Re: [Qemu-devel] Should the i8259 devices remain no-user?

[Markus Armbruster]

Paolo Bonzini <address@hidden> writes:

> Il 15/10/2013 14:43, Markus Armbruster ha scritto:
>> Paolo, or maybe Andreas,
>> 
>> To go beyond RFC with this series, I need to explain why isa-i8259 and
>> kvm-i8259 cannot_instantiate_with_device_add_yet, or drop that.  I'd
>> appreciate your help.
>> 
>> Both are derived from TYPE_PIC_COMMON, which is derived from
>> TYPE_ISA_DEVICE.
>> 
>> I figure isa-i8259 cannot_instantiate_with_device_add_yet, because it
>> sets global isa_pic and slave_pic.  slave_pic appears to be a lame way
>> to wire the slave PIC to the master PIC behind QOM's back.  isa_pic
>> appears to be a lame way to wire the master PIC to whatever it needs to
>> be wired to.  Is that a fair description?
>
> Yes.
>
>> If yes, is it sufficient reason for
>> cannot_instantiate_with_device_add_yet?
>> 
>> kvm-i8259 is the same device implemented with kernel support.  Does it
>> have its own reason for cannot_instantiate_with_device_add_yet?
>> 
>> If not, should it keep cannot_instantiate_with_device_add_yet for
>> symmetry with isa-i8259?
>
> Both i8259 implementations have to be matched with an appropriate array
> of qemu_irqs, such as the one returned by kvm_i8259_init.  I think this
> is the reason why kvm-i8259 cannot be instantiated from the command-line.

Let me try to elaborate, to make sure I understand.

Unlike ordinary ISA devices, the i8259 devices need additional wiring,
done by code.

For instance, board code like pc_q35_init(), pc_piix.c's pc_init1(),
mips_malta_init(), ... wire up their IRQ input lines.  The slave's IRQ
output line is wired to the master's IRQ2 in hw/intc/i8259.c for
isa-i8259, and the kernel for kvm-i8259.  The master's IRQ output line
is wired up by board code (it's complicated).

Correct?  If yes, I can turn it into a suitable comment.