* [PATCH mm] introduce reverse buddy concept to reduce buddy fragment
@ 2017-06-30 11:25 zhouxianrong
2017-07-03 7:48 ` Michal Hocko
0 siblings, 1 reply; 12+ messages in thread
From: zhouxianrong @ 2017-06-30 11:25 UTC (permalink / raw)
To: linux-mm
Cc: linux-kernel, akpm, vbabka, alexander.h.duyck, mhocko, mgorman,
l.stach, vdavydov.dev, hannes, minchan, npiggin, kirill.shutemov,
gi-oh.kim, luto, keescook, mark.rutland, mingo, heiko.carstens,
iamjoonsoo.kim, rientjes, ming.ling, jack, ebru.akagunduz,
bigeasy, Mi.Sophia.Wang, zhouxianrong, zhouxiyu, weidu.du,
fanghua3, won.ho.park
From: zhouxianrong <zhouxianrong@huawei.com>
when buddy is under fragment i find that still there are some pages
just like AFFA mode. A is allocated, F is free, AF is buddy pair for
oder n, FA is buddy pair for oder n as well. I want to compse the
FF as oder n + 1 and align to n other than n + 1. this patch broke
the rules of buddy stated as alignment to its length of oder. i think
we can do so except for kernel stack because the requirement comes from
buddy attribution rather than user. for kernel stack requirement i add
__GFP_NOREVERSEBUDDY for this purpose.
a sample just like blow.
Node 0, zone DMA
1389 1765 342 272 2 0 0 0 0 0 0
0 75 4398 1560 379 27 2 0 0 0 0
Node 0, zone Normal
20 24 14 2 0 0 0 0 0 0 0
0 6 228 3 0 0 0 0 0 0 0
the patch does not consider fallback allocation for now.
Signed-off-by: zhouxianrong <zhouxianrong@huawei.com>
---
include/linux/gfp.h | 8 +-
include/linux/mmzone.h | 2 +
include/linux/page-flags.h | 9 ++
include/linux/thread_info.h | 5 +-
mm/compaction.c | 17 ++++
mm/internal.h | 7 ++
mm/page_alloc.c | 222 +++++++++++++++++++++++++++++++++++++++----
mm/vmstat.c | 5 +-
8 files changed, 251 insertions(+), 24 deletions(-)
diff --git a/include/linux/gfp.h b/include/linux/gfp.h
index db373b9..f63d4d9 100644
--- a/include/linux/gfp.h
+++ b/include/linux/gfp.h
@@ -40,6 +40,7 @@
#define ___GFP_DIRECT_RECLAIM 0x400000u
#define ___GFP_WRITE 0x800000u
#define ___GFP_KSWAPD_RECLAIM 0x1000000u
+#define ___GFP_NOREVERSEBUDDY 0x2000000u
/* If the above are modified, __GFP_BITS_SHIFT may need updating */
/*
@@ -171,6 +172,10 @@
* __GFP_NOTRACK_FALSE_POSITIVE is an alias of __GFP_NOTRACK. It's a means of
* distinguishing in the source between false positives and allocations that
* cannot be supported (e.g. page tables).
+ *
+ * __GFP_NOREVERSEBUDDY does not allocate pages from reverse buddy list
+ * of current order. It make sure that allocation is alignment to same order
+ * with length order.
*/
#define __GFP_COLD ((__force gfp_t)___GFP_COLD)
#define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
@@ -178,9 +183,10 @@
#define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
#define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK)
#define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
+#define __GFP_NOREVERSEBUDDY ((__force gfp_t)___GFP_NOREVERSEBUDDY)
/* Room for N __GFP_FOO bits */
-#define __GFP_BITS_SHIFT 25
+#define __GFP_BITS_SHIFT 26
#define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
/*
diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 8e02b37..94237fe 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -89,7 +89,9 @@ enum {
struct free_area {
struct list_head free_list[MIGRATE_TYPES];
+ struct list_head free_list_reverse[MIGRATE_TYPES];
unsigned long nr_free;
+ unsigned long nr_free_reverse;
};
struct pglist_data;
diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
index 6b5818d..39d17d7 100644
--- a/include/linux/page-flags.h
+++ b/include/linux/page-flags.h
@@ -675,6 +675,15 @@ static inline int TestClearPageDoubleMap(struct page *page)
#define PAGE_KMEMCG_MAPCOUNT_VALUE (-512)
PAGE_MAPCOUNT_OPS(Kmemcg, KMEMCG)
+/*
+ * ReverseBuddy is enabled for the buddy allocator that allow allocating
+ * two adjacent same free order blocks other than buddy blocks and
+ * composing them as a order + 1 block. It is for reducing buddy
+ * fragment.
+ */
+#define PAGE_REVERSE_BUDDY_MAPCOUNT_VALUE (-1024)
+PAGE_MAPCOUNT_OPS(ReverseBuddy, REVERSE_BUDDY)
+
extern bool is_free_buddy_page(struct page *page);
__PAGEFLAG(Isolated, isolated, PF_ANY);
diff --git a/include/linux/thread_info.h b/include/linux/thread_info.h
index 5837387..b4a1605 100644
--- a/include/linux/thread_info.h
+++ b/include/linux/thread_info.h
@@ -28,9 +28,10 @@
#ifdef CONFIG_DEBUG_STACK_USAGE
# define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK | \
- __GFP_ZERO)
+ __GFP_NOREVERSEBUDDY | __GFP_ZERO)
#else
-# define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK)
+# define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK | \
+ __GFP_NOREVERSEBUDDY)
#endif
/*
diff --git a/mm/compaction.c b/mm/compaction.c
index 0fdfde0..a43f169 100644
--- a/mm/compaction.c
+++ b/mm/compaction.c
@@ -768,6 +768,20 @@ static bool too_many_isolated(struct zone *zone)
continue;
}
+ if (PageReverseBuddy(page)) {
+ unsigned long freepage_order = page_order_unsafe(page);
+
+ /*
+ * Without lock, we cannot be sure that what we got is
+ * a valid page order. Consider only values in the
+ * valid order range to prevent low_pfn overflow.
+ */
+ if (freepage_order > 0 &&
+ freepage_order < MAX_ORDER - 1)
+ low_pfn += (1UL << (freepage_order + 1)) - 1;
+ continue;
+ }
+
/*
* Regardless of being on LRU, compound pages such as THP and
* hugetlbfs are not to be compacted. We can potentially save
@@ -1005,6 +1019,9 @@ static bool suitable_migration_target(struct compact_control *cc,
return false;
}
+ if (PageReverseBuddy(page))
+ return false;
+
/* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
if (migrate_async_suitable(get_pageblock_migratetype(page)))
return true;
diff --git a/mm/internal.h b/mm/internal.h
index ccfc2a2..439b0a8 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -143,6 +143,13 @@ struct alloc_context {
return page_pfn ^ (1 << order);
}
+static inline unsigned long
+__find_reverse_buddy_pfn(unsigned long page_pfn, unsigned int order)
+{
+ return (page_pfn & (1 << order)) ? page_pfn + (1 << order) :
+ page_pfn - (1 << order);
+}
+
extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
unsigned long end_pfn, struct zone *zone);
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 9f9623d..ee1dc1b 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -715,6 +715,18 @@ static inline void rmv_page_order(struct page *page)
set_page_private(page, 0);
}
+static inline void set_reverse_page_order(struct page *page, unsigned int order)
+{
+ set_page_private(page, order);
+ __SetPageReverseBuddy(page);
+}
+
+static inline void rmv_reverse_page_order(struct page *page)
+{
+ __ClearPageReverseBuddy(page);
+ set_page_private(page, 0);
+}
+
/*
* This function checks whether a page is free && is the buddy
* we can do coalesce a page and its buddy if
@@ -758,6 +770,120 @@ static inline int page_is_buddy(struct page *page, struct page *buddy,
return 0;
}
+static inline int page_is_reverse_buddy(struct page *page, struct page *buddy,
+ unsigned int order)
+{
+ if (PageReverseBuddy(buddy) && page_order(buddy) == order) {
+ /*
+ * zone check is done late to avoid uselessly
+ * calculating zone/node ids for pages that could
+ * never merge.
+ */
+ if (page_zone_id(page) != page_zone_id(buddy))
+ return 0;
+
+ VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
+
+ return 1;
+ }
+ return 0;
+}
+
+static inline
+void set_reverse_free_area(struct zone *zone, struct page *page,
+ unsigned int order, int migratetype)
+{
+ unsigned long buddy_pfn;
+ unsigned long reserve_buddy_pfn;
+ struct page *reverse_buddy;
+ struct free_area *area;
+
+ if (order > MAX_ORDER - 3)
+ return;
+ if (unlikely(is_migrate_isolate(migratetype)))
+ return;
+ buddy_pfn = page_to_pfn(page);
+ reserve_buddy_pfn = __find_reverse_buddy_pfn(buddy_pfn, order);
+ if ((buddy_pfn ^ reserve_buddy_pfn) & ~(pageblock_nr_pages - 1))
+ return;
+ if (!pfn_valid_within(reserve_buddy_pfn))
+ return;
+ reverse_buddy = pfn_to_page(reserve_buddy_pfn);
+ if (page_zone_id(page) != page_zone_id(reverse_buddy))
+ return;
+ if (PageBuddy(reverse_buddy) &&
+ page_order(reverse_buddy) == order) {
+ area = &zone->free_area[order];
+ list_del(&page->lru);
+ rmv_page_order(page);
+ area->nr_free--;
+ set_pcppage_migratetype(page, migratetype);
+ list_del(&reverse_buddy->lru);
+ rmv_page_order(reverse_buddy);
+ area->nr_free--;
+ set_pcppage_migratetype(reverse_buddy, migratetype);
+ area++;
+ if (buddy_pfn < reserve_buddy_pfn) {
+ list_add(&page->lru,
+ &area->free_list_reverse[migratetype]);
+ area->nr_free_reverse++;
+ set_reverse_page_order(page, order);
+ set_reverse_page_order(reverse_buddy, order);
+ } else {
+ list_add(&reverse_buddy->lru,
+ &area->free_list_reverse[migratetype]);
+ area->nr_free_reverse++;
+ set_reverse_page_order(reverse_buddy, order);
+ set_reverse_page_order(page, order);
+ }
+ }
+}
+
+static inline
+void rmv_reverse_free_area(struct zone *zone, struct page *page,
+ unsigned int order, int migratetype)
+{
+ unsigned long pfn, buddy_pfn;
+ unsigned long reserve_buddy_pfn;
+ struct page *buddy, *reverse_buddy;
+ struct free_area *area;
+
+ pfn = page_to_pfn(page);
+ buddy_pfn = __find_buddy_index(pfn, order);
+ buddy = page + (buddy_pfn - pfn);
+ if (!pfn_valid_within(buddy_pfn))
+ return;
+ if (!page_is_reverse_buddy(page, buddy, order))
+ return;
+
+ area = &zone->free_area[order];
+ reserve_buddy_pfn = __find_reverse_buddy_pfn(buddy_pfn, order);
+ reverse_buddy = pfn_to_page(reserve_buddy_pfn);
+ VM_BUG_ON_PAGE(!PageReverseBuddy(reverse_buddy) ||
+ page_order(reverse_buddy) != order, reverse_buddy);
+ if (buddy_pfn < reserve_buddy_pfn) {
+ list_move(&buddy->lru, &area->free_list[migratetype]);
+ rmv_reverse_page_order(buddy);
+ area[1].nr_free_reverse--;
+ area->nr_free++;
+ set_page_order(buddy, order);
+ rmv_reverse_page_order(reverse_buddy);
+ list_add(&reverse_buddy->lru, &area->free_list[migratetype]);
+ area->nr_free++;
+ set_page_order(reverse_buddy, order);
+ } else {
+ list_move(&reverse_buddy->lru, &area->free_list[migratetype]);
+ rmv_reverse_page_order(reverse_buddy);
+ area[1].nr_free_reverse--;
+ area->nr_free++;
+ set_page_order(reverse_buddy, order);
+ rmv_reverse_page_order(buddy);
+ list_add(&buddy->lru, &area->free_list[migratetype]);
+ area->nr_free++;
+ set_page_order(buddy, order);
+ }
+}
+
/*
* Freeing function for a buddy system allocator.
