/* SPDX-License-Identifier: BSD-3-Clause
- * Copyright(c) 2019 Intel Corporation.
+ * Copyright(c) 2019-2020 Intel Corporation.
*/
#include <unistd.h>
#include <errno.h>
#include <stdlib.h>
#include <string.h>
-#include <poll.h>
#include <netinet/in.h>
#include <net/if.h>
#include <sys/socket.h>
#include <bpf/xsk.h>
#include <rte_ethdev.h>
-#include <rte_ethdev_driver.h>
-#include <rte_ethdev_vdev.h>
+#include <ethdev_driver.h>
+#include <ethdev_vdev.h>
#include <rte_kvargs.h>
#include <rte_bus_vdev.h>
#include <rte_string_fns.h>
#include <rte_branch_prediction.h>
#include <rte_common.h>
-#include <rte_config.h>
#include <rte_dev.h>
#include <rte_eal.h>
#include <rte_ether.h>
#include <rte_log.h>
#include <rte_memory.h>
#include <rte_memzone.h>
+#include <rte_mempool.h>
#include <rte_mbuf.h>
#include <rte_malloc.h>
#include <rte_ring.h>
+#include <rte_spinlock.h>
+
+#include "compat.h"
+
+#ifndef SO_PREFER_BUSY_POLL
+#define SO_PREFER_BUSY_POLL 69
+#endif
+#ifndef SO_BUSY_POLL_BUDGET
+#define SO_BUSY_POLL_BUDGET 70
+#endif
+
#ifndef SOL_XDP
#define SOL_XDP 283
#define PF_XDP AF_XDP
#endif
-static int af_xdp_logtype;
+RTE_LOG_REGISTER_DEFAULT(af_xdp_logtype, NOTICE);
#define AF_XDP_LOG(level, fmt, args...) \
rte_log(RTE_LOG_ ## level, af_xdp_logtype, \
#define ETH_AF_XDP_FRAME_SIZE 2048
#define ETH_AF_XDP_NUM_BUFFERS 4096
-#ifdef XDP_UMEM_UNALIGNED_CHUNK_FLAG
-#define ETH_AF_XDP_MBUF_OVERHEAD 128 /* sizeof(struct rte_mbuf) */
-#define ETH_AF_XDP_DATA_HEADROOM \
- (ETH_AF_XDP_MBUF_OVERHEAD + RTE_PKTMBUF_HEADROOM)
-#else
-#define ETH_AF_XDP_DATA_HEADROOM 0
-#endif
#define ETH_AF_XDP_DFLT_NUM_DESCS XSK_RING_CONS__DEFAULT_NUM_DESCS
#define ETH_AF_XDP_DFLT_START_QUEUE_IDX 0
#define ETH_AF_XDP_DFLT_QUEUE_COUNT 1
+#define ETH_AF_XDP_DFLT_BUSY_BUDGET 64
+#define ETH_AF_XDP_DFLT_BUSY_TIMEOUT 20
-#define ETH_AF_XDP_RX_BATCH_SIZE 32
-#define ETH_AF_XDP_TX_BATCH_SIZE 32
+#define ETH_AF_XDP_RX_BATCH_SIZE XSK_RING_CONS__DEFAULT_NUM_DESCS
+#define ETH_AF_XDP_TX_BATCH_SIZE XSK_RING_CONS__DEFAULT_NUM_DESCS
struct xsk_umem_info {
- struct xsk_ring_prod fq;
- struct xsk_ring_cons cq;
struct xsk_umem *umem;
struct rte_ring *buf_ring;
const struct rte_memzone *mz;
struct rte_mempool *mb_pool;
void *buffer;
+ uint8_t refcnt;
+ uint32_t max_xsks;
};
struct rx_stats {
struct rx_stats stats;
+ struct xsk_ring_prod fq;
+ struct xsk_ring_cons cq;
+
struct pkt_tx_queue *pair;
struct pollfd fds[1];
int xsk_queue_idx;
+ int busy_budget;
};
struct tx_stats {
int queue_cnt;
int max_queue_cnt;
int combined_queue_cnt;
+ bool shared_umem;
+ char prog_path[PATH_MAX];
+ bool custom_prog_configured;
struct rte_ether_addr eth_addr;
#define ETH_AF_XDP_IFACE_ARG "iface"
#define ETH_AF_XDP_START_QUEUE_ARG "start_queue"
#define ETH_AF_XDP_QUEUE_COUNT_ARG "queue_count"
+#define ETH_AF_XDP_SHARED_UMEM_ARG "shared_umem"
+#define ETH_AF_XDP_PROG_ARG "xdp_prog"
+#define ETH_AF_XDP_BUDGET_ARG "busy_budget"
static const char * const valid_arguments[] = {
ETH_AF_XDP_IFACE_ARG,
ETH_AF_XDP_START_QUEUE_ARG,
ETH_AF_XDP_QUEUE_COUNT_ARG,
+ ETH_AF_XDP_SHARED_UMEM_ARG,
+ ETH_AF_XDP_PROG_ARG,
+ ETH_AF_XDP_BUDGET_ARG,
NULL
};
.link_autoneg = ETH_LINK_AUTONEG
};
+/* List which tracks PMDs to facilitate sharing UMEMs across them. */
+struct internal_list {
+ TAILQ_ENTRY(internal_list) next;
+ struct rte_eth_dev *eth_dev;
+};
+
+TAILQ_HEAD(internal_list_head, internal_list);
+static struct internal_list_head internal_list =
+ TAILQ_HEAD_INITIALIZER(internal_list);
+
+static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
+
#if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
static inline int
reserve_fill_queue_zc(struct xsk_umem_info *umem, uint16_t reserve_size,
- struct rte_mbuf **bufs)
+ struct rte_mbuf **bufs, struct xsk_ring_prod *fq)
{
- struct xsk_ring_prod *fq = &umem->fq;
uint32_t idx;
uint16_t i;
uint64_t addr;
fq_addr = xsk_ring_prod__fill_addr(fq, idx++);
- addr = (uint64_t)bufs[i] - (uint64_t)umem->buffer;
+ addr = (uint64_t)bufs[i] - (uint64_t)umem->buffer -
+ umem->mb_pool->header_size;
*fq_addr = addr;
}
#else
static inline int
reserve_fill_queue_cp(struct xsk_umem_info *umem, uint16_t reserve_size,
- struct rte_mbuf **bufs __rte_unused)
+ struct rte_mbuf **bufs __rte_unused,
