Re: [PATCH v1] kbuild: Add KBUILD_VMLINUX_LIBS_PRELINK

From: Nathan Chancellor <nathan@kernel.org>
To: Josh Poimboeuf <jpoimboe@kernel.org>
Cc: Tiezhu Yang <yangtiezhu@loongson.cn>,
	Ard Biesheuvel <ardb@kernel.org>,
	Huacai Chen <chenhuacai@kernel.org>,
	linux-kbuild@vger.kernel.org, linux-efi@vger.kernel.org,
	loongarch@lists.linux.dev, linux-arm-kernel@lists.infradead.org,
	linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v1] kbuild: Add KBUILD_VMLINUX_LIBS_PRELINK
Date: Fri, 21 Nov 2025 17:01:01 -0700	[thread overview]
Message-ID: <20251122000101.GA1996391@ax162> (raw)
In-Reply-To: <hka6ukulmkmxp7mrcmxmllco2sqof2elextc7ggcikazfzmwni@a4pnogedcr7r>

On Fri, Nov 21, 2025 at 10:42:47AM -0800, Josh Poimboeuf wrote:
> On Fri, Nov 21, 2025 at 12:01:40AM -0700, Nathan Chancellor wrote:
> > > @@ -1199,6 +1199,7 @@ KBUILD_VMLINUX_OBJS := built-in.a $(patsubst %/, %/lib.a, $(filter %/, $(libs-y)
> > >  KBUILD_VMLINUX_LIBS := $(filter-out %/, $(libs-y))
> > >  
> > >  export KBUILD_VMLINUX_LIBS
> > > +export KBUILD_VMLINUX_LIBS_PRELINK
> > 
> > This variable name is a little confusing to me since they do get added
> > to vmlinux during linking, not before. I am not sure of a better one
> > though, maybe KBUILD_VMLINUX_LIBS_FINAL?
> 
> I think that name was my idea.  Not sure what I was thinking, I guess
> "prelink" was somehow meant to imply after the vmlinux.o whole-archive
> link but before the "final" link.

Yeah I could see where you were going with that but glad we agree final
is better heh.

> "final" is indeed better.
> 
> > It may also make sense to
> > introduce similar syntax to the existing libs-y syntax, maybe
> > 
> >   final-libs-y
> > 
> > or something like that?
> 
> Yeah, I suppose that would mirror how KBUILD_VMLINUX_LIBS is wired up.

On second thought, it might be less complicated to keep it this way...

> > Also, since these objects are no longer in KBUILD_VMLINUX_LIBS, does
> > this new variable need to be added to any of the other places in the
> > build system that use KBUILD_VMLINUX_LIBS for things such as build
> > dependencies?
> 
> Not sure about that one.  drivers/firmware/efi/libstub/lib.a is getting
> built regardless so there must already be an implicit dependency?

I think the explicit dependency currently comes from

  vmlinux_o: vmlinux.a $(KBUILD_VMLINUX_LIBS)

in Makefile. I think something like

diff --git a/scripts/Makefile.vmlinux b/scripts/Makefile.vmlinux
index cd788cac9d91..38b5f0ba9a4b 100644
--- a/scripts/Makefile.vmlinux
+++ b/scripts/Makefile.vmlinux
@@ -68,7 +68,7 @@ cmd_link_vmlinux =							\
 	$(if $(ARCH_POSTLINK), $(MAKE) -f $(ARCH_POSTLINK) $@, true)
 
 targets += vmlinux.unstripped .vmlinux.export.o
-vmlinux.unstripped: scripts/link-vmlinux.sh vmlinux.o .vmlinux.export.o $(KBUILD_LDS) FORCE
+vmlinux.unstripped: scripts/link-vmlinux.sh vmlinux.o .vmlinux.export.o $(KBUILD_LDS) $(KBUILD_VMLINUX_LIBS_FINAL) FORCE
 	+$(call if_changed_dep,link_vmlinux)
 ifdef CONFIG_DEBUG_INFO_BTF
 vmlinux.unstripped: $(RESOLVE_BTFIDS)

is needed to ensure that an updated drivers/firmware/efi/libstub/lib.a
results in vmlinux getting regenerated (at least according to my
testing).

Cheers,
Nathan

