Re: [Qemu-devel] [PATCH v3 14/16] target-s390x: Extend QMP command query-cpu-definitions

[Michael Mueller]

On Mon, 2 Mar 2015 16:11:00 -0300
Eduardo Habkost <address@hidden> wrote:

> On Mon, Mar 02, 2015 at 01:44:06PM +0100, Michael Mueller wrote:
> > This patch implements the QMP command 'query-cpu-definitions' in the S390
> > context. The command returns a list of cpu model names in the current host
> > context. A consumer may successfully request each listed cpu model as long
> > for a given accelerator this model is runnable.
> > 
> > The QMP type AccelCpuModelInfo is introduced and the type CpuDefinitionInfo
> > is extended by the optional field 'accelerators'. It contains a list of 
> > named
> > accelerators and some indication whether the associated cpu model is 
> > runnable
> > or the default cpu model. The default cpu model is used if either no 
> > specific
> > cpu model is requested during QEMU startup or if the cpu model with name
> > 'host' is requested.
> > 
> > request:
> >   {"execute": "query-cpu-definitions"}
> > 
> > answer:
> >   {"return":
> >     
> > [{"name":"2964-ga1","accelerators":[{"name":"kvm","runnable":false,"default":false}]},
> >      
> > {"name":"2828-ga1","accelerators":[{"name":"kvm","runnable":false,"default":false}]},
> >      
> > {"name":"2827-ga2","accelerators":[{"name":"kvm","runnable":true,"default":true}]},
> >      
> > {"name":"2827-ga1","accelerators":[{"name":"kvm","runnable":true,"default":false}]},
> >      
> > {"name":"2818-ga1","accelerators":[{"name":"kvm","runnable":true,"default":false}]},
> >      ...
> >      
> > {"name":"2064-ga1","accelerators":[{"runnable":true,"name":"kvm","default":false}]}
> >     ]
> >    }
> 
> On x86, being runnable or not is something that depends on the
> machine-type. I expect that to happen in other machines as soon as they
> start implementing backwards compatiblity.
> 
> I see two options to implement that: 1) adding a "machine-type" argument
> to query-cpu-definitions; 2) returning a machine-type-based dictionary
> on the "runnable" property. The former sounds better to me as it won't
> require enumerating all machine-types every time.
> 
> In that case, why we do need to enumerate all accelerators on every
> query, either? We could have both "machine-type" and "accel" arguments
> to query-cpu-definitions, so callers will just ask for the
> acceleratores/machine-types they are interested into.
> 
> e.g.:
> 
> request:
>   {"execute": "query-cpu-definitions",
>    "arguments": {"machine":"s390-virtio", "accel":"kvm"}}
> 
> answer:
>   {"return":
>     [{"name":"2964-ga1","runnable":true},
>      {"name":"2828-ga1","runnable":false}
>      ...
>     ]
>   }

I had some discussion with out libvirt people on that. I will compile a patch 
implementing this
to see how it fits..

> 
> We can also extend this to other variables, such as extra CPU flags that
> could make the CPU runnable or not. e.g.: want to know if "-machine
> foo,accel=bar -cpu xxx,+yyy,-zzz" is runnable? Send this request:
>   {"execute": "query-cpu-definitions",
>    "arguments": {"machine":"s390-virtio", "accel":"kvm", 
> "cpu":"xxx,+yyy,-zzz"}}
> and get this response:
>   {"return": [{"name":"xxx","runnable":false}]}
> or maybe being more explicit in the response about the extra CPU flags:
>   {"return": [{"name":"xxx,+yyy,-zzz","runnable":false}]}
> 
> 
> >   {"execute": "query-cpu-definitions"}
> > 
> > answer:
> >   {"return":
> >     
> > [{"name":"2964-ga1","accelerators":[{"name":"kvm","runnable":false,"default":false}]},
> >      
> > {"name":"2828-ga1","accelerators":[{"name":"kvm","runnable":false,"default":false}]},
> >      
> > {"name":"2827-ga2","accelerators":[{"name":"kvm","runnable":true,"default":true}]},
> >      
> > {"name":"2827-ga1","accelerators":[{"name":"kvm","runnable":true,"default":false}]},
> >      
> > {"name":"2818-ga1","accelerators":[{"name":"kvm","runnable":true,"default":false}]},
> > 
> > Signed-off-by: Michael Mueller <address@hidden>
> > [...]
>