[dpdk-dev] [PATCH v2 10/12] virtio: add Tx checksum offload support

Olivier Matz olivier.matz at 6wind.com
Fri Oct 7 18:36:35 CEST 2016


Hi Maxime,

On 10/07/2016 09:25 AM, Maxime Coquelin wrote:
> Hi Olivier,
> 
> On 10/03/2016 11:00 AM, Olivier Matz wrote:
>> Signed-off-by: Olivier Matz <olivier.matz at 6wind.com>
>> ---
>>  drivers/net/virtio/virtio_ethdev.c |  7 +++++
>>  drivers/net/virtio/virtio_ethdev.h |  1 +
>>  drivers/net/virtio/virtio_rxtx.c   | 57
>> +++++++++++++++++++++++++-------------
>>  3 files changed, 45 insertions(+), 20 deletions(-)
>>
>> diff --git a/drivers/net/virtio/virtio_ethdev.c
>> b/drivers/net/virtio/virtio_ethdev.c
>> index 43cb096..55024cd 100644
>> --- a/drivers/net/virtio/virtio_ethdev.c
>> +++ b/drivers/net/virtio/virtio_ethdev.c
>> @@ -1578,6 +1578,13 @@ virtio_dev_info_get(struct rte_eth_dev *dev,
>> struct rte_eth_dev_info *dev_info)
>>      dev_info->rx_offload_capa =
>>          DEV_RX_OFFLOAD_TCP_CKSUM |
>>          DEV_RX_OFFLOAD_UDP_CKSUM;
>> +    dev_info->tx_offload_capa = 0;
>> +
>> +    if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
>> +        dev_info->tx_offload_capa |=
>> +            DEV_TX_OFFLOAD_UDP_CKSUM |
>> +            DEV_TX_OFFLOAD_TCP_CKSUM;
>> +    }
>>  }
>>
>>  /*
>> diff --git a/drivers/net/virtio/virtio_ethdev.h
>> b/drivers/net/virtio/virtio_ethdev.h
>> index 2fc9218..202aa2e 100644
>> --- a/drivers/net/virtio/virtio_ethdev.h
>> +++ b/drivers/net/virtio/virtio_ethdev.h
>> @@ -62,6 +62,7 @@
>>       1u << VIRTIO_NET_F_CTRL_VQ      |    \
>>       1u << VIRTIO_NET_F_CTRL_RX      |    \
>>       1u << VIRTIO_NET_F_CTRL_VLAN      |    \
>> +     1u << VIRTIO_NET_F_CSUM      |    \
>>       1u << VIRTIO_NET_F_MRG_RXBUF      |    \
>>       1ULL << VIRTIO_F_VERSION_1)
>>
>> diff --git a/drivers/net/virtio/virtio_rxtx.c
>> b/drivers/net/virtio/virtio_rxtx.c
>> index eda678a..4ae11e7 100644
>> --- a/drivers/net/virtio/virtio_rxtx.c
>> +++ b/drivers/net/virtio/virtio_rxtx.c
>> @@ -213,13 +213,14 @@ static inline void
>>  virtqueue_enqueue_xmit(struct virtnet_tx *txvq, struct rte_mbuf *cookie,
>>                 uint16_t needed, int use_indirect, int can_push)
>>  {
>> +    struct virtio_tx_region *txr = txvq->virtio_net_hdr_mz->addr;
>>      struct vq_desc_extra *dxp;
>>      struct virtqueue *vq = txvq->vq;
>>      struct vring_desc *start_dp;
>>      uint16_t seg_num = cookie->nb_segs;
>>      uint16_t head_idx, idx;
>>      uint16_t head_size = vq->hw->vtnet_hdr_size;
>> -    unsigned long offs;
>> +    struct virtio_net_hdr *hdr;
>>
>>      head_idx = vq->vq_desc_head_idx;
>>      idx = head_idx;
>> @@ -230,10 +231,9 @@ virtqueue_enqueue_xmit(struct virtnet_tx *txvq,
>> struct rte_mbuf *cookie,
>>      start_dp = vq->vq_ring.desc;
>>
>>      if (can_push) {
>> -        /* put on zero'd transmit header (no offloads) */
>> -        void *hdr = rte_pktmbuf_prepend(cookie, head_size);
>> -
>> -        memset(hdr, 0, head_size);
>> +        /* prepend cannot fail, checked by caller */
>> +        hdr = (struct virtio_net_hdr *)
>> +            rte_pktmbuf_prepend(cookie, head_size);
>>      } else if (use_indirect) {
>>          /* setup tx ring slot to point to indirect
