_flow_ops;
3383 static const uint32_t *
3384 bnxt_dev_supported_ptypes_get_op(struct rte_eth_dev *dev)
3386 static const uint32_t ptypes[] = {
3387 RTE_PTYPE_L2_ETHER_VLAN,
3388 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
3389 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
3393 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
3394 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
3395 RTE_PTYPE_INNER_L4_ICMP,
3396 RTE_PTYPE_INNER_L4_TCP,
3397 RTE_PTYPE_INNER_L4_UDP,
3401 if (!dev->rx_pkt_burst)
3407 static int bnxt_map_regs(struct bnxt *bp, uint32_t *reg_arr, int count,
3410 uint32_t reg_base = *reg_arr & 0xfffff000;
3414 for (i = 0; i < count; i++) {
3415 if ((reg_arr[i] & 0xfffff000) != reg_base)
3418 win_off = BNXT_GRCPF_REG_WINDOW_BASE_OUT + (reg_win - 1) * 4;
3419 rte_write32(reg_base, (uint8_t *)bp->bar0 + win_off);
3423 static int bnxt_map_ptp_regs(struct bnxt *bp)
3425 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3429 reg_arr = ptp->rx_regs;
3430 rc = bnxt_map_regs(bp, reg_arr, BNXT_PTP_RX_REGS, 5);
3434 reg_arr = ptp->tx_regs;
3435 rc = bnxt_map_regs(bp, reg_arr, BNXT_PTP_TX_REGS, 6);
3439 for (i = 0; i < BNXT_PTP_RX_REGS; i++)
3440 ptp->rx_mapped_regs[i] = 0x5000 + (ptp->rx_regs[i] & 0xfff);
3442 for (i = 0; i < BNXT_PTP_TX_REGS; i++)
3443 ptp->tx_mapped_regs[i] = 0x6000 + (ptp->tx_regs[i] & 0xfff);
3448 static void bnxt_unmap_ptp_regs(struct bnxt *bp)
3450 rte_write32(0, (uint8_t *)bp->bar0 +
3451 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 16);
3452 rte_write32(0, (uint8_t *)bp->bar0 +
3453 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 20);
3456 static uint64_t bnxt_cc_read(struct bnxt *bp)
3460 ns = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3461 BNXT_GRCPF_REG_SYNC_TIME));
3462 ns |= (uint64_t)(rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3463 BNXT_GRCPF_REG_SYNC_TIME + 4))) << 32;
3467 static int bnxt_get_tx_ts(struct bnxt *bp, uint64_t *ts)
3469 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3472 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3473 ptp->tx_mapped_regs[BNXT_PTP_TX_FIFO]));
3474 if (fifo & BNXT_PTP_TX_FIFO_EMPTY)
3477 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3478 ptp->tx_mapped_regs[BNXT_PTP_TX_FIFO]));
3479 *ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3480 ptp->tx_mapped_regs[BNXT_PTP_TX_TS_L]));
3481 *ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3482 ptp->tx_mapped_regs[BNXT_PTP_TX_TS_H])) << 32;
3483 rte_read32((uint8_t *)bp->bar0 + ptp->tx_mapped_regs[BNXT_PTP_TX_SEQ]);
3488 static int bnxt_clr_rx_ts(struct bnxt *bp, uint64_t *last_ts)
3490 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3491 struct bnxt_pf_info *pf = bp->pf;
3496 if (!ptp || (bp->flags & BNXT_FLAG_CHIP_P5))
3499 port_id = pf->port_id;
3500 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3501 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3502 while ((fifo & BNXT_PTP_RX_FIFO_PENDING) && (i < BNXT_PTP_RX_PND_CNT)) {
3503 rte_write32(1 << port_id, (uint8_t *)bp->bar0 +
3504 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO_ADV]);
3505 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3506 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3507 *last_ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3508 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_L]));
3509 *last_ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3510 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_H])) << 32;
3514 if (i >= BNXT_PTP_RX_PND_CNT)
3520 static int bnxt_get_rx_ts(struct bnxt *bp, uint64_t *ts)
3522 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3523 struct bnxt_pf_info *pf = bp->pf;
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))
3532 port_id = pf->port_id;
3533 rte_write32(1 << port_id, (uint8_t *)bp->bar0 +
3534 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO_ADV]);
3536 fifo = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3537 ptp->rx_mapped_regs[BNXT_PTP_RX_FIFO]));
3538 if (fifo & BNXT_PTP_RX_FIFO_PENDING)
3539 return bnxt_clr_rx_ts(bp, ts);
3541 *ts = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3542 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_L]));
3543 *ts |= (uint64_t)rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
3544 ptp->rx_mapped_regs[BNXT_PTP_RX_TS_H])) << 32;
3550 bnxt_timesync_write_time(struct rte_eth_dev *dev, const struct timespec *ts)
3553 struct bnxt *bp = dev->data->dev_private;
3554 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3559 ns = rte_timespec_to_ns(ts);
3560 /* Set the timecounters to a new value. */
3562 ptp->tx_tstamp_tc.nsec = ns;
3563 ptp->rx_tstamp_tc.nsec = ns;
3569 bnxt_timesync_read_time(struct rte_eth_dev *dev, struct timespec *ts)
3571 struct bnxt *bp = dev->data->dev_private;
3572 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3573 uint64_t ns, systime_cycles = 0;
3579 if (BNXT_CHIP_P5(bp))
3580 rc = bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_CURRENT_TIME,
3583 systime_cycles = bnxt_cc_read(bp);
3585 ns = rte_timecounter_update(&ptp->tc, systime_cycles);
3586 *ts = rte_ns_to_timespec(ns);
3591 bnxt_timesync_enable(struct rte_eth_dev *dev)
3593 struct bnxt *bp = dev->data->dev_private;
3594 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3602 ptp->tx_tstamp_en = 1;
3603 ptp->rxctl = BNXT_PTP_MSG_EVENTS;
3605 rc = bnxt_hwrm_ptp_cfg(bp);
3609 memset(&ptp->tc, 0, sizeof(struct rte_timecounter));
3610 memset(&ptp->rx_tstamp_tc, 0, sizeof(struct rte_timecounter));
3611 memset(&ptp->tx_tstamp_tc, 0, sizeof(struct rte_timecounter));
3613 ptp->tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3614 ptp->tc.cc_shift = shift;
3615 ptp->tc.nsec_mask = (1ULL << shift) - 1;
3617 ptp->rx_tstamp_tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3618 ptp->rx_tstamp_tc.cc_shift = shift;
3619 ptp->rx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
3621 ptp->tx_tstamp_tc.cc_mask = BNXT_CYCLECOUNTER_MASK;
3622 ptp->tx_tstamp_tc.cc_shift = shift;
3623 ptp->tx_tstamp_tc.nsec_mask = (1ULL << shift) - 1;
3625 if (!BNXT_CHIP_P5(bp))
3626 bnxt_map_ptp_regs(bp);
3628 rc = bnxt_ptp_start(bp);
3634 bnxt_timesync_disable(struct rte_eth_dev *dev)
3636 struct bnxt *bp = dev->data->dev_private;
3637 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3643 ptp->tx_tstamp_en = 0;
3646 bnxt_hwrm_ptp_cfg(bp);
3648 if (!BNXT_CHIP_P5(bp))
3649 bnxt_unmap_ptp_regs(bp);
3657 bnxt_timesync_read_rx_timestamp(struct rte_eth_dev *dev,
3658 struct timespec *timestamp,
3659 uint32_t flags __rte_unused)
3661 struct bnxt *bp = dev->data->dev_private;
3662 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3663 uint64_t rx_tstamp_cycles = 0;
3669 if (BNXT_CHIP_P5(bp))
3670 rx_tstamp_cycles = ptp->rx_timestamp;
3672 bnxt_get_rx_ts(bp, &rx_tstamp_cycles);
3674 ns = rte_timecounter_update(&ptp->rx_tstamp_tc, rx_tstamp_cycles);
3675 *timestamp = rte_ns_to_timespec(ns);
3680 bnxt_timesync_read_tx_timestamp(struct rte_eth_dev *dev,
3681 struct timespec *timestamp)
3683 struct bnxt *bp = dev->data->dev_private;
3684 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3685 uint64_t tx_tstamp_cycles = 0;
3692 if (BNXT_CHIP_P5(bp))
3693 rc = bnxt_hwrm_port_ts_query(bp, BNXT_PTP_FLAGS_PATH_TX,
3696 rc = bnxt_get_tx_ts(bp, &tx_tstamp_cycles);
3698 ns = rte_timecounter_update(&ptp->tx_tstamp_tc, tx_tstamp_cycles);
3699 *timestamp = rte_ns_to_timespec(ns);
3705 bnxt_timesync_adjust_time(struct rte_eth_dev *dev, int64_t delta)
3707 struct bnxt *bp = dev->data->dev_private;
3708 struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
3713 ptp->tc.nsec += delta;
3714 ptp->tx_tstamp_tc.nsec += delta;
3715 ptp->rx_tstamp_tc.nsec += delta;
3721 bnxt_get_eeprom_length_op(struct rte_eth_dev *dev)
3723 struct bnxt *bp = dev->data->dev_private;
3725 uint32_t dir_entries;
3726 uint32_t entry_length;
3728 rc = is_bnxt_in_error(bp);
3732 PMD_DRV_LOG(INFO, PCI_PRI_FMT "\n",
3733 bp->pdev->addr.domain, bp->pdev->addr.bus,
3734 bp->pdev->addr.devid, bp->pdev->addr.function);
3736 rc = bnxt_hwrm_nvm_get_dir_info(bp, &dir_entries, &entry_length);
3740 return dir_entries * entry_length;
3744 bnxt_get_eeprom_op(struct rte_eth_dev *dev,
3745 struct rte_dev_eeprom_info *in_eeprom)
3747 struct bnxt *bp = dev->data->dev_private;
3752 rc = is_bnxt_in_error(bp);
3756 PMD_DRV_LOG(INFO, PCI_PRI_FMT " in_eeprom->offset = %d len = %d\n",
3757 bp->pdev->addr.domain, bp->pdev->addr.bus,
3758 bp->pdev->addr.devid, bp->pdev->addr.function,
3759 in_eeprom->offset, in_eeprom->length);
3761 if (in_eeprom->offset == 0) /* special offset value to get directory */
3762 return bnxt_get_nvram_directory(bp, in_eeprom->length,
3765 index = in_eeprom->offset >> 24;
3766 offset = in_eeprom->offset & 0xffffff;
