Re: [Qemu-devel] [PATCH v2 04/17] pc: route all memory devices through the machine hotplug handler

[David Hildenbrand <david@redhat.com>]

On 12.05.2018 16:47, Paolo Bonzini wrote:
> On 11/05/2018 15:19, David Hildenbrand wrote:
>> +    if (dev->parent_bus) {
>> +        if (object_dynamic_cast(OBJECT(dev), TYPE_MEMORY_DEVICE)) {
>> +            return HOTPLUG_HANDLER(machine);
>> +        }
>> +    }
>> +
> 
> How do you get here with a MemoryDevice that has !dev->parent_bus?
> 

Excellent question :)

This is for now (for pc and spapr) a theoretical case, but I
included it to make all hotplug handler look alike and also show for
other potential device (interfaces) how it should be handled.

I'll give you the s390x example I had in mind:

s390x cannot hotplug dimms. dimms, however are busless devices that
implement the MemoryDevice interface.

If we would simply always indicate this way that we have a hotplug
handler, e.g. the check in qdev_device_add() would not trigger:

...
if (bus) {
    qdev_set_parent_bus(dev, bus);
} else if (qdev_hotplug && !qdev_get_machine_hotplug_handler(dev)) {
    /* No bus, no machine hotplug handler --> device is not hotpluggable */
    error_setg(&err, "Device '%s' can not be hotplugged on this machine",
               driver);
    goto err_del_dev;
}
...

So the rational is "if its a busless device and I (the machine)
am not able to fully plug it, I must also not partially plug it."

However, right now I am not sure (due to qdev_hotplug) if this
is enough.


> Thanks,
> 
> Paolo
> 


-- 

Thanks,

David / dhildenb