Re: [PATCH] mm/page_alloc: don't reserve ZONE_HIGHMEM for ZONE_MOVABLE request

[Michal Hocko]

On Wed 04-04-18 09:24:06, Joonsoo Kim wrote:
> 2017-09-14 22:24 GMT+09:00 Michal Hocko <mhocko@xxxxxxxxxx>:
> > [Sorry for a later reply]
> >
> > On Wed 06-09-17 13:35:25, Joonsoo Kim wrote:
> >> From: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx>
> >>
> >> Freepage on ZONE_HIGHMEM doesn't work for kernel memory so it's not that
> >> important to reserve.
> >
> > I am still not convinced this is a good idea. I do agree that reserving
> > memory in both HIGHMEM and MOVABLE is just wasting memory but removing
> > the reserve from the highmem as well will result that an oom victim will
> > allocate from lower zones and that might have unexpected side effects.
> 
> Looks like you are confused.
> 
> This patch only affects the situation that ZONE_HIGHMEM and ZONE_MOVABLE is
> used at the same time. In that case, before this patch, ZONE_HIGHMEM has
> reserve for GFP_HIGHMEM | GFP_MOVABLE request, but, with this patch,  no reserve
> in ZONE_HIGHMEM for GFP_HIGHMEM | GFP_MOVABLE request. This perfectly
> matchs with your hope. :)

I have forgot all the details but my vague recollection is that the
concern was that GFP_HIGHUSER_MOVABLE etc. wouldn't keep any reserve in
the highmem zone and so emergency allocations - e.g. those during OOM
will have to fallback to kernel zones and might lead to hard to predict
results. Am I still confused and this will not happen after the patch?
-- 
Michal Hocko
SUSE Labs