svn commit: r255187 - in head/sys: conf crypto/aesni modules/aesni

[Bjoern A. Zeeb]

On Tue, 17 Sep 2013, John-Mark Gurney wrote:

> Bjoern A. Zeeb wrote this message on Tue, Sep 17, 2013 at 19:26 +0000:
>> On Tue, 3 Sep 2013, John-Mark Gurney wrote:
>>> + *
>>> + */
>>> +
>>> +#include <wmmintrin.h>
>>> +
>>
>>
>> This pulls in a header from user space, in fact from clang, and if
>> cross-building without building clang this header file is not
>> available and a buildkernel is failing.
>>
>> (paths shortend)
>> In file included from /sys/modules/aesni/../../crypto/aesni/aesni_wrap.c:40:
>> /sys/modules/aesni/../../crypto/aesni/aesencdec.h:30:10: fatal error:
>> 'wmmintrin.h' file not found
>> #include <wmmintrin.h> ^
>> 1 error generated.
>> --- aesni_wrap.o ---
>
> More details please...  cross building on what to what?  I will admit
> I haven't tried to build from arm or other non-x86 platform to an
> x86.. but I am pretty sure I tested cross building i386 on an amd64
> machine...
>
> Are you building w/ gcc?  it isn't clear from your message...  If you
> are, is your toolchain more recent than the 3rd?
>
> Also, all I did was turn off -nostdinc, I didn't add any other paths,
> which sounds like that the cross build environment isn't built
> properly if it can't access userland headers...

It should not access the base system headers, it needs the ones in
obj/ but these are not there when building WITHOUT_CLANG=
as that header file in my head checkout only exists in the clang source
(contrib/llvm/tools/clang/lib/Headers/wmmintrin.h).

The SVN revision is about r255569 or the following commit from Saturday.

And yes I am also building without buildtools using XCC= XCPP= XCXX=
using clang from the installed base system, which means the pollution from
that early stage doesn't happen either -- more pieces seem to rely on
this:(

And this is building i386 on i386 just reducing the compile time by 7
orders of magnitude not building clang twice but zero times as I don't
need it for the output install and the one installed in base is
exactly the same I'd build again for bootstrap so avoiding this.

I assume on an entirely clang-free system building with just gcc and
never seen clang the results would be the same as the header file
will never be available anywhere.

-- 
Bjoern A. Zeeb                             ????????? ??? ??????? ??????:
'??? ??? ???? ??????  ??????? ?? ?? ??????? ??????? ??? ????? ????? ????
?????? ?? ????? ????',  ????????? ?????????, "??? ????? ?? ?????", ?.???