+ struct xsk_ring_prod *fq)
{
- struct xsk_ring_prod *fq = &umem->fq;
void *addrs[reserve_size];
uint32_t idx;
uint16_t i;
static inline int
reserve_fill_queue(struct xsk_umem_info *umem, uint16_t reserve_size,
- struct rte_mbuf **bufs)
+ struct rte_mbuf **bufs, struct xsk_ring_prod *fq)
{
#if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
- return reserve_fill_queue_zc(umem, reserve_size, bufs);
+ return reserve_fill_queue_zc(umem, reserve_size, bufs, fq);
#else
- return reserve_fill_queue_cp(umem, reserve_size, bufs);
+ return reserve_fill_queue_cp(umem, reserve_size, bufs, fq);
#endif
}
{
struct pkt_rx_queue *rxq = queue;
struct xsk_ring_cons *rx = &rxq->rx;
+ struct xsk_ring_prod *fq = &rxq->fq;
struct xsk_umem_info *umem = rxq->umem;
uint32_t idx_rx = 0;
unsigned long rx_bytes = 0;
- int rcvd, i;
+ int i;
struct rte_mbuf *fq_bufs[ETH_AF_XDP_RX_BATCH_SIZE];
+ nb_pkts = xsk_ring_cons__peek(rx, nb_pkts, &idx_rx);
+
+ if (nb_pkts == 0) {
+ /* we can assume a kernel >= 5.11 is in use if busy polling is
+ * enabled and thus we can safely use the recvfrom() syscall
+ * which is only supported for AF_XDP sockets in kernels >=
+ * 5.11.
+ */
+ if (rxq->busy_budget) {
+ (void)recvfrom(xsk_socket__fd(rxq->xsk), NULL, 0,
+ MSG_DONTWAIT, NULL, NULL);
+ } else if (xsk_ring_prod__needs_wakeup(fq)) {
+ (void)poll(&rxq->fds[0], 1, 1000);
+ }
+
+ return 0;
+ }
+
/* allocate bufs for fill queue replenishment after rx */
if (rte_pktmbuf_alloc_bulk(umem->mb_pool, fq_bufs, nb_pkts)) {
AF_XDP_LOG(DEBUG,
"Failed to get enough buffers for fq.\n");
- return -1;
- }
-
- rcvd = xsk_ring_cons__peek(rx, nb_pkts, &idx_rx);
-
- if (rcvd == 0) {
-#if defined(XDP_USE_NEED_WAKEUP)
- if (xsk_ring_prod__needs_wakeup(&umem->fq))
- (void)poll(rxq->fds, 1, 1000);
-#endif
-
- goto out;
+ /* rollback cached_cons which is added by
+ * xsk_ring_cons__peek
+ */
+ rx->cached_cons -= nb_pkts;
+ return 0;
}
- for (i = 0; i < rcvd; i++) {
+ for (i = 0; i < nb_pkts; i++) {
const struct xdp_desc *desc;
uint64_t addr;
uint32_t len;
addr = xsk_umem__extract_addr(addr);
bufs[i] = (struct rte_mbuf *)
- xsk_umem__get_data(umem->buffer, addr);
- bufs[i]->data_off = offset - sizeof(struct rte_mbuf);
+ xsk_umem__get_data(umem->buffer, addr +
+ umem->mb_pool->header_size);
+ bufs[i]->data_off = offset - sizeof(struct rte_mbuf) -
+ rte_pktmbuf_priv_size(umem->mb_pool) -
+ umem->mb_pool->header_size;
rte_pktmbuf_pkt_len(bufs[i]) = len;
rte_pktmbuf_data_len(bufs[i]) = len;
rx_bytes += len;
}
- xsk_ring_cons__release(rx, rcvd);
-
- (void)reserve_fill_queue(umem, rcvd, fq_bufs);
+ xsk_ring_cons__release(rx, nb_pkts);
+ (void)reserve_fill_queue(umem, nb_pkts, fq_bufs, fq);
/* statistics */
- rxq->stats.rx_pkts += rcvd;
+ rxq->stats.rx_pkts += nb_pkts;
rxq->stats.rx_bytes += rx_bytes;
-out:
- if (rcvd != nb_pkts)
- rte_mempool_put_bulk(umem->mb_pool, (void **)&fq_bufs[rcvd],
- nb_pkts - rcvd);
-
- return rcvd;
+ return nb_pkts;
}
#else
static uint16_t
struct pkt_rx_queue *rxq = queue;
struct xsk_ring_cons *rx = &rxq->rx;
struct xsk_umem_info *umem = rxq->umem;
- struct xsk_ring_prod *fq = &umem->fq;
+ struct xsk_ring_prod *fq = &rxq->fq;
uint32_t idx_rx = 0;
unsigned long rx_bytes = 0;
- int rcvd, i;
+ int i;
uint32_t free_thresh = fq->size >> 1;
struct rte_mbuf *mbufs[ETH_AF_XDP_RX_BATCH_SIZE];
- if (unlikely(rte_pktmbuf_alloc_bulk(rxq->mb_pool, mbufs, nb_pkts) != 0))
- return 0;
+ if (xsk_prod_nb_free(fq, free_thresh) >= free_thresh)
+ (void)reserve_fill_queue(umem, nb_pkts, NULL, fq);
- rcvd = xsk_ring_cons__peek(rx, nb_pkts, &idx_rx);
- if (rcvd == 0) {
+ nb_pkts = xsk_ring_cons__peek(rx, nb_pkts, &idx_rx);
+ if (nb_pkts == 0) {
#if defined(XDP_USE_NEED_WAKEUP)
if (xsk_ring_prod__needs_wakeup(fq))
(void)poll(rxq->fds, 1, 1000);
#endif
-
- goto out;
+ return 0;
}
- if (xsk_prod_nb_free(fq, free_thresh) >= free_thresh)
- (void)reserve_fill_queue(umem, ETH_AF_XDP_RX_BATCH_SIZE, NULL);
+ if (unlikely(rte_pktmbuf_alloc_bulk(rxq->mb_pool, mbufs, nb_pkts))) {
+ /* rollback cached_cons which is added by
+ * xsk_ring_cons__peek
+ */
+ rx->cached_cons -= nb_pkts;
+ return 0;
+ }
- for (i = 0; i < rcvd; i++) {
+ for (i = 0; i < nb_pkts; i++) {
const struct xdp_desc *desc;