*
@@ -805,6 +931,7 @@ static inline void __free_one_page(struct page *page,
VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
VM_BUG_ON_PAGE(bad_range(zone, page), page);
+ rmv_reverse_free_area(zone, page, order, migratetype);
continue_merging:
while (order < max_order - 1) {
buddy_pfn = __find_buddy_pfn(pfn, order);
@@ -882,6 +1009,7 @@ static inline void __free_one_page(struct page *page,
list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
out:
zone->free_area[order].nr_free++;
+ set_reverse_free_area(zone, page, order, migratetype);
}
/*
@@ -1238,14 +1366,24 @@ void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
static void __free_pages_ok(struct page *page, unsigned int order)
{
+ bool reverse_buddy;
int migratetype;
unsigned long pfn = page_to_pfn(page);
if (!free_pages_prepare(page, order, true))
return;
+ reverse_buddy = order &&
+ ((pfn & ((1 << order) - 1)) == (1 << (order - 1)));
migratetype = get_pfnblock_migratetype(page, pfn);
- free_one_page(page_zone(page), page, pfn, order, migratetype);
+ if (!reverse_buddy)
+ free_one_page(page_zone(page), page, pfn, order, migratetype);
+ else {
+ free_one_page(page_zone(page), page,
+ pfn, order - 1, migratetype);
+ free_one_page(page_zone(page), page + (1 << (order - 1)),
+ pfn + (1 << (order - 1)), order - 1, migratetype);
+ }
}
static void __init __free_pages_boot_core(struct page *page, unsigned int order)
@@ -1651,6 +1789,25 @@ static inline void expand(struct zone *zone, struct page *page,
}
}
+static inline void expand_reverse(struct zone *zone, struct page *page,
+ int low, int high, struct free_area *area,
+ int migratetype)
+{
+ struct page *reverse_buddy;
+
+ reverse_buddy = page + (1 << (high - 1));
+ rmv_reverse_page_order(reverse_buddy);
+ set_pcppage_migratetype(reverse_buddy, migratetype);
+ if (high > low) {
+ area--;
+ high--;
+ expand(zone, page, low, high, area, migratetype);
+ list_add(&reverse_buddy->lru, &area->free_list[migratetype]);
+ area->nr_free++;
+ set_page_order(reverse_buddy, high);
+ }
+}
+
static void check_new_page_bad(struct page *page)
{
const char *bad_reason = NULL;
@@ -1785,25 +1942,45 @@ static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags
*/
static inline
struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
- int migratetype)
+ int migratetype, gfp_t gfp_flags)
{
+ bool reverse_buddy;
unsigned int current_order;
struct free_area *area;
struct page *page;
+ reverse_buddy = !!(gfp_flags & __GFP_NOREVERSEBUDDY);
/* Find a page of the appropriate size in the preferred list */
for (current_order = order; current_order < MAX_ORDER; ++current_order) {
area = &(zone->free_area[current_order]);
page = list_first_entry_or_null(&area->free_list[migratetype],
struct page, lru);
- if (!page)
+ if (page) {
+ list_del(&page->lru);
+ rmv_page_order(page);
+ area->nr_free--;
+ expand(zone, page, order,
+ current_order, area, migratetype);
+ set_pcppage_migratetype(page, migratetype);
+ return page;
+ }
+ if (current_order + reverse_buddy == 0 ||
+ current_order + reverse_buddy > MAX_ORDER - 2)
continue;
- list_del(&page->lru);
- rmv_page_order(page);
- area->nr_free--;
- expand(zone, page, order, current_order, area, migratetype);
- set_pcppage_migratetype(page, migratetype);
- return page;
+ area += reverse_buddy;
+ page = list_first_entry_or_null(
+ &area->free_list_reverse[migratetype],
+ struct page, lru);
+ if (page) {
+ list_del(&page->lru);
+ rmv_reverse_page_order(page);
+ area->nr_free_reverse--;
+ expand_reverse(zone, page, order,
+ current_order + reverse_buddy,
+ area, migratetype);
+ set_pcppage_migratetype(page, migratetype);
+ return page;
+ }
}
return NULL;
@@ -1828,13 +2005,13 @@ struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
#ifdef CONFIG_CMA
static struct page *__rmqueue_cma_fallback(struct zone *zone,
- unsigned int order)
+ unsigned int order, gfp_t gfp_flags)
{
- return __rmqueue_smallest(zone, order, MIGRATE_CMA);
+ return __rmqueue_smallest(zone, order, MIGRATE_CMA, gfp_flags);
}
#else
static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
- unsigned int order) { return NULL; }
+ unsigned int order, gfp_t gfp_flags) { return NULL; }
#endif
/*
@@ -2136,7 +2313,8 @@ static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
/* Remove an element from the buddy allocator from the fallback list */
static inline struct page *
-__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
+__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype,
+ gfp_t gfp_flags)
{
struct free_area *area;
unsigned int current_order;
@@ -2190,17 +2368,18 @@ static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
* Call me with the zone->lock already held.
*/
static struct page *__rmqueue(struct zone *zone, unsigned int order,
- int migratetype)
+ int migratetype, gfp_t gfp_flags)
{
struct page *page;
- page = __rmqueue_smallest(zone, order, migratetype);
+ page = __rmqueue_smallest(zone, order, migratetype, gfp_flags);
if (unlikely(!page)) {
if (migratetype == MIGRATE_MOVABLE)
- page = __rmqueue_cma_fallback(zone, order);
+ page = __rmqueue_cma_fallback(zone, order, gfp_flags);
if (!page)
- page = __rmqueue_fallback(zone, order, migratetype);
+ page = __rmqueue_fallback(zone, order, migratetype,
+ gfp_flags);
}
trace_mm_page_alloc_zone_locked(page, order, migratetype);
@@ -2221,7 +2400,7 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order,
spin_lock_irqsave(&zone->lock, flags);
for (i = 0; i < count; ++i) {
- struct page *page = __rmqueue(zone, order, migratetype);
+ struct page *page = __rmqueue(zone, order, migratetype, 0);
if (unlikely(page == NULL))
break;
@@ -2718,12 +2897,13 @@ struct page *rmqueue(struct zone *preferred_zone,
do {
page = NULL;
if (alloc_flags & ALLOC_HARDER) {
- page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
+ page = __rmqueue_smallest(zone, order,
+ MIGRATE_HIGHATOMIC, gfp_flags);
if (page)
trace_mm_page_alloc_zone_locked(page, order, migratetype);
}
if (!page)
- page = __rmqueue(zone, order, migratetype);
+ page = __rmqueue(zone, order, migratetype, gfp_flags);
} while (page && check_new_pages(page, order));
spin_unlock(&zone->lock);
if (!page)
@@ -5286,7 +5466,9 @@ static void __meminit zone_init_free_lists(struct zone *zone)
unsigned int order, t;
for_each_migratetype_order(order, t) {
INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
+ INIT_LIST_HEAD(&zone->free_area[order].free_list_reverse[t]);
zone->free_area[order].nr_free = 0;
+ zone->free_area[order].nr_free_reverse = 0;
}
}
diff --git a/mm/vmstat.c b/mm/vmstat.c
index 69f9aff..26007df 100644
--- a/mm/vmstat.c
+++ b/mm/vmstat.c
@@ -1146,10 +1146,13 @@ static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
{
int order;
- seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
+ seq_printf(m, "Node %d, zone %8s\n", pgdat->node_id, zone->name);
for (order = 0; order < MAX_ORDER; ++order)
seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
seq_putc(m, '\n');
+ for (order = 0; order < MAX_ORDER; ++order)
+ seq_printf(m, "%6lu ", zone->free_area[order].nr_free_reverse);
+ seq_putc(m, '\n');
}
/*
--
1.7.9.5
--
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>
^ permalink raw reply related [flat|nested] 12+ messages in thread
* Re: [PATCH mm] introduce reverse buddy concept to reduce buddy fragment
2017-06-30 11:25 [PATCH mm] introduce reverse buddy concept to reduce buddy fragment zhouxianrong
@ 2017-07-03 7:48 ` Michal Hocko
2017-07-03 9:01 ` zhouxianrong
2017-07-03 12:02 ` zhouxianrong
0 siblings, 2 replies; 12+ messages in thread
From: Michal Hocko @ 2017-07-03 7:48 UTC (permalink / raw)
To: zhouxianrong
Cc: linux-mm, linux-kernel, akpm, vbabka, alexander.h.duyck, mgorman,
l.stach, vdavydov.dev, hannes, minchan, npiggin, kirill.shutemov,
gi-oh.kim, luto, keescook, mark.rutland, mingo, heiko.carstens,
iamjoonsoo.kim, rientjes, ming.ling, jack, ebru.akagunduz,
bigeasy, Mi.Sophia.Wang, zhouxiyu, weidu.du, fanghua3,
won.ho.park
On Fri 30-06-17 19:25:41, zhouxianrong@huawei.com wrote:
> From: zhouxianrong <zhouxianrong@huawei.com>
>
> when buddy is under fragment i find that still there are some pages
> just like AFFA mode. A is allocated, F is free, AF is buddy pair for
> oder n, FA is buddy pair for oder n as well.
Could you quantify how often does this happen and how much of a problem
this actually is? Is there any specific workload that would suffer from
such an artificial fragmentation?
> I want to compse the
> FF as oder n + 1 and align to n other than n + 1. this patch broke
> the rules of buddy stated as alignment to its length of oder. i think
> we can do so except for kernel stack because the requirement comes from
> buddy attribution rather than user.
Why do you think the stack is a problem here?
> for kernel stack requirement i add
> __GFP_NOREVERSEBUDDY for this purpose.
>
> a sample just like blow.
>
> Node 0, zone DMA
> 1389 1765 342 272 2 0 0 0 0 0 0
> 0 75 4398 1560 379 27 2 0 0 0 0
> Node 0, zone Normal
> 20 24 14 2 0 0 0 0 0 0 0
> 0 6 228 3 0 0 0 0 0 0 0
>
> the patch does not consider fallback allocation for now.
The path is missing the crucial information required for any
optimization. Some numbers to compare how much it helps. The above
output of buddyinfo is pointless without any base to compare to. Also
which workloads would benefit from this change and how much? It is also
a non trivial amount of code in the guts of the page allocator so this
really needs _much_ better explanation.
I haven't looked closely on the code yet but a quick look at
set_reverse_free_area scared me away.
> Signed-off-by: zhouxianrong <zhouxianrong@huawei.com>
> ---
> include/linux/gfp.h | 8 +-
> include/linux/mmzone.h | 2 +
> include/linux/page-flags.h | 9 ++
> include/linux/thread_info.h | 5 +-
> mm/compaction.c | 17 ++++
> mm/internal.h | 7 ++
> mm/page_alloc.c | 222 +++++++++++++++++++++++++++++++++++++++----
> mm/vmstat.c | 5 +-
> 8 files changed, 251 insertions(+), 24 deletions(-)
>
> diff --git a/include/linux/gfp.h b/include/linux/gfp.h
> index db373b9..f63d4d9 100644
> --- a/include/linux/gfp.h
> +++ b/include/linux/gfp.h
> @@ -40,6 +40,7 @@
> #define ___GFP_DIRECT_RECLAIM 0x400000u
> #define ___GFP_WRITE 0x800000u
> #define ___GFP_KSWAPD_RECLAIM 0x1000000u
> +#define ___GFP_NOREVERSEBUDDY 0x2000000u
> /* If the above are modified, __GFP_BITS_SHIFT may need updating */
>
> /*
> @@ -171,6 +172,10 @@
> * __GFP_NOTRACK_FALSE_POSITIVE is an alias of __GFP_NOTRACK. It's a means of
> * distinguishing in the source between false positives and allocations that
> * cannot be supported (e.g. page tables).
> + *
> + * __GFP_NOREVERSEBUDDY does not allocate pages from reverse buddy list
> + * of current order. It make sure that allocation is alignment to same order
> + * with length order.
> */
> #define __GFP_COLD ((__force gfp_t)___GFP_COLD)
> #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
> @@ -178,9 +183,10 @@
> #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
> #define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK)
> #define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
> +#define __GFP_NOREVERSEBUDDY ((__force gfp_t)___GFP_NOREVERSEBUDDY)
>
> /* Room for N __GFP_FOO bits */
> -#define __GFP_BITS_SHIFT 25
> +#define __GFP_BITS_SHIFT 26
> #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
>
> /*
> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
> index 8e02b37..94237fe 100644
> --- a/include/linux/mmzone.h
> +++ b/include/linux/mmzone.h
> @@ -89,7 +89,9 @@ enum {
>
> struct free_area {
> struct list_head free_list[MIGRATE_TYPES];
> + struct list_head free_list_reverse[MIGRATE_TYPES];
> unsigned long nr_free;
> + unsigned long nr_free_reverse;
> };
>
> struct pglist_data;
> diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> index 6b5818d..39d17d7 100644
> --- a/include/linux/page-flags.h
> +++ b/include/linux/page-flags.h
> @@ -675,6 +675,15 @@ static inline int TestClearPageDoubleMap(struct page *page)
> #define PAGE_KMEMCG_MAPCOUNT_VALUE (-512)
> PAGE_MAPCOUNT_OPS(Kmemcg, KMEMCG)
>
> +/*
> + * ReverseBuddy is enabled for the buddy allocator that allow allocating
> + * two adjacent same free order blocks other than buddy blocks and
> + * composing them as a order + 1 block. It is for reducing buddy
> + * fragment.
> + */
> +#define PAGE_REVERSE_BUDDY_MAPCOUNT_VALUE (-1024)
> +PAGE_MAPCOUNT_OPS(ReverseBuddy, REVERSE_BUDDY)
> +
> extern bool is_free_buddy_page(struct page *page);
>
> __PAGEFLAG(Isolated, isolated, PF_ANY);
> diff --git a/include/linux/thread_info.h b/include/linux/thread_info.h
> index 5837387..b4a1605 100644
> --- a/include/linux/thread_info.h
> +++ b/include/linux/thread_info.h
> @@ -28,9 +28,10 @@
>
> #ifdef CONFIG_DEBUG_STACK_USAGE
> # define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK | \
> - __GFP_ZERO)
> + __GFP_NOREVERSEBUDDY | __GFP_ZERO)
> #else
> -# define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK)
> +# define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK | \
> + __GFP_NOREVERSEBUDDY)
> #endif
>
> /*
> diff --git a/mm/compaction.c b/mm/compaction.c
> index 0fdfde0..a43f169 100644
> --- a/mm/compaction.c
> +++ b/mm/compaction.c
> @@ -768,6 +768,20 @@ static bool too_many_isolated(struct zone *zone)
> continue;
> }
>
> + if (PageReverseBuddy(page)) {
> + unsigned long freepage_order = page_order_unsafe(page);
> +
> + /*
> + * Without lock, we cannot be sure that what we got is
> + * a valid page order. Consider only values in the
> + * valid order range to prevent low_pfn overflow.
