[PATCH] ARM: kbuild: Fix forced rebuild after 'make dtbs'

[PATCH] ARM: kbuild: Fix forced rebuild after 'make dtbs'

[Jason Cooper]

After this patch:

  f4d4ffc03efc kbuild: dtbs_install: new make target

was added the kernel tree, Linus Walleij noticed that 'make dtbs' forced
a following 'make zImage' to rebuild the entire tree, even though
nothing had changed.  His report:

  After this patch a while back I have observed the following behaviour
  of the kernel build:

  make zImage
  make zImage
  -> incremental build, just relink

  make zImage
  make dtbs
  make zImage
  -> The whole kernel gets rebuilt

  So now if I happen to recompile my device trees, I suddenly want
  the entire zImage to be rebuilt to? It's by definition not changes
  that affect the kernel build :-(

  I noticed this because my build scripts calls make dtbs && make
  zImage, and started to rebuild absolutely everything all the time.

To fix this, we make only the dtbs_install target depend on the prepare
target.  It's needed to make sure KERNELVERSION is calculated prior to
installing.

Fixes: f4d4ffc03efc: ("kbuild: dtbs_install: new make target")
Reported-by: Linus Walleij <linus.walleij@linaro.org>
Tested-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Jason Cooper <jason@lakedaemon.net>
---
Note: I've no idea if this is a 100% correct solution or not.  I know it's
definitely better than what we have currently.  If there is another way to
guarantee KERNELVERSION is set other than depending on 'prepare', I'd love to
hear it.

thx,

Jason.

 arch/arm/Makefile | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/arm/Makefile b/arch/arm/Makefile
index 034a94904d69..ae1c278128f8 100644
--- a/arch/arm/Makefile
+++ b/arch/arm/Makefile
@@ -312,7 +312,9 @@ $(INSTALL_TARGETS):
 	$(Q)$(MAKE) $(build)=$(boot)/dts MACHINE=$(MACHINE) $(boot)/dts/$@
 
 PHONY += dtbs dtbs_install
-dtbs dtbs_install: prepare scripts
+dtbs: scripts
+	$(Q)$(MAKE) $(build)=$(boot)/dts MACHINE=$(MACHINE) $@
+dtbs_install: prepare scripts
 	$(Q)$(MAKE) $(build)=$(boot)/dts MACHINE=$(MACHINE) $@
 
 # We use MRPROPER_FILES and CLEAN_FILES now
-- 
2.1.3

[PATCH] ARM: kbuild: Fix forced rebuild after 'make dtbs'

[Michal Marek]

Dne 2.11.2014 v 21:52 Jason Cooper napsal(a):
> After this patch:
> 
>   f4d4ffc03efc kbuild: dtbs_install: new make target
> 
> was added the kernel tree, Linus Walleij noticed that 'make dtbs' forced
> a following 'make zImage' to rebuild the entire tree, even though
> nothing had changed.  His report:
> 
>   After this patch a while back I have observed the following behaviour
>   of the kernel build:
> 
>   make zImage
>   make zImage
>   -> incremental build, just relink
> 
>   make zImage
>   make dtbs
>   make zImage
>   -> The whole kernel gets rebuilt
> 
>   So now if I happen to recompile my device trees, I suddenly want
>   the entire zImage to be rebuilt to? It's by definition not changes
>   that affect the kernel build :-(
> 
>   I noticed this because my build scripts calls make dtbs && make
>   zImage, and started to rebuild absolutely everything all the time.
> 
> To fix this, we make only the dtbs_install target depend on the prepare
> target.  It's needed to make sure KERNELVERSION is calculated prior to
> installing.

If a mere 'make prepare' causes a rebuild of the whole kernel, then
there is something fishy in the ARM Makefiles. However, if you only need
the KERNELRELEASE variable, then you do not need to depend on prepare.
The main Makefile makes sure that silentoldconfig is ran and therefore
KERNELRELEASE set for all targets except make *config.

