numam-dpdk/drivers/net/ionic/ionic_rxtx.c
Thomas Monjalon ce6427ddca replace cold attributes
The new macro __rte_cold, for compiler hinting,
is now used where appropriate for consistency.

Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Reviewed-by: David Christensen <drc@linux.vnet.ibm.com>
2020-04-16 18:30:58 +02:00

1083 lines
26 KiB
C

/* SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0)
* Copyright(c) 2018-2019 Pensando Systems, Inc. All rights reserved.
*/
#include <sys/queue.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <errno.h>
#include <stdint.h>
#include <stdarg.h>
#include <unistd.h>
#include <inttypes.h>
#include <rte_byteorder.h>
#include <rte_common.h>
#include <rte_cycles.h>
#include <rte_log.h>
#include <rte_debug.h>
#include <rte_interrupts.h>
#include <rte_pci.h>
#include <rte_memory.h>
#include <rte_memzone.h>
#include <rte_launch.h>
#include <rte_eal.h>
#include <rte_per_lcore.h>
#include <rte_lcore.h>
#include <rte_atomic.h>
#include <rte_branch_prediction.h>
#include <rte_mempool.h>
#include <rte_malloc.h>
#include <rte_mbuf.h>
#include <rte_ether.h>
#include <rte_ethdev_driver.h>
#include <rte_prefetch.h>
#include <rte_udp.h>
#include <rte_tcp.h>
#include <rte_sctp.h>
#include <rte_string_fns.h>
#include <rte_errno.h>
#include <rte_ip.h>
#include <rte_net.h>
#include "ionic_logs.h"
#include "ionic_mac_api.h"
#include "ionic_ethdev.h"
#include "ionic_lif.h"
#include "ionic_rxtx.h"
#define IONIC_RX_RING_DOORBELL_STRIDE (32 - 1)
/*********************************************************************
*
* TX functions
*
**********************************************************************/
void
ionic_txq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
struct rte_eth_txq_info *qinfo)
{
struct ionic_qcq *txq = dev->data->tx_queues[queue_id];
struct ionic_queue *q = &txq->q;
qinfo->nb_desc = q->num_descs;
qinfo->conf.offloads = txq->offloads;
qinfo->conf.tx_deferred_start = txq->deferred_start;
}
static inline void __rte_cold
ionic_tx_flush(struct ionic_cq *cq)
{
struct ionic_queue *q = cq->bound_q;
struct ionic_desc_info *q_desc_info;
struct rte_mbuf *txm, *next;
struct ionic_txq_comp *cq_desc_base = cq->base;
struct ionic_txq_comp *cq_desc;
u_int32_t comp_index = (u_int32_t)-1;
cq_desc = &cq_desc_base[cq->tail_idx];
while (color_match(cq_desc->color, cq->done_color)) {
cq->tail_idx = (cq->tail_idx + 1) & (cq->num_descs - 1);
/* Prefetch the next 4 descriptors (not really useful here) */
if ((cq->tail_idx & 0x3) == 0)
rte_prefetch0(&cq_desc_base[cq->tail_idx]);
if (cq->tail_idx == 0)
cq->done_color = !cq->done_color;
comp_index = cq_desc->comp_index;
cq_desc = &cq_desc_base[cq->tail_idx];
}
if (comp_index != (u_int32_t)-1) {
while (q->tail_idx != comp_index) {
q_desc_info = &q->info[q->tail_idx];
q->tail_idx = (q->tail_idx + 1) & (q->num_descs - 1);
/* Prefetch the next 4 descriptors */
if ((q->tail_idx & 0x3) == 0)
/* q desc info */
rte_prefetch0(&q->info[q->tail_idx]);
/*
* Note: you can just use rte_pktmbuf_free,
* but this loop is faster
*/
txm = q_desc_info->cb_arg;
while (txm != NULL) {
next = txm->next;
rte_pktmbuf_free_seg(txm);
txm = next;
}
}
}
}
void __rte_cold
ionic_dev_tx_queue_release(void *tx_queue)
{
struct ionic_qcq *txq = (struct ionic_qcq *)tx_queue;
IONIC_PRINT_CALL();
ionic_qcq_free(txq);
}
int __rte_cold
