net/bnxt: handle reset notify async event from FW
[dpdk.git] / drivers / net / bnxt / bnxt_ethdev.c
index 5e8bdff..6bc006a 100644 (file)
@@ -11,6 +11,7 @@
 #include <rte_ethdev_pci.h>
 #include <rte_malloc.h>
 #include <rte_cycles.h>
+#include <rte_alarm.h>
 
 #include "bnxt.h"
 #include "bnxt_cpr.h"
@@ -74,7 +75,8 @@ int bnxt_logtype_driver;
 #define BROADCOM_DEV_ID_57508 0x1750
 #define BROADCOM_DEV_ID_57504 0x1751
 #define BROADCOM_DEV_ID_57502 0x1752
-#define BROADCOM_DEV_ID_57500_VF 0x1807
+#define BROADCOM_DEV_ID_57500_VF1 0x1806
+#define BROADCOM_DEV_ID_57500_VF2 0x1807
 #define BROADCOM_DEV_ID_58802 0xd802
 #define BROADCOM_DEV_ID_58804 0xd804
 #define BROADCOM_DEV_ID_58808 0x16f0
@@ -126,7 +128,8 @@ static const struct rte_pci_id bnxt_pci_id_map[] = {
        { RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57508) },
        { RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57504) },
        { RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57502) },
-       { RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57500_VF) },
+       { RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57500_VF1) },
+       { RTE_PCI_DEVICE(PCI_VENDOR_ID_BROADCOM, BROADCOM_DEV_ID_57500_VF2) },
        { .vendor_id = 0, /* sentinel */ },
 };
 
@@ -164,6 +167,18 @@ static int bnxt_vlan_offload_set_op(struct rte_eth_dev *dev, int mask);
 static void bnxt_print_link_info(struct rte_eth_dev *eth_dev);
 static int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu);
 static int bnxt_dev_uninit(struct rte_eth_dev *eth_dev);
+static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev);
+static int bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev);
+
+int is_bnxt_in_error(struct bnxt *bp)
+{
+       if (bp->flags & BNXT_FLAG_FATAL_ERROR)
+               return -EIO;
+       if (bp->flags & BNXT_FLAG_FW_RESET)
+               return -EBUSY;
+
+       return 0;
+}
 
 /***********************/
 
@@ -189,21 +204,36 @@ static uint16_t  bnxt_rss_hash_tbl_size(const struct bnxt *bp)
        return bnxt_rss_ctxts(bp) * BNXT_RSS_ENTRIES_PER_CTX_THOR;
 }
 
-static void bnxt_free_mem(struct bnxt *bp)
+static void bnxt_free_mem(struct bnxt *bp, bool reconfig)
 {
        bnxt_free_filter_mem(bp);
        bnxt_free_vnic_attributes(bp);
        bnxt_free_vnic_mem(bp);
 
-       bnxt_free_stats(bp);
-       bnxt_free_tx_rings(bp);
-       bnxt_free_rx_rings(bp);
+       /* tx/rx rings are configured as part of *_queue_setup callbacks.
+        * If the number of rings change across fw update,
+        * we don't have much choice except to warn the user.
+        */
+       if (!reconfig) {
+               bnxt_free_stats(bp);
+               bnxt_free_tx_rings(bp);
+               bnxt_free_rx_rings(bp);
+       }
+       bnxt_free_async_cp_ring(bp);
 }
 
-static int bnxt_alloc_mem(struct bnxt *bp)
+static int bnxt_alloc_mem(struct bnxt *bp, bool reconfig)
 {
        int rc;
 
+       rc = bnxt_alloc_ring_grps(bp);
+       if (rc)
+               goto alloc_mem_err;
+
+       rc = bnxt_alloc_async_ring_struct(bp);
+       if (rc)
+               goto alloc_mem_err;
+
        rc = bnxt_alloc_vnic_mem(bp);
        if (rc)
                goto alloc_mem_err;
@@ -216,10 +246,14 @@ static int bnxt_alloc_mem(struct bnxt *bp)
        if (rc)
                goto alloc_mem_err;
 
+       rc = bnxt_alloc_async_cp_ring(bp);
+       if (rc)
+               goto alloc_mem_err;
+
        return 0;
 
 alloc_mem_err:
-       bnxt_free_mem(bp);
+       bnxt_free_mem(bp, reconfig);
        return rc;
 }
 
@@ -237,9 +271,6 @@ static int bnxt_init_chip(struct bnxt *bp)
        unsigned int i, j;
        int rc;
 
-       /* disable uio/vfio intr/eventfd mapping */
-       rte_intr_disable(intr_handle);
-
        if (bp->eth_dev->data->mtu > RTE_ETHER_MTU) {
                bp->eth_dev->data->dev_conf.rxmode.offloads |=
                        DEV_RX_OFFLOAD_JUMBO_FRAME;
@@ -416,7 +447,8 @@ static int bnxt_init_chip(struct bnxt *bp)
                        intr_handle->max_intr);
                for (queue_id = 0; queue_id < bp->eth_dev->data->nb_rx_queues;
                     queue_id++) {
-                       intr_handle->intr_vec[queue_id] = vec;
+                       intr_handle->intr_vec[queue_id] =
+                                                       vec + BNXT_RX_VEC_START;
                        if (vec < base + intr_handle->nb_efd - 1)
                                vec++;
                }
@@ -485,12 +517,17 @@ static int bnxt_init_nic(struct bnxt *bp)
  * Device configuration and status function
  */
 
