numam-dpdk/drivers/net/bnx2x/bnx2x_rxtx.c

512 lines
14 KiB
C
Raw Normal View History

/* SPDX-License-Identifier: BSD-3-Clause
* Copyright (c) 2013-2015 Brocade Communications Systems, Inc.
* Copyright (c) 2015-2018 Cavium Inc.
* All rights reserved.
* www.cavium.com
*/
#include "bnx2x.h"
#include "bnx2x_rxtx.h"
static const struct rte_memzone *
ring_dma_zone_reserve(struct rte_eth_dev *dev, const char *ring_name,
uint16_t queue_id, uint32_t ring_size, int socket_id)
{
return rte_eth_dma_zone_reserve(dev, ring_name, queue_id,
ring_size, BNX2X_PAGE_SIZE, socket_id);
}
static void
bnx2x_rx_queue_release(struct bnx2x_rx_queue *rx_queue)
{
uint16_t i;
struct rte_mbuf **sw_ring;
if (NULL != rx_queue) {
sw_ring = rx_queue->sw_ring;
if (NULL != sw_ring) {
for (i = 0; i < rx_queue->nb_rx_desc; i++) {
if (NULL != sw_ring[i])
rte_pktmbuf_free(sw_ring[i]);
}
rte_free(sw_ring);
}
rte_free(rx_queue);
}
}
void
bnx2x_dev_rx_queue_release(struct rte_eth_dev *dev, uint16_t queue_idx)
{
bnx2x_rx_queue_release(dev->data->rx_queues[queue_idx]);
}
int
bnx2x_dev_rx_queue_setup(struct rte_eth_dev *dev,
uint16_t queue_idx,
uint16_t nb_desc,
unsigned int socket_id,
__rte_unused const struct rte_eth_rxconf *rx_conf,
struct rte_mempool *mp)
{
uint16_t j, idx;
const struct rte_memzone *dma;
struct bnx2x_rx_queue *rxq;
uint32_t dma_size;
struct rte_mbuf *mbuf;
struct bnx2x_softc *sc = dev->data->dev_private;
struct bnx2x_fastpath *fp = &sc->fp[queue_idx];
struct eth_rx_cqe_next_page *nextpg;
rte_iova_t *rx_bd;
rte_iova_t busaddr;
/* First allocate the rx queue data structure */
rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct bnx2x_rx_queue),
RTE_CACHE_LINE_SIZE, socket_id);
if (NULL == rxq) {
PMD_DRV_LOG(ERR, sc, "rte_zmalloc for rxq failed!");
return -ENOMEM;
}
rxq->sc = sc;
rxq->mb_pool = mp;
rxq->queue_id = queue_idx;
rxq->port_id = dev->data->port_id;
rxq->nb_rx_pages = 1;
while (USABLE_RX_BD(rxq) < nb_desc)
rxq->nb_rx_pages <<= 1;
rxq->nb_rx_desc = TOTAL_RX_BD(rxq);
sc->rx_ring_size = USABLE_RX_BD(rxq);
rxq->nb_cq_pages = RCQ_BD_PAGES(rxq);
PMD_DRV_LOG(DEBUG, sc, "fp[%02d] req_bd=%u, usable_bd=%lu, "
"total_bd=%lu, rx_pages=%u, cq_pages=%u",
queue_idx, nb_desc, (unsigned long)USABLE_RX_BD(rxq),
(unsigned long)TOTAL_RX_BD(rxq), rxq->nb_rx_pages,
rxq->nb_cq_pages);
/* Allocate RX ring hardware descriptors */
dma_size = rxq->nb_rx_desc * sizeof(struct eth_rx_bd);
dma = ring_dma_zone_reserve(dev, "hw_ring", queue_idx, dma_size, socket_id);
if (NULL == dma) {
PMD_RX_LOG(ERR, "ring_dma_zone_reserve for rx_ring failed!");
bnx2x_rx_queue_release(rxq);
return -ENOMEM;
}
