[OE-core] [PATCH v2 1/3] module.bbclass: use Module.symvers for dependants

[Denys Dmytriyenko]

On Thu, Aug 18, 2016 at 08:56:24AM +0100, André Draszik wrote:
> When compiling multiple external kernel modules, where one
> depends on the other, there are two problems at the
> moment:
> 1) we get compile time warnings from the kernel build
>    system due to missing symbols (from modpost).
> 2) Any modules generated are missing dependency
>    information (in the .modinfo elf section) for any
>    dependencies outside the current source tree and
>    outside the kernel itself.
> 
> This is expected, but the kernel build system has a way to
> deal with this - the dependent module is expected to
> specify KBUILD_EXTRA_SYMBOLS (as a space-separated list)
> to point to any and all Module.symvers of kernel modules
> that are dependencies.
> 
> While 1) by itself is not really a big issue, 2) prevents
> the packaging process from generating cross-source tree
> package dependencies.
> 
> As a first step to solve the missing dependencies in
> packages created, we:
> 1) install Module.symvers of all external kernel module
>    builds (into a location that is automatically packaged
>    into the -dev package)
> 2) make use of KBUILD_EXTRA_SYMBOLS and pass the location
>    of all Module.symvers of all kernel-module-* packages
>    we depend on
> 
> This solves both problems mentioned above.
> 
> Signed-off-by: André Draszik <git at andred.net>
> ---
>  meta/classes/module.bbclass | 15 +++++++++++++++
>  1 file changed, 15 insertions(+)
> 
> diff --git a/meta/classes/module.bbclass b/meta/classes/module.bbclass
> index 01c9309..68e3d34 100644
> --- a/meta/classes/module.bbclass
> +++ b/meta/classes/module.bbclass
> @@ -8,6 +8,15 @@ EXTRA_OEMAKE += "KERNEL_SRC=${STAGING_KERNEL_DIR}"
>  
>  MODULES_INSTALL_TARGET ?= "modules_install"
>  
> +python __anonymous () {
> +    depends = d.getVar('DEPENDS', True)
> +    extra_symbols = []
> +    for dep in depends.split():
> +        if dep.startswith("kernel-module-"):
> +            extra_symbols.append("${STAGING_INCDIR}/" + dep + "/Module.symvers")
> +    d.setVar('KBUILD_EXTRA_SYMBOLS', " ".join(extra_symbols))
> +}
> +
>  module_do_compile() {
>  	unset CFLAGS CPPFLAGS CXXFLAGS LDFLAGS
>  	oe_runmake KERNEL_PATH=${STAGING_KERNEL_DIR}   \
> @@ -15,6 +24,7 @@ module_do_compile() {
>  		   CC="${KERNEL_CC}" LD="${KERNEL_LD}" \
>  		   AR="${KERNEL_AR}" \
>  	           O=${STAGING_KERNEL_BUILDDIR} \
> +		   KBUILD_EXTRA_SYMBOLS="${KBUILD_EXTRA_SYMBOLS}" \
>  		   ${MAKE_TARGETS}
>  }
>  
> @@ -24,6 +34,11 @@ module_do_install() {
>  	           CC="${KERNEL_CC}" LD="${KERNEL_LD}" \
>  	           O=${STAGING_KERNEL_BUILDDIR} \
>  	           ${MODULES_INSTALL_TARGET}
> +
> +	install -d -m0755 ${D}${includedir}/${BPN}
> +	cp -a --no-preserve=ownership ${B}/Module.symvers ${D}${includedir}/${BPN}

Hmm, why is Module.symvers expected to be in the root of ${B}? This seems like 
a very artificial assumption/requirement!

I have some out-of-tree modules with complicated hierarchies and it worked 
fine so far until this change, because there were only 2 well defined 
interfaces - make ${MAKE_TARGETS} and make ${MODULES_INSTALL_TARGET} - and 
both of them know where the resulting module .ko resides deep inside ${B} 
hierarchy.

I wonder if this should have been rolled into ${MODULES_INSTALL_TARGET} 
step...

-- 
Denys


> +	# it doesn't actually seem to matter which path is specified here
> +	sed -e 's:${B}/::g' -i ${D}${includedir}/${BPN}/Module.symvers
>  }
>  
>  EXPORT_FUNCTIONS do_compile do_install
> -- 
> 2.9.3
> 
> -- 
> _______________________________________________
> Openembedded-core mailing list
> Openembedded-core at lists.openembedded.org
> http://lists.openembedded.org/mailman/listinfo/openembedded-core