-static void bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
-                                 struct rte_eth_dev_info *dev_info)
+static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
+                               struct rte_eth_dev_info *dev_info)
 {
        struct bnxt *bp = eth_dev->data->dev_private;
        uint16_t max_vnics, i, j, vpool, vrxq;
        unsigned int max_rx_rings;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        /* MAC Specifics */
        dev_info->max_mac_addrs = bp->max_l2_ctx;
@@ -579,6 +616,8 @@ found:
 
        dev_info->vmdq_pool_base = 0;
        dev_info->vmdq_queue_base = 0;
+
+       return 0;
 }
 
 /* Configure the device based on the configuration provided */
@@ -593,6 +632,10 @@ static int bnxt_dev_configure_op(struct rte_eth_dev *eth_dev)
        bp->tx_nr_rings = eth_dev->data->nb_tx_queues;
        bp->rx_nr_rings = eth_dev->data->nb_rx_queues;
 
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
+
        if (BNXT_VF(bp) && (bp->flags & BNXT_FLAG_NEW_RM)) {
                rc = bnxt_hwrm_check_vf_rings(bp);
                if (rc) {
@@ -617,8 +660,8 @@ static int bnxt_dev_configure_op(struct rte_eth_dev *eth_dev)
        /* Inherit new configurations */
        if (eth_dev->data->nb_rx_queues > bp->max_rx_rings ||
            eth_dev->data->nb_tx_queues > bp->max_tx_rings ||
-           eth_dev->data->nb_rx_queues + eth_dev->data->nb_tx_queues >
-           bp->max_cp_rings ||
+           eth_dev->data->nb_rx_queues + eth_dev->data->nb_tx_queues
+               + BNXT_NUM_ASYNC_CPR(bp) > bp->max_cp_rings ||
            eth_dev->data->nb_rx_queues + eth_dev->data->nb_tx_queues >
            bp->max_stat_ctx)
                goto resource_error;
@@ -731,13 +774,11 @@ bnxt_transmit_function(__rte_unused struct rte_eth_dev *eth_dev)
 {
 #ifdef RTE_ARCH_X86
        /*
-        * Vector mode receive can be enabled only if scatter tx is not
-        * in use and tx offloads other than VLAN insertion are not
-        * in use.
+        * Vector mode transmit can be enabled only if not using scatter rx
+        * or tx offloads.
         */
        if (!eth_dev->data->scattered_rx &&
-           !(eth_dev->data->dev_conf.txmode.offloads &
-             ~DEV_TX_OFFLOAD_VLAN_INSERT)) {
+           !eth_dev->data->dev_conf.txmode.offloads) {
                PMD_DRV_LOG(INFO, "Using vector mode transmit for port %d\n",
                            eth_dev->data->port_id);
                return bnxt_xmit_pkts_vec;
@@ -784,8 +825,10 @@ static int bnxt_dev_start_op(struct rte_eth_dev *eth_dev)
 
        eth_dev->rx_pkt_burst = bnxt_receive_function(eth_dev);
        eth_dev->tx_pkt_burst = bnxt_transmit_function(eth_dev);
+
        bnxt_enable_int(bp);
        bp->flags |= BNXT_FLAG_INIT_DONE;
+       eth_dev->data->dev_started = 1;
        bp->dev_stopped = 0;
        return 0;
 
@@ -828,6 +871,11 @@ static void bnxt_dev_stop_op(struct rte_eth_dev *eth_dev)
        struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
        struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
 
+       eth_dev->data->dev_started = 0;
+       /* Prevent crashes when queues are still in use */
+       eth_dev->rx_pkt_burst = &bnxt_dummy_recv_pkts;
+       eth_dev->tx_pkt_burst = &bnxt_dummy_xmit_pkts;
+
        bnxt_disable_int(bp);
 
        /* disable uio/vfio intr/eventfd mapping */
@@ -882,6 +930,9 @@ static void bnxt_mac_addr_remove_op(struct rte_eth_dev *eth_dev,
        struct bnxt_filter_info *filter, *temp_filter;
        uint32_t i;
 
+       if (is_bnxt_in_error(bp))
+               return;
+
        /*
         * Loop through all VNICs from the specified filter flow pools to
         * remove the corresponding MAC addr filter
@@ -917,6 +968,10 @@ static int bnxt_mac_addr_add_op(struct rte_eth_dev *eth_dev,
        struct bnxt_filter_info *filter;
        int rc = 0;
 
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
+
        if (BNXT_VF(bp) & !BNXT_VF_IS_TRUSTED(bp)) {
                PMD_DRV_LOG(ERR, "Cannot add MAC address to a VF interface\n");
                return -ENOTSUP;
@@ -962,6 +1017,10 @@ int bnxt_link_update_op(struct rte_eth_dev *eth_dev, int wait_to_complete)
        struct rte_eth_link new;
        unsigned int cnt = BNXT_LINK_WAIT_CNT;
 
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
+
        memset(&new, 0, sizeof(new));
        do {
                /* Retrieve link info from hardware */
@@ -997,60 +1056,104 @@ out:
        return rc;
 }
 
-static void bnxt_promiscuous_enable_op(struct rte_eth_dev *eth_dev)
+static int bnxt_promiscuous_enable_op(struct rte_eth_dev *eth_dev)
 {
        struct bnxt *bp = eth_dev->data->dev_private;
        struct bnxt_vnic_info *vnic;
+       uint32_t old_flags;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        if (bp->vnic_info == NULL)
-               return;
+               return 0;
 
        vnic = &bp->vnic_info[0];
 
