Messages in this thread | | | Subject | Re: [PATCH] ARM: fix vdsomunge depends on glibc specific byteswap.h | From | "H. Nikolaus Schaller" <> | Date | Wed, 30 Sep 2015 19:25:30 +0200 |
| |
Hi,
Am 30.09.2015 um 18:17 schrieb Ard Biesheuvel <ard.biesheuvel@linaro.org>:
> On 30 September 2015 at 18:13, H. Nikolaus Schaller <hns@goldelico.com> wrote: >> >> Am 30.09.2015 um 18:02 schrieb Ard Biesheuvel <ard.biesheuvel@linaro.org>: >> >>> On 30 September 2015 at 17:56, H. Nikolaus Schaller <hns@goldelico.com> wrote: >>>> If the host toolchain is not glibc based then the arm kernel build >>>> fails with >>>> >>>> HOSTCC arch/arm/vdso/vdsomunge >>>> arch/arm/vdso/vdsomunge.c:48:22: fatal error: byteswap.h: No such file or directory >>>> >>>> Observed with omap2plus_defconfig and compile on Mac OS X with arm ELF >>>> cross-compiler. >>>> >>>> Reason: byteswap.h is a glibc only header. >>>> >>>> Changed to detect the host and include the right file as described at: >>>> >>>> https://bugs.freedesktop.org/show_bug.cgi?id=8882 >>>> >>>> Tested to compile on Mac OS X 10.9.5 host. >>>> >>>> Signed-off-by: H. Nikolaus Schaller <hns@goldelico.com> >>>> --- >>>> arch/arm/vdso/vdsomunge.c | 18 ++++++++++++++++++ >>>> 1 file changed, 18 insertions(+) >>>> >>>> diff --git a/arch/arm/vdso/vdsomunge.c b/arch/arm/vdso/vdsomunge.c >>>> index aedec81..5364cea 100644 >>>> --- a/arch/arm/vdso/vdsomunge.c >>>> +++ b/arch/arm/vdso/vdsomunge.c >>>> @@ -45,7 +45,25 @@ >>>> * it does. >>>> */ >>>> >>>> +#if defined(__linux__) >>>> #include <byteswap.h> >>>> +#elif defined(__OpenBSD__) >>>> +#include <sys/endian.h> >>>> +#define bswap_16 __swap16 >>>> +#define bswap_32 __swap32 >>>> +#define bswap_64 __swap64 >>>> +#elif defined(__APPLE__) >>>> +#include <libkern/OSByteOrder.h> >>>> +#define bswap_16 OSSwapInt16 >>>> +#define bswap_32 OSSwapInt32 >>>> +#define bswap_64 OSSwapInt64 >>>> +#else >>>> +#include <sys/endian.h> >>>> +#define bswap_16 bswap16 >>>> +#define bswap_32 bswap32 >>>> +#define bswap_64 bswap64 >>>> +#endif >>>> + >>> >>> Have you tried this? >>> >>> #define bswap_16 __builtin_bswap16 >>> #define bswap_32 __builtin_bswap32 >>> #define bswap_64 __builtin_bswap64 >>> >>> https://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html >> >> OS X host uses llvm and I am not sure if these builtins are >> always available. >> > > I am pretty sure recent clang supports these as well. Could you please try it?
What release date (= feature set) of a compiler would you no longer consider as "recent"?
> >> So it replaces an include file dependency by a compiler >> built-in dependency. IMHO my approach is more general >> and should cover a broader range of compile hosts. >> >> This is a tool compiled by HOSTCC to be run on the build >> host - so I think we can't assume to have a gcc host compiler >> (while for the CC cross-compiler we can). >> > > If my suggestion works, we'll support host side GCC and clang without > decorating the source code with lots of #ifdefs for OSes few people > care about. That would be an improvement imo
Indeed it would be the best solution.
But at least my Apple LLVM version 6.0 (clang-600.0.57) (based on LLVM 3.5svn) it does *not* work (see also for example http://sourceforge.net/p/flac/bugs/416/ ).
Clang 3.5 appears to provide all features of GCC 4.7.
I have checked and it appears that __builtin_bswap16 only became available in GCC 4.8 (released 2013-03-22). And maybe in Clang 3.7.
Unfortunately I could not find a reference what the minimum required gcc for the Linux HOSTCC currently is.
So if we require gcc >= 4.8 just by this small tool, this should be made well known and well decided.
Therefore I would propose to use the #ifdef approach now and revise 5 years after gcc 4.8 release (2013 ==> 2018) if __builtin_bswap16 (or something else) has become a widely available standard.
I can add this to the patch commit V2 for reference.
BR and thanks, Nikolaus Schaller
| |