On 14.06.2017 13:25, Paolo Bonzini wrote: > > > On 14/06/2017 13:18, Thomas Huth wrote: >> On 08.06.2017 16:18, Thomas Huth wrote: >>> There does not seem to be any target specific code in these files, so >>> we can put them into "common-obj" instead of "obj" to compile them only >>> once for all targets. >> >> Self-NACK: balloon.c uses kvm_enabled() which in turn depends on >> CONFIG_KVM ... and that flag is target-specific, so balloon.o can not be >> moved to common-obj right now. >> >> We should poison CONFIG_KVM for common code ... I'll have a look into >> that... > > No, your patch is okay. > > #if defined CONFIG_KVM || !defined NEED_CPU_H > #define kvm_enabled() (kvm_allowed) > ... > #else > #define kvm_enabled() (0) > > In other words, kvm_enabled() forces itself to be 0 only in per-target > files. This is done mostly to limit the number of required stubs: > target-independent APIs can be called from compile-once files and must > be stubbed; target-specific APIs can only be called from per-target > files, and their calls are culled by the compiler if !CONFIG_KVM.
OK, thanks for the explanation! ... but I now think my patch is still wrong, due to another reason: numa.o indirectly uses ram_addr_t variables, and these are target specific, as far as I can see. So at least that file should not be moved to common-obj-y. Thomas