Michal

[PATCH] ARM: kbuild: Fix forced rebuild after 'make dtbs'

[Russell King - ARM Linux]

On Tue, Nov 25, 2014 at 10:41:35PM +0100, Michal Marek wrote:
> Dne 2.11.2014 v 21:52 Jason Cooper napsal(a):
> > After this patch:
> > 
> >   f4d4ffc03efc kbuild: dtbs_install: new make target
> > 
> > was added the kernel tree, Linus Walleij noticed that 'make dtbs' forced
> > a following 'make zImage' to rebuild the entire tree, even though
> > nothing had changed.  His report:
> > 
> >   After this patch a while back I have observed the following behaviour
> >   of the kernel build:
> > 
> >   make zImage
> >   make zImage
> >   -> incremental build, just relink
> > 
> >   make zImage
> >   make dtbs
> >   make zImage
> >   -> The whole kernel gets rebuilt
> > 
> >   So now if I happen to recompile my device trees, I suddenly want
> >   the entire zImage to be rebuilt to? It's by definition not changes
> >   that affect the kernel build :-(
> > 
> >   I noticed this because my build scripts calls make dtbs && make
> >   zImage, and started to rebuild absolutely everything all the time.
> > 
> > To fix this, we make only the dtbs_install target depend on the prepare
> > target.  It's needed to make sure KERNELVERSION is calculated prior to
> > installing.
> 
> If a mere 'make prepare' causes a rebuild of the whole kernel, then
> there is something fishy in the ARM Makefiles. However, if you only need
> the KERNELRELEASE variable, then you do not need to depend on prepare.
> The main Makefile makes sure that silentoldconfig is ran and therefore
> KERNELRELEASE set for all targets except make *config.

Running make prepare and then re-running a build doesn't rebuild anything
for me.  However, I always build with O=  Also works if I do a make
dtbs too.  So, everything seems to work as expected here.

I think this needs a bit more debugging to see why the whole kernel is
being rebuilt - the kernel build system has methods to tell you why stuff
is being built, which would be a good place to start.

-- 
FTTC broadband for 0.8mile line: currently at 9.5Mbps down 400kbps up
according to speedtest.net.

[PATCH] ARM: kbuild: Fix forced rebuild after 'make dtbs'

[Jason Cooper]

On Tue, Nov 25, 2014 at 11:51:50PM +0000, Russell King - ARM Linux wrote:
> On Tue, Nov 25, 2014 at 10:41:35PM +0100, Michal Marek wrote:
> > Dne 2.11.2014 v 21:52 Jason Cooper napsal(a):
> > > After this patch:
> > > 
> > >   f4d4ffc03efc kbuild: dtbs_install: new make target
> > > 
> > > was added the kernel tree, Linus Walleij noticed that 'make dtbs' forced
> > > a following 'make zImage' to rebuild the entire tree, even though
> > > nothing had changed.  His report:
> > > 
> > >   After this patch a while back I have observed the following behaviour
> > >   of the kernel build:
> > > 
> > >   make zImage
> > >   make zImage
> > >   -> incremental build, just relink
> > > 
> > >   make zImage
> > >   make dtbs
> > >   make zImage
> > >   -> The whole kernel gets rebuilt
> > > 
> > >   So now if I happen to recompile my device trees, I suddenly want
> > >   the entire zImage to be rebuilt to? It's by definition not changes
> > >   that affect the kernel build :-(
> > > 
> > >   I noticed this because my build scripts calls make dtbs && make
> > >   zImage, and started to rebuild absolutely everything all the time.
> > > 
> > > To fix this, we make only the dtbs_install target depend on the prepare
> > > target.  It's needed to make sure KERNELVERSION is calculated prior to
> > > installing.
> > 
> > If a mere 'make prepare' causes a rebuild of the whole kernel, then
> > there is something fishy in the ARM Makefiles. However, if you only need
> > the KERNELRELEASE variable, then you do not need to depend on prepare.
> > The main Makefile makes sure that silentoldconfig is ran and therefore
> > KERNELRELEASE set for all targets except make *config.
> 
> Running make prepare and then re-running a build doesn't rebuild anything
> for me.  However, I always build with O=  Also works if I do a make
> dtbs too.  So, everything seems to work as expected here.
> 
> I think this needs a bit more debugging to see why the whole kernel is
> being rebuilt - the kernel build system has methods to tell you why stuff
> is being built, which would be a good place to start.

Linus, since you were the one experiencing the regression, care to take
a whack at that?

thx,

Jason.