3769 return bnxt_hwrm_get_nvram_item(bp, index - 1, offset,
3770 in_eeprom->length, in_eeprom->data);
3775 static bool bnxt_dir_type_is_ape_bin_format(uint16_t dir_type)
3778 case BNX_DIR_TYPE_CHIMP_PATCH:
3779 case BNX_DIR_TYPE_BOOTCODE:
3780 case BNX_DIR_TYPE_BOOTCODE_2:
3781 case BNX_DIR_TYPE_APE_FW:
3782 case BNX_DIR_TYPE_APE_PATCH:
3783 case BNX_DIR_TYPE_KONG_FW:
3784 case BNX_DIR_TYPE_KONG_PATCH:
3785 case BNX_DIR_TYPE_BONO_FW:
3786 case BNX_DIR_TYPE_BONO_PATCH:
3794 static bool bnxt_dir_type_is_other_exec_format(uint16_t dir_type)
3797 case BNX_DIR_TYPE_AVS:
3798 case BNX_DIR_TYPE_EXP_ROM_MBA:
3799 case BNX_DIR_TYPE_PCIE:
3800 case BNX_DIR_TYPE_TSCF_UCODE:
3801 case BNX_DIR_TYPE_EXT_PHY:
3802 case BNX_DIR_TYPE_CCM:
3803 case BNX_DIR_TYPE_ISCSI_BOOT:
3804 case BNX_DIR_TYPE_ISCSI_BOOT_IPV6:
3805 case BNX_DIR_TYPE_ISCSI_BOOT_IPV4N6:
3813 static bool bnxt_dir_type_is_executable(uint16_t dir_type)
3815 return bnxt_dir_type_is_ape_bin_format(dir_type) ||
3816 bnxt_dir_type_is_other_exec_format(dir_type);
3820 bnxt_set_eeprom_op(struct rte_eth_dev *dev,
3821 struct rte_dev_eeprom_info *in_eeprom)
3823 struct bnxt *bp = dev->data->dev_private;
3824 uint8_t index, dir_op;
3825 uint16_t type, ext, ordinal, attr;
3828 rc = is_bnxt_in_error(bp);
3832 PMD_DRV_LOG(INFO, PCI_PRI_FMT " in_eeprom->offset = %d len = %d\n",
3833 bp->pdev->addr.domain, bp->pdev->addr.bus,
3834 bp->pdev->addr.devid, bp->pdev->addr.function,
3835 in_eeprom->offset, in_eeprom->length);
3838 PMD_DRV_LOG(ERR, "NVM write not supported from a VF\n");
3842 type = in_eeprom->magic >> 16;
3844 if (type == 0xffff) { /* special value for directory operations */
3845 index = in_eeprom->magic & 0xff;
3846 dir_op = in_eeprom->magic >> 8;
3850 case 0x0e: /* erase */
3851 if (in_eeprom->offset != ~in_eeprom->magic)
3853 return bnxt_hwrm_erase_nvram_directory(bp, index - 1);
3859 /* Create or re-write an NVM item: */
3860 if (bnxt_dir_type_is_executable(type) == true)
3862 ext = in_eeprom->magic & 0xffff;
3863 ordinal = in_eeprom->offset >> 16;
3864 attr = in_eeprom->offset & 0xffff;
3866 return bnxt_hwrm_flash_nvram(bp, type, ordinal, ext, attr,
3867 in_eeprom->data, in_eeprom->length);
3870 static int bnxt_get_module_info(struct rte_eth_dev *dev,
3871 struct rte_eth_dev_module_info *modinfo)
3873 uint8_t module_info[SFF_DIAG_SUPPORT_OFFSET + 1];
3874 struct bnxt *bp = dev->data->dev_private;
3877 /* No point in going further if phy status indicates
3878 * module is not inserted or if it is powered down or
3879 * if it is of type 10GBase-T
3881 if (bp->link_info->module_status >
3882 HWRM_PORT_PHY_QCFG_OUTPUT_MODULE_STATUS_WARNINGMSG) {
3883 PMD_DRV_LOG(NOTICE, "Port %u : Module is not inserted or is powered down\n",
3884 dev->data->port_id);
3888 /* This feature is not supported in older firmware versions */
3889 if (bp->hwrm_spec_code < 0x10202) {
3890 PMD_DRV_LOG(NOTICE, "Port %u : Feature is not supported in older firmware\n",
3891 dev->data->port_id);
3895 rc = bnxt_hwrm_read_sfp_module_eeprom_info(bp, I2C_DEV_ADDR_A0, 0, 0,
3896 SFF_DIAG_SUPPORT_OFFSET + 1,
3902 switch (module_info[0]) {
3903 case SFF_MODULE_ID_SFP:
3904 modinfo->type = RTE_ETH_MODULE_SFF_8472;
3905 modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8472_LEN;
3906 if (module_info[SFF_DIAG_SUPPORT_OFFSET] == 0)
3907 modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8436_LEN;
3909 case SFF_MODULE_ID_QSFP:
3910 case SFF_MODULE_ID_QSFP_PLUS:
3911 modinfo->type = RTE_ETH_MODULE_SFF_8436;
3912 modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8436_LEN;
3914 case SFF_MODULE_ID_QSFP28:
3915 modinfo->type = RTE_ETH_MODULE_SFF_8636;
3916 modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8636_MAX_LEN;
3917 if (module_info[SFF8636_FLATMEM_OFFSET] & SFF8636_FLATMEM_MASK)
3918 modinfo->eeprom_len = RTE_ETH_MODULE_SFF_8636_LEN;
3921 PMD_DRV_LOG(NOTICE, "Port %u : Unsupported module\n", dev->data->port_id);
3925 PMD_DRV_LOG(INFO, "Port %u : modinfo->type = %d modinfo->eeprom_len = %d\n",
3926 dev->data->port_id, modinfo->type, modinfo->eeprom_len);
3931 static int bnxt_get_module_eeprom(struct rte_eth_dev *dev,
3932 struct rte_dev_eeprom_info *info)
3934 uint8_t pg_addr[5] = { I2C_DEV_ADDR_A0, I2C_DEV_ADDR_A0 };
3935 uint32_t offset = info->offset, length = info->length;
3936 uint8_t module_info[SFF_DIAG_SUPPORT_OFFSET + 1];
3937 struct bnxt *bp = dev->data->dev_private;
3938 uint8_t *data = info->data;
3939 uint8_t page = offset >> 7;
3940 uint8_t max_pages = 2;
3944 rc = bnxt_hwrm_read_sfp_module_eeprom_info(bp, I2C_DEV_ADDR_A0, 0, 0,
3945 SFF_DIAG_SUPPORT_OFFSET + 1,
3950 switch (module_info[0]) {
3951 case SFF_MODULE_ID_SFP:
3952 module_info[SFF_DIAG_SUPPORT_OFFSET] = 0;
3953 if (module_info[SFF_DIAG_SUPPORT_OFFSET]) {
3954 pg_addr[2] = I2C_DEV_ADDR_A2;
3955 pg_addr[3] = I2C_DEV_ADDR_A2;
3959 case SFF_MODULE_ID_QSFP28:
3960 rc = bnxt_hwrm_read_sfp_module_eeprom_info(bp, I2C_DEV_ADDR_A0, 0,
3961 SFF8636_OPT_PAGES_OFFSET,
3966 if (opt_pages & SFF8636_PAGE1_MASK) {
3967 pg_addr[2] = I2C_DEV_ADDR_A0;
3970 if (opt_pages & SFF8636_PAGE2_MASK) {
3971 pg_addr[3] = I2C_DEV_ADDR_A0;
3974 if (~module_info[SFF8636_FLATMEM_OFFSET] & SFF8636_FLATMEM_MASK) {
3975 pg_addr[4] = I2C_DEV_ADDR_A0;
3983 memset(data, 0, length);
3986 while (length && page < max_pages) {
3987 uint8_t raw_page = page ? page - 1 : 0;
3990 if (pg_addr[page] == I2C_DEV_ADDR_A2)
3994 chunk = RTE_MIN(length, 256 - offset);
3996 if (pg_addr[page]) {
3997 rc = bnxt_hwrm_read_sfp_module_eeprom_info(bp, pg_addr[page],
4007 page += 1 + (chunk > 128);
4010 return length ? -EINVAL : 0;
4017 static const struct eth_dev_ops bnxt_dev_ops = {
4018 .dev_infos_get = bnxt_dev_info_get_op,
4019 .dev_close = bnxt_dev_close_op,
4020 .dev_configure = bnxt_dev_configure_op,
4021 .dev_start = bnxt_dev_start_op,
4022 .dev_stop = bnxt_dev_stop_op,
4023 .dev_set_link_up = bnxt_dev_set_link_up_op,
4024 .dev_set_link_down = bnxt_dev_set_link_down_op,
4025 .stats_get = bnxt_stats_get_op,
4026 .stats_reset = bnxt_stats_reset_op,
4027 .rx_queue_setup = bnxt_rx_queue_setup_op,
4028 .rx_queue_release = bnxt_rx_queue_release_op,
4029 .tx_queue_setup = bnxt_tx_queue_setup_op,
4030 .tx_queue_release = bnxt_tx_queue_release_op,
4031 .rx_queue_intr_enable = bnxt_rx_queue_intr_enable_op,
4032 .rx_queue_intr_disable = bnxt_rx_queue_intr_disable_op,
4033 .reta_update = bnxt_reta_update_op,
4034 .reta_query = bnxt_reta_query_op,
4035 .rss_hash_update = bnxt_rss_hash_update_op,
4036 .rss_hash_conf_get = bnxt_rss_hash_conf_get_op,
4037 .link_update = bnxt_link_update_op,
4038 .promiscuous_enable = bnxt_promiscuous_enable_op,
4039 .promiscuous_disable = bnxt_promiscuous_disable_op,
4040 .allmulticast_enable = bnxt_allmulticast_enable_op,
4041 .allmulticast_disable = bnxt_allmulticast_disable_op,
4042 .mac_addr_add = bnxt_mac_addr_add_op,
4043 .mac_addr_remove = bnxt_mac_addr_remove_op,
4044 .flow_ctrl_get = bnxt_flow_ctrl_get_op,
4045 .flow_ctrl_set = bnxt_flow_ctrl_set_op,
4046 .udp_tunnel_port_add = bnxt_udp_tunnel_port_add_op,
4047 .udp_tunnel_port_del = bnxt_udp_tunnel_port_del_op,
4048 .vlan_filter_set = bnxt_vlan_filter_set_op,
4049 .vlan_offload_set = bnxt_vlan_offload_set_op,
4050 .vlan_tpid_set = bnxt_vlan_tpid_set_op,
4051 .vlan_pvid_set = bnxt_vlan_pvid_set_op,
4052 .mtu_set = bnxt_mtu_set_op,
4053 .mac_addr_set = bnxt_set_default_mac_addr_op,
4054 .xstats_get = bnxt_dev_xstats_get_op,
4055 .xstats_get_names = bnxt_dev_xstats_get_names_op,
4056 .xstats_reset = bnxt_dev_xstats_reset_op,
4057 .fw_version_get = bnxt_fw_version_get,
4058 .set_mc_addr_list = bnxt_dev_set_mc_addr_list_op,
4059 .rxq_info_get = bnxt_rxq_info_get_op,
4060 .txq_info_get = bnxt_txq_info_get_op,
4061 .rx_burst_mode_get = bnxt_rx_burst_mode_get,
4062 .tx_burst_mode_get = bnxt_tx_burst_mode_get,
4063 .dev_led_on = bnxt_dev_led_on_op,
4064 .dev_led_off = bnxt_dev_led_off_op,
4065 .rx_queue_start = bnxt_rx_queue_start,
4066 .rx_queue_stop = bnxt_rx_queue_stop,
4067 .tx_queue_start = bnxt_tx_queue_start,
4068 .tx_queue_stop = bnxt_tx_queue_stop,
4069 .flow_ops_get = bnxt_flow_ops_get_op,
4070 .dev_supported_ptypes_get = bnxt_dev_supported_ptypes_get_op,
4071 .get_eeprom_length = bnxt_get_eeprom_length_op,
4072 .get_eeprom = bnxt_get_eeprom_op,
4073 .set_eeprom = bnxt_set_eeprom_op,
4074 .get_module_info = bnxt_get_module_info,
4075 .get_module_eeprom = bnxt_get_module_eeprom,
4076 .timesync_enable = bnxt_timesync_enable,
4077 .timesync_disable = bnxt_timesync_disable,
4078 .timesync_read_time = bnxt_timesync_read_time,
4079 .timesync_write_time = bnxt_timesync_write_time,
4080 .timesync_adjust_time = bnxt_timesync_adjust_time,
4081 .timesync_read_rx_timestamp = bnxt_timesync_read_rx_timestamp,
4082 .timesync_read_tx_timestamp = bnxt_timesync_read_tx_timestamp,
