DPDK patches and discussions
 help / color / mirror / Atom feed
From: "Xie, Huawei" <huawei.xie@intel.com>
To: "Ouyang, Changchun" <changchun.ouyang@intel.com>,
	"dev@dpdk.org" <dev@dpdk.org>
Subject: Re: [dpdk-dev] [PATCH v2] virtio: Support mergeable buffer in virtio PMD
Date: Fri, 25 Jul 2014 08:05:59 +0000	[thread overview]
Message-ID: <C37D651A908B024F974696C65296B57B0F24595E@SHSMSX101.ccr.corp.intel.com> (raw)
In-Reply-To: <1406268195-24010-1-git-send-email-changchun.ouyang@intel.com>



> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Ouyang Changchun
> Sent: Friday, July 25, 2014 2:03 PM
> To: dev@dpdk.org
> Subject: [dpdk-dev] [PATCH v2] virtio: Support mergeable buffer in virtio PMD
> 
> v2 change:
> - Resolve conflicts wiht the tip code;
> - And resolve 2 issues:
>    -- fix mbuf leak when discard a uncompleted packet.
>    -- refine pkt.data to point to actual payload data start point.
> 
> v1 change:
> This patch supports mergeable buffer feature in DPDK based virtio PMD, which
> can
> receive jumbo frame with larger size, like 3K, 4K or even 9K.
> 
> Signed-off-by: Changchun Ouyang <changchun.ouyang@intel.com>
> ---
>  lib/librte_pmd_virtio/virtio_ethdev.c |  20 ++--
>  lib/librte_pmd_virtio/virtio_ethdev.h |   3 +
>  lib/librte_pmd_virtio/virtio_rxtx.c   | 206 +++++++++++++++++++++++++++++---
> --
>  3 files changed, 194 insertions(+), 35 deletions(-)
> 
> diff --git a/lib/librte_pmd_virtio/virtio_ethdev.c
> b/lib/librte_pmd_virtio/virtio_ethdev.c
> index b9f5529..535d798 100644
> --- a/lib/librte_pmd_virtio/virtio_ethdev.c
> +++ b/lib/librte_pmd_virtio/virtio_ethdev.c
> @@ -337,7 +337,7 @@ int virtio_dev_queue_setup(struct rte_eth_dev *dev,
>  		snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
>  			dev->data->port_id, queue_idx);
>  		vq->virtio_net_hdr_mz =
> rte_memzone_reserve_aligned(vq_name,
> -			vq_size * sizeof(struct virtio_net_hdr),
> +			vq_size * hw->vtnet_hdr_size,
>  			socket_id, 0, CACHE_LINE_SIZE);
>  		if (vq->virtio_net_hdr_mz == NULL) {
>  			rte_free(vq);
> @@ -346,7 +346,7 @@ int virtio_dev_queue_setup(struct rte_eth_dev *dev,
>  		vq->virtio_net_hdr_mem =
>  			vq->virtio_net_hdr_mz->phys_addr;
>  		memset(vq->virtio_net_hdr_mz->addr, 0,
> -			vq_size * sizeof(struct virtio_net_hdr));
> +			vq_size * hw->vtnet_hdr_size);
>  	} else if (queue_type == VTNET_CQ) {
>  		/* Allocate a page for control vq command, data and status */
>  		snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
> @@ -571,9 +571,6 @@ virtio_negotiate_features(struct virtio_hw *hw)
>  	mask |= VIRTIO_NET_F_GUEST_TSO4 | VIRTIO_NET_F_GUEST_TSO6 |
> VIRTIO_NET_F_GUEST_ECN;
>  	mask |= VTNET_LRO_FEATURES;
> 
> -	/* rx_mbuf should not be in multiple merged segments */
> -	mask |= VIRTIO_NET_F_MRG_RXBUF;
> -
>  	/* not negotiating INDIRECT descriptor table support */
>  	mask |= VIRTIO_RING_F_INDIRECT_DESC;
> 
> @@ -746,7 +743,6 @@ eth_virtio_dev_init(__rte_unused struct eth_driver
> *eth_drv,
>  	}
> 
>  	eth_dev->dev_ops = &virtio_eth_dev_ops;