[PATCH] ARM: kbuild: Fix forced rebuild after 'make dtbs'

[Linus Walleij]

On Wed, Nov 26, 2014 at 1:57 AM, Jason Cooper <jason@lakedaemon.net> wrote:
> On Tue, Nov 25, 2014 at 11:51:50PM +0000, Russell King - ARM Linux wrote:

>> I think this needs a bit more debugging to see why the whole kernel is
>> being rebuilt - the kernel build system has methods to tell you why stuff
>> is being built, which would be a good place to start.
>
> Linus, since you were the one experiencing the regression, care to take
> a whack at that?

OK this full example should be possible to repeat for anyone
with a cross compiler, and uses a separate build dir, and yeah
there was something fishy: the bug only appears if you specify
CONFIG_DEBUG_SECTION_MISMATCH=y:

make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 u300_defconfig
make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 zImage
make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 zImage

-> no rebuild, just relink

make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 u300_defconfig
make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 zImage CONFIG_DEBUG_SECTION_MISMATCH=y
make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 zImage CONFIG_DEBUG_SECTION_MISMATCH=y

-> no rebuild, just relink

make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 u300_defconfig
make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 zImage
make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 dtbs
make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 zImage

-> no rebuild, just relink

make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 u300_defconfig
make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 zImage CONFIG_DEBUG_SECTION_MISMATCH=y
make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 dtbs
make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 zImage CONFIG_DEBUG_SECTION_MISMATCH=y

-> Rebuilds everything.

This doesn't occur before the offending commit. So it only happens
when specifying extra
environment variables on the command line.

So I can get rid of the rebuilding by making my scripts less
strict by removing CONFIG_DEBUG_SECTION_MISMATCH=y.

But is this really correct?

Two last examples with more detail:

make[1]: Entering directory `/home/linus/linux/build-u300'
  CHK     include/config/kernel.release
  GEN     ./Makefile
  CHK     include/generated/uapi/linux/version.h
  CHK     include/generated/utsrelease.h
  Using .. as source for kernel
make[2]: `include/generated/mach-types.h' is up to date.
  CALL    ../scripts/checksyscalls.sh - due to target missing
  CHK     include/generated/compile.h
  Kernel: arch/arm/boot/Image is ready
  Kernel: arch/arm/boot/zImage is ready
make[1]: Leaving directory `/home/linus/linux/build-u300'

make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
KBUILD_OUTPUT=build-u300 zImage CONFIG_DEBUG_SECTION_MISMATCH=y V=2

make[1]: Entering directory `/home/linus/linux/build-u300'
  CHK     include/config/kernel.release
  GEN     ./Makefile
  CHK     include/generated/uapi/linux/version.h
  CHK     include/generated/utsrelease.h
  Using .. as source for kernel
  CC      scripts/mod/empty.o - due to command line change
  CC      scripts/mod/devicetable-offsets.s - due to command line change
  MKELF   scripts/mod/elfconfig.h - due to: scripts/mod/empty.o
  HOSTCC  scripts/mod/modpost.o - due to: scripts/mod/elfconfig.h
  HOSTCC  scripts/mod/sumversion.o - due to: scripts/mod/elfconfig.h
  GEN     scripts/mod/devicetable-offsets.h - due to:
scripts/mod/devicetable-offsets.s
  HOSTCC  scripts/mod/file2alias.o - due to: scripts/mod/elfconfig.h
scripts/mod/devicetable-offsets.h
  HOSTLD  scripts/mod/modpost - due to: scripts/mod/modpost.o
scripts/mod/file2alias.o scripts/mod/sumversion.o
make[2]: `include/generated/mach-types.h' is up to date.
  CC      kernel/bounds.s - due to command line change
  GEN     include/generated/bounds.h - due to: kernel/bounds.s
  CC      arch/arm/kernel/asm-offsets.s - due to: include/generated/bounds.h
  GEN     include/generated/asm-offsets.h - due to:
arch/arm/kernel/asm-offsets.s
  CALL    ../scripts/checksyscalls.sh - due to target missing
  CC      init/main.o - due to: include/generated/bounds.h
  CC      arch/arm/common/firmware.o - due to: include/generated/bounds.h
  CC      arch/arm/mm/dma-mapping.o - due to: include/generated/bounds.h
  CC      arch/arm/kernel/elf.o - due to: include/generated/bounds.h
  CC      arch/arm/mach-u300/core.o - due to: include/generated/bounds.h
  AS      arch/arm/kernel/entry-common.o - due to:
include/generated/asm-offsets.h
  LD      arch/arm/common/built-in.o - due to: arch/arm/common/firmware.o
  CHK     include/generated/compile.h
  CC      arch/arm/kernel/irq.o - due to: include/generated/bounds.h
  CC      arch/arm/mach-u300/dummyspichip.o - due to: include/generated/bounds.h
  CC      init/do_mounts.o - due to: include/generated/bounds.h
  CC      arch/arm/mm/extable.o - due to: include/generated/bounds.h
  CC      arch/arm/kernel/opcodes.o - due to: include/generated/bounds.h
(...)

