Re: [Qemu-devel] [PATCH v6 02/17] Add accelerator id and model name to CPUState

[Eduardo Habkost]

On Wed, May 06, 2015 at 11:59:38AM +0200, Michael Mueller wrote:
> On Tue, 5 May 2015 10:26:02 -0300
> Eduardo Habkost <address@hidden> wrote:
> 
> > On Mon, Apr 27, 2015 at 04:53:16PM +0200, Michael Mueller wrote:
> > > The patch defines ids per accelerator and adds the accel_id and
> > > the model_name to the CPUState. The accel_id is initialized by
> > > common code, the model name needs to be initialized by target
> > > specific code.
> > > 
> > > Signed-off-by: Michael Mueller <address@hidden>
> > > Acked-by: Christian Borntraeger <address@hidden>
> > > ---
> > >  include/qom/cpu.h |  5 +++++
> > >  qapi-schema.json  |  9 +++++++++
> > >  qom/cpu.c         | 14 ++++++++++++++
> > >  3 files changed, 28 insertions(+)
> > > 
> > > diff --git a/include/qom/cpu.h b/include/qom/cpu.h
> > > index 9dafb48..4ffc050 100644
> > > --- a/include/qom/cpu.h
> > > +++ b/include/qom/cpu.h
> > > @@ -236,6 +236,8 @@ struct kvm_run;
> > >   * @mem_io_pc: Host Program Counter at which the memory was accessed.
> > >   * @mem_io_vaddr: Target virtual address at which the memory was 
> > > accessed.
> > >   * @kvm_fd: vCPU file descriptor for KVM.
> > > + * @accel_id: accelerator id of this CPU.
> > > + * @model_name: model name of this CPU
> > >   *
> > >   * State of one CPU core or thread.
> > >   */
> > > @@ -313,6 +315,9 @@ struct CPUState {
> > >         (absolute value) offset as small as possible.  This reduces code
> > >         size, especially for hosts without large memory offsets.  */
> > >      volatile sig_atomic_t tcg_exit_req;
> > > +
> > > +    AccelId accel_id;
> > 
> > This can be a AccelState pointer, set on initialization, because we have
> > another user case for having a AccelState pointer: query-cpu-definition
> > implementations may create temporary CPU objects with a different accel
> > object to be able to probe for accel-specific data.
> > 
> > (The pointer may become a link QOM property later.)
> > 
> > 
> > > +    char *model_name;
> > >  };
> > >  
> > >  QTAILQ_HEAD(CPUTailQ, CPUState);
> > > diff --git a/qapi-schema.json b/qapi-schema.json
> > > index ac9594d..540e520 100644
> > > --- a/qapi-schema.json
> > > +++ b/qapi-schema.json
> > > @@ -2515,6 +2515,15 @@
> > >  ##
> > >  { 'command': 'query-machines', 'returns': ['MachineInfo'] }
> > >  
> > > +# @AccelId
> > > +#
> > > +# Defines accelerator ids
> > > +#
> > > +# Since: 2.4
> > > +##
> > > +{ 'enum': 'AccelId',
> > > +  'data': ['qtest', 'tcg', 'kvm', 'xen'] }
> > > +
> > 
> > Not sure if it is better to have an enum or simply a string here.
> > 
> > >  ##
> > >  # @CpuDefinitionInfo:
> > >  #
> > > diff --git a/qom/cpu.c b/qom/cpu.c
> > > index 108bfa2..457afc7 100644
> > > --- a/qom/cpu.c
> > > +++ b/qom/cpu.c
> > > @@ -67,6 +67,20 @@ CPUState *cpu_generic_init(const char *typename, const 
> > > char *cpu_model)
> > >          goto out;
> > >      }
> > >  
> > > +    if (tcg_enabled()) {
> > > +        cpu->accel_id = ACCEL_ID_TCG;
> > > +    } else if (kvm_enabled()) {
> > > +        cpu->accel_id = ACCEL_ID_KVM;
> > > +    }
> > > +#ifdef CONFIG_XEN
> > > +    else if (xen_enabled()) {
> > > +        cpu->accel_id = ACCEL_ID_XEN;
> > > +    }
> > > +#endif
> > > +    else {
> > > +        cpu->accel_id = ACCEL_ID_QTEST;
> > > +    }
> > 
> > You can simply use ACCEL_GET_CLASS(current_machine->accelerator)->name
> > here. If we really want an enum, we can add an AccelId field to
> > AccelClass, and initialize it properly on the accel class_init
> > functions.
> 
> The AccelClass (ac = ACCEL_GET_CLASS(current_machine->accelerator) would be 
> ok though.
> That will allow to access the ac->accel_id (not yet there) at places where 
> required.
> 
> I'm just not sure how to access current_machine here.

It is a global variable declared in hw/boards.h. But it makes sense only
if !CONFIG_USER.

But I am starting to wonder if we shouldn't simply expose accel info as
a /mmachine property, instead of per-CPU information. It may become
per-CPU in the future, but right now we really don't support having
multiple accelerators so why are we trying to pretend we do?

> 
> > 
> > CONFIG_USER may require some special code when returning the accelerator
> > ID as "tcg" because IIRC it doesn't use the QOM accel classes (yet).
> > 
> > > +
> > >      object_property_set_bool(OBJECT(cpu), true, "realized", &err);
> > >  
> > >  out:
> > > -- 
> > > 1.8.3.1
> > > 
> > 
> 

-- 
Eduardo