Re: [PATCH] include/sysemu: Poison all accelerator CONFIG switches in common code

[Peter Maydell]

On Tue, 16 Mar 2021 at 12:54, Thomas Huth <thuth@redhat.com> wrote:
>
> We are already poisoning CONFIG_KVM since this switch is not working
> in common code. Do the same with the other accelerator switches, too
> (except for CONFIG_TCG, which is special, since it is also defined in
> config-host.h).
>
> Signed-off-by: Thomas Huth <thuth@redhat.com>
> ---
>  include/exec/poison.h | 4 ++++
>  include/sysemu/hax.h  | 4 ++++
>  include/sysemu/hvf.h  | 4 ++++
>  include/sysemu/whpx.h | 4 ++++
>  4 files changed, 16 insertions(+)
>
> diff --git a/include/exec/poison.h b/include/exec/poison.h
> index 4cd3f8abb4..3250fc1d52 100644
> --- a/include/exec/poison.h
> +++ b/include/exec/poison.h
> @@ -88,8 +88,12 @@
>  #pragma GCC poison CONFIG_SPARC_DIS
>  #pragma GCC poison CONFIG_XTENSA_DIS
>
> +#pragma GCC poison CONFIG_HAX
> +#pragma GCC poison CONFIG_HVF
>  #pragma GCC poison CONFIG_LINUX_USER
>  #pragma GCC poison CONFIG_KVM
>  #pragma GCC poison CONFIG_SOFTMMU
> +#pragma GCC poison CONFIG_WHPX
> +#pragma GCC poison CONFIG_XEN
>
>  #endif
> diff --git a/include/sysemu/hax.h b/include/sysemu/hax.h
> index 12fb54f990..247f0661d1 100644
> --- a/include/sysemu/hax.h
> +++ b/include/sysemu/hax.h
> @@ -24,6 +24,8 @@
>
>  int hax_sync_vcpus(void);
>
> +#ifdef NEED_CPU_H
> +
>  #ifdef CONFIG_HAX
>
>  int hax_enabled(void);
> @@ -34,4 +36,6 @@ int hax_enabled(void);
>
>  #endif /* CONFIG_HAX */
>
> +#endif /* NEED_CPU_H */
> +
>  #endif /* QEMU_HAX_H */

Now we only get hax_enabled() if NEED_CPU_H was defined
(which is different behaviour from kvm_enabled()).

I think we should structure all of these accel-specific
headers the same way we do kvm.h:
 * if we're being included from non-target-specific code, or
   if we're in target-specific code and we know KVM is present,
   you get the versions that reference functions/bools/etc
 * if we're being included from target-specific code where we
   know KVM is not present there are macros that evaluate
   to always-false/always-fails
 * there's a stub file that proivdes the functions/bools
   for the KVM-not-actually-present case

My argument here is basically consistency. We know already
that the structure we use in kvm.h works, and it's the most
solidly used/tested of the accelerators. Structuring the
other accelerators differently for no particular reason
is confusing to people trying to read or modify the code.

thanks
-- PMM