* [dpdk-dev] [Bug 650] af_xdp：the packets in complete_queue can't match the tx_queue
@ 2021-03-09 10:22 bugzilla
2021-03-10 9:33 ` bugzilla
0 siblings, 1 reply; 2+ messages in thread
From: bugzilla @ 2021-03-09 10:22 UTC (permalink / raw)
To: dev
https://bugs.dpdk.org/show_bug.cgi?id=650
Bug ID: 650
Summary: af_xdp：the packets in complete_queue can't match the
tx_queue
Product: DPDK
Version: 20.05
Hardware: x86
OS: Linux
Status: UNCONFIRMED
Severity: normal
Priority: Normal
Component: other
Assignee: dev@dpdk.org
Reporter: huangying-c@360.cn
Target Milestone: ---
kernel:5.9.1
=================================================
NIC:
driver: ixgbe
version: 5.9.1
firmware-version: 0x800006c5, 15.0.27
expansion-rom-version:
bus-info: 0000:03:00.0
supports-statistics: yes
supports-test: yes
supports-eeprom-access: yes
supports-register-dump: yes
supports-priv-flags: yes
ethtool -l eth2
Channel parameters for eth2:
Pre-set maximums:
RX: 0
TX: 0
Other: 1
Combined: 63
Current hardware settings:
RX: 0
TX: 0
Other: 1
Combined: 1
==================================================
appplication args:
EAL = --syslog local1 --log-level=pmd.net.af_xdp:8
--vdev=net_af_xdp,iface=eth2,start_queue=0,queue_count=1 -l0-1
==================================================
I wrote a DNS server using AF_XDP Drivre, and when I stress tested it, I found
that the number of packets that have been sent has been consistently lower (in
complete_queue) than the number that have been put into tx_queue, and this has
always been the case, resulting in poor performance during stress tests.
Then, I added logging to the code.
I'm so sorry, my English is not good, so the Chinese description is attached
英语不好，所以附上中文描述
我用af_xdp
drivre写了个dns的服务，当我压力测试的时候发现发送完成的数据包的数量一直比放进tx_queue的包的数量要少，而且一直是这样，如此一来在压力测试的时候，性能很低。
log is following :
af_xdp_tx_zc(): want to send nb_pkts 8 /*here I want to send 8 packets to the
af_xdp_tx_zc */
pull_umem_cq(): send complete n 3 packets
af_xdp_tx_zc(): submit to tx_ring count 8 packets /*here put the 8 packets into
tx_ring*/
af_xdp_rx_zc(): rcvd 5 packets
reserve_fill_queue_zc(): reserve size 5 for fill queue
af_xdp_tx_zc(): want to send nb_pkts 5
pull_umem_cq(): send complete n 5 packets //here the send completely packets is
5 and not 8
the added log in rte_eth_af_xdp.c:af_xdp_tx_zc
static uint16_t
af_xdp_tx_zc(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
{
struct pkt_tx_queue *txq = queue;
struct xsk_umem_info *umem = txq->umem;
struct rte_mbuf *mbuf;
unsigned long tx_bytes = 0;
int i;
uint32_t idx_tx;
uint16_t count = 0;
struct xdp_desc *desc;
uint64_t addr, offset;
AF_XDP_LOG(DEBUG, "want to send nb_pkts %u\n", nb_pkts); //the log is added
by me
pull_umem_cq(umem, nb_pkts); // here, check last tx sended packets
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)) {
AF_XDP_LOG(DEBUG, "=======nb_pkts %u, count %u\n", nb_pkts,
count);
kick_tx(txq);
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 -
umem->mb_pool->header_size;
offset = rte_pktmbuf_mtod(mbuf, uint64_t) -
(uint64_t)mbuf +
umem->mb_pool->header_size;
offset = offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT;
desc->addr = addr | offset;
count++;
} else {
struct rte_mbuf *local_mbuf =
rte_pktmbuf_alloc(umem->mb_pool);
void *pkt;
if (local_mbuf == NULL)
goto out;
if (!xsk_ring_prod__reserve(&txq->tx, 1, &idx_tx)) {
rte_pktmbuf_free(local_mbuf);
kick_tx(txq);
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 -
umem->mb_pool->header_size;
offset = rte_pktmbuf_mtod(local_mbuf, uint64_t) -
(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;
rte_memcpy(pkt, rte_pktmbuf_mtod(mbuf, void *),
desc->len);
rte_pktmbuf_free(mbuf);
count++;
}
tx_bytes += mbuf->pkt_len;
}
kick_tx(txq);
AF_XDP_LOG(DEBUG, "submit to tx_ring count %u packets\n", count);//the log
is added by me
out:
xsk_ring_prod__submit(&txq->tx, count);
txq->stats.tx_pkts += count;
txq->stats.tx_bytes += tx_bytes;
txq->stats.tx_dropped += nb_pkts - count;
return count;
}
the added log in rte_eth_af_xdp.c:af_xdp_tx_zc
static void
pull_umem_cq(struct xsk_umem_info *umem, int size)
{
struct xsk_ring_cons *cq = &umem->cq;
size_t i, n;
uint32_t idx_cq = 0;
n = xsk_ring_cons__peek(cq, size, &idx_cq);
for (i = 0; i < n; i++) {
uint64_t addr;
addr = *xsk_ring_cons__comp_addr(cq, idx_cq++);
#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 + umem->mb_pool->header_size));
#else
rte_ring_enqueue(umem->buf_ring, (void *)addr);
#endif
}
AF_XDP_LOG(DEBUG, "send complete n %lu packets\n", n);//here the log
xsk_ring_cons__release(cq, n);
}
--
You are receiving this mail because:
You are the assignee for the bug.
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2021-03-10 9:33 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-03-09 10:22 [dpdk-dev] [Bug 650] af_xdp：the packets in complete_queue can't match the tx_queue bugzilla
2021-03-10 9:33 ` bugzilla
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).