Re: [Linaro-mm-sig] [PATCHv23 12/16] mm: trigger page reclaim inalloc_contig_range() to stabilise watermarks

[Barry Song]

2012/2/23 Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>:
> alloc_contig_range() performs memory allocation so it also should keep
> track on keeping the correct level of memory watermarks. This commit adds
> a call to *_slowpath style reclaim to grab enough pages to make sure that
> the final collection of contiguous pages from freelists will not starve
> the system.
>
> Signed-off-by: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>
> Signed-off-by: Kyungmin Park <kyungmin.park@xxxxxxxxxxx>
> CC: Michal Nazarewicz <mina86@xxxxxxxxxx>
> Tested-by: Rob Clark <rob.clark@xxxxxxxxxx>
> Tested-by: Ohad Ben-Cohen <ohad@xxxxxxxxxx>
> Tested-by: Benjamin Gaignard <benjamin.gaignard@xxxxxxxxxx>
> Tested-by: Robert Nelson <robertcnelson@xxxxxxxxx>
> ---
> Âinclude/linux/mmzone.h | Â Â9 +++++++
> Âmm/page_alloc.c    Â|  62 ++++++++++++++++++++++++++++++++++++++++++++++++
> Â2 files changed, 71 insertions(+), 0 deletions(-)
>
> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
> index 4781f30..77db8c0 100644
> --- a/include/linux/mmzone.h
> +++ b/include/linux/mmzone.h
> @@ -63,8 +63,10 @@ enum {
>
> Â#ifdef CONFIG_CMA
> Â# Âdefine is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
> +# Âdefine cma_wmark_pages(zone) Â Â Â Âzone->min_cma_pages
> Â#else
> Â# Âdefine is_migrate_cma(migratetype) false
> +# Âdefine cma_wmark_pages(zone) 0
> Â#endif
>
> Â#define for_each_migratetype_order(order, type) \
> @@ -371,6 +373,13 @@ struct zone {
> Â Â Â Â/* see spanned/present_pages for more description */
>    Âseqlock_t        span_seqlock;
> Â#endif
> +#ifdef CONFIG_CMA
> + Â Â Â /*
> + Â Â Â Â* CMA needs to increase watermark levels during the allocation
> + Â Â Â Â* process to make sure that the system is not starved.
> + Â Â Â Â*/
> +    unsigned long      min_cma_pages;
> +#endif
>    Âstruct free_area    Âfree_area[MAX_ORDER];
>
> Â#ifndef CONFIG_SPARSEMEM
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 7a0d286..39cd74f 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -5092,6 +5092,11 @@ static void __setup_per_zone_wmarks(void)
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Âlow + (min >> 2);
> Â Â Â Â Â Â Â Âzone->watermark[WMARK_HIGH] = min_wmark_pages(zone) +
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Âlow + (min >> 1);
> +
> + Â Â Â Â Â Â Â zone->watermark[WMARK_MIN] += cma_wmark_pages(zone);
> + Â Â Â Â Â Â Â zone->watermark[WMARK_LOW] += cma_wmark_pages(zone);
> + Â Â Â Â Â Â Â zone->watermark[WMARK_HIGH] += cma_wmark_pages(zone);
> +
> Â Â Â Â Â Â Â Âsetup_zone_migrate_reserve(zone);
> Â Â Â Â Â Â Â Âspin_unlock_irqrestore(&zone->lock, flags);
> Â Â Â Â}
> @@ -5695,6 +5700,56 @@ static int __alloc_contig_migrate_range(unsigned long start, unsigned long end)
> Â Â Â Âreturn ret > 0 ? 0 : ret;
> Â}
>
> +/*
> + * Update zone's cma pages counter used for watermark level calculation.
> + */
> +static inline void __update_cma_watermarks(struct zone *zone, int count)
> +{
> + Â Â Â unsigned long flags;
> + Â Â Â spin_lock_irqsave(&zone->lock, flags);
> + Â Â Â zone->min_cma_pages += count;
> + Â Â Â spin_unlock_irqrestore(&zone->lock, flags);
> + Â Â Â setup_per_zone_wmarks();
> +}
> +
> +/*
> + * Trigger memory pressure bump to reclaim some pages in order to be able to
> + * allocate 'count' pages in single page units. Does similar work as
> + *__alloc_pages_slowpath() function.
> + */
> +static int __reclaim_pages(struct zone *zone, gfp_t gfp_mask, int count)
> +{
> + Â Â Â enum zone_type high_zoneidx = gfp_zone(gfp_mask);
> + Â Â Â struct zonelist *zonelist = node_zonelist(0, gfp_mask);
> + Â Â Â int did_some_progress = 0;
> + Â Â Â int order = 1;
> + Â Â Â unsigned long watermark;
> +
> + Â Â Â /*
> + Â Â Â Â* Increase level of watermarks to force kswapd do his job
> + Â Â Â Â* to stabilise at new watermark level.
> + Â Â Â Â*/
> + Â Â Â __update_cma_watermarks(zone, count);
> +
> + Â Â Â /* Obey watermarks as if the page was being allocated */
> + Â Â Â watermark = low_wmark_pages(zone) + count;
> + Â Â Â while (!zone_watermark_ok(zone, 0, watermark, 0, 0)) {
> + Â Â Â Â Â Â Â wake_all_kswapd(order, zonelist, high_zoneidx, zone_idx(zone));
> +
> + Â Â Â Â Â Â Â did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
> + Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â NULL);
> + Â Â Â Â Â Â Â if (!did_some_progress) {
> + Â Â Â Â Â Â Â Â Â Â Â /* Exhausted what can be done so it's blamo time */
> + Â Â Â Â Â Â Â Â Â Â Â out_of_memory(zonelist, gfp_mask, order, NULL);

out_of_memory() has got another param in the newest next/master tree,
out_of_memory(zonelist, gfp_mask, order, NULL, false) should be OK.

-barry
--
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/