Re: [PATCH v3 25/36] docs/devel: introduce some key concepts for QOM development

[Alex Bennée]

Philippe Mathieu-Daudé <philmd@linaro.org> writes:

> On 27/6/23 18:09, Alex Bennée wrote:
>> Using QOM correctly is increasingly important to maintaining a modern
>> code base. However the current documentation skips some important
>> concepts before launching into a simple example. Lets:
>>    - at least mention properties
>>    - mention TYPE_OBJECT and TYPE_DEVICE
>>    - talk about why we have realize/unrealize
>>    - mention the QOM tree
>>    - lightly re-arrange the order we mention things
>> Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
>> Message-Id: <20230619171437.357374-6-alex.bennee@linaro.org>
>> ---
>> v3
>>    - moved around as per Paolo's review
>> ---
>>   docs/devel/qom.rst | 58 +++++++++++++++++++++++++++++++++++++++++-----
>>   1 file changed, 52 insertions(+), 6 deletions(-)
>
>
>> +Creating a QOM class
>> +====================
>> +
>> +A simple minimal device implementation may look something like bellow:
>>     .. code-block:: c
>>      :caption: Creating a minimal type
>> @@ -48,6 +66,12 @@ In the above example, we create a simple type that is 
>> described by #TypeInfo.
>>   #TypeInfo describes information about the type including what it inherits
>>   from, the instance and class size, and constructor/destructor hooks.
>>   +The TYPE_DEVICE class is the parent class for all modern devices
>> +implemented in QEMU and adds some specific methods to handle QEMU
>> +device model. This includes managing the lifetime of devices from
>> +creation through to when they become visible to the guest and
>> +eventually unrealized.
>
> Good enough but we are mixing QOM vs QDev...

Yeah, but one relies on the other. From the point of view of someone
coming new to the code I think we do want to mention them both together.
Most people implementing QOM classes will be for devices I think.

Maybe we should enumerate all the types that have TYPE_OBJECT as their
parent?

>
>>   Alternatively several static types could be registered using helper macro
>>   DEFINE_TYPES()
>
>
>> +.. _device-life-cycle:
>> +
>> +Device Life-cycle
>> +=================
>> +
>> +As class initialisation cannot fail devices have an two additional
>> +methods to handle the creation of dynamic devices. The ``realize``
>> +function is called with ``Error **`` pointer which should be set if
>> +the device cannot complete its setup. Otherwise on successful
>> +completion of the ``realize`` method the device object is added to the
>> +QOM tree and made visible to the guest.
>> +
>> +The reverse function is ``unrealize`` and should be were clean-up
>> +code lives to tidy up after the system is done with the device.
>
> Worth mentioning hotplug devices must implement it?
>
>> +All devices can be instantiated by C code, however only some can
>> +created dynamically via the command line or monitor.
>>   +Likewise only some can be unplugged after creation and need an
>> +explicit ``unrealize`` implementation.
>
> Ah, here we go.
>
>> This is determined by the
>> +``user_creatable`` variable in the root ``DeviceClass`` structure.
>> +Devices can only be unplugged if their ``parent_bus`` has a registered
>> +``HotplugHandler``.
>
> TODO on top, mentions the reset() handlers are called after realize(),
> and can be called multiple times.

Where is this TODO? I'm wary of talking about reset too much as the old
reset handler is a deprecated API.

>
>>   API Reference
>> --------------
>> +=============
>>     See the :ref:`QOM API<qom-api>` and :ref:`QDEV API<qdev-api>`
>>   documents for the complete API description.
>
> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>


-- 
Alex Bennée
Virtualisation Tech Lead @ Linaro