Re: [PATCH v2 11/28] kbuild: lto: postpone objtool

[Sami Tolvanen]

On Thu, Sep 03, 2020 at 03:19:43PM -0700, Kees Cook wrote:
> On Thu, Sep 03, 2020 at 01:30:36PM -0700, Sami Tolvanen wrote:
> > With LTO, LLVM bitcode won't be compiled into native code until
> > modpost_link, or modfinal for modules. This change postpones calls
> > to objtool until after these steps.
> > 
> > Signed-off-by: Sami Tolvanen <samitolvanen@xxxxxxxxxx>
> 
> For a "fail fast" style of building, it makes sense to have objtool run
> as early as possible, so it makes sense to keep the current behavior in
> non-LTO mode. I do wonder, though, if there is a real benefit to having
> "fail fast" case. I imagine a lot of automated builds are using
> --keep-going with make, and actually waiting until the end to do the
> validation means more code will get build-tested before objtool rejects
> the results. *shrug*
> 
> > ---
> >  arch/Kconfig              |  2 +-
> >  scripts/Makefile.build    |  2 ++
> >  scripts/Makefile.modfinal | 24 ++++++++++++++++++++++--
> >  scripts/link-vmlinux.sh   | 23 ++++++++++++++++++++++-
> >  4 files changed, 47 insertions(+), 4 deletions(-)
> > 
> > diff --git a/arch/Kconfig b/arch/Kconfig
> > index 71392e4a8900..7a418907e686 100644
> > --- a/arch/Kconfig
> > +++ b/arch/Kconfig
> > @@ -599,7 +599,7 @@ config LTO_CLANG
> >  	depends on $(success,$(NM) --help | head -n 1 | grep -qi llvm)
> >  	depends on $(success,$(AR) --help | head -n 1 | grep -qi llvm)
> >  	depends on ARCH_SUPPORTS_LTO_CLANG
> > -	depends on !FTRACE_MCOUNT_RECORD
> > +	depends on HAVE_OBJTOOL_MCOUNT || !(X86_64 && DYNAMIC_FTRACE)
> >  	depends on !KASAN
> >  	depends on !GCOV_KERNEL
> >  	select LTO
> > diff --git a/scripts/Makefile.build b/scripts/Makefile.build
> > index c348e6d6b436..b8f1f0d65a73 100644
> > --- a/scripts/Makefile.build
> > +++ b/scripts/Makefile.build
> > @@ -218,6 +218,7 @@ cmd_record_mcount = $(if $(findstring $(strip $(CC_FLAGS_FTRACE)),$(_c_flags)),
> >  endif # USE_RECORDMCOUNT
> >  
> >  ifdef CONFIG_STACK_VALIDATION
> > +ifndef CONFIG_LTO_CLANG
> >  ifneq ($(SKIP_STACK_VALIDATION),1)
> >  
> >  __objtool_obj := $(objtree)/tools/objtool/objtool
> > @@ -253,6 +254,7 @@ objtool_obj = $(if $(patsubst y%,, \
> >  	$(__objtool_obj))
> >  
> >  endif # SKIP_STACK_VALIDATION
> > +endif # CONFIG_LTO_CLANG
> >  endif # CONFIG_STACK_VALIDATION
> >  
> >  # Rebuild all objects when objtool changes, or is enabled/disabled.
> > diff --git a/scripts/Makefile.modfinal b/scripts/Makefile.modfinal
> > index 1005b147abd0..909bd509edb4 100644
> > --- a/scripts/Makefile.modfinal
> > +++ b/scripts/Makefile.modfinal
> > @@ -34,10 +34,30 @@ ifdef CONFIG_LTO_CLANG
> >  # With CONFIG_LTO_CLANG, reuse the object file we compiled for modpost to
> >  # avoid a second slow LTO link
> >  prelink-ext := .lto
> > -endif
> > +
> > +# ELF processing was skipped earlier because we didn't have native code,
> > +# so let's now process the prelinked binary before we link the module.
> > +
> > +ifdef CONFIG_STACK_VALIDATION
> > +ifneq ($(SKIP_STACK_VALIDATION),1)
> > +cmd_ld_ko_o +=								\
> > +	$(objtree)/tools/objtool/objtool				\
> > +		$(if $(CONFIG_UNWINDER_ORC),orc generate,check)		\
> > +		--module						\
> > +		$(if $(CONFIG_FRAME_POINTER),,--no-fp)			\
> > +		$(if $(CONFIG_GCOV_KERNEL),--no-unreachable,)		\
> > +		$(if $(CONFIG_RETPOLINE),--retpoline,)			\
> > +		$(if $(CONFIG_X86_SMAP),--uaccess,)			\
> > +		$(if $(USE_OBJTOOL_MCOUNT),--mcount,)			\
> > +		$(@:.ko=$(prelink-ext).o);
> > +
> > +endif # SKIP_STACK_VALIDATION
> > +endif # CONFIG_STACK_VALIDATION
> 
> I wonder if objtool_args could be reused here instead of having two
> places to keep in sync? It looks like that might mean moving things
> around a bit before this patch, since I can't quite see if
> Makefile.build's variables are visible to Makefile.modfinal?

It doesn't look like they are. I suppose we could move objtool_args to
Makefile.lib. Masahiro, any thoughts?

Sami