Re: some kernel headers broken in current git ?

[Gabriel C <nix.or.die@googlemail.com>]

H. Peter Anvin wrote:
> Gabriel C wrote:
>> Hi,
>>
>> usually I'll wait for rc1 and test compile external module to see which are broken and what need fixing
>> but while I need virtualbox for some tests I test compile it on current git and it failed badly.
>>
>> Maybe something is missing from x86 merge ?
>>
>> Here is what I get :
>>
>> ...
>>
>> /linux/memobj-r0drv-linux.c
>> In file included from /lib/modules/2.6.23-g4fa4d23f-dirty/build/include/asm/atomic_32.h:265,
>>                  from /lib/modules/2.6.23-g4fa4d23f-dirty/build/include/asm/atomic.h:2,
>>                  from /lib/modules/2.6.23-g4fa4d23f-dirty/build/include/asm/spinlock_32.h:4,
>>                  from /lib/modules/2.6.23-g4fa4d23f-dirty/build/include/asm/spinlock.h:2,
>>                  from /lib/modules/2.6.23-g4fa4d23f-dirty/build/include/linux/spinlock.h:87,
>>                  from /work/crazy/VBox/stable/virtualbox/src/VirtualBox-1.5.2_OSE/src/VBox/Runtime/r0drv/linux/the-linux-kernel.h:53,
>>                  from /work/crazy/VBox/stable/virtualbox/src/VirtualBox-1.5.2_OSE/src/VBox/Runtime/r0drv/linux/alloc-r0drv-linux.c:22:
>> /lib/modules/2.6.23-g4fa4d23f-dirty/build/include/asm-generic/atomic.h:23: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'atomic_long_t'
> 
> I have been unable to make heads or tails of the maze of twisty 
> dependencies that VirtualBox wants, but the fact that it gets to line 23 
> of <asm-generic/atomic.h> means it has gotten past:
> 
>      21  #if BITS_PER_LONG == 64
>      22
>      23  typedef atomic64_t atomic_long_t;
> 
> BITS_PER_LONG was originally set in <asm/types.h>:
> 
>      39  #ifdef CONFIG_X86_32
>      40  # define BITS_PER_LONG 32
>      41  #else
>      42  # define BITS_PER_LONG 64
>      43  #endif

User land does not know anything about 'CONFIG_X86_32' right ?

I just changed some things manually to test , s/CONFIG_X86_32/__i386__/ in asm/types.h and worked fine 
but subarch headers are still not included probably for the same reason. ( manually changed to test as well )

After doing so , the part filed compiled.

but next error a bit later :

..


In file included from /work/crazy/VBox/stable/virtualbox/src/VirtualBox-1.5.2_OSE/include/iprt/string.h:25,
                 from /work/crazy/VBox/stable/virtualbox/src/VirtualBox-1.5.2_OSE/src/VBox/Runtime/assert.cpp:24:
/lib/modules/2.6.23-rc0/build/include/linux/string.h:70: error: declaration of C function '__kernel_size_t strlen(const char*)' conflicts with
/lib/modules/2.6.23-rc0/build/include/asm/string_64.h:53: error: previous declaration 'size_t strlen(const char*)' here
/lib/modules/2.6.23-rc0/build/include/linux/string.h:101: error: declaration of C function 'int memcmp(const void*, const void*, __kernel_size_t)' conflicts with
/lib/modules/2.6.23-rc0/build/include/asm/string_64.h:52: error: previous declaration 'int memcmp(const void*, const void*, size_t)' here

..

And again both 32/64 things are defined at the same time.


> 
> The most obvious reason for failure is that the symbol CONFIG_X86_32 
> isn't being defined where expected.  From that point on everything goes 
> to hell.

Yes you are right but as I said above I don't think user land understands CONFIG_X86_32.

> 
> Have you done "make oldconfig && make prepare" in your kernel tree since 
> you last updated it?

Yes I always do that.

> 
> 	-hpa
> 
> 

Gabriel