Re: [Qemu-devel] [PATCH v5 2/2] enable multi-function hot-add

[Cao jin <caoj.fnst@cn.fujitsu.com>]

Hi Michael
     Thanks for your quick response

On 10/27/2015 05:47 PM, Michael S. Tsirkin wrote:
> On Tue, Oct 27, 2015 at 05:39:01PM +0800, Cao jin wrote:
>>>>> static
>>>>> bool pcie_has_upstream_port(PCIDevice *dev)
>>>>> {
>>>>>      PCIDevice *parent_dev = pci_bridge_get_device(pci_dev->bus);
>>>>>
>>>>>      /*
>>>>>       * Device associated with an upstream port.
>>>>>       * As there are several types of these, it's easier to check the
>>>>>       * parent device: upstream ports are always connected to
>>>>>       * root or downstream ports.
>>>>>       */
>>>>>      return parent_dev &&
>>>>> 	    pci_is_express(parent_dev) &&
>>>>> 		parent_dev->exp.exp_cap &&
>>>>> 		(pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
>>>>> 		pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
>>>>> }
>>>>>
>>>>
>>>> Assume my understanding is right, which means both ARI and non-ARI device
>>>> have the upstream port(root port or downstream port), could the existence of
>>>> upstream port be the judgment condition?
>>>
>>> This tells us whether we are behind a port that
>>> can address devices in slot != 0.
>>>
>>>
>>
>> Seems I find something, according to spec: Endpoints are classified as
>> either legacy, PCI Express, or Root Complex Integrated Endpoints. I think
>> this is also what you means in comment "As there are several types of these"
>> And "we are behind a port" means the device is not a Root Complex
>> Integrated, only the other two types can be behind a port.
>>
>> Am I right about this?
>
> No, device can be a bridge itself.

Ah yes, you are saying a PCIe-->PCI(PCI-X) bridge, right?

> It's simple: root and downstream
> ports only have 1 slot (and it's a physical one) so you know there's
> always 1 device there. It could have multiple functions.
>
Yes, I am aware.
It seems there is no big and obvious error about other understandings in 
last mail:) I just want to get myself totally understood about ARI and 
your implementation before using your code. Thank you very much for your 
time and patience:)

And the v6 is on the way

-- 
Yours Sincerely,

Cao Jin