4085 static uint32_t bnxt_map_reset_regs(struct bnxt *bp, uint32_t reg)
4089 /* Only pre-map the reset GRC registers using window 3 */
4090 rte_write32(reg & 0xfffff000, (uint8_t *)bp->bar0 +
4091 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 8);
4093 offset = BNXT_GRCP_WINDOW_3_BASE + (reg & 0xffc);
4098 int bnxt_map_fw_health_status_regs(struct bnxt *bp)
4100 struct bnxt_error_recovery_info *info = bp->recovery_info;
4101 uint32_t reg_base = 0xffffffff;
4104 /* Only pre-map the monitoring GRC registers using window 2 */
4105 for (i = 0; i < BNXT_FW_STATUS_REG_CNT; i++) {
4106 uint32_t reg = info->status_regs[i];
4108 if (BNXT_FW_STATUS_REG_TYPE(reg) != BNXT_FW_STATUS_REG_TYPE_GRC)
4111 if (reg_base == 0xffffffff)
4112 reg_base = reg & 0xfffff000;
4113 if ((reg & 0xfffff000) != reg_base)
4116 /* Use mask 0xffc as the Lower 2 bits indicates
4117 * address space location
4119 info->mapped_status_regs[i] = BNXT_GRCP_WINDOW_2_BASE +
4123 if (reg_base == 0xffffffff)
4126 rte_write32(reg_base, (uint8_t *)bp->bar0 +
4127 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
4132 static void bnxt_write_fw_reset_reg(struct bnxt *bp, uint32_t index)
4134 struct bnxt_error_recovery_info *info = bp->recovery_info;
4135 uint32_t delay = info->delay_after_reset[index];
4136 uint32_t val = info->reset_reg_val[index];
4137 uint32_t reg = info->reset_reg[index];
4138 uint32_t type, offset;
4141 type = BNXT_FW_STATUS_REG_TYPE(reg);
4142 offset = BNXT_FW_STATUS_REG_OFF(reg);
4145 case BNXT_FW_STATUS_REG_TYPE_CFG:
4146 ret = rte_pci_write_config(bp->pdev, &val, sizeof(val), offset);
4148 PMD_DRV_LOG(ERR, "Failed to write %#x at PCI offset %#x",
4153 case BNXT_FW_STATUS_REG_TYPE_GRC:
4154 offset = bnxt_map_reset_regs(bp, offset);
4155 rte_write32(val, (uint8_t *)bp->bar0 + offset);
4157 case BNXT_FW_STATUS_REG_TYPE_BAR0:
4158 rte_write32(val, (uint8_t *)bp->bar0 + offset);
4161 /* wait on a specific interval of time until core reset is complete */
4163 rte_delay_ms(delay);
4166 static void bnxt_dev_cleanup(struct bnxt *bp)
4168 bp->eth_dev->data->dev_link.link_status = 0;
4169 bp->link_info->link_up = 0;
4170 if (bp->eth_dev->data->dev_started)
4171 bnxt_dev_stop(bp->eth_dev);
4173 bnxt_uninit_resources(bp, true);
4177 bnxt_check_fw_reset_done(struct bnxt *bp)
4179 int timeout = bp->fw_reset_max_msecs;
4184 rc = rte_pci_read_config(bp->pdev, &val, sizeof(val), PCI_SUBSYSTEM_ID_OFFSET);
4186 PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x", PCI_SUBSYSTEM_ID_OFFSET);
4192 } while (timeout--);
4194 if (val == 0xffff) {
4195 PMD_DRV_LOG(ERR, "Firmware reset aborted, PCI config space invalid\n");
4202 static int bnxt_restore_vlan_filters(struct bnxt *bp)
4204 struct rte_eth_dev *dev = bp->eth_dev;
4205 struct rte_vlan_filter_conf *vfc;
4209 for (vlan_id = 1; vlan_id <= RTE_ETHER_MAX_VLAN_ID; vlan_id++) {
4210 vfc = &dev->data->vlan_filter_conf;
4211 vidx = vlan_id / 64;
4212 vbit = vlan_id % 64;
4214 /* Each bit corresponds to a VLAN id */
4215 if (vfc->ids[vidx] & (UINT64_C(1) << vbit)) {
4216 rc = bnxt_add_vlan_filter(bp, vlan_id);
4225 static int bnxt_restore_mac_filters(struct bnxt *bp)
4227 struct rte_eth_dev *dev = bp->eth_dev;
4228 struct rte_eth_dev_info dev_info;
4229 struct rte_ether_addr *addr;
4235 if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp))
4238 rc = bnxt_dev_info_get_op(dev, &dev_info);
4242 /* replay MAC address configuration */
4243 for (i = 1; i < dev_info.max_mac_addrs; i++) {
4244 addr = &dev->data->mac_addrs[i];
4246 /* skip zero address */
4247 if (rte_is_zero_ether_addr(addr))
4251 pool_mask = dev->data->mac_pool_sel[i];
4254 if (pool_mask & 1ULL) {
4255 rc = bnxt_mac_addr_add_op(dev, addr, i, pool);
4261 } while (pool_mask);
4267 static int bnxt_restore_filters(struct bnxt *bp)
4269 struct rte_eth_dev *dev = bp->eth_dev;
4272 if (dev->data->all_multicast) {
4273 ret = bnxt_allmulticast_enable_op(dev);
4277 if (dev->data->promiscuous) {
4278 ret = bnxt_promiscuous_enable_op(dev);
4283 ret = bnxt_restore_mac_filters(bp);
4287 ret = bnxt_restore_vlan_filters(bp);
4288 /* TODO restore other filters as well */
4292 static int bnxt_check_fw_ready(struct bnxt *bp)
4294 int timeout = bp->fw_reset_max_msecs;
4298 rc = bnxt_hwrm_poll_ver_get(bp);
4301 rte_delay_ms(BNXT_FW_READY_WAIT_INTERVAL);
4302 timeout -= BNXT_FW_READY_WAIT_INTERVAL;
4303 } while (rc && timeout > 0);
4306 PMD_DRV_LOG(ERR, "FW is not Ready after reset\n");
4311 static void bnxt_dev_recover(void *arg)
4313 struct bnxt *bp = arg;
4316 pthread_mutex_lock(&bp->err_recovery_lock);
4318 if (!bp->fw_reset_min_msecs) {
4319 rc = bnxt_check_fw_reset_done(bp);
4324 /* Clear Error flag so that device re-init should happen */
4325 bp->flags &= ~BNXT_FLAG_FATAL_ERROR;
4327 rc = bnxt_check_fw_ready(bp);
4331 rc = bnxt_init_resources(bp, true);
4334 "Failed to initialize resources after reset\n");
4337 /* clear reset flag as the device is initialized now */
4338 bp->flags &= ~BNXT_FLAG_FW_RESET;
4340 rc = bnxt_dev_start_op(bp->eth_dev);
4342 PMD_DRV_LOG(ERR, "Failed to start port after reset\n");
4346 rc = bnxt_restore_filters(bp);
4350 PMD_DRV_LOG(INFO, "Recovered from FW reset\n");
4351 pthread_mutex_unlock(&bp->err_recovery_lock);
4355 bnxt_dev_stop(bp->eth_dev);
4357 bp->flags |= BNXT_FLAG_FATAL_ERROR;
4358 bnxt_uninit_resources(bp, false);
4359 if (bp->eth_dev->data->dev_conf.intr_conf.rmv)
4360 rte_eth_dev_callback_process(bp->eth_dev,
4361 RTE_ETH_EVENT_INTR_RMV,
4363 pthread_mutex_unlock(&bp->err_recovery_lock);
4364 PMD_DRV_LOG(ERR, "Failed to recover from FW reset\n");
4367 void bnxt_dev_reset_and_resume(void *arg)
4369 struct bnxt *bp = arg;
4370 uint32_t us = US_PER_MS * bp->fw_reset_min_msecs;
4374 bnxt_dev_cleanup(bp);
4376 bnxt_wait_for_device_shutdown(bp);
4378 /* During some fatal firmware error conditions, the PCI config space
4379 * register 0x2e which normally contains the subsystem ID will become
4380 * 0xffff. This register will revert back to the normal value after
4381 * the chip has completed core reset. If we detect this condition,
4382 * we can poll this config register immediately for the value to revert.
4384 if (bp->flags & BNXT_FLAG_FATAL_ERROR) {
4385 rc = rte_pci_read_config(bp->pdev, &val, sizeof(val), PCI_SUBSYSTEM_ID_OFFSET);
4387 PMD_DRV_LOG(ERR, "Failed to read PCI offset 0x%x", PCI_SUBSYSTEM_ID_OFFSET);
4390 if (val == 0xffff) {
4391 bp->fw_reset_min_msecs = 0;
4396 rc = rte_eal_alarm_set(us, bnxt_dev_recover, (void *)bp);
4398 PMD_DRV_LOG(ERR, "Error setting recovery alarm");
4401 uint32_t bnxt_read_fw_status_reg(struct bnxt *bp, uint32_t index)
4403 struct bnxt_error_recovery_info *info = bp->recovery_info;
4404 uint32_t reg = info->status_regs[index];
4405 uint32_t type, offset, val = 0;
4408 type = BNXT_FW_STATUS_REG_TYPE(reg);
4409 offset = BNXT_FW_STATUS_REG_OFF(reg);
4412 case BNXT_FW_STATUS_REG_TYPE_CFG:
4413 ret = rte_pci_read_config(bp->pdev, &val, sizeof(val), offset);
4415 PMD_DRV_LOG(ERR, "Failed to read PCI offset %#x",
4418 case BNXT_FW_STATUS_REG_TYPE_GRC:
4419 offset = info->mapped_status_regs[index];
4421 case BNXT_FW_STATUS_REG_TYPE_BAR0:
4422 val = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
4430 static int bnxt_fw_reset_all(struct bnxt *bp)
4432 struct bnxt_error_recovery_info *info = bp->recovery_info;
4436 if (info->flags & BNXT_FLAG_ERROR_RECOVERY_HOST) {
4437 /* Reset through primary function driver */
4438 for (i = 0; i < info->reg_array_cnt; i++)
4439 bnxt_write_fw_reset_reg(bp, i);
4440 /* Wait for time specified by FW after triggering reset */
4441 rte_delay_ms(info->primary_func_wait_period_after_reset);
4442 } else if (info->flags & BNXT_FLAG_ERROR_RECOVERY_CO_CPU) {
4443 /* Reset with the help of Kong processor */
4444 rc = bnxt_hwrm_fw_reset(bp);
4446 PMD_DRV_LOG(ERR, "Failed to reset FW\n");
4452 static void bnxt_fw_reset_cb(void *arg)
4454 struct bnxt *bp = arg;
4455 struct bnxt_error_recovery_info *info = bp->recovery_info;
4458 /* Only Primary function can do FW reset */
4459 if (bnxt_is_primary_func(bp) &&
4460 bnxt_is_recovery_enabled(bp)) {
4461 rc = bnxt_fw_reset_all(bp);
4463 PMD_DRV_LOG(ERR, "Adapter recovery failed\n");
4468 /* if recovery method is ERROR_RECOVERY_CO_CPU, KONG will send
4469 * EXCEPTION_FATAL_ASYNC event to all the functions
4470 * (including MASTER FUNC). After receiving this Async, all the active
4471 * drivers should treat this case as FW initiated recovery
4473 if (info->flags & BNXT_FLAG_ERROR_RECOVERY_HOST) {
4474 bp->fw_reset_min_msecs = BNXT_MIN_FW_READY_TIMEOUT;
4475 bp->fw_reset_max_msecs = BNXT_MAX_FW_RESET_TIMEOUT;
4477 /* To recover from error */
4478 rte_eal_alarm_set(US_PER_MS, bnxt_dev_reset_and_resume,
4483 /* Driver should poll FW heartbeat, reset_counter with the frequency
4484 * advertised by FW in HWRM_ERROR_RECOVERY_QCFG.
