On 11/08/2011 12:25 AM, Michal Marek wrote:
> On 8.11.2011 00:33, Ralf Baechle wrote:
>> On Mon, Nov 07, 2011 at 10:19:00PM +0100, Michal Marek wrote:
>>
>>> Wild guess - does this patch help?
>>>
>>>
>>> diff --git a/Kbuild b/Kbuild
>>> index 4caab4f..77c191a 100644
>>> --- a/Kbuild
>>> +++ b/Kbuild
>>> @@ -94,7 +94,7 @@ targets += missing-syscalls
>>>   quiet_cmd_syscalls = CALL    $<
>>>         cmd_syscalls = $(CONFIG_SHELL) $<  $(CC) $(c_flags)
>>>
>>> -missing-syscalls: scripts/checksyscalls.sh $(offsets-file) FORCE
>>> +missing-syscalls: scripts/checksyscalls.sh $(offsets-file) $(bounds-file) FORCE
>>>   	$(call cmd,syscalls)
>>>
>>>   # Keep these two files during make clean
>>
>> No, it didn't.
>>
>>> If not, please attach logs of make V=1 with clean Linus' tree and with
>>> 5f7efb4c6da9f90cb306923ced2a6494d065a595 reverted.
>>
>> $ git checkout 31555213f03bca37d2c02e10946296052f4ecfcd
>> $ git revert 5f7efb4c6da9f90cb306923ced2a6494d065a595
>> $ make ARCH=mips ip27_defconfig
>> $ make ARCH=mips V=1 2>&1 | tee log
>
> Thanks, can you also post a log without the revert?
>

The problem is that compiler options meant to be used only for the 
compiling done by scripts/checksyscalls.sh are now leaking into the 
compilation of other parts of the kernel (asm-offsets.c), where they 
wreak havoc.

Something like the attached is what I think needs to be done.

David Daney