[libvirt][RFC PATCH] add a new 'default' option for attribute mode in numatune

[Zhong, Luyao]

On 11/20/2020 6:08 PM, Martin Kletzander wrote:
> On Fri, Nov 20, 2020 at 01:17:29PM +0800, Zhong, Luyao wrote:
>> On 11/18/2020 8:42 PM, Martin Kletzander wrote:
>>> So let's finish this sooner rather than later.  Let's remove the
>>> "migratable/movable" attribute and add another mode for the way memory
>>> allocations are going to be treated.  But let's name it something 
>>> else than
>>> "default" and let's explain the name properly in man pages and
>>> documentation.
>>> Since naming is hard and names I come up with are usually bad I can only
>>> suggest
>>> a lowest bottom fallback if we can't come up with anything else =) Let's
>>> say
>>> something like "restrictive".
>>>
>> I have no better name than yours. So "restrictive" is good I think, I
>> could use it and send out the patch first, then other reviewers might
>> come up with new name otherwise we keep it.
>>
> 
> I actually thought of another way yesterday.  What if not specifying any 
> mode,
> but specifying nodeset means we'll use cpuset.mems, but don't tell 
> QEMU?  That
> sounds to me like something that could make sense for both of us and 
> could be
> easily explainable in the documentation to your users.
> 

"not specifying any mode" will be treated as and formatted to "strict" 
mode in libvirt since the enumeration type is initialed to the first 
value: "strict".[1]

[1]https://github.com/libvirt/libvirt/blob/master/include/libvirt/libvirt-domain.h#L2845

And I need a place(a variable with new value) to hold my config, right? 
Otherwise in libvirt what value can indicates this config? That's one of 
the reasons why I have to introduce a new option for mode.



> Basically just like your patch, but "no mode" means "default mode" and 
> without
> any "migratable" or anything.  We might need to figure out how to deal with
> virsh parameters, but that should be easy to do.
> 
> Thanks for your patience ;)