[Cc: akpm]
On Wed, 21 Mar 2012 15:59:34 +0100 Andrea Arcangeli <aarcange@xxxxxxxxxx> wrote:
>
> On Wed, Mar 21, 2012 at 10:36:14AM -0400, Mark Salter wrote:
> > I'm seeing a build failure in linux-next:
> >
> > CC init/main.o
> > In file included from /es/linux/linux-next/arch/c6x/include/asm/pgtable.h:76:0,
> > from /es/linux/linux-next/include/linux/mm.h:44,
> > from /es/linux/linux-next/include/linux/ring_buffer.h:5,
> > from /es/linux/linux-next/include/linux/ftrace_event.h:4,
> > from /es/linux/linux-next/include/trace/syscall.h:6,
> > from /es/linux/linux-next/include/linux/syscalls.h:78,
> > from /es/linux/linux-next/init/main.c:16:
> > /es/linux/linux-next/include/asm-generic/pgtable.h: In function 'pmd_none_or_trans_huge_or_clear_bad':
> > /es/linux/linux-next/include/asm-generic/pgtable.h:476:4: error: implicit declaration of function 'pmd_clear_bad' [-Werror=implicit-function-declaration]
> >
> >
> > This patch added some functions to asm-generic/pgtable.h which should
> > have been placed in the CONFIG_MMU conditional block:
> >
> > Author: Andrea Arcangeli <aarcange@xxxxxxxxxx>
> > Date: Wed Mar 21 10:48:00 2012 +1100
> >
> > mm: thp: fix pmd_bad() triggering in code paths holding mmap_sem read mode
> >
> >
> > The following patch fixes the build problem for me:
>
> Thanks for noticing this problem.
>
>
>
> >
> > diff --git a/include/asm-generic/pgtable.h b/include/asm-generic/pgtable.h
> > index 202c010..8ba3ba5 100644
> > --- a/include/asm-generic/pgtable.h
> > +++ b/include/asm-generic/pgtable.h
> > @@ -342,6 +342,64 @@ static inline void ptep_modify_prot_commit(struct mm_struct *mm,
> > __ptep_modify_prot_commit(mm, addr, ptep, pte);
> > }
> > #endif /* __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION */
> > +
> > +/*
> > + * This function is meant to be used by sites walking pagetables with
> > + * the mmap_sem hold in read mode to protect against MADV_DONTNEED and
> > + * transhuge page faults. MADV_DONTNEED can convert a transhuge pmd
> > + * into a null pmd and the transhuge page fault can convert a null pmd
> > + * into an hugepmd or into a regular pmd (if the hugepage allocation
> > + * fails). While holding the mmap_sem in read mode the pmd becomes
> > + * stable and stops changing under us only if it's not null and not a
> > + * transhuge pmd. When those races occurs and this function makes a
> > + * difference vs the standard pmd_none_or_clear_bad, the result is
> > + * undefined so behaving like if the pmd was none is safe (because it
> > + * can return none anyway). The compiler level barrier() is critically
> > + * important to compute the two checks atomically on the same pmdval.
> > + */
> > +static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
> > +{
> > + /* depend on compiler for an atomic pmd read */
> > + pmd_t pmdval = *pmd;
> > + /*
> > + * The barrier will stabilize the pmdval in a register or on
> > + * the stack so that it will stop changing under the code.
> > + */
> > +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> > + barrier();
> > +#endif
> > + if (pmd_none(pmdval))
> > + return 1;
> > + if (unlikely(pmd_bad(pmdval))) {
> > + if (!pmd_trans_huge(pmdval))
> > + pmd_clear_bad(pmd);
>
> Problem is, this fixes MMU=n but it'll break x86 with MMU=y and THP=n.
>
> These functions shall be placed after pmd_trans_huge you see at the
> end of asm-generic/pgtable.h .
>
> The simplest fix is that you add #ifdef CONFIG_MMU around it instead
> of moving (I guess you can keep pmd_trans_huge and the rest at the end
> of the file inside CONFIG_MMU too as it shall never be called as it
> all takes pmds/ptes as parameter).
>
> Thanks,
> Andrea
--
Cheers,
Stephen Rothwell sfr@xxxxxxxxxxxxxxxx
http://www.canb.auug.org.au/~sfr/
Attachment:
pgpz_LjJvogPB.pgp
Description: PGP signature
