Hi,
>> -static int usb_ehci_initfn(PCIDevice *dev)
>> +static void usb_ehci_initfn(EHCIState *s, DeviceState *dev)
>> {
>> - EHCIState *s = DO_UPCAST(EHCIState, dev, dev);
>> +static int usb_ehci_pci_initfn(PCIDevice *dev)
>> +{
>> + EHCIPCIState *i = DO_UPCAST(EHCIPCIState, pcidev, dev);
>
> Same as discussed for Peter's patchset, this should be using a QOM cast
> macro and may need an abstract base type if there is no unique type
> matching EHCIPCIState struct.
--verbose please.
This is exactly like it used to be. PCIDeviceClass->init() gets passed
in a PCIDevice pointer and uses DO_UPCAST to get the container struct
carrying the ehci state (EHCIState before the patch, EHCIPCIState now).
This is common practice all over the tree.
> Should I send you a follow-up to squash if this is the approach we are
> going to take?
If there is a new, more QOM-ish way to do the same feel free to send
patches. I see that as independent cleanup though, not as something
which should be squashed into this patch.
cheers,
Gerd