fp->rx_desc_mapping = rxq->rx_ring_phys_addr = (uint64_t)dma->iova;
rxq->rx_ring = (uint64_t*)dma->addr;
memset((void *)rxq->rx_ring, 0, dma_size);
/* Link the RX chain pages. */
for (j = 1; j <= rxq->nb_rx_pages; j++) {
rx_bd = &rxq->rx_ring[TOTAL_RX_BD_PER_PAGE * j - 2];
busaddr = rxq->rx_ring_phys_addr + BNX2X_PAGE_SIZE * (j % rxq->nb_rx_pages);
*rx_bd = busaddr;
}
/* Allocate software ring */
dma_size = rxq->nb_rx_desc * sizeof(struct bnx2x_rx_entry);
rxq->sw_ring = rte_zmalloc_socket("sw_ring", dma_size,
RTE_CACHE_LINE_SIZE,
socket_id);
if (NULL == rxq->sw_ring) {
PMD_RX_LOG(ERR, "rte_zmalloc for sw_ring failed!");
bnx2x_rx_queue_release(rxq);
return -ENOMEM;
}
/* Initialize software ring entries */
for (idx = 0; idx < rxq->nb_rx_desc; idx = NEXT_RX_BD(idx)) {
mbuf = rte_mbuf_raw_alloc(mp);
if (NULL == mbuf) {
PMD_RX_LOG(ERR, "RX mbuf alloc failed queue_id=%u, idx=%d",
(unsigned)rxq->queue_id, idx);
bnx2x_rx_queue_release(rxq);
return -ENOMEM;
}
rxq->sw_ring[idx] = mbuf;
rxq->rx_ring[idx] =
rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf));
}
rxq->pkt_first_seg = NULL;
rxq->pkt_last_seg = NULL;
rxq->rx_bd_head = 0;
rxq->rx_bd_tail = rxq->nb_rx_desc;
/* Allocate CQ chain. */
dma_size = BNX2X_RX_CHAIN_PAGE_SZ * rxq->nb_cq_pages;
dma = ring_dma_zone_reserve(dev, "bnx2x_rcq", queue_idx, dma_size, socket_id);
if (NULL == dma) {
PMD_RX_LOG(ERR, "RCQ alloc failed");
return -ENOMEM;
}
fp->rx_comp_mapping = rxq->cq_ring_phys_addr = (uint64_t)dma->iova;
rxq->cq_ring = (union eth_rx_cqe*)dma->addr;
/* Link the CQ chain pages. */
for (j = 1; j <= rxq->nb_cq_pages; j++) {
nextpg = &rxq->cq_ring[TOTAL_RCQ_ENTRIES_PER_PAGE * j - 1].next_page_cqe;
busaddr = rxq->cq_ring_phys_addr + BNX2X_PAGE_SIZE * (j % rxq->nb_cq_pages);
nextpg->addr_hi = rte_cpu_to_le_32(U64_HI(busaddr));
nextpg->addr_lo = rte_cpu_to_le_32(U64_LO(busaddr));
}
rxq->rx_cq_head = 0;
rxq->rx_cq_tail = TOTAL_RCQ_ENTRIES(rxq);
dev->data->rx_queues[queue_idx] = rxq;
if (!sc->rx_queues) sc->rx_queues = dev->data->rx_queues;
return 0;
}
static void
bnx2x_tx_queue_release(struct bnx2x_tx_queue *tx_queue)
{
uint16_t i;
struct rte_mbuf **sw_ring;
if (NULL != tx_queue) {
sw_ring = tx_queue->sw_ring;
if (NULL != sw_ring) {
for (i = 0; i < tx_queue->nb_tx_desc; i++) {
if (NULL != sw_ring[i])
rte_pktmbuf_free(sw_ring[i]);
}
rte_free(sw_ring);
}
rte_free(tx_queue);
}
}
void
bnx2x_dev_tx_queue_release(struct rte_eth_dev *dev, uint16_t queue_idx)
{
bnx2x_tx_queue_release(dev->data->tx_queues[queue_idx]);
}
static uint16_t
bnx2x_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
{
struct bnx2x_tx_queue *txq;
struct bnx2x_softc *sc;
struct bnx2x_fastpath *fp;
