Re: Fixup the page of buddy_higher address's calculation

[Michal Hocko]

On Fri 24-08-12 17:08:36, Li Haifeng wrote:
> 2012/8/24 Michal Hocko <mhocko@xxxxxxx>:
> > On Fri 24-08-12 10:08:20, Li Haifeng wrote:
> > [...]
> >> Subject: [PATCH] Fix the page address of higher page's buddy calculation
> >>
> >> Calculate the page address of higher page's buddy should be based
> >> higher_page with the offset between index of higher page and
> >> index of higher page's buddy.
> >
> > Sorry for insisting but could you add an information about when this has
> > been introduced (I have mentioned the commit in the other email) and the
> > effect of the bug so that we can consider whether this is worth
> > backporting to stable trees.
> >
> >> Signed-off-by: Haifeng Li <omycle@xxxxxxxxx>
> >> Signed-off-by: Gavin Shan <shangw@xxxxxxxxxxxxxxxxxx>
> >
> > Other than that
> > Reviewed-by: Michal Hocko <mhocko@xxxxxxx>
> >
> >> ---
> >>  mm/page_alloc.c |    2 +-
> >>  1 files changed, 1 insertions(+), 1 deletions(-)
> >>
> >> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> >> index cdef1d4..642cd62 100644
> >> --- a/mm/page_alloc.c
> >> +++ b/mm/page_alloc.c
> >> @@ -536,7 +536,7 @@ static inline void __free_one_page(struct page *page,
> >>                 combined_idx = buddy_idx & page_idx;
> >>                 higher_page = page + (combined_idx - page_idx);
> >>                 buddy_idx = __find_buddy_index(combined_idx, order + 1);
> >> -               higher_buddy = page + (buddy_idx - combined_idx);
> >> +               higher_buddy = higher_page + (buddy_idx - combined_idx);
> >>                 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
> >>                         list_add_tail(&page->lru,
> >>                                 &zone->free_area[order].free_list[migratetype]);
> >> --
> >> 1.7.5.4
> >
> > --
> > Michal Hocko
> > SUSE Labs
> 
> I am sorry Michal. I misinterpreted what you mean.
> 
> And the post blow is OK?
> ------------------------------------------>
> Subject: [PATCH] Fix the page address of higher page's buddy calculation
> 
> The heuristic method for buddy has been introduced since
> 43506fad(mm/page_alloc.c: simplify calculation of combined index
> of adjacent buddy lists). 

Maybe I just misunderstood you here but the heuristic has been
introduced by 6dda9d55 (page allocator: reduce fragmentation in buddy
allocator by adding buddies that are merging to the tail of the free
lists) and the commit you are mentioning broke it.

> But the page address of higher page's buddy was wrongly calculated,
> which will lead page_is_buddy to fail for ever. IOW, the heuristic
> method would be disabled with the wrong page address of higher page's
> buddy.
> 
> Calculating the page address of higher page's buddy should be based
> higher_page with the offset between index of higher page and
> index of higher page's buddy.
>

I would also add CC: stable [2.6.38+] although it is not clear how much
the heuristic is helpful. Anyway it's a regression and should be fix
IMHO.

> Signed-off-by: Haifeng Li <omycle@xxxxxxxxx>
> Signed-off-by: Gavin Shan <shangw@xxxxxxxxxxxxxxxxxx>
> Reviewed-by: Michal Hocko <mhocko@xxxxxxx>
> ---
>  mm/page_alloc.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
> 
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index ddbc17d..0754a3c 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -579,7 +579,7 @@ static inline void __free_one_page(struct page *page,
>                 combined_idx = buddy_idx & page_idx;
>                 higher_page = page + (combined_idx - page_idx);
>                 buddy_idx = __find_buddy_index(combined_idx, order + 1);
> -               higher_buddy = page + (buddy_idx - combined_idx);
> +               higher_buddy = higher_page + (buddy_idx - combined_idx);
>                 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
>                         list_add_tail(&page->lru,
>                                 &zone->free_area[order].free_list[migratetype]);
> --
> 1.7.5.4

-- 
Michal Hocko
SUSE Labs
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/