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

[Scott Wood <scottwood@freescale.com>]

On 10/11/2012 10:56:51 AM, Alexander Graf wrote:
>=20
> On 11.10.2012, at 17:50, Scott Wood wrote:
>=20
> > On 10/11/2012 08:04:58 AM, Alexander Graf wrote:
> >> On 11.10.2012, at 05:32, Tabi Timur-B04825 wrote:
> >> > On Wed, Oct 10, 2012 at 9:47 PM, Stephen Rothwell =20
> <sfr@canb.auug.org.au> wrote:
> >> >
> >> >> Commit 549d62d889b4 ("KVM: PPC: use definitions in epapr header
> >> >> for hcalls") from the kvm-ppc tree added an include of =20
> asm/epapr_hcall.h
> >> >> to the user visible part of asm/kvm_para.h so asm/epapr_hcall.h =20
> became a
> >> >> user visible header file.
> >> >
> >> > Any real user-space code that tries to call any of the functions =20
> in
> >> > epapr_hcall.h will cause an exception.
> >
> > Actually it'll cause a linker error, as of the "Don't use hardcoded =20
> opcode for ePAPR hcall invocation" patch.
> >
> >> > Claiming that kernel header files that KVM needs are suddenly
> >> > user-space header files doesn't make much sense to me, but I =20
> guess
> >> > it's not my decision.
> >> This is about kvm_host.h, which is the part that is exported to =20
> user space. It usually contains constants and structs that are =20
> required for the API.
> >
> > s/kvm_host/kvm_para/
> >
> > I'm not sure anything in kvm_para.h or epapr_hcalls.h would be of =20
> use to userspace, but it would be of use to the guest OS, so it's =20
> still not kernel-internal.
> >
> > The variable externs should be #ifdef __KERNEL__ as David points =20
> out.  The asm stuff could still be useful in a guest OS if the guest =20
> provides epapr_hypercall_start.
>=20
> I don't think we should worry about guest kernel code potentially =20
> having use for this. Let's make the user space headers be great for =20
> user space usage, not for guest code usage :).

Well, that asm stuff started out as a guest header (from Freescale's =20
Topaz hypervisor), not a Linux header. :-)

-Scott=