Re: [PATCH] mm,page_alloc,cma: conditionally prefer cma pageblocks for movable allocations

[Minchan Kim]

On Fri, Mar 06, 2020 at 03:01:02PM -0500, Rik van Riel wrote:
> Posting this one for Roman so I can deal with any upstream feedback and
> create a v2 if needed, while scratching my head over the next piece of
> this puzzle :)
> 
> ---8<---
> 
> From: Roman Gushchin <guro@xxxxxx>
> 
> Currently a cma area is barely used by the page allocator because
> it's used only as a fallback from movable, however kswapd tries
> hard to make sure that the fallback path isn't used.
> 
> This results in a system evicting memory and pushing data into swap,
> while lots of CMA memory is still available. This happens despite the
> fact that alloc_contig_range is perfectly capable of moving any movable
> allocations out of the way of an allocation.
> 
> To effectively use the cma area let's alter the rules: if the zone
> has more free cma pages than the half of total free pages in the zone,
> use cma pageblocks first and fallback to movable blocks in the case of
> failure.
> 
> Signed-off-by: Rik van Riel <riel@xxxxxxxxxxx>
> Co-developed-by: Rik van Riel <riel@xxxxxxxxxxx>
> Signed-off-by: Roman Gushchin <guro@xxxxxx>
> ---
>  mm/page_alloc.c | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
> 
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 3c4eb750a199..0fb3c1719625 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -2711,6 +2711,18 @@ __rmqueue(struct zone *zone, unsigned int order, int migratetype,
>  {
>  	struct page *page;
>  
> +	/*
> +	 * Balance movable allocations between regular and CMA areas by
> +	 * allocating from CMA when over half of the zone's free memory
> +	 * is in the CMA area.
> +	 */
> +	if (migratetype == MIGRATE_MOVABLE &&
> +	    zone_page_state(zone, NR_FREE_CMA_PAGES) >
> +	    zone_page_state(zone, NR_FREE_PAGES) / 2) {

Can't we move the check to caller so that only one atomic operation
per pcp refill?

rmqueue_bulk:
    spin_lock(zone->lock);
    cma_first = FREE_CMA > FREE_PAGE / 2;
    for (i, i < count; ++i) {
        __rmqueue(zone, order, migratetype, alloc_flags, cma_first);
    }

As a long term solution, I am looking forward to seeing cma zone
approach but this is also good as stop-gap solution.
Actually, in the android, vendors have used their customization to
make CMA area utilization high(i.e., CMA first and then movable)
but more restricted allocation pathes. So, I really want to see
this patch in upstream to make CMA utilization higher. A good side
about this patch is quite simple.

About the CMA allocation failure ratio, there is no good idea
to solve the issue perfectly. Even we go with cma zone approach,
it could happen. If so, I'd like to expose the symptom more
aggressively so that we could hear the pain and find the solution
actively rather than relying on luck.

Thus,
Acked-by: Minchan Kim <minchan@xxxxxxxxxx>