> -	eth_dev->rx_pkt_burst = &virtio_recv_pkts;
>  	eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
> 
>  	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
> @@ -801,10 +797,13 @@ eth_virtio_dev_init(__rte_unused struct eth_driver
> *eth_drv,
>  	virtio_negotiate_features(hw);
> 
>  	/* Setting up rx_header size for the device */
> -	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
> +	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
> +		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
>  		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
> -	else
> +	} else {
> +		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
>  		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
> +	}
> 
>  	/* Allocate memory for storing MAC addresses */
>  	eth_dev->data->mac_addrs = rte_zmalloc("virtio", ETHER_ADDR_LEN,
> 0);
> @@ -1009,7 +1008,7 @@ static void virtio_dev_free_mbufs(struct rte_eth_dev
> *dev)
> 
>  		while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
>  					dev->data->rx_queues[i])) != NULL) {
> -			rte_pktmbuf_free_seg(buf);
> +			rte_pktmbuf_free(buf);
>  			mbuf_num++;
>  		}
> 
> @@ -1028,7 +1027,8 @@ static void virtio_dev_free_mbufs(struct rte_eth_dev
> *dev)
>  		mbuf_num = 0;
>  		while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
>  					dev->data->tx_queues[i])) != NULL) {
> -			rte_pktmbuf_free_seg(buf);
> +			rte_pktmbuf_free(buf);
> +
>  			mbuf_num++;
>  		}
> 
> diff --git a/lib/librte_pmd_virtio/virtio_ethdev.h
> b/lib/librte_pmd_virtio/virtio_ethdev.h
> index 858e644..d2e1eed 100644
> --- a/lib/librte_pmd_virtio/virtio_ethdev.h
> +++ b/lib/librte_pmd_virtio/virtio_ethdev.h
> @@ -104,6 +104,9 @@ int  virtio_dev_tx_queue_setup(struct rte_eth_dev *dev,
> uint16_t tx_queue_id,
>  uint16_t virtio_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
>  		uint16_t nb_pkts);
> 
> +uint16_t virtio_recv_mergeable_pkts(void *rx_queue, struct rte_mbuf
> **rx_pkts,
> +		uint16_t nb_pkts);
> +
>  uint16_t virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
>  		uint16_t nb_pkts);
> 
> diff --git a/lib/librte_pmd_virtio/virtio_rxtx.c
> b/lib/librte_pmd_virtio/virtio_rxtx.c
> index fcd8bd1..3d81b34 100644
> --- a/lib/librte_pmd_virtio/virtio_rxtx.c
> +++ b/lib/librte_pmd_virtio/virtio_rxtx.c
> @@ -146,6 +146,7 @@ static inline int
>  virtqueue_enqueue_recv_refill(struct virtqueue *vq, struct rte_mbuf *cookie)
>  {
>  	struct vq_desc_extra *dxp;
> +	struct virtio_hw *hw = vq->hw;
>  	struct vring_desc *start_dp;
>  	uint16_t needed = 1;
>  	uint16_t head_idx, idx;
> @@ -165,9 +166,11 @@ virtqueue_enqueue_recv_refill(struct virtqueue *vq,
> struct rte_mbuf *cookie)
>  	dxp->ndescs = needed;
> 
>  	start_dp = vq->vq_ring.desc;
> -	start_dp[idx].addr  =
> -		(uint64_t) (cookie->buf_physaddr + RTE_PKTMBUF_HEADROOM
> - sizeof(struct virtio_net_hdr));
> -	start_dp[idx].len   = cookie->buf_len - RTE_PKTMBUF_HEADROOM +
> sizeof(struct virtio_net_hdr);
> +	start_dp[idx].addr =
> +		(uint64_t)(cookie->buf_physaddr + RTE_PKTMBUF_HEADROOM