uint16_t nb_tx_pkts;
uint16_t nb_pkt_sent = 0;
uint32_t ret;
txq = p_txq;
sc = txq->sc;
fp = &sc->fp[txq->queue_id];
if ((unlikely((txq->nb_tx_desc - txq->nb_tx_avail) >
txq->tx_free_thresh)))
bnx2x_txeof(sc, fp);
nb_tx_pkts = RTE_MIN(nb_pkts, txq->nb_tx_avail / BDS_PER_TX_PKT);
if (unlikely(nb_tx_pkts == 0))
return 0;
while (nb_tx_pkts--) {
struct rte_mbuf *m = *tx_pkts++;
assert(m != NULL);
ret = bnx2x_tx_encap(txq, m);
fp->tx_db.data.prod += ret;
nb_pkt_sent++;
}
bnx2x_update_fp_sb_idx(fp);
mb();
DOORBELL(sc, txq->queue_id, fp->tx_db.raw);
mb();
if ((txq->nb_tx_desc - txq->nb_tx_avail) >
txq->tx_free_thresh)
bnx2x_txeof(sc, fp);
return nb_pkt_sent;
}
int
bnx2x_dev_tx_queue_setup(struct rte_eth_dev *dev,
uint16_t queue_idx,
uint16_t nb_desc,
unsigned int socket_id,
const struct rte_eth_txconf *tx_conf)
{
uint16_t i;
unsigned int tsize;
const struct rte_memzone *tz;
struct bnx2x_tx_queue *txq;
struct eth_tx_next_bd *tx_n_bd;
uint64_t busaddr;
struct bnx2x_softc *sc = dev->data->dev_private;
struct bnx2x_fastpath *fp = &sc->fp[queue_idx];
/* First allocate the tx queue data structure */
txq = rte_zmalloc("ethdev TX queue", sizeof(struct bnx2x_tx_queue),
RTE_CACHE_LINE_SIZE);
if (txq == NULL)
return -ENOMEM;
txq->sc = sc;
txq->nb_tx_pages = 1;
while (USABLE_TX_BD(txq) < nb_desc)
txq->nb_tx_pages <<= 1;
txq->nb_tx_desc = TOTAL_TX_BD(txq);
sc->tx_ring_size = TOTAL_TX_BD(txq);
txq->tx_free_thresh = tx_conf->tx_free_thresh ?
tx_conf->tx_free_thresh : DEFAULT_TX_FREE_THRESH;
txq->tx_free_thresh = min(txq->tx_free_thresh,
txq->nb_tx_desc - BDS_PER_TX_PKT);
PMD_DRV_LOG(DEBUG, sc, "fp[%02d] req_bd=%u, thresh=%u, usable_bd=%lu, "
"total_bd=%lu, tx_pages=%u",
queue_idx, nb_desc, txq->tx_free_thresh,
(unsigned long)USABLE_TX_BD(txq),
(unsigned long)TOTAL_TX_BD(txq), txq->nb_tx_pages);
/* Allocate TX ring hardware descriptors */
tsize = txq->nb_tx_desc * sizeof(union eth_tx_bd_types);
tz = ring_dma_zone_reserve(dev, "tx_hw_ring", queue_idx, tsize, socket_id);
if (tz == NULL) {
bnx2x_tx_queue_release(txq);
return -ENOMEM;
}
fp->tx_desc_mapping = txq->tx_ring_phys_addr = (uint64_t)tz->iova;
txq->tx_ring = (union eth_tx_bd_types *) tz->addr;
memset(txq->tx_ring, 0, tsize);
/* Allocate software ring */
tsize = txq->nb_tx_desc * sizeof(struct rte_mbuf *);
txq->sw_ring = rte_zmalloc("tx_sw_ring", tsize,
RTE_CACHE_LINE_SIZE);
if (txq->sw_ring == NULL) {
bnx2x_tx_queue_release(txq);
return -ENOMEM;
}
/* PMD_DRV_LOG(DEBUG, sc, "sw_ring=%p hw_ring=%p dma_addr=0x%"PRIx64,
txq->sw_ring, txq->tx_ring, txq->tx_ring_phys_addr); */
/* Link TX pages */
for (i = 1; i <= txq->nb_tx_pages; i++) {
