Philippe Mathieu-Daudé <phi...@redhat.com> writes: > On 3/16/20 7:23 AM, Markus Armbruster wrote: >> Paolo Bonzini <pbonz...@redhat.com> writes: >> >>> On 13/03/20 23:16, BALATON Zoltan wrote: >>>>> >>>>> + pci_dev = pci_create_simple(pci_bus, -1, "cmd646-ide"); >>>>> + pci_ide_create_devs(pci_dev); >>>> >>>> Additionally, I think it may also make sense to move pci_ide_create_devs >>>> call into the realize methods of these IDE controllers so boards do not >>>> need to do it explicitely. These calls always follow the creation of the >>>> device immediately so could just be done internally in IDE device and >>>> simplify it further. I can attempt to prepare additional patches for >>>> that but first I'd like to hear if anyone has anything against that to >>>> avoid doing useless work. >>> >>> No, it's better to do it separately. I think that otherwise you could >>> add another IDE controller with -device, and both controllers would try >>> to add the drives. >> >> Correct. >> >> Creating device frontends for -drive if=ide is the board's job. Boards >> may delegate to suitable helpers. I'd very much prefer these helpers >> not to live with device model code. Board and device model code should >> be cleanly separated to to reduce the temptation to muddle their >> responsibilities. It's separation of concerns. >> >> I actually wish we had separate sub-trees for boards and devices instead >> of keeping both in hw/. > > Never too late! > > To be clear, you suggest: > > - one dir with machines, boards, system-on-module > - one dir with devices, cpu, system-on-chips > > Correct?
In QOM terms: * One sub-tree with descendants of TYPE_DEVICE * One sub-tree with descendants of TYPE_MACHINE