ionic_dev_tx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
{
struct ionic_qcq *txq;
IONIC_PRINT_CALL();
txq = eth_dev->data->tx_queues[tx_queue_id];
/*
* Note: we should better post NOP Tx desc and wait for its completion
* before disabling Tx queue
*/
ionic_qcq_disable(txq);
ionic_tx_flush(&txq->cq);
ionic_lif_txq_deinit(txq);
eth_dev->data->tx_queue_state[tx_queue_id] =
RTE_ETH_QUEUE_STATE_STOPPED;
return 0;
}
int __rte_cold
ionic_dev_tx_queue_setup(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id,
uint16_t nb_desc, uint32_t socket_id __rte_unused,
const struct rte_eth_txconf *tx_conf)
{
struct ionic_lif *lif = IONIC_ETH_DEV_TO_LIF(eth_dev);
struct ionic_qcq *txq;
uint64_t offloads;
int err;
IONIC_PRINT_CALL();
IONIC_PRINT(DEBUG, "Configuring TX queue %u with %u buffers",
tx_queue_id, nb_desc);
if (tx_queue_id >= lif->ntxqcqs) {
IONIC_PRINT(DEBUG, "Queue index %u not available "
"(max %u queues)",
tx_queue_id, lif->ntxqcqs);
return -EINVAL;
}
offloads = tx_conf->offloads | eth_dev->data->dev_conf.txmode.offloads;
/* Validate number of receive descriptors */
if (!rte_is_power_of_2(nb_desc) || nb_desc < IONIC_MIN_RING_DESC)
return -EINVAL; /* or use IONIC_DEFAULT_RING_DESC */
/* Free memory prior to re-allocation if needed... */
if (eth_dev->data->tx_queues[tx_queue_id] != NULL) {
void *tx_queue = eth_dev->data->tx_queues[tx_queue_id];
ionic_dev_tx_queue_release(tx_queue);
eth_dev->data->tx_queues[tx_queue_id] = NULL;
}
err = ionic_tx_qcq_alloc(lif, tx_queue_id, nb_desc, &txq);
if (err) {
IONIC_PRINT(DEBUG, "Queue allocation failure");
return -EINVAL;
}
/* Do not start queue with rte_eth_dev_start() */
txq->deferred_start = tx_conf->tx_deferred_start;
txq->offloads = offloads;
eth_dev->data->tx_queues[tx_queue_id] = txq;
return 0;
}
/*
* Start Transmit Units for specified queue.
*/
int __rte_cold
ionic_dev_tx_queue_start(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
{
struct ionic_qcq *txq;
int err;
IONIC_PRINT_CALL();
txq = eth_dev->data->tx_queues[tx_queue_id];
err = ionic_lif_txq_init(txq);
if (err)
return err;
ionic_qcq_enable(txq);
eth_dev->data->tx_queue_state[tx_queue_id] =
RTE_ETH_QUEUE_STATE_STARTED;
return 0;
}
static void
ionic_tx_tcp_pseudo_csum(struct rte_mbuf *txm)
{
struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(txm, struct ether_hdr *);
char *l3_hdr = ((char *)eth_hdr) + txm->l2_len;
struct rte_tcp_hdr *tcp_hdr = (struct rte_tcp_hdr *)
(l3_hdr + txm->l3_len);
if (txm->ol_flags & PKT_TX_IP_CKSUM) {
struct rte_ipv4_hdr *ipv4_hdr = (struct rte_ipv4_hdr *)l3_hdr;
ipv4_hdr->hdr_checksum = 0;
tcp_hdr->cksum = 0;
tcp_hdr->cksum = rte_ipv4_udptcp_cksum(ipv4_hdr, tcp_hdr);
} else {
struct rte_ipv6_hdr *ipv6_hdr = (struct rte_ipv6_hdr *)l3_hdr;
tcp_hdr->cksum = 0;
tcp_hdr->cksum = rte_ipv6_udptcp_cksum(ipv6_hdr, tcp_hdr);
}
}
static void
ionic_tx_tcp_inner_pseudo_csum(struct rte_mbuf *txm)
{
struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(txm, struct ether_hdr *);
char *l3_hdr = ((char *)eth_hdr) + txm->outer_l2_len +
txm->outer_l3_len + txm->l2_len;
struct rte_tcp_hdr *tcp_hdr = (struct rte_tcp_hdr *)
(l3_hdr + txm->l3_len);
if (txm->ol_flags & PKT_TX_IPV4) {
struct rte_ipv4_hdr *ipv4_hdr = (struct rte_ipv4_hdr *)l3_hdr;
ipv4_hdr->hdr_checksum = 0;
tcp_hdr->cksum = 0;
tcp_hdr->cksum = rte_ipv4_udptcp_cksum(ipv4_hdr, tcp_hdr);
