Re: [PATCH 12/12] hw/i386/pc: Move PC-machine specific declarations to 'pc_internal.h'

[Philippe Mathieu-Daudé]

On 12/16/19 4:41 PM, Paolo Bonzini wrote:
> On 16/12/19 16:37, Philippe Mathieu-Daudé wrote:
> > On 12/15/19 10:58 AM, Michael S. Tsirkin wrote:
> > > On Fri, Dec 13, 2019 at 05:47:28PM +0100, Philippe Mathieu-Daudé wrote:
> > > > On 12/13/19 5:17 PM, Philippe Mathieu-Daudé wrote:
> > > > > Historically, QEMU started with only one X86 machine: the PC.
> > > > > The 'hw/i386/pc.h' header was used to store all X86 and PC
> > > > > declarations. Since we have now multiple machines based on the
> > > > > X86 architecture, move the PC-specific declarations in a new
> > > > > header.
> > > > > We use 'internal' in the name to explicit this header is restricted
> > > > > to the X86 architecture. Other architecture can not access it.
> > > > >
> > > > > Signed-off-by: Philippe Mathieu-Daudé <address@hidden>
> > > > > ---
> > > > > Maybe name it 'pc_machine.h'?
> > > >
> > > > I forgot to describe here (and in the cover), what's follow after this
> > > > patch.
> > > >
> > > > Patch #13 moves PCMachineClass to
> > > >
> > > > If you ignore PCMachineState, "hw/i386/pc.h" now only contains 76
> > > > lines, and
> > > > it is easier to see what is PC machine specific, what is X86
> > > > specific, and
> > > > what is device generic (not X86 related at all):
> > > >
> > > > - GSI is common to X86 (Paolo sent [3], [6])
> > > > - IOAPIC is common to X86
> > > > - i8259 is multiarch (Paolo [2])
> > > > - PCI_HOST definitions and pc_pci_hole64_start() are X86
> > > > - pc_machine_is_smm_enabled() is X86 (Paolo sent [5])
> > > > - hpet
> > > > - tsc (Paolo sent [3])
> > > > - 3 more functions
> > > >
> > > > So we can move half of this file to "pc_internal.h" and the other to
> > > >
> > > > One problem is the Q35 MCH north bridge which directly sets the PCI
> > > > PCMachineState->bus in q35_host_realize(). This seems a QOM violation
> > > > and is
> > > > probably easily fixable.
> > > >
> > > > Maybe I can apply Paolo's patches instead of this #12, move X86-generic
> > > > declarations to "hw/i386/x86.h", and directly git-move what's left of
> > > > "hw/i386/pc.h" to "pc_internal.h".
> > >
> > > Yea that sounds a bit better.
> >
> > OK, I'll wait for Paolo's next pull get in, then continue based on that,
> > including Paolo's "x86: allow building without PC machine types" series.
> >
> > (Thanks Paolo for picking most of this series!)
>
> FWIW I don't think kvm_i8259_init should be in sysemu/kvm.h, since it's
> x86-specific and that would be something like the same mistake already
> done with hw/i386/pc.h.

Hmm OK.
So to follow your reasoning, 
kvm_pc_gsi_handler/kvm_pc_setup_irq_routing() are x86-specific and could 
be moved out.
I'll figure that out when I rework the last patches.