Re: [PATCH v8a 5/6] xen/arm: split vgic driver into generic and vgic-v2 driver

[Julien Grall <julien.grall@linaro.org>]

On 07/03/2014 02:02 PM, Ian Campbell wrote:
> On Thu, 2014-07-03 at 13:57 +0100, Julien Grall wrote:
> 
>>> +struct vgic_ops {
>>> +    /* Initialize vGIC */
>>> +    int (*vcpu_init)(struct vcpu *v);
>>> +    /* Domain specific initialization of vGIC */
>>> +    int (*domain_init)(struct domain *d);
>>> +    /* SGI handler of vGIC */
>>> +    int (*send_sgi)(struct vcpu *v, register_t sgir);
>>
>> By reviewing the VGIC-v3 support, I still don't think this is the right
>> callback to add. You bypass the VGIC common emulation with your
>> vgic_emulate...
>>
>> I would introduce a callback to emulate_sysreg rather than this send_sgi.
> 
> Why? The vgic will either be v2 or v3, so either MMIO or sysreg, once
> the thing has been decoded then you want to send an SGI I think, hence
> the callback. Passing a register_t does seem odd though, I'd have
> thought it would take an SGI number and any other flags which would then
> be interpreted for either v2 or v3 as appropriate.

The decoding depends on the vgic emulation. For now this function is
badly implement in vgic-v3.c.

What I was trying to say is send_sgi can be handled internally. If you
are looking to the calls of this function, it's only happen within the
file vgic-v2.c (or vgic-v3.c)

But, the sysreg emulation is called outside the vgic code. So we should
add a callaback for this.

Regards,

-- 
Julien Grall