> +		- hw->vtnet_hdr_size);
> +	start_dp[idx].len =
> +		cookie->buf_len - RTE_PKTMBUF_HEADROOM + hw-
> >vtnet_hdr_size;
>  	start_dp[idx].flags =  VRING_DESC_F_WRITE;
>  	idx = start_dp[idx].next;
>  	vq->vq_desc_head_idx = idx;
> @@ -184,8 +187,10 @@ virtqueue_enqueue_xmit(struct virtqueue *txvq, struct
> rte_mbuf *cookie)
>  {
>  	struct vq_desc_extra *dxp;
>  	struct vring_desc *start_dp;
> -	uint16_t needed = 2;
> +	uint16_t seg_num = cookie->pkt.nb_segs;
> +	uint16_t needed = 1 + seg_num;
>  	uint16_t head_idx, idx;
> +	uint16_t head_size = txvq->hw->vtnet_hdr_size;
> 
>  	if (unlikely(txvq->vq_free_cnt == 0))
>  		return -ENOSPC;
> @@ -198,19 +203,25 @@ virtqueue_enqueue_xmit(struct virtqueue *txvq,
> struct rte_mbuf *cookie)
>  	idx = head_idx;
>  	dxp = &txvq->vq_descx[idx];
>  	if (dxp->cookie != NULL)
> -		rte_pktmbuf_free_seg(dxp->cookie);
> +		rte_pktmbuf_free(dxp->cookie);
>  	dxp->cookie = (void *)cookie;
>  	dxp->ndescs = needed;
> 
>  	start_dp = txvq->vq_ring.desc;
> -	start_dp[idx].addr  =
> -		txvq->virtio_net_hdr_mem + idx * sizeof(struct virtio_net_hdr);
> -	start_dp[idx].len   = sizeof(struct virtio_net_hdr);
> +	start_dp[idx].addr =
> +		txvq->virtio_net_hdr_mem + idx * head_size;
> +	start_dp[idx].len = (uint32_t)head_size;
>  	start_dp[idx].flags = VRING_DESC_F_NEXT;
> -	idx = start_dp[idx].next;
> -	start_dp[idx].addr  = RTE_MBUF_DATA_DMA_ADDR(cookie);
> -	start_dp[idx].len   = cookie->pkt.data_len;
> -	start_dp[idx].flags = 0;
> +
> +	for (; ((seg_num > 0) && (cookie != NULL)); seg_num--) {
> +		idx = start_dp[idx].next;
> +		start_dp[idx].addr  = RTE_MBUF_DATA_DMA_ADDR(cookie);
> +		start_dp[idx].len   = cookie->pkt.data_len;
> +		start_dp[idx].flags = VRING_DESC_F_NEXT;
> +		cookie = cookie->pkt.next;
> +	}
> +
> +	start_dp[idx].flags &= ~VRING_DESC_F_NEXT;
>  	idx = start_dp[idx].next;
>  	txvq->vq_desc_head_idx = idx;
>  	if (txvq->vq_desc_head_idx == VQ_RING_DESC_CHAIN_END)
> @@ -284,7 +295,7 @@ virtio_dev_vring_start(struct virtqueue *vq, int
> queue_type)
>  			error = virtqueue_enqueue_recv_refill(vq, m);
> 
>  			if (error) {
> -				rte_pktmbuf_free_seg(m);
> +				rte_pktmbuf_free(m);
>  				break;
>  			}
>  			nbufs++;
> @@ -423,7 +434,7 @@ virtio_discard_rxbuf(struct virtqueue *vq, struct
> rte_mbuf *m)
>  	error = virtqueue_enqueue_recv_refill(vq, m);
>  	if (unlikely(error)) {
>  		RTE_LOG(ERR, PMD, "cannot requeue discarded mbuf");
> -		rte_pktmbuf_free_seg(m);
> +		rte_pktmbuf_free(m);
>  	}
>  }
> 
> @@ -471,17 +482,158 @@ virtio_recv_pkts(void *rx_queue, struct rte_mbuf
> **rx_pkts, uint16_t nb_pkts)
> 
>  		rxm->pkt.in_port = rxvq->port_id;
>  		rxm->pkt.data = (char *)rxm->buf_addr +
> RTE_PKTMBUF_HEADROOM;
> +
>  		rxm->pkt.nb_segs = 1;
>  		rxm->pkt.next = NULL;
> -		rxm->pkt.pkt_len  = (uint32_t)(len[i]
> -					       - sizeof(struct virtio_net_hdr));
> -		rxm->pkt.data_len = (uint16_t)(len[i]
> -					       - sizeof(struct virtio_net_hdr));
> +		rxm->pkt.pkt_len = (uint32_t)(len[i] - hw->vtnet_hdr_size);
