DPDK patches and discussions
 help / color / mirror / Atom feed
From: "Sanford, Robert" <rsanford@akamai.com>
To: Declan Doherty <declan.doherty@intel.com>, "dev@dpdk.org" <dev@dpdk.org>
Subject: Re: [dpdk-dev] [PATCH 4/6] bond: free mbufs if transmission fails in bonding tx_burst functions
Date: Wed, 20 Aug 2014 15:54:02 -0500	[thread overview]
Message-ID: <D01A81B6.27E1%rsanford@akamai.com> (raw)

Hi Declan,

I have a problem with the TX-burst logic of this patch. I believe that for
packets that we *don't* enqueue to the device, we should *NOT* free them.
The API expects that the caller will free them or try again to send them.

Here is one way to accomplish selective freeing: Move mbuf pointers of
packets successfully enqueued, to the beginning of the caller's mbuf
vector; move mbuf pointers of packets not enqueued, to the end of the
caller's mbuf list. Although possibly re-ordered, the caller will be able
to free/resend all (and only) mbufs that we failed to enqueue.

Here is a code snippet to do this:

  uint16_t failed_index = nb_pkts;

  /* Send packet burst on each slave device */
  for (i = 0; i < num_of_slaves; i++) {
    if (slave_nb_pkts[i] > 0) {
      num_tx_slave = rte_eth_tx_burst(slaves[i],
          bd_tx_q->queue_id, slave_bufs[i], slave_nb_pkts[i]);

    
      /* Move sent packets to the beginning of the caller's array. */
      if (likely(num_tx_slave)) {
        memcpy(&bufs[num_tx_total], slave_bufs[i],
            num_tx_slave * sizeof(void *));
        num_tx_total += num_tx_slave;
      }

    
      /* Move failed packets to the end of the caller's array,
       * so that the caller can free or resend the correct ones.
       */
      if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
        uint16_t num_failed = slave_nb_pkts[i] - num_tx_slave;
        failed_index -= num_failed;
        memcpy(&bufs[failed_index], &slave_bufs[i][num_tx_slave],
            num_failed * sizeof(void *));
      }
    }
  }       
 