} else {
struct rte_ipv6_hdr *ipv6_hdr = (struct rte_ipv6_hdr *)l3_hdr;
tcp_hdr->cksum = 0;
tcp_hdr->cksum = rte_ipv6_udptcp_cksum(ipv6_hdr, tcp_hdr);
}
}
static void
ionic_tx_tso_post(struct ionic_queue *q, struct ionic_txq_desc *desc,
struct rte_mbuf *txm,
rte_iova_t addr, uint8_t nsge, uint16_t len,
uint32_t hdrlen, uint32_t mss,
bool encap,
uint16_t vlan_tci, bool has_vlan,
bool start, bool done)
{
uint8_t flags = 0;
flags |= has_vlan ? IONIC_TXQ_DESC_FLAG_VLAN : 0;
flags |= encap ? IONIC_TXQ_DESC_FLAG_ENCAP : 0;
flags |= start ? IONIC_TXQ_DESC_FLAG_TSO_SOT : 0;
flags |= done ? IONIC_TXQ_DESC_FLAG_TSO_EOT : 0;
desc->cmd = encode_txq_desc_cmd(IONIC_TXQ_DESC_OPCODE_TSO,
flags, nsge, addr);
desc->len = len;
desc->vlan_tci = vlan_tci;
desc->hdr_len = hdrlen;
desc->mss = mss;
ionic_q_post(q, done, NULL, done ? txm : NULL);
}
static struct ionic_txq_desc *
ionic_tx_tso_next(struct ionic_queue *q, struct ionic_txq_sg_elem **elem)
{
struct ionic_txq_desc *desc_base = q->base;
struct ionic_txq_sg_desc *sg_desc_base = q->sg_base;
struct ionic_txq_desc *desc = &desc_base[q->head_idx];
struct ionic_txq_sg_desc *sg_desc = &sg_desc_base[q->head_idx];
*elem = sg_desc->elems;
return desc;
}
static int
ionic_tx_tso(struct ionic_queue *q, struct rte_mbuf *txm,
uint64_t offloads __rte_unused, bool not_xmit_more)
{
struct ionic_tx_stats *stats = IONIC_Q_TO_TX_STATS(q);
struct ionic_txq_desc *desc;
struct ionic_txq_sg_elem *elem;
struct rte_mbuf *txm_seg;
uint64_t desc_addr = 0;
uint16_t desc_len = 0;
uint8_t desc_nsge;
uint32_t hdrlen;
uint32_t mss = txm->tso_segsz;
uint32_t frag_left = 0;
uint32_t left;
uint32_t seglen;
uint32_t len;
uint32_t offset = 0;
bool start, done;
bool encap;
bool has_vlan = !!(txm->ol_flags & PKT_TX_VLAN_PKT);
uint16_t vlan_tci = txm->vlan_tci;
uint64_t ol_flags = txm->ol_flags;
encap = ((ol_flags & PKT_TX_OUTER_IP_CKSUM) ||
(ol_flags & PKT_TX_OUTER_UDP_CKSUM)) &&
((ol_flags & PKT_TX_OUTER_IPV4) ||
(ol_flags & PKT_TX_OUTER_IPV6));
/* Preload inner-most TCP csum field with IP pseudo hdr
* calculated with IP length set to zero. HW will later
* add in length to each TCP segment resulting from the TSO.
*/
if (encap) {
ionic_tx_tcp_inner_pseudo_csum(txm);
hdrlen = txm->outer_l2_len + txm->outer_l3_len +
txm->l2_len + txm->l3_len + txm->l4_len;
} else {
ionic_tx_tcp_pseudo_csum(txm);
hdrlen = txm->l2_len + txm->l3_len + txm->l4_len;
}
seglen = hdrlen + mss;
left = txm->data_len;
desc = ionic_tx_tso_next(q, &elem);
start = true;
/* Chop data up into desc segments */
while (left > 0) {
len = RTE_MIN(seglen, left);
frag_left = seglen - len;
desc_addr = rte_cpu_to_le_64(rte_mbuf_data_iova_default(txm));
desc_len = len;
desc_nsge = 0;
left -= len;
offset += len;
if (txm->nb_segs > 1 && frag_left > 0)
continue;
done = (txm->nb_segs == 1 && left == 0);
ionic_tx_tso_post(q, desc, txm,
desc_addr, desc_nsge, desc_len,
hdrlen, mss,
encap,
vlan_tci, has_vlan,
start, done && not_xmit_more);
desc = ionic_tx_tso_next(q, &elem);
start = false;
seglen = mss;
}
/* Chop frags into desc segments */
txm_seg = txm->next;
while (txm_seg != NULL) {
offset = 0;
left = txm_seg->data_len;
stats->frags++;
while (left > 0) {
rte_iova_t data_iova;
data_iova = rte_mbuf_data_iova(txm_seg);
elem->addr = rte_cpu_to_le_64(data_iova) + offset;
if (frag_left > 0) {
