From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 116CFA00BE; Fri, 12 Jun 2020 15:34:37 +0200 (CEST) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 0CC5D1BECC; Fri, 12 Jun 2020 15:34:10 +0200 (CEST) Received: from relay.smtp.broadcom.com (unknown [192.19.232.149]) by dpdk.org (Postfix) with ESMTP id AA0A71BEB7 for ; Fri, 12 Jun 2020 15:34:06 +0200 (CEST) Received: from dhcp-10-123-153-55.dhcp.broadcom.net (dhcp-10-123-153-55.dhcp.broadcom.net [10.123.153.55]) by relay.smtp.broadcom.com (Postfix) with ESMTP id 1F5FA1BD7E1; Fri, 12 Jun 2020 06:34:04 -0700 (PDT) DKIM-Filter: OpenDKIM Filter v2.10.3 relay.smtp.broadcom.com 1F5FA1BD7E1 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=broadcom.com; s=dkimrelay; t=1591968846; bh=S5lWBQX32K2JouY3XAi1VjxBTAk4tKqGVkow+eA0v/8=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=Xh0F8B4HbbZg2o0m9WMoO8v0oF55MBNt4UKWV11nC+4jSbzJvC5VAAhE9aiAfjlPo guz9Ey0lk4Z47+faPOTV7+g0h2c3xDq5O/wJCmuxK0M5b/jSJg2JEw10eLI0TCGjVD 4nTuzKFM2LViznF7LLGSQdLCeqL3bEgPgO/OhXZo= From: Somnath Kotur To: dev@dpdk.org Cc: ferruh.yigit@intel.com Date: Fri, 12 Jun 2020 18:58:46 +0530 Message-Id: <20200612132934.16488-3-somnath.kotur@broadcom.com> X-Mailer: git-send-email 2.10.1.613.g2cc2e70 In-Reply-To: <20200612132934.16488-1-somnath.kotur@broadcom.com> References: <20200612132934.16488-1-somnath.kotur@broadcom.com> Subject: [dpdk-dev] [PATCH 02/50] net/bnxt: Infrastructure support for VF-reps data path X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Added code to support Tx/Rx from a VF representor port. The VF-reps use the RX/TX rings of the Trusted VF/PF. For each VF-rep, the Trusted VF/PF driver issues a VFR_ALLOC FW cmd that returns "cfa_code" and "cfa_action" values. The FW sets up the filter tables in such a way that VF traffic by default (in absence of other rules) gets punted to the parent function i.e. either the Trusted VF or the PF. The cfa_code value in the RX-compl informs the driver of the source VF. For traffic being transmitted from the VF-rep, the TX BD is tagged with a cfa_action value that informs the HW to punt it to the corresponding VF. Reviewed-by: Venkat Duvvuru Reviewed-by: Ajit Kumar Khaparde Signed-off-by: Somnath Kotur Signed-off-by: Venkat Duvvuru --- drivers/net/bnxt/bnxt.h | 30 ++- drivers/net/bnxt/bnxt_ethdev.c | 150 +++++++++---- drivers/net/bnxt/bnxt_reps.c | 476 ++++++++++++++++++++++++++++++++++++++--- drivers/net/bnxt/bnxt_reps.h | 11 + drivers/net/bnxt/bnxt_rxr.c | 22 +- drivers/net/bnxt/bnxt_rxr.h | 1 + drivers/net/bnxt/bnxt_txq.h | 1 + drivers/net/bnxt/bnxt_txr.c | 4 +- 8 files changed, 616 insertions(+), 79 deletions(-) diff --git a/drivers/net/bnxt/bnxt.h b/drivers/net/bnxt/bnxt.h index 9b7b87c..443d9fe 100644 --- a/drivers/net/bnxt/bnxt.h +++ b/drivers/net/bnxt/bnxt.h @@ -495,6 +495,7 @@ struct bnxt_mark_info { struct bnxt_rep_info { struct rte_eth_dev *vfr_eth_dev; + pthread_mutex_t vfr_lock; }; /* address space location of register */ @@ -755,7 +756,8 @@ struct bnxt { uint16_t fw_reset_max_msecs; uint16_t switch_domain_id; uint16_t num_reps; - struct bnxt_rep_info rep_info[BNXT_MAX_VF_REPS]; + struct bnxt_rep_info *rep_info; + uint16_t *cfa_code_map; /* Struct to hold adapter error recovery related info */ struct bnxt_error_recovery_info *recovery_info; #define BNXT_MARK_TABLE_SZ (sizeof(struct bnxt_mark_info) * 64 * 1024) @@ -780,12 +782,28 @@ struct bnxt { * Structure to store private data for each VF representor instance */ struct bnxt_vf_representor { - uint16_t switch_domain_id; - uint16_t vf_id; + uint16_t switch_domain_id; + uint16_t vf_id; + uint16_t tx_cfa_action; + uint16_t rx_cfa_code; /* Private data store of associated PF/Trusted VF */ - struct bnxt *parent_priv; - uint8_t mac_addr[RTE_ETHER_ADDR_LEN]; - uint8_t dflt_mac_addr[RTE_ETHER_ADDR_LEN]; + struct rte_eth_dev *parent_dev; + uint8_t mac_addr[RTE_ETHER_ADDR_LEN]; + uint8_t