--
Regards,
Robert


>Fixing a number of corner cases that if transmission failed on slave
>devices then this
>could lead to leaked mbufs
>
>
>Signed-off-by: Declan Doherty <declan.doherty@intel.com>
>---
> app/test/test_link_bonding.c           |    4 +-
> lib/librte_pmd_bond/rte_eth_bond_pmd.c |   46
>+++++++++++++++++++++++++-------
> 2 files changed, 38 insertions(+), 12 deletions(-)
>
>diff --git a/app/test/test_link_bonding.c b/app/test/test_link_bonding.c
>index 02823b6..3c265ee 100644
>--- a/app/test/test_link_bonding.c
>+++ b/app/test/test_link_bonding.c
>@@ -3415,7 +3415,7 @@ test_broadcast_tx_burst(void)
> 	/* Send burst on bonded port */
> 	nb_tx = rte_eth_tx_burst(test_params->bonded_port_id, 0, pkts_burst,
> 			burst_size);
>-	if (nb_tx != burst_size * test_params->bonded_slave_count) {
>+	if (nb_tx != burst_size) {
> 		printf("Bonded Port (%d) rx burst failed, packets transmitted value
>(%u) not as expected (%d)\n",
> 				test_params->bonded_port_id,
> 				nb_tx, burst_size);
>@@ -3770,7 +3770,7 @@
>test_broadcast_verify_slave_link_status_change_behaviour(void)
> 	}
> 
> 	if (rte_eth_tx_burst(test_params->bonded_port_id, 0, &pkt_burst[0][0],
>-			burst_size) != (burst_size * slave_count)) {
>+			burst_size) != burst_size) {
> 		printf("rte_eth_tx_burst failed\n");
> 		return -1;
> 	}
>diff --git a/lib/librte_pmd_bond/rte_eth_bond_pmd.c
>b/lib/librte_pmd_bond/rte_eth_bond_pmd.c
>index 70123fc..ae9726e 100644
>--- a/lib/librte_pmd_bond/rte_eth_bond_pmd.c
>+++ b/lib/librte_pmd_bond/rte_eth_bond_pmd.c
>@@ -101,7 +101,7 @@ bond_ethdev_tx_burst_round_robin(void *queue, struct
>rte_mbuf **bufs,
> 	uint8_t num_of_slaves;
> 	uint8_t slaves[RTE_MAX_ETHPORTS];
> 
>-	uint16_t num_tx_total = 0;
>+	uint16_t num_tx_total = 0, num_tx_slave;
> 
> 	static int slave_idx = 0;
> 	int i, cs_idx = 0;
>@@ -130,9 +130,17 @@ bond_ethdev_tx_burst_round_robin(void *queue, struct
>rte_mbuf **bufs,
> 
> 	/* Send packet burst on each slave device */
> 	for (i = 0; i < num_of_slaves; i++)
>-		if (slave_nb_pkts[i] > 0)
>-			num_tx_total += rte_eth_tx_burst(slaves[i],
>+		if (slave_nb_pkts[i] > 0) {
>+			num_tx_slave = rte_eth_tx_burst(slaves[i],
> 					bd_tx_q->queue_id, slave_bufs[i], slave_nb_pkts[i]);
>+			num_tx_total += num_tx_slave;
>+
>+			/* if tx burst fails, free unsent mbufs */
>+			while (unlikely(num_tx_slave < slave_nb_pkts[i])) {
>+				rte_pktmbuf_free(slave_bufs[i][num_tx_slave]);
>+				num_tx_slave++;
>+			}
>+		}
> 
> 	return num_tx_total;
> }
>@@ -283,7 +291,7 @@ bond_ethdev_tx_burst_balance(void *queue, struct
>rte_mbuf **bufs,
> 	uint8_t num_of_slaves;
> 	uint8_t slaves[RTE_MAX_ETHPORTS];
> 
>-	uint16_t num_tx_total = 0;
>+	uint16_t num_tx_total = 0, num_tx_slave = 0;
> 
> 	int i, op_slave_id;
> 
>@@ -315,11 +323,19 @@ bond_ethdev_tx_burst_balance(void *queue, struct
>rte_mbuf **bufs,
> 	/* Send packet burst on each slave device */
> 	for (i = 0; i < num_of_slaves; i++) {
> 		if (slave_nb_pkts[i] > 0) {
>-			num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
>+			num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
> 					slave_bufs[i], slave_nb_pkts[i]);
>+			num_tx_total += num_tx_slave;
>+
>+			/* if tx burst fails, free unsent mbufs */
>+			while (unlikely(num_tx_slave < slave_nb_pkts[i])) {
>+				rte_pktmbuf_free(slave_bufs[i][num_tx_slave]);
>+				num_tx_slave++;
>+			}
> 		}
> 	}
> 
>+
> 	return num_tx_total;
> }
> 
>@@ -333,7 +349,7 @@ bond_ethdev_tx_burst_broadcast(void *queue, struct
>rte_mbuf **bufs,
> 	uint8_t num_of_slaves;
> 	uint8_t slaves[RTE_MAX_ETHPORTS];
> 
>-	uint16_t num_tx_total = 0;
>+	uint16_t num_tx_slave = 0, max_tx_pkts = 0;
> 
> 	int i;
> 
>@@ -354,11 +370,21 @@ bond_ethdev_tx_burst_broadcast(void *queue, struct
>rte_mbuf **bufs,
> 		rte_mbuf_refcnt_update(bufs[i], num_of_slaves - 1);
> 
> 	/* Transmit burst on each active slave */
>-	for (i = 0; i < num_of_slaves; i++)
>-		num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
>-				bufs, nb_pkts);
>+	for (i = 0; i < num_of_slaves; i++) {
>+		num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
>+					bufs, nb_pkts);
> 
>-	return num_tx_total;
>+		if (max_tx_pkts < num_tx_slave)
>+			max_tx_pkts = num_tx_slave;
>+
>+		/* if tx burst fails, free unsent mbufs */
>+		while (unlikely(num_tx_slave < nb_pkts)) {
>+			rte_pktmbuf_free(bufs[num_tx_slave]);
>+			num_tx_slave++;
>+		}
>+	}
>+
>+	return max_tx_pkts;
> }
> 
> void
>-- 
>1.7.0.7
>

             reply	other threads:[~2014-08-20 20:50 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-08-20 20:54 Sanford, Robert [this message]
2014-08-29 15:36 ` Thomas Monjalon
  -- strict thread matches above, loose matches on Subject: below --
2014-08-19 13:51 [dpdk-dev] [PATCH 0/6] link bonding Declan Doherty
2014-08-19 13:51 ` [dpdk-dev] [PATCH 4/6] bond: free mbufs if transmission fails in bonding tx_burst functions Declan Doherty

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=D01A81B6.27E1%rsanford@akamai.com \
    --to=rsanford@akamai.com \
    --cc=declan.doherty@intel.com \
    --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).