len = RTE_MIN(frag_left, left);
frag_left -= len;
elem->len = len;
elem++;
desc_nsge++;
} else {
len = RTE_MIN(mss, left);
frag_left = mss - len;
data_iova = rte_mbuf_data_iova(txm_seg);
desc_addr = rte_cpu_to_le_64(data_iova);
desc_len = len;
desc_nsge = 0;
}
left -= len;
offset += len;
if (txm_seg->next != NULL && frag_left > 0)
continue;
done = (txm_seg->next == NULL && left == 0);
ionic_tx_tso_post(q, desc, txm_seg,
desc_addr, desc_nsge, desc_len,
hdrlen, mss,
encap,
vlan_tci, has_vlan,
start, done && not_xmit_more);
desc = ionic_tx_tso_next(q, &elem);
start = false;
}
txm_seg = txm_seg->next;
}
stats->tso++;
return 0;
}
static int
ionic_tx(struct ionic_queue *q, struct rte_mbuf *txm,
uint64_t offloads, bool not_xmit_more)
{
struct ionic_txq_desc *desc_base = q->base;
struct ionic_txq_sg_desc *sg_desc_base = q->sg_base;
struct ionic_txq_desc *desc = &desc_base[q->head_idx];
struct ionic_txq_sg_desc *sg_desc = &sg_desc_base[q->head_idx];
struct ionic_txq_sg_elem *elem = sg_desc->elems;
struct ionic_tx_stats *stats = IONIC_Q_TO_TX_STATS(q);
struct rte_mbuf *txm_seg;
bool encap;
bool has_vlan;
uint64_t ol_flags = txm->ol_flags;
uint64_t addr = rte_cpu_to_le_64(rte_mbuf_data_iova_default(txm));
uint8_t opcode = IONIC_TXQ_DESC_OPCODE_CSUM_NONE;
uint8_t flags = 0;
if ((ol_flags & PKT_TX_IP_CKSUM) &&
(offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)) {
opcode = IONIC_TXQ_DESC_OPCODE_CSUM_HW;
flags |= IONIC_TXQ_DESC_FLAG_CSUM_L3;
if (((ol_flags & PKT_TX_TCP_CKSUM) &&
(offloads & DEV_TX_OFFLOAD_TCP_CKSUM)) ||
((ol_flags & PKT_TX_UDP_CKSUM) &&
(offloads & DEV_TX_OFFLOAD_UDP_CKSUM)))
flags |= IONIC_TXQ_DESC_FLAG_CSUM_L4;
} else {
stats->no_csum++;
}
has_vlan = (ol_flags & PKT_TX_VLAN_PKT);
encap = ((ol_flags & PKT_TX_OUTER_IP_CKSUM) ||
(ol_flags & PKT_TX_OUTER_UDP_CKSUM)) &&
((ol_flags & PKT_TX_OUTER_IPV4) ||
(ol_flags & PKT_TX_OUTER_IPV6));
flags |= has_vlan ? IONIC_TXQ_DESC_FLAG_VLAN : 0;
flags |= encap ? IONIC_TXQ_DESC_FLAG_ENCAP : 0;
desc->cmd = encode_txq_desc_cmd(opcode, flags, txm->nb_segs - 1, addr);
desc->len = txm->data_len;
desc->vlan_tci = txm->vlan_tci;
txm_seg = txm->next;
while (txm_seg != NULL) {
elem->len = txm_seg->data_len;
elem->addr = rte_cpu_to_le_64(rte_mbuf_data_iova(txm_seg));
stats->frags++;
elem++;
txm_seg = txm_seg->next;
}
ionic_q_post(q, not_xmit_more, NULL, txm);
return 0;
}
uint16_t
ionic_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
uint16_t nb_pkts)
{
struct ionic_qcq *txq = (struct ionic_qcq *)tx_queue;
struct ionic_queue *q = &txq->q;
struct ionic_cq *cq = &txq->cq;
struct ionic_tx_stats *stats = IONIC_Q_TO_TX_STATS(q);
uint32_t next_q_head_idx;
uint32_t bytes_tx = 0;
uint16_t nb_tx = 0;
int err;
bool last;
/* Cleaning old buffers */
ionic_tx_flush(cq);
if (unlikely(ionic_q_space_avail(q) < nb_pkts)) {
stats->stop += nb_pkts;
return 0;
}
while (nb_tx < nb_pkts) {
last = (nb_tx == (nb_pkts - 1));
next_q_head_idx = (q->head_idx + 1) & (q->num_descs - 1);
if ((next_q_head_idx & 0x3) == 0) {
struct ionic_txq_desc *desc_base = q->base;
rte_prefetch0(&desc_base[next_q_head_idx]);
rte_prefetch0(&q->info[next_q_head_idx]);
}
if (tx_pkts[nb_tx]->ol_flags & PKT_TX_TCP_SEG)
err = ionic_tx_tso(q, tx_pkts[nb_tx], txq->offloads,
last);
else
err = ionic_tx(q, tx_pkts[nb_tx], txq->offloads, last);
if (err) {
stats->drop += nb_pkts - nb_tx;