> + */
> + if (freepage_order > 0 &&
> + freepage_order < MAX_ORDER - 1)
> + low_pfn += (1UL << (freepage_order + 1)) - 1;
> + continue;
> + }
> +
> /*
> * Regardless of being on LRU, compound pages such as THP and
> * hugetlbfs are not to be compacted. We can potentially save
> @@ -1005,6 +1019,9 @@ static bool suitable_migration_target(struct compact_control *cc,
> return false;
> }
>
> + if (PageReverseBuddy(page))
> + return false;
> +
> /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
> if (migrate_async_suitable(get_pageblock_migratetype(page)))
> return true;
> diff --git a/mm/internal.h b/mm/internal.h
> index ccfc2a2..439b0a8 100644
> --- a/mm/internal.h
> +++ b/mm/internal.h
> @@ -143,6 +143,13 @@ struct alloc_context {
> return page_pfn ^ (1 << order);
> }
>
> +static inline unsigned long
> +__find_reverse_buddy_pfn(unsigned long page_pfn, unsigned int order)
> +{
> + return (page_pfn & (1 << order)) ? page_pfn + (1 << order) :
> + page_pfn - (1 << order);
> +}
> +
> extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
> unsigned long end_pfn, struct zone *zone);
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 9f9623d..ee1dc1b 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -715,6 +715,18 @@ static inline void rmv_page_order(struct page *page)
> set_page_private(page, 0);
> }
>
> +static inline void set_reverse_page_order(struct page *page, unsigned int order)
> +{
> + set_page_private(page, order);
> + __SetPageReverseBuddy(page);
> +}
> +
> +static inline void rmv_reverse_page_order(struct page *page)
> +{
> + __ClearPageReverseBuddy(page);
> + set_page_private(page, 0);
> +}
> +
> /*
> * This function checks whether a page is free && is the buddy
> * we can do coalesce a page and its buddy if
> @@ -758,6 +770,120 @@ static inline int page_is_buddy(struct page *page, struct page *buddy,
> return 0;
> }
>
> +static inline int page_is_reverse_buddy(struct page *page, struct page *buddy,
> + unsigned int order)
> +{
> + if (PageReverseBuddy(buddy) && page_order(buddy) == order) {
> + /*
> + * zone check is done late to avoid uselessly
> + * calculating zone/node ids for pages that could
> + * never merge.
> + */
> + if (page_zone_id(page) != page_zone_id(buddy))
> + return 0;
> +
> + VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
> +
> + return 1;
> + }
> + return 0;
> +}
> +
> +static inline
> +void set_reverse_free_area(struct zone *zone, struct page *page,
> + unsigned int order, int migratetype)
> +{
> + unsigned long buddy_pfn;
> + unsigned long reserve_buddy_pfn;
> + struct page *reverse_buddy;
> + struct free_area *area;
> +
> + if (order > MAX_ORDER - 3)
> + return;
> + if (unlikely(is_migrate_isolate(migratetype)))
> + return;
> + buddy_pfn = page_to_pfn(page);
> + reserve_buddy_pfn = __find_reverse_buddy_pfn(buddy_pfn, order);
> + if ((buddy_pfn ^ reserve_buddy_pfn) & ~(pageblock_nr_pages - 1))
> + return;
> + if (!pfn_valid_within(reserve_buddy_pfn))
> + return;
> + reverse_buddy = pfn_to_page(reserve_buddy_pfn);
> + if (page_zone_id(page) != page_zone_id(reverse_buddy))
> + return;
> + if (PageBuddy(reverse_buddy) &&
> + page_order(reverse_buddy) == order) {
> + area = &zone->free_area[order];
> + list_del(&page->lru);
> + rmv_page_order(page);
> + area->nr_free--;
> + set_pcppage_migratetype(page, migratetype);
> + list_del(&reverse_buddy->lru);
> + rmv_page_order(reverse_buddy);
> + area->nr_free--;
> + set_pcppage_migratetype(reverse_buddy, migratetype);
> + area++;
> + if (buddy_pfn < reserve_buddy_pfn) {
> + list_add(&page->lru,
> + &area->free_list_reverse[migratetype]);
> + area->nr_free_reverse++;
> + set_reverse_page_order(page, order);
> + set_reverse_page_order(reverse_buddy, order);
> + } else {
> + list_add(&reverse_buddy->lru,
> + &area->free_list_reverse[migratetype]);
> + area->nr_free_reverse++;
> + set_reverse_page_order(reverse_buddy, order);
> + set_reverse_page_order(page, order);
> + }
> + }
> +}
> +
> +static inline
> +void rmv_reverse_free_area(struct zone *zone, struct page *page,
> + unsigned int order, int migratetype)
> +{
> + unsigned long pfn, buddy_pfn;
> + unsigned long reserve_buddy_pfn;
> + struct page *buddy, *reverse_buddy;
> + struct free_area *area;
> +
> + pfn = page_to_pfn(page);
> + buddy_pfn = __find_buddy_index(pfn, order);
> + buddy = page + (buddy_pfn - pfn);
> + if (!pfn_valid_within(buddy_pfn))
> + return;
> + if (!page_is_reverse_buddy(page, buddy, order))
> + return;
> +
> + area = &zone->free_area[order];
> + reserve_buddy_pfn = __find_reverse_buddy_pfn(buddy_pfn, order);
> + reverse_buddy = pfn_to_page(reserve_buddy_pfn);
> + VM_BUG_ON_PAGE(!PageReverseBuddy(reverse_buddy) ||
> + page_order(reverse_buddy) != order, reverse_buddy);
> + if (buddy_pfn < reserve_buddy_pfn) {
> + list_move(&buddy->lru, &area->free_list[migratetype]);
> + rmv_reverse_page_order(buddy);
> + area[1].nr_free_reverse--;
> + area->nr_free++;
> + set_page_order(buddy, order);
> + rmv_reverse_page_order(reverse_buddy);
> + list_add(&reverse_buddy->lru, &area->free_list[migratetype]);
> + area->nr_free++;
> + set_page_order(reverse_buddy, order);
> + } else {
> + list_move(&reverse_buddy->lru, &area->free_list[migratetype]);
> + rmv_reverse_page_order(reverse_buddy);
> + area[1].nr_free_reverse--;
> + area->nr_free++;
> + set_page_order(reverse_buddy, order);
> + rmv_reverse_page_order(buddy);
> + list_add(&buddy->lru, &area->free_list[migratetype]);
> + area->nr_free++;
> + set_page_order(buddy, order);
> + }
> +}
> +
> /*
> * Freeing function for a buddy system allocator.
> *
> @@ -805,6 +931,7 @@ static inline void __free_one_page(struct page *page,
> VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
> VM_BUG_ON_PAGE(bad_range(zone, page), page);
>
> + rmv_reverse_free_area(zone, page, order, migratetype);
> continue_merging:
> while (order < max_order - 1) {
> buddy_pfn = __find_buddy_pfn(pfn, order);
> @@ -882,6 +1009,7 @@ static inline void __free_one_page(struct page *page,
> list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
> out:
> zone->free_area[order].nr_free++;
> + set_reverse_free_area(zone, page, order, migratetype);
> }
>
> /*
> @@ -1238,14 +1366,24 @@ void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
>
> static void __free_pages_ok(struct page *page, unsigned int order)
> {
> + bool reverse_buddy;
> int migratetype;
> unsigned long pfn = page_to_pfn(page);
>
> if (!free_pages_prepare(page, order, true))
> return;
>
> + reverse_buddy = order &&
> + ((pfn & ((1 << order) - 1)) == (1 << (order - 1)));
> migratetype = get_pfnblock_migratetype(page, pfn);
> - free_one_page(page_zone(page), page, pfn, order, migratetype);
> + if (!reverse_buddy)
> + free_one_page(page_zone(page), page, pfn, order, migratetype);
> + else {
> + free_one_page(page_zone(page), page,
> + pfn, order - 1, migratetype);
> + free_one_page(page_zone(page), page + (1 << (order - 1)),
> + pfn + (1 << (order - 1)), order - 1, migratetype);
> + }
> }
>
> static void __init __free_pages_boot_core(struct page *page, unsigned int order)
> @@ -1651,6 +1789,25 @@ static inline void expand(struct zone *zone, struct page *page,
> }
> }
>
> +static inline void expand_reverse(struct zone *zone, struct page *page,
> + int low, int high, struct free_area *area,
> + int migratetype)
> +{
> + struct page *reverse_buddy;
> +
> + reverse_buddy = page + (1 << (high - 1));
> + rmv_reverse_page_order(reverse_buddy);
> + set_pcppage_migratetype(reverse_buddy, migratetype);
> + if (high > low) {
> + area--;
> + high--;
> + expand(zone, page, low, high, area, migratetype);
> + list_add(&reverse_buddy->lru, &area->free_list[migratetype]);
> + area->nr_free++;
> + set_page_order(reverse_buddy, high);
> + }
> +}
> +
> static void check_new_page_bad(struct page *page)
> {
> const char *bad_reason = NULL;
> @@ -1785,25 +1942,45 @@ static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags
> */
> static inline
> struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
> - int migratetype)
> + int migratetype, gfp_t gfp_flags)
> {
> + bool reverse_buddy;
> unsigned int current_order;
> struct free_area *area;
> struct page *page;
>
> + reverse_buddy = !!(gfp_flags & __GFP_NOREVERSEBUDDY);
> /* Find a page of the appropriate size in the preferred list */
> for (current_order = order; current_order < MAX_ORDER; ++current_order) {
> area = &(zone->free_area[current_order]);
> page = list_first_entry_or_null(&area->free_list[migratetype],
> struct page, lru);
> - if (!page)
> + if (page) {
> + list_del(&page->lru);
> + rmv_page_order(page);
> + area->nr_free--;
> + expand(zone, page, order,
> + current_order, area, migratetype);
> + set_pcppage_migratetype(page, migratetype);
> + return page;
> + }
> + if (current_order + reverse_buddy == 0 ||
> + current_order + reverse_buddy > MAX_ORDER - 2)
> continue;
> - list_del(&page->lru);
> - rmv_page_order(page);
> - area->nr_free--;
> - expand(zone, page, order, current_order, area, migratetype);
> - set_pcppage_migratetype(page, migratetype);
> - return page;
> + area += reverse_buddy;
> + page = list_first_entry_or_null(
> + &area->free_list_reverse[migratetype],
> + struct page, lru);
> + if (page) {
> + list_del(&page->lru);
> + rmv_reverse_page_order(page);
> + area->nr_free_reverse--;
> + expand_reverse(zone, page, order,
> + current_order + reverse_buddy,
> + area, migratetype);
> + set_pcppage_migratetype(page, migratetype);
> + return page;
> + }
> }
>
> return NULL;
> @@ -1828,13 +2005,13 @@ struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
>
> #ifdef CONFIG_CMA
> static struct page *__rmqueue_cma_fallback(struct zone *zone,
> - unsigned int order)
> + unsigned int order, gfp_t gfp_flags)
> {
> - return __rmqueue_smallest(zone, order, MIGRATE_CMA);
> + return __rmqueue_smallest(zone, order, MIGRATE_CMA, gfp_flags);
> }
> #else
> static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
> - unsigned int order) { return NULL; }
> + unsigned int order, gfp_t gfp_flags) { return NULL; }
> #endif
>
> /*
> @@ -2136,7 +2313,8 @@ static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
>
> /* Remove an element from the buddy allocator from the fallback list */
> static inline struct page *
> -__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
> +__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype,
> + gfp_t gfp_flags)
> {
> struct free_area *area;
> unsigned int current_order;
> @@ -2190,17 +2368,18 @@ static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
> * Call me with the zone->lock already held.
> */
> static struct page *__rmqueue(struct zone *zone, unsigned int order,
> - int migratetype)
> + int migratetype, gfp_t gfp_flags)
> {
> struct page *page;
>
> - page = __rmqueue_smallest(zone, order, migratetype);
> + page = __rmqueue_smallest(zone, order, migratetype, gfp_flags);
> if (unlikely(!page)) {
> if (migratetype == MIGRATE_MOVABLE)
> - page = __rmqueue_cma_fallback(zone, order);
> + page = __rmqueue_cma_fallback(zone, order, gfp_flags);
>
> if (!page)
> - page = __rmqueue_fallback(zone, order, migratetype);
> + page = __rmqueue_fallback(zone, order, migratetype,
> + gfp_flags);
> }
>
> trace_mm_page_alloc_zone_locked(page, order, migratetype);
> @@ -2221,7 +2400,7 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order,
>
> spin_lock_irqsave(&zone->lock, flags);
> for (i = 0; i < count; ++i) {
> - struct page *page = __rmqueue(zone, order, migratetype);
> + struct page *page = __rmqueue(zone, order, migratetype, 0);
> if (unlikely(page == NULL))
> break;
>
> @@ -2718,12 +2897,13 @@ struct page *rmqueue(struct zone *preferred_zone,
> do {
> page = NULL;
> if (alloc_flags & ALLOC_HARDER) {
> - page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
> + page = __rmqueue_smallest(zone, order,
> + MIGRATE_HIGHATOMIC, gfp_flags);
> if (page)
> trace_mm_page_alloc_zone_locked(page, order, migratetype);
> }
> if (!page)
> - page = __rmqueue(zone, order, migratetype);
> + page = __rmqueue(zone, order, migratetype, gfp_flags);
> } while (page && check_new_pages(page, order));
> spin_unlock(&zone->lock);
> if (!page)
> @@ -5286,7 +5466,9 @@ static void __meminit zone_init_free_lists(struct zone *zone)
> unsigned int order, t;
> for_each_migratetype_order(order, t) {
> INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
> + INIT_LIST_HEAD(&zone->free_area[order].free_list_reverse[t]);
> zone->free_area[order].nr_free = 0;
> + zone->free_area[order].nr_free_reverse = 0;
> }
> }
>
> diff --git a/mm/vmstat.c b/mm/vmstat.c
> index 69f9aff..26007df 100644
> --- a/mm/vmstat.c
> +++ b/mm/vmstat.c
> @@ -1146,10 +1146,13 @@ static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
> {
> int order;
>
> - seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
> + seq_printf(m, "Node %d, zone %8s\n", pgdat->node_id, zone->name);
> for (order = 0; order < MAX_ORDER; ++order)
> seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
> seq_putc(m, '\n');
> + for (order = 0; order < MAX_ORDER; ++order)
> + seq_printf(m, "%6lu ", zone->free_area[order].nr_free_reverse);
> + seq_putc(m, '\n');
> }
>
> /*
> --
> 1.7.9.5
>
--
Michal Hocko
SUSE Labs
--
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>
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH mm] introduce reverse buddy concept to reduce buddy fragment
2017-07-03 7:48 ` Michal Hocko
@ 2017-07-03 9:01 ` zhouxianrong
2017-07-03 12:02 ` zhouxianrong
1 sibling, 0 replies; 12+ messages in thread
From: zhouxianrong @ 2017-07-03 9:01 UTC (permalink / raw)
To: Michal Hocko
Cc: linux-mm, linux-kernel, akpm, vbabka, alexander.h.duyck, mgorman,
l.stach, vdavydov.dev, hannes, minchan, npiggin, kirill.shutemov,
gi-oh.kim, luto, keescook, mark.rutland, mingo, heiko.carstens,
iamjoonsoo.kim, rientjes, ming.ling, jack, ebru.akagunduz,
bigeasy, Mi.Sophia.Wang, zhouxiyu, weidu.du, fanghua3,
won.ho.park
On 2017/7/3 15:48, Michal Hocko wrote:
> On Fri 30-06-17 19:25:41, zhouxianrong@huawei.com wrote:
>> From: zhouxianrong <zhouxianrong@huawei.com>
>>
>> when buddy is under fragment i find that still there are some pages
>> just like AFFA mode. A is allocated, F is free, AF is buddy pair for
>> oder n, FA is buddy pair for oder n as well.