uint64_t addr;
uint32_t len;
bufs[i] = mbufs[i];
}
- xsk_ring_cons__release(rx, rcvd);
+ xsk_ring_cons__release(rx, nb_pkts);
/* statistics */
- rxq->stats.rx_pkts += rcvd;
+ rxq->stats.rx_pkts += nb_pkts;
rxq->stats.rx_bytes += rx_bytes;
-out:
- if (rcvd != nb_pkts)
- rte_mempool_put_bulk(rxq->mb_pool, (void **)&mbufs[rcvd],
- nb_pkts - rcvd);
-
- return rcvd;
+ return nb_pkts;
}
#endif
static uint16_t
-eth_af_xdp_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
+af_xdp_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
{
- nb_pkts = RTE_MIN(nb_pkts, ETH_AF_XDP_RX_BATCH_SIZE);
-
#if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
return af_xdp_rx_zc(queue, bufs, nb_pkts);
#else
#endif
}
+static uint16_t
+eth_af_xdp_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
+{
+ uint16_t nb_rx;
+
+ if (likely(nb_pkts <= ETH_AF_XDP_RX_BATCH_SIZE))
+ return af_xdp_rx(queue, bufs, nb_pkts);
+
+ /* Split larger batch into smaller batches of size
+ * ETH_AF_XDP_RX_BATCH_SIZE or less.
+ */
+ nb_rx = 0;
+ while (nb_pkts) {
+ uint16_t ret, n;
+
+ n = (uint16_t)RTE_MIN(nb_pkts, ETH_AF_XDP_RX_BATCH_SIZE);
+ ret = af_xdp_rx(queue, &bufs[nb_rx], n);
+ nb_rx = (uint16_t)(nb_rx + ret);
+ nb_pkts = (uint16_t)(nb_pkts - ret);
+ if (ret < n)
+ break;
+ }
+
+ return nb_rx;
+}
+
static void
-pull_umem_cq(struct xsk_umem_info *umem, int size)
+pull_umem_cq(struct xsk_umem_info *umem, int size, struct xsk_ring_cons *cq)
{
- struct xsk_ring_cons *cq = &umem->cq;
size_t i, n;
uint32_t idx_cq = 0;
#if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
addr = xsk_umem__extract_addr(addr);
rte_pktmbuf_free((struct rte_mbuf *)
- xsk_umem__get_data(umem->buffer, addr));
+ xsk_umem__get_data(umem->buffer,
+ addr + umem->mb_pool->header_size));
#else
rte_ring_enqueue(umem->buf_ring, (void *)addr);
#endif
}
static void
-kick_tx(struct pkt_tx_queue *txq)
+kick_tx(struct pkt_tx_queue *txq, struct xsk_ring_cons *cq)
{
struct xsk_umem_info *umem = txq->umem;
-#if defined(XDP_USE_NEED_WAKEUP)
- if (xsk_ring_prod__needs_wakeup(&txq->tx))
-#endif
+ pull_umem_cq(umem, XSK_RING_CONS__DEFAULT_NUM_DESCS, cq);
+
+ if (tx_syscall_needed(&txq->tx))
while (send(xsk_socket__fd(txq->pair->xsk), NULL,
0, MSG_DONTWAIT) < 0) {
/* some thing unexpected */
/* pull from completion queue to leave more space */
if (errno == EAGAIN)
- pull_umem_cq(umem, ETH_AF_XDP_TX_BATCH_SIZE);
+ pull_umem_cq(umem,
+ XSK_RING_CONS__DEFAULT_NUM_DESCS,
+ cq);
}
-#ifndef XDP_UMEM_UNALIGNED_CHUNK_FLAG
- pull_umem_cq(umem, ETH_AF_XDP_TX_BATCH_SIZE);
-#endif
}
#if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
uint16_t count = 0;
struct xdp_desc *desc;
uint64_t addr, offset;
+ struct xsk_ring_cons *cq = &txq->pair->cq;
+ uint32_t free_thresh = cq->size >> 1;
- pull_umem_cq(umem, nb_pkts);
+ if (xsk_cons_nb_avail(cq, free_thresh) >= free_thresh)
+ pull_umem_cq(umem, XSK_RING_CONS__DEFAULT_NUM_DESCS, cq);
for (i = 0; i < nb_pkts; i++) {
mbuf = bufs[i];
if (mbuf->pool == umem->mb_pool) {
if (!xsk_ring_prod__reserve(&txq->tx, 1, &idx_tx)) {
- kick_tx(txq);
- goto out;
+ kick_tx(txq, cq);
+ if (!xsk_ring_prod__reserve(&txq->tx, 1,
+ &idx_tx))
+ goto out;
}
desc = xsk_ring_prod__tx_desc(&txq->tx, idx_tx);
desc->len = mbuf->pkt_len;
- addr = (uint64_t)mbuf - (uint64_t)umem->buffer;
+ addr = (uint64_t)mbuf - (uint64_t)umem->buffer -
+ umem->mb_pool->header_size;
offset = rte_pktmbuf_mtod(mbuf, uint64_t) -
- (uint64_t)mbuf;
+ (uint64_t)mbuf +
+ umem->mb_pool->header_size;
offset = offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT;
desc->addr = addr | offset;
count++;
if (!xsk_ring_prod__reserve(&txq->tx, 1, &idx_tx)) {
rte_pktmbuf_free(local_mbuf);
- kick_tx(txq);
+ kick_tx(txq, cq);
goto out;
}
desc = xsk_ring_prod__tx_desc(&txq->tx, idx_tx);
desc->len = mbuf->pkt_len;
- addr = (uint64_t)local_mbuf - (uint64_t)umem->buffer;
+ addr = (uint64_t)local_mbuf - (uint64_t)umem->buffer -
+ umem->mb_pool->header_size;
offset = rte_pktmbuf_mtod(local_mbuf, uint64_t) -
- (uint64_t)local_mbuf;
+ (uint64_t)local_mbuf +
+ umem->mb_pool->header_size;
pkt = xsk_umem__get_data(umem->buffer, addr + offset);
offset = offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT;
desc->addr = addr | offset;