>>           * descriptor list stored in reserved region.
>> @@ -241,14 +241,11 @@ virtqueue_enqueue_xmit(struct virtnet_tx *txvq,
>> struct rte_mbuf *cookie,
>>           * the first slot in indirect ring is already preset
>>           * to point to the header in reserved region
>>           */
>> -        struct virtio_tx_region *txr = txvq->virtio_net_hdr_mz->addr;
>> -
>> -        offs = idx * sizeof(struct virtio_tx_region)
>> -            + offsetof(struct virtio_tx_region, tx_indir);
>> -
>> -        start_dp[idx].addr  = txvq->virtio_net_hdr_mem + offs;
>> +        start_dp[idx].addr  = txvq->virtio_net_hdr_mem +
>> +            RTE_PTR_DIFF(&txr[idx].tx_indir, txr);
>>          start_dp[idx].len   = (seg_num + 1) * sizeof(struct vring_desc);
>>          start_dp[idx].flags = VRING_DESC_F_INDIRECT;
>> +        hdr = (struct virtio_net_hdr *)&txr[idx].tx_hdr;
>>
>>          /* loop below will fill in rest of the indirect elements */
>>          start_dp = txr[idx].tx_indir;
>> @@ -257,15 +254,40 @@ virtqueue_enqueue_xmit(struct virtnet_tx *txvq,
>> struct rte_mbuf *cookie,
>>          /* setup first tx ring slot to point to header
>>           * stored in reserved region.
>>           */
>> -        offs = idx * sizeof(struct virtio_tx_region)
>> -            + offsetof(struct virtio_tx_region, tx_hdr);
>> -
>> -        start_dp[idx].addr  = txvq->virtio_net_hdr_mem + offs;
>> +        start_dp[idx].addr  = txvq->virtio_net_hdr_mem +
>> +            RTE_PTR_DIFF(&txr[idx].tx_hdr, txr);
>>          start_dp[idx].len   = vq->hw->vtnet_hdr_size;
>>          start_dp[idx].flags = VRING_DESC_F_NEXT;
>> +        hdr = (struct virtio_net_hdr *)&txr[idx].tx_hdr;
>> +
>>          idx = start_dp[idx].next;
>>      }
>>
>> +    /* Checksum Offload */
>> +    switch (cookie->ol_flags & PKT_TX_L4_MASK) {
>> +    case PKT_TX_UDP_CKSUM:
>> +        hdr->csum_start = cookie->l2_len + cookie->l3_len;
>> +        hdr->csum_offset = 6;
>> +        hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
>> +        break;
>> +
>> +    case PKT_TX_TCP_CKSUM:
>> +        hdr->csum_start = cookie->l2_len + cookie->l3_len;
>> +        hdr->csum_offset = 16;
>> +        hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
>> +        break;
>> +
>> +    default:
>> +        hdr->csum_start = 0;
>> +        hdr->csum_offset = 0;
>> +        hdr->flags = 0;
>> +        break;
>> +    }
>> +
>> +    hdr->gso_type = 0;
>> +    hdr->gso_size = 0;
>> +    hdr->hdr_len = 0;
> 
> In he case we don't use any offload, have you measured the performance
> regression
> with current code when using a dedicated descriptor for the header?
> I haven't tested you series, but I would think it is more than 15% for
> 64 bytes packets based on my trials to use a single descriptor.
> 
> Indeed, without your series, when using a dedicated desc for the header,
> the header is not accessed in the virtio transmit path. It is zeroed at
> init time.
> 
> Could we keep the same behaviour when offloading features aren't
> negotiated?

You're right, it could have a performance impact. I'll try to restore
the initial behavior when offload is disabled.

Thanks,
Olivier


More information about the dev mailing list