>
> Could you quantify how often does this happen and how much of a problem
> this actually is? Is there any specific workload that would suffer from
> such an artificial fragmentation?
>
>> I want to compse the
>> FF as oder n + 1 and align to n other than n + 1. this patch broke
>> the rules of buddy stated as alignment to its length of oder. i think
>> we can do so except for kernel stack because the requirement comes from
>> buddy attribution rather than user.
>
> Why do you think the stack is a problem here?
>
>> for kernel stack requirement i add
>> __GFP_NOREVERSEBUDDY for this purpose.
>>
>> a sample just like blow.
>>
>> Node 0, zone DMA
>> 1389 1765 342 272 2 0 0 0 0 0 0
>> 0 75 4398 1560 379 27 2 0 0 0 0
>> Node 0, zone Normal
>> 20 24 14 2 0 0 0 0 0 0 0
>> 0 6 228 3 0 0 0 0 0 0 0
>>
at the sample moment if we have not this patch, the aspect should like below:
Node 0, zone DMA
(1389 + 75 * 2) (1765 + 4398 * 2) (342 + 1560 * 2) (272 + 379 * 2) (2 + 27 * 2) (0 + 2 * 2) 0 0 0 0 0
Node 0, zone Normal
(20 + 6 * 2) (24 + 228 * 2) (14 + 3 * 2) 2 0 0 0 0 0 0 0
i find out AFFA mode in lower order free_list and move FF into higher order free_list_reverse.
>> the patch does not consider fallback allocation for now.
>
> The path is missing the crucial information required for any
> optimization. Some numbers to compare how much it helps. The above
> output of buddyinfo is pointless without any base to compare to. Also
> which workloads would benefit from this change and how much? It is also
> a non trivial amount of code in the guts of the page allocator so this
> really needs _much_ better explanation.
>
> I haven't looked closely on the code yet but a quick look at
> set_reverse_free_area scared me away.
>
>> Signed-off-by: zhouxianrong <zhouxianrong@huawei.com>
>> ---
>> include/linux/gfp.h | 8 +-
>> include/linux/mmzone.h | 2 +
>> include/linux/page-flags.h | 9 ++
>> include/linux/thread_info.h | 5 +-
>> mm/compaction.c | 17 ++++
>> mm/internal.h | 7 ++
>> mm/page_alloc.c | 222 +++++++++++++++++++++++++++++++++++++++----
>> mm/vmstat.c | 5 +-
>> 8 files changed, 251 insertions(+), 24 deletions(-)
>>
>> diff --git a/include/linux/gfp.h b/include/linux/gfp.h
>> index db373b9..f63d4d9 100644
>> --- a/include/linux/gfp.h
>> +++ b/include/linux/gfp.h
>> @@ -40,6 +40,7 @@
>> #define ___GFP_DIRECT_RECLAIM 0x400000u
>> #define ___GFP_WRITE 0x800000u
>> #define ___GFP_KSWAPD_RECLAIM 0x1000000u
>> +#define ___GFP_NOREVERSEBUDDY 0x2000000u
>> /* If the above are modified, __GFP_BITS_SHIFT may need updating */
>>
>> /*
>> @@ -171,6 +172,10 @@
>> * __GFP_NOTRACK_FALSE_POSITIVE is an alias of __GFP_NOTRACK. It's a means of
>> * distinguishing in the source between false positives and allocations that
>> * cannot be supported (e.g. page tables).
>> + *
>> + * __GFP_NOREVERSEBUDDY does not allocate pages from reverse buddy list
>> + * of current order. It make sure that allocation is alignment to same order
>> + * with length order.
>> */
>> #define __GFP_COLD ((__force gfp_t)___GFP_COLD)
>> #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
>> @@ -178,9 +183,10 @@
>> #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
>> #define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK)
>> #define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
>> +#define __GFP_NOREVERSEBUDDY ((__force gfp_t)___GFP_NOREVERSEBUDDY)
>>
>> /* Room for N __GFP_FOO bits */
>> -#define __GFP_BITS_SHIFT 25
>> +#define __GFP_BITS_SHIFT 26
>> #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
>>
>> /*
>> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
>> index 8e02b37..94237fe 100644
>> --- a/include/linux/mmzone.h
>> +++ b/include/linux/mmzone.h
>> @@ -89,7 +89,9 @@ enum {
>>
>> struct free_area {
>> struct list_head free_list[MIGRATE_TYPES];
>> + struct list_head free_list_reverse[MIGRATE_TYPES];
>> unsigned long nr_free;
>> + unsigned long nr_free_reverse;
>> };
>>
>> struct pglist_data;
>> diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
>> index 6b5818d..39d17d7 100644
>> --- a/include/linux/page-flags.h
>> +++ b/include/linux/page-flags.h
>> @@ -675,6 +675,15 @@ static inline int TestClearPageDoubleMap(struct page *page)
>> #define PAGE_KMEMCG_MAPCOUNT_VALUE (-512)
>> PAGE_MAPCOUNT_OPS(Kmemcg, KMEMCG)
>>
>> +/*
>> + * ReverseBuddy is enabled for the buddy allocator that allow allocating
>> + * two adjacent same free order blocks other than buddy blocks and
>> + * composing them as a order + 1 block. It is for reducing buddy
>> + * fragment.
>> + */
>> +#define PAGE_REVERSE_BUDDY_MAPCOUNT_VALUE (-1024)
>> +PAGE_MAPCOUNT_OPS(ReverseBuddy, REVERSE_BUDDY)
>> +
>> extern bool is_free_buddy_page(struct page *page);
>>
>> __PAGEFLAG(Isolated, isolated, PF_ANY);
>> diff --git a/include/linux/thread_info.h b/include/linux/thread_info.h
>> index 5837387..b4a1605 100644
>> --- a/include/linux/thread_info.h
>> +++ b/include/linux/thread_info.h
>> @@ -28,9 +28,10 @@
>>
>> #ifdef CONFIG_DEBUG_STACK_USAGE
>> # define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK | \
>> - __GFP_ZERO)
>> + __GFP_NOREVERSEBUDDY | __GFP_ZERO)
>> #else
>> -# define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK)
>> +# define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK | \
>> + __GFP_NOREVERSEBUDDY)
>> #endif
>>
>> /*
>> diff --git a/mm/compaction.c b/mm/compaction.c
>> index 0fdfde0..a43f169 100644
>> --- a/mm/compaction.c
>> +++ b/mm/compaction.c
>> @@ -768,6 +768,20 @@ static bool too_many_isolated(struct zone *zone)
>> continue;
>> }
>>
>> + if (PageReverseBuddy(page)) {
>> + unsigned long freepage_order = page_order_unsafe(page);
>> +
>> + /*
>> + * Without lock, we cannot be sure that what we got is
>> + * a valid page order. Consider only values in the
>> + * valid order range to prevent low_pfn overflow.
>> + */
>> + if (freepage_order > 0 &&
>> + freepage_order < MAX_ORDER - 1)
>> + low_pfn += (1UL << (freepage_order + 1)) - 1;
>> + continue;
>> + }
>> +
>> /*
>> * Regardless of being on LRU, compound pages such as THP and
>> * hugetlbfs are not to be compacted. We can potentially save
>> @@ -1005,6 +1019,9 @@ static bool suitable_migration_target(struct compact_control *cc,
>> return false;
>> }
>>
>> + if (PageReverseBuddy(page))
>> + return false;
>> +
>> /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
>> if (migrate_async_suitable(get_pageblock_migratetype(page)))
>> return true;
>> diff --git a/mm/internal.h b/mm/internal.h
>> index ccfc2a2..439b0a8 100644
>> --- a/mm/internal.h
>> +++ b/mm/internal.h
>> @@ -143,6 +143,13 @@ struct alloc_context {
>> return page_pfn ^ (1 << order);
>> }
>>
>> +static inline unsigned long
>> +__find_reverse_buddy_pfn(unsigned long page_pfn, unsigned int order)
>> +{
>> + return (page_pfn & (1 << order)) ? page_pfn + (1 << order) :
>> + page_pfn - (1 << order);
>> +}
>> +
>> extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
>> unsigned long end_pfn, struct zone *zone);
>>
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index 9f9623d..ee1dc1b 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -715,6 +715,18 @@ static inline void rmv_page_order(struct page *page)
>> set_page_private(page, 0);
>> }
>>
>> +static inline void set_reverse_page_order(struct page *page, unsigned int order)
>> +{
>> + set_page_private(page, order);
>> + __SetPageReverseBuddy(page);
>> +}
>> +
>> +static inline void rmv_reverse_page_order(struct page *page)
>> +{
>> + __ClearPageReverseBuddy(page);
>> + set_page_private(page, 0);
>> +}
>> +
>> /*
>> * This function checks whether a page is free && is the buddy
>> * we can do coalesce a page and its buddy if
>> @@ -758,6 +770,120 @@ static inline int page_is_buddy(struct page *page, struct page *buddy,
>> return 0;
>> }
>>
>> +static inline int page_is_reverse_buddy(struct page *page, struct page *buddy,
>> + unsigned int order)
>> +{
>> + if (PageReverseBuddy(buddy) && page_order(buddy) == order) {
>> + /*
>> + * zone check is done late to avoid uselessly
>> + * calculating zone/node ids for pages that could
>> + * never merge.
>> + */
>> + if (page_zone_id(page) != page_zone_id(buddy))
>> + return 0;
>> +
>> + VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
>> +
>> + return 1;
>> + }
>> + return 0;
>> +}
>> +
>> +static inline
>> +void set_reverse_free_area(struct zone *zone, struct page *page,
>> + unsigned int order, int migratetype)
>> +{
>> + unsigned long buddy_pfn;
>> + unsigned long reserve_buddy_pfn;
>> + struct page *reverse_buddy;
>> + struct free_area *area;
>> +
>> + if (order > MAX_ORDER - 3)
>> + return;
>> + if (unlikely(is_migrate_isolate(migratetype)))
>> + return;
>> + buddy_pfn = page_to_pfn(page);
>> + reserve_buddy_pfn = __find_reverse_buddy_pfn(buddy_pfn, order);
>> + if ((buddy_pfn ^ reserve_buddy_pfn) & ~(pageblock_nr_pages - 1))
>> + return;
>> + if (!pfn_valid_within(reserve_buddy_pfn))
>> + return;
>> + reverse_buddy = pfn_to_page(reserve_buddy_pfn);
>> + if (page_zone_id(page) != page_zone_id(reverse_buddy))
>> + return;
>> + if (PageBuddy(reverse_buddy) &&
>> + page_order(reverse_buddy) == order) {
>> + area = &zone->free_area[order];
>> + list_del(&page->lru);
>> + rmv_page_order(page);
>> + area->nr_free--;
>> + set_pcppage_migratetype(page, migratetype);
>> + list_del(&reverse_buddy->lru);
>> + rmv_page_order(reverse_buddy);
>> + area->nr_free--;
>> + set_pcppage_migratetype(reverse_buddy, migratetype);
>> + area++;
>> + if (buddy_pfn < reserve_buddy_pfn) {
>> + list_add(&page->lru,
>> + &area->free_list_reverse[migratetype]);
>> + area->nr_free_reverse++;
>> + set_reverse_page_order(page, order);
>> + set_reverse_page_order(reverse_buddy, order);
>> + } else {
>> + list_add(&reverse_buddy->lru,
>> + &area->free_list_reverse[migratetype]);
>> + area->nr_free_reverse++;
>> + set_reverse_page_order(reverse_buddy, order);
>> + set_reverse_page_order(page, order);
>> + }
>> + }
>> +}
>> +
>> +static inline
>> +void rmv_reverse_free_area(struct zone *zone, struct page *page,
>> + unsigned int order, int migratetype)
>> +{
>> + unsigned long pfn, buddy_pfn;
>> + unsigned long reserve_buddy_pfn;
>> + struct page *buddy, *reverse_buddy;
>> + struct free_area *area;
>> +
>> + pfn = page_to_pfn(page);
>> + buddy_pfn = __find_buddy_index(pfn, order);
>> + buddy = page + (buddy_pfn - pfn);
>> + if (!pfn_valid_within(buddy_pfn))
>> + return;
>> + if (!page_is_reverse_buddy(page, buddy, order))
>> + return;
>> +
>> + area = &zone->free_area[order];
>> + reserve_buddy_pfn = __find_reverse_buddy_pfn(buddy_pfn, order);
>> + reverse_buddy = pfn_to_page(reserve_buddy_pfn);
>> + VM_BUG_ON_PAGE(!PageReverseBuddy(reverse_buddy) ||
>> + page_order(reverse_buddy) != order, reverse_buddy);
>> + if (buddy_pfn < reserve_buddy_pfn) {
>> + list_move(&buddy->lru, &area->free_list[migratetype]);
>> + rmv_reverse_page_order(buddy);
>> + area[1].nr_free_reverse--;
>> + area->nr_free++;
>> + set_page_order(buddy, order);
>> + rmv_reverse_page_order(reverse_buddy);
>> + list_add(&reverse_buddy->lru, &area->free_list[migratetype]);
>> + area->nr_free++;
>> + set_page_order(reverse_buddy, order);
>> + } else {
>> + list_move(&reverse_buddy->lru, &area->free_list[migratetype]);
>> + rmv_reverse_page_order(reverse_buddy);
>> + area[1].nr_free_reverse--;
>> + area->nr_free++;
>> + set_page_order(reverse_buddy, order);
>> + rmv_reverse_page_order(buddy);
>> + list_add(&buddy->lru, &area->free_list[migratetype]);
>> + area->nr_free++;
>> + set_page_order(buddy, order);
>> + }
>> +}
>> +
>> /*
>> * Freeing function for a buddy system allocator.