+       old_flags = vnic->flags;
        vnic->flags |= BNXT_VNIC_INFO_PROMISC;
-       bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
+       rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
+       if (rc != 0)
+               vnic->flags = old_flags;
+
+       return rc;
 }
 
-static void bnxt_promiscuous_disable_op(struct rte_eth_dev *eth_dev)
+static int bnxt_promiscuous_disable_op(struct rte_eth_dev *eth_dev)
 {
        struct bnxt *bp = eth_dev->data->dev_private;
        struct bnxt_vnic_info *vnic;
+       uint32_t old_flags;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        if (bp->vnic_info == NULL)
-               return;
+               return 0;
 
        vnic = &bp->vnic_info[0];
 
+       old_flags = vnic->flags;
        vnic->flags &= ~BNXT_VNIC_INFO_PROMISC;
-       bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
+       rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
+       if (rc != 0)
+               vnic->flags = old_flags;
+
+       return rc;
 }
 
-static void bnxt_allmulticast_enable_op(struct rte_eth_dev *eth_dev)
+static int bnxt_allmulticast_enable_op(struct rte_eth_dev *eth_dev)
 {
        struct bnxt *bp = eth_dev->data->dev_private;
        struct bnxt_vnic_info *vnic;
+       uint32_t old_flags;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        if (bp->vnic_info == NULL)
-               return;
+               return 0;
 
        vnic = &bp->vnic_info[0];
 
+       old_flags = vnic->flags;
        vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
-       bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
+       rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
+       if (rc != 0)
+               vnic->flags = old_flags;
+
+       return rc;
 }
 
-static void bnxt_allmulticast_disable_op(struct rte_eth_dev *eth_dev)
+static int bnxt_allmulticast_disable_op(struct rte_eth_dev *eth_dev)
 {
        struct bnxt *bp = eth_dev->data->dev_private;
        struct bnxt_vnic_info *vnic;
+       uint32_t old_flags;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        if (bp->vnic_info == NULL)
-               return;
+               return 0;
 
        vnic = &bp->vnic_info[0];
 
+       old_flags = vnic->flags;
        vnic->flags &= ~BNXT_VNIC_INFO_ALLMULTI;
-       bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
+       rc = bnxt_hwrm_cfa_l2_set_rx_mask(bp, vnic, 0, NULL);
+       if (rc != 0)
+               vnic->flags = old_flags;
+
+       return rc;
 }
 
 /* Return bnxt_rx_queue pointer corresponding to a given rxq. */
@@ -1093,7 +1196,11 @@ static int bnxt_reta_update_op(struct rte_eth_dev *eth_dev,
        struct bnxt_vnic_info *vnic = &bp->vnic_info[0];
        uint16_t tbl_size = bnxt_rss_hash_tbl_size(bp);
        uint16_t idx, sft;
-       int i;
+       int i, rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        if (!vnic->rss_table)
                return -EINVAL;
@@ -1149,6 +1256,11 @@ static int bnxt_reta_query_op(struct rte_eth_dev *eth_dev,
        struct bnxt_vnic_info *vnic = &bp->vnic_info[0];
        uint16_t tbl_size = bnxt_rss_hash_tbl_size(bp);
        uint16_t idx, sft, i;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        /* Retrieve from the default VNIC */
        if (!vnic)
@@ -1195,6 +1307,11 @@ static int bnxt_rss_hash_update_op(struct rte_eth_dev *eth_dev,
        struct bnxt_vnic_info *vnic;
        uint16_t hash_type = 0;
        unsigned int i;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        /*
         * If RSS enablement were different than dev_configure,
@@ -1248,9 +1365,13 @@ static int bnxt_rss_hash_conf_get_op(struct rte_eth_dev *eth_dev,
 {
        struct bnxt *bp = eth_dev->data->dev_private;
        struct bnxt_vnic_info *vnic = &bp->vnic_info[0];
-       int len;
+       int len, rc;
        uint32_t hash_types;
 
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
+
        /* RSS configuration is the same for all VNICs */
        if (vnic && vnic->rss_hash_key) {
                if (rss_conf->rss_key) {
@@ -1308,6 +1429,10 @@ static int bnxt_flow_ctrl_get_op(struct rte_eth_dev *dev,
        struct rte_eth_link link_info;
        int rc;
 
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
+
        rc = bnxt_get_hwrm_link_config(bp, &link_info);
        if (rc)
                return rc;
@@ -1337,6 +1462,11 @@ static int bnxt_flow_ctrl_set_op(struct rte_eth_dev *dev,
                               struct rte_eth_fc_conf *fc_conf)
 {
        struct bnxt *bp = dev->data->dev_private;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        if (!BNXT_SINGLE_PF(bp) || BNXT_VF(bp)) {
                PMD_DRV_LOG(ERR, "Flow Control Settings cannot be modified\n");
@@ -1396,6 +1526,10 @@ bnxt_udp_tunnel_port_add_op(struct rte_eth_dev *eth_dev,
        uint16_t tunnel_type = 0;
        int rc = 0;
 
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
+
        switch (udp_tunnel->prot_type) {
        case RTE_TUNNEL_TYPE_VXLAN:
                if (bp->vxlan_port_cnt) {
@@ -1445,6 +1579,10 @@ bnxt_udp_tunnel_port_del_op(struct rte_eth_dev *eth_dev,
        uint16_t port = 0;
        int rc = 0;
 
