Re: [dpdk-dev] [PATCH] doc: note KNI alternatives and deprecation plan

[Ferruh Yigit <ferruh.yigit@intel.com>]

On 8/20/2021 1:58 PM, David Marchand wrote:
> On Wed, Jun 23, 2021 at 7:32 PM Ferruh Yigit <ferruh.yigit@intel.com> wrote:
>>
>> Add a note that KNI kernel module will be moved to dpdk-kmods git repo
>> and there is a long term plan to deprecate it.
>>
>> Also add some more details on the alternatives to KNI and cons of the
>> KNI against these alternatives.
> 
> Some suggestions on wording below, feel free to ignore if this sounds
> like French ;-).
> 
> 
>>
>> Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
>> ---
>>   doc/guides/nics/tap.rst                       |  2 +
>>   .../prog_guide/kernel_nic_interface.rst       | 38 +++++++++++++++++--
>>   2 files changed, 37 insertions(+), 3 deletions(-)
>>
>> diff --git a/doc/guides/nics/tap.rst b/doc/guides/nics/tap.rst
>> index 3ce696b605d1..07315fe32422 100644
>> --- a/doc/guides/nics/tap.rst
>> +++ b/doc/guides/nics/tap.rst
>> @@ -1,6 +1,8 @@
>>   ..  SPDX-License-Identifier: BSD-3-Clause
>>       Copyright(c) 2016 Intel Corporation.
>>
>> +.. _TunTap_PMD:
>> +
>>   Tun|Tap Poll Mode Driver
>>   ========================
>>
>> diff --git a/doc/guides/prog_guide/kernel_nic_interface.rst b/doc/guides/prog_guide/kernel_nic_interface.rst
>> index 1ce03ec1a374..29f8c92fd9d6 100644
>> --- a/doc/guides/prog_guide/kernel_nic_interface.rst
>> +++ b/doc/guides/prog_guide/kernel_nic_interface.rst
>> @@ -6,16 +6,48 @@
>>   Kernel NIC Interface
>>   ====================
>>
>> +.. Note::
>> +
> 
>> +   KNI kernel module will be removed from main git repository to `dpdk-kmods <https://git.dpdk.org/dpdk-kmods/>`_
> The* KNI kernel module will be moved* (Olivier comment) to ...
> 
>> +   repository by the `DPDK technical board decision <https://mails.dpdk.org/archives/dev/2021-January/197077.html>`_.
> repository following* the `...`_.
> 
> I'd move this first part in doc/guides/rel_notes/deprecation.rst

Perhaps it is better to clarify the decisions first in the deprecation
notice, and document it later referencing it, instead of documenting it first.

I will try to separate the decision part from the KNI documentation.

> and/or release notes.
> Then, reword this as:
> 
>> +   Also there is a `long term plan <https://mails.dpdk.org/archives/dev/2021-May/209026.html>`_ to deprecate the KNI.
> 
> The `long term plan...`_ is* to deprecate KNI, see <XXX link to
> deprecation notice/release notes>.
> 
> 
>> +
>> +   :ref:`virtio_user_as_exceptional_path` alternative is preferred way for
> 
> is one/a* preferred way
> 

ack

>> +   interfacing with Linux network stack as it is being in-kernel solution and
> 
> interfacing with the* Linux network stack as it is a* in-kernel and
>

ack
  
>> +   similar performance expectations.
> 
> has* similar performance expectations.
> 

ack

> 
>> +
>>   The DPDK Kernel NIC Interface (KNI) allows userspace applications access to the Linux* control plane.
>>
>> -The benefits of using the DPDK KNI are:
>> +KNI allows an interface with the kernel network stack and allows management of
> 
> KNI provides* an interface ...
> 

ack

>> +DPDK ports using standard Linux net tools such as ``ethtool``, ``ifconfig`` and
>> +``tcpdump``.
>> +
>> +Main use case of KNI is get/receive exception packets from/to Linux network
> 
> The* main use case of KNI is to* ...
> 

ack

>> +stack while main datapath IO is done bypassing the networking stack.
>> +
>> +There are other alternatives to KNI, all are available in the upstream Linux:
>> +
>> +#. :ref:`TunTap_PMD` as wrapper to `Linux tun/tap
>> +   <https://www.kernel.org/doc/Documentation/networking/tuntap.txt>`_
>> +
>> +#. :ref:`virtio_user_as_exceptional_path`
>> +
> 
> +1 to Olivier comment, for making virtio-user first.
> 
> 
>> +The benefits of using the DPDK KNI against alternatives are:
>>
>>   *   Faster than existing Linux TUN/TAP interfaces
>>       (by eliminating system calls and copy_to_user()/copy_from_user() operations.
>>
>> -*   Allows management of DPDK ports using standard Linux net tools such as ethtool, ifconfig and tcpdump.
>> +The cons of the DPDK KNI are:
> 
> Not sure why we need an extra DPDK, KNI alone is enough.
> 

ack

> 
>> +
>> +* It is out-of-tree Linux kernel module and it can't be distributed as binary as
>> +  part of OSV DPDK packages. This makes it harder to consume, although it is
> 
> OSV as in OS vendors, right? ok for me.
> 
> 
>> +  always possible to compile it from the source code.
>> +
>> +* As it shares memory between userspace and kernelspace, and kernel part
>> +  directly uses input provided by userspace, it is not safe. This makes hard to
>> +  upstream the module.
>>
>> -*   Allows an interface with the kernel network stack.
>> +* Only a subset of control commands are supported by KNI.
> 
> of net devices control commands*
> 

ack

>>
>>   The components of an application using the DPDK Kernel NIC Interface are shown in :numref:`figure_kernel_nic_intf`.
>>
>> --
>> 2.31.1
>>
> 
>