tx_bytes += mbuf->pkt_len;
}
- kick_tx(txq);
+ kick_tx(txq, cq);
out:
xsk_ring_prod__submit(&txq->tx, count);
unsigned long tx_bytes = 0;
int i;
uint32_t idx_tx;
+ struct xsk_ring_cons *cq = &txq->pair->cq;
- nb_pkts = RTE_MIN(nb_pkts, ETH_AF_XDP_TX_BATCH_SIZE);
-
- pull_umem_cq(umem, nb_pkts);
+ pull_umem_cq(umem, nb_pkts, cq);
nb_pkts = rte_ring_dequeue_bulk(umem->buf_ring, addrs,
nb_pkts, NULL);
return 0;
if (xsk_ring_prod__reserve(&txq->tx, nb_pkts, &idx_tx) != nb_pkts) {
- kick_tx(txq);
+ kick_tx(txq, cq);
rte_ring_enqueue_bulk(umem->buf_ring, addrs, nb_pkts, NULL);
return 0;
}
xsk_ring_prod__submit(&txq->tx, nb_pkts);
- kick_tx(txq);
+ kick_tx(txq, cq);
txq->stats.tx_pkts += nb_pkts;
txq->stats.tx_bytes += tx_bytes;
return nb_pkts;
}
+
+static uint16_t
+af_xdp_tx_cp_batch(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
+{
+ uint16_t nb_tx;
+
+ if (likely(nb_pkts <= ETH_AF_XDP_TX_BATCH_SIZE))
+ return af_xdp_tx_cp(queue, bufs, nb_pkts);
+
+ nb_tx = 0;
+ while (nb_pkts) {
+ uint16_t ret, n;
+
+ /* Split larger batch into smaller batches of size
+ * ETH_AF_XDP_TX_BATCH_SIZE or less.
+ */
+ n = (uint16_t)RTE_MIN(nb_pkts, ETH_AF_XDP_TX_BATCH_SIZE);
+ ret = af_xdp_tx_cp(queue, &bufs[nb_tx], n);
+ nb_tx = (uint16_t)(nb_tx + ret);
+ nb_pkts = (uint16_t)(nb_pkts - ret);
+ if (ret < n)
+ break;
+ }
+
+ return nb_tx;
+}
#endif
static uint16_t
#if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
return af_xdp_tx_zc(queue, bufs, nb_pkts);
#else
- return af_xdp_tx_cp(queue, bufs, nb_pkts);
+ return af_xdp_tx_cp_batch(queue, bufs, nb_pkts);
#endif
}
}
/* This function gets called when the current port gets stopped. */
-static void
+static int
eth_dev_stop(struct rte_eth_dev *dev)
{
dev->data->dev_link.link_status = ETH_LINK_DOWN;
+ return 0;
+}
+
+/* Find ethdev in list */
+static inline struct internal_list *
+find_internal_resource(struct pmd_internals *port_int)
+{
+ int found = 0;
+ struct internal_list *list = NULL;
+
+ if (port_int == NULL)
+ return NULL;
+
+ pthread_mutex_lock(&internal_list_lock);
+
+ TAILQ_FOREACH(list, &internal_list, next) {
+ struct pmd_internals *list_int =
+ list->eth_dev->data->dev_private;
+ if (list_int == port_int) {
+ found = 1;
+ break;
+ }
+ }
+
+ pthread_mutex_unlock(&internal_list_lock);
+
+ if (!found)
+ return NULL;
+
+ return list;
+}
+
+/* Check if the netdev,qid context already exists */
+static inline bool
+ctx_exists(struct pkt_rx_queue *rxq, const char *ifname,
+ struct pkt_rx_queue *list_rxq, const char *list_ifname)
+{
+ bool exists = false;
+
+ if (rxq->xsk_queue_idx == list_rxq->xsk_queue_idx &&
+ !strncmp(ifname, list_ifname, IFNAMSIZ)) {
+ AF_XDP_LOG(ERR, "ctx %s,%i already exists, cannot share umem\n",
+ ifname, rxq->xsk_queue_idx);
+ exists = true;
+ }
+
+ return exists;
+}
+
+/* Get a pointer to an existing UMEM which overlays the rxq's mb_pool */
+static inline int
+get_shared_umem(struct pkt_rx_queue *rxq, const char *ifname,
+ struct xsk_umem_info **umem)
+{
+ struct internal_list *list;
+ struct pmd_internals *internals;
+ int i = 0, ret = 0;
+ struct rte_mempool *mb_pool = rxq->mb_pool;
+
+ if (mb_pool == NULL)
+ return ret;
+
+ pthread_mutex_lock(&internal_list_lock);
+
+ TAILQ_FOREACH(list, &internal_list, next) {
+ internals = list->eth_dev->data->dev_private;
+ for (i = 0; i < internals->queue_cnt; i++) {
+ struct pkt_rx_queue *list_rxq =
+ &internals->rx_queues[i];
+ if (rxq == list_rxq)
+ continue;
+ if (mb_pool == internals->rx_queues[i].mb_pool) {
+ if (ctx_exists(rxq, ifname, list_rxq,
+ internals->if_name)) {
+ ret = -1;
+ goto out;
+ }
+ if (__atomic_load_n(
+ &internals->rx_queues[i].umem->refcnt,
+ __ATOMIC_ACQUIRE)) {
+ *umem = internals->rx_queues[i].umem;
+ goto out;
+ }
+ }
+ }
+ }
+
+out:
+ pthread_mutex_unlock(&internal_list_lock);
+
+ return ret;
}
static int
eth_dev_configure(struct rte_eth_dev *dev)
{
+ struct pmd_internals *internal = dev->data->dev_private;
+
/* rx/tx must be paired */
if (dev->data->nb_rx_queues != dev->data->nb_tx_queues)
return -EINVAL;
+ if (internal->shared_umem) {
+ struct internal_list *list = NULL;
+ const char *name = dev->device->name;
+
+ /* Ensure PMD is not already inserted into the list */