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
+
        switch (udp_tunnel->prot_type) {
        case RTE_TUNNEL_TYPE_VXLAN:
                if (!bp->vxlan_port_cnt) {
@@ -1598,6 +1736,11 @@ static int bnxt_vlan_filter_set_op(struct rte_eth_dev *eth_dev,
                uint16_t vlan_id, int on)
 {
        struct bnxt *bp = eth_dev->data->dev_private;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        /* These operations apply to ALL existing MAC/VLAN filters */
        if (on)
@@ -1612,6 +1755,11 @@ bnxt_vlan_offload_set_op(struct rte_eth_dev *dev, int mask)
        struct bnxt *bp = dev->data->dev_private;
        uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
        unsigned int i;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        if (mask & ETH_VLAN_FILTER_MASK) {
                if (!(rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)) {
@@ -1653,6 +1801,10 @@ bnxt_set_default_mac_addr_op(struct rte_eth_dev *dev,
        struct bnxt_filter_info *filter;
        int rc;
 
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
+
        if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp))
                return -EPERM;
 
@@ -1692,6 +1844,11 @@ bnxt_dev_set_mc_addr_list_op(struct rte_eth_dev *eth_dev,
        char *mc_addr_list = (char *)mc_addr_set;
        struct bnxt_vnic_info *vnic;
        uint32_t off = 0, i = 0;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        vnic = &bp->vnic_info[0];
 
@@ -1777,10 +1934,18 @@ static int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
        uint32_t rc = 0;
        uint32_t i;
 
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
+
        new_pkt_size = new_mtu + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN +
                       VLAN_TAG_SIZE * BNXT_NUM_VLANS;
 
-       bnxt_dev_info_get_op(eth_dev, &dev_info);
+       rc = bnxt_dev_info_get_op(eth_dev, &dev_info);
+       if (rc != 0) {
+               PMD_DRV_LOG(ERR, "Error during getting ethernet device info\n");
+               return rc;
+       }
 
        if (new_mtu < RTE_ETHER_MIN_MTU || new_mtu > BNXT_MAX_MTU) {
                PMD_DRV_LOG(ERR, "MTU requested must be within (%d, %d)\n",
@@ -1850,6 +2015,10 @@ bnxt_vlan_pvid_set_op(struct rte_eth_dev *dev, uint16_t pvid, int on)
        uint16_t vlan = bp->vlan;
        int rc;
 
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
+
        if (!BNXT_SINGLE_PF(bp) || BNXT_VF(bp)) {
                PMD_DRV_LOG(ERR,
                        "PVID cannot be modified for this function\n");
@@ -1867,6 +2036,11 @@ static int
 bnxt_dev_led_on_op(struct rte_eth_dev *dev)
 {
        struct bnxt *bp = dev->data->dev_private;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        return bnxt_hwrm_port_led_cfg(bp, true);
 }
@@ -1875,6 +2049,11 @@ static int
 bnxt_dev_led_off_op(struct rte_eth_dev *dev)
 {
        struct bnxt *bp = dev->data->dev_private;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        return bnxt_hwrm_port_led_cfg(bp, false);
 }
@@ -1882,6 +2061,7 @@ bnxt_dev_led_off_op(struct rte_eth_dev *dev)
 static uint32_t
 bnxt_rx_queue_count_op(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 {
+       struct bnxt *bp = (struct bnxt *)dev->data->dev_private;
        uint32_t desc = 0, raw_cons = 0, cons;
        struct bnxt_cp_ring_info *cpr;
        struct bnxt_rx_queue *rxq;
@@ -1889,6 +2069,11 @@ bnxt_rx_queue_count_op(struct rte_eth_dev *dev, uint16_t rx_queue_id)
        uint16_t cmp_type;
        uint8_t cmp = 1;
        bool valid;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        rxq = dev->data->rx_queues[rx_queue_id];
        cpr = rxq->cp_ring;
@@ -1933,10 +2118,15 @@ bnxt_rx_descriptor_status_op(void *rx_queue, uint16_t offset)
        struct bnxt_sw_rx_bd *rx_buf;
        struct rx_pkt_cmpl *rxcmp;
        uint32_t cons, cp_cons;
+       int rc;
 
        if (!rxq)
                return -EINVAL;
 
+       rc = is_bnxt_in_error(rxq->bp);
+       if (rc)
+               return rc;
+
        cpr = rxq->cp_ring;
        rxr = rxq->rx_ring;
 
@@ -1971,10 +2161,15 @@ bnxt_tx_descriptor_status_op(void *tx_queue, uint16_t offset)
        struct bnxt_sw_tx_bd *tx_buf;
        struct tx_pkt_cmpl *txcmp;
        uint32_t cons, cp_cons;
+       int rc;
 
        if (!txq)
                return -EINVAL;
 
+       rc = is_bnxt_in_error(txq->bp);
+       if (rc)
+               return rc;
+
        cpr = txq->cp_ring;
        txr = txq->tx_ring;
 
@@ -2804,6 +2999,10 @@ bnxt_filter_ctrl_op(struct rte_eth_dev *dev __rte_unused,
 {
        int ret = 0;
 
+       ret = is_bnxt_in_error(dev->data->dev_private);
+       if (ret)
+               return ret;
+
        switch (filter_type) {
        case RTE_ETH_FILTER_TUNNEL:
                PMD_DRV_LOG(ERR,
@@ -3119,6 +3318,10 @@ bnxt_get_eeprom_length_op(struct rte_eth_dev *dev)
        uint32_t dir_entries;
        uint32_t entry_length;
 
