DPDK patches and discussions
 help / color / mirror / Atom feed
From: "Xie, Huawei" <huawei.xie@intel.com>
To: Stephen Hemminger <stephen@networkplumber.org>
Cc: "dev@dpdk.org" <dev@dpdk.org>
Subject: Re: [dpdk-dev] [PATCH 1/5] virtio: clean up space checks on xmit
Date: Mon, 19 Oct 2015 16:27:08 +0000	[thread overview]
Message-ID: <C37D651A908B024F974696C65296B57B4B12CC27@SHSMSX101.ccr.corp.intel.com> (raw)
In-Reply-To: <20151019084835.7bbeb2c3@xeon-e3>

On 10/19/2015 11:48 PM, Stephen Hemminger wrote:
> On Mon, 19 Oct 2015 08:02:07 +0000
> "Xie, Huawei" <huawei.xie@intel.com> wrote:
>
>> On 10/19/2015 1:16 PM, Stephen Hemminger wrote:
>>> The space check for transmit ring only needs a single conditional.
>>> I.e only need to recheck for space if there was no space in first check.
>> I see you reorganized the code. It is more clear and readable with your
>> change, but no check is removed.
>> We could remove the check after virtio_xmit_cleanup. In current
>> implementation, we assume the virtio_xmit_cleanup and vq_ring_free_chain
>> always succeed.
> The only case that matters is if virtio_xmit_cleanup can not
> free up any slots.
Is there the case in current implementation virtio_xmit_cleanup could
free up any descriptors, i.e, dxp->ndesc is zero?
    vq->free_cnt += dxp->ndesc
>
>>> This can help performance and simplifies loop.
>>>
>>> Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
>>> ---
>>>  drivers/net/virtio/virtio_rxtx.c | 66 ++++++++++++++++------------------------
>>>  1 file changed, 27 insertions(+), 39 deletions(-)
>>>
>>> diff --git a/drivers/net/virtio/virtio_rxtx.c b/drivers/net/virtio/virtio_rxtx.c
>>> index c5b53bb..5b50ed0 100644
>>> --- a/drivers/net/virtio/virtio_rxtx.c
>>> +++ b/drivers/net/virtio/virtio_rxtx.c
>>> @@ -745,7 +745,6 @@ uint16_t
>>>  virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
>>>  {
>>>  	struct virtqueue *txvq = tx_queue;
>>> -	struct rte_mbuf *txm;
>>>  	uint16_t nb_used, nb_tx;
>>>  	int error;
>>>  
>>> @@ -759,57 +758,46 @@ virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
>>>  	if (likely(nb_used > txvq->vq_nentries - txvq->vq_free_thresh))
>>>  		virtio_xmit_cleanup(txvq, nb_used);
>>>  
>>> -	nb_tx = 0;
>>> +	for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
>>> +		struct rte_mbuf *txm = tx_pkts[nb_tx];
>>> +		int need = txm->nb_segs - txvq->vq_free_cnt + 1;
>>>  
>>> -	while (nb_tx < nb_pkts) {
>>> -		/* Need one more descriptor for virtio header. */
>>> -		int need = tx_pkts[nb_tx]->nb_segs - txvq->vq_free_cnt + 1;
>>> -
>>> -		/*Positive value indicates it need free vring descriptors */
>>> +		/* Positive value indicates it need free vring descriptors */
>> As you fix the comment, if it is correct, could you do s/need/needs/ as
>> well, :)?
>>>  		if (unlikely(need > 0)) {
>>>  			nb_used = VIRTQUEUE_NUSED(txvq);
>>>  			virtio_rmb();
>>>  			need = RTE_MIN(need, (int)nb_used);
>>>  
>>>  			virtio_xmit_cleanup(txvq, need);
>>> -			need = (int)tx_pkts[nb_tx]->nb_segs -
>>> -				txvq->vq_free_cnt + 1;
>>> -		}
>>> -
>>> -		/*
>>> -		 * Zero or negative value indicates it has enough free
>>> -		 * descriptors to use for transmitting.
>>> -		 */
>>> -		if (likely(need <= 0)) {
>>> -			txm = tx_pkts[nb_tx];
>>> -
>>> -			/* Do VLAN tag insertion */
>>> -			if (unlikely(txm->ol_flags & PKT_TX_VLAN_PKT)) {
>>> -				error = rte_vlan_insert(&txm);
>>> -				if (unlikely(error)) {
>>> -					rte_pktmbuf_free(txm);
>>> -					++nb_tx;
>>> -					continue;
>>> -				}
>> Still need is rechecked here. It could be removed if virtio_xmit_cleanup
>> ensures need would be <= 0 after the call.
>>> +			need = txm->nb_segs - txvq->vq_free_cnt + 1;
>>> +			if (unlikely(need > 0)) {
>>> +				PMD_TX_LOG(ERR,
>>> +					   "No free tx descriptors to transmit");
>>> +				break;
>>>  			}
>>> +		}
>>>  
>>> -			/* Enqueue Packet buffers */
>>> -			error = virtqueue_enqueue_xmit(txvq, txm);
>>> +		/* Do VLAN tag insertion */
>>> +		if (unlikely(txm->ol_flags & PKT_TX_VLAN_PKT)) {
>>> +			error = rte_vlan_insert(&txm);
>>>  			if (unlikely(error)) {
>>> -				if (error == ENOSPC)
>>> -					PMD_TX_LOG(ERR, "virtqueue_enqueue Free count = 0");
>>> -				else if (error == EMSGSIZE)
>>> -					PMD_TX_LOG(ERR, "virtqueue_enqueue Free count < 1");
>>> -				else
>>> -					PMD_TX_LOG(ERR, "virtqueue_enqueue error: %d", error);
>>> -				break;
>>> +				rte_pktmbuf_free(txm);
>>> +				continue;
>>>  			}
>>> -			nb_tx++;
>>> -			txvq->bytes += txm->pkt_len;
>>> -		} else {
>>> -			PMD_TX_LOG(ERR, "No free tx descriptors to transmit");
>>> +		}
>>> +
>>> +		/* Enqueue Packet buffers */
>>> +		error = virtqueue_enqueue_xmit(txvq, txm);
>>> +		if (unlikely(error)) {
>>> +			if (error == ENOSPC)
>>> +				PMD_TX_LOG(ERR, "virtqueue_enqueue Free count = 0");
>>> +			else if (error == EMSGSIZE)
>>> +				PMD_TX_LOG(ERR, "virtqueue_enqueue Free count < 1");
>>> +			else
>>> +				PMD_TX_LOG(ERR, "virtqueue_enqueue error: %d", error);
>>>  			break;
>>>  		}
>>> +		txvq->bytes += txm->pkt_len;
>>>  	}
>>>  
>>>  	txvq->packets += nb_tx;
>


