Re: [PATCH 2/2] KVM: Device assignemnt with VT-d

[Avi Kivity <avi@qumranet.com>]

Amit Shah wrote:
> * On Friday 22 Aug 2008 23:48:42 Avi Kivity wrote:
>   
>> Amit Shah wrote:
>>     
>>> diff --git a/include/linux/kvm.h b/include/linux/kvm.h
>>> index d9ef7d3..2956e35 100644
>>> --- a/include/linux/kvm.h
>>> +++ b/include/linux/kvm.h
>>> @@ -495,4 +495,6 @@ struct kvm_assigned_irq {
>>>         __u32 flags;
>>>  };
>>>
>>> +#define KVM_DEV_ASSIGN_USE_VTD (1 << 1)
>>> +
>>>  #endif
>>>       
>> (1 >> 0)?
>>     
>
> I kept the 1st field reserved for no particular implementation in mind as of 
> now.
>
>   

Why?

>> This is a userspace inteface, so use a generic name like iommu.  We also
>> need a KVM_CAP so userspace can check whether an iommu is present or not.
>>     
>
> We could have multiple hardware IOMMU implementations, like Intel's VT-d and 
> AMD's IOMMU.
>
>   

Not in userspace.  Userspace sees either iommu or no iommu; it doesn't 
care about the iommu model.

> Also, is KVM_CAP_foo needed for this? This is the only #define that'll be used 
> and we can simply do something like
>
> #ifdef KVM_DEV_ASSIGN_USE_VTD
> flags |= KVM_DEV_ASSIGN_USE_VTD
> #endif
>
> ?
>   

That only detects if the headers have the flag, not if the kernel 
actually supports it (and whether there is an iommu in the host).  We 
need run-time detection.

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.