Re: [dpdk-dev] [PATCH v2] net/vhost: merge vhost stats loop in vhost Tx/Rx

["Gaoxiang Liu" <gaoxiangliu0@163.com>]

Hi, Maxime
I agree with you.The inline should be added to vhost_update_single_packet_xstats function.
I will fix it in [PATCH v3].

Thanks,
Gaoxiang



发自 网易邮箱大师




---- 回复的原邮件 ----
| 发件人 | Maxime Coquelin<maxime.coquelin@redhat.com> |
| 日期 | 2021年10月15日 20:16 |
| 收件人 | Gaoxiang Liu<gaoxiangliu0@163.com>、chenbo.xia@intel.com<chenbo.xia@intel.com> |
| 抄送至 | dev@dpdk.org<dev@dpdk.org>、liugaoxiang@huawei.com<liugaoxiang@huawei.com> |
| 主题 | Re: [PATCH v2] net/vhost: merge vhost stats loop in vhost Tx/Rx |
Hi,

On 9/28/21 03:43, Gaoxiang Liu wrote:
> To improve performance in vhost Tx/Rx, merge vhost stats loop.
> eth_vhost_tx has 2 loop of send num iteraion.
> It can be merge into one.
> eth_vhost_rx has the same issue as Tx.
>
> Fixes: 4d6cf2ac93dc ("net/vhost: add extended statistics")

Please remove the Fixes tag, this is an optimization, not a fix.

>
> Signed-off-by: Gaoxiang Liu <gaoxiangliu0@163.com>
> ---
>
> v2:
>   * Fix coding style issues.
> ---
>   drivers/net/vhost/rte_eth_vhost.c | 62 ++++++++++++++-----------------
>   1 file changed, 28 insertions(+), 34 deletions(-)
>
> diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
> index a202931e9a..a4129980f2 100644
> --- a/drivers/net/vhost/rte_eth_vhost.c
> +++ b/drivers/net/vhost/rte_eth_vhost.c
> @@ -336,38 +336,29 @@ vhost_count_xcast_packets(struct vhost_queue *vq,
>   }
>  
>   static void
> -vhost_update_packet_xstats(struct vhost_queue *vq, struct rte_mbuf **bufs,
> -                  uint16_t count, uint64_t nb_bytes,
> -                  uint64_t nb_missed)
> +vhost_update_single_packet_xstats(struct vhost_queue *vq, struct rte_mbuf *buf)

I tried to build without and with your patch, and I think that what can
explain most of the performance difference is that without your patch
the function is not inlined, whereas it is implicitely inlined with your
patch applied.

I agree with your patch, but I think we might add __rte_always_inline to
this function to make it explicit. What do you think?

Other than that:

Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>

Thanks,
Maxime


>   {
>        uint32_t pkt_len = 0;
> -     uint64_t i = 0;
>        uint64_t index;
>        struct vhost_stats *pstats = &vq->stats;
>  
> -     pstats->xstats[VHOST_BYTE] += nb_bytes;
> -     pstats->xstats[VHOST_MISSED_PKT] += nb_missed;
> -     pstats->xstats[VHOST_UNICAST_PKT] += nb_missed;
> -
> -     for (i = 0; i < count ; i++) {
> -          pstats->xstats[VHOST_PKT]++;
> -          pkt_len = bufs[i]->pkt_len;
> -          if (pkt_len == 64) {
> -               pstats->xstats[VHOST_64_PKT]++;
> -          } else if (pkt_len > 64 && pkt_len < 1024) {
> -               index = (sizeof(pkt_len) * 8)
> -                    - __builtin_clz(pkt_len) - 5;
> -               pstats->xstats[index]++;
> -          } else {
> -               if (pkt_len < 64)
> -                    pstats->xstats[VHOST_UNDERSIZE_PKT]++;
> -               else if (pkt_len <= 1522)
> -                    pstats->xstats[VHOST_1024_TO_1522_PKT]++;
> -               else if (pkt_len > 1522)
> -                    pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
> -          }
> -          vhost_count_xcast_packets(vq, bufs[i]);
> +     pstats->xstats[VHOST_PKT]++;
> +     pkt_len = buf->pkt_len;
> +     if (pkt_len == 64) {
> +          pstats->xstats[VHOST_64_PKT]++;
> +     } else if (pkt_len > 64 && pkt_len < 1024) {
> +          index = (sizeof(pkt_len) * 8)
> +               - __builtin_clz(pkt_len) - 5;
> +          pstats->xstats[index]++;
> +     } else {
> +          if (pkt_len < 64)
> +               pstats->xstats[VHOST_UNDERSIZE_PKT]++;
> +          else if (pkt_len <= 1522)
> +               pstats->xstats[VHOST_1024_TO_1522_PKT]++;
> +          else if (pkt_len > 1522)
> +               pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
>        }
> +     vhost_count_xcast_packets(vq, buf);
>   }
>  
>   static uint16_t
> @@ -376,7 +367,6 @@ eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
>        struct vhost_queue *r = q;
>        uint16_t i, nb_rx = 0;
>        uint16_t nb_receive = nb_bufs;
> -     uint64_t nb_bytes = 0;
>  
>        if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
>             return 0;
> @@ -411,11 +401,11 @@ eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
>             if (r->internal->vlan_strip)
>                  rte_vlan_strip(bufs[i]);
>  
> -          nb_bytes += bufs[i]->pkt_len;
> -     }
> +          r->stats.bytes += bufs[i]->pkt_len;
> +          r->stats.xstats[VHOST_BYTE] += bufs[i]->pkt_len;
>  
> -     r->stats.bytes += nb_bytes;
> -     vhost_update_packet_xstats(r, bufs, nb_rx, nb_bytes, 0);
> +          vhost_update_single_packet_xstats(r, bufs[i]);
> +     }
>  
>   out:
>        rte_atomic32_set(&r->while_queuing, 0);
> @@ -471,16 +461,20 @@ eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
>                  break;
>        }
>  
> -     for (i = 0; likely(i < nb_tx); i++)
> +     for (i = 0; likely(i < nb_tx); i++) {
>             nb_bytes += bufs[i]->pkt_len;
> +          vhost_update_single_packet_xstats(r, bufs[i]);
> +     }
>  
>        nb_missed = nb_bufs - nb_tx;
>  
>        r->stats.pkts += nb_tx;
>        r->stats.bytes += nb_bytes;
> -     r->stats.missed_pkts += nb_bufs - nb_tx;
> +     r->stats.missed_pkts += nb_missed;
>  
> -     vhost_update_packet_xstats(r, bufs, nb_tx, nb_bytes, nb_missed);
> +     r->stats.xstats[VHOST_BYTE] += nb_bytes;
> +     r->stats.xstats[VHOST_MISSED_PKT] += nb_missed;
> +     r->stats.xstats[VHOST_UNICAST_PKT] += nb_missed;
>  
>        /* According to RFC2863, ifHCOutUcastPkts, ifHCOutMulticastPkts and
>         * ifHCOutBroadcastPkts counters are increased when packets are not
>