if (nb_tx > 0)
ionic_q_flush(q);
break;
}
bytes_tx += tx_pkts[nb_tx]->pkt_len;
nb_tx++;
}
stats->packets += nb_tx;
stats->bytes += bytes_tx;
return nb_tx;
}
/*********************************************************************
*
* TX prep functions
*
**********************************************************************/
#define IONIC_TX_OFFLOAD_MASK ( \
PKT_TX_IPV4 | \
PKT_TX_IPV6 | \
PKT_TX_VLAN | \
PKT_TX_IP_CKSUM | \
PKT_TX_TCP_SEG | \
PKT_TX_L4_MASK)
#define IONIC_TX_OFFLOAD_NOTSUP_MASK \
(PKT_TX_OFFLOAD_MASK ^ IONIC_TX_OFFLOAD_MASK)
uint16_t
ionic_prep_pkts(void *tx_queue __rte_unused, struct rte_mbuf **tx_pkts,
uint16_t nb_pkts)
{
struct rte_mbuf *txm;
uint64_t offloads;
int i = 0;
for (i = 0; i < nb_pkts; i++) {
txm = tx_pkts[i];
if (txm->nb_segs > IONIC_TX_MAX_SG_ELEMS) {
rte_errno = -EINVAL;
break;
}
offloads = txm->ol_flags;
if (offloads & IONIC_TX_OFFLOAD_NOTSUP_MASK) {
rte_errno = -ENOTSUP;
break;
}
}
return i;
}
/*********************************************************************
*
* RX functions
*
**********************************************************************/
static void ionic_rx_recycle(struct ionic_queue *q, uint32_t q_desc_index,
struct rte_mbuf *mbuf);
void
ionic_rxq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
struct rte_eth_rxq_info *qinfo)
{
struct ionic_qcq *rxq = dev->data->rx_queues[queue_id];
struct ionic_queue *q = &rxq->q;
qinfo->mp = rxq->mb_pool;
qinfo->scattered_rx = dev->data->scattered_rx;
qinfo->nb_desc = q->num_descs;
qinfo->conf.rx_deferred_start = rxq->deferred_start;
qinfo->conf.offloads = rxq->offloads;
}
static void __rte_cold
ionic_rx_empty(struct ionic_queue *q)
{
struct ionic_qcq *rxq = IONIC_Q_TO_QCQ(q);
struct ionic_desc_info *cur;
struct rte_mbuf *mbuf;
while (q->tail_idx != q->head_idx) {
cur = &q->info[q->tail_idx];
mbuf = cur->cb_arg;
rte_mempool_put(rxq->mb_pool, mbuf);
q->tail_idx = (q->tail_idx + 1) & (q->num_descs - 1);
}
}
void __rte_cold
ionic_dev_rx_queue_release(void *rx_queue)
{
struct ionic_qcq *rxq = (struct ionic_qcq *)rx_queue;
IONIC_PRINT_CALL();
ionic_rx_empty(&rxq->q);
ionic_qcq_free(rxq);
}
int __rte_cold
ionic_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
uint16_t rx_queue_id,
uint16_t nb_desc,
uint32_t socket_id __rte_unused,
const struct rte_eth_rxconf *rx_conf,
struct rte_mempool *mp)
{
struct ionic_lif *lif = IONIC_ETH_DEV_TO_LIF(eth_dev);
struct ionic_qcq *rxq;
uint64_t offloads;
int err;
IONIC_PRINT_CALL();
IONIC_PRINT(DEBUG, "Configuring RX queue %u with %u buffers",
rx_queue_id, nb_desc);
if (rx_queue_id >= lif->nrxqcqs) {
IONIC_PRINT(ERR,
"Queue index %u not available (max %u queues)",
rx_queue_id, lif->nrxqcqs);
return -EINVAL;
}
offloads = rx_conf->offloads | eth_dev->data->dev_conf.rxmode.offloads;
/* Validate number of receive descriptors */
if (!rte_is_power_of_2(nb_desc) ||
nb_desc < IONIC_MIN_RING_DESC ||
nb_desc > IONIC_MAX_RING_DESC) {
IONIC_PRINT(ERR,
"Bad number of descriptors (%u) for queue %u (min: %u)",
nb_desc, rx_queue_id, IONIC_MIN_RING_DESC);
return -EINVAL; /* or use IONIC_DEFAULT_RING_DESC */
}
if (rx_conf->offloads & DEV_RX_OFFLOAD_SCATTER)
eth_dev->data->scattered_rx = 1;
/* Free memory prior to re-allocation if needed... */
if (eth_dev->data->rx_queues[rx_queue_id] != NULL) {
void *rx_queue = eth_dev->data->rx_queues[rx_queue_id];
