Re: [LSF/MM TOPIC] Improving CMA

[Gioh Kim <gioh.kim@lge.com>]

2014-11-27 i??i?? 3:12i?? Joonsoo Kim i?'(e??) i?' e,?:
> On Mon, Nov 24, 2014 at 05:54:14PM -0800, Laura Abbott wrote:
>> There have been a number of patch series posted designed to improve various
>> aspects of CMA. A sampling:
>>
>> https://lkml.org/lkml/2014/10/15/623
>> http://marc.info/?l=linux-mm&m=141571797202006&w=2
>> https://lkml.org/lkml/2014/6/26/549
>>
>> As far as I can tell, these are all trying to fix real problems with CMA but
>> none of them have moved forward very much from what I can tell. The goal of
>> this session would be to come out with an agreement on what are the biggest
>> problems with CMA and the best ways to solve them.
>
> I also tried to solve problem from CMA, that is, reserved memory
> utilization.
>
> https://lkml.org/lkml/2014/5/28/64
>
> While playing that patchset, I found serious problem about free page
> counting, so I stopped to develop it for a while and tried to fix it.
> Now, it is fixed by me and I can continue my patchset.
>
> https://lkml.org/lkml/2014/10/31/69
>
> I heard that Minchan suggests new CMA zone like movable zone, and, I
> think that it would be the way to go. But, it would be a long-term goal
> and I'd like to solve utilization problem with my patchset for now.
> It is the biggest issue and it already forces someone to develop
> out of tree solution. It's not good that out of tree solution is used
> more and more in the product so I'd like to fix it quickly at first
> stage.
>
> I think that CMA have big potential. If we fix problems of CMA
> completely, it can be used for many places. One such case in my mind
> is hugetlb or THP. Until now, hugetlb uses reserved approach, that is
> very inefficient. System administrator carefully set the number of
> reserved hugepage according to whole system workload. And application
> can't use it freely, because it is very limited and managed resource.
> If we use CMA for hugetlb, we can easily allocate hugepage and
> application can use hugepages more freely.
>
> Anyway, I'd like to attend LSF/MM and discuss this topic.
>
> Thanks.
>

Until now, I've used CMA with 2 out-of-tree patches:
1. https://lkml.org/lkml/2012/8/31/313 : Laura's patch
2. https://lkml.org/lkml/2014/5/28/64 : Joonsoo's patch

And one merged patch by me: https://lkml.org/lkml/2014/9/4/78

With them, my platform could've worked but it still had free-page-counting problem.

I think if Joonsoo's patch [2] is merged into mainline, CMA can be stable and useful.
Allocation latency Minchan mentioned is not problem for my platform.
CMA allocation is not often and limited to only one drivers.

Allocation guarantee is, I hope, fixed with my patch (https://lkml.org/lkml/2014/9/4/78) at least in my platform.
My platform had worked for several hours but it lacks heavy load test.
I have a plan to use CMA for massive product next year.

I'd like to attend LSF/MM and discuss this topic too.

--
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>