Re: [V8 PATCH 7/8] pvh dom0: add check for pvh in vioapic_range

[George Dunlap <george.dunlap@eu.citrix.com>]

On 04/07/2014 07:59 AM, Jan Beulich wrote:
>>>> On 05.04.14 at 03:00, <mukesh.rathor@oracle.com> wrote:
>> On Tue, 01 Apr 2014 16:09:15 +0100
>> "Jan Beulich" <JBeulich@suse.com> wrote:
>>
>>>>>> On 01.04.14 at 16:40, <george.dunlap@eu.citrix.com> wrote:
>>>> On 03/24/2014 09:34 AM, Jan Beulich wrote:
>>>>>>>> On 22.03.14 at 02:39, <mukesh.rathor@oracle.com> wrote:
>>>>>> --- a/xen/arch/x86/hvm/vioapic.c
>>>>>> +++ b/xen/arch/x86/hvm/vioapic.c
>>>>>> @@ -238,8 +238,13 @@ static int vioapic_write(
>>>>>>
>>>>>>    static int vioapic_range(struct vcpu *v, unsigned long addr)
>>>>>>    {
>>>>>> -    struct hvm_hw_vioapic *vioapic = domain_vioapic(v->domain);
>>>>>> +    struct hvm_hw_vioapic *vioapic;
>>>>>> +
>>>>>> +    /* pvh uses event channel callback */
>>>>>> +    if ( is_pvh_vcpu(v) )
>>>>>> +        return 0;
>>>>>>
>>>>>> +    vioapic = domain_vioapic(v->domain);
>>>>> I can see why the extra check is needed, but I can't see why you
>>>>> convert the initializer to an assignment: Afaict domain_vioapic()
>>>>> is safe even if d->arch.hvm_domain.vioapic == NULL.
>>>> Or better yet, just make it something like:
>>>>
>>>> return vioapic && ((addr >= [...original range check]))
>>>>
>>>> That way we don't have to have a PVH-specific hook at all.  If a
>>>> domain doesn't have a vioapic for any reason, return 0.
>>> No, vioapic isn't going to be NULL for PVH:
>>>
>>> #define domain_vioapic(d)
>>> (&(d)->arch.hvm_domain.vioapic->hvm_hw_vioapic)
>> No, viopaic is NULL for PVH, hence the patch. So, can prob just check
>> for the ptr like George suggests and remove the pvh check.
> Sorry, no - I agree that d->arch.hvm_domain.vioapic is NULL for
> PVH, but that doesn't mean the result of domain_vioapic(d) is too
> (see the quoted #define above).

I interpreted Mukesh saying "check for the ptr" as, "check 
d->arch_hvm.domain.vioapic", not "check domain_vioapic(d)".

  -George