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
+
        PMD_DRV_LOG(INFO, "%04x:%02x:%02x:%02x\n",
                bp->pdev->addr.domain, bp->pdev->addr.bus,
                bp->pdev->addr.devid, bp->pdev->addr.function);
@@ -3137,6 +3340,11 @@ bnxt_get_eeprom_op(struct rte_eth_dev *dev,
        struct bnxt *bp = dev->data->dev_private;
        uint32_t index;
        uint32_t offset;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        PMD_DRV_LOG(INFO, "%04x:%02x:%02x:%02x in_eeprom->offset = %d "
                "len = %d\n", bp->pdev->addr.domain,
@@ -3208,6 +3416,11 @@ bnxt_set_eeprom_op(struct rte_eth_dev *dev,
        struct bnxt *bp = dev->data->dev_private;
        uint8_t index, dir_op;
        uint16_t type, ext, ordinal, attr;
+       int rc;
+
+       rc = is_bnxt_in_error(bp);
+       if (rc)
+               return rc;
 
        PMD_DRV_LOG(INFO, "%04x:%02x:%02x:%02x in_eeprom->offset = %d "
                "len = %d\n", bp->pdev->addr.domain,
@@ -3319,6 +3532,89 @@ static const struct eth_dev_ops bnxt_dev_ops = {
        .timesync_read_tx_timestamp = bnxt_timesync_read_tx_timestamp,
 };
 
+static void bnxt_dev_cleanup(struct bnxt *bp)
+{
+       bnxt_set_hwrm_link_config(bp, false);
+       bp->link_info.link_up = 0;
+       if (bp->dev_stopped == 0)
+               bnxt_dev_stop_op(bp->eth_dev);
+
+       bnxt_uninit_resources(bp, true);
+}
+
+static int bnxt_restore_filters(struct bnxt *bp)
+{
+       struct rte_eth_dev *dev = bp->eth_dev;
+       int ret = 0;
+
+       if (dev->data->all_multicast)
+               ret = bnxt_allmulticast_enable_op(dev);
+       if (dev->data->promiscuous)
+               ret = bnxt_promiscuous_enable_op(dev);
+
+       /* TODO restore other filters as well */
+       return ret;
+}
+
+static void bnxt_dev_recover(void *arg)
+{
+       struct bnxt *bp = arg;
+       int timeout = bp->fw_reset_max_msecs;
+       int rc = 0;
+
+       do {
+               rc = bnxt_hwrm_ver_get(bp);
+               if (rc == 0)
+                       break;
+               rte_delay_ms(BNXT_FW_READY_WAIT_INTERVAL);
+               timeout -= BNXT_FW_READY_WAIT_INTERVAL;
+       } while (rc && timeout);
+
+       if (rc) {
+               PMD_DRV_LOG(ERR, "FW is not Ready after reset\n");
+               goto err;
+       }
+
+       rc = bnxt_init_resources(bp, true);
+       if (rc) {
+               PMD_DRV_LOG(ERR,
+                           "Failed to initialize resources after reset\n");
+               goto err;
+       }
+       /* clear reset flag as the device is initialized now */
+       bp->flags &= ~BNXT_FLAG_FW_RESET;
+
+       rc = bnxt_dev_start_op(bp->eth_dev);
+       if (rc) {
+               PMD_DRV_LOG(ERR, "Failed to start port after reset\n");
+               goto err;
+       }
+
+       rc = bnxt_restore_filters(bp);
+       if (rc)
+               goto err;
+
+       PMD_DRV_LOG(INFO, "Recovered from FW reset\n");
+       return;
+err:
+       bp->flags |= BNXT_FLAG_FATAL_ERROR;
+       bnxt_uninit_resources(bp, false);
+       PMD_DRV_LOG(ERR, "Failed to recover from FW reset\n");
+}
+
+void bnxt_dev_reset_and_resume(void *arg)
+{
+       struct bnxt *bp = arg;
+       int rc;
+
+       bnxt_dev_cleanup(bp);
+
+       rc = rte_eal_alarm_set(US_PER_MS * bp->fw_reset_min_msecs,
+                              bnxt_dev_recover, (void *)bp);
+       if (rc)
+               PMD_DRV_LOG(ERR, "Error setting recovery alarm");
+}
+
 static bool bnxt_vf_pciid(uint16_t id)
 {
        if (id == BROADCOM_DEV_ID_57304_VF ||
@@ -3329,7 +3625,8 @@ static bool bnxt_vf_pciid(uint16_t id)
            id == BROADCOM_DEV_ID_STRATUS_NIC_VF1 ||
            id == BROADCOM_DEV_ID_STRATUS_NIC_VF2 ||
            id == BROADCOM_DEV_ID_58802_VF ||
-           id == BROADCOM_DEV_ID_57500_VF)
+           id == BROADCOM_DEV_ID_57500_VF1 ||
+           id == BROADCOM_DEV_ID_57500_VF2)
                return true;
        return false;
 }
@@ -3391,8 +3688,9 @@ static int bnxt_alloc_ctx_mem_blk(__rte_unused struct bnxt *bp,
        valid_bits = PTU_PTE_VALID;
 
        if (rmem->nr_pages > 1) {
-               snprintf(mz_name, RTE_MEMZONE_NAMESIZE, "bnxt_ctx_pg_tbl%s_%x",
-                        suffix, idx);
+               snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
+                        "bnxt_ctx_pg_tbl%s_%x_%d",
+                        suffix, idx, bp->eth_dev->data->port_id);
                mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
                mz = rte_memzone_lookup(mz_name);
                if (!mz) {
@@ -3428,7 +3726,8 @@ static int bnxt_alloc_ctx_mem_blk(__rte_unused struct bnxt *bp,
                rmem->pg_tbl_mz = mz;
        }
 
