Re: [dpdk-dev] [RFC] Accelerating Data Movement for DPDK vHost with DMA Engines

[Jerin Jacob <jerinjacobk@gmail.com>]

On Fri, Apr 17, 2020 at 2:56 PM Fu, Patrick <patrick.fu@intel.com> wrote:
>
>
>
> > -----Original Message-----
> > From: Maxime Coquelin <maxime.coquelin@redhat.com>
> > Sent: Friday, April 17, 2020 4:40 PM
> > To: Fu, Patrick <patrick.fu@intel.com>; Jerin Jacob <jerinjacobk@gmail.com>
> > Cc: dev@dpdk.org; Ye, Xiaolong <xiaolong.ye@intel.com>; Hu, Jiayu
> > <jiayu.hu@intel.com>; Wang, Zhihong <zhihong.wang@intel.com>; Liang,
> > Cunming <cunming.liang@intel.com>
> > Subject: Re: [dpdk-dev] [RFC] Accelerating Data Movement for DPDK vHost
> > with DMA Engines
> >
> >
> >
> > On 4/17/20 10:29 AM, Fu, Patrick wrote:
> > > Hi Jerin,
> > >
> > >> -----Original Message-----
> > >> From: Jerin Jacob <jerinjacobk@gmail.com>
> > >> Sent: Friday, April 17, 2020 4:02 PM
> > >> To: Fu, Patrick <patrick.fu@intel.com>
> > >> Cc: dev@dpdk.org; Maxime Coquelin <maxime.coquelin@redhat.com>;
> > Ye,
> > >> Xiaolong <xiaolong.ye@intel.com>; Hu, Jiayu <jiayu.hu@intel.com>;
> > >> Wang, Zhihong <zhihong.wang@intel.com>; Liang, Cunming
> > >> <cunming.liang@intel.com>
> > >> Subject: Re: [dpdk-dev] [RFC] Accelerating Data Movement for DPDK
> > >> vHost with DMA Engines
> > >>
> > >> On Fri, Apr 17, 2020 at 12:56 PM Fu, Patrick <patrick.fu@intel.com> wrote:
> > >>>
> > >>> Background
> > >>> ====================================
> > >>> DPDK vhost library implements a user-space VirtIO net backend
> > >>> allowing
> > >> host applications to directly communicate with VirtIO front-end in
> > >> VMs and containers. However, every vhost enqueue/dequeue operation
> > >> requires to copy packet buffers between guest and host memory. The
> > >> overhead of copying large bulk of data makes the vhost backend become
> > >> the I/O bottleneck. DMA engines, including un-core DMA accelerator,
> > >> like Crystal Beach DMA (CBDMA) and Data Streaming Accelerator (DSA),
> > >> and discrete card general purpose DMA, are extremely efficient in
> > >> data movement within system memory. Therefore, we propose a set of
> > >> asynchronous DMA data movement API in vhost library for DMA
> > >> acceleration. With offloading packet copies in vhost data-path from
> > >> the CPU to the DMA engine, which can not only accelerate data transfers,
> > but also save precious CPU core resources.
> > >>>
> > >>> New API Overview
> > >>> ====================================
> > >>> The proposed APIs in the vhost library support various DMA engines
> > >>> to
> > >> accelerate data transfers in the data-path. For the higher
> > >> performance, DMA engines work in an asynchronous manner, where
> > DMA
> > >> data transfers and CPU computations are executed in parallel. The
> > >> proposed API consists of control path API and data path API. The
> > >> control path API includes Registration API and DMA operation
> > >> callback, and the data path API includes asynchronous API. To remove
> > >> the dependency of vendor specific DMA engines, the DMA operation
> > >> callback provides generic DMA data transfer abstractions. To support
> > >> asynchronous DMA data movement, the new async API provides
> > >> asynchronous ring operation semantic in data-path. To enable/disable
> > >> DMA acceleration for virtqueues, users need to use registration API
> > >> is to register/unregister DMA callback implementations to the vhost
> > >> library and bind DMA channels to virtqueues. The DMA channels used by
> > >> virtqueues are provided by DPDK applications, which is backed by virtual
> > or physical DMA devices.
> > >>> The proposed APIs are consisted of 3 sub-sets:
> > >>> 1. DMA Registration APIs
> > >>> 2. DMA Operation Callbacks
> > >>> 3. Async Data APIs
> > >>>
> > >>> DMA Registration APIs
> > >>> ====================================
> > >>> DMA acceleration is per queue basis. DPDK applications need to
> > >>> explicitly
> > >> decide whether a virtqueue needs DMA acceleration and which DMA
> > >> channel to use. In addition, a DMA channel is dedicated to a
> > >> virtqueue and a DMA channel cannot be bound to multiple virtqueues at
> > >> the same time. To enable DMA acceleration for a virtqueue, DPDK
> > >> applications need to implement DMA operation callbacks for a specific
> > >> DMA type (e.g. CBDMA) first, then register the callbacks to the vhost
> > >> library and bind a DMA channel to a virtqueue, and finally use the
> > >> new async API to perform data-path operations on the virtqueue.
> > >>> The definitions of registration API are shown below:
> > >>> int rte_vhost_async_channel_register(int vid, uint16_t queue_id,
> > >>>                                         struct rte_vdma_device_ops
> > >>> *ops);
> > >>>
> > >>> int rte_vhost_async_channel_unregister(int vid, uint16_t queue_id);
> > >>
> > >> We already have multiple DMA implementation over raw dev.
> > >> Why not make a new dmadev class for DMA acceleration and use it by
> > >> virtio and any other clients?
> > >
> > > I believe it doesn't conflict. The purpose of this RFC is to create an async
> > data path in vhost-user and provide a way for applications to work with this
> > new path. dmadev is another topic which could be discussed separately. If
> > we do have the dmadev available in the future, this vhost async data path
> > could certainly be backed by the new dma abstraction without major
> > interface change.
> >
> > Maybe that one advantage of a dmadev class is that it would be easier and
> > more transparent for the application to consume.
> >
> > The application would register some DMA devices, pass them to the Vhost
> > library, and then rte_vhost_submit_enqueue_burst and
> > rte_vhost_poll_enqueue_completed would call the dmadev callbacks directly.
> >
> > Do you think that could work?
> >
> Yes, this is a workable model. As I said in previous reply, I have no objection to make the dmadev. However, what we currently want to do is creating the async data path for vhost, and we actually have no preference to the underlying DMA device model. I believe our current design of the API proto type /data structures are quite common for various DMA acceleration solutions and there is no blocker for any new DMA device to adapt to these APIs or extend to a new one.

IMO, as a driver writer,  we should not be writing TWO DMA driver. One
for vhost and other one for rawdev.
If vhost is the first consumer of DMA needed then I think, it make
sense to add dmadev first.
The rawdev DMA driver to dmadev DMA driver conversion will be the
driver owner job.
I think, it makes sense to define the dmadev API and then costume by
virtio to avoid integration issues.



>
> Thanks,
>
> Patrick
>