Re: [PATCH v5 3/8] qapi: Restrict balloon-related commands to machine code

[Philippe Mathieu-Daudé]

On 9/14/20 1:21 PM, Markus Armbruster wrote:
> David Hildenbrand <david@redhat.com> writes:
> 
>>> Am 14.09.2020 um 11:42 schrieb Philippe Mathieu-Daudé <philmd@redhat.com>:
>>>
>>> +Laurent and David
>>>
>>>> On 9/14/20 11:16 AM, Markus Armbruster wrote:
>>>> Philippe Mathieu-Daudé <philmd@redhat.com> writes:
>>>>
>>>>> Only qemu-system-FOO and qemu-storage-daemon provide QMP
>>>>> monitors, therefore such declarations and definitions are
>>>>> irrelevant for user-mode emulation.
>>>>>
>>>>> Restricting the balloon-related commands to machine.json
>>>>> allows pulling less declarations/definitions to user-mode.
>>>>
>>>> How this affects user mode is not obvious to (ignorant) me.  Can you
>>>> provide a clue?
>>>
>>> I guess this was discussed with David at some point.
>>>
>>> Maybe the QMP commands are not exposed via HMP, making this
>>> code unreachable?
>>>
>>> Anyhow user-mode binaries don't use the memory ballooning feature,
>>> this is specific to system-mode emulation.
>>>
>>> Laurent/David, do you have some more trivial explanation?
>>
>> Agreed.
>>
>> No memory ballooning device -> no memory ballooning :)
> 
> I understand why user mode doesn't need device models.  What I don't
> understand offhand is how balloon-related stuff in misc.json ends up
> pulling "declarations/definitions to user-mode".  What exactly is being
> pulled where before the series, and no more afterwards?
> 
> Is it just the code generated for the QAPI stuff you move, or is it
> more?

As of this series, this is only QAPI generated code.
(code which pulls in unwanted declarations/definitions
that should be poisoned, but we can't because of this).

I didn't feel the need to enumerate what is removed from
user-mode, because from the diff (and no link failure)
I thought it was obvious. I can do, but that would be
simply copy/pasting the QAPI changes.

Part 3 start to remove things, but I kept that separated.