mbox series

[v1,0/2] introduce asynchronous data path for vhost

Message ID 1591869725-13331-1-git-send-email-patrick.fu@intel.com (mailing list archive)
Headers show
Series introduce asynchronous data path for vhost | expand

Message

Patrick Fu June 11, 2020, 10:02 a.m. UTC
From: Patrick Fu <patrick.fu@intel.com>

Performing large memory copies usually takes up a major part of CPU
cycles and becomes the hot spot in vhost-user enqueue operation. To
offload expensive memory operations from the CPU, this patch set
proposes to leverage DMA engines, e.g., I/OAT, a DMA engine in the
Intel's processor, to accelerate large copies.

Large copies are offloaded from the CPU to the DMA in an asynchronous
manner. The CPU just submits copy jobs to the DMA but without waiting
for its copy completion. Thus, there is no CPU intervention during
data transfer; we can save precious CPU cycles and improve the overall
throughput for vhost-user based applications, like OVS. During packet
transmission, it offloads large copies to the DMA and performs small
copies by the CPU, due to startup overheads associated with the DMA.

This patch set construct a general framework that applications can
leverage to attach DMA channels with vhost-user transmit queues. Four
new RTE APIs are introduced to vhost library for applications to
register and use the asynchronous data path. In addition, two new DMA
operation callbacks are defined, by which vhost-user asynchronous data
path can interact with DMA hardware. Currently only enqueue operation
for split queue is implemented, but the frame is flexible to extend
support for dequeue & packed queue.

Patrick Fu (2):
  vhost: introduce async data path registration API
  vhost: introduce async enqueue for split ring

 lib/librte_vhost/Makefile          |   3 +-
 lib/librte_vhost/rte_vhost.h       |   1 +
 lib/librte_vhost/rte_vhost_async.h | 172 ++++++++++++
 lib/librte_vhost/socket.c          |  20 ++
 lib/librte_vhost/vhost.c           |  74 ++++-
 lib/librte_vhost/vhost.h           |  30 ++-
 lib/librte_vhost/vhost_user.c      |  28 +-
 lib/librte_vhost/virtio_net.c      | 538 ++++++++++++++++++++++++++++++++++++-
 8 files changed, 857 insertions(+), 9 deletions(-)
 create mode 100644 lib/librte_vhost/rte_vhost_async.h

Comments

Maxime Coquelin June 26, 2020, 2:42 p.m. UTC | #1
Hi Patrick,

On 6/11/20 12:02 PM, patrick.fu@intel.com wrote:
> From: Patrick Fu <patrick.fu@intel.com>
> 
> Performing large memory copies usually takes up a major part of CPU
> cycles and becomes the hot spot in vhost-user enqueue operation. To
> offload expensive memory operations from the CPU, this patch set
> proposes to leverage DMA engines, e.g., I/OAT, a DMA engine in the
> Intel's processor, to accelerate large copies.
> 
> Large copies are offloaded from the CPU to the DMA in an asynchronous
> manner. The CPU just submits copy jobs to the DMA but without waiting
> for its copy completion. Thus, there is no CPU intervention during
> data transfer; we can save precious CPU cycles and improve the overall
> throughput for vhost-user based applications, like OVS. During packet
> transmission, it offloads large copies to the DMA and performs small
> copies by the CPU, due to startup overheads associated with the DMA.
> 
> This patch set construct a general framework that applications can
> leverage to attach DMA channels with vhost-user transmit queues. Four
> new RTE APIs are introduced to vhost library for applications to
> register and use the asynchronous data path. In addition, two new DMA
> operation callbacks are defined, by which vhost-user asynchronous data
> path can interact with DMA hardware. Currently only enqueue operation
> for split queue is implemented, but the frame is flexible to extend
> support for dequeue & packed queue.


Thanks for this big rework of the Vhost DMA series.
It looks overall good to me and is consistent with what design you
suggested few months back.

I don't see a big risk to integrate it in v20.08 once the few comments
are taken into account.

I'll try to make another pass before you send a v2.

Maxime

> Patrick Fu (2):
>   vhost: introduce async data path registration API
>   vhost: introduce async enqueue for split ring
> 
>  lib/librte_vhost/Makefile          |   3 +-
>  lib/librte_vhost/rte_vhost.h       |   1 +
>  lib/librte_vhost/rte_vhost_async.h | 172 ++++++++++++
>  lib/librte_vhost/socket.c          |  20 ++
>  lib/librte_vhost/vhost.c           |  74 ++++-
>  lib/librte_vhost/vhost.h           |  30 ++-
>  lib/librte_vhost/vhost_user.c      |  28 +-
>  lib/librte_vhost/virtio_net.c      | 538 ++++++++++++++++++++++++++++++++++++-
>  8 files changed, 857 insertions(+), 9 deletions(-)
>  create mode 100644 lib/librte_vhost/rte_vhost_async.h
>