Re: [PATCH bpf-next] kbuild: Restore ability to build out-of-tree modules

[Leon Romanovsky]

On Tue, Dec 01, 2020 at 01:44:46PM -0800, Andrii Nakryiko wrote:
> On Tue, Dec 1, 2020 at 11:32 AM Leon Romanovsky <leon@xxxxxxxxxx> wrote:
> >
> > On Tue, Dec 01, 2020 at 10:01:23AM -0800, Andrii Nakryiko wrote:
> > > On Tue, Dec 1, 2020 at 6:37 AM Leon Romanovsky <leon@xxxxxxxxxx> wrote:
> > > >
> > > > From: Leon Romanovsky <leonro@xxxxxxxxxx>
> > > >
> > > > The out-of-tree modules are built without vmlinux target and request
> > > > to recompile that target unconditionally causes to the following
> > > > compilation error.
> > > >
> > > > [root@server kernel]# make
> > > > <..>
> > > > make -f ./scripts/Makefile.modpost
> > > > make -f ./scripts/Makefile.modfinal
> > > > make[3]: *** No rule to make target 'vmlinux', needed by '/my_temp/out-of-tree-module/kernel/test.ko'.  Stop.
> > > > make[2]: *** [scripts/Makefile.modpost:117: __modpost] Error 2
> > > > make[1]: *** [Makefile:1703: modules] Error 2
> > > > make[1]: Leaving directory '/usr/src/kernels/5.10.0-rc5_for_upstream_base_2020_11_29_11_34'
> > > > make: *** [Makefile:80: modules] Error 2
> > > >
> > > > As a solution separate between build paths that has vmlinux target and paths without.
> > > >
> > > > Fixes: 5f9ae91f7c0d ("kbuild: Build kernel module BTFs if BTF is enabled and pahole supports it")
> > > > Reported-by: Edward Srouji <edwards@xxxxxxxxxx>
> > > > Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxx>
> > > > ---
> > >
> > > e732b538f455 ("kbuild: Skip module BTF generation for out-of-tree
> > > external modules") ([0]) was supposed to take care of this. Did you
> > > try it?
> >
> > My tree doesn't have this patch yet, so my questions can be stupid:
> > 1. Will it print "Skipping BTF generation for ... due to unavailability
> > of vmlinux" line if my .config doesn't have "CONFIG_DEBUG_INFO_BTF_MODULES"?
> > I hope it is not.
>
> No, it shouldn't. cmd_btf_ko is only executed if
> CONFIG_DEBUG_INFO_BTF_MODULES is set.
>
> > 2. Reliance on existence of vmlinux can be problematic, no one promises
> > us that "make clean" is called before and there are no other leftovers
> > from previous builds.
>
> In such a case, the worst thing that can happen would be that the
> kernel module will get BTF that doesn't match actual vmlinux BTF, and
> when attempted to load into the kernel BTF will be ignored (with a
> warning). It's not ideal, but I don't know how else we can handle this
> short of just not supporting BTF for out-of-tree modules, which a
> bunch of folks would be disappointed about, I think. I'm open to
> suggestions on how to do it better, though.

You can rely on MODPOST_VMLINUX variable to differentiate between builds
of modules vs. vmlinux. Proper message during kernel boot that vmlinux
BTF doesn't match module can be enough, because in real world you won't
load that recompiled vmlinux without "make install" first.

So right now, from out-of-tree modules perspective, you are testing against
some pseudo-random file.

BTW, I tried your patch with and without vmlinux, it worked, so we are
talking about better-to-fix and not must-to-fix thing.

Thanks

>
>
> >
> > And in general, the idea that such invasive change in build infrastructure
> > came without any Ack from relevant maintainers doesn't look right to me.
> >
> > Thanks
> >
> > >
> > >   [0] https://patchwork.kernel.org/project/netdevbpf/patch/20201121070829.2612884-1-andrii@xxxxxxxxxx/
> > >
> > >
> > > > Not proficient enough in Makefile, but it fixes the issue.
> > > > ---
> > > >  scripts/Makefile.modfinal | 5 +++++
> > > >  scripts/Makefile.modpost  | 4 ++++
> > > >  2 files changed, 9 insertions(+)
> > > >
> > > > diff --git a/scripts/Makefile.modfinal b/scripts/Makefile.modfinal
> > > > index 02b892421f7a..8a7d0604e7d0 100644
> > > > --- a/scripts/Makefile.modfinal
> > > > +++ b/scripts/Makefile.modfinal
> > > > @@ -48,9 +48,14 @@ if_changed_except = $(if $(call newer_prereqs_except,$(2))$(cmd-check),      \
> > > >         $(cmd);                                                              \
> > > >         printf '%s\n' 'cmd_$@ := $(make-cmd)' > $(dot-target).cmd, @:)
> > > >
> > > > +ifdef MODPOST_VMLINUX
> > > >  # Re-generate module BTFs if either module's .ko or vmlinux changed
> > > >  $(modules): %.ko: %.o %.mod.o scripts/module.lds vmlinux FORCE
> > > >         +$(call if_changed_except,ld_ko_o,vmlinux)
> > > > +else
> > > > +$(modules): %.ko: %.o %.mod.o scripts/module.lds FORCE
> > > > +       +$(call if_changed_except,ld_ko_o)
> > > > +endif
> > > >  ifdef CONFIG_DEBUG_INFO_BTF_MODULES
> > > >         +$(if $(newer-prereqs),$(call cmd,btf_ko))
> > > >  endif
> > > > diff --git a/scripts/Makefile.modpost b/scripts/Makefile.modpost
> > > > index f54b6ac37ac2..f5aa5b422ad7 100644
> > > > --- a/scripts/Makefile.modpost
> > > > +++ b/scripts/Makefile.modpost
> > > > @@ -114,8 +114,12 @@ targets += $(output-symdump)
> > > >
> > > >  __modpost: $(output-symdump)
> > > >  ifneq ($(KBUILD_MODPOST_NOFINAL),1)
> > > > +ifdef MODPOST_VMLINUX
> > > > +       $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.modfinal MODPOST_VMLINUX=1
> > > > +else
> > > >         $(Q)$(MAKE) -f $(srctree)/scripts/Makefile.modfinal
> > > >  endif
> > > > +endif
> > > >
> > > >  PHONY += FORCE
> > > >  FORCE:
> > > > --
> > > > 2.28.0
> > > >