dflt_mac_addr[RTE_ETHER_ADDR_LEN]; + struct bnxt_rx_queue **rx_queues; + unsigned int rx_nr_rings; + unsigned int tx_nr_rings; + uint64_t tx_pkts[BNXT_MAX_VF_REP_RINGS]; + uint64_t tx_bytes[BNXT_MAX_VF_REP_RINGS]; + uint64_t rx_pkts[BNXT_MAX_VF_REP_RINGS]; + uint64_t rx_bytes[BNXT_MAX_VF_REP_RINGS]; + uint64_t rx_drop_pkts[BNXT_MAX_VF_REP_RINGS]; + uint64_t rx_drop_bytes[BNXT_MAX_VF_REP_RINGS]; +}; + +struct bnxt_vf_rep_tx_queue { + struct bnxt_tx_queue *txq; + struct bnxt_vf_representor *bp; }; int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu); diff --git a/drivers/net/bnxt/bnxt_ethdev.c b/drivers/net/bnxt/bnxt_ethdev.c index 4911745..4202904 100644 --- a/drivers/net/bnxt/bnxt_ethdev.c +++ b/drivers/net/bnxt/bnxt_ethdev.c @@ -137,6 +137,7 @@ static void bnxt_cancel_fw_health_check(struct bnxt *bp); static int bnxt_restore_vlan_filters(struct bnxt *bp); static void bnxt_dev_recover(void *arg); static void bnxt_free_error_recovery_info(struct bnxt *bp); +static void bnxt_free_rep_info(struct bnxt *bp); int is_bnxt_in_error(struct bnxt *bp) { @@ -5243,7 +5244,7 @@ bnxt_init_locks(struct bnxt *bp) static int bnxt_init_resources(struct bnxt *bp, bool reconfig_dev) { - int rc; + int rc = 0; rc = bnxt_init_fw(bp); if (rc) @@ -5642,6 +5643,8 @@ bnxt_uninit_locks(struct bnxt *bp) { pthread_mutex_destroy(&bp->flow_lock); pthread_mutex_destroy(&bp->def_cp_lock); + if (bp->rep_info) + pthread_mutex_destroy(&bp->rep_info->vfr_lock); } static int @@ -5664,6 +5667,7 @@ bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev) bnxt_uninit_locks(bp); bnxt_free_flow_stats_info(bp); + bnxt_free_rep_info(bp); rte_free(bp->ptp_cfg); bp->ptp_cfg = NULL; return rc; @@ -5703,56 +5707,73 @@ static int bnxt_pci_remove_dev_with_reps(struct rte_eth_dev *eth_dev) return ret; } -static int bnxt_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, - struct rte_pci_device *pci_dev) +static void bnxt_free_rep_info(struct bnxt *bp) { - char name[RTE_ETH_NAME_MAX_LEN]; - struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 }; - struct rte_eth_dev *backing_eth_dev, *vf_rep_eth_dev; - uint16_t num_rep; - int i, ret = 0; - struct bnxt *backing_bp; + rte_free(bp->rep_info); + bp->rep_info = NULL; + rte_free(bp->cfa_code_map); + bp->cfa_code_map = NULL; +} - if (pci_dev->device.devargs) { - ret = rte_eth_devargs_parse(pci_dev->device.devargs->args, - ð_da); - if (ret) - return ret; - } +static int bnxt_init_rep_info(struct bnxt *bp) +{ + int i = 0, rc; - num_rep = eth_da.nb_representor_ports; - PMD_DRV_LOG(DEBUG, "nb_representor_ports = %d\n", - num_rep); + if (bp->rep_info) + return 0; - /* We could come here after first level of probe is already invoked - * as part of an application bringup(OVS-DPDK vswitchd), so first check - * for already allocated eth_dev for the backing device (PF/Trusted VF) - */ - backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name); - if (backing_eth_dev == NULL) { - ret = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name, - sizeof(struct bnxt), - eth_dev_pci_specific_init, pci_dev, - bnxt_dev_init, NULL); + bp->rep_info = rte_zmalloc("bnxt_rep_info", + sizeof(bp->rep_info[0]) * BNXT_MAX_VF_REPS, + 0); + if (!bp->rep_info) { + PMD_DRV_LOG(ERR, "Failed to alloc memory for rep info\n"); + return -ENOMEM; + } + bp->cfa_code_map = rte_zmalloc("bnxt_cfa_code_map", + sizeof(*bp->cfa_code_map) * + BNXT_MAX_CFA_CODE, 0); + if (!bp->cfa_code_map) { + PMD_DRV_LOG(ERR, "Failed to alloc memory for cfa_code_map\n"); + bnxt_free_rep_info(bp); + return -ENOMEM; + } - if (ret || !num_rep) - return ret; + for (i = 0; i < BNXT_MAX_CFA_CODE; i++) + bp->cfa_code_map[i] = BNXT_VF_IDX_INVALID; + + rc = pthread_mutex_init(&bp->rep_info->vfr_lock, NULL); + if (rc) { + PMD_DRV_LOG(ERR, "Unable to initialize vfr_lock\n"); + bnxt_free_rep_info(bp); + return rc; } + return rc; +} + +static int