Re: [RFC PATCH 1/1] mm/percpu.c: fix memory leakage issue when allocate a odd alignment area

From: zijun_hu
Date: Tue Oct 11 2016 - 20:29:02 EST

On 2016/10/12 1:22, Michal Hocko wrote:
> On Tue 11-10-16 21:24:50, zijun_hu wrote:
>> From: zijun_hu <zijun_hu@xxxxxxx>
>> the LSB of a chunk->map element is used for free/in-use flag of a area
>> and the other bits for offset, the sufficient and necessary condition of
>> this usage is that both size and alignment of a area must be even numbers
>> however, pcpu_alloc() doesn't force its @align parameter a even number
>> explicitly, so a odd @align maybe causes a series of errors, see below
>> example for concrete descriptions.
> Is or was there any user who would use a different than even (or power of 2)
> alighment? If not is this really worth handling?

it seems only a power of 2 alignment except 1 can make sure it work very well,
that is a strict limit, maybe this more strict limit should be checked

i don't know since there are too many sources and too many users and too many
use cases. even if nobody, i can't be sure that it doesn't happens in the future

it is worth since below reasons
1) if it is used in right ways, this patch have no impact; otherwise, it can alert
user by warning message and correct the behavior.
is it better that a warning message and correcting than resulting in many terrible
error silently under a special case by change?
it can make program more stronger.

2) does any alignment but 1 means a power of 2 alignment conventionally and implicitly?
if not, is it better that adjusting both @align and @size uniformly based on the sufficient
necessary condition than mixing supposing one part is right and correcting the other?
i find that there is BUG_ON(!is_power_of_2(align)) statement in mm/vmalloc.c

3) this simple fix can make the function applicable in wider range, it hints the reader
that the lowest requirement for alignment is a even number

4) for char a[10][10]; char (*p)[10]; if a user want to allocate a @size = 10 and
@align = 10 memory block, should we reject the user's request?