THe kbuild documentation already notes that, where there's an obj-y target in a subdirectory, it will be orphaned if the subdirectory uses obj-m. Suggest a way of forcing the directory to obj-y when it's based on a config option which could be 'm'. Linus made several suggestions as to how this is usually done here: https://lore.kernel.org/linux-kselftest/CAHk-=wgK07PQ_DBBbAbSJ41t__6de1xZ6q7RRu-JS=2SU0wqLQ@xxxxxxxxxxxxxx/ This only documents the first one, which seems most common. Signed-off-by: David Gow <davidgow@xxxxxxxxxx> --- Documentation/kbuild/makefiles.rst | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/Documentation/kbuild/makefiles.rst b/Documentation/kbuild/makefiles.rst index e67eb261c9b0..d9eb24799f52 100644 --- a/Documentation/kbuild/makefiles.rst +++ b/Documentation/kbuild/makefiles.rst @@ -254,6 +254,17 @@ from that directory will be linked into vmlinux. If the Makefile in that directory specifies obj-y, those objects will be left orphan. It is very likely a bug of the Makefile or of dependencies in Kconfig. +This means that, if a directory contains both obj-y and obj-m targets, +it should be added with obj-y. If this directory nevertheless should +only be built when a config option is enabled (typically to support +a module which requires one or two files to nevertheless be built-in), +'subst m,y' can be used to ensure obj-y is used. + +Example:: + + #drivers/Makefile + obj-$(subst m,y,$(CONFIG_HYPERV)) += hv/ + Kbuild also supports dedicated syntax, subdir-y and subdir-m, for descending into subdirectories. It is a good fit when you know they do not contain kernel-space objects at all. A typical usage is to let -- 2.39.2.722.g9855ee24e9-goog
