Re: [PATCH v2 3/5] kbuild: Use relative path for $(objtree)

[Randy Dunlap <rdunlap@infradead.org>]

On 06/09/14 15:14, Michal Marek wrote:
> Dne 9.6.2014 23:23, Randy Dunlap napsal(a):
>> On 06/05/14 08:56, Michal Marek wrote:
>>> On Wed, Jun 04, 2014 at 03:12:33PM +0200, Michal Marek wrote:
>>>> On 2014-06-04 11:43, Michal Marek wrote:
>>>>> On 2014-06-04 11:03, Richard Weinberger wrote:
>>>>>> To reproduce run:
>>>>>> make defconfig ARCH=um O=/mnt/o && make linux ARCH=um O=/mnt/
>>>>>>
>>>>>> If there is anything in UML which needs fixing, please tell. :-)
>>>>>
>>>>> I'll have a look, thanks for the report.
>>>>
>>>> Findings so far: For some reason, syscalls_32.h is generated in the
>>>> source tree (which is wrong) and syscalls_64.h is not generated at all.
>>>> Looking further.
>>>
>>> Can you try the below patch? The same pattern is used in the rules for
>>> tools/ and tools/% in the main Makefile, need to look into that as well.
>>> But UML should work now.
>>>
>>> Michal
>>>
>>> From d4bc590f8716f7dde6b7bca319097ac30a8cb0b4 Mon Sep 17 00:00:00 2001
>>> From: Michal Marek <mmarek@suse.cz>
>>> Date: Thu, 5 Jun 2014 17:44:44 +0200
>>> Subject: [PATCH] um: Fix for relative objtree when generating x86 headers
>>>
>>> In an O= build, rely on the generated Makefile to call the main Makefile
>>> properly. When building in the source tree, we do not need to specify
>>> the -C and O= either. This fixes the problem when $(objtree) is a
>>> relative path and the -C changes the directory.
>>>
>>> Reported-by: Richard Weinberger <richard.weinberger@gmail.com>
>>> Signed-off-by: Michal Marek <mmarek@suse.cz>
>>> ---
>>>  arch/um/Makefile | 3 +--
>>>  1 file changed, 1 insertion(+), 2 deletions(-)
>>>
>>> diff --git a/arch/um/Makefile b/arch/um/Makefile
>>> index 36e658a..e4b1a96 100644
>>> --- a/arch/um/Makefile
>>> +++ b/arch/um/Makefile
>>> @@ -111,8 +111,7 @@ endef
>>>  KBUILD_KCONFIG := $(HOST_DIR)/um/Kconfig
>>>  
>>>  archheaders:
>>> -	$(Q)$(MAKE) -C '$(srctree)' KBUILD_SRC= \
>>> -		ARCH=$(HEADER_ARCH) O='$(objtree)' archheaders
>>> +	$(Q)$(MAKE) KBUILD_SRC= ARCH=$(HEADER_ARCH) archheaders
>>>  
>>>  archprepare: include/generated/user_constants.h
>>>  
>>>
>>
>> I still get this build error when building uml for i386:
>>
>>   CC      arch/x86/um/user-offsets.s
>> ../arch/x86/um/user-offsets.c:14:29: fatal error: asm/syscalls_32.h: No such file or directory
>> compilation terminated.
>> make[2]: *** [arch/x86/um/user-offsets.s] Error 1
> 
> Thanks for testing the patch. I cannot reproduce it though:
> 
>   make ARCH=um SUBARCH=i386 O=/dev/shm/li defconfig
>   make ARCH=um SUBARCH=i386 O=/dev/shm/li

That doesn't work for me either.  Must be a difference somewhere else.
I am using linux-next of 20140606 (latest that I know of) with only your
recent patch applied to it.

> works fine (at least it gets to the point when it starts compiling the
> actual kernel source). This is the previous kbuild/kbuild branch (commit
> 9da0763) with the above fix applied.
> 
> Also, you pointed out i386 -- Does it mean that the fix worked for you
> on x86_64?

Yes, it did work on x86_64 for some reason.


-- 
~Randy