Extending kbuild syntax

Over the last few weeks I have pondered with the idea to extend the current kbuild syntax.
The idea have existed for long but only recently I started to think how to do this in a truely flexible manner.

Two areas are in need for a bit of attention to improve current kbuild files in the kernel.

The first issue is the EXTRA_CFLAGS.
They are often conditionally assigned like in the following example:

ifeq ($(DEBUG),y)
EXTRA_CFLAGS := -DDEBUG
endif

Introducing the following new variable could make this a oneliner:
ccflags-y

ccflags-$(DEBUG) := -DDEBUG

grep -r -C 1 -B 1 EXTRA_CFLAGS shows that the above is a very common pattern especially in drivers/

The kbuild variable is named 'ccflags' to match the CC prefix for the C compiler used in non-verose output.
And then it does not clash with current cflags-y usage too.

The second is the more controversial suggestion.
In several Makefile we have simple if expression of the variants:
if ($(CONFIG_FOO),y)
obj-$(CONFIG_BAR) += fubar.o
endif

The pattern varies over this theme.
The suggestion here is to introduce a few helpers:

obj-y-if-$(CONFIG_FOO) += fubar.o
This one shall read:
if $(CONFIG_FOO) is y or m then set += to obj-y

In several cases we will need it to be more complicated like the this:
obj-y-ify-$(CONFIG_FOO) += fubar.o
This one shall read:
if $(CONFIG_FOO) is y (thats the ify thing) then += to obj-y

And we can mix it like the following:
obj-$(CONFIG_BAR)-ify-$(CONFIG_FOO) += fubar.o This one shall read:
if $(CONFIG_FOO) equal y then += to obj-$(CONFIG_BAR)

The full list of new variables are (for obj-y):
obj-y-if-y
obj-y-if-m

obj-y-ify-y
obj-y-ifm-m

obj-y-ifn-

And similar for obj-m:
obj-m-if-y
obj-m-if-m

obj-m-ify-y
obj-m-ifm-m

obj-m-ifn-

The 'y' and 'm' can be replaced by $(CONFIG_FOO) but the one right to the if are not supposed to be replaced.

To better express how to use it I have tried to update a few Makefiles to use the new syntax. See below.

On MAJOR drawback is the linking order.
I have no way to keep the link order if the new syntax are sued. The .o files will either be put before .o files specified with obj-y or obj-m or after the same.

For some places this does not matter but for other places (fs/Makefile) I will expect it to matter a lot.

Comments?
Is this just to magic for bare humans to grasp.
Or mabe the linking order is so important that we do not want it?

PS. The coming x86 merge triggered this. I would like to make the Makefiles readable and the above can help here.

Sam

 fs/Makefile             |   10 +++-------
 kernel/Makefile         |    8 ++------
 mm/Makefile             |    7 +++----
 sound/Makefile          |    4 +---
 sound/core/seq/Makefile |    9 +++------
 sound/i2c/Makefile      |    4 +---
 sound/isa/sb/Makefile   |    6 ++----
 sound/oss/Makefile      |    4 +---
 8 files changed, 16 insertions(+), 36 deletions(-)

index 720c29d..1b0a48e 100644
@@ -13,11 +13,8 @@ obj-y :=	open.o read_write.o file_table.o super.o \
 		pnode.o drop_caches.o splice.o sync.o utimes.o \
 		stack.o
 
-ifeq ($(CONFIG_BLOCK),y)
-obj-y +=	buffer.o bio.o block_dev.o direct-io.o mpage.o ioprio.o
-else
-obj-y +=	no-block.o
-endif
+obj-y-ify-$(CONFIG_BLOCK) := buffer.o bio.o block_dev.o direct-io.o mpage.o ioprio.o
+obj-y-ifn-$(CONFIG_BLOCK) := no-block.o
 
 obj-$(CONFIG_INOTIFY)		+= inotify.o
 obj-$(CONFIG_INOTIFY_USER)	+= inotify_user.o

@@ -28,8 +25,7 @@ obj-$(CONFIG_TIMERFD)		+= timerfd.o
 obj-$(CONFIG_EVENTFD)		+= eventfd.o
 obj-$(CONFIG_COMPAT)		+= compat.o compat_ioctl.o
 