ionic_dev_rx_queue_release(rx_queue);
eth_dev->data->rx_queues[rx_queue_id] = NULL;
}
err = ionic_rx_qcq_alloc(lif, rx_queue_id, nb_desc, &rxq);
if (err) {
IONIC_PRINT(ERR, "Queue allocation failure");
return -EINVAL;
}
rxq->mb_pool = mp;
/*
* Note: the interface does not currently support
* DEV_RX_OFFLOAD_KEEP_CRC, please also consider ETHER_CRC_LEN
* when the adapter will be able to keep the CRC and subtract
* it to the length for all received packets:
* if (eth_dev->data->dev_conf.rxmode.offloads &
* DEV_RX_OFFLOAD_KEEP_CRC)
* rxq->crc_len = ETHER_CRC_LEN;
*/
/* Do not start queue with rte_eth_dev_start() */
rxq->deferred_start = rx_conf->rx_deferred_start;
rxq->offloads = offloads;
eth_dev->data->rx_queues[rx_queue_id] = rxq;
return 0;
}
static void
ionic_rx_clean(struct ionic_queue *q,
uint32_t q_desc_index, uint32_t cq_desc_index,
void *cb_arg, void *service_cb_arg)
{
struct ionic_rxq_comp *cq_desc_base = q->bound_cq->base;
struct ionic_rxq_comp *cq_desc = &cq_desc_base[cq_desc_index];
struct rte_mbuf *rxm = cb_arg;
struct rte_mbuf *rxm_seg;
struct ionic_qcq *rxq = IONIC_Q_TO_QCQ(q);
uint32_t max_frame_size =
rxq->lif->eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
uint64_t pkt_flags = 0;
uint32_t pkt_type;
struct ionic_rx_stats *stats = IONIC_Q_TO_RX_STATS(q);
struct ionic_rx_service *recv_args = (struct ionic_rx_service *)
service_cb_arg;
uint32_t buf_size = (uint16_t)
(rte_pktmbuf_data_room_size(rxq->mb_pool) -
RTE_PKTMBUF_HEADROOM);
uint32_t left;
if (!recv_args) {
stats->no_cb_arg++;
/* Flush */
rte_pktmbuf_free(rxm);
/*
* Note: rte_mempool_put is faster with no segs
* rte_mempool_put(rxq->mb_pool, rxm);
*/
return;
}
if (cq_desc->status) {
stats->bad_cq_status++;
ionic_rx_recycle(q, q_desc_index, rxm);
return;
}
if (recv_args->nb_rx >= recv_args->nb_pkts) {
stats->no_room++;
ionic_rx_recycle(q, q_desc_index, rxm);
return;
}
if (cq_desc->len > max_frame_size ||
cq_desc->len == 0) {
stats->bad_len++;
ionic_rx_recycle(q, q_desc_index, rxm);
return;
}
rxm->data_off = RTE_PKTMBUF_HEADROOM;
rte_prefetch1((char *)rxm->buf_addr + rxm->data_off);
rxm->nb_segs = 1; /* cq_desc->num_sg_elems */
rxm->pkt_len = cq_desc->len;
rxm->port = rxq->lif->port_id;
left = cq_desc->len;
rxm->data_len = RTE_MIN(buf_size, left);
left -= rxm->data_len;
rxm_seg = rxm->next;
while (rxm_seg && left) {
rxm_seg->data_len = RTE_MIN(buf_size, left);
left -= rxm_seg->data_len;
rxm_seg = rxm_seg->next;
rxm->nb_segs++;
}
/* RSS */
pkt_flags |= PKT_RX_RSS_HASH;
rxm->hash.rss = cq_desc->rss_hash;
/* Vlan Strip */
if (cq_desc->csum_flags & IONIC_RXQ_COMP_CSUM_F_VLAN) {
pkt_flags |= PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED;
rxm->vlan_tci = cq_desc->vlan_tci;
}
/* Checksum */
if (cq_desc->csum_flags & IONIC_RXQ_COMP_CSUM_F_CALC) {
if (cq_desc->csum_flags & IONIC_RXQ_COMP_CSUM_F_IP_OK)
pkt_flags |= PKT_RX_IP_CKSUM_GOOD;
else if (cq_desc->csum_flags & IONIC_RXQ_COMP_CSUM_F_IP_BAD)
pkt_flags |= PKT_RX_IP_CKSUM_BAD;
if ((cq_desc->csum_flags & IONIC_RXQ_COMP_CSUM_F_TCP_OK) ||
(cq_desc->csum_flags & IONIC_RXQ_COMP_CSUM_F_UDP_OK))
pkt_flags |= PKT_RX_L4_CKSUM_GOOD;
else if ((cq_desc->csum_flags &
IONIC_RXQ_COMP_CSUM_F_TCP_BAD) ||
(cq_desc->csum_flags &
IONIC_RXQ_COMP_CSUM_F_UDP_BAD))
pkt_flags |= PKT_RX_L4_CKSUM_BAD;
}
rxm->ol_flags = pkt_flags;
/* Packet Type */