  reply	other threads:[~2015-10-19 16:27 UTC|newest]

Thread overview: 35+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-10-19  5:16 [dpdk-dev] [PATCH v2 0/5] virtio: Tx performance improvements Stephen Hemminger
2015-10-19  5:16 ` [dpdk-dev] [PATCH 1/5] virtio: clean up space checks on xmit Stephen Hemminger
2015-10-19  8:02   ` Xie, Huawei
2015-10-19 15:48     ` Stephen Hemminger
2015-10-19 16:27       ` Xie, Huawei [this message]
2015-10-19  5:16 ` [dpdk-dev] [PATCH 2/5] virtio: don't use unlikely for normal tx stuff Stephen Hemminger
2015-10-19  5:16 ` [dpdk-dev] [PATCH 3/5] virtio: use indirect ring elements Stephen Hemminger
2015-10-19 13:19   ` Xie, Huawei
2015-10-19 15:47     ` Stephen Hemminger
2015-10-19 16:18       ` Xie, Huawei
2015-10-30 18:01   ` Thomas Monjalon
2015-10-19  5:16 ` [dpdk-dev] [PATCH 4/5] virtio: use any layout on transmit Stephen Hemminger
2015-10-19 16:28   ` Xie, Huawei
2015-10-19 16:43     ` Stephen Hemminger
2015-10-19 16:56       ` Xie, Huawei
2015-10-26 23:47         ` Stephen Hemminger
2015-10-19 17:19     ` Stephen Hemminger
2015-10-19  5:16 ` [dpdk-dev] [PATCH 5/5] virtio: optimize transmit enqueue Stephen Hemminger
2015-10-20  1:48   ` Xie, Huawei
2015-10-21 13:18 ` [dpdk-dev] [PATCH v2 0/5] virtio: Tx performance improvements Thomas Monjalon
2015-10-22 10:38   ` Xie, Huawei
2015-10-22 12:13     ` Xie, Huawei
2015-10-22 16:04     ` Stephen Hemminger
2015-10-23  9:00       ` Xie, Huawei
2015-10-26 23:52         ` Stephen Hemminger
2015-10-27  1:56           ` Xie, Huawei
2015-10-27  2:23             ` Stephen Hemminger
2015-10-27  2:38               ` Xie, Huawei
2015-10-26 14:05 ` Xie, Huawei
2016-01-05  8:10   ` Xie, Huawei
2016-01-06 12:03     ` Thomas Monjalon
2016-01-14 13:49       ` Xie, Huawei
2016-03-04  6:18         ` Xie, Huawei
2016-03-04 18:17           ` Stephen Hemminger
2016-03-08  1:38             ` Xie, Huawei

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=C37D651A908B024F974696C65296B57B4B12CC27@SHSMSX101.ccr.corp.intel.com \
    --to=huawei.xie@intel.com \
    --cc=dev@dpdk.org \
    --cc=stephen@networkplumber.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).