>
> Hi,
> In some vendores (like Mellanox, that reflects the linux behavior in this
> case) the Rx and Tx configurations must be the same, therefore it is
> not enough to configure the max_rx_pkt_len to JUMBO_FRAME_MAX_SIZE, we also
> need to configure the MTU size in order to receive
> large packets.
Interesting, didn't know that.
Does it mean that l3fwd and bunch of other sample apps that use max_rx_pkt_len
to setup MTU,
don't work properly with mlx devices (not able to run with jumbo-frame enabled)?
> In order to avoid from adding another configuration to the command line, we
> can configure the MTU to be equal to max_rx_pkt_len, and it
> won't change the functionality of the test.
I don't have any objections with new option for MTU size (-m) if you need it.
It was just unclear to me why it was necessary for you.
But yes, setting MTU to the correlated with max_rx_ptk_len value sounds like a
good idea.
>
> In addition there are PMDs that need to enable the scatter-gather so it will
> be functional for RX frames bigger then mbuf size. We can add
> the configuration and avoid changing the mbuf size.
You mean check if MTU is bigger than mbuf size, and if so setup
(DEV_RX_OFFLOAD_JUMBO_FRAME | DEV_RX_OFFLOAD_SCATTER)
?
Konstantin
>
> What do you think about this solution?
>
> Thanks,
> Noa.
>
> -----Original Message-----
> From: Ananyev, Konstantin [mailto:konstantin.anan...@intel.com]
> Sent: Thursday, December 20, 2018 2:18 AM
> To: Noa Ezra <n...@mellanox.com>
> Cc: dev@dpdk.org
> Subject: RE: [PATCH] examples/ip_fragmentation: support bigger packets
>
> Hi,
>
> >
> > Adding MTU and mbuf size configuration to the application's command
> > line, in order to be able to receive all packet sizes by the NIC and
> > DPDK application.
> > The maximum transmission unit (MTU) is the largest size packet in
> > bytes that can be sent on the network, therefore before adding MTU
> > parameter, the NIC could not receive packets larger than 1500 bytes,
> > which is the default MTU size.
>
> I wonder why is that?
> Currently ip_fragmentation sets max_rx_pkt_len to up to 9.5KB:
>
> static struct rte_eth_conf port_conf = {
> .rxmode = {
> .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE, ...
>
> local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
> dev_info.max_rx_pktlen,
> local_port_conf.rxmode.max_rx_pkt_len);
>
> That in theory should be enough to enable jumbo frame RX.
> Did you find it is not working as expected, if yes on which NIC?
>
> > The mbuf is the memory buffer that contains the packet. Before adding
> > mbuf parameter, the DPDK application could not receive packets larger
> > than 2KB, which is the default mbuf size.
>
> Again why is that?
> All packets that do support scatter-gather RX should be able to RX frames
> bigger then mbuf size (if properly configured of course).
> Are you trying to make it work on NICs with no multi-segment support for
> RX/TX?
> But then how you plan to do TX (usually it is symmetric for RX/TX)?
>
> >
> > Signed-off-by: Noa Ezra <n...@mellanox.com>
> > ---
> > doc/guides/sample_app_ug/ip_frag.rst | 18 ++++++++-
> > examples/ip_fragmentation/main.c | 77
> > +++++++++++++++++++++++++++++++++---
> > 2 files changed, 88 insertions(+), 7 deletions(-)
> >
> > diff --git a/doc/guides/sample_app_ug/ip_frag.rst
> > b/doc/guides/sample_app_ug/ip_frag.rst
> > index 7914a97..13933c7 100644
> > --- a/doc/guides/sample_app_ug/ip_frag.rst
> > +++ b/doc/guides/sample_app_ug/ip_frag.rst
> > @@ -53,7 +53,7 @@ Application usage:
> >
> > .. code-block:: console
> >
> > - ./build/ip_fragmentation [EAL options] -- -p PORTMASK [-q NQ]
> > + ./build/ip_fragmentation [EAL options] -- -p PORTMASK [-q NQ] [-b
> > + MBUFSIZE] [-m MTUSIZE]
> >
> > where:
> >
> > @@ -61,6 +61,15 @@ where:
> >
> > * -q NQ is the number of queue (=ports) per lcore (the default is 1)
> >
> > +* -b MBUFSIZE is the mbuf size in bytes (the default is 2048)
> > +
> > +* -m MTUSIZE is the mtu size in bytes (the default is 1500)
> > +
> > +The MTU is the maximum size of a single data unit that can be
> > +transmitted over the network, therefore it must be greater than the
> > requested max packet size, otherwise the NIC won't be able to get
> the packet.
> > +The mbuf is a buffer that is used by the DPDK application to store
> > +message buffers.If not using scatter then the mbuf size must be
> > +greater than the requested max packet size, otherwise the DPDK will not be
> > able to receive the packet.
> > To run the example in linuxapp environment with 2 lcores (2,4) over 2
> > ports(0,2) with 1 RX queue per lcore:
> >
> > .. code-block:: console
> > @@ -96,6 +105,13 @@ To run the example in linuxapp environment with 1
> > lcore (4) over 2 ports(0,2) wi
> >
> > ./build/ip_fragmentation -l 4 -n 3 -- -p 5 -q 2
> >
> > +To run the example with defined MTU size 4000 bytes and mbuf size 9000
> > byes:
> > +
> > +.. code-block:: console
> > +
> > + ./build/ip_fragmentation -l 4 -n 3 -- -p 5 -m 4000 -b 9000
> > +
> > +
> > To test the application, flows should be set up in the flow generator
> > that match the values in the l3fwd_ipv4_route_array and/or
> > l3fwd_ipv6_route_array table.
> >
> > diff --git a/examples/ip_fragmentation/main.c
> > b/examples/ip_fragmentation/main.c
> > index 17a877d..0cf23b4 100644
> > --- a/examples/ip_fragmentation/main.c
> > +++ b/examples/ip_fragmentation/main.c
> > @@ -111,6 +111,10 @@
> >
> > static int rx_queue_per_lcore = 1;
> >
> > +static int mbuf_size = RTE_MBUF_DEFAULT_BUF_SIZE;
> > +
> > +static int mtu_size = ETHER_MTU;
> > +
> > #define MBUF_TABLE_SIZE (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
> >
> > struct mbuf_table {
> > @@ -425,7 +429,6 @@ struct rte_lpm6_config lpm6_config = {
> > * Read packet from RX queues
> > */
> > for (i = 0; i < qconf->n_rx_queue; i++) {
> > -
> > portid = qconf->rx_queue_list[i].portid;
> > nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
> > MAX_PKT_BURST);
> > @@ -455,9 +458,11 @@ struct rte_lpm6_config lpm6_config = { static
> > void print_usage(const char *prgname) {
> > - printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
> > + printf("%s [EAL options] -- -p PORTMASK [-q NQ] [-b MBUFSIZE] [-m
> > MTUSIZE]\n"
> > " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
> > - " -q NQ: number of queue (=ports) per lcore (default is 1)\n",
> > + " -q NQ: number of queue (=ports) per lcore (default is 1)\n"
> > + " -b MBUFSIZE: set the data size of mbuf\n"
> > + " -m MTUSIZE: set the MTU size\n",
> > prgname);
> > }
> >
> > @@ -496,6 +501,38 @@ struct rte_lpm6_config lpm6_config = {
> > return n;
> > }
> >
> > +static int
> > +parse_mbufsize(const char *q_arg)
> > +{
> > + char *end = NULL;
> > + unsigned long mbuf;
> > +
> > + /* parse hexadecimal string */
>
> You expect decimal string below.
>
> > + mbuf = strtoul(q_arg, &end, 10);
> > + if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
> > + return -1;
>
> You probably need to set and test errno too.
>
> > + if (mbuf == 0)
> > + return -1;
> > +
> > + return mbuf;
> > +}
>
>
> These 2 parse functions looks identical.
> Why you need two of them?
>
> > +
> > +static int
> > +parse_mtusize(const char *q_arg)
> > +{
> > + char *end = NULL;
> > + unsigned long mtu;
> > +
> > + /* parse hexadecimal string */
> > + mtu = strtoul(q_arg, &end, 10);
> > + if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
> > + return -1;
> > + if (mtu == 0)
> > + return -1;
> > +
> > + return mtu;
> > +}
> > +
> > /* Parse the argument given in the command line of the application */
> > static int parse_args(int argc, char **argv) @@ -510,7 +547,7 @@
> > struct rte_lpm6_config lpm6_config = {
> >
> > argvopt = argv;
> >
> > - while ((opt = getopt_long(argc, argvopt, "p:q:",
> > + while ((opt = getopt_long(argc, argvopt, "p:q:b:m:",
> > lgopts, &option_index)) != EOF) {
> >
> > switch (opt) {
> > @@ -534,6 +571,26 @@ struct rte_lpm6_config lpm6_config = {
> > }
> > break;
> >
> > + /* mbufsize */
> > + case 'b':
> > + mbuf_size = parse_mbufsize(optarg);
> > + if (mbuf_size < 0) {
> > + printf("invalid mbuf size\n");
> > + print_usage(prgname);
> > + return -1;
> > + }
> > + break;
> > +
> > + /* mtusize */
> > + case 'm':
> > + mtu_size = parse_mtusize(optarg);
> > + if (mtu_size < 0) {
> > + printf("invalid mtu size\n");
> > + print_usage(prgname);
> > + return -1;
> > + }
> > + break;
> > +
> > /* long options */
> > case 0:
> > print_usage(prgname);
> > @@ -777,9 +834,8 @@ struct rte_lpm6_config lpm6_config = {
> > RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on
> > socket %i\n",
> > socket);
> > snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
> > -
> > mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
> > - 0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
> > + 0, mbuf_size, socket);
> > if (mp == NULL) {
> > RTE_LOG(ERR, IP_FRAG, "Cannot create direct
> > mempool\n");
> > return -1;
> > @@ -892,6 +948,15 @@ struct rte_lpm6_config lpm6_config = {
> > dev_info.max_rx_pktlen,
> > local_port_conf.rxmode.max_rx_pkt_len);
> >
> > + /* set the mtu to the maximum received packet size */
> > + ret = rte_eth_dev_set_mtu(portid, mtu_size);
> > + if (ret < 0) {
> > + printf("\n");
> > + rte_exit(EXIT_FAILURE, "Set MTU failed: "
> > + "err=%d, port=%d\n",
> > + ret, portid);
> > + }
> > +
> > /* get the lcore_id for this port */
> > while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
> > qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
> > --
> > 1.8.3.1