switch (cq_desc->pkt_type_color & IONIC_RXQ_COMP_PKT_TYPE_MASK) {
case IONIC_PKT_TYPE_IPV4:
pkt_type = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4;
break;
case IONIC_PKT_TYPE_IPV6:
pkt_type = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6;
break;
case IONIC_PKT_TYPE_IPV4_TCP:
pkt_type = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4 |
RTE_PTYPE_L4_TCP;
break;
case IONIC_PKT_TYPE_IPV6_TCP:
pkt_type = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6 |
RTE_PTYPE_L4_TCP;
break;
case IONIC_PKT_TYPE_IPV4_UDP:
pkt_type = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4 |
RTE_PTYPE_L4_UDP;
break;
case IONIC_PKT_TYPE_IPV6_UDP:
pkt_type = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6 |
RTE_PTYPE_L4_UDP;
break;
default:
{
struct rte_ether_hdr *eth_h = rte_pktmbuf_mtod(rxm,
struct rte_ether_hdr *);
uint16_t ether_type = eth_h->ether_type;
if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_ARP))
pkt_type = RTE_PTYPE_L2_ETHER_ARP;
else
pkt_type = RTE_PTYPE_UNKNOWN;
break;
}
}
rxm->packet_type = pkt_type;
recv_args->rx_pkts[recv_args->nb_rx] = rxm;
recv_args->nb_rx++;
stats->packets++;
stats->bytes += rxm->pkt_len;
}
static void
ionic_rx_recycle(struct ionic_queue *q, uint32_t q_desc_index,
struct rte_mbuf *mbuf)
{
struct ionic_rxq_desc *desc_base = q->base;
struct ionic_rxq_desc *old = &desc_base[q_desc_index];
struct ionic_rxq_desc *new = &desc_base[q->head_idx];
new->addr = old->addr;
new->len = old->len;
ionic_q_post(q, true, ionic_rx_clean, mbuf);
}
static int __rte_cold
ionic_rx_fill(struct ionic_qcq *rxq, uint32_t len)
{
struct ionic_queue *q = &rxq->q;
struct ionic_rxq_desc *desc_base = q->base;
struct ionic_rxq_sg_desc *sg_desc_base = q->sg_base;
struct ionic_rxq_desc *desc;
struct ionic_rxq_sg_desc *sg_desc;
struct ionic_rxq_sg_elem *elem;
rte_iova_t dma_addr;
uint32_t i, j, nsegs, buf_size, size;
bool ring_doorbell;
buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mb_pool) -
RTE_PKTMBUF_HEADROOM);
/* Initialize software ring entries */
for (i = ionic_q_space_avail(q); i; i--) {
struct rte_mbuf *rxm = rte_mbuf_raw_alloc(rxq->mb_pool);
struct rte_mbuf *prev_rxm_seg;
if (rxm == NULL) {
IONIC_PRINT(ERR, "RX mbuf alloc failed");
return -ENOMEM;
}
nsegs = (len + buf_size - 1) / buf_size;
desc = &desc_base[q->head_idx];
dma_addr = rte_cpu_to_le_64(rte_mbuf_data_iova_default(rxm));
desc->addr = dma_addr;
desc->len = buf_size;
size = buf_size;
desc->opcode = (nsegs > 1) ? IONIC_RXQ_DESC_OPCODE_SG :
IONIC_RXQ_DESC_OPCODE_SIMPLE;
rxm->next = NULL;
prev_rxm_seg = rxm;
sg_desc = &sg_desc_base[q->head_idx];
elem = sg_desc->elems;
for (j = 0; j < nsegs - 1 && j < IONIC_RX_MAX_SG_ELEMS; j++) {
struct rte_mbuf *rxm_seg;
rte_iova_t data_iova;
rxm_seg = rte_mbuf_raw_alloc(rxq->mb_pool);
if (rxm_seg == NULL) {
IONIC_PRINT(ERR, "RX mbuf alloc failed");
return -ENOMEM;
}
data_iova = rte_mbuf_data_iova(rxm_seg);
dma_addr = rte_cpu_to_le_64(data_iova);
elem->addr = dma_addr;
elem->len = buf_size;
size += buf_size;
elem++;
rxm_seg->next = NULL;
prev_rxm_seg->next = rxm_seg;
prev_rxm_seg = rxm_seg;
}
if (size < len)
IONIC_PRINT(ERR, "Rx SG size is not sufficient (%d < %d)",
size, len);
ring_doorbell = ((q->head_idx + 1) &
IONIC_RX_RING_DOORBELL_STRIDE) == 0;
ionic_q_post(q, ring_doorbell, ionic_rx_clean, rxm);
}
return 0;
}
/*
* Start Receive Units for specified queue.
