On Sat, 10 May 2014 00:16:48 +0800 Jianyu Zhan <nasa4836@xxxxxxxxx> wrote:
> mlocked_vma_newpage() is only called in fault path by
> page_add_new_anon_rmap(), which is called on a *new* page.
> And such page is initially only visible via the pagetables, and the
> pte is locked while calling page_add_new_anon_rmap(), so we need not
> use an irq-safe mod_zone_page_state() here, using a light-weight version
> __mod_zone_page_state() would be OK.
>
> ...
>
> --- a/mm/internal.h
> +++ b/mm/internal.h
> @@ -196,7 +196,7 @@ static inline int mlocked_vma_newpage(struct vm_area_struct *vma,
> return 0;
>
> if (!TestSetPageMlocked(page)) {
> - mod_zone_page_state(page_zone(page), NR_MLOCK,
> + __mod_zone_page_state(page_zone(page), NR_MLOCK,
> hpage_nr_pages(page));
> count_vm_event(UNEVICTABLE_PGMLOCKED);
> }
The comment over __mod_zone_page_state() says "For use when we know
that interrupts are disabled". But that is not the case here.
Please fix this up, probably by altering the __mod_zone_page_state()
documentation.
Your proposed change to (the very poorly named) mlocked_vma_newpage()
makes that function more dangerous - people might add new callsites
without knowing that they're adding races. This risk can be reduced by
a) adding documentation explaining why __mod_zone_page_state() is
used and why it is safe and
b) moving mlocked_vma_newpage() out of internal.h and placing it in
mm/rmap.c immediately before page_add_new_anon_rmap(). Or, probably
better, by removing mlocked_vma_newpage() altogether and open-coding
its logic into page_add_new_anon_rmap().
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>