Re: [Qemu-devel] [PATCH v2] New cpu-max field in query-machines QMP command output

[Michal Novotny]

On 04/09/2013 03:42 PM, Luiz Capitulino wrote:
> On Tue, 09 Apr 2013 15:32:57 +0200
> Michal Novotny <address@hidden> wrote:
>
>> On 04/09/2013 03:15 PM, Luiz Capitulino wrote:
>>> On Tue, 09 Apr 2013 15:09:21 +0200
>>> Michal Novotny <address@hidden> wrote:
>>>
>>>> On 04/09/2013 03:06 PM, Luiz Capitulino wrote:
>>>>> On Mon, 08 Apr 2013 11:14:32 -0600
>>>>> Eric Blake <address@hidden> wrote:
>>>>>
>>>>>> On 04/08/2013 10:41 AM, Michal Novotny wrote:
>>>>>>> Alter the query-machines QMP command to output information about
>>>>>>> maximum number of CPUs for each machine type with default value
>>>>>>> set to 1 in case the number of max_cpus is not set.
>>>>>>>
>>>>>>> Signed-off-by: Michal Novotny <address@hidden>
>>>>>>> ---
>>>>>>>  qapi-schema.json | 4 +++-
>>>>>>>  vl.c             | 1 +
>>>>>>>  2 files changed, 4 insertions(+), 1 deletion(-)
>>>>>>>
>>>>>>> diff --git a/qapi-schema.json b/qapi-schema.json
>>>>>>> index db542f6..689ca8d 100644
>>>>>>> --- a/qapi-schema.json
>>>>>>> +++ b/qapi-schema.json
>>>>>>> @@ -2861,11 +2861,13 @@
>>>>>>>  #
>>>>>>>  # @default: #optional whether the machine is default
>>>>>>>  #
>>>>>>> +# @cpu-max: maximum number of CPUs supported by the machine type
>>>>>> Typically, when adding a field in a later version than the original
>>>>>> introduction of the datatype, we add '(since 1.5)' to make it obvious
>>>>>> when to expect the field.  However, as nothing (currently) enforces this
>>>>>> rule, I think such an addition is minor enough that it wouldn't
>>>>>> invalidate the use of my:
>>>>> Oh, it turns out I was making some confusion with this patch and
>>>>> didn't realize it was extending the query-machines command.
>>>>>
>>>>> I don't mean there's anything wrong with it, but my question is: doesn't
>>>>> this patch invalidates query-cpu-max?
>>>> Not really, the query-cpu-max shows maximum number of CPUs supported per
>>>> currently running machine (machine type) however the extension to
>>>> query-machines shows the maximum number of CPUs supported per each of
>>>> the displayed machine types. This usually doesn't matter on x86
>>>> architectures however it does matter on ARM, for example, as ARM has
>>>> some CPU models supporting up to 4 CPUs (highbank model) whereas most of
>>>> the ARM processors does have only one CPU.
>>> Can't we eliminate the query-cpu-max command if we extend the
>>> query-machines command to show which machine type is in use?
>> It may be useful as query-cpu-max is showing maximum number of CPUs
>> supported per machine type QEMU currently emulates. We don't have to
>> know the machine type and look for this number manually in the
>> query-machines output if we have query-cpu-max.
> QMP is a machine protocol, "manual" in that context is not a problem.
> It's actually wanted.


By manually in this context I mean if you have already running machine
and you know the machine type then you have to run query-machines and
look for your CPU type if you want know maximum number of CPUs supported
by this machine type.


>
>> We would have to both
>> know the machine type and look for the number manually to get max-cpu
>> count for currently running machine if we get rid of query-cpu-max command.
>>
>> Whether something will use the value of CPU-max for currently running
>> machine (query-cpu-max output) is another question.
> What you mean by that? Why are we adding a command if we don't even know
> if it's going to be used?


The original reason was to allow libvirt to know max. number of CPUs
however libvirt queries the machines and puts the aliases and CPU models
into capabilities which is basically reason why we could drop
query-cpu-max command as new implementation was necessary (the one
extending MachineInfo by cpu-max).


>
> If I remember correctly, it was said that libivrt was asking for this
> command. If this is not the case let's drop it, because the command is
> bad (ie. this should be done through QOM).

Originally yes however the implementation of extending MachineInfo seems
to be better for them so they don't need query-cpu-max anymore.


>
> I've applied this patch, but again, if query-cpu-max is not required by
> any QMP user, please post patches reverting it.

Ok, I will create a revert patch and I'll send it.

Michal

-- 
Michal Novotny <address@hidden>, RHCE, Red Hat
Virtualization | libvirt-php bindings | php-virt-control.org