+ list = find_internal_resource(internal);
+ if (list)
+ return 0;
+
+ list = rte_zmalloc_socket(name, sizeof(*list), 0,
+ dev->device->numa_node);
+ if (list == NULL)
+ return -1;
+
+ list->eth_dev = dev;
+ pthread_mutex_lock(&internal_list_lock);
+ TAILQ_INSERT_TAIL(&internal_list, list, next);
+ pthread_mutex_unlock(&internal_list_lock);
+ }
+
return 0;
}
dev_info->max_tx_queues = internals->queue_cnt;
dev_info->min_mtu = RTE_ETHER_MIN_MTU;
- dev_info->max_mtu = ETH_AF_XDP_FRAME_SIZE - ETH_AF_XDP_DATA_HEADROOM;
+#if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
+ dev_info->max_mtu = getpagesize() -
+ sizeof(struct rte_mempool_objhdr) -
+ sizeof(struct rte_mbuf) -
+ RTE_PKTMBUF_HEADROOM - XDP_PACKET_HEADROOM;
+#else
+ dev_info->max_mtu = ETH_AF_XDP_FRAME_SIZE - XDP_PACKET_HEADROOM;
+#endif
+ dev_info->default_rxportconf.burst_size = ETH_AF_XDP_DFLT_BUSY_BUDGET;
+ dev_info->default_txportconf.burst_size = ETH_AF_XDP_DFLT_BUSY_BUDGET;
dev_info->default_rxportconf.nb_queues = 1;
dev_info->default_txportconf.nb_queues = 1;
dev_info->default_rxportconf.ring_size = ETH_AF_XDP_DFLT_NUM_DESCS;
xdp_umem_destroy(struct xsk_umem_info *umem)
{
#if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
- rte_mempool_free(umem->mb_pool);
umem->mb_pool = NULL;
#else
rte_memzone_free(umem->mz);
#endif
rte_free(umem);
- umem = NULL;
}
-static void
+static int
eth_dev_close(struct rte_eth_dev *dev)
{
struct pmd_internals *internals = dev->data->dev_private;
struct pkt_rx_queue *rxq;
int i;
+ if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+ return 0;
+
AF_XDP_LOG(INFO, "Closing AF_XDP ethdev on numa socket %u\n",
rte_socket_id());
if (rxq->umem == NULL)
break;
xsk_socket__delete(rxq->xsk);
- (void)xsk_umem__delete(rxq->umem->umem);
- xdp_umem_destroy(rxq->umem);
+
+ if (__atomic_sub_fetch(&rxq->umem->refcnt, 1, __ATOMIC_ACQUIRE)
+ == 0) {
+ (void)xsk_umem__delete(rxq->umem->umem);
+ xdp_umem_destroy(rxq->umem);
+ }
/* free pkt_tx_queue */
rte_free(rxq->pair);
dev->data->mac_addrs = NULL;
remove_xdp_program(internals);
+
+ if (internals->shared_umem) {
+ struct internal_list *list;
+
+ /* Remove ethdev from list used to track and share UMEMs */
+ list = find_internal_resource(internals);
+ if (list) {
+ pthread_mutex_lock(&internal_list_lock);
+ TAILQ_REMOVE(&internal_list, list, next);
+ pthread_mutex_unlock(&internal_list_lock);
+ rte_free(list);
+ }
+ }
+
+ return 0;
}
static void
}
#if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
-static inline uint64_t get_base_addr(struct rte_mempool *mp)
+static inline uintptr_t get_base_addr(struct rte_mempool *mp, uint64_t *align)
{
struct rte_mempool_memhdr *memhdr;
+ uintptr_t memhdr_addr, aligned_addr;
memhdr = STAILQ_FIRST(&mp->mem_list);
- return (uint64_t)memhdr->addr & ~(getpagesize() - 1);
+ memhdr_addr = (uintptr_t)memhdr->addr;
+ aligned_addr = memhdr_addr & ~(getpagesize() - 1);
+ *align = memhdr_addr - aligned_addr;
+
+ return aligned_addr;
}
static struct
-xsk_umem_info *xdp_umem_configure(struct pmd_internals *internals __rte_unused,
+xsk_umem_info *xdp_umem_configure(struct pmd_internals *internals,
struct pkt_rx_queue *rxq)
{
- struct xsk_umem_info *umem;
+ struct xsk_umem_info *umem = NULL;
int ret;
struct xsk_umem_config usr_config = {
- .fill_size = ETH_AF_XDP_DFLT_NUM_DESCS,
+ .fill_size = ETH_AF_XDP_DFLT_NUM_DESCS * 2,
.comp_size = ETH_AF_XDP_DFLT_NUM_DESCS,
.flags = XDP_UMEM_UNALIGNED_CHUNK_FLAG};
void *base_addr = NULL;
struct rte_mempool *mb_pool = rxq->mb_pool;
+ uint64_t umem_size, align = 0;
+
+ if (internals->shared_umem) {
+ if (get_shared_umem(rxq, internals->if_name, &umem) < 0)
+ return NULL;
+
+ if (umem != NULL &&
+ __atomic_load_n(&umem->refcnt, __ATOMIC_ACQUIRE) <
+ umem->max_xsks) {
+ AF_XDP_LOG(INFO, "%s,qid%i sharing UMEM\n",
+ internals->if_name, rxq->xsk_queue_idx);
+ __atomic_fetch_add(&umem->refcnt, 1, __ATOMIC_ACQUIRE);
+ }
+ }
- usr_config.frame_size = rte_pktmbuf_data_room_size(mb_pool) +
- ETH_AF_XDP_MBUF_OVERHEAD +
- mb_pool->private_data_size;
- usr_config.frame_headroom = ETH_AF_XDP_DATA_HEADROOM +
- mb_pool->private_data_size;
-
- umem = rte_zmalloc_socket("umem", sizeof(*umem), 0, rte_socket_id());
if (umem == NULL) {
