On Fri, Apr 24, 2009 at 07:41:28AM +0200, Manuel Lauss wrote: > Move Makefile information on all Alchemy boards to a separate file > in the arch subdir. > > Signed-off-by: Manuel Lauss <mano@xxxxxxxxxxxxxxxxxxxxxxx> > --- > Applies on top of my Alchemy-gpio patches; builds and runs fine on a few > different alchemy systems. It seems nicer to not have to modify the main > mips makefile when adding new alchemy boards. What do you all think? > > arch/mips/Makefile | 104 +------------------------------------------ > arch/mips/alchemy/Makefile | 106 ++++++++++++++++++++++++++++++++++++++++++++ > 2 files changed, 108 insertions(+), 102 deletions(-) > create mode 100644 arch/mips/alchemy/Makefile I think Sam Ravnbord has written something like the grand plan for the cleanup a few days ago; I append his mail below. Ralf Date: Tue, 21 Apr 2009 21:43:43 +0200 From: Sam Ravnborg <sam@xxxxxxxxxxxx> To: Ralf Baechle <ralf@xxxxxxxxxxxxxx> Cc: linux-arch@xxxxxxxxxxxxxxx Subject: Re: introduce arch/$ARCH/Kbuild ? Message-ID: <20090421194343.GA24535@xxxxxxxxxxxxxxxxxxx> References: <20090416183701.GA5810@xxxxxxxxxxxxxxxxxxx> <20090421131430.GA25098@xxxxxxxxxxxxxx> In-Reply-To: <20090421131430.GA25098@xxxxxxxxxxxxxx> > > Most of arch/mips uses -Werror these days and while painful at times it > keeps everybody on their toes hopefully cleaner, less buggy code. So if > your solution allows adding -Werror to all subdirs automatically with > a mechanism to remove -Werror from a few selected dirs then I'm interested. It is already present in mainline - so go wild. I took a look at mips. mips supports an impressive amount of platform/boards. The has resulted in lines like the following in the arch Makefile: core-$(CONFIG_SGI_IP32) += arch/mips/sgi-ip32/ cflags-$(CONFIG_SGI_IP32) += -I$(srctree)/arch/mips/include/asm/mach-ip32 load-$(CONFIG_SGI_IP32) += 0xffffffff80004000 But this is less then optimal. If two people add a paltform you will have a merge issue. And centralize information like that is also questionable. mips would be better suited if you had all sgi_ip32 information located in a single directory. How about a setup like this: arch/mips/sgi_ip32/Platform: platfrom-y += arch/mips/sgi-ip32/ cflags-$(CONFIG_SGI_IP32) += -I$(srctree)/arch/mips/include/asm/mach-ip32 load-$(CONFIG_SGI_IP32) += 0xffffffff80004000 arch/mips/Kbuild.platforms: #All platforms listed in alphabetic order platforms-y += lasat/ platforms-y += sgi_ip32/ #include the platform specific files include $(patsubst %, arch/misp/%Platform) arch/mips/Kbuild: subdir-ccflags-y := -Werror include arch/mips/Kbuild.platforms obj-y += $(platform-y) obj-y += arch/mips/kernel/ arch/mips/mm/ arch/mips/math-emu/ arch/mips/Makefile: core-y += arch/mips/ include arch/mips/Kbuild.platforms The above does a few things: 1) It decentralize the plaform stuff (to the Platform files) 2) In troduces a arch/mips/Kbuild file that specify everything that is linked in as core-y 3) It adds a single subdir-ccflags-y := -Werror that covers all platforms and the core part of the kernel (Everything specified in arch/mips/Kbuild) 4) It reuses Kbuild.platforms in Kbuild and in Makefile. In Makefile it is used to find ccflags-y and load-y definitions. In Kbuild it is used to find the objects to add to obj-y. The above is entirely untested - but I hope to have adressed the principles. Sam