-       snprintf(mz_name, RTE_MEMZONE_NAMESIZE, "bnxt_ctx_%s_%x", suffix, idx);
+       snprintf(mz_name, RTE_MEMZONE_NAMESIZE, "bnxt_ctx_%s_%x_%d",
+                suffix, idx, bp->eth_dev->data->port_id);
        mz = rte_memzone_lookup(mz_name);
        if (!mz) {
                mz = rte_memzone_reserve_aligned(mz_name,
@@ -3758,138 +4057,47 @@ static int bnxt_setup_mac_addr(struct rte_eth_dev *eth_dev)
        return rc;
 }
 
-#define ALLOW_FUNC(x)  \
-       { \
-               uint32_t arg = (x); \
-               bp->pf.vf_req_fwd[((arg) >> 5)] &= \
-               ~rte_cpu_to_le_32(1 << ((arg) & 0x1f)); \
-       }
-static int
-bnxt_dev_init(struct rte_eth_dev *eth_dev)
+static int bnxt_restore_dflt_mac(struct bnxt *bp)
 {
-       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
-       static int version_printed;
-       struct bnxt *bp;
-       uint16_t mtu;
-       int rc;
-
-       if (version_printed++ == 0)
-               PMD_DRV_LOG(INFO, "%s\n", bnxt_version);
-
-       rte_eth_copy_pci_info(eth_dev, pci_dev);
-
-       bp = eth_dev->data->dev_private;
-
-       bp->dev_stopped = 1;
-
-       eth_dev->dev_ops = &bnxt_dev_ops;
-       eth_dev->rx_pkt_burst = &bnxt_recv_pkts;
-       eth_dev->tx_pkt_burst = &bnxt_xmit_pkts;
+       int rc = 0;
 
-       /*
-        * For secondary processes, we don't initialise any further
-        * as primary has already done this work.
-        */
-       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+       /* MAC is already configured in FW */
+       if (!bnxt_check_zero_bytes(bp->dflt_mac_addr, RTE_ETHER_ADDR_LEN))
                return 0;
 
-       if (bnxt_vf_pciid(pci_dev->id.device_id))
-               bp->flags |= BNXT_FLAG_VF;
-
-       if (pci_dev->id.device_id == BROADCOM_DEV_ID_57508 ||
-           pci_dev->id.device_id == BROADCOM_DEV_ID_57504 ||
-           pci_dev->id.device_id == BROADCOM_DEV_ID_57502 ||
-           pci_dev->id.device_id == BROADCOM_DEV_ID_57500_VF)
-               bp->flags |= BNXT_FLAG_THOR_CHIP;
-
-       rc = bnxt_init_board(eth_dev);
-       if (rc) {
-               PMD_DRV_LOG(ERR,
-                       "Board initialization failed rc: %x\n", rc);
-               goto error;
-       }
-
-       rc = bnxt_alloc_hwrm_resources(bp);
-       if (rc) {
-               PMD_DRV_LOG(ERR,
-                       "hwrm resource allocation failure rc: %x\n", rc);
-               goto error_free;
-       }
-       rc = bnxt_hwrm_ver_get(bp);
-       if (rc)
-               goto error_free;
-
-       rc = bnxt_hwrm_func_reset(bp);
-       if (rc) {
-               PMD_DRV_LOG(ERR, "hwrm chip reset failure rc: %x\n", rc);
-               rc = -EIO;
-               goto error_free;
-       }
-
-       rc = bnxt_hwrm_queue_qportcfg(bp);
-       if (rc) {
-               PMD_DRV_LOG(ERR, "hwrm queue qportcfg failed\n");
-               goto error_free;
-       }
-       /* Get the MAX capabilities for this function */
-       rc = bnxt_hwrm_func_qcaps(bp);
-       if (rc) {
-               PMD_DRV_LOG(ERR, "hwrm query capability failure rc: %x\n", rc);
-               goto error_free;
-       }
-
-       rc = bnxt_alloc_stats_mem(bp);
+       /* Restore the old MAC configured */
+       rc = bnxt_hwrm_set_mac(bp);
        if (rc)
-               goto error_free;
-
-       if (bp->max_tx_rings == 0) {
-               PMD_DRV_LOG(ERR, "No TX rings available!\n");
-               rc = -EBUSY;
-               goto error_free;
-       }
+               PMD_DRV_LOG(ERR, "Failed to restore MAC address\n");
 
-       rc = bnxt_setup_mac_addr(eth_dev);
-       if (rc)
-               goto error_free;
+       return rc;
+}
 
-       /* THOR does not support ring groups.
-        * But we will use the array to save RSS context IDs.
-        */
-       if (BNXT_CHIP_THOR(bp)) {
-               bp->max_ring_grps = BNXT_MAX_RSS_CTXTS_THOR;
-       } else if (bp->max_ring_grps < bp->rx_cp_nr_rings) {
-               /* 1 ring is for default completion ring */
-               PMD_DRV_LOG(ERR, "Insufficient resource: Ring Group\n");
-               rc = -ENOSPC;
-               goto error_free;
-       }
+static void bnxt_config_vf_req_fwd(struct bnxt *bp)
+{
+       if (!BNXT_PF(bp))
+               return;
 