- AF_XDP_LOG(ERR, "Failed to allocate umem info");
- return NULL;
- }
+ usr_config.frame_size =
+ rte_mempool_calc_obj_size(mb_pool->elt_size,
+ mb_pool->flags, NULL);
+ usr_config.frame_headroom = mb_pool->header_size +
+ sizeof(struct rte_mbuf) +
+ rte_pktmbuf_priv_size(mb_pool) +
+ RTE_PKTMBUF_HEADROOM;
+
+ umem = rte_zmalloc_socket("umem", sizeof(*umem), 0,
+ rte_socket_id());
+ if (umem == NULL) {
+ AF_XDP_LOG(ERR, "Failed to allocate umem info");
+ return NULL;
+ }
- umem->mb_pool = mb_pool;
- base_addr = (void *)get_base_addr(mb_pool);
+ umem->mb_pool = mb_pool;
+ base_addr = (void *)get_base_addr(mb_pool, &align);
+ umem_size = (uint64_t)mb_pool->populated_size *
+ (uint64_t)usr_config.frame_size +
+ align;
+
+ ret = xsk_umem__create(&umem->umem, base_addr, umem_size,
+ &rxq->fq, &rxq->cq, &usr_config);
+ if (ret) {
+ AF_XDP_LOG(ERR, "Failed to create umem");
+ goto err;
+ }
+ umem->buffer = base_addr;
- ret = xsk_umem__create(&umem->umem, base_addr,
- mb_pool->populated_size * usr_config.frame_size,
- &umem->fq, &umem->cq,
- &usr_config);
+ if (internals->shared_umem) {
+ umem->max_xsks = mb_pool->populated_size /
+ ETH_AF_XDP_NUM_BUFFERS;
+ AF_XDP_LOG(INFO, "Max xsks for UMEM %s: %u\n",
+ mb_pool->name, umem->max_xsks);
+ }
- if (ret) {
- AF_XDP_LOG(ERR, "Failed to create umem");
- goto err;
+ __atomic_store_n(&umem->refcnt, 1, __ATOMIC_RELEASE);
}
- umem->buffer = base_addr;
#else
static struct
.fill_size = ETH_AF_XDP_DFLT_NUM_DESCS,
.comp_size = ETH_AF_XDP_DFLT_NUM_DESCS,
.frame_size = ETH_AF_XDP_FRAME_SIZE,
- .frame_headroom = ETH_AF_XDP_DATA_HEADROOM };
+ .frame_headroom = 0 };
char ring_name[RTE_RING_NAMESIZE];
char mz_name[RTE_MEMZONE_NAMESIZE];
int ret;
for (i = 0; i < ETH_AF_XDP_NUM_BUFFERS; i++)
rte_ring_enqueue(umem->buf_ring,
- (void *)(i * ETH_AF_XDP_FRAME_SIZE +
- ETH_AF_XDP_DATA_HEADROOM));
+ (void *)(i * ETH_AF_XDP_FRAME_SIZE));
snprintf(mz_name, sizeof(mz_name), "af_xdp_umem_%s_%u",
internals->if_name, rxq->xsk_queue_idx);
ret = xsk_umem__create(&umem->umem, mz->addr,
ETH_AF_XDP_NUM_BUFFERS * ETH_AF_XDP_FRAME_SIZE,
- &umem->fq, &umem->cq,
+ &rxq->fq, &rxq->cq,
&usr_config);
if (ret) {
return NULL;
}
+static int
+load_custom_xdp_prog(const char *prog_path, int if_index)
+{
+ int ret, prog_fd = -1;
+ struct bpf_object *obj;
+ struct bpf_map *map;
+
+ ret = bpf_prog_load(prog_path, BPF_PROG_TYPE_XDP, &obj, &prog_fd);
+ if (ret) {
+ AF_XDP_LOG(ERR, "Failed to load program %s\n", prog_path);
+ return ret;
+ }
+
+ /*
+ * The loaded program must provision for a map of xsks, such that some
+ * traffic can be redirected to userspace. When the xsk is created,
+ * libbpf inserts it into the map.
+ */
+ map = bpf_object__find_map_by_name(obj, "xsks_map");
+ if (!map) {
+ AF_XDP_LOG(ERR, "Failed to find xsks_map in %s\n", prog_path);
+ return -1;
+ }
+
+ /* Link the program with the given network device */
+ ret = bpf_set_link_xdp_fd(if_index, prog_fd,
+ XDP_FLAGS_UPDATE_IF_NOEXIST);
+ if (ret) {
+ AF_XDP_LOG(ERR, "Failed to set prog fd %d on interface\n",
+ prog_fd);
+ return -1;
+ }
+
+ AF_XDP_LOG(INFO, "Successfully loaded XDP program %s with fd %d\n",
+ prog_path, prog_fd);
+
+ return 0;
+}
+
+/* Detect support for busy polling through setsockopt(). */
+static int
+configure_preferred_busy_poll(struct pkt_rx_queue *rxq)
+{
+ int sock_opt = 1;
+ int fd = xsk_socket__fd(rxq->xsk);
+ int ret = 0;
+
+ ret = setsockopt(fd, SOL_SOCKET, SO_PREFER_BUSY_POLL,
+ (void *)&sock_opt, sizeof(sock_opt));
+ if (ret < 0) {
+ AF_XDP_LOG(DEBUG, "Failed to set SO_PREFER_BUSY_POLL\n");
+ goto err_prefer;
+ }
+
+ sock_opt = ETH_AF_XDP_DFLT_BUSY_TIMEOUT;
+ ret = setsockopt(fd, SOL_SOCKET, SO_BUSY_POLL, (void *)&sock_opt,
+ sizeof(sock_opt));
+ if (ret < 0) {
+ AF_XDP_LOG(DEBUG, "Failed to set SO_BUSY_POLL\n");
+ goto err_timeout;
+ }
+
+ sock_opt = rxq->busy_budget;
+ ret = setsockopt(fd, SOL_SOCKET, SO_BUSY_POLL_BUDGET,
+ (void *)&sock_opt, sizeof(sock_opt));
+ if (ret < 0) {
+ AF_XDP_LOG(DEBUG, "Failed to set SO_BUSY_POLL_BUDGET\n");
+ } else {
+ AF_XDP_LOG(INFO, "Busy polling budget set to: %u\n",
+ rxq->busy_budget);
+ return 0;
+ }
+
+ /* setsockopt failure - attempt to restore xsk to default state and
+ * proceed without busy polling support.