>> *
>> @@ -805,6 +931,7 @@ static inline void __free_one_page(struct page *page,
>> VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
>> VM_BUG_ON_PAGE(bad_range(zone, page), page);
>>
>> + rmv_reverse_free_area(zone, page, order, migratetype);
>> continue_merging:
>> while (order < max_order - 1) {
>> buddy_pfn = __find_buddy_pfn(pfn, order);
>> @@ -882,6 +1009,7 @@ static inline void __free_one_page(struct page *page,
>> list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
>> out:
>> zone->free_area[order].nr_free++;
>> + set_reverse_free_area(zone, page, order, migratetype);
>> }
>>
>> /*
>> @@ -1238,14 +1366,24 @@ void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
>>
>> static void __free_pages_ok(struct page *page, unsigned int order)
>> {
>> + bool reverse_buddy;
>> int migratetype;
>> unsigned long pfn = page_to_pfn(page);
>>
>> if (!free_pages_prepare(page, order, true))
>> return;
>>
>> + reverse_buddy = order &&
>> + ((pfn & ((1 << order) - 1)) == (1 << (order - 1)));
>> migratetype = get_pfnblock_migratetype(page, pfn);
>> - free_one_page(page_zone(page), page, pfn, order, migratetype);
>> + if (!reverse_buddy)
>> + free_one_page(page_zone(page), page, pfn, order, migratetype);
>> + else {
>> + free_one_page(page_zone(page), page,
>> + pfn, order - 1, migratetype);
>> + free_one_page(page_zone(page), page + (1 << (order - 1)),
>> + pfn + (1 << (order - 1)), order - 1, migratetype);
>> + }
>> }
>>
>> static void __init __free_pages_boot_core(struct page *page, unsigned int order)
>> @@ -1651,6 +1789,25 @@ static inline void expand(struct zone *zone, struct page *page,
>> }
>> }
>>
>> +static inline void expand_reverse(struct zone *zone, struct page *page,
>> + int low, int high, struct free_area *area,
>> + int migratetype)
>> +{
>> + struct page *reverse_buddy;
>> +
>> + reverse_buddy = page + (1 << (high - 1));
>> + rmv_reverse_page_order(reverse_buddy);
>> + set_pcppage_migratetype(reverse_buddy, migratetype);
>> + if (high > low) {
>> + area--;
>> + high--;
>> + expand(zone, page, low, high, area, migratetype);
>> + list_add(&reverse_buddy->lru, &area->free_list[migratetype]);
>> + area->nr_free++;
>> + set_page_order(reverse_buddy, high);
>> + }
>> +}
>> +
>> static void check_new_page_bad(struct page *page)
>> {
>> const char *bad_reason = NULL;
>> @@ -1785,25 +1942,45 @@ static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags
>> */
>> static inline
>> struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
>> - int migratetype)
>> + int migratetype, gfp_t gfp_flags)
>> {
>> + bool reverse_buddy;
>> unsigned int current_order;
>> struct free_area *area;
>> struct page *page;
>>
>> + reverse_buddy = !!(gfp_flags & __GFP_NOREVERSEBUDDY);
>> /* Find a page of the appropriate size in the preferred list */
>> for (current_order = order; current_order < MAX_ORDER; ++current_order) {
>> area = &(zone->free_area[current_order]);
>> page = list_first_entry_or_null(&area->free_list[migratetype],
>> struct page, lru);
>> - if (!page)
>> + if (page) {
>> + list_del(&page->lru);
>> + rmv_page_order(page);
>> + area->nr_free--;
>> + expand(zone, page, order,
>> + current_order, area, migratetype);
>> + set_pcppage_migratetype(page, migratetype);
>> + return page;
>> + }
>> + if (current_order + reverse_buddy == 0 ||
>> + current_order + reverse_buddy > MAX_ORDER - 2)
>> continue;
>> - list_del(&page->lru);
>> - rmv_page_order(page);
>> - area->nr_free--;
>> - expand(zone, page, order, current_order, area, migratetype);
>> - set_pcppage_migratetype(page, migratetype);
>> - return page;
>> + area += reverse_buddy;
>> + page = list_first_entry_or_null(
>> + &area->free_list_reverse[migratetype],
>> + struct page, lru);
>> + if (page) {
>> + list_del(&page->lru);
>> + rmv_reverse_page_order(page);
>> + area->nr_free_reverse--;
>> + expand_reverse(zone, page, order,
>> + current_order + reverse_buddy,
>> + area, migratetype);
>> + set_pcppage_migratetype(page, migratetype);
>> + return page;
>> + }
>> }
>>
>> return NULL;
>> @@ -1828,13 +2005,13 @@ struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
>>
>> #ifdef CONFIG_CMA
>> static struct page *__rmqueue_cma_fallback(struct zone *zone,
>> - unsigned int order)
>> + unsigned int order, gfp_t gfp_flags)
>> {
>> - return __rmqueue_smallest(zone, order, MIGRATE_CMA);
>> + return __rmqueue_smallest(zone, order, MIGRATE_CMA, gfp_flags);
>> }
>> #else
>> static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
>> - unsigned int order) { return NULL; }
>> + unsigned int order, gfp_t gfp_flags) { return NULL; }
>> #endif
>>
>> /*
>> @@ -2136,7 +2313,8 @@ static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
>>
>> /* Remove an element from the buddy allocator from the fallback list */
>> static inline struct page *
>> -__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
>> +__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype,
>> + gfp_t gfp_flags)
>> {
>> struct free_area *area;
>> unsigned int current_order;
>> @@ -2190,17 +2368,18 @@ static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
>> * Call me with the zone->lock already held.
>> */
>> static struct page *__rmqueue(struct zone *zone, unsigned int order,
>> - int migratetype)
>> + int migratetype, gfp_t gfp_flags)
>> {
>> struct page *page;
>>
>> - page = __rmqueue_smallest(zone, order, migratetype);
>> + page = __rmqueue_smallest(zone, order, migratetype, gfp_flags);
>> if (unlikely(!page)) {
>> if (migratetype == MIGRATE_MOVABLE)
>> - page = __rmqueue_cma_fallback(zone, order);
>> + page = __rmqueue_cma_fallback(zone, order, gfp_flags);
>>
>> if (!page)
>> - page = __rmqueue_fallback(zone, order, migratetype);
>> + page = __rmqueue_fallback(zone, order, migratetype,
>> + gfp_flags);
>> }
>>
>> trace_mm_page_alloc_zone_locked(page, order, migratetype);
>> @@ -2221,7 +2400,7 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order,
>>
>> spin_lock_irqsave(&zone->lock, flags);
>> for (i = 0; i < count; ++i) {
>> - struct page *page = __rmqueue(zone, order, migratetype);
>> + struct page *page = __rmqueue(zone, order, migratetype, 0);
>> if (unlikely(page == NULL))
>> break;
>>
>> @@ -2718,12 +2897,13 @@ struct page *rmqueue(struct zone *preferred_zone,
>> do {
>> page = NULL;
>> if (alloc_flags & ALLOC_HARDER) {
>> - page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
>> + page = __rmqueue_smallest(zone, order,
>> + MIGRATE_HIGHATOMIC, gfp_flags);
>> if (page)
>> trace_mm_page_alloc_zone_locked(page, order, migratetype);
>> }
>> if (!page)
>> - page = __rmqueue(zone, order, migratetype);
>> + page = __rmqueue(zone, order, migratetype, gfp_flags);
>> } while (page && check_new_pages(page, order));
>> spin_unlock(&zone->lock);
>> if (!page)
>> @@ -5286,7 +5466,9 @@ static void __meminit zone_init_free_lists(struct zone *zone)
>> unsigned int order, t;
>> for_each_migratetype_order(order, t) {
>> INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
>> + INIT_LIST_HEAD(&zone->free_area[order].free_list_reverse[t]);
>> zone->free_area[order].nr_free = 0;
>> + zone->free_area[order].nr_free_reverse = 0;
>> }
>> }
>>
>> diff --git a/mm/vmstat.c b/mm/vmstat.c
>> index 69f9aff..26007df 100644
>> --- a/mm/vmstat.c
>> +++ b/mm/vmstat.c
>> @@ -1146,10 +1146,13 @@ static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
>> {
>> int order;
>>
>> - seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
>> + seq_printf(m, "Node %d, zone %8s\n", pgdat->node_id, zone->name);
>> for (order = 0; order < MAX_ORDER; ++order)
>> seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
>> seq_putc(m, '\n');
>> + for (order = 0; order < MAX_ORDER; ++order)
>> + seq_printf(m, "%6lu ", zone->free_area[order].nr_free_reverse);
>> + seq_putc(m, '\n');
>> }
>>
>> /*
>> --
>> 1.7.9.5
>>
>
--
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>
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH mm] introduce reverse buddy concept to reduce buddy fragment
2017-07-03 7:48 ` Michal Hocko
2017-07-03 9:01 ` zhouxianrong
@ 2017-07-03 12:02 ` zhouxianrong
2017-07-03 15:33 ` Michal Hocko
1 sibling, 1 reply; 12+ messages in thread
From: zhouxianrong @ 2017-07-03 12:02 UTC (permalink / raw)
To: Michal Hocko
Cc: linux-mm, linux-kernel, akpm, vbabka, alexander.h.duyck, mgorman,
l.stach, vdavydov.dev, hannes, minchan, npiggin, kirill.shutemov,
gi-oh.kim, luto, keescook, mark.rutland, mingo, heiko.carstens,
iamjoonsoo.kim, rientjes, ming.ling, jack, ebru.akagunduz,
bigeasy, Mi.Sophia.Wang, zhouxiyu, weidu.du, fanghua3,
won.ho.park
On 2017/7/3 15:48, Michal Hocko wrote:
> On Fri 30-06-17 19:25:41, zhouxianrong@huawei.com wrote:
>> From: zhouxianrong <zhouxianrong@huawei.com>
>>
>> when buddy is under fragment i find that still there are some pages
>> just like AFFA mode. A is allocated, F is free, AF is buddy pair for
>> oder n, FA is buddy pair for oder n as well.
>
> Could you quantify how often does this happen and how much of a problem
> this actually is? Is there any specific workload that would suffer from
> such an artificial fragmentation?
>
>> I want to compse the
>> FF as oder n + 1 and align to n other than n + 1. this patch broke
>> the rules of buddy stated as alignment to its length of oder. i think
>> we can do so except for kernel stack because the requirement comes from
>> buddy attribution rather than user.
>
> Why do you think the stack is a problem here?
>
>> for kernel stack requirement i add
>> __GFP_NOREVERSEBUDDY for this purpose.
>>
>> a sample just like blow.
>>
>> Node 0, zone DMA
>> 1389 1765 342 272 2 0 0 0 0 0 0
>> 0 75 4398 1560 379 27 2 0 0 0 0
>> Node 0, zone Normal
>> 20 24 14 2 0 0 0 0 0 0 0
>> 0 6 228 3 0 0 0 0 0 0 0
at the sample moment if we have not this patch, the aspect should like below:
Node 0, zone DMA
(1389 + 75 * 2) (1765 + 4398 * 2) (342 + 1560 * 2) (272 + 379 * 2) (2 + 27 * 2) (0 + 2 * 2) 0 0 0 0 0
Node 0, zone Normal
(20 + 6 * 2) (24 + 228 * 2) (14 + 3 * 2) 2 0 0 0 0 0 0 0
i find out AFFA mode in lower order free_list and move FF into higher order free_list_reverse.
now we only consider the DMA zone. let we see the difference.
Node 0, zone DMA
(1389 + 75 * 2) (1765 + 4398 * 2) (342 + 1560 * 2) (272 + 379 * 2) (2 + 27 * 2) (0 + 2 * 2) 0 0 0 0 0
it is equal to
Node 0, zone DMA
1539 10561 3804 1302 58 4 0 0 0 0 0 -------------- 1)
after applying this patch
Node 0, zone DMA
1389 1765 342 272 2 0 0 0 0 0 0
0 75 4398 1560 379 27 2 0 0 0 0
it is equivalent to
Node 0, zone DMA
(1389 + 0) (1765 + 75) (342 + 4398) (272 + 1560) (2 + 379) (0 + 27) (0 + 2) 0 0 0 0
it is equal to
Node 0, zone DMA
1389 1840 4740 1832 381 27 2 0 0 0 0 -------------- 2)
let's write 1) and 2) together and compare them
1539 10561 3804 1302 58 4 0 0 0 0 0 ----- 1)
1389 1840 4740 1832 381 27 2 0 0 0 0 ----- 2)
from above i think after applying the patch the result is better.