4485 * When the driver detects heartbeat stop or change in reset_counter,
4486 * it has to trigger a reset to recover from the error condition.
4487 * A “primary function” is the function who will have the privilege to
4488 * initiate the chimp reset. The primary function will be elected by the
4489 * firmware and will be notified through async message.
4491 static void bnxt_check_fw_health(void *arg)
4493 struct bnxt *bp = arg;
4494 struct bnxt_error_recovery_info *info = bp->recovery_info;
4495 uint32_t val = 0, wait_msec;
4497 if (!info || !bnxt_is_recovery_enabled(bp) ||
4498 is_bnxt_in_error(bp))
4501 val = bnxt_read_fw_status_reg(bp, BNXT_FW_HEARTBEAT_CNT_REG);
4502 if (val == info->last_heart_beat)
4505 info->last_heart_beat = val;
4507 val = bnxt_read_fw_status_reg(bp, BNXT_FW_RECOVERY_CNT_REG);
4508 if (val != info->last_reset_counter)
4511 info->last_reset_counter = val;
4513 rte_eal_alarm_set(US_PER_MS * info->driver_polling_freq,
4514 bnxt_check_fw_health, (void *)bp);
4518 /* Stop DMA to/from device */
4519 bp->flags |= BNXT_FLAG_FATAL_ERROR;
4520 bp->flags |= BNXT_FLAG_FW_RESET;
4524 PMD_DRV_LOG(ERR, "Detected FW dead condition\n");
4526 if (bnxt_is_primary_func(bp))
4527 wait_msec = info->primary_func_wait_period;
4529 wait_msec = info->normal_func_wait_period;
4531 rte_eal_alarm_set(US_PER_MS * wait_msec,
4532 bnxt_fw_reset_cb, (void *)bp);
4535 void bnxt_schedule_fw_health_check(struct bnxt *bp)
4537 uint32_t polling_freq;
4539 pthread_mutex_lock(&bp->health_check_lock);
4541 if (!bnxt_is_recovery_enabled(bp))
4544 if (bp->flags & BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED)
4547 polling_freq = bp->recovery_info->driver_polling_freq;
4549 rte_eal_alarm_set(US_PER_MS * polling_freq,
4550 bnxt_check_fw_health, (void *)bp);
4551 bp->flags |= BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED;
4554 pthread_mutex_unlock(&bp->health_check_lock);
4557 static void bnxt_cancel_fw_health_check(struct bnxt *bp)
4559 rte_eal_alarm_cancel(bnxt_check_fw_health, (void *)bp);
4560 bp->flags &= ~BNXT_FLAG_FW_HEALTH_CHECK_SCHEDULED;
4563 static bool bnxt_vf_pciid(uint16_t device_id)
4565 switch (device_id) {
4566 case BROADCOM_DEV_ID_57304_VF:
4567 case BROADCOM_DEV_ID_57406_VF:
4568 case BROADCOM_DEV_ID_5731X_VF:
4569 case BROADCOM_DEV_ID_5741X_VF:
4570 case BROADCOM_DEV_ID_57414_VF:
4571 case BROADCOM_DEV_ID_STRATUS_NIC_VF1:
4572 case BROADCOM_DEV_ID_STRATUS_NIC_VF2:
4573 case BROADCOM_DEV_ID_58802_VF:
4574 case BROADCOM_DEV_ID_57500_VF1:
4575 case BROADCOM_DEV_ID_57500_VF2:
4576 case BROADCOM_DEV_ID_58818_VF:
4584 /* Phase 5 device */
4585 static bool bnxt_p5_device(uint16_t device_id)
4587 switch (device_id) {
4588 case BROADCOM_DEV_ID_57508:
4589 case BROADCOM_DEV_ID_57504:
4590 case BROADCOM_DEV_ID_57502:
4591 case BROADCOM_DEV_ID_57508_MF1:
4592 case BROADCOM_DEV_ID_57504_MF1:
4593 case BROADCOM_DEV_ID_57502_MF1:
4594 case BROADCOM_DEV_ID_57508_MF2:
4595 case BROADCOM_DEV_ID_57504_MF2:
4596 case BROADCOM_DEV_ID_57502_MF2:
4597 case BROADCOM_DEV_ID_57500_VF1:
4598 case BROADCOM_DEV_ID_57500_VF2:
4599 case BROADCOM_DEV_ID_58812:
4600 case BROADCOM_DEV_ID_58814:
4601 case BROADCOM_DEV_ID_58818:
4602 case BROADCOM_DEV_ID_58818_VF:
4610 bool bnxt_stratus_device(struct bnxt *bp)
4612 uint16_t device_id = bp->pdev->id.device_id;
4614 switch (device_id) {
4615 case BROADCOM_DEV_ID_STRATUS_NIC:
4616 case BROADCOM_DEV_ID_STRATUS_NIC_VF1:
4617 case BROADCOM_DEV_ID_STRATUS_NIC_VF2:
4625 static int bnxt_map_pci_bars(struct rte_eth_dev *eth_dev)
4627 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
4628 struct bnxt *bp = eth_dev->data->dev_private;
4630 /* enable device (incl. PCI PM wakeup), and bus-mastering */
4631 bp->bar0 = (void *)pci_dev->mem_resource[0].addr;
4632 bp->doorbell_base = (void *)pci_dev->mem_resource[2].addr;
4633 if (!bp->bar0 || !bp->doorbell_base) {
4634 PMD_DRV_LOG(ERR, "Unable to access Hardware\n");
4638 bp->eth_dev = eth_dev;
4644 static int bnxt_alloc_ctx_mem_blk(struct bnxt *bp,
4645 struct bnxt_ctx_pg_info *ctx_pg,
4650 struct bnxt_ring_mem_info *rmem = &ctx_pg->ring_mem;
4651 const struct rte_memzone *mz = NULL;
4652 char mz_name[RTE_MEMZONE_NAMESIZE];
4653 rte_iova_t mz_phys_addr;
4654 uint64_t valid_bits = 0;
4661 rmem->nr_pages = RTE_ALIGN_MUL_CEIL(mem_size, BNXT_PAGE_SIZE) /
4663 rmem->page_size = BNXT_PAGE_SIZE;
4664 rmem->pg_arr = ctx_pg->ctx_pg_arr;
4665 rmem->dma_arr = ctx_pg->ctx_dma_arr;
4666 rmem->flags = BNXT_RMEM_VALID_PTE_FLAG;
4668 valid_bits = PTU_PTE_VALID;
4670 if (rmem->nr_pages > 1) {
4671 snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4672 "bnxt_ctx_pg_tbl%s_%x_%d",
4673 suffix, idx, bp->eth_dev->data->port_id);
4674 mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4675 mz = rte_memzone_lookup(mz_name);
4677 mz = rte_memzone_reserve_aligned(mz_name,
4679 bp->eth_dev->device->numa_node,
4681 RTE_MEMZONE_SIZE_HINT_ONLY |
4682 RTE_MEMZONE_IOVA_CONTIG,
4688 memset(mz->addr, 0, mz->len);
4689 mz_phys_addr = mz->iova;
4691 rmem->pg_tbl = mz->addr;
4692 rmem->pg_tbl_map = mz_phys_addr;
4693 rmem->pg_tbl_mz = mz;
4696 snprintf(mz_name, RTE_MEMZONE_NAMESIZE, "bnxt_ctx_%s_%x_%d",
4697 suffix, idx, bp->eth_dev->data->port_id);
4698 mz = rte_memzone_lookup(mz_name);
4700 mz = rte_memzone_reserve_aligned(mz_name,
4702 bp->eth_dev->device->numa_node,
4704 RTE_MEMZONE_SIZE_HINT_ONLY |
4705 RTE_MEMZONE_IOVA_CONTIG,
4711 memset(mz->addr, 0, mz->len);
4712 mz_phys_addr = mz->iova;
4714 for (sz = 0, i = 0; sz < mem_size; sz += BNXT_PAGE_SIZE, i++) {
4715 rmem->pg_arr[i] = ((char *)mz->addr) + sz;
4716 rmem->dma_arr[i] = mz_phys_addr + sz;
4718 if (rmem->nr_pages > 1) {
4719 if (i == rmem->nr_pages - 2 &&
4720 (rmem->flags & BNXT_RMEM_RING_PTE_FLAG))
4721 valid_bits |= PTU_PTE_NEXT_TO_LAST;
4722 else if (i == rmem->nr_pages - 1 &&
4723 (rmem->flags & BNXT_RMEM_RING_PTE_FLAG))
4724 valid_bits |= PTU_PTE_LAST;
4726 rmem->pg_tbl[i] = rte_cpu_to_le_64(rmem->dma_arr[i] |
4732 if (rmem->vmem_size)
4733 rmem->vmem = (void **)mz->addr;
4734 rmem->dma_arr[0] = mz_phys_addr;
4738 static void bnxt_free_ctx_mem(struct bnxt *bp)
4742 if (!bp->ctx || !(bp->ctx->flags & BNXT_CTX_FLAG_INITED))
4745 bp->ctx->flags &= ~BNXT_CTX_FLAG_INITED;
4746 rte_memzone_free(bp->ctx->qp_mem.ring_mem.mz);
4747 rte_memzone_free(bp->ctx->srq_mem.ring_mem.mz);
4748 rte_memzone_free(bp->ctx->cq_mem.ring_mem.mz);
4749 rte_memzone_free(bp->ctx->vnic_mem.ring_mem.mz);
4750 rte_memzone_free(bp->ctx->stat_mem.ring_mem.mz);
4751 rte_memzone_free(bp->ctx->qp_mem.ring_mem.pg_tbl_mz);
4752 rte_memzone_free(bp->ctx->srq_mem.ring_mem.pg_tbl_mz);
4753 rte_memzone_free(bp->ctx->cq_mem.ring_mem.pg_tbl_mz);
4754 rte_memzone_free(bp->ctx->vnic_mem.ring_mem.pg_tbl_mz);
4755 rte_memzone_free(bp->ctx->stat_mem.ring_mem.pg_tbl_mz);
4757 for (i = 0; i < bp->ctx->tqm_fp_rings_count + 1; i++) {
4758 if (bp->ctx->tqm_mem[i])
4759 rte_memzone_free(bp->ctx->tqm_mem[i]->ring_mem.mz);
4766 #define bnxt_roundup(x, y) ((((x) + ((y) - 1)) / (y)) * (y))
4768 #define min_t(type, x, y) ({ \
4769 type __min1 = (x); \
4770 type __min2 = (y); \
4771 __min1 < __min2 ? __min1 : __min2; })
4773 #define max_t(type, x, y) ({ \
4774 type __max1 = (x); \
4775 type __max2 = (y); \
4776 __max1 > __max2 ? __max1 : __max2; })
4778 #define clamp_t(type, _x, min, max) min_t(type, max_t(type, _x, min), max)
4780 int bnxt_alloc_ctx_mem(struct bnxt *bp)
4782 struct bnxt_ctx_pg_info *ctx_pg;
4783 struct bnxt_ctx_mem_info *ctx;
4784 uint32_t mem_size, ena, entries;
4785 uint32_t entries_sp, min;
4788 rc = bnxt_hwrm_func_backing_store_qcaps(bp);
4790 PMD_DRV_LOG(ERR, "Query context mem capability failed\n");
4794 if (!ctx || (ctx->flags & BNXT_CTX_FLAG_INITED))
4797 ctx_pg = &ctx->qp_mem;
4798 ctx_pg->entries = ctx->qp_min_qp1_entries + ctx->qp_max_l2_entries;
4799 if (ctx->qp_entry_size) {
4800 mem_size = ctx->qp_entry_size * ctx_pg->entries;
4801 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "qp_mem", 0);
4806 ctx_pg = &ctx->srq_mem;
4807 ctx_pg->entries = ctx->srq_max_l2_entries;
4808 if (ctx->srq_entry_size) {
4809 mem_size = ctx->srq_entry_size * ctx_pg->entries;
4810 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "srq_mem", 0);
4815 ctx_pg = &ctx->cq_mem;
4816 ctx_pg->entries = ctx->cq_max_l2_entries;
4817 if (ctx->cq_entry_size) {
4818 mem_size = ctx->cq_entry_size * ctx_pg->entries;
4819 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "cq_mem", 0);
4824 ctx_pg = &ctx->vnic_mem;
4825 ctx_pg->entries = ctx->vnic_max_vnic_entries +
4826 ctx->vnic_max_ring_table_entries;
4827 if (ctx->vnic_entry_size) {
4828 mem_size = ctx->vnic_entry_size * ctx_pg->entries;
4829 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "vnic_mem", 0);
4834 ctx_pg = &ctx->stat_mem;
4835 ctx_pg->entries = ctx->stat_max_entries;
4836 if (ctx->stat_entry_size) {
4837 mem_size = ctx->stat_entry_size * ctx_pg->entries;
4838 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size, "stat_mem", 0);
4843 min = ctx->tqm_min_entries_per_ring;
4845 entries_sp = ctx->qp_max_l2_entries +
4846 ctx->vnic_max_vnic_entries +
4847 2 * ctx->qp_min_qp1_entries + min;
4848 entries_sp = bnxt_roundup(entries_sp, ctx->tqm_entries_multiple);
4850 entries = ctx->qp_max_l2_entries + ctx->qp_min_qp1_entries;
4851 entries = bnxt_roundup(entries, ctx->tqm_entries_multiple);
4852 entries = clamp_t(uint32_t, entries, min,
4853 ctx->tqm_max_entries_per_ring);
4854 for (i = 0, ena = 0; i < ctx->tqm_fp_rings_count + 1; i++) {
4855 /* i=0 is for TQM_SP. i=1 to i=8 applies to RING0 to RING7.