+ */
+ sock_opt = 0;
+ ret = setsockopt(fd, SOL_SOCKET, SO_BUSY_POLL, (void *)&sock_opt,
+ sizeof(sock_opt));
+ if (ret < 0) {
+ AF_XDP_LOG(ERR, "Failed to unset SO_BUSY_POLL\n");
+ return -1;
+ }
+
+err_timeout:
+ sock_opt = 0;
+ ret = setsockopt(fd, SOL_SOCKET, SO_PREFER_BUSY_POLL,
+ (void *)&sock_opt, sizeof(sock_opt));
+ if (ret < 0) {
+ AF_XDP_LOG(ERR, "Failed to unset SO_PREFER_BUSY_POLL\n");
+ return -1;
+ }
+
+err_prefer:
+ rxq->busy_budget = 0;
+ return 0;
+}
+
static int
xsk_configure(struct pmd_internals *internals, struct pkt_rx_queue *rxq,
int ring_size)
struct xsk_socket_config cfg;
struct pkt_tx_queue *txq = rxq->pair;
int ret = 0;
- int reserve_size = ETH_AF_XDP_DFLT_NUM_DESCS / 2;
+ int reserve_size = ETH_AF_XDP_DFLT_NUM_DESCS;
struct rte_mbuf *fq_bufs[reserve_size];
rxq->umem = xdp_umem_configure(internals, rxq);
cfg.bind_flags |= XDP_USE_NEED_WAKEUP;
#endif
- ret = xsk_socket__create(&rxq->xsk, internals->if_name,
- rxq->xsk_queue_idx, rxq->umem->umem, &rxq->rx,
- &txq->tx, &cfg);
+ if (strnlen(internals->prog_path, PATH_MAX) &&
+ !internals->custom_prog_configured) {
+ ret = load_custom_xdp_prog(internals->prog_path,
+ internals->if_index);
+ if (ret) {
+ AF_XDP_LOG(ERR, "Failed to load custom XDP program %s\n",
+ internals->prog_path);
+ goto err;
+ }
+ internals->custom_prog_configured = 1;
+ }
+
+ if (internals->shared_umem)
+ ret = create_shared_socket(&rxq->xsk, internals->if_name,
+ rxq->xsk_queue_idx, rxq->umem->umem, &rxq->rx,
+ &txq->tx, &rxq->fq, &rxq->cq, &cfg);
+ else
+ ret = xsk_socket__create(&rxq->xsk, internals->if_name,
+ rxq->xsk_queue_idx, rxq->umem->umem, &rxq->rx,
+ &txq->tx, &cfg);
+
if (ret) {
AF_XDP_LOG(ERR, "Failed to create xsk socket.\n");
goto err;
}
#if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
- if (rte_pktmbuf_alloc_bulk(rxq->umem->mb_pool, fq_bufs, reserve_size)) {
+ ret = rte_pktmbuf_alloc_bulk(rxq->umem->mb_pool, fq_bufs, reserve_size);
+ if (ret) {
AF_XDP_LOG(DEBUG, "Failed to get enough buffers for fq.\n");
goto err;
}
#endif
- ret = reserve_fill_queue(rxq->umem, reserve_size, fq_bufs);
+
+ if (rxq->busy_budget) {
+ ret = configure_preferred_busy_poll(rxq);
+ if (ret) {
+ AF_XDP_LOG(ERR, "Failed configure busy polling.\n");
+ goto err;
+ }
+ }
+
+ ret = reserve_fill_queue(rxq->umem, reserve_size, fq_bufs, &rxq->fq);
if (ret) {
xsk_socket__delete(rxq->xsk);
AF_XDP_LOG(ERR, "Failed to reserve fill queue.\n");
return 0;
err:
- xdp_umem_destroy(rxq->umem);
+ if (__atomic_sub_fetch(&rxq->umem->refcnt, 1, __ATOMIC_ACQUIRE) == 0)
+ xdp_umem_destroy(rxq->umem);
return ret;
}
/* Now get the space available for data in the mbuf */
buf_size = rte_pktmbuf_data_room_size(mb_pool) -
RTE_PKTMBUF_HEADROOM;
- data_size = ETH_AF_XDP_FRAME_SIZE - ETH_AF_XDP_DATA_HEADROOM;
+ data_size = ETH_AF_XDP_FRAME_SIZE;
if (data_size > buf_size) {
AF_XDP_LOG(ERR, "%s: %d bytes will not fit in mbuf (%d bytes)\n",
goto err;
}
+ if (!rxq->busy_budget)
+ AF_XDP_LOG(DEBUG, "Preferred busy polling not enabled\n");
+
rxq->fds[0].fd = xsk_socket__fd(rxq->xsk);
rxq->fds[0].events = POLLIN;
.stats_reset = eth_stats_reset,
};
+/** parse busy_budget argument */
+static int
+parse_budget_arg(const char *key __rte_unused,
+ const char *value, void *extra_args)
+{
+ int *i = (int *)extra_args;
+ char *end;
+
+ *i = strtol(value, &end, 10);
+ if (*i < 0 || *i > UINT16_MAX) {
+ AF_XDP_LOG(ERR, "Invalid busy_budget %i, must be >= 0 and <= %u\n",
+ *i, UINT16_MAX);
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
/** parse integer from integer argument */
static int
parse_integer_arg(const char *key __rte_unused,
return 0;
}
+/** parse xdp prog argument */
+static int
+parse_prog_arg(const char *key __rte_unused,
+ const char *value, void *extra_args)
+{
+ char *path = extra_args;
+
+ if (strnlen(value, PATH_MAX) == PATH_MAX) {
+ AF_XDP_LOG(ERR, "Invalid path %s, should be less than %u bytes.\n",
+ value, PATH_MAX);
+ return -EINVAL;
+ }
+
+ if (access(value, F_OK) != 0) {
+ AF_XDP_LOG(ERR, "Error accessing %s: %s\n",
