Re: [Qemu-devel] [PATCH 1/3] arm: gic: add GICType

[Peter Xu]

On Tue, Mar 01, 2016 at 03:20:59PM +0100, Andrea Bolognani wrote:
> On Tue, 2016-02-23 at 18:52 +0800, Peter Xu wrote:
> > +{ 'enum': 'GICType', 'data': [ 'gicv2', 'gicv3', 'gicv2-kvm', 'gicv3-kvm' 
> > ] }
> 
> Wouldn't this conflate the use of accel= and kernel_irqchip= options?

AFAIU, it's not a problem. Let me paste some lines from the original
RFC thread which explains the definition of the entries:

- gicv2:      GIC version 2 without kernel IRQ chip
- gicv2-kvm:  GIC version 2 with kernel IRQ chip
- gicv3:      GIC version 3 without kernel IRQ chip (not supported)
- gicv3-kvm:  GIC version 3 with kernel IRQ chip

(from https://lists.gnu.org/archive/html/qemu-devel/2016-02/msg02882.html)

So... what I understand is that, we are not talking about "accel="
at all. Instead, we are talking about "kernel_irqchip=" only. Or
say, all these GIC version information we provide from QEMU does not
tell whether KVM is supported or not (for "accel=", it is provided
by another QMP message named "query-kvm"). We are only talking about
which kind of GIC we support. In our case, for each version, it
could be supported either in userspace, or in kernel.

> 
> IIUC, depending on the hardware, you might find yourself in the
> following situation:
> 
>   accel=tcg,gic-version=3                     unavailable
>   accel=kvm,kernel_irqchip=off,gic-version=3  unavailable

As explained above, IIUC, both of these two "unavailable" ones
correspond to "gicv3" entry of the results.

>   accel=kvm,gic-version=3                     available

And this one corresponds to "gicv3-kvm" entry.

> 
> so I'd expect the output to be something like
> 
>   [ "v2": { "tcg": true,
>             "kvm-without-kernel-irqchip": true,
>             "kvm": true },
>     "v3": { "tcg": false,
>             "kvm-without-kernel-irqchip": false,
>             "kvm": true } ]

Actually, this reminded me about the "kernel_irqchip=split" case. Do
we need to consider that one? AFAIK, splitted irqchip is only used
for x86 currently. Whether ARM will possibly support splitted kernel
irqchip one day just like x86? If so, I would prefer to change the
query result layout from array to dict, like:

[ "v2": { "emulated": true,
          "split": false,
          "kernel": true },
  "v3": { "emulated": false,
          "split": false,
          "kernel": true } ]

Since the matrix is big enough (2x3) to consider drop the array
format (I'd admit maybe dict is always the best one...).

Peter

> 
> Since libvirt currently doesn't have support for the
> kernel_irqchip= option, it would only take the "tcg" and "kvm"
> values into account; on the other hand, if at some point
> libvirt will grow support for that option it would be able to
> retrieve all the required information.
> 
> Cheers.
> 
> -- 
> Andrea Bolognani
> Software Engineer - Virtualization Team