[ovs-dev] [PATCH 0/2] Deprecate the Linux OOT driver

Gregory Rose gvrose8192 at gmail.com
Wed Jan 6 19:09:29 UTC 2021



On 1/6/2021 10:55 AM, Ilya Maximets wrote:
> On 1/6/21 7:09 PM, Gregory Rose wrote:
>>
>>
>> On 1/6/2021 4:34 AM, Ilya Maximets wrote:
>>> On 1/5/21 7:33 PM, Greg Rose wrote:
>>>> As agreed in (1) deprecate the Linux OOT driver.
>>>>
>>>> github Build and Test here:
>>>> https://github.com/gvrose8192/ovs-experimental/actions/runs/463987690
>>>>
>>>> 1. https://mail.openvswitch.org/pipermail/ovs-dev/2020-December/378831.html
>>>>
>>>> Greg Rose (2):
>>>>     doc: Deprecate the Linux Out of Tree drivers
>>>>     acinclude: Warn when --with-linux parameter is supplied
>>>>
>>>>    Documentation/faq/releases.rst | 7 ++++++-
>>>>    NEWS                           | 3 +++
>>>>    acinclude.m4                   | 1 +
>>>>    3 files changed, 10 insertions(+), 1 deletion(-)
>>>>
>>>
>>> Thanks for the patches!
>>>
>>> Few general comments:
>>>
>>> 1. Word 'driver' sounds weird to me.  'kennel module' is more commonly
>>>      used term, I think.  We're using term 'driver' for windows datapath,
>>>      but it seems like windows-specific thing.  In Linux world 'driver' is
>>>      usually something that talks directly to hardware and that is not
>>>      the case for openvswitch.ko and other parts.
>>>      I'd say that we need to do s/kernel driver/kernel module/ in this
>>>      patch set.
>>
>> Sure, not a problem.
>>
>>>
>>> 2. We need to specify the date of removal in NEWS and docs.  I'd say
>>>      that we could state that OOT kernel module will be removed in 2.16.
>>>      BTW, from the development point of view it might be good to remove
>>>      it as soon as 2.15 branched/released.
>>
>> Do we actually want to remove it or just leave it deprecated but still there?
>>
>> And when we say remove it does that mean just disable the '--with-linux'
>> configure option or would we be removing all the code as well?  I just
>> want to make sure we define removal the same way.
> 
> My understanding is to completely remove all the related code and documentation.
> This includes 'datapath' directory and configuration stuff from m4 files.

OK, good to be on the same page.

> 
> Users will be able to build kernel module from the 2.15 branch.  This way
> we will not need to maintain duplicate of the code on newer branches.
> 
> One problem here is that OVS 2.15 will reach EOL relatively soon, and projected
> EOL for kernel 5.4 is Dec 2025.  We might actually postpone complete removal
> until 2.18.  This way we will have OVS 2.17 LTS with kernel module included.
> And it will be supported until Feb 2025.  And we might actually increase
> support time on branch-2.17 just for kernel module until kernel 5.4 reaches EOL.
> After that we can safely remove OOT module in OVS 2.18, because there will
> be no supported upstream kernel at this point that OOT module supports.
> 
> Thoughts?

I'm fine with this plan of action.  I will develop and maintain a side
branch with the final removal of the Linux kernel datapath so that it
will be ready to go when needed.  That way I can test it early and be
on the lookout for unwanted side effects.

> 
> If that's OK, we can just state in NEWS and documentation that OOT kernel
> module will be removed in OVS 2.18 and keep it until then.

I will  make it so.  I'll send out a V2 patch set.

Thanks,

- Greg


More information about the dev mailing list