Sorry for responding to myself here, of course right after sending this email I found again the CPP debugging options that I was looking for only to realize that the following happened: io/mmap.c does include input.h which includes libfrog/projectsh which includes xfs.h which includes linux.h. Because linux.h has a header guard, the second inclusion *after* sys/mman.h had no chance of re-defining MAP_SYNC, I should be able to submit a patch addressing that. On 7/20/22 12:49, Florian Fainelli wrote: > Hi, > > While testing musl-libc builds with MIPS-based platforms, I ran into the following build failure affecting the mmap.c file only: > > mmap.c: In function 'mmap_f': > mmap.c:196:12: error: 'MAP_SYNC' undeclared (first use in this function); did you mean 'MS_SYNC'? > 196 | flags = MAP_SYNC | MAP_SHARED_VALIDATE; > | ^~~~~~~~ > | MS_SYNC > mmap.c:196:12: note: each undeclared identifier is reported only once for each function it appears in > make[4]: *** [../include/buildrules:81: mmap.o] Error 1 > > musl-libc does explicitly un-define MAP_SYNC since MIPS does not support it: > > https://git.musl-libc.org/cgit/musl/tree/arch/mips/bits/mman.h#n21 > > however the include chain for io/mmap.c should still have the linux.h compatibility shim provide a suitable definition for MAP_SYNC so this did not really make sense to me: > > #include <sys/mman.h> > -> bits/mman.h (where the undefined happens) > #include "io.h" > -> include/xfs.h > -> include/linux.h > > I have confirmed that at every stage of the inclusion chain the MAP_SYNC and HAVE_MAP_SYNC are defined or undefined as intended, up until the top-level inclusion of include/xfs.h in io/io.h where MAP_SYNC is not defined after we included xfs.h which does not quite make sense to me at all. The build command line for that file is: > > mipsel-buildroot-linux-musl-gcc -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -Os -g0 -DDANGEROUS_COMMANDS_ENABLED -D_FILE_OFFSET_BITS=64 -Wno-address-of-packed-member -g -O2 -DDEBUG -DVERSION=\"5.14.2\" -DLOCALEDIR=\"/usr/share/locale\" -DPACKAGE=\"xfsprogs\" -I../include -I../libxfs -I.. -D_LGPL_SOURCE -D_GNU_SOURCE -funsigned-char -fno-strict-aliasing -Wall -DHAVE_MNTENT -DHAVE_FSETXATTR -DENABLE_BLKID -DHAVE_GETFSMAP -DHAVE_FALLOCATE -DHAVE_FADVISE -DHAVE_MADVISE -DHAVE_MINCORE -DHAVE_SENDFILE -DHAVE_FIEMAP -DHAVE_COPY_FILE_RANGE -DHAVE_SYNC_FILE_RANGE -DHAVE_SYNCFS -DHAVE_FALLOCATE -DHAVE_PREADV -DHAVE_PWRITEV -DHAVE_READDIR -DHAVE_MREMAP -DHAVE_STATFS_FLAGS -c mmap.c > > People in other projects have attempted to fix it > > https://github.com/openwrt/packages/blob/master/utils/xfsprogs/patches/140-mman.patch > > but this feels just plain wrong. The only way that I could get it to build was to do the following, but this does not really work on my x86_64 glibc host: > > diff --git a/include/linux.h b/include/linux.h > index 3d9f4e3dca80..7ca9fab16705 100644 > --- a/include/linux.h > +++ b/include/linux.h > @@ -252,8 +252,11 @@ struct fsxattr { > #endif > > #ifndef HAVE_MAP_SYNC > +#include <sys/mman.h> > #define MAP_SYNC 0 > +#ifndef MAP_SHARED_VALIDATE > #define MAP_SHARED_VALIDATE 0 > +#endif > #else > #include <asm-generic/mman.h> > #include <asm-generic/mman-common.h> > > Any clues from the mmap.i attached filed what could be going wrong? > > Thank you for reading me! -- Florian