tx_n_bd = &txq->tx_ring[TOTAL_TX_BD_PER_PAGE * i - 1].next_bd;
busaddr = txq->tx_ring_phys_addr + BNX2X_PAGE_SIZE * (i % txq->nb_tx_pages);
tx_n_bd->addr_hi = rte_cpu_to_le_32(U64_HI(busaddr));
tx_n_bd->addr_lo = rte_cpu_to_le_32(U64_LO(busaddr));
/* PMD_DRV_LOG(DEBUG, sc, "link tx page %lu",
* (TOTAL_TX_BD_PER_PAGE * i - 1));
*/
}
txq->queue_id = queue_idx;
txq->port_id = dev->data->port_id;
txq->tx_pkt_tail = 0;
txq->tx_pkt_head = 0;
txq->tx_bd_tail = 0;
txq->tx_bd_head = 0;
txq->nb_tx_avail = txq->nb_tx_desc;
dev->data->tx_queues[queue_idx] = txq;
if (!sc->tx_queues) sc->tx_queues = dev->data->tx_queues;
return 0;
}
static inline void
bnx2x_upd_rx_prod_fast(struct bnx2x_softc *sc, struct bnx2x_fastpath *fp,
uint16_t rx_bd_prod, uint16_t rx_cq_prod)
{
union {
struct ustorm_eth_rx_producers rx_prods;
uint32_t val;
} val = { {0} };
val.rx_prods.bd_prod = rx_bd_prod;
val.rx_prods.cqe_prod = rx_cq_prod;
REG_WR(sc, fp->ustorm_rx_prods_offset, val.val);
}
static uint16_t
bnx2x_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
{
struct bnx2x_rx_queue *rxq = p_rxq;
struct bnx2x_softc *sc = rxq->sc;
struct bnx2x_fastpath *fp = &sc->fp[rxq->queue_id];
uint32_t nb_rx = 0;
uint16_t hw_cq_cons, sw_cq_cons, sw_cq_prod;
uint16_t bd_cons, bd_prod;
struct rte_mbuf *new_mb;
uint16_t rx_pref;
struct eth_fast_path_rx_cqe *cqe_fp;
uint16_t len, pad, bd_len, buf_len;
struct rte_mbuf *rx_mb = NULL;
static bool log_once = true;
rte_spinlock_lock(&(fp)->rx_mtx);
hw_cq_cons = le16toh(*fp->rx_cq_cons_sb);
if ((hw_cq_cons & USABLE_RCQ_ENTRIES_PER_PAGE) ==
USABLE_RCQ_ENTRIES_PER_PAGE) {
++hw_cq_cons;
}
bd_cons = rxq->rx_bd_head;
bd_prod = rxq->rx_bd_tail;
sw_cq_cons = rxq->rx_cq_head;
sw_cq_prod = rxq->rx_cq_tail;
if (sw_cq_cons == hw_cq_cons) {
rte_spinlock_unlock(&(fp)->rx_mtx);
return 0;
}
while (nb_rx < nb_pkts && sw_cq_cons != hw_cq_cons) {
bd_prod &= MAX_RX_BD(rxq);
bd_cons &= MAX_RX_BD(rxq);
cqe_fp = &rxq->cq_ring[sw_cq_cons & MAX_RX_BD(rxq)].fast_path_cqe;
if (unlikely(CQE_TYPE_SLOW(cqe_fp->type_error_flags & ETH_FAST_PATH_RX_CQE_TYPE))) {
PMD_RX_LOG(ERR, "slowpath event during traffic processing");
break;
}
if (unlikely(cqe_fp->type_error_flags & ETH_FAST_PATH_RX_CQE_PHY_DECODE_ERR_FLG)) {
PMD_RX_LOG(ERR, "flags 0x%x rx packet %u",
cqe_fp->type_error_flags, sw_cq_cons);
goto next_rx;
}
len = cqe_fp->pkt_len_or_gro_seg_len;
pad = cqe_fp->placement_offset;
bd_len = cqe_fp->len_on_bd;
buf_len = rxq->sw_ring[bd_cons]->buf_len;
/* Check for sufficient buffer length */
if (unlikely(buf_len < len + (pad + RTE_PKTMBUF_HEADROOM))) {
if (unlikely(log_once)) {
