Re: [PATCH RFC 01/10] mm: Add pmd support for _PAGE_SPECIAL

[Matthew Wilcox]

On Fri, Jan 10, 2020 at 07:03:04PM +0000, Joao Martins wrote:
> +++ b/arch/x86/include/asm/pgtable.h
> @@ -293,6 +293,15 @@ static inline int pgd_devmap(pgd_t pgd)
>  {
>  	return 0;
>  }
> +#endif
> +
> +#ifdef CONFIG_ARCH_HAS_PTE_SPECIAL
> +static inline int pmd_special(pmd_t pmd)
> +{
> +	return !!(pmd_flags(pmd) & _PAGE_SPECIAL);
> +}
> +#endif

The ifdef/endif don't make much sense here; x86 does have PTE_SPECIAL,
and this is an x86 header file, so that can be assumed.

> +++ b/mm/gup.c
> @@ -2079,6 +2079,9 @@ static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
>  		return __gup_device_huge_pmd(orig, pmdp, addr, end, pages, nr);
>  	}
>  
> +	if (pmd_special(orig))
> +		return 0;

Here, you're calling it unconditionally.  I think you need a pmd_special()
conditionally defined in include/asm-generic/pgtable.h

+#ifndef CONFIG_ARCH_HAS_PTE_SPECIAL
+static inline bool pmd_special(pmd_t pmd)
+{
+	return false;
+}
+#endif

(oh, and plese use bool instead of int; I know that's different from
pte_special(), but pte_special() predates bool and nobody's done the work
to convert it yet)

> +++ b/mm/huge_memory.c
> @@ -791,6 +791,8 @@ static void insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr,
>  	entry = pmd_mkhuge(pfn_t_pmd(pfn, prot));
>  	if (pfn_t_devmap(pfn))
>  		entry = pmd_mkdevmap(entry);
> +	else if (pfn_t_special(pfn))
> +		entry = pmd_mkspecial(entry);

Again, we'll need a generic one.

> @@ -823,8 +825,7 @@ vm_fault_t vmf_insert_pfn_pmd(struct vm_fault *vmf, pfn_t pfn, bool write)
>  	 * but we need to be consistent with PTEs and architectures that
>  	 * can't support a 'special' bit.
>  	 */
> -	BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) &&
> -			!pfn_t_devmap(pfn));
> +	BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));

Should that rather be ...

+	BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) &&
+			!pfn_t_devmap(pfn) && !pfn_t_special(pfn));

I also think this comment needs adjusting:

        /*
         * There is no pmd_special() but there may be special pmds, e.g.
         * in a direct-access (dax) mapping, so let's just replicate the
         * !CONFIG_ARCH_HAS_PTE_SPECIAL case from vm_normal_page() here.
         */