kvm-bindings: fix feature inconsistencies

Turns out that there is a flaw in upstream kvm-bindings that
wasn't caught so far. Some parts of the serde feature also need
the fam-wrappers feature.

This was first discovered here [0] [1].

[0] #350
[1] https://buildkite.com/rust-vmm/kvm-ci/builds/1463/steps/canvas?jid=01995738-26ed-4e8c-9e14-c3f23d6de434

Signed-off-by: Philipp Schuster <[email protected]>
On-behalf-of: SAP [email protected]

Author: phip1611

kvm-bindings/src/x86_64/serialize.rs

@@ -9,6 +9,7 @@ use super::bindings::{
     kvm_nested_state__bindgen_ty_1, kvm_pit_channel_state, kvm_pit_state2, kvm_regs, kvm_segment,
     kvm_sregs, kvm_vcpu_events, kvm_xcr, kvm_xcrs, kvm_xsave,
 };
+#[cfg(feature = "fam-wrappers")]
 use super::fam_wrappers::kvm_xsave2;
 use super::nested::{KvmNestedStateBuffer, kvm_nested_state__data};
 use serde::{Deserialize, Deserializer, Serialize, Serializer};
@@ -33,13 +34,15 @@ serde_impls!(
     kvm_msrs,
     kvm_cpuid2,
     kvm_xsave,
-    kvm_xsave2,
     kvm_irqchip,
     kvm_irq_routing,
     kvm_irq_routing_entry,
     KvmNestedStateBuffer
 );
 
+#[cfg(feature = "fam-wrappers")]
+serde_impls!(kvm_xsave2);
+
 // SAFETY: zerocopy's derives explicitly disallow deriving for unions where
 // the fields have different sizes, due to the smaller fields having padding.
 // Miri however does not complain about these implementations (e.g. about
@@ -200,6 +203,7 @@ mod tests {
         is_serde::<kvm_vcpu_events>();
         is_serde::<kvm_debugregs>();
         is_serde::<kvm_xsave>();
+        #[cfg(feature = "fam-wrappers")]
         is_serde::<kvm_xsave2>();
         is_serde::<kvm_xcr>();
         is_serde::<kvm_xcrs>();
@@ -235,6 +239,7 @@ mod tests {
         is_serde_json::<kvm_vcpu_events>();
         is_serde_json::<kvm_debugregs>();
         is_serde_json::<kvm_xsave>();
+        #[cfg(feature = "fam-wrappers")]
         is_serde_json::<kvm_xsave2>();
         is_serde_json::<kvm_xcr>();
         is_serde_json::<kvm_xcrs>();