bnxt_rep_port_probe(struct rte_pci_device *pci_dev, + struct rte_eth_devargs eth_da, + struct rte_eth_dev *backing_eth_dev) +{ + struct rte_eth_dev *vf_rep_eth_dev; + char name[RTE_ETH_NAME_MAX_LEN]; + struct bnxt *backing_bp; + uint16_t num_rep; + int i, ret = 0; + num_rep = eth_da.nb_representor_ports; if (num_rep > BNXT_MAX_VF_REPS) { PMD_DRV_LOG(ERR, "nb_representor_ports = %d > %d MAX VF REPS\n", - eth_da.nb_representor_ports, BNXT_MAX_VF_REPS); - ret = -EINVAL; - return ret; + num_rep, BNXT_MAX_VF_REPS); + return -EINVAL; } - /* probe representor ports now */ - if (!backing_eth_dev) - backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name); - if (backing_eth_dev == NULL) { - ret = -ENODEV; - return ret; + if (num_rep > RTE_MAX_ETHPORTS) { + PMD_DRV_LOG(ERR, + "nb_representor_ports = %d > %d MAX ETHPORTS\n", + num_rep, RTE_MAX_ETHPORTS); + return -EINVAL; } + backing_bp = backing_eth_dev->data->dev_private; if (!(BNXT_PF(backing_bp) || BNXT_VF_IS_TRUSTED(backing_bp))) { @@ -5761,14 +5782,17 @@ static int bnxt_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, /* Returning an error is not an option. * Applications are not handling this correctly */ - return ret; + return 0; } - for (i = 0; i < eth_da.nb_representor_ports; i++) { + if (bnxt_init_rep_info(backing_bp)) + return 0; + + for (i = 0; i < num_rep; i++) { struct bnxt_vf_representor representor = { .vf_id = eth_da.representor_ports[i], .switch_domain_id = backing_bp->switch_domain_id, - .parent_priv = backing_bp + .parent_dev = backing_eth_dev }; if (representor.vf_id >= BNXT_MAX_VF_REPS) { @@ -5809,6 +5833,48 @@ static int bnxt_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, return ret; } +static int bnxt_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, + struct rte_pci_device *pci_dev) +{ + struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 }; + struct rte_eth_dev *backing_eth_dev; + uint16_t num_rep; + int ret = 0; + + if (pci_dev->device.devargs) { + ret = rte_eth_devargs_parse(pci_dev->device.devargs->args, + ð_da); + if (ret) + return ret; + } + + num_rep = eth_da.nb_representor_ports; + PMD_DRV_LOG(DEBUG, "nb_representor_ports = %d\n", + num_rep); + + /* We could come here after first level of probe is already invoked + * as part of an application bringup(OVS-DPDK vswitchd), so first check + * for already allocated eth_dev for the backing device (PF/Trusted VF) + */ + backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name); + if (backing_eth_dev == NULL) { + ret = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name, + sizeof(struct bnxt), + eth_dev_pci_specific_init, pci_dev, + bnxt_dev_init, NULL); + + if (ret || !num_rep) + return ret; + + backing_eth_dev = rte_eth_dev_allocated(pci_dev->device.name); + } + + /* probe representor ports now */ + ret = bnxt_rep_port_probe(pci_dev, eth_da, backing_eth_dev); + + return ret; +} + static int bnxt_pci_remove(struct rte_pci_device *pci_dev) { struct rte_eth_dev *eth_dev; diff --git a/drivers/net/bnxt/bnxt_reps.c b/drivers/net/bnxt/bnxt_reps.c index 7033d62..39f3d23 100644 --- a/drivers/net/bnxt/bnxt_reps.c +++ b/drivers/net/bnxt/bnxt_reps.c @@ -6,6 +6,11 @@ #include "bnxt.h" #include "bnxt_ring.h" #include "bnxt_reps.h" +#include "bnxt_rxq.h" +#include "bnxt_rxr.h" +#include "bnxt_txq.h" +#include "bnxt_txr.h" +#include "bnxt_hwrm.h" #include "hsi_struct_def_dpdk.h" static const struct eth_dev_ops bnxt_vf_rep_dev_ops = { @@ -13,25 +18,128 @@ static const struct eth_dev_ops bnxt_vf_rep_dev_ops = { .dev_configure = bnxt_vf_rep_dev_configure_op, .dev_start = bnxt_vf_rep_dev_start_op, .rx_queue_setup = bnxt_vf_rep_rx_queue_setup_op, + .rx_queue_release = bnxt_vf_rep_rx_queue_release_op, .tx_queue_setup = bnxt_vf_rep_tx_queue_setup_op, + .tx_queue_release = bnxt_vf_rep_tx_queue_release_op, .link_update = bnxt_vf_rep_link_update_op, .dev_close = bnxt_vf_rep_dev_close_op, - .dev_stop = bnxt_vf_rep_dev_stop_op + .dev_stop = bnxt_vf_rep_dev_stop_op, + .stats_get = bnxt_vf_rep_stats_get_op, + .stats_reset = bnxt_vf_rep_stats_reset_op, }; -static uint16_t -bnxt_vf_rep_rx_burst(__rte_unused void *rx_queue, - __rte_unused struct rte_mbuf **rx_pkts, - __rte_unused uint16_t nb_pkts) +uint16_t +bnxt_vfr_recv(struct bnxt *bp, uint16_t cfa_code, uint16_t queue_id, + struct rte_mbuf *mbuf) { + struct bnxt_sw_rx_bd *prod_rx_buf; + struct bnxt_rx_ring_info *rep_rxr; + struct bnxt_rx_queue *rep_rxq; + struct rte_eth_dev *vfr_eth_dev; + struct bnxt_vf_representor *vfr_bp; + uint16_t vf_id; + uint16_t mask; + uint8_t que; + + vf_id = bp->cfa_code_map[cfa_code]; + /* cfa_code is invalid OR vf_id > MAX REP. Assume normal Rx */ + if (vf_id == BNXT_VF_IDX_INVALID || vf_id > BNXT_MAX_VF_REPS) + return 1; + vfr_eth_dev = bp->rep_info[vf_id].vfr_eth_dev; + if (!vfr_eth_dev) + return 1; + vfr_bp = vfr_eth_dev->data->dev_private; + if (vfr_bp->rx_cfa_code != cfa_code) { + /* cfa_code not meant for this VF rep!!?? */ + return 1; + } + /* If rxq_id happens to be > max rep_queue, use rxq0 */ + que = queue_id < BNXT_MAX_VF_REP_RINGS ? queue_id : 0; + rep_rxq = vfr_bp->rx_queues[que]; + rep_rxr = rep_rxq->rx_ring; + mask = rep_rxr->rx_ring_struct->ring_mask; + + /* Put this mbuf on the RxQ of the Representor */ + prod_rx_buf = + &rep_rxr->rx_buf_ring[rep_rxr->rx_prod++ & mask]; + if (!prod_rx_buf->mbuf) { + prod_rx_buf->mbuf = mbuf; + vfr_bp->rx_bytes[que] += mbuf->pkt_len; + vfr_bp->rx_pkts[que]++; + } else { + vfr_bp->rx_drop_bytes[que] += mbuf->pkt_len; + vfr_bp->rx_drop_pkts[que]++; + rte_free(mbuf); /* Representor Rx ring full, drop pkt */ + } + return 0; } static uint16_t -bnxt_vf_rep_tx_burst(__rte_unused void *tx_queue, - __rte_unused struct rte_mbuf **tx_pkts, +bnxt_vf_rep_rx_burst(void *rx_queue, + struct rte_mbuf **rx_pkts, + uint16_t nb_pkts) +{ + struct bnxt_rx_queue *rxq = rx_queue; + struct bnxt_sw_rx_bd *cons_rx_buf; + struct bnxt_rx_ring_info *rxr; + uint16_t nb_rx_pkts = 0; + uint16_t mask, i; + + if (!rxq) + return 0; + + rxr = rxq->rx_ring; + mask = rxr->rx_ring_struct->ring_mask; + for (i = 0; i < nb_pkts; i++) { + cons_rx_buf = &rxr->rx_buf_ring[rxr->rx_cons & mask]; + if (!cons_rx_buf->mbuf) + return nb_rx_pkts; + rx_pkts[nb_rx_pkts] = cons_rx_buf->mbuf; + rx_pkts[nb_rx_pkts]->port = rxq->port_id; + cons_rx_buf->mbuf = NULL; + nb_rx_pkts++; + rxr->rx_cons++; + } + + return nb_rx_pkts; +} + +static uint16_t +bnxt_vf_rep_tx_burst(void *tx_queue, + struct rte_mbuf **tx_pkts, __rte_unused uint16_t nb_pkts) { + struct bnxt_vf_rep_tx_queue *vfr_txq = tx_queue; + struct bnxt_tx_queue *ptxq; + struct bnxt *parent; + struct bnxt_vf_representor *vf_rep_bp; + int qid; + int rc; + int i; + + if (!vfr_txq) + return 0; + + qid = vfr_txq->txq->queue_id; + vf_rep_bp = vfr_txq->bp; + parent = vf_rep_bp->parent_dev->data->dev_private; + pthread_mutex_lock(&parent->rep_info->vfr_lock); + ptxq = parent->tx_queues[qid]; + + ptxq->tx_cfa_action = vf_rep_bp->tx_cfa_action; + + for (i = 0; i < nb_pkts; i++) { + vf_rep_bp->tx_bytes[qid] += tx_pkts[i]->pkt_len; + vf_rep_bp->tx_pkts[qid]++; + } + + rc = bnxt_xmit_pkts(ptxq, tx_pkts, nb_pkts); + ptxq->tx_cfa_action = 0; + pthread_mutex_unlock(&parent->rep_info->vfr_lock); + + return rc; + return 0; } @@ -45,7 +153,7 @@ int bnxt_vf_representor_init(struct rte_eth_dev *eth_dev, void *params) vf_rep_bp->vf_id = rep_params->vf_id; vf_rep_bp->switch_domain_id = rep_params->switch_domain_id; - vf_rep_bp->parent_priv = rep_params->parent_priv; + vf_rep_bp->parent_dev = rep_params->parent_dev; eth_dev->data->dev_flags |= RTE_ETH_DEV_REPRESENTOR; eth_dev->data->representor_id = rep_params->vf_id; @@ -63,7 +171,7 @@ int bnxt_vf_representor_init(struct rte_eth_dev *eth_dev, void *params) eth_dev->rx_pkt_burst = bnxt_vf_rep_rx_burst; eth_dev->tx_pkt_burst = bnxt_vf_rep_tx_burst; /* Link