Re: [Qemu-devel] Re: [PATCH 1/3] x86: fix warning without CONFIG_KVM

[Jamie Lokier]

Anthony Liguori wrote:
> Jan Kiszka wrote:
> >Isn't it cleaner to provide an empty static inline in the !CONFIG_KVM
> >case? Not all compilers might be smart enough to optimize this function
> >away.
> 
> kvm_enabled() is (0) when !CONFIG_KVM so that's why this is a warning 
> and not an error.  The function call disappears during linking from dead 
> code elimination.  There is no routine overhead when !CONFIG_KVM.

I think it's in reference to:

>  static void host_cpuid(uint32_t function, uint32_t *eax, uint32_t *ebx,
>                         uint32_t *ecx, uint32_t *edx)

This compiles to an empty function if !CONFIG_KVM.  Current GCC will
remove the function call by auto-inlining, but some compilers
(including older GCC) will produce an empty function.

If you're saying !CONFIG_KVM results in the _call_ to host_cpuid()
being removed, I'm thinking that the option to provide the guest with
the same features as the host CPU is still useful if KVM is
disabled...  I have a real use-case for this.  It's installing some
versions of Windows in a guest, to be used with KVM after
installation, because the install process is more reliable (and
faster!) with KVM disabled but after that it works.

-- Jamie