Re: 2.6.22-rc4-mm1

Re: 2.6.22-rc4-mm1

[KAMEZAWA Hiroyuki]

Question.

While writing memory unplug, I noticed this code.
==
static int
fixup_anon_page(pte_t *pte, unsigned long start, unsigned long end, void *priv)
{
        struct vm_area_struct *vma = priv;
        struct page *page = vm_normal_page(vma, start, *pte);

        if (page && PageAnon(page))
                page->index = linear_page_index(vma, start);

        return 0;
}

static int fixup_anon_pages(struct vm_area_struct *vma)
{
        struct mm_walk walk = {
                .pte_entry = fixup_anon_page,
        };

        return walk_page_range(vma->vm_mm,
                        vma->vm_start, vma->vm_end, &walk, vma);
}
==

I think that 'pte' passed to fixup_anon_page() by walk_page_range()
is not guaranteed to be 'Present'.

Then, vm_normal_page() will show print_bad_pte().

If this never occur now, I'll add my own check code for memory migration by kernel here.

(Sorry, I can't find who should be CCed.)

-Kame


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: 2.6.22-rc4-mm1

[Andrew Morton]

On Thu, 7 Jun 2007 21:47:06 +0900 KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:

> Question.
> 
> While writing memory unplug, I noticed this code.
> ==
> static int
> fixup_anon_page(pte_t *pte, unsigned long start, unsigned long end, void *priv)
> {
>         struct vm_area_struct *vma = priv;
>         struct page *page = vm_normal_page(vma, start, *pte);
> 
>         if (page && PageAnon(page))
>                 page->index = linear_page_index(vma, start);
> 
>         return 0;
> }
> 
> static int fixup_anon_pages(struct vm_area_struct *vma)
> {
>         struct mm_walk walk = {
>                 .pte_entry = fixup_anon_page,
>         };
> 
>         return walk_page_range(vma->vm_mm,
>                         vma->vm_start, vma->vm_end, &walk, vma);
> }

I assume the above is your code - it's not in the tree?

> 
> I think that 'pte' passed to fixup_anon_page() by walk_page_range()
> is not guaranteed to be 'Present'.

yup - the pagewalker only checks for !pte_none().

> Then, vm_normal_page() will show print_bad_pte().

> If this never occur now, I'll add my own check code for memory migration by kernel here.

Yes, you'll need to perform additional filtering where appropriate.

> (Sorry, I can't find who should be CCed.)

Matt and David did most of the work here.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: 2.6.22-rc4-mm1

[KAMEZAWA Hiroyuki]

On Thu, 7 Jun 2007 08:34:58 -0700
Andrew Morton <akpm@linux-foundation.org> wrote:
> 
> I assume the above is your code - it's not in the tree?
> 
Ah, that code was disappeared in -mm2. 

But it informed me that I should consider memory unplug v.s. sys_mremap case... 

Thanks, anyway.

-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>