If I revert commit f4d4ffc03efc:

make[1]: Entering directory `/home/linus/linux/build-u300'
  CHK     include/config/kernel.release
  GEN     ./Makefile
  CHK     include/generated/uapi/linux/version.h
  Using .. as source for kernel
  CHK     include/generated/utsrelease.h
make[2]: `include/generated/mach-types.h' is up to date.
  CC      scripts/mod/empty.o - due to command line change
  CC      scripts/mod/devicetable-offsets.s - due to command line change
  CALL    ../scripts/checksyscalls.sh - due to target missing
  MKELF   scripts/mod/elfconfig.h - due to: scripts/mod/empty.o
  HOSTCC  scripts/mod/modpost.o - due to: scripts/mod/elfconfig.h
  HOSTCC  scripts/mod/sumversion.o - due to: scripts/mod/elfconfig.h
  GEN     scripts/mod/devicetable-offsets.h - due to:
scripts/mod/devicetable-offsets.s
  HOSTCC  scripts/mod/file2alias.o - due to: scripts/mod/elfconfig.h
scripts/mod/devicetable-offsets.h
  HOSTLD  scripts/mod/modpost - due to: scripts/mod/modpost.o
scripts/mod/file2alias.o scripts/mod/sumversion.o
  CHK     include/generated/compile.h
  Kernel: arch/arm/boot/Image is ready
  Kernel: arch/arm/boot/zImage is ready
make[1]: Leaving directory `/home/linus/linux/build-u300'

So in both cases a changed command line is detected and
empty.o and devicetable-offsets.s is rebuilt.

But after the commit the rebuild goes on to rebuild
kernel/bounds.s which generates bounds.h which in
turn rebuilds the whole world.

I don't know why the behaviour changed like this :(

Yours,
Linus Walleij

[PATCH] ARM: kbuild: Fix forced rebuild after 'make dtbs'

[Russell King - ARM Linux]

On Wed, Nov 26, 2014 at 02:57:42PM +0100, Linus Walleij wrote:
> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
> KBUILD_OUTPUT=build-u300 u300_defconfig
> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
> KBUILD_OUTPUT=build-u300 zImage CONFIG_DEBUG_SECTION_MISMATCH=y
> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
> KBUILD_OUTPUT=build-u300 dtbs
> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
> KBUILD_OUTPUT=build-u300 zImage CONFIG_DEBUG_SECTION_MISMATCH=y
> 
> -> Rebuilds everything.
> 
> This doesn't occur before the offending commit. So it only happens
> when specifying extra environment variables on the command line.

I suspect if you also provide the CONFIG_DEBUG_SECTION_MISMATCH=y on the
dtbs line, everything will work properly.

The problem is that dtbs line executes a prepare, which I guess rebuilds
the bounds stuff.  So, the dtbs target results in it being rebuilt without
the section mismatch, and then you re-execute a make with it, causing
the bounds stuff to be rebuilt again.

-- 
FTTC broadband for 0.8mile line: currently at 9.5Mbps down 400kbps up
according to speedtest.net.