4856 * i > 8 is other ext rings.
4858 ctx_pg = ctx->tqm_mem[i];
4859 ctx_pg->entries = i ? entries : entries_sp;
4860 if (ctx->tqm_entry_size) {
4861 mem_size = ctx->tqm_entry_size * ctx_pg->entries;
4862 rc = bnxt_alloc_ctx_mem_blk(bp, ctx_pg, mem_size,
4867 if (i < BNXT_MAX_TQM_LEGACY_RINGS)
4868 ena |= HWRM_FUNC_BACKING_STORE_CFG_INPUT_ENABLES_TQM_SP << i;
4870 ena |= HWRM_FUNC_BACKING_STORE_CFG_INPUT_ENABLES_TQM_RING8;
4873 ena |= FUNC_BACKING_STORE_CFG_INPUT_DFLT_ENABLES;
4874 rc = bnxt_hwrm_func_backing_store_cfg(bp, ena);
4877 "Failed to configure context mem: rc = %d\n", rc);
4879 ctx->flags |= BNXT_CTX_FLAG_INITED;
4884 static int bnxt_alloc_stats_mem(struct bnxt *bp)
4886 struct rte_pci_device *pci_dev = bp->pdev;
4887 char mz_name[RTE_MEMZONE_NAMESIZE];
4888 const struct rte_memzone *mz = NULL;
4889 uint32_t total_alloc_len;
4890 rte_iova_t mz_phys_addr;
4892 if (pci_dev->id.device_id == BROADCOM_DEV_ID_NS2)
4895 snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4896 "bnxt_" PCI_PRI_FMT "-%s", pci_dev->addr.domain,
4897 pci_dev->addr.bus, pci_dev->addr.devid,
4898 pci_dev->addr.function, "rx_port_stats");
4899 mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4900 mz = rte_memzone_lookup(mz_name);
4902 RTE_CACHE_LINE_ROUNDUP(sizeof(struct rx_port_stats) +
4903 sizeof(struct rx_port_stats_ext) + 512);
4905 mz = rte_memzone_reserve(mz_name, total_alloc_len,
4908 RTE_MEMZONE_SIZE_HINT_ONLY |
4909 RTE_MEMZONE_IOVA_CONTIG);
4913 memset(mz->addr, 0, mz->len);
4914 mz_phys_addr = mz->iova;
4916 bp->rx_mem_zone = (const void *)mz;
4917 bp->hw_rx_port_stats = mz->addr;
4918 bp->hw_rx_port_stats_map = mz_phys_addr;
4920 snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
4921 "bnxt_" PCI_PRI_FMT "-%s", pci_dev->addr.domain,
4922 pci_dev->addr.bus, pci_dev->addr.devid,
4923 pci_dev->addr.function, "tx_port_stats");
4924 mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
4925 mz = rte_memzone_lookup(mz_name);
4927 RTE_CACHE_LINE_ROUNDUP(sizeof(struct tx_port_stats) +
4928 sizeof(struct tx_port_stats_ext) + 512);
4930 mz = rte_memzone_reserve(mz_name,
4934 RTE_MEMZONE_SIZE_HINT_ONLY |
4935 RTE_MEMZONE_IOVA_CONTIG);
4939 memset(mz->addr, 0, mz->len);
4940 mz_phys_addr = mz->iova;
4942 bp->tx_mem_zone = (const void *)mz;
4943 bp->hw_tx_port_stats = mz->addr;
4944 bp->hw_tx_port_stats_map = mz_phys_addr;
4945 bp->flags |= BNXT_FLAG_PORT_STATS;
4947 /* Display extended statistics if FW supports it */
4948 if (bp->hwrm_spec_code < HWRM_SPEC_CODE_1_8_4 ||
4949 bp->hwrm_spec_code == HWRM_SPEC_CODE_1_9_0 ||
4950 !(bp->flags & BNXT_FLAG_EXT_STATS_SUPPORTED))
4953 bp->hw_rx_port_stats_ext = (void *)
4954 ((uint8_t *)bp->hw_rx_port_stats +
4955 sizeof(struct rx_port_stats));
4956 bp->hw_rx_port_stats_ext_map = bp->hw_rx_port_stats_map +
4957 sizeof(struct rx_port_stats);
4958 bp->flags |= BNXT_FLAG_EXT_RX_PORT_STATS;
4960 if (bp->hwrm_spec_code < HWRM_SPEC_CODE_1_9_2 ||
4961 bp->flags & BNXT_FLAG_EXT_STATS_SUPPORTED) {
4962 bp->hw_tx_port_stats_ext = (void *)
4963 ((uint8_t *)bp->hw_tx_port_stats +
4964 sizeof(struct tx_port_stats));
4965 bp->hw_tx_port_stats_ext_map =
4966 bp->hw_tx_port_stats_map +
4967 sizeof(struct tx_port_stats);
4968 bp->flags |= BNXT_FLAG_EXT_TX_PORT_STATS;
4974 static int bnxt_setup_mac_addr(struct rte_eth_dev *eth_dev)
4976 struct bnxt *bp = eth_dev->data->dev_private;
4979 eth_dev->data->mac_addrs = rte_zmalloc("bnxt_mac_addr_tbl",
4980 RTE_ETHER_ADDR_LEN *
4983 if (eth_dev->data->mac_addrs == NULL) {
4984 PMD_DRV_LOG(ERR, "Failed to alloc MAC addr tbl\n");
4988 if (!BNXT_HAS_DFLT_MAC_SET(bp)) {
4992 /* Generate a random MAC address, if none was assigned by PF */
4993 PMD_DRV_LOG(INFO, "VF MAC address not assigned by Host PF\n");
4994 bnxt_eth_hw_addr_random(bp->mac_addr);
4996 "Assign random MAC:" RTE_ETHER_ADDR_PRT_FMT "\n",
4997 bp->mac_addr[0], bp->mac_addr[1], bp->mac_addr[2],
4998 bp->mac_addr[3], bp->mac_addr[4], bp->mac_addr[5]);
5000 rc = bnxt_hwrm_set_mac(bp);
5005 /* Copy the permanent MAC from the FUNC_QCAPS response */
5006 memcpy(ð_dev->data->mac_addrs[0], bp->mac_addr, RTE_ETHER_ADDR_LEN);
5011 static int bnxt_restore_dflt_mac(struct bnxt *bp)
5015 /* MAC is already configured in FW */
5016 if (BNXT_HAS_DFLT_MAC_SET(bp))
5019 /* Restore the old MAC configured */
5020 rc = bnxt_hwrm_set_mac(bp);
5022 PMD_DRV_LOG(ERR, "Failed to restore MAC address\n");
5027 static void bnxt_config_vf_req_fwd(struct bnxt *bp)
5032 memset(bp->pf->vf_req_fwd, 0, sizeof(bp->pf->vf_req_fwd));
5034 if (!(bp->fw_cap & BNXT_FW_CAP_LINK_ADMIN))
5035 BNXT_HWRM_CMD_TO_FORWARD(HWRM_PORT_PHY_QCFG);
5036 BNXT_HWRM_CMD_TO_FORWARD(HWRM_FUNC_CFG);
5037 BNXT_HWRM_CMD_TO_FORWARD(HWRM_FUNC_VF_CFG);
5038 BNXT_HWRM_CMD_TO_FORWARD(HWRM_CFA_L2_FILTER_ALLOC);
5039 BNXT_HWRM_CMD_TO_FORWARD(HWRM_OEM_CMD);
5042 static void bnxt_alloc_error_recovery_info(struct bnxt *bp)
5044 struct bnxt_error_recovery_info *info = bp->recovery_info;
5047 if (!(bp->fw_cap & BNXT_FW_CAP_HCOMM_FW_STATUS))
5048 memset(info, 0, sizeof(*info));
5052 if (!(bp->fw_cap & BNXT_FW_CAP_ERROR_RECOVERY))
5055 info = rte_zmalloc("bnxt_hwrm_error_recovery_qcfg",
5058 bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5060 bp->recovery_info = info;
5063 static void bnxt_check_fw_status(struct bnxt *bp)
5067 if (!(bp->recovery_info &&
5068 (bp->fw_cap & BNXT_FW_CAP_HCOMM_FW_STATUS)))
5071 fw_status = bnxt_read_fw_status_reg(bp, BNXT_FW_STATUS_REG);
5072 if (fw_status != BNXT_FW_STATUS_HEALTHY)
5073 PMD_DRV_LOG(ERR, "Firmware not responding, status: %#x\n",
5077 static int bnxt_map_hcomm_fw_status_reg(struct bnxt *bp)
5079 struct bnxt_error_recovery_info *info = bp->recovery_info;
5080 uint32_t status_loc;
5083 rte_write32(HCOMM_STATUS_STRUCT_LOC, (uint8_t *)bp->bar0 +
5084 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
5085 sig_ver = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
5086 BNXT_GRCP_WINDOW_2_BASE +
5087 offsetof(struct hcomm_status,
5089 /* If the signature is absent, then FW does not support this feature */
5090 if ((sig_ver & HCOMM_STATUS_SIGNATURE_MASK) !=
5091 HCOMM_STATUS_SIGNATURE_VAL)
5095 info = rte_zmalloc("bnxt_hwrm_error_recovery_qcfg",
5099 bp->recovery_info = info;
5101 memset(info, 0, sizeof(*info));
5104 status_loc = rte_le_to_cpu_32(rte_read32((uint8_t *)bp->bar0 +
5105 BNXT_GRCP_WINDOW_2_BASE +
5106 offsetof(struct hcomm_status,
5109 /* Only pre-map the FW health status GRC register */
5110 if (BNXT_FW_STATUS_REG_TYPE(status_loc) != BNXT_FW_STATUS_REG_TYPE_GRC)
5113 info->status_regs[BNXT_FW_STATUS_REG] = status_loc;
5114 info->mapped_status_regs[BNXT_FW_STATUS_REG] =
5115 BNXT_GRCP_WINDOW_2_BASE + (status_loc & BNXT_GRCP_OFFSET_MASK);
5117 rte_write32((status_loc & BNXT_GRCP_BASE_MASK), (uint8_t *)bp->bar0 +
5118 BNXT_GRCPF_REG_WINDOW_BASE_OUT + 4);
5120 bp->fw_cap |= BNXT_FW_CAP_HCOMM_FW_STATUS;
5125 /* This function gets the FW version along with the
5126 * capabilities(MAX and current) of the function, vnic,
5127 * error recovery, phy and other chip related info
5129 static int bnxt_get_config(struct bnxt *bp)
5136 rc = bnxt_map_hcomm_fw_status_reg(bp);
5140 rc = bnxt_hwrm_ver_get(bp, DFLT_HWRM_CMD_TIMEOUT);
5142 bnxt_check_fw_status(bp);
5146 rc = bnxt_hwrm_func_reset(bp);
5150 rc = bnxt_hwrm_vnic_qcaps(bp);
5154 rc = bnxt_hwrm_queue_qportcfg(bp);
5158 /* Get the MAX capabilities for this function.