*/
int __rte_cold
ionic_dev_rx_queue_start(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
{
uint32_t frame_size = eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
struct ionic_qcq *rxq;
int err;
IONIC_PRINT_CALL();
IONIC_PRINT(DEBUG, "Allocating RX queue buffers (size: %u)",
frame_size);
rxq = eth_dev->data->rx_queues[rx_queue_id];
err = ionic_lif_rxq_init(rxq);
if (err)
return err;
/* Allocate buffers for descriptor rings */
if (ionic_rx_fill(rxq, frame_size) != 0) {
IONIC_PRINT(ERR, "Could not alloc mbuf for queue:%d",
rx_queue_id);
return -1;
}
ionic_qcq_enable(rxq);
eth_dev->data->rx_queue_state[rx_queue_id] =
RTE_ETH_QUEUE_STATE_STARTED;
return 0;
}
static inline void __rte_cold
ionic_rxq_service(struct ionic_cq *cq, uint32_t work_to_do,
void *service_cb_arg)
{
struct ionic_queue *q = cq->bound_q;
struct ionic_desc_info *q_desc_info;
struct ionic_rxq_comp *cq_desc_base = cq->base;
struct ionic_rxq_comp *cq_desc;
bool more;
uint32_t curr_q_tail_idx, curr_cq_tail_idx;
uint32_t work_done = 0;
if (work_to_do == 0)
return;
cq_desc = &cq_desc_base[cq->tail_idx];
while (color_match(cq_desc->pkt_type_color, cq->done_color)) {
curr_cq_tail_idx = cq->tail_idx;
cq->tail_idx = (cq->tail_idx + 1) & (cq->num_descs - 1);
if (cq->tail_idx == 0)
cq->done_color = !cq->done_color;
/* Prefetch the next 4 descriptors */
if ((cq->tail_idx & 0x3) == 0)
rte_prefetch0(&cq_desc_base[cq->tail_idx]);
do {
more = (q->tail_idx != cq_desc->comp_index);
q_desc_info = &q->info[q->tail_idx];
curr_q_tail_idx = q->tail_idx;
q->tail_idx = (q->tail_idx + 1) & (q->num_descs - 1);
/* Prefetch the next 4 descriptors */
if ((q->tail_idx & 0x3) == 0)
/* q desc info */
rte_prefetch0(&q->info[q->tail_idx]);
ionic_rx_clean(q, curr_q_tail_idx, curr_cq_tail_idx,
q_desc_info->cb_arg, service_cb_arg);
} while (more);
if (++work_done == work_to_do)
break;
cq_desc = &cq_desc_base[cq->tail_idx];
}
}
/*
* Stop Receive Units for specified queue.
*/
int __rte_cold
ionic_dev_rx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
{
struct ionic_qcq *rxq;
IONIC_PRINT_CALL();
rxq = eth_dev->data->rx_queues[rx_queue_id];
ionic_qcq_disable(rxq);
/* Flush */
ionic_rxq_service(&rxq->cq, -1, NULL);
ionic_lif_rxq_deinit(rxq);
eth_dev->data->rx_queue_state[rx_queue_id] =
RTE_ETH_QUEUE_STATE_STOPPED;
return 0;
}
uint16_t
ionic_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
uint16_t nb_pkts)
{
struct ionic_qcq *rxq = (struct ionic_qcq *)rx_queue;
uint32_t frame_size =
rxq->lif->eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
struct ionic_cq *cq = &rxq->cq;
struct ionic_rx_service service_cb_arg;
service_cb_arg.rx_pkts = rx_pkts;
service_cb_arg.nb_pkts = nb_pkts;
service_cb_arg.nb_rx = 0;
ionic_rxq_service(cq, nb_pkts, &service_cb_arg);
ionic_rx_fill(rxq, frame_size);
return service_cb_arg.nb_rx;
}