Re: [PATCH] vhost: support sync copy when data buffer is small

[Maxime Coquelin <maxime.coquelin@redhat.com>]

Hi Jiayu,

On 5/23/22 02:38, Hu, Jiayu wrote:
> Hi Maxime,
> 
>> -----Original Message-----
>> From: Maxime Coquelin <maxime.coquelin@redhat.com>
>> Sent: Tuesday, April 19, 2022 5:10 PM
>> To: Ma, WenwuX <wenwux.ma@intel.com>; Xia, Chenbo
>> <chenbo.xia@intel.com>; dev@dpdk.org
>> Cc: Hu, Jiayu <jiayu.hu@intel.com>; Wang, Yinan <yinan.wang@intel.com>;
>> He, Xingguang <xingguang.he@intel.com>; Ling, WeiX <weix.ling@intel.com>
>> Subject: Re: [PATCH] vhost: support sync copy when data buffer is small
>>
>> Hi Wenwu,
>>
>> On 4/11/22 15:38, Wenwu Ma wrote:
>>> In async datapath, if the length of data buffer is less than 256, the
>>> data will be copied by CPU instead of DMA.
>>>
>>> Signed-off-by: Wenwu Ma <wenwux.ma@intel.com>
>>> ---
>>>    lib/vhost/vhost.h      |  6 ++-
>>>    lib/vhost/virtio_net.c | 96 ++++++++++++++++++++++++++++++------------
>>>    2 files changed, 73 insertions(+), 29 deletions(-)
>>>
>>
>> As I mentioned in the past, let's have a DMA-only solution working before
>> trying to support CPU copy for small packets.
> 
> Thanks for your suggestion.
> 
> For "DMA-only solution working", do you mean community reaches an
> agreement on OVS DSA design? The reason of asking this question is
> that we will have some following patches around async vhost in 22.11,
> like vhostpmd supporting async vhost data-path. Regarding your suggestion,
> do you think they also need to be deferred until "DMA-only solution working"?

No, I was just referring to CPU copy for small buffers features.

Thanks,
Maxime

> Thanks,
> Jiayu
> 
>>
>> Thanks,
>> Maxime
>