> +		rxm->pkt.data_len = (uint16_t)(len[i] - hw->vtnet_hdr_size);
> 
>  		VIRTIO_DUMP_PACKET(rxm, rxm->pkt.data_len);
> 
>  		rx_pkts[nb_rx++] = rxm;
> -		rxvq->bytes += len[i] - sizeof(struct virtio_net_hdr);
> +		rxvq->bytes += rx_pkts[nb_rx - 1]->pkt.pkt_len;
> +	}
> +
> +	rxvq->packets += nb_rx;
> +
> +	/* Allocate new mbuf for the used descriptor */
> +	error = ENOSPC;
> +	while (likely(!virtqueue_full(rxvq))) {
> +		new_mbuf = rte_rxmbuf_alloc(rxvq->mpool);
> +		if (unlikely(new_mbuf == NULL)) {
> +			struct rte_eth_dev *dev
> +				= &rte_eth_devices[rxvq->port_id];
> +			dev->data->rx_mbuf_alloc_failed++;
> +			break;
> +		}
> +		error = virtqueue_enqueue_recv_refill(rxvq, new_mbuf);
> +		if (unlikely(error)) {
> +			rte_pktmbuf_free(new_mbuf);
> +			break;
> +		}
> +		nb_enqueued++;
> +	}
> +
> +	if (likely(nb_enqueued)) {
> +		if (unlikely(virtqueue_kick_prepare(rxvq))) {
> +			virtqueue_notify(rxvq);
> +			PMD_RX_LOG(DEBUG, "Notified\n");
> +		}
> +	}
> +
> +	vq_update_avail_idx(rxvq);
> +
> +	return nb_rx;
> +}
> +
> +uint16_t
> +virtio_recv_mergeable_pkts(void *rx_queue,
> +			struct rte_mbuf **rx_pkts,
> +			uint16_t nb_pkts)
> +{
> +	struct virtqueue *rxvq = rx_queue;
> +	struct virtio_hw *hw = rxvq->hw;
> +	struct rte_mbuf *rxm, *new_mbuf;
> +	uint16_t nb_used, num, nb_rx = 0;
> +	uint32_t len[VIRTIO_MBUF_BURST_SZ];
> +	struct rte_mbuf *rcv_pkts[VIRTIO_MBUF_BURST_SZ];
> +	struct rte_mbuf *prev;
> +	int error;
> +	uint32_t i = 0, nb_enqueued = 0;
> +	uint32_t seg_num = 0;
> +	uint16_t extra_idx = 0;
> +
> +	nb_used = VIRTQUEUE_NUSED(rxvq);
> +
> +	rmb();
> +
> +	if (nb_used == 0)
> +		return 0;
> +
> +	while (i < nb_used) {
> +		struct virtio_net_hdr_mrg_rxbuf *header;
> +		char *head_ptr;
> +
> +		if (nb_rx >= nb_pkts)
> +			break;
> +
> +		num = virtqueue_dequeue_burst_rx(rxvq, rcv_pkts, len, 1);
> +		i += num;
> +
> +		PMD_RX_LOG(DEBUG, "used:%d dequeue:%d\n", nb_used,
> num);
> +		PMD_RX_LOG(DEBUG, "packet len:%d\n", len[i]);
> +
> +		rxm = rcv_pkts[0];
> +		extra_idx = 0;
> +
> +		if (unlikely(len[0]
> +			     < (uint32_t)hw->vtnet_hdr_size + ETHER_HDR_LEN))
> {
> +			PMD_RX_LOG(ERR, "Packet drop\n");
> +			nb_enqueued++;
> +			virtio_discard_rxbuf(rxvq, rxm);
> +			rxvq->errors++;
> +			continue;
> +		}
> +
> +		head_ptr = (char *)rxm->pkt.data;
> +		head_ptr -= hw->vtnet_hdr_size;
> +		header = (struct virtio_net_hdr_mrg_rxbuf *)head_ptr;
> +		seg_num = header->num_buffers;
> +
> +		if (seg_num == 0)
> +			seg_num = 1;
> +
> +		rxm->pkt.data = (char *)rxm->buf_addr +
> RTE_PKTMBUF_HEADROOM;
> +		rxm->pkt.nb_segs = seg_num;
> +		rxm->pkt.next = NULL;
> +		rxm->pkt.pkt_len = (uint32_t)(len[0] - hw->vtnet_hdr_size);
> +		rxm->pkt.data_len = (uint16_t)(len[0] - hw->vtnet_hdr_size);
> +
> +		rxm->pkt.in_port = rxvq->port_id;
> +		rx_pkts[nb_rx] = rxm;
> +
> +		prev = rxm;
> +
> +		VIRTIO_DUMP_PACKET(rxm, rxm->pkt.data_len);
Here it might cause segmentation fault when debug is enabled.
> +
> +		/*
> +		 * Get extra segments for current uncompleted packet.
