Re: Reset problem vs. MMIO emulation, hypercalls, etc...

[Avi Kivity <avi@redhat.com>]

On 08/08/2012 12:09 AM, Benjamin Herrenschmidt wrote:
> On Tue, 2012-08-07 at 16:13 +0300, Avi Kivity wrote:
>> 
> 
>> Peter has started to fix up this naming mess in qemu.  I guess we should
>> do the same for the kernel (except for ABIs) and document it, because it
>> keeps generating confusion. 
> 
> Ok so our current situation is that the XICS and MPIC are emulated in
> userspace entirely but the link between them and the VCPU is the
> asynchronous EE line so we are fine.
> 
> I'm currently working on moving the XICS into the kernel for performance
> reasons, however, just like ARM VGIC, I can't seem to find a way to
> "fit" in the "generic" irqchip code in there. It's just not generic at
> all and quite x86 centric :-)

The generic code is for the two apic architectures: x64 and ia64.  Don't
try to shoehorn it in, you'll damage both the shoe and your foot.

> 
> So for now I'm just doing my own version of CREATE_IRQCHIP to create it
> and KVM_INTERRUPT to trigger the various interrupts. None of the mapping
> stuff (which we really don't need).

You mean KVM_IRQ_LINE.  KVM_INTERRUPT is a synchronous vcpu ioctl.

> 
> That's a bit of a problem vs. some of the code qemu-side such as in
> virtio-pci which does seem to be written around the model exposed by the
> x86 stuff and relies on doing such mappings so I think we'll have to
> butcher some of that.

Can you elaborate? virtio-pci is pci-centric, there should be nothing
x86 specific there.


-- 
error compiling committee.c: too many arguments to function