>>
>> the patch does not consider fallback allocation for now.
>
> The path is missing the crucial information required for any
> optimization. Some numbers to compare how much it helps. The above
> output of buddyinfo is pointless without any base to compare to. Also
> which workloads would benefit from this change and how much? It is also
> a non trivial amount of code in the guts of the page allocator so this
> really needs _much_ better explanation.
>
> I haven't looked closely on the code yet but a quick look at
> set_reverse_free_area scared me away.
>
>> Signed-off-by: zhouxianrong <zhouxianrong@huawei.com>
>> ---
>> include/linux/gfp.h | 8 +-
>> include/linux/mmzone.h | 2 +
>> include/linux/page-flags.h | 9 ++
>> include/linux/thread_info.h | 5 +-
>> mm/compaction.c | 17 ++++
>> mm/internal.h | 7 ++
>> mm/page_alloc.c | 222 +++++++++++++++++++++++++++++++++++++++----
>> mm/vmstat.c | 5 +-
>> 8 files changed, 251 insertions(+), 24 deletions(-)
>>
>> diff --git a/include/linux/gfp.h b/include/linux/gfp.h
>> index db373b9..f63d4d9 100644
>> --- a/include/linux/gfp.h
>> +++ b/include/linux/gfp.h
>> @@ -40,6 +40,7 @@
>> #define ___GFP_DIRECT_RECLAIM 0x400000u
>> #define ___GFP_WRITE 0x800000u
>> #define ___GFP_KSWAPD_RECLAIM 0x1000000u
>> +#define ___GFP_NOREVERSEBUDDY 0x2000000u
>> /* If the above are modified, __GFP_BITS_SHIFT may need updating */
>>
>> /*
>> @@ -171,6 +172,10 @@
>> * __GFP_NOTRACK_FALSE_POSITIVE is an alias of __GFP_NOTRACK. It's a means of
>> * distinguishing in the source between false positives and allocations that
>> * cannot be supported (e.g. page tables).
>> + *
>> + * __GFP_NOREVERSEBUDDY does not allocate pages from reverse buddy list
>> + * of current order. It make sure that allocation is alignment to same order
>> + * with length order.
>> */
>> #define __GFP_COLD ((__force gfp_t)___GFP_COLD)
>> #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
>> @@ -178,9 +183,10 @@
>> #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
>> #define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK)
>> #define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
>> +#define __GFP_NOREVERSEBUDDY ((__force gfp_t)___GFP_NOREVERSEBUDDY)
>>
>> /* Room for N __GFP_FOO bits */
>> -#define __GFP_BITS_SHIFT 25
>> +#define __GFP_BITS_SHIFT 26
>> #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
>>
>> /*
>> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
>> index 8e02b37..94237fe 100644
>> --- a/include/linux/mmzone.h
>> +++ b/include/linux/mmzone.h
>> @@ -89,7 +89,9 @@ enum {
>>
>> struct free_area {
>> struct list_head free_list[MIGRATE_TYPES];
>> + struct list_head free_list_reverse[MIGRATE_TYPES];
>> unsigned long nr_free;
>> + unsigned long nr_free_reverse;
>> };
>>
>> struct pglist_data;
>> diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
>> index 6b5818d..39d17d7 100644
>> --- a/include/linux/page-flags.h
>> +++ b/include/linux/page-flags.h
>> @@ -675,6 +675,15 @@ static inline int TestClearPageDoubleMap(struct page *page)
>> #define PAGE_KMEMCG_MAPCOUNT_VALUE (-512)
>> PAGE_MAPCOUNT_OPS(Kmemcg, KMEMCG)
>>
>> +/*
>> + * ReverseBuddy is enabled for the buddy allocator that allow allocating
>> + * two adjacent same free order blocks other than buddy blocks and
>> + * composing them as a order + 1 block. It is for reducing buddy
>> + * fragment.
>> + */
>> +#define PAGE_REVERSE_BUDDY_MAPCOUNT_VALUE (-1024)
>> +PAGE_MAPCOUNT_OPS(ReverseBuddy, REVERSE_BUDDY)
>> +
>> extern bool is_free_buddy_page(struct page *page);
>>
>> __PAGEFLAG(Isolated, isolated, PF_ANY);
>> diff --git a/include/linux/thread_info.h b/include/linux/thread_info.h
>> index 5837387..b4a1605 100644
>> --- a/include/linux/thread_info.h
>> +++ b/include/linux/thread_info.h
>> @@ -28,9 +28,10 @@
>>
>> #ifdef CONFIG_DEBUG_STACK_USAGE
>> # define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK | \
>> - __GFP_ZERO)
>> + __GFP_NOREVERSEBUDDY | __GFP_ZERO)
>> #else
>> -# define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK)
>> +# define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK | \
>> + __GFP_NOREVERSEBUDDY)
>> #endif
>>
>> /*
>> diff --git a/mm/compaction.c b/mm/compaction.c
>> index 0fdfde0..a43f169 100644
>> --- a/mm/compaction.c
>> +++ b/mm/compaction.c
>> @@ -768,6 +768,20 @@ static bool too_many_isolated(struct zone *zone)
>> continue;
>> }
>>
>> + if (PageReverseBuddy(page)) {
>> + unsigned long freepage_order = page_order_unsafe(page);
>> +
>> + /*
>> + * Without lock, we cannot be sure that what we got is
>> + * a valid page order. Consider only values in the
>> + * valid order range to prevent low_pfn overflow.
>> + */
>> + if (freepage_order > 0 &&
>> + freepage_order < MAX_ORDER - 1)
>> + low_pfn += (1UL << (freepage_order + 1)) - 1;
>> + continue;
>> + }
>> +
>> /*
>> * Regardless of being on LRU, compound pages such as THP and
>> * hugetlbfs are not to be compacted. We can potentially save
>> @@ -1005,6 +1019,9 @@ static bool suitable_migration_target(struct compact_control *cc,
>> return false;
>> }
>>
>> + if (PageReverseBuddy(page))
>> + return false;
>> +
>> /* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
>> if (migrate_async_suitable(get_pageblock_migratetype(page)))
>> return true;
>> diff --git a/mm/internal.h b/mm/internal.h
>> index ccfc2a2..439b0a8 100644
>> --- a/mm/internal.h
>> +++ b/mm/internal.h
>> @@ -143,6 +143,13 @@ struct alloc_context {
>> return page_pfn ^ (1 << order);
>> }
>>
>> +static inline unsigned long
>> +__find_reverse_buddy_pfn(unsigned long page_pfn, unsigned int order)
>> +{
>> + return (page_pfn & (1 << order)) ? page_pfn + (1 << order) :
>> + page_pfn - (1 << order);
>> +}
>> +
>> extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
>> unsigned long end_pfn, struct zone *zone);
>>
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index 9f9623d..ee1dc1b 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -715,6 +715,18 @@ static inline void rmv_page_order(struct page *page)
>> set_page_private(page, 0);
>> }
>>
>> +static inline void set_reverse_page_order(struct page *page, unsigned int order)
>> +{
>> + set_page_private(page, order);
>> + __SetPageReverseBuddy(page);
>> +}
>> +
>> +static inline void rmv_reverse_page_order(struct page *page)
>> +{
>> + __ClearPageReverseBuddy(page);
>> + set_page_private(page, 0);
>> +}
>> +
>> /*
>> * This function checks whether a page is free && is the buddy
>> * we can do coalesce a page and its buddy if
>> @@ -758,6 +770,120 @@ static inline int page_is_buddy(struct page *page, struct page *buddy,
>> return 0;
>> }
>>
>> +static inline int page_is_reverse_buddy(struct page *page, struct page *buddy,
>> + unsigned int order)
>> +{
>> + if (PageReverseBuddy(buddy) && page_order(buddy) == order) {
>> + /*
>> + * zone check is done late to avoid uselessly
>> + * calculating zone/node ids for pages that could
>> + * never merge.
>> + */
>> + if (page_zone_id(page) != page_zone_id(buddy))
>> + return 0;
>> +
>> + VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
>> +
>> + return 1;
>> + }
>> + return 0;
>> +}
>> +
>> +static inline
>> +void set_reverse_free_area(struct zone *zone, struct page *page,
>> + unsigned int order, int migratetype)
>> +{
>> + unsigned long buddy_pfn;
>> + unsigned long reserve_buddy_pfn;
>> + struct page *reverse_buddy;
>> + struct free_area *area;
>> +
>> + if (order > MAX_ORDER - 3)
>> + return;
>> + if (unlikely(is_migrate_isolate(migratetype)))
>> + return;
>> + buddy_pfn = page_to_pfn(page);
>> + reserve_buddy_pfn = __find_reverse_buddy_pfn(buddy_pfn, order);
>> + if ((buddy_pfn ^ reserve_buddy_pfn) & ~(pageblock_nr_pages - 1))
>> + return;
>> + if (!pfn_valid_within(reserve_buddy_pfn))
>> + return;
>> + reverse_buddy = pfn_to_page(reserve_buddy_pfn);
>> + if (page_zone_id(page) != page_zone_id(reverse_buddy))
>> + return;
>> + if (PageBuddy(reverse_buddy) &&
>> + page_order(reverse_buddy) == order) {
>> + area = &zone->free_area[order];
>> + list_del(&page->lru);
>> + rmv_page_order(page);
>> + area->nr_free--;
>> + set_pcppage_migratetype(page, migratetype);
>> + list_del(&reverse_buddy->lru);
>> + rmv_page_order(reverse_buddy);
>> + area->nr_free--;
>> + set_pcppage_migratetype(reverse_buddy, migratetype);
>> + area++;
>> + if (buddy_pfn < reserve_buddy_pfn) {
>> + list_add(&page->lru,
>> + &area->free_list_reverse[migratetype]);
>> + area->nr_free_reverse++;
>> + set_reverse_page_order(page, order);
>> + set_reverse_page_order(reverse_buddy, order);
>> + } else {
>> + list_add(&reverse_buddy->lru,
>> + &area->free_list_reverse[migratetype]);
>> + area->nr_free_reverse++;
>> + set_reverse_page_order(reverse_buddy, order);
>> + set_reverse_page_order(page, order);
>> + }
>> + }
>> +}
>> +
>> +static inline
>> +void rmv_reverse_free_area(struct zone *zone, struct page *page,
>> + unsigned int order, int migratetype)
>> +{
>> + unsigned long pfn, buddy_pfn;
>> + unsigned long reserve_buddy_pfn;
>> + struct page *buddy, *reverse_buddy;
>> + struct free_area *area;
>> +
>> + pfn = page_to_pfn(page);
>> + buddy_pfn = __find_buddy_index(pfn, order);
>> + buddy = page + (buddy_pfn - pfn);
>> + if (!pfn_valid_within(buddy_pfn))
>> + return;
>> + if (!page_is_reverse_buddy(page, buddy, order))
>> + return;
>> +
>> + area = &zone->free_area[order];
>> + reserve_buddy_pfn = __find_reverse_buddy_pfn(buddy_pfn, order);
>> + reverse_buddy = pfn_to_page(reserve_buddy_pfn);
>> + VM_BUG_ON_PAGE(!PageReverseBuddy(reverse_buddy) ||
>> + page_order(reverse_buddy) != order, reverse_buddy);
>> + if (buddy_pfn < reserve_buddy_pfn) {
>> + list_move(&buddy->lru, &area->free_list[migratetype]);
>> + rmv_reverse_page_order(buddy);
>> + area[1].nr_free_reverse--;
>> + area->nr_free++;
>> + set_page_order(buddy, order);
>> + rmv_reverse_page_order(reverse_buddy);
>> + list_add(&reverse_buddy->lru, &area->free_list[migratetype]);
>> + area->nr_free++;
>> + set_page_order(reverse_buddy, order);
>> + } else {
>> + list_move(&reverse_buddy->lru, &area->free_list[migratetype]);
>> + rmv_reverse_page_order(reverse_buddy);
>> + area[1].nr_free_reverse--;
>> + area->nr_free++;
>> + set_page_order(reverse_buddy, order);
>> + rmv_reverse_page_order(buddy);
>> + list_add(&buddy->lru, &area->free_list[migratetype]);
>> + area->nr_free++;
>> + set_page_order(buddy, order);
>> + }
>> +}
>> +
>> /*
>> * Freeing function for a buddy system allocator.
