lkml.org 
[lkml]   [2022]   [Apr]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Date
SubjectRe: [PATCH] mm/page_alloc: give priority to free cma-pages from pcplist to buddy
On Sun, Apr 24, 2022 at 3:28 PM <lipeifeng@oppo.com> wrote:
>
> From: lipeifeng <lipeifeng@oppo.com>
>
> Cma-pages will be fallback to movable pages in many scenarios.when cma
> pages are freed to pcplist, we give priority to free it from pcplist
> to buddy in order to avoids cma-pages to be used as movable-pages soon
> if there is enough free-movable-pages, which saves more cma-pages in
> buddy to decrease pages migration when cma_alloc.
>
> Signed-off-by: lipeifeng <lipeifeng@oppo.com>
> ---

+ Christoph, Marek, Robin as it is cma-related.


> mm/page_alloc.c | 28 ++++++++++++++++++++++++----
> 1 file changed, 24 insertions(+), 4 deletions(-)
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 3589feb..69369ed 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -3372,7 +3372,7 @@ static int nr_pcp_high(struct per_cpu_pages *pcp, struct zone *zone)
> }
>
> static void free_unref_page_commit(struct page *page, unsigned long pfn,
> - int migratetype, unsigned int order)
> + int migratetype, unsigned int order, bool fast_free)
> {
> struct zone *zone = page_zone(page);
> struct per_cpu_pages *pcp;
> @@ -3382,7 +3382,10 @@ static void free_unref_page_commit(struct page *page, unsigned long pfn,
> __count_vm_event(PGFREE);
> pcp = this_cpu_ptr(zone->per_cpu_pageset);
> pindex = order_to_pindex(migratetype, order);
> - list_add(&page->lru, &pcp->lists[pindex]);
> + if (fast_free)
> + list_add_tail(&page->lru, &pcp->lists[pindex]);
> + else
> + list_add(&page->lru, &pcp->lists[pindex]);

Ok. This is interesting, we used to have a separate cma pcp list but
now MIGRATE_CMA is an outsider so cma pages are placed in the
MIGRATE_MOVABLE list.

enum migratetype {
MIGRATE_UNMOVABLE,
MIGRATE_MOVABLE,
MIGRATE_RECLAIMABLE,
MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
#ifdef CONFIG_CMA
...
MIGRATE_CMA,
#endif
#ifdef CONFIG_MEMORY_ISOLATION
MIGRATE_ISOLATE, /* can't allocate from here */
#endif
MIGRATE_TYPES
};

#define NR_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1
+ NR_PCP_THP))

> pcp->count += 1 << order;
> high = nr_pcp_high(pcp, zone);
> if (pcp->count >= high) {
> @@ -3400,6 +3403,7 @@ void free_unref_page(struct page *page, unsigned int order)
> unsigned long flags;
> unsigned long pfn = page_to_pfn(page);
> int migratetype;
> + bool fast_free = false;
>
> if (!free_unref_page_prepare(page, pfn, order))
> return;
> @@ -3419,9 +3423,15 @@ void free_unref_page(struct page *page, unsigned int order)
> }
> migratetype = MIGRATE_MOVABLE;
> }
> + /*
> + * Give priority to free cma-pages to buddy in order to
> + * decrease pages migration when cma_alloc.
> + */
> + if (migratetype == MIGRATE_CMA)
> + fast_free = true;
>
> local_lock_irqsave(&pagesets.lock, flags);
> - free_unref_page_commit(page, pfn, migratetype, order);
> + free_unref_page_commit(page, pfn, migratetype, order, fast_free);
> local_unlock_irqrestore(&pagesets.lock, flags);
> }
>
> @@ -3459,6 +3469,8 @@ void free_unref_page_list(struct list_head *list)
>
> local_lock_irqsave(&pagesets.lock, flags);
> list_for_each_entry_safe(page, next, list, lru) {
> + bool fast_free = false;
> +
> pfn = page_private(page);
> set_page_private(page, 0);
>
> @@ -3467,11 +3479,19 @@ void free_unref_page_list(struct list_head *list)
> * to the MIGRATE_MOVABLE pcp list.
> */
> migratetype = get_pcppage_migratetype(page);
> +
> + /*
> + * Give priority to free cma-pages to buddy in order to
> + * decrease pages migration when cma_alloc.
> + */
> + if (migratetype == MIGRATE_CMA)
> + fast_free = true;
> +
> if (unlikely(migratetype >= MIGRATE_PCPTYPES))
> migratetype = MIGRATE_MOVABLE;
>
> trace_mm_page_free_batched(page);
> - free_unref_page_commit(page, pfn, migratetype, 0);
> + free_unref_page_commit(page, pfn, migratetype, 0, fast_free);

i'd call get_pcppage_migratetype() again in free_unref_page_commit()
rather than adding a parameter to increase a couple cross functions.

>
> /*
> * Guard against excessive IRQ disabled times when we get
> --
> 2.7.4
>

Thanks
Barry

\
 
 \ /
  Last update: 2022-04-24 06:34    [W:0.857 / U:0.092 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site