Re: [PATCH 5/7] hw/ide/piix: Use generic ide_init_ioport()

[Mark Cave-Ayland]

On 10/02/2023 16:34, Bernhard Beschow wrote:

> Fishing out the ISA bus is still a hack IMO, for two reasons: First, IIUC, 
> QOM'ified devices shall only care about its children while looking up one's 
> parent bus violates this rule. Second, using the global machine pointer to scan 
> for the ISA bus just trades one global for another. That's why I'm only doing 
> this for user-created instances. If we deprecated user-created piix IDE devices 
> we could eventually get rid of this hack.

As far as I can tell the solution to QOMified devices finding their parent bus is 
easy: turn DeviceState::parent_bus into a QOM link property called "parent-bus" or 
similar which accepts TYPE_BUS, and then any object of TYPE_DEVICE can locate its 
parent bus using object_property_get_link() with a standardised property name.

I think it may be impossible to completely remove parent bus references, since buses 
like PCI currently make use of upwards hierarchy navigation for things like IRQ 
mapping and PCI-PCI bridge traversal.


ATB,

Mark.