state. Inherited from PF or trusted VF */ - parent_bp = vf_rep_bp->parent_priv; + parent_bp = vf_rep_bp->parent_dev->data->dev_private; link = &parent_bp->eth_dev->data->dev_link; eth_dev->data->dev_link.link_speed = link->link_speed; @@ -94,18 +202,18 @@ int bnxt_vf_representor_uninit(struct rte_eth_dev *eth_dev) uint16_t vf_id; eth_dev->data->mac_addrs = NULL; - - parent_bp = rep->parent_priv; - if (parent_bp) { - parent_bp->num_reps--; - vf_id = rep->vf_id; - if (parent_bp->rep_info) { - memset(&parent_bp->rep_info[vf_id], 0, - sizeof(parent_bp->rep_info[vf_id])); - /* mark that this representor has been freed */ - } - } eth_dev->dev_ops = NULL; + + parent_bp = rep->parent_dev->data->dev_private; + if (!parent_bp) + return 0; + + parent_bp->num_reps--; + vf_id = rep->vf_id; + if (parent_bp->rep_info) + memset(&parent_bp->rep_info[vf_id], 0, + sizeof(parent_bp->rep_info[vf_id])); + /* mark that this representor has been freed */ return 0; } @@ -117,7 +225,7 @@ int bnxt_vf_rep_link_update_op(struct rte_eth_dev *eth_dev, int wait_to_compl) struct rte_eth_link *link; int rc; - parent_bp = rep->parent_priv; + parent_bp = rep->parent_dev->data->dev_private; rc = bnxt_link_update_op(parent_bp->eth_dev, wait_to_compl); /* Link state. Inherited from PF or trusted VF */ @@ -132,16 +240,134 @@ int bnxt_vf_rep_link_update_op(struct rte_eth_dev *eth_dev, int wait_to_compl) return rc; } +static int bnxt_vfr_alloc(struct bnxt_vf_representor *vfr) +{ + int rc = 0; + struct bnxt *parent_bp; + + if (!vfr || !vfr->parent_dev) { + PMD_DRV_LOG(ERR, + "No memory allocated for representor\n"); + return -ENOMEM; + } + + parent_bp = vfr->parent_dev->data->dev_private; + + /* Check if representor has been already allocated in FW */ + if (vfr->tx_cfa_action && vfr->rx_cfa_code) + return 0; + + /* + * Alloc VF rep rules in CFA after default VNIC is created. + * Otherwise the FW will create the VF-rep rules with + * default drop action. + */ + + /* + * This is where we need to replace invoking an HWRM cmd + * with the new TFLIB ULP API to do more/less the same job + rc = bnxt_hwrm_cfa_vfr_alloc(parent_bp, + vfr->vf_id, + &vfr->tx_cfa_action, + &vfr->rx_cfa_code); + */ + if (!rc) { + parent_bp->cfa_code_map[vfr->rx_cfa_code] = vfr->vf_id; + PMD_DRV_LOG(DEBUG, "allocated representor %d in FW\n", + vfr->vf_id); + } else { + PMD_DRV_LOG(ERR, + "Failed to alloc representor %d in FW\n", + vfr->vf_id); + } + + return rc; +} + +static void bnxt_vf_rep_free_rx_mbufs(struct bnxt_vf_representor *rep_bp) +{ + struct bnxt_rx_queue *rxq; + unsigned int i; + + for (i = 0; i < rep_bp->rx_nr_rings; i++) { + rxq = rep_bp->rx_queues[i]; + bnxt_rx_queue_release_mbufs(rxq); + } +} + int bnxt_vf_rep_dev_start_op(struct rte_eth_dev *eth_dev) { - bnxt_vf_rep_link_update_op(eth_dev, 1); + struct bnxt_vf_representor *rep_bp = eth_dev->data->dev_private; + int rc; - return 0; + rc = bnxt_vfr_alloc(rep_bp); + + if (!rc) { + eth_dev->rx_pkt_burst = &bnxt_vf_rep_rx_burst; + eth_dev->tx_pkt_burst = &bnxt_vf_rep_tx_burst; + + bnxt_vf_rep_link_update_op(eth_dev, 1); + } else { + eth_dev->data->dev_link.link_status = 0; + bnxt_vf_rep_free_rx_mbufs(rep_bp); + } + + return rc; +} + +static int bnxt_vfr_free(struct bnxt_vf_representor *vfr) +{ + int rc = 0; + struct bnxt *parent_bp; + + if (!vfr || !vfr->parent_dev) { + PMD_DRV_LOG(ERR, + "No memory allocated for representor\n"); + return -ENOMEM; + } + + parent_bp = vfr->parent_dev->data->dev_private; + + /* Check if representor has been already freed in FW */ + if (!vfr->tx_cfa_action && !vfr->rx_cfa_code) + return 0; + + /* + * This is where we need to replace invoking an HWRM cmd + * with the new TFLIB ULP API to do more/less the same job + rc = bnxt_hwrm_cfa_vfr_free(parent_bp, + vfr->vf_id); + */ + if (rc) { + PMD_DRV_LOG(ERR, + "Failed to free representor %d in FW\n", + vfr->vf_id); + return