Re: [Qemu-devel] [PATCH 4/4] i8254: convert to qdev

[Anthony Liguori]

On 03/06/2011 10:47 AM, Blue Swirl wrote:
> > The PIT is not an ISA device.  Modelling it as such is worse than leaving it
> > unmodelled.
> >      
> No. These days, PIT is part of Super I/O chip, which is accessed via
> LPC bus. LPC is from software point of view equal to ISA. Therefore,
> in absence of LPC, ISA is correct.
>    

In the system we model, the PIT is part of the PIIX3.  The right way to 
model it is as a DeviceState that's no_user=1 and created as part of the 
initialized of PIIX3 (for the PC at least).

LPC is still an expansion bus and it's primarily used for discrete 
components like a TPM.  For components that are all part of a Super I/O 
chip, there really just isn't a bus in the middle.

> > > -PITState *pit_init(int base, qemu_irq irq);
> > > -void pit_set_gate(PITState *pit, int channel, int val);
> > > -int pit_get_gate(PITState *pit, int channel);
> > > -int pit_get_initial_count(PITState *pit, int channel);
> > > -int pit_get_mode(PITState *pit, int channel);
> > > -int pit_get_out(PITState *pit, int channel, int64_t current_time);
> > > +void pit_set_gate(ISADevice *dev, int channel, int val);
> > > +int pit_get_gate(ISADevice *dev, int channel);
> > > +int pit_get_initial_count(ISADevice *dev, int channel);
> > > +int pit_get_mode(ISADevice *dev, int channel);
> > > +int pit_get_out(ISADevice *dev, int channel, int64_t current_time);
> > >
> > >        
> > Making these functions take an ISADevice hurts type safety.  They should
> > take a PITState.
> >      
> These functions should be removed anyway. For example, pcspk probably
> should be merged with i825, or a signal like interface could be used.
>    

It's tough to do that without doing a real simulation.  The PC speaker 
uses one of the PIT channel outputs to drive its tone output.  But we 
don't want to do something like use a qemu_irq with a timer programmed 
to generate the right frequency wave to drive the tone.

So having the PC speaker hold a reference to the PIT and providing a 
high level method interface makes sense to me.

Regards,

Anthony Liguori