>> *
>> @@ -805,6 +931,7 @@ static inline void __free_one_page(struct page *page,
>> VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
>> VM_BUG_ON_PAGE(bad_range(zone, page), page);
>>
>> + rmv_reverse_free_area(zone, page, order, migratetype);
>> continue_merging:
>> while (order < max_order - 1) {
>> buddy_pfn = __find_buddy_pfn(pfn, order);
>> @@ -882,6 +1009,7 @@ static inline void __free_one_page(struct page *page,
>> list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
>> out:
>> zone->free_area[order].nr_free++;
>> + set_reverse_free_area(zone, page, order, migratetype);
>> }
>>
>> /*
>> @@ -1238,14 +1366,24 @@ void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
>>
>> static void __free_pages_ok(struct page *page, unsigned int order)
>> {
>> + bool reverse_buddy;
>> int migratetype;
>> unsigned long pfn = page_to_pfn(page);
>>
>> if (!free_pages_prepare(page, order, true))
>> return;
>>
>> + reverse_buddy = order &&
>> + ((pfn & ((1 << order) - 1)) == (1 << (order - 1)));
>> migratetype = get_pfnblock_migratetype(page, pfn);
>> - free_one_page(page_zone(page), page, pfn, order, migratetype);
>> + if (!reverse_buddy)
>> + free_one_page(page_zone(page), page, pfn, order, migratetype);
>> + else {
>> + free_one_page(page_zone(page), page,
>> + pfn, order - 1, migratetype);
>> + free_one_page(page_zone(page), page + (1 << (order - 1)),
>> + pfn + (1 << (order - 1)), order - 1, migratetype);
>> + }
>> }
>>
>> static void __init __free_pages_boot_core(struct page *page, unsigned int order)
>> @@ -1651,6 +1789,25 @@ static inline void expand(struct zone *zone, struct page *page,
>> }
>> }
>>
>> +static inline void expand_reverse(struct zone *zone, struct page *page,
>> + int low, int high, struct free_area *area,
>> + int migratetype)
>> +{
>> + struct page *reverse_buddy;
>> +
>> + reverse_buddy = page + (1 << (high - 1));
>> + rmv_reverse_page_order(reverse_buddy);
>> + set_pcppage_migratetype(reverse_buddy, migratetype);
>> + if (high > low) {
>> + area--;
>> + high--;
>> + expand(zone, page, low, high, area, migratetype);
>> + list_add(&reverse_buddy->lru, &area->free_list[migratetype]);
>> + area->nr_free++;
>> + set_page_order(reverse_buddy, high);
>> + }
>> +}
>> +
>> static void check_new_page_bad(struct page *page)
>> {
>> const char *bad_reason = NULL;
>> @@ -1785,25 +1942,45 @@ static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags
>> */
>> static inline
>> struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
>> - int migratetype)
>> + int migratetype, gfp_t gfp_flags)
>> {
>> + bool reverse_buddy;
>> unsigned int current_order;
>> struct free_area *area;
>> struct page *page;
>>
>> + reverse_buddy = !!(gfp_flags & __GFP_NOREVERSEBUDDY);
>> /* Find a page of the appropriate size in the preferred list */
>> for (current_order = order; current_order < MAX_ORDER; ++current_order) {
>> area = &(zone->free_area[current_order]);
>> page = list_first_entry_or_null(&area->free_list[migratetype],
>> struct page, lru);
>> - if (!page)
>> + if (page) {
>> + list_del(&page->lru);
>> + rmv_page_order(page);
>> + area->nr_free--;
>> + expand(zone, page, order,
>> + current_order, area, migratetype);
>> + set_pcppage_migratetype(page, migratetype);
>> + return page;
>> + }
>> + if (current_order + reverse_buddy == 0 ||
>> + current_order + reverse_buddy > MAX_ORDER - 2)
>> continue;
>> - list_del(&page->lru);
>> - rmv_page_order(page);
>> - area->nr_free--;
>> - expand(zone, page, order, current_order, area, migratetype);
>> - set_pcppage_migratetype(page, migratetype);
>> - return page;
>> + area += reverse_buddy;
>> + page = list_first_entry_or_null(
>> + &area->free_list_reverse[migratetype],
>> + struct page, lru);
>> + if (page) {
>> + list_del(&page->lru);
>> + rmv_reverse_page_order(page);
>> + area->nr_free_reverse--;
>> + expand_reverse(zone, page, order,
>> + current_order + reverse_buddy,
>> + area, migratetype);
>> + set_pcppage_migratetype(page, migratetype);
>> + return page;
>> + }
>> }
>>
>> return NULL;
>> @@ -1828,13 +2005,13 @@ struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
>>
>> #ifdef CONFIG_CMA
>> static struct page *__rmqueue_cma_fallback(struct zone *zone,
>> - unsigned int order)
>> + unsigned int order, gfp_t gfp_flags)
>> {
>> - return __rmqueue_smallest(zone, order, MIGRATE_CMA);
>> + return __rmqueue_smallest(zone, order, MIGRATE_CMA, gfp_flags);
>> }
>> #else
>> static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
>> - unsigned int order) { return NULL; }
>> + unsigned int order, gfp_t gfp_flags) { return NULL; }
>> #endif
>>
>> /*
>> @@ -2136,7 +2313,8 @@ static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
>>
>> /* Remove an element from the buddy allocator from the fallback list */
>> static inline struct page *
>> -__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
>> +__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype,
>> + gfp_t gfp_flags)
>> {
>> struct free_area *area;
>> unsigned int current_order;
>> @@ -2190,17 +2368,18 @@ static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
>> * Call me with the zone->lock already held.
>> */
>> static struct page *__rmqueue(struct zone *zone, unsigned int order,
>> - int migratetype)
>> + int migratetype, gfp_t gfp_flags)
>> {
>> struct page *page;
>>
>> - page = __rmqueue_smallest(zone, order, migratetype);
>> + page = __rmqueue_smallest(zone, order, migratetype, gfp_flags);
>> if (unlikely(!page)) {
>> if (migratetype == MIGRATE_MOVABLE)
>> - page = __rmqueue_cma_fallback(zone, order);
>> + page = __rmqueue_cma_fallback(zone, order, gfp_flags);
>>
>> if (!page)
>> - page = __rmqueue_fallback(zone, order, migratetype);
>> + page = __rmqueue_fallback(zone, order, migratetype,
>> + gfp_flags);
>> }
>>
>> trace_mm_page_alloc_zone_locked(page, order, migratetype);
>> @@ -2221,7 +2400,7 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order,
>>
>> spin_lock_irqsave(&zone->lock, flags);
>> for (i = 0; i < count; ++i) {
>> - struct page *page = __rmqueue(zone, order, migratetype);
>> + struct page *page = __rmqueue(zone, order, migratetype, 0);
>> if (unlikely(page == NULL))
>> break;
>>
>> @@ -2718,12 +2897,13 @@ struct page *rmqueue(struct zone *preferred_zone,
>> do {
>> page = NULL;
>> if (alloc_flags & ALLOC_HARDER) {
>> - page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
>> + page = __rmqueue_smallest(zone, order,
>> + MIGRATE_HIGHATOMIC, gfp_flags);
>> if (page)
>> trace_mm_page_alloc_zone_locked(page, order, migratetype);
>> }
>> if (!page)
>> - page = __rmqueue(zone, order, migratetype);
>> + page = __rmqueue(zone, order, migratetype, gfp_flags);
>> } while (page && check_new_pages(page, order));
>> spin_unlock(&zone->lock);
>> if (!page)
>> @@ -5286,7 +5466,9 @@ static void __meminit zone_init_free_lists(struct zone *zone)
>> unsigned int order, t;
>> for_each_migratetype_order(order, t) {
>> INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
>> + INIT_LIST_HEAD(&zone->free_area[order].free_list_reverse[t]);
>> zone->free_area[order].nr_free = 0;
>> + zone->free_area[order].nr_free_reverse = 0;
>> }
>> }
>>
>> diff --git a/mm/vmstat.c b/mm/vmstat.c
>> index 69f9aff..26007df 100644
>> --- a/mm/vmstat.c
>> +++ b/mm/vmstat.c
>> @@ -1146,10 +1146,13 @@ static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
>> {
>> int order;
>>
>> - seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
>> + seq_printf(m, "Node %d, zone %8s\n", pgdat->node_id, zone->name);
>> for (order = 0; order < MAX_ORDER; ++order)
>> seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
>> seq_putc(m, '\n');
>> + for (order = 0; order < MAX_ORDER; ++order)
>> + seq_printf(m, "%6lu ", zone->free_area[order].nr_free_reverse);
>> + seq_putc(m, '\n');
>> }
>>
>> /*
>> --
>> 1.7.9.5
>>
>
--
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>
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH mm] introduce reverse buddy concept to reduce buddy fragment
2017-07-03 12:02 ` zhouxianrong
@ 2017-07-03 15:33 ` Michal Hocko
2017-07-04 1:21 ` zhouxianrong
0 siblings, 1 reply; 12+ messages in thread
From: Michal Hocko @ 2017-07-03 15:33 UTC (permalink / raw)
To: zhouxianrong
Cc: linux-mm, linux-kernel, akpm, vbabka, alexander.h.duyck, mgorman,
l.stach, vdavydov.dev, hannes, minchan, npiggin, kirill.shutemov,
gi-oh.kim, luto, keescook, mark.rutland, mingo, heiko.carstens,
iamjoonsoo.kim, rientjes, ming.ling, jack, ebru.akagunduz,
bigeasy, Mi.Sophia.Wang, zhouxiyu, weidu.du, fanghua3,
won.ho.park
On Mon 03-07-17 20:02:16, zhouxianrong wrote:
[...]
> from above i think after applying the patch the result is better.
You haven't described your testing methodology, nor the workload that was
tested. As such this data is completely meaningless.
--
Michal Hocko
SUSE Labs
--
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>
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH mm] introduce reverse buddy concept to reduce buddy fragment
2017-07-03 15:33 ` Michal Hocko
@ 2017-07-04 1:21 ` zhouxianrong
2017-07-04 6:52 ` Michal Hocko
0 siblings, 1 reply; 12+ messages in thread
From: zhouxianrong @ 2017-07-04 1:21 UTC (permalink / raw)
To: Michal Hocko
Cc: linux-mm, linux-kernel, akpm, vbabka, alexander.h.duyck, mgorman,
l.stach, vdavydov.dev, hannes, minchan, npiggin, kirill.shutemov,
gi-oh.kim, luto, keescook, mark.rutland, mingo, heiko.carstens,
iamjoonsoo.kim, rientjes, ming.ling, jack, ebru.akagunduz,
bigeasy, Mi.Sophia.Wang, zhouxiyu, weidu.du, fanghua3,
won.ho.park
the test was done as follows:
1. the environment is android 7.0 and kernel is 4.1 and managed memory is 3.5GB
2. every 4s startup one apk, total 100 more apks need to startup
3. after finishing step 2, sample buddyinfo once and get the result
On 2017/7/3 23:33, Michal Hocko wrote:
> On Mon 03-07-17 20:02:16, zhouxianrong wrote:
> [...]
>> from above i think after applying the patch the result is better.
>
> You haven't described your testing methodology, nor the workload that was
> tested. As such this data is completely meaningless.
>
--
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>
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH mm] introduce reverse buddy concept to reduce buddy fragment
2017-07-04 1:21 ` zhouxianrong
@ 2017-07-04 6:52 ` Michal Hocko
2017-07-04 7:00 ` zhouxianrong
2017-07-04 8:04 ` zhouxianrong
0 siblings, 2 replies; 12+ messages in thread
From: Michal Hocko @ 2017-07-04 6:52 UTC (permalink / raw)
To: zhouxianrong
Cc: linux-mm, linux-kernel, akpm, vbabka, alexander.h.duyck, mgorman,
l.stach, vdavydov.dev, hannes, minchan, npiggin, kirill.shutemov,
gi-oh.kim, luto, keescook, mark.rutland, mingo, heiko.carstens,
iamjoonsoo.kim, rientjes, ming.ling, jack, ebru.akagunduz,
bigeasy, Mi.Sophia.Wang, zhouxiyu, weidu.du, fanghua3,
won.ho.park
On Tue 04-07-17 09:21:00, zhouxianrong wrote:
> the test was done as follows:
>
> 1. the environment is android 7.0 and kernel is 4.1 and managed memory is 3.5GB
There have been many changes in the compaction proper since than. Do you
see the same problem with the current upstream kernel?
> 2. every 4s startup one apk, total 100 more apks need to startup
> 3. after finishing step 2, sample buddyinfo once and get the result
How stable are those results?
--
Michal Hocko
SUSE Labs
--
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>
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH mm] introduce reverse buddy concept to reduce buddy fragment
2017-07-04 6:52 ` Michal Hocko
@ 2017-07-04 7:00 ` zhouxianrong
2017-07-04 8:04 ` zhouxianrong
1 sibling, 0 replies; 12+ messages in thread
From: zhouxianrong @ 2017-07-04 7:00 UTC (permalink / raw)
To: Michal Hocko
Cc: linux-mm, linux-kernel, akpm, vbabka, alexander.h.duyck, mgorman,
l.stach, vdavydov.dev, hannes, minchan, npiggin, kirill.shutemov,
gi-oh.kim, luto, keescook, mark.rutland, mingo, heiko.carstens,
iamjoonsoo.kim, rientjes, ming.ling, jack, ebru.akagunduz,
bigeasy, Mi.Sophia.Wang, zhouxiyu, weidu.du, fanghua3,
won.ho.park
i do the test again. after minutes i tell you the result.
On 2017/7/4 14:52, Michal Hocko wrote:
> On Tue 04-07-17 09:21:00, zhouxianrong wrote:
>> the test was done as follows:
>>
>> 1. the environment is android 7.0 and kernel is 4.1 and managed memory is 3.5GB
>
> There have been many changes in the compaction proper since than. Do you
> see the same problem with the current upstream kernel?
>
>> 2. every 4s startup one apk, total 100 more apks need to startup
>> 3. after finishing step 2, sample buddyinfo once and get the result
>
> How stable are those results?
>
--
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>
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH mm] introduce reverse buddy concept to reduce buddy fragment
2017-07-04 6:52 ` Michal Hocko
2017-07-04 7:00 ` zhouxianrong
@ 2017-07-04 8:04 ` zhouxianrong
2017-07-04 11:24 ` Michal Hocko
1 sibling, 1 reply; 12+ messages in thread
From: zhouxianrong @ 2017-07-04 8:04 UTC (permalink / raw)
To: Michal Hocko
Cc: linux-mm, linux-kernel, akpm, vbabka, alexander.h.duyck, mgorman,
l.stach, vdavydov.dev, hannes, minchan, npiggin, kirill.shutemov,
gi-oh.kim, luto, keescook, mark.rutland, mingo, heiko.carstens,
iamjoonsoo.kim, rientjes, ming.ling, jack, ebru.akagunduz,
bigeasy, Mi.Sophia.Wang, zhouxiyu, weidu.du, fanghua3,
won.ho.park
every 2s i sample /proc/buddyinfo in the whole test process.
the last about 90 samples were sampled after the test was done.