-nfsd-$(CONFIG_NFSD)		:= nfsctl.o
-obj-y				+= $(nfsd-y) $(nfsd-m)
+obj-y-if-$(CONFIG_NFSD)		+= nfsctl.o
 
 obj-$(CONFIG_BINFMT_AOUT)	+= binfmt_aout.o
 obj-$(CONFIG_BINFMT_EM86)	+= binfmt_em86.o

index 2a99983..5d7706e 100644
@@ -15,13 +15,9 @@ obj-$(CONFIG_STACKTRACE) += stacktrace.o
 obj-y += time/
 obj-$(CONFIG_DEBUG_MUTEXES) += mutex-debug.o
 obj-$(CONFIG_LOCKDEP) += lockdep.o
-ifeq ($(CONFIG_PROC_FS),y)
-obj-$(CONFIG_LOCKDEP) += lockdep_proc.o
-endif
+obj-$(CONFIG_LOCKDEP)-if-$(CONFIG_PROC_FS) += lockdep_proc.o
 obj-$(CONFIG_FUTEX) += futex.o
-ifeq ($(CONFIG_COMPAT),y)
-obj-$(CONFIG_FUTEX) += futex_compat.o
-endif
+obj-$(CONFIG_FUTEX)-if-$(CONFIG_COMPAT) += futex_compat.o
 obj-$(CONFIG_RT_MUTEXES) += rtmutex.o
 obj-$(CONFIG_DEBUG_RT_MUTEXES) += rtmutex-debug.o
 obj-$(CONFIG_RT_MUTEX_TESTER) += rtmutex-tester.o

index 245e33a..3a55991 100644
@@ -2,16 +2,15 @@
 # Makefile for the linux memory manager.
 #
 
-mmu-y			:= nommu.o
-mmu-$(CONFIG_MMU)	:= fremap.o highmem.o madvise.o memory.o mincore.o \
+obj-y-ifn-$(CONFIG_MMU) := nommu.o
+obj-$(CONFIG_MMU)	:= fremap.o highmem.o madvise.o memory.o mincore.o \
 			   mlock.o mmap.o mprotect.o mremap.o msync.o rmap.o \
 			   vmalloc.o
 
 obj-y			:= bootmem.o filemap.o mempool.o oom_kill.o fadvise.o \
 			   page_alloc.o page-writeback.o pdflush.o \
 			   readahead.o swap.o truncate.o vmscan.o \
-			   prio_tree.o util.o mmzone.o vmstat.o backing-dev.o \
-			   $(mmu-y)
+			   prio_tree.o util.o mmzone.o vmstat.o backing-dev.o
 
 obj-$(CONFIG_BOUNCE)	+= bounce.o
 obj-$(CONFIG_SWAP)	+= page_io.o swap_state.o swapfile.o thrash.o

index 3ead922..cc2bba0 100644
@@ -11,8 +11,6 @@ obj-$(CONFIG_SND_AOA) += aoa/
 # This one must be compilable even if sound is configured out
 obj-$(CONFIG_AC97_BUS) += ac97_bus.o
 
-ifeq ($(CONFIG_SND),y)
-  obj-y += last.o
-endif
+objy-ify-$(CONFIG_SND) += last.o
 
 soundcore-objs  := sound_core.o

index 402e2b4..1e676d9 100644
@@ -4,9 +4,7 @@
 #
 
 obj-$(CONFIG_SND) += instr/
-ifeq ($(CONFIG_SND_SEQUENCER_OSS),y)
-  obj-$(CONFIG_SND_SEQUENCER) += oss/
-endif
+obj-$(CONFIG_SND_SEQUENCER)-ify-$(CONFIG_SND_SEQUENCER_OSS) += oss/
 
 snd-seq-device-objs := seq_device.o
 snd-seq-objs := seq.o seq_lock.o seq_clientmgr.o seq_memory.o seq_queue.o \

@@ -28,9 +26,8 @@ snd-seq-virmidi-objs := seq_virmidi.o
 sequencer = $(if $(subst y,,$(CONFIG_SND_SEQUENCER)),$(if $(1),m),$(if $(CONFIG_SND_SEQUENCER),$(1)))
 
 obj-$(CONFIG_SND_SEQUENCER) += snd-seq.o snd-seq-device.o
-ifeq ($(CONFIG_SND_SEQUENCER_OSS),y)
-obj-$(CONFIG_SND_SEQUENCER) += snd-seq-midi-event.o
-endif
+
+obj-$(CONFIG_SND_SEQUENCER)-ify-$(CONFIG_SND_SEQUENCER_OSS) += snd-seq-midi-event.o
 obj-$(CONFIG_SND_SEQ_DUMMY) += snd-seq-dummy.o
 
 # Toplevel Module Dependency

index 45902d4..521708f 100644
@@ -7,9 +7,7 @@ snd-i2c-objs := i2c.o
 snd-cs8427-objs := cs8427.o
 snd-tea6330t-objs := tea6330t.o
 