> +		 */
> +		if (VIRTQUEUE_NUSED(rxvq) >= seg_num - 1) {
> +			uint32_t rx_num = virtqueue_dequeue_burst_rx(rxvq,
> +				rcv_pkts, len, seg_num - 1);
> +			i += rx_num;
> +		} else {
> +			PMD_RX_LOG(ERR, "No enough segments for
> packet.\n");
> +			nb_enqueued++;
> +			virtio_discard_rxbuf(rxvq, rxm);
> +			rxvq->errors++;
> +			break;
Here we should figure out if virtio net spec specify the behavior. If the backend wants to en-queue a merge able packet, but there isn't enough available descriptors, could it do it in a non-atomic way, i.e., firstly transfer some of them, update the used->idx and later transfer the rest of them? 
If that is the case,  the handling here will cause the content of next segment later en-queued to be treated as virtio merge-able header.

> +		}
> +
> +		while (extra_idx < seg_num - 1) {
> +			rxm = rcv_pkts[extra_idx];
> +
> +			rxm->pkt.data =
> +				(char *)rxm->buf_addr +
> RTE_PKTMBUF_HEADROOM
> +				- hw->vtnet_hdr_size;
> +			rxm->pkt.next = NULL;
> +			rxm->pkt.pkt_len = (uint32_t)(len[extra_idx]);
> +			rxm->pkt.data_len = (uint16_t)(len[extra_idx]);
> +
> +			if (prev)
> +				prev->pkt.next = rxm;
> +
> +			prev = rxm;
> +			rx_pkts[nb_rx]->pkt.pkt_len += rxm->pkt.pkt_len;
> +
> +			extra_idx++;
> +
> +			VIRTIO_DUMP_PACKET(rxm, rxm->pkt.data_len);
> +		};
> +
> +		rxvq->bytes += rx_pkts[nb_rx]->pkt.pkt_len;
> +		nb_rx++;
>  	}
> 
>  	rxvq->packets += nb_rx;
> @@ -498,11 +650,12 @@ virtio_recv_pkts(void *rx_queue, struct rte_mbuf
> **rx_pkts, uint16_t nb_pkts)
>  		}
>  		error = virtqueue_enqueue_recv_refill(rxvq, new_mbuf);
>  		if (unlikely(error)) {
> -			rte_pktmbuf_free_seg(new_mbuf);
> +			rte_pktmbuf_free(new_mbuf);
>  			break;
>  		}
>  		nb_enqueued++;
>  	}
> +
>  	if (likely(nb_enqueued)) {
>  		if (unlikely(virtqueue_kick_prepare(rxvq))) {
>  			virtqueue_notify(rxvq);
> @@ -536,12 +689,15 @@ virtio_xmit_pkts(void *tx_queue, struct rte_mbuf
> **tx_pkts, uint16_t nb_pkts)
>  	num = (uint16_t)(likely(nb_used < VIRTIO_MBUF_BURST_SZ) ? nb_used :
> VIRTIO_MBUF_BURST_SZ);
> 
>  	while (nb_tx < nb_pkts) {
> -		if (virtqueue_full(txvq) && num) {
> -			virtqueue_dequeue_pkt_tx(txvq);
> -			num--;
> +		int need = tx_pkts[nb_tx]->pkt.nb_segs - txvq->vq_free_cnt;
> +		if ((need > 0) && (num > 0)) {
> +			do {
> +				virtqueue_dequeue_pkt_tx(txvq);
> +				num--;
> +			} while (num > 0);
>  		}
> 
> -		if (!virtqueue_full(txvq)) {
> +		if (tx_pkts[nb_tx]->pkt.nb_segs <= txvq->vq_free_cnt) {
>  			txm = tx_pkts[nb_tx];
>  			/* Enqueue Packet buffers */
>  			error = virtqueue_enqueue_xmit(txvq, txm);
> @@ -555,7 +711,7 @@ virtio_xmit_pkts(void *tx_queue, struct rte_mbuf
> **tx_pkts, uint16_t nb_pkts)
>  				break;
>  			}
>  			nb_tx++;
> -			txvq->bytes += txm->pkt.data_len;
> +			txvq->bytes += txm->pkt.pkt_len;
>  		} else {
>  			PMD_TX_LOG(ERR, "No free tx descriptors to transmit");
>  			break;
> --
> 1.8.4.2

  reply	other threads:[~2014-07-25  8:04 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-07-25  6:03 Ouyang Changchun
2014-07-25  8:05 ` Xie, Huawei [this message]
2014-07-30  5:33 ` Fu, JingguoX

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=C37D651A908B024F974696C65296B57B0F24595E@SHSMSX101.ccr.corp.intel.com \
    --to=huawei.xie@intel.com \
    --cc=changchun.ouyang@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).