Re: [Qemu-devel] [PATCH v3 5/7] hmp: Add info commands for preconfig

[Markus Armbruster]

Igor Mammedov <address@hidden> writes:

> On Mon, 11 Jun 2018 15:40:16 -0300
> Eduardo Habkost <address@hidden> wrote:
>
>> On Mon, Jun 11, 2018 at 02:01:52PM +0200, Markus Armbruster wrote:
>> > * Eduardo, why does "info numa" have no QMP equivalent?
>> 
>> Nobody ever asked for one, which seems to qualify as "only for
>> human users".
>> 
>> Should we add an equivalent QMP command even if we don't expect
>> anybody to use it?

The hard requirement for QMP from day one was "provide everything
machine clients need".  To avoid speculation and endless arguments about
what might be needed / not needed, we resolved to approximate this by
"provide everything, except stuff that's *clearly* of no use to
machines".

When you think a command is such an exception, you should explain why in
its commit message.

Note that HMP need not provide the functionality in the exact same
packaging.  For instance, it's fine to have building blocks in QMP, and
just a high-level command in HMP.  However, the latter must be
implemented with the building blocks to make it obvious that QMP
provides the same functionality.

For additional references, see
Message-ID: <address@hidden>
https://lists.nongnu.org/archive/html/qemu-devel/2018-06/msg02176.html

> we inderectly can fetch numa info via QMP, using 
>   query-hotpluggable-cpus
> for CPU mapping and
>   query-memory-devices
> for (NV|PC)-dimm devices, however there is no QMP way for getting
> for numa mapping of initial RAM nor configured numa nodes
> (not counting querying CLI options).

Sounds like most of the building blocks are already there.  The
"obviousness" isn't.

> So perhaps we need info 'numa' equivalent for QMP which would give
> the same amount of information as HMP in one query.

I'd appreciate patches to get us to "QMP has the building blocks, and
HMP is implemented with them".

> Maybe libvirt side as actual users know better if it's really needed (CCed)

Always welcome.