-       if (BNXT_HAS_RING_GRPS(bp)) {
-               bp->grp_info = rte_zmalloc("bnxt_grp_info",
-                                       sizeof(*bp->grp_info) *
-                                               bp->max_ring_grps, 0);
-               if (!bp->grp_info) {
-                       PMD_DRV_LOG(ERR,
-                               "Failed to alloc %zu bytes for grp info tbl.\n",
-                               sizeof(*bp->grp_info) * bp->max_ring_grps);
-                       rc = -ENOMEM;
-                       goto error_free;
-               }
+#define ALLOW_FUNC(x)  \
+       { \
+               uint32_t arg = (x); \
+               bp->pf.vf_req_fwd[((arg) >> 5)] &= \
+               ~rte_cpu_to_le_32(1 << ((arg) & 0x1f)); \
        }
 
        /* Forward all requests if firmware is new enough */
        if (((bp->fw_ver >= ((20 << 24) | (6 << 16) | (100 << 8))) &&
-           (bp->fw_ver < ((20 << 24) | (7 << 16)))) ||
+            (bp->fw_ver < ((20 << 24) | (7 << 16)))) ||
            ((bp->fw_ver >= ((20 << 24) | (8 << 16))))) {
                memset(bp->pf.vf_req_fwd, 0xff, sizeof(bp->pf.vf_req_fwd));
        } else {
                PMD_DRV_LOG(WARNING,
-                       "Firmware too old for VF mailbox functionality\n");
+                           "Firmware too old for VF mailbox functionality\n");
                memset(bp->pf.vf_req_fwd, 0, sizeof(bp->pf.vf_req_fwd));
        }
 
        /*
-        * The following are used for driver cleanup.  If we disallow these,
+        * The following are used for driver cleanup. If we disallow these,
         * VF drivers can't clean up cleanly.
         */
        ALLOW_FUNC(HWRM_FUNC_DRV_UNRGTR);
@@ -3901,70 +4109,195 @@ bnxt_dev_init(struct rte_eth_dev *eth_dev)
        ALLOW_FUNC(HWRM_STAT_CTX_FREE);
        ALLOW_FUNC(HWRM_PORT_PHY_QCFG);
        ALLOW_FUNC(HWRM_VNIC_TPA_CFG);
-       rc = bnxt_hwrm_func_driver_register(bp);
-       if (rc) {
-               PMD_DRV_LOG(ERR,
-                       "Failed to register driver");
-               rc = -EBUSY;
-               goto error_free;
-       }
+}
 
-       PMD_DRV_LOG(INFO,
-               DRV_MODULE_NAME " found at mem %" PRIx64 ", node addr %pM\n",
-               pci_dev->mem_resource[0].phys_addr,
-               pci_dev->mem_resource[0].addr);
+static int bnxt_init_fw(struct bnxt *bp)
+{
+       uint16_t mtu;
+       int rc = 0;
+
+       rc = bnxt_hwrm_ver_get(bp);
+       if (rc)
+               return rc;
+
+       rc = bnxt_hwrm_func_reset(bp);
+       if (rc)
+               return -EIO;
+
+       rc = bnxt_hwrm_queue_qportcfg(bp);
+       if (rc)
+               return rc;
+
+       /* Get the MAX capabilities for this function */
+       rc = bnxt_hwrm_func_qcaps(bp);
+       if (rc)
+               return rc;
 
        rc = bnxt_hwrm_func_qcfg(bp, &mtu);
-       if (rc) {
-               PMD_DRV_LOG(ERR, "hwrm func qcfg failed\n");
-               goto error_free;
-       }
+       if (rc)
+               return rc;
 
        if (mtu >= RTE_ETHER_MIN_MTU && mtu <= BNXT_MAX_MTU &&
-           mtu != eth_dev->data->mtu)
-               eth_dev->data->mtu = mtu;
+           mtu != bp->eth_dev->data->mtu)
+               bp->eth_dev->data->mtu = mtu;
+
+       bnxt_hwrm_port_led_qcaps(bp);
+
+       return 0;
+}
+
+static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev)
+{
+       int rc;
+
+       rc = bnxt_init_fw(bp);
+       if (rc)
+               return rc;
+
+       if (!reconfig_dev) {
+               rc = bnxt_setup_mac_addr(bp->eth_dev);
+               if (rc)
+                       return rc;
+       } else {
+               rc = bnxt_restore_dflt_mac(bp);
+               if (rc)
+                       return rc;
+       }
+
+       bnxt_config_vf_req_fwd(bp);
+
+       rc = bnxt_hwrm_func_driver_register(bp);
+       if (rc) {
+               PMD_DRV_LOG(ERR, "Failed to register driver");
+               return -EBUSY;
+       }
 
        if (BNXT_PF(bp)) {
-               //if (bp->pf.active_vfs) {
-                       // TODO: Deallocate VF resources?
-               //}
                if (bp->pdev->max_vfs) {
                        rc = bnxt_hwrm_allocate_vfs(bp, bp->pdev->max_vfs);
                        if (rc) {
                                PMD_DRV_LOG(ERR, "Failed to allocate VFs\n");
-                               goto error_free;
+                               return rc;
                        }
                } else {
                        rc = bnxt_hwrm_allocate_pf_only(bp);
                        if (rc) {
                                PMD_DRV_LOG(ERR,
-                                       "Failed to allocate PF resources\n");
-                               goto error_free;
+                                           "Failed to allocate PF resources");
+                               return rc;
                        }
                }
        }
 