+ value, strerror(errno));
+ return -EINVAL;
+ }
+
+ strlcpy(path, value, PATH_MAX);
+
+ return 0;
+}
+
static int
xdp_get_channels_info(const char *if_name, int *max_queues,
int *combined_queues)
channels.cmd = ETHTOOL_GCHANNELS;
ifr.ifr_data = (void *)&channels;
- strncpy(ifr.ifr_name, if_name, IFNAMSIZ);
+ strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
ret = ioctl(fd, SIOCETHTOOL, &ifr);
if (ret) {
if (errno == EOPNOTSUPP) {
static int
parse_parameters(struct rte_kvargs *kvlist, char *if_name, int *start_queue,
- int *queue_cnt)
+ int *queue_cnt, int *shared_umem, char *prog_path,
+ int *busy_budget)
{
int ret;
goto free_kvlist;
}
+ ret = rte_kvargs_process(kvlist, ETH_AF_XDP_SHARED_UMEM_ARG,
+ &parse_integer_arg, shared_umem);
+ if (ret < 0)
+ goto free_kvlist;
+
+ ret = rte_kvargs_process(kvlist, ETH_AF_XDP_PROG_ARG,
+ &parse_prog_arg, prog_path);
+ if (ret < 0)
+ goto free_kvlist;
+
+ ret = rte_kvargs_process(kvlist, ETH_AF_XDP_BUDGET_ARG,
+ &parse_budget_arg, busy_budget);
+ if (ret < 0)
+ goto free_kvlist;
+
free_kvlist:
rte_kvargs_free(kvlist);
return ret;
static struct rte_eth_dev *
init_internals(struct rte_vdev_device *dev, const char *if_name,
- int start_queue_idx, int queue_cnt)
+ int start_queue_idx, int queue_cnt, int shared_umem,
+ const char *prog_path, int busy_budget)
{
const char *name = rte_vdev_device_name(dev);
const unsigned int numa_node = dev->device.numa_node;
internals->start_queue_idx = start_queue_idx;
internals->queue_cnt = queue_cnt;
strlcpy(internals->if_name, if_name, IFNAMSIZ);
+ strlcpy(internals->prog_path, prog_path, PATH_MAX);
+ internals->custom_prog_configured = 0;
+
+#ifndef ETH_AF_XDP_SHARED_UMEM
+ if (shared_umem) {
+ AF_XDP_LOG(ERR, "Shared UMEM feature not available. "
+ "Check kernel and libbpf version\n");
+ goto err_free_internals;
+ }
+#endif
+ internals->shared_umem = shared_umem;
if (xdp_get_channels_info(if_name, &internals->max_queue_cnt,
&internals->combined_queue_cnt)) {
internals->rx_queues[i].pair = &internals->tx_queues[i];
internals->rx_queues[i].xsk_queue_idx = start_queue_idx + i;
internals->tx_queues[i].xsk_queue_idx = start_queue_idx + i;
+ internals->rx_queues[i].busy_budget = busy_budget;
}
ret = get_iface_info(if_name, &internals->eth_addr,
eth_dev->data->dev_private = internals;
eth_dev->data->dev_link = pmd_link;
eth_dev->data->mac_addrs = &internals->eth_addr;
+ eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
eth_dev->dev_ops = &ops;
eth_dev->rx_pkt_burst = eth_af_xdp_rx;
eth_dev->tx_pkt_burst = eth_af_xdp_tx;
- /* Let rte_eth_dev_close() release the port resources. */
- eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
#if defined(XDP_UMEM_UNALIGNED_CHUNK_FLAG)
AF_XDP_LOG(INFO, "Zero copy between umem and mbuf enabled.\n");
char if_name[IFNAMSIZ] = {'\0'};
int xsk_start_queue_idx = ETH_AF_XDP_DFLT_START_QUEUE_IDX;
int xsk_queue_cnt = ETH_AF_XDP_DFLT_QUEUE_COUNT;
+ int shared_umem = 0;
+ char prog_path[PATH_MAX] = {'\0'};
+ int busy_budget = -1;
struct rte_eth_dev *eth_dev = NULL;
const char *name;
dev->device.numa_node = rte_socket_id();
if (parse_parameters(kvlist, if_name, &xsk_start_queue_idx,
- &xsk_queue_cnt) < 0) {
+ &xsk_queue_cnt, &shared_umem, prog_path,
+ &busy_budget) < 0) {
AF_XDP_LOG(ERR, "Invalid kvargs value\n");
return -EINVAL;
}
return -EINVAL;
}
+ busy_budget = busy_budget == -1 ? ETH_AF_XDP_DFLT_BUSY_BUDGET :
+ busy_budget;
+
eth_dev = init_internals(dev, if_name, xsk_start_queue_idx,
- xsk_queue_cnt);
+ xsk_queue_cnt, shared_umem, prog_path,
+ busy_budget);
if (eth_dev == NULL) {
AF_XDP_LOG(ERR, "Failed to init internals\n");
return -1;
RTE_PMD_REGISTER_PARAM_STRING(net_af_xdp,
"iface=<string> "
"start_queue=<int> "
- "queue_count=<int> ");
-
-RTE_INIT(af_xdp_init_log)
-{
- af_xdp_logtype = rte_log_register("pmd.net.af_xdp");
- if (af_xdp_logtype >= 0)
- rte_log_set_level(af_xdp_logtype, RTE_LOG_NOTICE);
-}
+ "queue_count=<int> "
+ "shared_umem=<int> "
+ "xdp_prog=<string> "
+ "busy_budget=<int>");