Re: [PATCH] kbuild: use $(obj)/ instead of $(src)/ for COPY

[Nicolas Schier <nsc@kernel.org>]

On Thu, Oct 09, 2025 at 02:16:20PM -0700, Hong, Yifan wrote:
> On Thu, Oct 9, 2025 at 2:16 AM Nicolas Schier <nsc@kernel.org> wrote:
> >
> > On Mon, Oct 06, 2025 at 07:38:38PM +0000, HONG Yifan wrote:
> > > Similar to
> > > commit 9a0ebe5011f4 ("kbuild: use $(obj)/ instead of $(src)/ for common pattern rules")
> > >
> > > This change updates the COPY rule to use $(obj) instead of $(src). This
> > > makes Kbuild rules like
> > >
> > >     always-y += libfoo/.foo.o.cmd
> >
> > This is a strange example.  Why should we ship any prebuilt .*.o.cmd file?
> 
> When one ships the .o file, it might be beneficial to also ship the
> accompanying .o.cmd file
> so that compdb may work, I guess. Though, I just get this example from one
> of the SoC manufacturers, so I actually don't know the true reasoning
> behind it. I agree that it
> isn't a good example.
> 
> Still, this applies to any file that does NOT match any existing
> pattern rules in Kbuild. It might be more
> generic if I had said instead
> 
>     always-y += libfoo/foo.xyz
> 
> ... and we were providing a libfoo/foo.xyz_shipped in the source tree.
> 
> >
> > >
> > > work when the user provides libfoo/.foo.o.cmd_shipped, even when obj and
> > > src is different and src is an absolute path. This is useful when foo.o
> > > and .foo.o.cmd are checked-in as prebuilts.
> > >
> > > (Admittedly, `always-y += libfoo/.foo.o.cmd` is not recommended in
> > > kbuild/modules.rst, "Several Subdirectories".)
> > >
> > > For example, if
> > >
> > >     obj=.
> > >     src=/some/path
> > >
> > > then the original rule
> > >
> > >     $(obj)/%: $(src)/%_shipped
> > >
> > > expands to
> > >
> > >     ./%: /some/path/%_shipped
> > >
> > > And when matching against the above example, the stem is just `bar.o`
> > > [^1] so the following is looked up:
> > >
> > >     libfoo/.foo.o.cmd: libfoo//some/path/.foo.o.cmd_shipped
> > >
> > > ... and it cannot be matched.
> > >
> > > With this change, the rule expands to
> > >
> > >     ./%: ./%_shipped
> > >
> > > ... and it should work, at least for files that does not have a more
> > > specific pattern rule.
> > >
> > > NOTE: that after this change, code like
> > >
> > >     bar-y += libfoo/foo.o
> > >
> > > ... with libfoo/foo.o_shipped provided still DOES NOT work, because
> > > the pattern rule $(obj)/%.o takes priority. For .o_shipped files,
> > > the user still needs an explicit `$(obj)/%.o: $(obj)/%.o_shipped` rule
> > > in its own Kbuild file.
> > >
> > > [^1]: https://www.gnu.org/software/make/manual/html_node/Pattern-Match.html
> > >
> > > Signed-off-by: HONG Yifan <elsk@google.com>
> > > ---
> > >  scripts/Makefile.lib | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
> > > index 1d581ba5df66..e066b7b00bcc 100644
> > > --- a/scripts/Makefile.lib
> > > +++ b/scripts/Makefile.lib
> > > @@ -272,7 +272,7 @@ endef
> > >  quiet_cmd_copy = COPY    $@
> > >        cmd_copy = cat $< > $@
> > >
> > > -$(obj)/%: $(src)/%_shipped
> > > +$(obj)/%: $(obj)/%_shipped
> > >       $(call cmd,copy)
> >
> > No, I don't see a reason to support *_shipped files from the build tree.
> > The purpose of *_shipped is to deliver prebuilt files with the source
> > tree to allow or simplify building the corresponding output files.
> 
> The goal is not to support .o_shipped files from the build tree,
> though it is indeed one
> side effect of the patch. My goal is that we support .o_shipped files
> from the source tree when all
> of the following is true:
> 
> - We are building with O= / MO=, so $(obj) and $(src) are different,
> and in this case $(obj) is `.` and
>   $(src) is an absolute path);
> - We have a `xxx-y += libfoo/bar.o` line; in other words, the
> bar.o_shipped file is in a **subdirectory**
>   libfoo below the directory of the Kbuild file. As I said in the
> commit message, this is NOT recommended
>   (kbuild/modules.rst, "Several Subdirectories"), but it is a
> documented way to include a .o file from
>   elsewhere.
> 
> And as I said in the commit message at the end, unfortunately this
> patch still can't achieve this goal for .o files,
> only for files that don't match any existing pattern rules, like .cmd
> files for example.
> 
> Would you please suggest how we can support .o_shipped files in a
> subdirectory when building with
> $(obj) = . and $(src) = <some absolute path>? Thank you!

Thanks for the clarifications.  Please note that shipping .o_shipped
files with upstream Linux will not happen out of obvious reasons, and I
am not willing to invest time into supporting any closed-source
approach.

You might want to try to add the missing rules to your local modules'
Kbuild file, e.g. something like

$(obj)/%.o: $(obj)/%.o.cmd

$(obj)/%: $(src)/%_shipped | $(obj)/%.o.cmd
	$(call cmd,copy)

(not tested, no support).

Good luck and kind regards,
Nicolas