rc; + } + + parent_bp->cfa_code_map[vfr->rx_cfa_code] = BNXT_VF_IDX_INVALID; + PMD_DRV_LOG(DEBUG, "freed representor %d in FW\n", + vfr->vf_id); + vfr->tx_cfa_action = 0; + vfr->rx_cfa_code = 0; + + return rc; } void bnxt_vf_rep_dev_stop_op(struct rte_eth_dev *eth_dev) { - eth_dev = eth_dev; + struct bnxt_vf_representor *vfr_bp = eth_dev->data->dev_private; + + /* Avoid crashes as we are about to free queues */ + eth_dev->rx_pkt_burst = &bnxt_dummy_recv_pkts; + eth_dev->tx_pkt_burst = &bnxt_dummy_xmit_pkts; + + bnxt_vfr_free(vfr_bp); + + if (eth_dev->data->dev_started) + eth_dev->data->dev_link.link_status = 0; + + bnxt_vf_rep_free_rx_mbufs(vfr_bp); } void bnxt_vf_rep_dev_close_op(struct rte_eth_dev *eth_dev) @@ -159,7 +385,7 @@ int bnxt_vf_rep_dev_info_get_op(struct rte_eth_dev *eth_dev, int rc = 0; /* MAC Specifics */ - parent_bp = rep_bp->parent_priv; + parent_bp = rep_bp->parent_dev->data->dev_private; if (!parent_bp) { PMD_DRV_LOG(ERR, "Rep parent NULL!\n"); return rc; @@ -257,7 +483,13 @@ int bnxt_vf_rep_dev_info_get_op(struct rte_eth_dev *eth_dev, int bnxt_vf_rep_dev_configure_op(__rte_unused struct rte_eth_dev *eth_dev) { + struct bnxt_vf_representor *rep_bp = eth_dev->data->dev_private; + PMD_DRV_LOG(DEBUG, "Representor dev_configure_op\n"); + rep_bp->rx_queues = (void *)eth_dev->data->rx_queues; + rep_bp->tx_nr_rings = eth_dev->data->nb_tx_queues; + rep_bp->rx_nr_rings = eth_dev->data->nb_rx_queues; + return 0; } @@ -269,9 +501,94 @@ int bnxt_vf_rep_rx_queue_setup_op(struct rte_eth_dev *eth_dev, rx_conf, __rte_unused struct rte_mempool *mp) { - eth_dev = eth_dev; + struct bnxt_vf_representor *rep_bp = eth_dev->data->dev_private; + struct bnxt *parent_bp = rep_bp->parent_dev->data->dev_private; + struct bnxt_rx_queue *parent_rxq; + struct bnxt_rx_queue *rxq; + struct bnxt_sw_rx_bd *buf_ring; + int rc = 0; + + if (queue_idx >= BNXT_MAX_VF_REP_RINGS) { + PMD_DRV_LOG(ERR, + "Cannot create Rx ring %d. %d rings available\n", + queue_idx, BNXT_MAX_VF_REP_RINGS); + return -EINVAL; + } + + if (!nb_desc || nb_desc > MAX_RX_DESC_CNT) { + PMD_DRV_LOG(ERR, "nb_desc %d is invalid\n", nb_desc); + return -EINVAL; + } + + parent_rxq = parent_bp->rx_queues[queue_idx]; + if (!parent_rxq) { + PMD_DRV_LOG(ERR, "Parent RxQ has not been configured yet\n"); + return -EINVAL; + } + + if (nb_desc != parent_rxq->nb_rx_desc) { + PMD_DRV_LOG(ERR, "nb_desc %d do not match parent rxq", nb_desc); + return -EINVAL; + } + + if (eth_dev->data->rx_queues) { + rxq = eth_dev->data->rx_queues[queue_idx]; + if (rxq) + bnxt_rx_queue_release_op(rxq); + } + + rxq = rte_zmalloc_socket("bnxt_vfr_rx_queue", + sizeof(struct bnxt_rx_queue), + RTE_CACHE_LINE_SIZE, socket_id); + if (!rxq) { + PMD_DRV_LOG(ERR, "bnxt_vfr_rx_queue allocation failed!\n"); + return -ENOMEM; + } + + rxq->nb_rx_desc = nb_desc; + + rc = bnxt_init_rx_ring_struct(rxq, socket_id); + if (rc) + goto out; + + buf_ring = rte_zmalloc_socket("bnxt_rx_vfr_buf_ring", + sizeof(struct bnxt_sw_rx_bd) * + rxq->rx_ring->rx_ring_struct->ring_size, + RTE_CACHE_LINE_SIZE, socket_id); + if (!buf_ring) { + PMD_DRV_LOG(ERR, "bnxt_rx_vfr_buf_ring allocation failed!\n"); + rc = -ENOMEM; + goto out; + } + + rxq->rx_ring->rx_buf_ring = buf_ring; + rxq->queue_id = queue_idx; + rxq->port_id = eth_dev->data->port_id; + eth_dev->data->rx_queues[queue_idx] = rxq; return 0; + +out: + if (rxq) + bnxt_rx_queue_release_op(rxq); + + return rc; +} + +void bnxt_vf_rep_rx_queue_release_op(void *rx_queue) +{ + struct bnxt_rx_queue *rxq = (struct bnxt_rx_queue *)rx_queue; + + if (!rxq) + return; + + bnxt_rx_queue_release_mbufs(rxq); + + bnxt_free_ring(rxq->rx_ring->rx_ring_struct); + bnxt_free_ring(rxq->rx_ring->ag_ring_struct); + bnxt_free_ring(rxq->cp_ring->cp_ring_struct); + + rte_free(rxq); } int bnxt_vf_rep_tx_queue_setup_op(struct rte_eth_dev *eth_dev, @@ -281,7 +598,112 @@ int bnxt_vf_rep_tx_queue_setup_op(struct