On Mon, 5 May 2008, Nick Piggin wrote: > > Index: linux-2.6/include/asm-x86/pgtable_32.h > =================================================================== > --- linux-2.6.orig/include/asm-x86/pgtable_32.h > +++ linux-2.6/include/asm-x86/pgtable_32.h > @@ -133,7 +133,12 @@ extern int pmd_bad(pmd_t pmd); > * pgd_offset() returns a (pgd_t *) > * pgd_index() is used get the offset into the pgd page's array of pgd_t's; > */ > -#define pgd_offset(mm, address) ((mm)->pgd + pgd_index((address))) > +#define pgd_offset(mm, address) \ > +({ \ > + pgd_t *ret = ((mm)->pgd + pgd_index((address))); \ > + smp_read_barrier_depends(); /* see mm/memory.c:__pte_alloc */ \ > + ret; \ > +}) Is there some fundamental reason this needs to be a macro? It is really ugly, and it would be much nicer to make this an inline function if at all possible. Yeah, maybe it requires some more #include's, but .. (Especially since it apparently gets worse, and the pgd load needs a ACCESS_ONCE() too - the code generated is the same, but the source gets more and more involved) That said, I *also* think that it's sad that you do this at all, since smp_read_barrier_depends() is a no-op on x86, so why should we have it in an x86-specific header file? In short, I think the fixes are real, but the patch itself is really just confusing things for no apparent good reason. Linus -- To unsubscribe from this list: send the line "unsubscribe linux-arch" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html