5159 * This function also allocates context memory for TQM rings and
5160 * informs the firmware about this allocated backing store memory.
5162 rc = bnxt_hwrm_func_qcaps(bp);
5166 rc = bnxt_hwrm_func_qcfg(bp, &mtu);
5170 rc = bnxt_hwrm_cfa_adv_flow_mgmt_qcaps(bp);
5174 bnxt_hwrm_port_mac_qcfg(bp);
5176 bnxt_hwrm_parent_pf_qcfg(bp);
5178 bnxt_hwrm_port_phy_qcaps(bp);
5180 bnxt_alloc_error_recovery_info(bp);
5181 /* Get the adapter error recovery support info */
5182 rc = bnxt_hwrm_error_recovery_qcfg(bp);
5184 bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5186 bnxt_hwrm_port_led_qcaps(bp);
5192 bnxt_init_locks(struct bnxt *bp)
5196 err = pthread_mutex_init(&bp->flow_lock, NULL);
5198 PMD_DRV_LOG(ERR, "Unable to initialize flow_lock\n");
5202 err = pthread_mutex_init(&bp->def_cp_lock, NULL);
5204 PMD_DRV_LOG(ERR, "Unable to initialize def_cp_lock\n");
5208 err = pthread_mutex_init(&bp->health_check_lock, NULL);
5210 PMD_DRV_LOG(ERR, "Unable to initialize health_check_lock\n");
5214 err = pthread_mutex_init(&bp->err_recovery_lock, NULL);
5216 PMD_DRV_LOG(ERR, "Unable to initialize err_recovery_lock\n");
5221 static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev)
5225 rc = bnxt_get_config(bp);
5229 if (!reconfig_dev) {
5230 rc = bnxt_setup_mac_addr(bp->eth_dev);
5234 rc = bnxt_restore_dflt_mac(bp);
5239 bnxt_config_vf_req_fwd(bp);
5241 rc = bnxt_hwrm_func_driver_register(bp);
5243 PMD_DRV_LOG(ERR, "Failed to register driver");
5248 if (bp->pdev->max_vfs) {
5249 rc = bnxt_hwrm_allocate_vfs(bp, bp->pdev->max_vfs);
5251 PMD_DRV_LOG(ERR, "Failed to allocate VFs\n");
5255 rc = bnxt_hwrm_allocate_pf_only(bp);
5258 "Failed to allocate PF resources");
5264 rc = bnxt_alloc_mem(bp, reconfig_dev);
5268 rc = bnxt_setup_int(bp);
5272 rc = bnxt_request_int(bp);
5276 rc = bnxt_init_ctx_mem(bp);
5278 PMD_DRV_LOG(ERR, "Failed to init adv_flow_counters\n");
5286 bnxt_parse_devarg_flow_xstat(__rte_unused const char *key,
5287 const char *value, void *opaque_arg)
5289 struct bnxt *bp = opaque_arg;
5290 unsigned long flow_xstat;
5293 if (!value || !opaque_arg) {
5295 "Invalid parameter passed to flow_xstat devarg.\n");
5299 flow_xstat = strtoul(value, &end, 10);
5300 if (end == NULL || *end != '\0' ||
5301 (flow_xstat == ULONG_MAX && errno == ERANGE)) {
5303 "Invalid parameter passed to flow_xstat devarg.\n");
5307 if (BNXT_DEVARG_FLOW_XSTAT_INVALID(flow_xstat)) {
5309 "Invalid value passed to flow_xstat devarg.\n");
5313 bp->flags |= BNXT_FLAG_FLOW_XSTATS_EN;
5314 if (BNXT_FLOW_XSTATS_EN(bp))
5315 PMD_DRV_LOG(INFO, "flow_xstat feature enabled.\n");
5321 bnxt_parse_devarg_max_num_kflows(__rte_unused const char *key,
5322 const char *value, void *opaque_arg)
5324 struct bnxt *bp = opaque_arg;
5325 unsigned long max_num_kflows;
5328 if (!value || !opaque_arg) {
5330 "Invalid parameter passed to max_num_kflows devarg.\n");
5334 max_num_kflows = strtoul(value, &end, 10);
5335 if (end == NULL || *end != '\0' ||
5336 (max_num_kflows == ULONG_MAX && errno == ERANGE)) {
5338 "Invalid parameter passed to max_num_kflows devarg.\n");
5342 if (bnxt_devarg_max_num_kflow_invalid(max_num_kflows)) {
5344 "Invalid value passed to max_num_kflows devarg.\n");
5348 bp->max_num_kflows = max_num_kflows;
5349 if (bp->max_num_kflows)
5350 PMD_DRV_LOG(INFO, "max_num_kflows set as %ldK.\n",
5357 bnxt_parse_devarg_app_id(__rte_unused const char *key,
5358 const char *value, void *opaque_arg)
5360 struct bnxt *bp = opaque_arg;
5361 unsigned long app_id;
5364 if (!value || !opaque_arg) {
5366 "Invalid parameter passed to app-id "
5371 app_id = strtoul(value, &end, 10);
5372 if (end == NULL || *end != '\0' ||
5373 (app_id == ULONG_MAX && errno == ERANGE)) {
5375 "Invalid parameter passed to app_id "
5380 if (BNXT_DEVARG_APP_ID_INVALID(app_id)) {
5381 PMD_DRV_LOG(ERR, "Invalid app-id(%d) devargs.\n",
5386 bp->app_id = app_id;
5387 PMD_DRV_LOG(INFO, "app-id=%d feature enabled.\n", (uint16_t)app_id);
5393 bnxt_parse_devarg_rep_is_pf(__rte_unused const char *key,
5394 const char *value, void *opaque_arg)
5396 struct bnxt_representor *vfr_bp = opaque_arg;
5397 unsigned long rep_is_pf;
5400 if (!value || !opaque_arg) {
5402 "Invalid parameter passed to rep_is_pf devargs.\n");
5406 rep_is_pf = strtoul(value, &end, 10);
5407 if (end == NULL || *end != '\0' ||
5408 (rep_is_pf == ULONG_MAX && errno == ERANGE)) {
5410 "Invalid parameter passed to rep_is_pf devargs.\n");
5414 if (BNXT_DEVARG_REP_IS_PF_INVALID(rep_is_pf)) {
5416 "Invalid value passed to rep_is_pf devargs.\n");
5420 vfr_bp->flags |= rep_is_pf;
5421 if (BNXT_REP_PF(vfr_bp))
5422 PMD_DRV_LOG(INFO, "PF representor\n");
5424 PMD_DRV_LOG(INFO, "VF representor\n");
5430 bnxt_parse_devarg_rep_based_pf(__rte_unused const char *key,
5431 const char *value, void *opaque_arg)
5433 struct bnxt_representor *vfr_bp = opaque_arg;
5434 unsigned long rep_based_pf;
5437 if (!value || !opaque_arg) {
5439 "Invalid parameter passed to rep_based_pf "
5444 rep_based_pf = strtoul(value, &end, 10);
5445 if (end == NULL || *end != '\0' ||
5446 (rep_based_pf == ULONG_MAX && errno == ERANGE)) {
5448 "Invalid parameter passed to rep_based_pf "
5453 if (BNXT_DEVARG_REP_BASED_PF_INVALID(rep_based_pf)) {
5455 "Invalid value passed to rep_based_pf devargs.\n");
5459 vfr_bp->rep_based_pf = rep_based_pf;
5460 vfr_bp->flags |= BNXT_REP_BASED_PF_VALID;
5462 PMD_DRV_LOG(INFO, "rep-based-pf = %d\n", vfr_bp->rep_based_pf);
5468 bnxt_parse_devarg_rep_q_r2f(__rte_unused const char *key,
5469 const char *value, void *opaque_arg)
5471 struct bnxt_representor *vfr_bp = opaque_arg;
5472 unsigned long rep_q_r2f;
5475 if (!value || !opaque_arg) {
5477 "Invalid parameter passed to rep_q_r2f "
5482 rep_q_r2f = strtoul(value, &end, 10);
5483 if (end == NULL || *end != '\0' ||
5484 (rep_q_r2f == ULONG_MAX && errno == ERANGE)) {
5486 "Invalid parameter passed to rep_q_r2f "
5491 if (BNXT_DEVARG_REP_Q_R2F_INVALID(rep_q_r2f)) {
5493 "Invalid value passed to rep_q_r2f devargs.\n");
5497 vfr_bp->rep_q_r2f = rep_q_r2f;
5498 vfr_bp->flags |= BNXT_REP_Q_R2F_VALID;
5499 PMD_DRV_LOG(INFO, "rep-q-r2f = %d\n", vfr_bp->rep_q_r2f);
5505 bnxt_parse_devarg_rep_q_f2r(__rte_unused const char *key,
5506 const char *value, void *opaque_arg)
5508 struct bnxt_representor *vfr_bp = opaque_arg;
5509 unsigned long rep_q_f2r;
5512 if (!value || !opaque_arg) {
5514 "Invalid parameter passed to rep_q_f2r "
5519 rep_q_f2r = strtoul(value, &end, 10);
5520 if (end == NULL || *end != '\0' ||
5521 (rep_q_f2r == ULONG_MAX && errno == ERANGE)) {
5523 "Invalid parameter passed to rep_q_f2r "
5528 if (BNXT_DEVARG_REP_Q_F2R_INVALID(rep_q_f2r)) {
5530 "Invalid value passed to rep_q_f2r devargs.\n");
5534 vfr_bp->rep_q_f2r = rep_q_f2r;
5535 vfr_bp->flags |= BNXT_REP_Q_F2R_VALID;
5536 PMD_DRV_LOG(INFO, "rep-q-f2r = %d\n", vfr_bp->rep_q_f2r);
5542 bnxt_parse_devarg_rep_fc_r2f(__rte_unused const char *key,
5543 const char *value, void *opaque_arg)
5545 struct bnxt_representor *vfr_bp = opaque_arg;
5546 unsigned long rep_fc_r2f;
5549 if (!value || !opaque_arg) {
5551 "Invalid parameter passed to rep_fc_r2f "
5556 rep_fc_r2f = strtoul(value, &end, 10);
5557 if (end == NULL || *end != '\0' ||
5558 (rep_fc_r2f == ULONG_MAX && errno == ERANGE)) {
5560 "Invalid parameter passed to rep_fc_r2f "
5565 if (BNXT_DEVARG_REP_FC_R2F_INVALID(rep_fc_r2f)) {
5567 "Invalid value passed to rep_fc_r2f devargs.\n");
5571 vfr_bp->flags |= BNXT_REP_FC_R2F_VALID;
5572 vfr_bp->rep_fc_r2f = rep_fc_r2f;
5573 PMD_DRV_LOG(INFO, "rep-fc-r2f = %lu\n", rep_fc_r2f);
5579 bnxt_parse_devarg_rep_fc_f2r(__rte_unused const char *key,
5580 const char *value, void *opaque_arg)
5582 struct bnxt_representor *vfr_bp = opaque_arg;
5583 unsigned long rep_fc_f2r;
5586 if (!value || !opaque_arg) {
5588 "Invalid parameter passed to rep_fc_f2r "
5593 rep_fc_f2r = strtoul(value, &end, 10);
5594 if (end == NULL || *end != '\0' ||
5595 (rep_fc_f2r == ULONG_MAX && errno == ERANGE)) {
5597 "Invalid parameter passed to rep_fc_f2r "
5602 if (BNXT_DEVARG_REP_FC_F2R_INVALID(rep_fc_f2r)) {
5604 "Invalid value passed to rep_fc_f2r devargs.\n");
5608 vfr_bp->flags |= BNXT_REP_FC_F2R_VALID;
5609 vfr_bp->rep_fc_f2r = rep_fc_f2r;
5610 PMD_DRV_LOG(INFO, "rep-fc-f2r = %lu\n", rep_fc_f2r);
5616 bnxt_parse_dev_args(struct bnxt *bp, struct rte_devargs *devargs)
5618 struct rte_kvargs *kvlist;
5621 if (devargs == NULL)
5624 kvlist = rte_kvargs_parse(devargs->args, bnxt_dev_args);
5629 * Handler for "flow_xstat" devarg.