PMD_DRV_LOG(ERR, sc, "mbuf size %d is not enough to hold Rx packet length more than %d",
buf_len - RTE_PKTMBUF_HEADROOM,
buf_len -
(pad + RTE_PKTMBUF_HEADROOM));
log_once = false;
}
goto next_rx;
}
new_mb = rte_mbuf_raw_alloc(rxq->mb_pool);
if (unlikely(!new_mb)) {
PMD_RX_LOG(ERR, "mbuf alloc fail fp[%02d]", fp->index);
rte_eth_devices[rxq->port_id].data->
rx_mbuf_alloc_failed++;
goto next_rx;
}
rx_mb = rxq->sw_ring[bd_cons];
rxq->sw_ring[bd_cons] = new_mb;
rxq->rx_ring[bd_prod] =
rte_cpu_to_le_64(rte_mbuf_data_iova_default(new_mb));
rx_pref = NEXT_RX_BD(bd_cons) & MAX_RX_BD(rxq);
rte_prefetch0(rxq->sw_ring[rx_pref]);
if ((rx_pref & 0x3) == 0) {
rte_prefetch0(&rxq->rx_ring[rx_pref]);
rte_prefetch0(&rxq->sw_ring[rx_pref]);
}
rx_mb->data_off = pad + RTE_PKTMBUF_HEADROOM;
rx_mb->nb_segs = 1;
rx_mb->next = NULL;
rx_mb->pkt_len = len;
rx_mb->data_len = bd_len;
rx_mb->port = rxq->port_id;
rte_prefetch1(rte_pktmbuf_mtod(rx_mb, void *));
/*
* If we received a packet with a vlan tag,
* attach that information to the packet.
*/
if (cqe_fp->pars_flags.flags & PARSING_FLAGS_VLAN) {
rx_mb->vlan_tci = cqe_fp->vlan_tag;
rx_mb->ol_flags |= RTE_MBUF_F_RX_VLAN | RTE_MBUF_F_RX_VLAN_STRIPPED;
}
rx_pkts[nb_rx] = rx_mb;
nb_rx++;
/* limit spinning on the queue */
if (unlikely(nb_rx == sc->rx_budget)) {
PMD_RX_LOG(ERR, "Limit spinning on the queue");
break;
}
next_rx:
bd_cons = NEXT_RX_BD(bd_cons);
bd_prod = NEXT_RX_BD(bd_prod);
sw_cq_prod = NEXT_RCQ_IDX(sw_cq_prod);
sw_cq_cons = NEXT_RCQ_IDX(sw_cq_cons);
}
rxq->rx_bd_head = bd_cons;
rxq->rx_bd_tail = bd_prod;
rxq->rx_cq_head = sw_cq_cons;
rxq->rx_cq_tail = sw_cq_prod;
bnx2x_upd_rx_prod_fast(sc, fp, bd_prod, sw_cq_prod);
rte_spinlock_unlock(&(fp)->rx_mtx);
return nb_rx;
}
static uint16_t
bnx2x_rxtx_pkts_dummy(__rte_unused void *p_rxq,
__rte_unused struct rte_mbuf **rx_pkts,
__rte_unused uint16_t nb_pkts)
{
return 0;
}
void bnx2x_dev_rxtx_init_dummy(struct rte_eth_dev *dev)
{
dev->rx_pkt_burst = bnx2x_rxtx_pkts_dummy;
dev->tx_pkt_burst = bnx2x_rxtx_pkts_dummy;
}
void bnx2x_dev_rxtx_init(struct rte_eth_dev *dev)
{
dev->rx_pkt_burst = bnx2x_recv_pkts;
dev->tx_pkt_burst = bnx2x_xmit_pkts;
}
void
bnx2x_dev_clear_queues(struct rte_eth_dev *dev)
{
struct bnx2x_softc *sc = dev->data->dev_private;
uint8_t i;
PMD_INIT_FUNC_TRACE(sc);
for (i = 0; i < dev->data->nb_tx_queues; i++) {
struct bnx2x_tx_queue *txq = dev->data->tx_queues[i];
if (txq != NULL) {
bnx2x_tx_queue_release(txq);
dev->data->tx_queues[i] = NULL;
}
}
for (i = 0; i < dev->data->nb_rx_queues; i++) {
struct bnx2x_rx_queue *rxq = dev->data->rx_queues[i];
if (rxq != NULL) {
bnx2x_rx_queue_release(rxq);
dev->data->rx_queues[i] = NULL;
}
}
}