On 23.02.2026 20:00, Andrew Cooper wrote: > On 02/02/2026 4:26 pm, Jan Beulich wrote: >> On 02.02.2026 16:47, Andrew Cooper wrote: >>> On 07/01/2026 2:17 pm, Jan Beulich wrote: >>>> diff --git a/xen/arch/x86/arch.mk b/xen/arch/x86/arch.mk >>>> index 0203138a819a..be6c76d2934b 100644 >>>> --- a/xen/arch/x86/arch.mk >>>> +++ b/xen/arch/x86/arch.mk >>>> @@ -4,6 +4,7 @@ >>>> export XEN_IMG_OFFSET := 0x200000 >>>> >>>> ARCH_LIBS-y += arch/x86/lib/lib.a >>>> +ALL_LIBS-y += arch/x86/lib/cpu-policy/lib.a >>> This wants to extend ARCH_LIBS-y surely? Is this a rebasing oversight? >> No, this was deliberate. The functions here are different from those in >> arch/x86/lib/lib.a. We don't need to fear collision with "common code" >> ones. Hence I preferred to use the more "normal" placement into what's >> passed to the linker. > > I agree that we don't have the explicit ordering requirement that we > have with arch/x86/lib/lib.a. > > But, it still reads as bogus to be putting arch/x86/lib/cpu-policy/lib.a > in the non-ARCH list. > > What difference is there having this a little earlier in the linker > arguments? Nothing AFAICT.
Indeed. The sole reason why I'd prefer things as presented is that putting stuff in ARCH_LIBS should imo be the special case (i.e. requiring a special reason), while putting things in ALL_LIBS should be the default. Jan