5630 * Invoked as for ex: "-a 0000:00:0d.0,flow_xstat=1"
5632 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_FLOW_XSTAT,
5633 bnxt_parse_devarg_flow_xstat, bp);
5638 * Handler for "max_num_kflows" devarg.
5639 * Invoked as for ex: "-a 000:00:0d.0,max_num_kflows=32"
5641 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_MAX_NUM_KFLOWS,
5642 bnxt_parse_devarg_max_num_kflows, bp);
5648 * Handler for "app-id" devarg.
5649 * Invoked as for ex: "-a 000:00:0d.0,app-id=1"
5651 rte_kvargs_process(kvlist, BNXT_DEVARG_APP_ID,
5652 bnxt_parse_devarg_app_id, bp);
5654 rte_kvargs_free(kvlist);
5658 static int bnxt_alloc_switch_domain(struct bnxt *bp)
5662 if (BNXT_PF(bp) || BNXT_VF_IS_TRUSTED(bp)) {
5663 rc = rte_eth_switch_domain_alloc(&bp->switch_domain_id);
5666 "Failed to alloc switch domain: %d\n", rc);
5669 "Switch domain allocated %d\n",
5670 bp->switch_domain_id);
5676 /* Allocate and initialize various fields in bnxt struct that
5677 * need to be allocated/destroyed only once in the lifetime of the driver
5679 static int bnxt_drv_init(struct rte_eth_dev *eth_dev)
5681 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
5682 struct bnxt *bp = eth_dev->data->dev_private;
5685 bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
5687 if (bnxt_vf_pciid(pci_dev->id.device_id))
5688 bp->flags |= BNXT_FLAG_VF;
5690 if (bnxt_p5_device(pci_dev->id.device_id))
5691 bp->flags |= BNXT_FLAG_CHIP_P5;
5693 if (pci_dev->id.device_id == BROADCOM_DEV_ID_58802 ||
5694 pci_dev->id.device_id == BROADCOM_DEV_ID_58804 ||
5695 pci_dev->id.device_id == BROADCOM_DEV_ID_58808 ||
5696 pci_dev->id.device_id == BROADCOM_DEV_ID_58802_VF)
5697 bp->flags |= BNXT_FLAG_STINGRAY;
5699 if (BNXT_TRUFLOW_EN(bp)) {
5700 /* extra mbuf field is required to store CFA code from mark */
5701 static const struct rte_mbuf_dynfield bnxt_cfa_code_dynfield_desc = {
5702 .name = RTE_PMD_BNXT_CFA_CODE_DYNFIELD_NAME,
5703 .size = sizeof(bnxt_cfa_code_dynfield_t),
5704 .align = __alignof__(bnxt_cfa_code_dynfield_t),
5706 bnxt_cfa_code_dynfield_offset =
5707 rte_mbuf_dynfield_register(&bnxt_cfa_code_dynfield_desc);
5708 if (bnxt_cfa_code_dynfield_offset < 0) {
5710 "Failed to register mbuf field for TruFlow mark\n");
5715 rc = bnxt_map_pci_bars(eth_dev);
5718 "Failed to initialize board rc: %x\n", rc);
5722 rc = bnxt_alloc_pf_info(bp);
5726 rc = bnxt_alloc_link_info(bp);
5730 rc = bnxt_alloc_parent_info(bp);
5734 rc = bnxt_alloc_hwrm_resources(bp);
5737 "Failed to allocate response buffer rc: %x\n", rc);
5740 rc = bnxt_alloc_leds_info(bp);
5744 rc = bnxt_alloc_cos_queues(bp);
5748 rc = bnxt_init_locks(bp);
5752 rc = bnxt_alloc_switch_domain(bp);
5760 bnxt_dev_init(struct rte_eth_dev *eth_dev, void *params __rte_unused)
5762 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
5763 static int version_printed;
5767 if (version_printed++ == 0)
5768 PMD_DRV_LOG(INFO, "%s\n", bnxt_version);
5770 eth_dev->dev_ops = &bnxt_dev_ops;
5771 eth_dev->rx_queue_count = bnxt_rx_queue_count_op;
5772 eth_dev->rx_descriptor_status = bnxt_rx_descriptor_status_op;
5773 eth_dev->tx_descriptor_status = bnxt_tx_descriptor_status_op;
5774 eth_dev->rx_pkt_burst = &bnxt_recv_pkts;
5775 eth_dev->tx_pkt_burst = &bnxt_xmit_pkts;
5778 * For secondary processes, we don't initialise any further
5779 * as primary has already done this work.
5781 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
5784 rte_eth_copy_pci_info(eth_dev, pci_dev);
5785 eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
5787 bp = eth_dev->data->dev_private;
5789 /* Parse dev arguments passed on when starting the DPDK application. */
5790 rc = bnxt_parse_dev_args(bp, pci_dev->device.devargs);
5794 rc = bnxt_drv_init(eth_dev);
5798 rc = bnxt_init_resources(bp, false);
5802 rc = bnxt_alloc_stats_mem(bp);
5807 "Found %s device at mem %" PRIX64 ", node addr %pM\n",
5809 pci_dev->mem_resource[0].phys_addr,
5810 pci_dev->mem_resource[0].addr);
5815 bnxt_dev_uninit(eth_dev);
5820 static void bnxt_free_ctx_mem_buf(struct bnxt_ctx_mem_buf_info *ctx)
5829 ctx->dma = RTE_BAD_IOVA;
5830 ctx->ctx_id = BNXT_CTX_VAL_INVAL;
5833 static void bnxt_unregister_fc_ctx_mem(struct bnxt *bp)
5835 bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_RX,
5836 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
5837 bp->flow_stat->rx_fc_out_tbl.ctx_id,
5838 bp->flow_stat->max_fc,
5841 bnxt_hwrm_cfa_counter_cfg(bp, BNXT_DIR_TX,
5842 CFA_COUNTER_CFG_IN_COUNTER_TYPE_FC,
5843 bp->flow_stat->tx_fc_out_tbl.ctx_id,
5844 bp->flow_stat->max_fc,
5847 if (bp->flow_stat->rx_fc_in_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5848 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->rx_fc_in_tbl.ctx_id);
5849 bp->flow_stat->rx_fc_in_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5851 if (bp->flow_stat->rx_fc_out_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5852 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->rx_fc_out_tbl.ctx_id);
5853 bp->flow_stat->rx_fc_out_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5855 if (bp->flow_stat->tx_fc_in_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5856 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->tx_fc_in_tbl.ctx_id);
5857 bp->flow_stat->tx_fc_in_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5859 if (bp->flow_stat->tx_fc_out_tbl.ctx_id != BNXT_CTX_VAL_INVAL)
5860 bnxt_hwrm_ctx_unrgtr(bp, bp->flow_stat->tx_fc_out_tbl.ctx_id);
5861 bp->flow_stat->tx_fc_out_tbl.ctx_id = BNXT_CTX_VAL_INVAL;
5864 static void bnxt_uninit_fc_ctx_mem(struct bnxt *bp)
5866 bnxt_unregister_fc_ctx_mem(bp);
5868 bnxt_free_ctx_mem_buf(&bp->flow_stat->rx_fc_in_tbl);
5869 bnxt_free_ctx_mem_buf(&bp->flow_stat->rx_fc_out_tbl);
5870 bnxt_free_ctx_mem_buf(&bp->flow_stat->tx_fc_in_tbl);
5871 bnxt_free_ctx_mem_buf(&bp->flow_stat->tx_fc_out_tbl);
5874 static void bnxt_uninit_ctx_mem(struct bnxt *bp)
5876 if (BNXT_FLOW_XSTATS_EN(bp))
5877 bnxt_uninit_fc_ctx_mem(bp);
5881 bnxt_free_error_recovery_info(struct bnxt *bp)
5883 rte_free(bp->recovery_info);
5884 bp->recovery_info = NULL;
5885 bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY;
5889 bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev)
5894 bnxt_free_mem(bp, reconfig_dev);
5896 bnxt_hwrm_func_buf_unrgtr(bp);
5897 if (bp->pf != NULL) {
5898 rte_free(bp->pf->vf_req_buf);
5899 bp->pf->vf_req_buf = NULL;
5902 rc = bnxt_hwrm_func_driver_unregister(bp);
5903 bp->flags &= ~BNXT_FLAG_REGISTERED;
5904 bnxt_free_ctx_mem(bp);
5905 if (!reconfig_dev) {
5906 bnxt_free_hwrm_resources(bp);
5907 bnxt_free_error_recovery_info(bp);
5910 bnxt_uninit_ctx_mem(bp);
5912 bnxt_free_flow_stats_info(bp);
5913 if (bp->rep_info != NULL)
5914 bnxt_free_switch_domain(bp);
5915 bnxt_free_rep_info(bp);
5916 rte_free(bp->ptp_cfg);
5922 bnxt_dev_uninit(struct rte_eth_dev *eth_dev)
5924 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
5927 PMD_DRV_LOG(DEBUG, "Calling Device uninit\n");
5929 if (eth_dev->state != RTE_ETH_DEV_UNUSED)
5930 bnxt_dev_close_op(eth_dev);
5935 static int bnxt_pci_remove_dev_with_reps(struct rte_eth_dev *eth_dev)
5937 struct bnxt *bp = eth_dev->data->dev_private;
5938 struct rte_eth_dev *vf_rep_eth_dev;
5944 for (i = 0; i < bp->num_reps; i++) {
5945 vf_rep_eth_dev = bp->rep_info[i].vfr_eth_dev;
5946 if (!vf_rep_eth_dev)
5948 PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR pci remove\n",
5949 vf_rep_eth_dev->data->port_id);
5950 rte_eth_dev_destroy(vf_rep_eth_dev, bnxt_representor_uninit);
5952 PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci remove\n",
5953 eth_dev->data->port_id);
5954 ret = rte_eth_dev_destroy(eth_dev, bnxt_dev_uninit);
5959 static void bnxt_free_rep_info(struct bnxt *bp)
5961 rte_free(bp->rep_info);
5962 bp->rep_info = NULL;
5963 rte_free(bp->cfa_code_map);
