From: Matteo Lanzuisi <m.lanzuisi@resi.it>
To: "dev@dpdk.org" <dev@dpdk.org>
Subject: [dpdk-dev] Symmetry for TCP packets on X710 Intel
Date: Tue, 2 Oct 2018 19:19:17 +0200 [thread overview]
Message-ID: <384bb9db-1023-65f7-6d18-f1311db9228a@resi.it> (raw)
Hi all,
I got a huge problem: I found that using normal hashing and
configuration on X710 hardware is not sufficient, so I followed some
threads and read the code inside the testpmd application to configure a
symmetric hashing for IP src / IP dst on TCP packets so that all packets
in a flow are sent to the same lcore.
This is not working (printing mbuf.hash.rss gives different values for
different directions) and I cannot understand why.
I am using dpdk 18.02.2 on RedHat 7.5.
The code I'm using (just the part needed by this thread) is:
// haskey passed to i40e
static uint8_t hashkey[40] = {0x6D, 0x5A, 0x6D, 0x5A, 0x6D, 0x5A, 0x6D,
0x5A,
0x6D, 0x5A, 0x6D, 0x5A, 0x6D, 0x5A, 0x6D,
0x5A,
0x6D, 0x5A, 0x6D, 0x5A, 0x6D, 0x5A, 0x6D,
0x5A,
0x6D, 0x5A, 0x6D, 0x5A, 0x6D, 0x5A, 0x6D,
0x5A,
0x6D, 0x5A, 0x6D, 0x5A, 0x6D, 0x5A, 0x6D,
0x5A};
struct rte_eth_hash_filter_info info;
unsigned int ftype, idx, offset;
// my port configuration
struct rte_eth_conf port_conf = {
.rxmode =
{
.mq_mode = ETH_MQ_RX_RSS,
.enable_scatter = 1
}
};
const uint16_t rx_rings = 8, tx_rings = 1;
const uint16_t rx_ring_size = 2048;
const uint16_t tx_ring_size = 512;
port_conf.rx_adv_conf.rss_conf.rss_key = hashkey;
port_conf.rx_adv_conf.rss_conf.rss_key_len = 40;
port_conf.rx_adv_conf.rss_conf.rss_hf = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4
| ETH_RSS_NONFRAG_IPV4_TCP | ETH_RSS_NONFRAG_IPV4_UDP |
ETH_RSS_NONFRAG_IPV4_SCTP | ETH_RSS_NONFRAG_IPV4_OTHER | ETH_RSS_IPV6 |
ETH_RSS_FRAG_IPV6 | ETH_RSS_NONFRAG_IPV6_TCP | ETH_RSS_NONFRAG_IPV6_UDP
| ETH_RSS_NONFRAG_IPV6_SCTP | ETH_RSS_NONFRAG_IPV6_OTHER;
if ((retval = rte_eth_dev_configure(port_num, rx_rings, tx_rings,
&port_conf)) != 0)
return retval;
for (q = 0; q < rx_rings; q++)
{
retval = rte_eth_rx_queue_setup(port_num, q, rx_ring_size,
rte_eth_dev_socket_id(port_num), NULL, pkt_mbuf_pool[q]);
if (retval < 0) return retval;
}
printf("rx... %hu rings...", rx_rings);
for (q = 0; q < tx_rings; q++)
{
retval = rte_eth_tx_queue_setup(port_num, q, tx_ring_size,
rte_eth_dev_socket_id(port_num), NULL);
if (retval < 0) return retval;
}
printf("tx... %hu rings...", tx_rings);
// we need ports in promiscuous mode
rte_eth_promiscuous_enable(port_num);
if ((retval = rte_eth_dev_set_mtu(port_num, 9000)) != 0)
{
printf("MTU not set for port %hhu... %d\n", port_num, ritorno);
}
else
{
printf("MTU set for port %hhu\n", port_num);
}
// specific commands for X710
// select per ipv4 tcp - src ipv4
memset(&info, 0, sizeof (info));
info.info_type = RTE_ETH_HASH_FILTER_INPUT_SET_SELECT;
info.info.input_set_conf.flow_type = RTE_ETH_FLOW_NONFRAG_IPV4_TCP;
info.info.input_set_conf.field[0] = RTE_ETH_INPUT_SET_L3_SRC_IP4;
info.info.input_set_conf.inset_size = 1;
info.info.input_set_conf.op = RTE_ETH_INPUT_SET_SELECT;
ritorno = rte_eth_dev_filter_ctrl(port_num, RTE_ETH_FILTER_HASH,
RTE_ETH_FILTER_SET, &info);
if (ritorno < 0)
{
printf("Failure: set select ipv4 tcp (src ipv4) for port %hhu\n",
port_num);
}
// add per ipv4 tcp - dst ipv4
memset(&info, 0, sizeof (info));
info.info_type = RTE_ETH_HASH_FILTER_INPUT_SET_SELECT;
info.info.input_set_conf.flow_type = RTE_ETH_FLOW_NONFRAG_IPV4_TCP;
info.info.input_set_conf.field[0] = RTE_ETH_INPUT_SET_L3_DST_IP4;
info.info.input_set_conf.inset_size = 1;
info.info.input_set_conf.op = RTE_ETH_INPUT_SET_ADD;
ritorno = rte_eth_dev_filter_ctrl(port_num, RTE_ETH_FILTER_HASH,
RTE_ETH_FILTER_SET, &info);
if (ritorno < 0)
{
printf("Failure: set add ipv4 tcp (dst ipv4) for port %hhu\n",
port_num);
}
// hash global config ipv4 tcp
memset(&info, 0, sizeof (info));
info.info_type = RTE_ETH_HASH_FILTER_GLOBAL_CONFIG;
info.info.global_conf.hash_func = RTE_ETH_HASH_FUNCTION_DEFAULT;
ftype = RTE_ETH_FLOW_NONFRAG_IPV4_TCP;
idx = ftype / UINT64_BIT;
offset = ftype % UINT64_BIT;
info.info.global_conf.valid_bit_mask[idx] |= (1ULL << offset);
info.info.global_conf.sym_hash_enable_mask[idx] |= (1ULL << offset);
ritorno = rte_eth_dev_filter_ctrl(port_num, RTE_ETH_FILTER_HASH,
RTE_ETH_FILTER_SET, &info);
if (ritorno < 0)
{
printf("Cannot set global hash configurations for port %hhu proto
ipv4 tcp\n", port_num);
}
retval = rte_eth_dev_start(port_num);
if (retval < 0) return retval;
Even turning on rx debug for i40e doesn't help.
can anyone help me?
Thank you
Matteo
next reply other threads:[~2018-10-02 17:19 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-10-02 17:19 Matteo Lanzuisi [this message]
2018-10-04 9:35 ` Matteo Lanzuisi
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=384bb9db-1023-65f7-6d18-f1311db9228a@resi.it \
--to=m.lanzuisi@resi.it \
--cc=dev@dpdk.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).