[libvirt] [Qemu-devel] [RFC] qmp: query-device-slots command

[Laine Stump]

On 12/14/2016 12:17 PM, Marcel Apfelbaum wrote:
> On 12/13/2016 06:00 PM, Eduardo Habkost wrote:
>> On Tue, Dec 13, 2016 at 04:15:18PM +0200, Marcel Apfelbaum wrote:
>>> On 12/13/2016 02:42 PM, Eduardo Habkost wrote:
>>>> On Tue, Dec 13, 2016 at 12:04:17PM +0100, Markus Armbruster wrote:
>>>>> Quick interface review only:
>>>>>
>>>>> Eduardo Habkost <ehabkost at redhat.com> writes:
>>>>>
>>>>>> This adds a new command to QMP: query-device-slots. It will allow
>>>>>> management software to query possible slots where devices can be
>>>>>> plugged.
>>>>>>
>>>>>> This implementation of the command will return:
>>>>>>
>>>>>> * Multiple PCI slots per bus, in the case of PCI buses;
>>>>>> * One slot per bus in the case of the other buses;
>>>>>
>>>>> Umm, that doesn't seem right.  For instance, a SCSI bus has multiple
>>>>> slots.  The slot address is the SCSI ID.  An IDE bus may have one
>>>>> (SATA)
>>>>> or two (PATA).  For more examples, see qdev-device-use.txt section
>>>>> "Specifying Bus and Address on Bus".
>>>>
>>>> Yes, I should have clarified that: this version changes only PCI
>>>> to expose multiple slots, but other buses also need be changed to
>>>> implement BusClass::enumerate_slots() properly, too.
>>>>
>>>>>
>>>>>> * One slot for each entry from query-hotpluggable-cpus.
>>>>>> In the example below, I am not sure if the PCIe ports are all
>>>>>> supposed to report all slots, but I didn't find any existing
>>>>>> field in PCIBus that would help me figure out the actual number
>>>>>> of slots in a given PCI bus.
>>>>>>
>>>>>> Git tree
>>>>>> --------
>>>>>>
>>>>>> This patch needs the previous query-machines series I am working
>>>>>> on. The full tree can be found on the git tree at:
>>>>>>
>>>>>>   git://github.com/ehabkost/qemu-hacks.git
>>>>>> work/query-machines-bus-info
>>>>>>
>>>>>> Example output
>>>>>> --------------
>>>>>>
>>>>>> The following output was returned by QEMU when running it as:
>>>>>>
>>>>>>  $ qemu-system-x86_64 -machine q35 \
>>>>>>    -readconfig docs/q35-chipset.cfg \
>>>>>>    -smp 4,maxcpus=8,sockets=2,cores=2,threads=2
>>>>>>
>>>>>>  {
>>>>>>     "return": [
>>>>>
>>>>> Are you sure >3000 lines of example output make sense here?
>>>>
>>>> I'm not. :)
>>>>
>>>> That's why I need feedback from the PCI experts. I believe most
>>>> of the PCI ports on q35 accept only one device,

I don't think of it as "one device" but as "only slot 0 can be used" 
(since, as Marcel pointed out, a different device can be placed on each 
of the 8 functions of that slot).

> but I see no code
>>>> implementing that restriction, and no obvious PCIBus or
>>>> PCIBusClass field indicating that.
>>>
>>> Hi,
>>>
>>> Sadly there is no such restriction today as far as I know.
>>> We trust the user that he knows what he is doing :)
>>> and we let libvirt implement the restriction.
>>>
>>> I do agree we should fix that.
>>
>> OK, this matches what I saw on the current code.
>>
>> But I would like to understand what exactly should be the rules
>> if we fix that. Should we report only 1 slot if
>> pci_bus_has_pcie_upstream_port(bus)?
>>
>
> We should report 1 slot for a bus exposed by a PCIe Root Port
> or PCIe Downstream Port.
>
> Be aware that 1 slot doesn't necessarily refer to only one device,
> we can have a multi-function device.

Is that a fixed thing? I mean, is it always possible to use all 8 
functions of a slot (assuming no hotplug of course)?

>
> Thanks,
> Marcel
>
>> (Note that pci_bus_has_pcie_upstream_port() doesn't exist in the
>> current code, it is added by my patch)
>>
>