5964 bp->cfa_code_map = NULL;
5967 static int bnxt_init_rep_info(struct bnxt *bp)
5974 bp->rep_info = rte_zmalloc("bnxt_rep_info",
5975 sizeof(bp->rep_info[0]) * BNXT_MAX_VF_REPS(bp),
5977 if (!bp->rep_info) {
5978 PMD_DRV_LOG(ERR, "Failed to alloc memory for rep info\n");
5981 bp->cfa_code_map = rte_zmalloc("bnxt_cfa_code_map",
5982 sizeof(*bp->cfa_code_map) *
5983 BNXT_MAX_CFA_CODE, 0);
5984 if (!bp->cfa_code_map) {
5985 PMD_DRV_LOG(ERR, "Failed to alloc memory for cfa_code_map\n");
5986 bnxt_free_rep_info(bp);
5990 for (i = 0; i < BNXT_MAX_CFA_CODE; i++)
5991 bp->cfa_code_map[i] = BNXT_VF_IDX_INVALID;
5993 rc = pthread_mutex_init(&bp->rep_info->vfr_lock, NULL);
5995 PMD_DRV_LOG(ERR, "Unable to initialize vfr_lock\n");
5996 bnxt_free_rep_info(bp);
6000 rc = pthread_mutex_init(&bp->rep_info->vfr_start_lock, NULL);
6002 PMD_DRV_LOG(ERR, "Unable to initialize vfr_start_lock\n");
6003 bnxt_free_rep_info(bp);
6010 static int bnxt_rep_port_probe(struct rte_pci_device *pci_dev,
6011 struct rte_eth_devargs *eth_da,
6012 struct rte_eth_dev *backing_eth_dev,
6013 const char *dev_args)
6015 struct rte_eth_dev *vf_rep_eth_dev;
6016 char name[RTE_ETH_NAME_MAX_LEN];
6017 struct bnxt *backing_bp = backing_eth_dev->data->dev_private;
6018 uint16_t max_vf_reps = BNXT_MAX_VF_REPS(backing_bp);
6022 struct rte_kvargs *kvlist = NULL;
6024 if (eth_da->type == RTE_ETH_REPRESENTOR_NONE)
6026 if (eth_da->type != RTE_ETH_REPRESENTOR_VF) {
6027 PMD_DRV_LOG(ERR, "unsupported representor type %d\n",
6031 num_rep = eth_da->nb_representor_ports;
6032 if (num_rep > max_vf_reps) {
6033 PMD_DRV_LOG(ERR, "nb_representor_ports = %d > %d MAX VF REPS\n",
6034 num_rep, max_vf_reps);
6038 if (num_rep >= RTE_MAX_ETHPORTS) {
6040 "nb_representor_ports = %d > %d MAX ETHPORTS\n",
6041 num_rep, RTE_MAX_ETHPORTS);
6045 if (!(BNXT_PF(backing_bp) || BNXT_VF_IS_TRUSTED(backing_bp))) {
6047 "Not a PF or trusted VF. No Representor support\n");
6048 /* Returning an error is not an option.
6049 * Applications are not handling this correctly
6054 if (bnxt_init_rep_info(backing_bp))
6057 for (i = 0; i < num_rep; i++) {
6058 struct bnxt_representor representor = {
6059 .vf_id = eth_da->representor_ports[i],
6060 .switch_domain_id = backing_bp->switch_domain_id,
6061 .parent_dev = backing_eth_dev
6064 if (representor.vf_id >= max_vf_reps) {
6065 PMD_DRV_LOG(ERR, "VF-Rep id %d >= %d MAX VF ID\n",
6066 representor.vf_id, max_vf_reps);
6070 /* representor port net_bdf_port */
6071 snprintf(name, sizeof(name), "net_%s_representor_%d",
6072 pci_dev->device.name, eth_da->representor_ports[i]);
6074 kvlist = rte_kvargs_parse(dev_args, bnxt_dev_args);
6077 * Handler for "rep_is_pf" devarg.
6078 * Invoked as for ex: "-a 000:00:0d.0,
6079 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6081 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_IS_PF,
6082 bnxt_parse_devarg_rep_is_pf,
6083 (void *)&representor);
6089 * Handler for "rep_based_pf" devarg.
6090 * Invoked as for ex: "-a 000:00:0d.0,
6091 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6093 ret = rte_kvargs_process(kvlist,
6094 BNXT_DEVARG_REP_BASED_PF,
6095 bnxt_parse_devarg_rep_based_pf,
6096 (void *)&representor);
6102 * Handler for "rep_based_pf" devarg.
6103 * Invoked as for ex: "-a 000:00:0d.0,
6104 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6106 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_Q_R2F,
6107 bnxt_parse_devarg_rep_q_r2f,
6108 (void *)&representor);
6114 * Handler for "rep_based_pf" devarg.
6115 * Invoked as for ex: "-a 000:00:0d.0,
6116 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6118 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_Q_F2R,
6119 bnxt_parse_devarg_rep_q_f2r,
6120 (void *)&representor);
6126 * Handler for "rep_based_pf" devarg.
6127 * Invoked as for ex: "-a 000:00:0d.0,
6128 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6130 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_FC_R2F,
6131 bnxt_parse_devarg_rep_fc_r2f,
6132 (void *)&representor);
6138 * Handler for "rep_based_pf" devarg.
6139 * Invoked as for ex: "-a 000:00:0d.0,
6140 * rep-based-pf=<pf index> rep-is-pf=<VF=0 or PF=1>"
6142 ret = rte_kvargs_process(kvlist, BNXT_DEVARG_REP_FC_F2R,
6143 bnxt_parse_devarg_rep_fc_f2r,
6144 (void *)&representor);
6151 ret = rte_eth_dev_create(&pci_dev->device, name,
6152 sizeof(struct bnxt_representor),
6154 bnxt_representor_init,
6157 PMD_DRV_LOG(ERR, "failed to create bnxt vf "
6158 "representor %s.", name);
6162 vf_rep_eth_dev = rte_eth_dev_allocated(name);
6163 if (!vf_rep_eth_dev) {
6164 PMD_DRV_LOG(ERR, "Failed to find the eth_dev"
6165 " for VF-Rep: %s.", name);
6170 PMD_DRV_LOG(DEBUG, "BNXT Port:%d VFR pci probe\n",
6171 backing_eth_dev->data->port_id);
6172 backing_bp->rep_info[representor.vf_id].vfr_eth_dev =
6174 backing_bp->num_reps++;
6178 rte_kvargs_free(kvlist);
6182 /* If num_rep > 1, then rollback already created
6183 * ports, since we'll be failing the probe anyway
6186 bnxt_pci_remove_dev_with_reps(backing_eth_dev);
6188 rte_kvargs_free(kvlist);
6193 static int bnxt_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
6194 struct rte_pci_device *pci_dev)
6196 struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
6197 struct rte_eth_dev *backing_eth_dev;
6201 if (pci_dev->device.devargs) {
6202 ret = rte_eth_devargs_parse(pci_dev->device.devargs->args,
6208 num_rep = eth_da.nb_representor_ports;
6209 PMD_DRV_LOG(DEBUG, "nb_representor_ports = %d\n",
6212 /* We could come here after first level of probe is already invoked
6213 * as part of an application bringup(OVS-DPDK vswitchd), so first check
6214 * for already allocated eth_dev for the backing device (PF/Trusted VF)
6216 backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6217 if (backing_eth_dev == NULL) {
6218 ret = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
6219 sizeof(struct bnxt),
6220 eth_dev_pci_specific_init, pci_dev,
6221 bnxt_dev_init, NULL);
6223 if (ret || !num_rep)
6226 backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6228 PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci probe\n",
6229 backing_eth_dev->data->port_id);
6234 /* probe representor ports now */
6235 ret = bnxt_rep_port_probe(pci_dev, ð_da, backing_eth_dev,
6236 pci_dev->device.devargs->args);
6241 static int bnxt_pci_remove(struct rte_pci_device *pci_dev)
6243 struct rte_eth_dev *eth_dev;
6245 eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
6247 return 0; /* Invoked typically only by OVS-DPDK, by the
6248 * time it comes here the eth_dev is already
6249 * deleted by rte_eth_dev_close(), so returning
6250 * +ve value will at least help in proper cleanup
6253 PMD_DRV_LOG(DEBUG, "BNXT Port:%d pci remove\n", eth_dev->data->port_id);
6254 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
6255 if (eth_dev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
6256 return rte_eth_dev_destroy(eth_dev,
6257 bnxt_representor_uninit);
6259 return rte_eth_dev_destroy(eth_dev,
6262 return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
6266 static struct rte_pci_driver bnxt_rte_pmd = {
6267 .id_table = bnxt_pci_id_map,
6268 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
6269 RTE_PCI_DRV_INTR_RMV |
6270 RTE_PCI_DRV_PROBE_AGAIN, /* Needed in case of VF-REPs
6273 .probe = bnxt_pci_probe,
6274 .remove = bnxt_pci_remove,
6278 is_device_supported(struct rte_eth_dev *dev, struct rte_pci_driver *drv)
6280 if (strcmp(dev->device->driver->name, drv->driver.name))
6286 bool is_bnxt_supported(struct rte_eth_dev *dev)
6288 return is_device_supported(dev, &bnxt_rte_pmd);
6291 RTE_LOG_REGISTER_SUFFIX(bnxt_logtype_driver, driver, NOTICE);
6292 RTE_PMD_REGISTER_PCI(net_bnxt, bnxt_rte_pmd);
6293 RTE_PMD_REGISTER_PCI_TABLE(net_bnxt, bnxt_pci_id_map);