Re: A proposal - binary

[Zachary Amsden <zach@vmware.com>]

Alan Cox wrote:
> Ar Iau, 2006-08-03 am 22:29 +0200, ysgrifennodd Willy Tarreau:
>   
>> I think that the issue Zach tried to cover is the current inability to
>> keep the same binary module across multiple kernel versions. That's why
>> he compared modules<->kernel to ELF<->glibc. In that sense, he's right.
>>     
>
> I think thats why he's wrong.
>
> The interface for a hypedvisor is
>
>       Kernel -> Something -> Hypedvisor
>
> The kernel->something interface can change randomly by day of week, who
> cares. A better analogy would be a device driver - we recompile device
> drivers each kernel variant, which change their internal interfaces, we 
> redesign their locking but we don't have to change the hardware.
>
> Ditto talking to the hypedvisor. The ABI is the hypedvisor syscall/trap
> interface not the kernel module interface. As such insmod is just fine.
>   

Yes, the module issue is completely tangential.  We would like to have 
the ability to load a hypervisor module at run-time, and this may be 
slightly nicer from a GPL point of view, by allowing us to publish a GPL 
module that interfaces to the kernel.  But the Something layer really is 
more like firmware, and merely making a GPL'd module interface to it 
doesn't actually change the underlying legal / technical ramifications 
that Alan pointed out.

Zach