-ifeq ($(subst m,y,$(CONFIG_L3)),y)
-  obj-$(CONFIG_L3) += l3/
-endif
+obj-$(CONFIG_L3)-ifm-$(CONFIG_L3) += l3/
 
 obj-$(CONFIG_SND) += other/
 

index 556e669..4681c46 100644
@@ -29,10 +29,8 @@ obj-$(CONFIG_SND_SB8) += snd-sb8.o
 obj-$(CONFIG_SND_SB16) += snd-sb16.o
 obj-$(CONFIG_SND_SBAWE) += snd-sbawe.o
 obj-$(CONFIG_SND_ES968) += snd-es968.o
-ifeq ($(CONFIG_SND_SB16_CSP),y)
-  obj-$(CONFIG_SND_SB16) += snd-sb16-csp.o
-  obj-$(CONFIG_SND_SBAWE) += snd-sb16-csp.o
-endif
+obj-$(CONFIG_SND_SB16)-ify-$(CONFIG_SND_SB16_CSP)  += snd-sb16-csp.o
+obj-$(CONFIG_SND_SBAWE)-ify-$(CONFIG_SND_SB16_CSP) += snd-sb16-csp.o
 obj-$(call sequencer,$(CONFIG_SND_SBAWE)) += snd-emu8000-synth.o
 
 obj-m := $(sort $(obj-m))

index 1200670..de54e00 100644
@@ -29,9 +29,7 @@ obj-$(CONFIG_SOUND_VIDC)	+= vidc_mod.o
 obj-$(CONFIG_SOUND_WAVEARTIST)	+= waveartist.o
 
 obj-$(CONFIG_SOUND_VIA82CXXX)	+= via82cxxx_audio.o ac97_codec.o
-ifeq ($(CONFIG_MIDI_VIA82CXXX),y)
-  obj-$(CONFIG_SOUND_VIA82CXXX) += sound.o uart401.o
-endif
+obj-$(CONFIG_SOUND_VIA82CXXX)-ify-$(CONFIG_MIDI_VIA82CXXX) += sound.o uart401.o
 obj-$(CONFIG_SOUND_MSNDCLAS)	+= msnd.o msnd_classic.o
 obj-$(CONFIG_SOUND_MSNDPIN)	+= msnd.o msnd_pinnacle.o
 obj-$(CONFIG_SOUND_VWSND)	+= vwsnd.o

IMHO for people who are not kbuild junkies the pattern is more clear with the current syntax.

But you should better ask some guinea pigs who have not already seen as many kernel Makefiles as I have...

Some of the cases have the following pattern:

config X86_POWERNOW_K8_ACPI

        bool
        depends on X86_POWERNOW_K8 && ACPI_PROCESSOR
        depends on !(X86_POWERNOW_K8 = y && ACPI_PROCESSOR = m)
        default y

Your suggested syntax has to be enhanced with three additional variables for handling such cases.

The complicated cases can be handled either in kconfig or in kbuild, and I think kconfig is the better place for them:

Handling it in kconfig has the advantage that we also get a variable we can use in C code.

For all the *-if-m and *-m-if{,n}-* cases this is a huge advantage over having to express the whole dependency in each #if.

Compare
#if defined(CONFIG_ACPI_PROCESSOR) || (defined(CONFIG_ACPI_PROCESSOR_MODULE && defined(MODULE))
with
#ifdef CONFIG_X86_POWERNOW_K8_ACPI

cu
Adrian

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed

Target group are for kernel developers and people who have actually read Documentation/kbuild/makefiles.txt

But point taken - this is too 'magic'.
I will try to think of something that looks a bit more straightforward.

The kbuild enhancements are for the cases where it makes less sense to express this in Kconfig.
They are not thought as replacing Kconfig dependencies in any way.

I will try to come up with a new proposal later today.

Sam

Hi Sam,

ACK. Also ACK for asflags, if done the same way :-)

Yes, but please bear in mind, what the developers are trying to express in these cases.

This is "feature FOO of module BAR" where the feature itself cannot be a module. The composition scheme described in section 3.3 is at least equally useful. And that is used today. Maybe the documentation of that scheme is not prominent enough :-)

This is the only one needed, because it is cumbersome to express negative rules in kbuild to include stubs (e.g. nommu stuff). But again this can be done with composition rules right now, but is order dependent. If we could get rid of this requirement, I would be happy already. So kbuild is just lacking an "else" clause here.

Best Regards

Ingo Oeser

Perhaps this would work?

ifeq (${CONFIG_NO_MMU},)
obj-m += magic_mmu.o
endif

Sometimes substitution looks acceptable: