DPDK usage discussions
 help / color / mirror / Atom feed
From: Stephen Hemminger <stephen@networkplumber.org>
To: Lokesh Chakka <lvenkatakumarchakka@gmail.com>
Cc: users <users@dpdk.org>
Subject: Re: reuse the packets after tx burst
Date: Wed, 14 Aug 2024 07:59:02 -0700	[thread overview]
Message-ID: <20240814075902.187cda3f@hermes.local> (raw)
In-Reply-To: <CACh--sjR8781bSyZKxwQ4S8ZbnV0+7rpzzSwAoJxtbQFf80Ckw@mail.gmail.com>

On Wed, 14 Aug 2024 17:57:40 +0530
Lokesh Chakka <lvenkatakumarchakka@gmail.com> wrote:

> hi stephen,
> 
> Thanks for your response. However it seems the solution is not working. The
> following is the modified snippet....
> 
> while( condition )
> {
>      for( pkt_count=0; pkt_count<num_of_pkts_per_queue; pkt_count++ )
>      {
>           rte_mbuf_refcnt_set( mbuf[pkt_count], 2 );// setting to two -
> first and second iteration
>      }
>      for( pkt_count=0; pkt_count<5; pkt_count++ )
>      {
>           fprintf( stderr, "%s %d %u\n", __func__, __LINE__,
> rte_mbuf_refcnt_read( mbuf[pkt_count] ) );//able to print two- first and
> second iteration
>      }
>      if( rte_eth_tx_burst( port_id, 0, mbuf, num_of_pkts_per_queue ) !=
> num_of_pkts_per_queue )
>      {
>           fprintf( stderr, "%s %d %d %s\n", __func__, __LINE__, rte_errno,
> rte_strerror(rte_errno) );//failing second time
>           rte_exit( EXIT_FAILURE, "%s %d rte_eth_tx_burst port id: %u\n",
> __func__, __LINE__, port_id );
>      }
>      fprintf( stderr, "%s %d port: %u sent %u packets\n", __func__,
> __LINE__, port_id, num_of_pkts_per_queue );//able to send once - first
> iteration only
>      for( pkt_count=0; pkt_count<5; pkt_count++ )
>      {
>           fprintf( stderr, "%s %d %u\n", __func__, __LINE__,
> rte_mbuf_refcnt_read( mbuf[pkt_count] ) );//refcnt is printing one
>      }
> }
> 
> There seems to be some more gap in my understanding. Could you please help
> understand the issue?
> 
> Thanks & Regards
> --
> Lokesh Chakka.
> 
> 
> On Mon, Aug 12, 2024 at 8:22 PM Stephen Hemminger <
> stephen@networkplumber.org> wrote:  
> 
> > On Mon, 12 Aug 2024 15:55:50 +0530
> > Lokesh Chakka <lvenkatakumarchakka@gmail.com> wrote:
> >  
> > > hello,
> > >
> > > Here is a small piece of code :
> > >
> > > while( condition )
> > > {
> > >
> > >       if( rte_eth_tx_burst( port_id, 0, mbuf, num_of_pkts_per_queue ) !=
> > > num_of_pkts_per_queue )
> > >       {
> > >               fprintf( stderr, "%d %s\n", rte_errno,  
> > rte_strerror(rte_errno) );  
> > >               rte_exit( EXIT_FAILURE, "%s %d rte_eth_tx_burst port id:  
> > %u\n",  
> > > __func__, __LINE__, port_id );//second iteration failing.
> > >       }
> > >       fprintf( stderr, "%s %d port: %u packet: %c sent %u packets\n",
> > > __func__, __LINE__, port_id, argv[3][0], num_of_pkts_per_queue
> > > );//printing once
> > >       for( pkt_count=0; pkt_count<num_of_pkts_per_queue; pkt_count++ )
> > >       {//want to send same data again...!!!
> > >               mbuf[pkt_count]->pkt_len = mbuf[pkt_count]->data_len =  
> > dev_info.max_mtu;  
> > >
> > >       }
> > >
> > > }
> > >
> > > Can someone help me understand how to reuse the packets again to send the
> > > same data ?
> > >  
> >
> > When packet is passed to tx_burst, the ownership of that mbuf is passed
> > to the driver. The driver will free it after it is sent.
> >
> > One option would be to increase the reference count on the packet before
> > sending.
> > Using rte_mbuf_refcnt_update() function to add one to refcount.
> > Then the driver will decrement refcount and the refcount will still be one
> > (not freed).
> >
> > Assume this is some kind of packet generator.
> >  


Don't use refcnt_set(), you want to add an additional refcount not force it to two.
The device driver may hold onto the mbuf for a while after it was passed to tx_burst
so can't assume a value of two.

If the number actually queued is less that the number requested, that means the device
transmit queue is now full. Need to handle that case.

Something like this might get you started:


volatile running = true;

void flood(uint16_t port_id, struct rte_mbufs *mbufs[], uint16_t num_pkts)
{

	while (running) {
		/* Add additional reference to retain the packets */
		for (uint16_t i = 0; i < num_pkts; i++)
			rte_mbuf_refcnt_update(mbufs[i], 1);

		uint16_t sent = rte_eth_tx_burst(port_id, queue_id, mbufs, num_pkts);

		if (sent < num_pkts) {
			/* device transmit queue was full, drop our reference */
			for (uint16_t i = sent; i < num_pkts; i++)
				rte_pktmbuf_free(mbufs[i]);
		}
	}

	/* Don't leak original version */
	rte_pktmbuf_free_bulk(mbufs, num_pkts);

}
			


  reply	other threads:[~2024-08-14 14:59 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-08-12 10:25 Lokesh Chakka
2024-08-12 14:52 ` Stephen Hemminger
2024-08-14 12:27   ` Lokesh Chakka
2024-08-14 14:59     ` Stephen Hemminger [this message]
2024-08-15 16:18       ` Lokesh Chakka
2024-08-16  7:12         ` Lokesh Chakka

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=20240814075902.187cda3f@hermes.local \
    --to=stephen@networkplumber.org \
    --cc=lvenkatakumarchakka@gmail.com \
    --cc=users@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).