[PATCH] ARM: kbuild: Fix forced rebuild after 'make dtbs'

[Linus Walleij]

On Wed, Nov 26, 2014 at 3:41 PM, Russell King - ARM Linux
<linux@arm.linux.org.uk> wrote:
> On Wed, Nov 26, 2014 at 02:57:42PM +0100, Linus Walleij wrote:
>> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
>> KBUILD_OUTPUT=build-u300 u300_defconfig
>> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
>> KBUILD_OUTPUT=build-u300 zImage CONFIG_DEBUG_SECTION_MISMATCH=y
>> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
>> KBUILD_OUTPUT=build-u300 dtbs
>> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
>> KBUILD_OUTPUT=build-u300 zImage CONFIG_DEBUG_SECTION_MISMATCH=y
>>
>> -> Rebuilds everything.
>>
>> This doesn't occur before the offending commit. So it only happens
>> when specifying extra environment variables on the command line.
>
> I suspect if you also provide the CONFIG_DEBUG_SECTION_MISMATCH=y on the
> dtbs line, everything will work properly.

Yay, it works! :)

> The problem is that dtbs line executes a prepare, which I guess rebuilds
> the bounds stuff.  So, the dtbs target results in it being rebuilt without
> the section mismatch, and then you re-execute a make with it, causing
> the bounds stuff to be rebuilt again.

Yep. Not very intuitive to require passing section mismatch debug
flags to DTB rebuilding but whatever, it's not so bad I can't live with
it.

Strange that it worked before.

Yours,
Linus Walleij

[PATCH] ARM: kbuild: Fix forced rebuild after 'make dtbs'

[Michal Marek]

On 2014-11-28 15:10, Linus Walleij wrote:
> On Wed, Nov 26, 2014 at 3:41 PM, Russell King - ARM Linux
> <linux@arm.linux.org.uk> wrote:
>> On Wed, Nov 26, 2014 at 02:57:42PM +0100, Linus Walleij wrote:
>>> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
>>> KBUILD_OUTPUT=build-u300 u300_defconfig
>>> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
>>> KBUILD_OUTPUT=build-u300 zImage CONFIG_DEBUG_SECTION_MISMATCH=y
>>> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
>>> KBUILD_OUTPUT=build-u300 dtbs
>>> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
>>> KBUILD_OUTPUT=build-u300 zImage CONFIG_DEBUG_SECTION_MISMATCH=y
>>>
>>> -> Rebuilds everything.
>>>
>>> This doesn't occur before the offending commit. So it only happens
>>> when specifying extra environment variables on the command line.
>>
>> I suspect if you also provide the CONFIG_DEBUG_SECTION_MISMATCH=y on the
>> dtbs line, everything will work properly.
> 
> Yay, it works! :)
> 
>> The problem is that dtbs line executes a prepare, which I guess rebuilds
>> the bounds stuff.  So, the dtbs target results in it being rebuilt without
>> the section mismatch, and then you re-execute a make with it, causing
>> the bounds stuff to be rebuilt again.
> 
> Yep. Not very intuitive to require passing section mismatch debug
> flags to DTB rebuilding but whatever, it's not so bad I can't live with
> it.

Still, the dependency on the 'prepare' target is superfluous, if the
only requirement is that $(KERNELRELEASE) be set properly.

Michal

[PATCH] ARM: kbuild: Fix forced rebuild after 'make dtbs'

[Grant Likely]

On Mon, 01 Dec 2014 14:55:48 +0100
, Michal Marek <mmarek@suse.cz>
 wrote:
> On 2014-11-28 15:10, Linus Walleij wrote:
> > On Wed, Nov 26, 2014 at 3:41 PM, Russell King - ARM Linux
> > <linux@arm.linux.org.uk> wrote:
> >> On Wed, Nov 26, 2014 at 02:57:42PM +0100, Linus Walleij wrote:
> >>> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
> >>> KBUILD_OUTPUT=build-u300 u300_defconfig
> >>> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
> >>> KBUILD_OUTPUT=build-u300 zImage CONFIG_DEBUG_SECTION_MISMATCH=y
> >>> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
> >>> KBUILD_OUTPUT=build-u300 dtbs
> >>> make -f Makefile -j5 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
> >>> KBUILD_OUTPUT=build-u300 zImage CONFIG_DEBUG_SECTION_MISMATCH=y
> >>>
> >>> -> Rebuilds everything.
> >>>
> >>> This doesn't occur before the offending commit. So it only happens
> >>> when specifying extra environment variables on the command line.
> >>
> >> I suspect if you also provide the CONFIG_DEBUG_SECTION_MISMATCH=y on the
> >> dtbs line, everything will work properly.
> > 
> > Yay, it works! :)
> > 
> >> The problem is that dtbs line executes a prepare, which I guess rebuilds
> >> the bounds stuff.  So, the dtbs target results in it being rebuilt without
> >> the section mismatch, and then you re-execute a make with it, causing
> >> the bounds stuff to be rebuilt again.
> > 
> > Yep. Not very intuitive to require passing section mismatch debug
> > flags to DTB rebuilding but whatever, it's not so bad I can't live with
> > it.
> 
> Still, the dependency on the 'prepare' target is superfluous, if the
> only requirement is that $(KERNELRELEASE) be set properly.

