On Mon, Feb 18, 2019 at 05:52:09PM +0100, Arnd Bergmann wrote: > On Mon, Feb 18, 2019 at 5:44 PM Michael S. Tsirkin <mst@xxxxxxxxxx> wrote: > > > > On Tue, Feb 19, 2019 at 12:28:55AM +0800, kbuild test robot wrote: > > > tree: https://git.kernel.org/pub/scm/linux/kernel/git/arnd/asm-generic.git master > > > head: 3c4fcc2648569c3fe573849f9254c8c3499ec648 > > > commit: 9ac483109d19095e22414e46d38a6817520883b9 [2/6] arch: move common mmap flags to linux/mman.h > > > config: x86_64-kexec (attached as .config) > > > compiler: gcc-7 (Debian 7.3.0-1) 7.3.0 > > > reproduce: > > > git checkout 9ac483109d19095e22414e46d38a6817520883b9 > > > # save the attached .config to linux build tree > > > make ARCH=x86_64 > > > > > > All errors (new ones prefixed by >>): > > > > > > arch/x86//mm/mpx.c: In function 'mpx_mmap': > > > >> arch/x86//mm/mpx.c:57:26: error: 'MAP_PRIVATE' undeclared (first use in this function); did you mean 'DQF_PRIVATE'? > > > MAP_ANONYMOUS | MAP_PRIVATE, VM_MPX, 0, &populate, NULL); > > > ^~~~~~~~~~~ > > > DQF_PRIVATE > > > arch/x86//mm/mpx.c:57:26: note: each undeclared identifier is reported only once for each function it appears in > > > > Arnd I think you are missing: > > x86/mpx: tweak header name > > > > The whole file is going away eventually: > > https://lkml.kernel.org/r/7de5232f-4bc1-5c85-00d5-cb646462577f@xxxxxxxxx > > > > but to avoid breaking bisect I think it's needed. > > Right, I misunderstood the earlier mails and assumed it wasn't really > needed even for building the current tree. Added it in now as well. I see that > there are various other files including asm/mman.h: I did my best to check the rest, yes. I don't think any of them are broken. > arch/arm/mm/hugetlbpage.c:#include <asm/mman.h> > arch/arm64/mm/hugetlbpage.c:#include <asm/mman.h> > arch/c6x/include/asm/cacheflush.h:#include <asm/mman.h> > arch/ia64/include/asm/pgtable.h:#include <asm/mman.h> > arch/ia64/mm/hugetlbpage.c:#include <asm/mman.h> > arch/mips/kernel/linux32.c:#include <asm/mman.h> > arch/mips/mm/hugetlbpage.c:#include <asm/mman.h> > arch/parisc/mm/hugetlbpage.c:#include <asm/mman.h> > arch/powerpc/include/asm/mman.h:#include <uapi/asm/mman.h> > arch/powerpc/mm/hugetlbpage-radix.c:#include <asm/mman.h> > arch/powerpc/mm/pkeys.c:#include <asm/mman.h> > arch/powerpc/mm/slice.c:#include <asm/mman.h> > arch/sh/mm/hugetlbpage.c:#include <asm/mman.h> > arch/sparc/include/asm/mman.h:#include <uapi/asm/mman.h> > arch/sparc/mm/hugetlbpage.c:#include <asm/mman.h> > arch/um/kernel/syscall.c:#include <asm/mman.h> > arch/x86/ia32/sys_ia32.c:#include <asm/mman.h> > arch/x86/mm/hugetlbpage.c:#include <asm/mman.h> > arch/x86/um/shared/sysdep/kernel-offsets.h:#include <asm/mman.h> > fs/eventpoll.c:#include <asm/mman.h> > include/uapi/linux/mman.h:#include <asm/mman.h> > mm/filemap.c:#include <asm/mman.h> > tools/include/uapi/linux/mman.h:#include <asm/mman.h> > virt/kvm/arm/arm.c:#include <asm/mman.h> > > Did you check if we need to do the same change on any of the > other files? A lot of them are architecture specific and might not be > caught by the regular build testing. > > Arnd First let's check for direct issues: $ git grep -e MAP_SHARED -e MAP_PRIVATE `git grep -l asm/mman.h` arch/ia64/include/asm/pgtable.h: * version gets used for a shared memory segment with MAP_SHARED on. include/uapi/linux/mman.h:#define MAP_SHARED 0x01 /* Share changes */ include/uapi/linux/mman.h:#define MAP_PRIVATE 0x02 /* Changes are private */ include/uapi/linux/mman.h:#define MAP_SHARED_VALIDATE 0x03 /* share + validate extension flags */ Now it's possible that there's an indirect include. Let's find headers - but note that any asm/mman.h header itself is not a problem. $git grep -l asm/mman.h|fgrep -e .h|grep -v mman.h arch/c6x/include/asm/cacheflush.h arch/ia64/include/asm/pgtable.h arch/x86/um/shared/sysdep/kernel-offsets.h Analysis (only looking at arch specific files): $ git grep -e MAP_SHARED -e MAP_PRIVATE `git grep -l asm/cacheflush.h arch/c6x/` $ git grep -e MAP_SHARED -e MAP_PRIVATE `git grep -l asm/pgtable.h arch/ia64/` $ git grep MAP_ `git grep -l kernel-offsets.h` Having said that, once these patches are in I think the next step would be to: - patch CodingStyle to document the rule (already enforced by checkpatch.pl) that everyone should include linux/foo.h in preference to asm/foo.h - write a script to find and fix violations - they are probably not limited to mman.h -- MST