rte_eth_dev *eth_dev, __rte_unused const struct rte_eth_txconf * tx_conf) { - eth_dev = eth_dev; + struct bnxt_vf_representor *rep_bp = eth_dev->data->dev_private; + struct bnxt *parent_bp = rep_bp->parent_dev->data->dev_private; + struct bnxt_tx_queue *parent_txq, *txq; + struct bnxt_vf_rep_tx_queue *vfr_txq; + + if (queue_idx >= BNXT_MAX_VF_REP_RINGS) { + PMD_DRV_LOG(ERR, + "Cannot create Tx rings %d. %d rings available\n", + queue_idx, BNXT_MAX_VF_REP_RINGS); + return -EINVAL; + } + + if (!nb_desc || nb_desc > MAX_TX_DESC_CNT) { + PMD_DRV_LOG(ERR, "nb_desc %d is invalid", nb_desc); + return -EINVAL; + } + + parent_txq = parent_bp->tx_queues[queue_idx]; + if (!parent_txq) { + PMD_DRV_LOG(ERR, "Parent TxQ has not been configured yet\n"); + return -EINVAL; + } + if (nb_desc != parent_txq->nb_tx_desc) { + PMD_DRV_LOG(ERR, "nb_desc %d do not match parent txq", nb_desc); + return -EINVAL; + } + + if (eth_dev->data->tx_queues) { + vfr_txq = eth_dev->data->tx_queues[queue_idx]; + bnxt_vf_rep_tx_queue_release_op(vfr_txq); + vfr_txq = NULL; + } + + vfr_txq = rte_zmalloc_socket("bnxt_vfr_tx_queue", + sizeof(struct bnxt_vf_rep_tx_queue), + RTE_CACHE_LINE_SIZE, socket_id); + if (!vfr_txq) { + PMD_DRV_LOG(ERR, "bnxt_vfr_tx_queue allocation failed!"); + return -ENOMEM; + } + txq = rte_zmalloc_socket("bnxt_tx_queue", + sizeof(struct bnxt_tx_queue), + RTE_CACHE_LINE_SIZE, socket_id); + if (!txq) { + PMD_DRV_LOG(ERR, "bnxt_tx_queue allocation failed!"); + rte_free(vfr_txq); + return -ENOMEM; + } + + txq->nb_tx_desc = nb_desc; + txq->queue_id = queue_idx; + txq->port_id = eth_dev->data->port_id; + vfr_txq->txq = txq; + vfr_txq->bp = rep_bp; + eth_dev->data->tx_queues[queue_idx] = vfr_txq; + + return 0; +} + +void bnxt_vf_rep_tx_queue_release_op(void *tx_queue) +{ + struct bnxt_vf_rep_tx_queue *vfr_txq = tx_queue; + + if (!vfr_txq) + return; + + rte_free(vfr_txq->txq); + rte_free(vfr_txq); +} + +int bnxt_vf_rep_stats_get_op(struct rte_eth_dev *eth_dev, + struct rte_eth_stats *stats) +{ + struct bnxt_vf_representor *rep_bp = eth_dev->data->dev_private; + int i; + + memset(stats, 0, sizeof(*stats)); + for (i = 0; i < BNXT_MAX_VF_REP_RINGS; i++) { + stats->obytes += rep_bp->tx_bytes[i]; + stats->opackets += rep_bp->tx_pkts[i]; + stats->ibytes += rep_bp->rx_bytes[i]; + stats->ipackets += rep_bp->rx_pkts[i]; + stats->imissed += rep_bp->rx_drop_pkts[i]; + + stats->q_ipackets[i] = rep_bp->rx_pkts[i]; + stats->q_ibytes[i] = rep_bp->rx_bytes[i]; + stats->q_opackets[i] = rep_bp->tx_pkts[i]; + stats->q_obytes[i] = rep_bp->tx_bytes[i]; + stats->q_errors[i] = rep_bp->rx_drop_pkts[i]; + } + + return 0; +} + +int bnxt_vf_rep_stats_reset_op(struct rte_eth_dev *eth_dev) +{ + struct bnxt_vf_representor *rep_bp = eth_dev->data->dev_private; + int i; + + for (i = 0; i < BNXT_MAX_VF_REP_RINGS; i++) { + rep_bp->tx_pkts[i] = 0; + rep_bp->tx_bytes[i] = 0; + rep_bp->rx_pkts[i] = 0; + rep_bp->rx_bytes[i] = 0; + rep_bp->rx_drop_pkts[i] = 0; + } return 0; } diff --git a/drivers/net/bnxt/bnxt_reps.h b/drivers/net/bnxt/bnxt_reps.h index f4c033a..c8a3c7d 100644 --- a/drivers/net/bnxt/bnxt_reps.h +++ b/drivers/net/bnxt/bnxt_reps.h @@ -9,6 +9,12 @@ #include #include +#define BNXT_MAX_CFA_CODE 65536 +#define BNXT_VF_IDX_INVALID 0xffff + +uint16_t +bnxt_vfr_recv(struct bnxt *bp, uint16_t cfa_code, uint16_t queue_id, + struct rte_mbuf *mbuf); int bnxt_vf_representor_init(struct rte_eth_dev *eth_dev, void *params); int bnxt_vf_representor_uninit(struct rte_eth_dev *eth_dev); int bnxt_vf_rep_dev_info_get_op(struct rte_eth_dev *eth_dev, @@ -30,6 +36,11 @@ int bnxt_vf_rep_tx_queue_setup_op(struct rte_eth_dev *eth_dev, __rte_unused unsigned int socket_id, __rte_unused const struct rte_eth_txconf * tx_conf); +void bnxt_vf_rep_rx_queue_release_op(void *rx_queue); +void bnxt_vf_rep_tx_queue_release_op(void *tx_queue); void bnxt_vf_rep_dev_stop_op(struct rte_eth_dev *eth_dev); void bnxt_vf_rep_dev_close_op(struct rte_eth_dev *eth_dev); +int bnxt_vf_rep_stats_get_op(struct rte_eth_dev *eth_dev, + struct rte_eth_stats *stats); +int bnxt_vf_rep_stats_reset_op(struct rte_eth_dev *eth_dev); #endif /* _BNXT_REPS_H_ */ diff --git a/drivers/net/bnxt/bnxt_rxr.c b/drivers/net/bnxt/bnxt_rxr.c index 3bcc624..0ecf199 100644 --- a/drivers/net/bnxt/bnxt_rxr.c +++ b/drivers/net/bnxt/bnxt_rxr.c @@ -12,6 +12,7 @@ #include #include "bnxt.h" +#include "bnxt_reps.h" #include "bnxt_ring.h" #include "bnxt_rxr.h" #include "bnxt_rxq.h" @@ -538,7 +539,7 @@ void bnxt_set_mark_in_mbuf(struct bnxt *bp, } static int bnxt_rx_pkt(struct rte_mbuf **rx_pkt, - struct bnxt_rx_queue *rxq, uint32_t *raw_cons) + struct bnxt_rx_queue *rxq, uint32_t *raw_cons) { struct bnxt_cp_ring_info *cpr = rxq->cp_ring; struct bnxt_rx_ring_info *rxr = rxq->rx_ring; @@ -734,6 +735,20 @@ static int bnxt_rx_pkt(struct rte_mbuf **rx_pkt, rx: *rx_pkt = mbuf; + if ((BNXT_VF_IS_TRUSTED(rxq->bp) || BNXT_PF(rxq->bp)) && + rxq->bp->cfa_code_map && rxcmp1->cfa_code) { + if (!bnxt_vfr_recv(rxq->bp, rxcmp1->cfa_code, rxq->queue_id, + mbuf)) { + /* Now return an error so that nb_rx_pkts is not + * incremented. + * This packet was meant to be given to the representor. + * So no need to account the packet and give it to + * parent Rx burst function. + */ + rc = -ENODEV; + } + } + next_rx: *raw_cons = tmp_raw_cons; @@ -750,6 +765,7 @@ uint16_t bnxt_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint32_t raw_cons = cpr->cp_raw_cons; uint32_t cons; int nb_rx_pkts = 0; + int nb_rep_rx_pkts = 0; struct rx_pkt_cmpl *rxcmp; uint16_t prod = rxr->rx_prod; uint16_t ag_prod = rxr->ag_prod; @@ -783,6 +799,8 @@ uint16_t bnxt_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, nb_rx_pkts++; if (rc == -EBUSY) /* partial completion */ break; + if (rc == -ENODEV) /* completion for representor */ + nb_rep_rx_pkts++; } else if (!BNXT_NUM_ASYNC_CPR(rxq->bp)) { evt = bnxt_event_hwrm_resp_handler(rxq->bp, @@ -801,7 +819,7 @@ uint16_t bnxt_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, } cpr->cp_raw_cons = raw_cons; - if (!nb_rx_pkts && !evt) { + if (!nb_rx_pkts && !nb_rep_rx_pkts && !evt) { /* * For PMD, there is no need to keep on pushing to REARM * the doorbell if there are no new completions diff --git a/drivers/net/bnxt/bnxt_rxr.h b/drivers/net/bnxt/bnxt_rxr.h index 811dcd8..e60c97f 100644 --- a/drivers/net/bnxt/bnxt_rxr.h +++ b/drivers/net/bnxt/bnxt_rxr.h @@ -188,6 +188,7 @@ struct bnxt_sw_rx_bd { struct bnxt_rx_ring_info { uint16_t rx_prod; uint16_t ag_prod; + uint16_t rx_cons; /* Needed for representor */ struct bnxt_db_info rx_db; struct bnxt_db_info ag_db; diff --git a/drivers/net/bnxt/bnxt_txq.h b/drivers/net/bnxt/bnxt_txq.h index 37a3f95..69ff89a 100644 --- a/drivers/net/bnxt/bnxt_txq.h +++ b/drivers/net/bnxt/bnxt_txq.h @@ -29,6 +29,7 @@ struct bnxt_tx_queue { struct bnxt *bp; int index; int tx_wake_thresh; + uint32_t tx_cfa_action; struct bnxt_tx_ring_info *tx_ring; unsigned int cp_nr_rings; diff --git a/drivers/net/bnxt/bnxt_txr.c b/drivers/net/bnxt/bnxt_txr.c index 1602140..d7e193d 100644 --- a/drivers/net/bnxt/bnxt_txr.c +++ b/drivers/net/bnxt/bnxt_txr.c @@ -131,7 +131,7 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt, PKT_TX_VLAN_PKT | PKT_TX_OUTER_IP_CKSUM | PKT_TX_TUNNEL_GRE | PKT_TX_TUNNEL_VXLAN | PKT_TX_TUNNEL_GENEVE | PKT_TX_IEEE1588_TMST | - PKT_TX_QINQ_PKT)) + PKT_TX_QINQ_PKT) || txq->tx_cfa_action) long_bd = true; nr_bds = long_bd + tx_pkt->nb_segs; @@ -184,7 +184,7 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt, if (long_bd) { txbd->flags_type |= TX_BD_LONG_TYPE_TX_BD_LONG; vlan_tag_flags = 0; - cfa_action = 0; + cfa_action = txq->tx_cfa_action; /* HW can accelerate only outer vlan in QinQ mode */ if (tx_buf->mbuf->ol_flags & PKT_TX_QINQ_PKT) { vlan_tag_flags = TX_BD_LONG_CFA_META_KEY_VLAN_TAG | -- 2.7.4