Re: linux-next: manual merge of the kvm-ppc tree with the powerpc-merge tree

On 10/11/2012 12:24:59 PM, Timur Tabi wrote:

Scott Wood wrote:

quoted

quoted

quoted

My concern is that when I think of a user-space header file, I =20

think

quoted

quoted

quoted

of a
user-space application that calls ioctls.  I know that KVM guest
kernels
run as user-space processes, but that does not seem like a =20

reason to

quoted

quoted

quoted

combine all of the header files that the KVM guest kernel needs =20

with

quoted

quoted

quoted

"real" user-space header files.

=20

quoted

So where should guest headers go?

=20
I admit that I don't have any answers, especially since this whole =20
thing
is new to me.  Like I said, I don't know much about KVM internals, so =20
I
just don't understand why KVM guests need to have access to these =20
kernel
header files as if they're user header files.  The guests are still =20
Linux
kernels (or other OSes that think they're running as privileged code).

For hypercalls and other paravirt.  That's the point -- they're not =20
kernel headers.  They're guest API headers.

-scott=

Scott Wood wrote:

For hypercalls and other paravirt.  That's the point -- they're not  
kernel headers.  They're guest API headers.

Well, IMHO, guest API != user-space.  Maybe we should have a parallel
mechanism to uapi to make the kvm guest header files available.

-- 
Timur Tabi
Linux kernel developer at Freescale