Re: [PATCH 06/12] KVM: x86: API changes for SMM support

[Paolo Bonzini <pbonzini@redhat.com>]

On 21/05/2015 18:26, Radim Krčmář wrote:
> 2015-05-21 16:59+0200, Paolo Bonzini:
>> On 21/05/2015 16:49, Radim Krčmář wrote:
>>> 2015-05-08 13:20+0200, Paolo Bonzini:
>>>> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
>>>> @@ -202,7 +202,7 @@ struct kvm_run {
>>>>  	__u32 exit_reason;
>>>>  	__u8 ready_for_interrupt_injection;
>>>>  	__u8 if_flag;
>>>> -	__u8 padding2[2];
>>>> +	__u16 flags;
>>>
>>> (It got lost last review and I'd really like to know ...
>>>  what is the advantage of giving both bytes to flags?)
>>
>> No advantage.  You just should leave padding2[1] in the middle so that
>> the offset of &run->padding2[0] doesn't change.
> 
> I don't get that.  The position of padding should be decided by
> comparing probabilities of extending 'if_flag' and 'flags'.
> 
>>                                                  Since it's not obvious
>> I gave two bytes to flags, but I can do it either way.
> 
> if_flag seems to be set in stone as one bit, so I'd vote for
> 
>   __u8 flags;
>   __u8 padding2;
> 
> (Or 'padding3', to prevent the same class of errors that removing it
>  altogether does;  which we didn't do for other tailed padding).

You're right that we didn't do it.  I'll change it to flags + padding2.

Paolo

> For there isn't much space left in struct kvm ...
>