-       bnxt_hwrm_port_led_qcaps(bp);
-
-       rc = bnxt_setup_int(bp);
+       rc = bnxt_alloc_mem(bp, reconfig_dev);
        if (rc)
-               goto error_free;
+               return rc;
 
-       rc = bnxt_alloc_mem(bp);
+       rc = bnxt_setup_int(bp);
        if (rc)
-               goto error_free;
+               return rc;
 
        bnxt_init_nic(bp);
 
        rc = bnxt_request_int(bp);
+       if (rc)
+               return rc;
+
+       return 0;
+}
+
+static int
+bnxt_dev_init(struct rte_eth_dev *eth_dev)
+{
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
+       static int version_printed;
+       struct bnxt *bp;
+       int rc;
+
+       if (version_printed++ == 0)
+               PMD_DRV_LOG(INFO, "%s\n", bnxt_version);
+
+       rte_eth_copy_pci_info(eth_dev, pci_dev);
+
+       bp = eth_dev->data->dev_private;
+
+       bp->dev_stopped = 1;
+
+       eth_dev->dev_ops = &bnxt_dev_ops;
+       eth_dev->rx_pkt_burst = &bnxt_recv_pkts;
+       eth_dev->tx_pkt_burst = &bnxt_xmit_pkts;
+
+       /*
+        * For secondary processes, we don't initialise any further
+        * as primary has already done this work.
+        */
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return 0;
+
+       if (bnxt_vf_pciid(pci_dev->id.device_id))
+               bp->flags |= BNXT_FLAG_VF;
+
+       if (pci_dev->id.device_id == BROADCOM_DEV_ID_57508 ||
+           pci_dev->id.device_id == BROADCOM_DEV_ID_57504 ||
+           pci_dev->id.device_id == BROADCOM_DEV_ID_57502 ||
+           pci_dev->id.device_id == BROADCOM_DEV_ID_57500_VF1 ||
+           pci_dev->id.device_id == BROADCOM_DEV_ID_57500_VF2)
+               bp->flags |= BNXT_FLAG_THOR_CHIP;
+
+       if (pci_dev->id.device_id == BROADCOM_DEV_ID_58802 ||
+           pci_dev->id.device_id == BROADCOM_DEV_ID_58804 ||
+           pci_dev->id.device_id == BROADCOM_DEV_ID_58808 ||
+           pci_dev->id.device_id == BROADCOM_DEV_ID_58802_VF)
+               bp->flags |= BNXT_FLAG_STINGRAY;
+
+       rc = bnxt_init_board(eth_dev);
+       if (rc) {
+               PMD_DRV_LOG(ERR,
+                           "Failed to initialize board rc: %x\n", rc);
+               return rc;
+       }
+
+       rc = bnxt_alloc_hwrm_resources(bp);
+       if (rc) {
+               PMD_DRV_LOG(ERR,
+                           "Failed to allocate hwrm resource rc: %x\n", rc);
+               goto error_free;
+       }
+       rc = bnxt_init_resources(bp, false);
        if (rc)
                goto error_free;
 
+       rc = bnxt_alloc_stats_mem(bp);
+       if (rc)
+               goto error_free;
+
+       PMD_DRV_LOG(INFO,
+                   DRV_MODULE_NAME "found at mem %" PRIX64 ", node addr %pM\n",
+                   pci_dev->mem_resource[0].phys_addr,
+                   pci_dev->mem_resource[0].addr);
+
        return 0;
 
 error_free:
        bnxt_dev_uninit(eth_dev);
-error:
+       return rc;
+}
+
+static int
+bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev)
+{
+       int rc;
+
+       bnxt_disable_int(bp);
+       bnxt_free_int(bp);
+       bnxt_free_mem(bp, reconfig_dev);
+       bnxt_hwrm_func_buf_unrgtr(bp);
+       rc = bnxt_hwrm_func_driver_unregister(bp, 0);
+       bp->flags &= ~BNXT_FLAG_REGISTERED;
+       bnxt_free_ctx_mem(bp);
+       if (!reconfig_dev)
+               bnxt_free_hwrm_resources(bp);
+
        return rc;
 }
 
@@ -3978,18 +4311,13 @@ bnxt_dev_uninit(struct rte_eth_dev *eth_dev)
                return -EPERM;
 
        PMD_DRV_LOG(DEBUG, "Calling Device uninit\n");
-       bnxt_disable_int(bp);
-       bnxt_free_int(bp);
-       bnxt_free_mem(bp);
 
-       bnxt_hwrm_func_buf_unrgtr(bp);
+       rc = bnxt_uninit_resources(bp, false);
 
        if (bp->grp_info != NULL) {
                rte_free(bp->grp_info);
                bp->grp_info = NULL;
        }
-       rc = bnxt_hwrm_func_driver_unregister(bp, 0);
-       bnxt_free_hwrm_resources(bp);
 
        if (bp->tx_mem_zone) {
                rte_memzone_free((const struct rte_memzone *)bp->tx_mem_zone);
@@ -4005,7 +4333,6 @@ bnxt_dev_uninit(struct rte_eth_dev *eth_dev)
                bnxt_dev_close_op(eth_dev);
        if (bp->pf.vf_info)
                rte_free(bp->pf.vf_info);
-       bnxt_free_ctx_mem(bp);
        eth_dev->dev_ops = NULL;
        eth_dev->rx_pkt_burst = NULL;
        eth_dev->tx_pkt_burst = NULL;