So, I've lost track of where things stand on this issue. What is the
current proposed fix? Make dtbs depend on 'scripts' and dtbs_install
depend on 'prepare scripts'? Or is there a better way to make sure
KERNELRELEASE is set for the dtbs_install target?

g.

[PATCH] ARM: kbuild: Fix forced rebuild after 'make dtbs'

[Russell King - ARM Linux]

On Thu, Dec 04, 2014 at 11:22:01AM +0000, Grant Likely wrote:
> On Mon, 01 Dec 2014 14:55:48 +0100, Michal Marek <mmarek@suse.cz> wrote:
> > Still, the dependency on the 'prepare' target is superfluous, if the
> > only requirement is that $(KERNELRELEASE) be set properly.
> 
> So, I've lost track of where things stand on this issue. What is the
> current proposed fix? Make dtbs depend on 'scripts' and dtbs_install
> depend on 'prepare scripts'? Or is there a better way to make sure
> KERNELRELEASE is set for the dtbs_install target?

Marek's first reply in the thread contains the solution:

  if you only need
  the KERNELRELEASE variable, then you do not need to depend on prepare.
  The main Makefile makes sure that silentoldconfig is ran and therefore
  KERNELRELEASE set for all targets except make *config.

which seems to suggest that "prepare" is not needed for either target.

-- 
FTTC broadband for 0.8mile line: currently at 9.5Mbps down 400kbps up
according to speedtest.net.

[PATCH] ARM: kbuild: Fix forced rebuild after 'make dtbs'

[Michal Marek]

On 2014-11-26 00:51, Russell King - ARM Linux wrote:
> On Tue, Nov 25, 2014 at 10:41:35PM +0100, Michal Marek wrote:
>> Dne 2.11.2014 v 21:52 Jason Cooper napsal(a):
>>> After this patch:
>>>
>>>   f4d4ffc03efc kbuild: dtbs_install: new make target
>>>
>>> was added the kernel tree, Linus Walleij noticed that 'make dtbs' forced
>>> a following 'make zImage' to rebuild the entire tree, even though
>>> nothing had changed.  His report:
>>>
>>>   After this patch a while back I have observed the following behaviour
>>>   of the kernel build:
>>>
>>>   make zImage
>>>   make zImage
>>>   -> incremental build, just relink
>>>
>>>   make zImage
>>>   make dtbs
>>>   make zImage
>>>   -> The whole kernel gets rebuilt
>>>
>>>   So now if I happen to recompile my device trees, I suddenly want
>>>   the entire zImage to be rebuilt to? It's by definition not changes
>>>   that affect the kernel build :-(
>>>
>>>   I noticed this because my build scripts calls make dtbs && make
>>>   zImage, and started to rebuild absolutely everything all the time.
>>>
>>> To fix this, we make only the dtbs_install target depend on the prepare
>>> target.  It's needed to make sure KERNELVERSION is calculated prior to
>>> installing.
>>
>> If a mere 'make prepare' causes a rebuild of the whole kernel, then
>> there is something fishy in the ARM Makefiles. However, if you only need
>> the KERNELRELEASE variable, then you do not need to depend on prepare.
>> The main Makefile makes sure that silentoldconfig is ran and therefore
>> KERNELRELEASE set for all targets except make *config.
> 
> Running make prepare and then re-running a build doesn't rebuild anything
> for me.  However, I always build with O=  Also works if I do a make
> dtbs too.  So, everything seems to work as expected here.

Good :).


> I think this needs a bit more debugging to see why the whole kernel is
> being rebuilt - the kernel build system has methods to tell you why stuff
> is being built, which would be a good place to start.

Right, this is make V=2.

Michal