RE: [PATCH v4 0/5] vhost: support async dequeue data path

["Ding, Xuan" <xuan.ding@intel.com>]

Hi Maxime,

> -----Original Message-----
> From: Maxime Coquelin <maxime.coquelin@redhat.com>
> Sent: Friday, May 6, 2022 3:53 AM
> To: Ding, Xuan <xuan.ding@intel.com>; Xia, Chenbo <chenbo.xia@intel.com>
> Cc: dev@dpdk.org; Hu, Jiayu <jiayu.hu@intel.com>; Jiang, Cheng1
> <cheng1.jiang@intel.com>; Pai G, Sunil <sunil.pai.g@intel.com>;
> liangma@liangbit.com
> Subject: Re: [PATCH v4 0/5] vhost: support async dequeue data path
> 
> Hi Xuan,
> 
> On 5/5/22 08:23, xuan.ding@intel.com wrote:
> > From: Xuan Ding <xuan.ding@intel.com>
> >
> > The presence of asynchronous path allows applications to offload
> > memory copies to DMA engine, so as to save CPU cycles and improve the
> > copy performance. This patch set implements vhost async dequeue data
> > path for split ring. The code is based on latest enqueue changes [1].
> >
> > This patch set is a new design and implementation of [2]. Since dmadev
> > was introduced in DPDK 21.11, to simplify application logics, this
> > patch integrates dmadev in vhost. With dmadev integrated, vhost
> > supports M:N mapping between vrings and DMA virtual channels.
> > Specifically, one vring can use multiple different DMA channels and
> > one DMA channel can be shared by multiple vrings at the same time.
> >
> > A new asynchronous dequeue function is introduced:
> >          1) rte_vhost_async_try_dequeue_burst(int vid, uint16_t queue_id,
> >                  struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts,
> >                  uint16_t count, int *nr_inflight,
> >                  uint16_t dma_id, uint16_t vchan_id)
> >
> >          Receive packets from the guest and offloads copies to DMA
> > virtual channel.
> >
> > [1] https://mails.dpdk.org/archives/dev/2022-February/234555.html
> > [2] https://mails.dpdk.org/archives/dev/2021-September/218591.html
> >
> > v3->v4:
> > * fix CI build warnings
> > * adjust some indentation
> > * pass vq instead of queue_id
> >
> > v2->v3:
> > * fix mbuf not updated correctly for large packets
> >
> > v1->v2:
> > * fix a typo
> > * fix a bug in desc_to_mbuf filling
> >
> > RFC v3 -> v1:
> > * add sync and async path descriptor to mbuf refactoring
> > * add API description in docs
> >
> > RFC v2 -> RFC v3:
> > * rebase to latest DPDK version
> >
> > RFC v1 -> RFC v2:
> > * fix one bug in example
> > * rename vchan to vchan_id
> > * check if dma_id and vchan_id valid
> > * rework all the logs to new standard
> >
> > Xuan Ding (5):
> >    vhost: prepare sync for descriptor to mbuf refactoring
> >    vhost: prepare async for descriptor to mbuf refactoring
> >    vhost: merge sync and async descriptor to mbuf filling
> >    vhost: support async dequeue for split ring
> >    examples/vhost: support async dequeue data path
> >
> >   doc/guides/prog_guide/vhost_lib.rst    |   7 +
> >   doc/guides/rel_notes/release_22_07.rst |   4 +
> >   doc/guides/sample_app_ug/vhost.rst     |   9 +-
> >   examples/vhost/main.c                  | 292 +++++++++++-----
> >   examples/vhost/main.h                  |  35 +-
> >   examples/vhost/virtio_net.c            |  16 +-
> >   lib/vhost/rte_vhost_async.h            |  33 ++
> >   lib/vhost/version.map                  |   3 +
> >   lib/vhost/vhost.h                      |   1 +
> >   lib/vhost/virtio_net.c                 | 467 ++++++++++++++++++++++---
> >   10 files changed, 716 insertions(+), 151 deletions(-)
> >
> 
> I applied your other series about unsafe API to get DMA inflight packets, so I
> have some conflicts when applying this series.
> 
> Could you please rebase on top of next-virtio/main branch and repost?

Sure, I will rebase on top of latest main branch, please see v5.

Thanks,
Xuan

> 
> Thanks in avance,
> Maxime