Node 0, zone DMA
4706 2099 838 266 50 5 3 2 1 2 38
0 395 1261 211 57 6 1 0 0 0 0
Node 0, zone DMA
4691 2107 833 265 50 5 3 2 1 2 38
0 395 1260 211 57 6 1 0 0 0 0
Node 0, zone DMA
1815 1437 791 266 51 6 2 2 1 2 38
0 244 1275 211 57 6 1 0 0 0 0
Node 0, zone DMA
1465 1030 796 267 51 6 2 2 1 2 38
0 246 1279 211 57 6 1 0 0 0 0
Node 0, zone DMA
1378 1114 791 260 51 6 2 2 1 2 38
0 183 1282 216 54 3 1 0 0 0 0
Node 0, zone DMA
2605 2021 619 260 51 6 2 2 1 2 38
0 307 1330 220 54 3 1 0 0 0 0
Node 0, zone DMA
2465 2026 618 260 51 6 2 2 1 2 38
0 312 1330 220 54 3 1 0 0 0 0
Node 0, zone DMA
758 766 462 224 43 6 2 2 1 2 38
0 148 1082 194 50 3 1 0 0 0 0
Node 0, zone DMA
912 939 472 224 43 6 2 2 1 2 38
0 174 1086 194 50 3 1 0 0 0 0
Node 0, zone DMA
502 1049 428 226 44 6 2 2 1 2 38
0 187 1092 198 50 3 1 0 0 0 0
Node 0, zone DMA
747 1338 671 228 46 6 2 2 1 2 38
0 222 1180 204 51 3 1 0 0 0 0
Node 0, zone DMA
675 1351 667 226 46 6 2 2 1 2 38
0 220 1180 204 52 3 1 0 0 0 0
Node 0, zone DMA
865 787 266 220 45 6 2 2 1 2 38
0 116 984 203 51 3 1 0 0 0 0
Node 0, zone DMA
1915 1233 351 261 47 6 2 2 1 2 38
0 179 1191 230 53 3 1 0 0 0 0
Node 0, zone DMA
2078 1348 402 258 46 6 2 2 1 2 38
0 183 1233 228 53 3 1 0 0 0 0
Node 0, zone DMA
2940 1129 457 259 46 6 2 2 1 2 38
0 209 1239 229 54 3 1 0 0 0 0
Node 0, zone DMA
2906 1127 457 259 46 6 2 2 1 2 38
0 222 1240 230 54 3 1 0 0 0 0
Node 0, zone DMA
1540 1093 475 256 46 6 2 2 1 2 38
0 293 1234 227 54 3 1 0 0 0 0
Node 0, zone DMA
1060 1071 487 257 46 6 2 2 1 2 38
0 297 1238 227 54 3 1 0 0 0 0
Node 0, zone DMA
1693 869 405 267 46 6 2 2 1 2 38
0 243 1480 230 54 3 1 0 0 0 0
Node 0, zone DMA
1720 928 426 269 46 6 2 2 1 2 38
0 260 1485 230 54 3 1 0 0 0 0
Node 0, zone DMA
546 601 393 269 46 6 2 2 1 2 38
0 193 1314 230 54 3 1 0 0 0 0
Node 0, zone DMA
583 336 28 42 20 3 2 2 2 2 36
0 15 43 5 20 1 1 0 0 0 0
Node 0, zone DMA
592 382 27 39 21 3 2 2 2 2 36
0 17 27 5 21 1 1 0 0 0 0
Node 0, zone DMA
3534 1510 212 78 35 6 3 3 1 2 32
0 122 333 66 28 2 1 0 0 0 0
Node 0, zone DMA
3411 1521 212 78 35 6 3 3 1 2 32
0 123 334 67 28 2 1 0 0 0 0
Node 0, zone DMA
1521 1521 211 79 35 6 3 3 1 2 32
0 118 336 68 28 2 1 0 0 0 0
Node 0, zone DMA
3 1 2 0 3 3 1 1 1 3 18
0 2 0 3 0 0 0 0 0 0 0
Node 0, zone DMA
1 3 0 29 7 4 1 1 7 5 4
0 0 162 70 19 0 1 0 0 0 0
Node 0, zone DMA
388 113 162 128 22 3 2 2 14 6 2
0 32 1189 192 19 2 0 0 0 0 0
Node 0, zone DMA
468 541 176 171 35 3 2 3 6 5 2
0 122 959 275 46 1 0 0 0 0 0
Node 0, zone DMA
1251 1625 421 160 27 12 3 2 3 1 0
0 211 1160 234 30 9 1 0 0 0 0
Node 0, zone DMA
19 107 36 24 86 20 4 1 0 0 0
0 14 137 460 118 21 1 0 0 0 0
Node 0, zone DMA
452 217 69 141 91 22 8 0 8 0 0
0 39 632 267 149 23 1 0 0 0 0
Node 0, zone DMA
1073 581 321 116 16 29 8 0 9 3 1
0 124 782 218 73 26 2 0 0 0 0
Node 0, zone DMA
146 218 75 44 9 5 3 1 9 2 0
0 34 412 105 15 3 2 0 0 0 0
Node 0, zone DMA
1766 1850 555 157 4 1 0 0 0 0 0
0 331 1575 212 8 2 0 0 0 0 0
Node 0, zone DMA
635 303 97 241 18 0 0 0 6 0 0
0 59 736 341 17 5 0 0 0 0 0
Node 0, zone DMA
270 75 25 68 42 3 1 0 9 1 0
0 9 315 246 50 7 0 0 0 0 0
Node 0, zone DMA
564 133 36 72 2 0 4 1 11 1 0
0 25 334 146 5 6 0 0 0 0 0
Node 0, zone DMA
704 120 31 73 2 1 0 0 12 0 0
0 31 338 155 7 6 0 0 0 0 0
Node 0, zone DMA
840 159 137 109 7 1 1 0 10 0 0
0 47 674 189 13 6 0 0 0 0 0
Node 0, zone DMA
368 89 160 117 21 0 0 0 1 0 0
0 23 1242 481 45 3 0 0 0 0 0
Node 0, zone DMA
725 255 87 33 25 1 0 0 2 0 1
0 39 491 450 54 3 0 0 0 0 0
Node 0, zone DMA
2054 586 125 98 10 2 1 1 3 1 1
0 138 425 267 50 3 0 0 0 0 0
Node 0, zone DMA
1789 669 84 29 3 3 2 1 2 1 1
0 165 315 178 11 0 0 0 0 0 0
Node 0, zone DMA
2216 2117 797 152 29 1 0 0 0 1 0
0 251 1111 320 15 0 0 0 0 0 0
Node 0, zone DMA
52 361 80 62 4 0 0 0 0 0 0
0 306 2840 270 22 0 0 0 0 0 0
Node 0, zone DMA
776 282 77 32 0 0 0 0 0 0 0
0 326 2839 288 22 0 0 0 0 0 0
Node 0, zone DMA
416 511 182 35 7 0 0 0 0 0 0
0 340 2821 328 37 0 0 0 0 0 0
Node 0, zone DMA
1037 578 187 39 1 0 0 0 0 0 0
0 343 2834 331 38 1 0 0 0 0 0
Node 0, zone DMA
24 0 0 0 0 0 0 0 0 0 0
0 280 2884 332 38 1 0 0 0 0 0
Node 0, zone DMA
130 132 0 0 0 0 0 0 0 0 0
0 237 2573 561 39 1 0 0 0 0 0
Node 0, zone DMA
50 3 1 1 1 1 1 1 0 0 0
0 47 2580 561 39 1 0 0 0 0 0
Node 0, zone DMA
1526 70 7 18 1 1 1 1 0 0 0
0 25 2152 568 39 1 0 0 0 0 0
Node 0, zone DMA
1615 119 8 17 1 0 1 1 0 0 0
0 40 2144 568 39 1 0 0 0 0 0
Node 0, zone DMA
4 5 1 18 1 0 0 1 0 0 0
0 11 1714 568 39 1 0 0 0 0 0
Node 0, zone DMA
154 86 14 103 85 0 0 1 0 0 1
0 28 48 834 93 3 0 1 0 1 0
Node 0, zone DMA
18 62 19 103 85 0 0 1 0 0 1
0 26 59 834 93 3 0 1 0 1 0
Node 0, zone DMA
1366 484 66 65 86 0 0 1 0 0 1
0 108 155 703 94 3 0 1 0 1 0
Node 0, zone DMA
882 771 101 69 86 0 0 1 0 0 1
0 101 242 706 94 3 0 1 0 1 0
Node 0, zone DMA
396 797 108 70 85 0 0 1 0 0 1
0 112 245 706 94 3 0 1 0 1 0
Node 0, zone DMA
1261 365 53 9 41 0 0 0 0 0 1
0 41 49 481 94 3 0 0 0 1 0
Node 0, zone DMA
943 437 62 10 35 0 0 0 0 0 1
0 40 38 501 94 3 0 0 0 1 0
Node 0, zone DMA
2367 864 113 33 36 0 0 0 1 0 1
0 109 123 499 100 3 0 0 0 1 0
Node 0, zone DMA
2598 969 121 32 37 0 0 0 1 0 1
0 105 117 501 100 3 0 0 0 1 0
Node 0, zone DMA
1849 903 120 32 36 0 0 0 1 0 1
0 86 124 502 100 3 0 0 0 1 0
Node 0, zone DMA
226 127 54 19 20 8 1 1 3 1 1
0 18 161 333 131 6 0 0 0 0 0
Node 0, zone DMA
130 337 234 32 20 7 2 2 2 1 1
0 70 399 352 131 6 0 0 0 0 0
Node 0, zone DMA
688 414 347 38 23 7 2 2 2 1 1
0 130 652 328 133 6 0 0 0 0 0
Node 0, zone DMA
1212 131 348 38 23 7 2 2 2 1 1
0 133 660 328 133 6 0 0 0 0 0
Node 0, zone DMA
846 123 348 38 23 7 2 2 2 1 1
0 138 660 328 133 6 0 0 0 0 0
Node 0, zone DMA
216 285 73 17 32 7 2 1 1 1 1
0 91 95 292 153 8 0 0 0 0 0
Node 0, zone DMA
41 253 76 17 32 7 2 1 1 1 1
0 75 89 308 153 8 0 0 0 0 0
Node 0, zone DMA
1837 600 174 13 33 7 2 1 1 1 1
0 103 184 273 155 8 0 0 0 0 0
Node 0, zone DMA
1570 883 184 22 33 7 2 1 1 1 1
0 112 219 274 155 8 0 0 0 0 0
Node 0, zone DMA
90 54 32 21 33 7 2 1 1 1 1
0 19 114 276 155 8 0 0 0 0 0
Node 0, zone DMA
1029 43 3 3 0 0 0 0 0 0 0
0 87 610 343 158 8 0 0 0 0 0
Node 0, zone DMA
153 142 2 3 0 0 0 0 0 0 0
0 103 680 350 158 8 0 0 0 0 0
Node 0, zone DMA
1555 797 157 46 15 2 0 0 0 0 0
0 70 744 434 177 8 0 0 0 0 0
Node 0, zone DMA
1538 824 187 49 13 2 0 0 0 0 0
0 68 712 434 174 8 0 0 0 0 0
Node 0, zone DMA
808 19 4 54 3 1 0 0 0 0 0
0 15 586 440 174 8 0 0 0 0 0
Node 0, zone DMA
934 702 48 3 1 1 0 0 0 0 0
0 170 392 469 178 8 0 0 0 0 0
Node 0, zone DMA
1012 729 36 3 1 1 0 0 0 0 0
0 169 416 470 178 8 0 0 0 0 0
Node 0, zone DMA
1204 891 188 39 1 1 0 0 1 0 0
0 180 484 512 179 8 0 0 0 0 0
Node 0, zone DMA
1712 951 195 38 1 1 0 0 1 0 0
0 192 487 512 179 8 0 0 0 0 0
Node 0, zone DMA
1679 951 205 11 1 1 0 0 1 0 0
0 194 505 517 179 8 0 0 0 0 0
Node 0, zone DMA
193 34 1 8 1 1 1 1 0 0 0
0 12 205 525 177 8 0 0 0 0 0
Node 0, zone DMA
974 25 1 2 0 0 0 0 0 0 0
0 19 152 525 178 8 0 0 0 0 0
Node 0, zone DMA
1500 1279 94 2 0 0 0 0 2 0 0
0 209 209 500 178 8 0 0 0 0 0
Node 0, zone DMA
1092 1375 106 3 0 0 0 0 2 0 0
0 244 216 500 178 8 0 0 0 0 0
Node 0, zone DMA
609 1376 109 6 2 1 2 0 1 0 0
0 240 216 491 178 8 0 0 0 0 0
Node 0, zone DMA
280 1906 396 8 2 1 2 0 2 0 0
0 303 396 499 179 8 0 0 0 0 0
Node 0, zone DMA
52 1833 388 9 2 1 2 0 2 0 0
0 300 393 489 179 8 0 0 0 0 0
Node 0, zone DMA
512 1889 349 9 0 1 2 0 2 0 0
0 293 408 491 179 8 0 0 0 0 0
Node 0, zone DMA
559 1891 348 9 0 1 2 0 2 0 0
0 304 409 491 179 8 0 0 0 0 0
Node 0, zone DMA
18 1088 365 10 0 1 2 0 2 0 0
0 314 419 491 178 8 0 0 0 0 0
Node 0, zone DMA
1153 1850 411 10 1 1 2 0 2 0 0
0 317 439 492 178 8 0 0 0 0 0
Node 0, zone DMA
695 1849 401 10 1 1 2 0 2 0 0
0 317 440 492 178 8 0 0 0 0 0
Node 0, zone DMA
266 1924 351 12 1 1 2 0 2 0 0
0 307 458 492 178 8 0 0 0 0 0
Node 0, zone DMA
88 1842 347 10 1 1 2 0 2 0 0
0 301 444 493 178 8 0 0 0 0 0
Node 0, zone DMA
414 1856 348 10 0 1 2 0 2 0 0
0 300 454 492 178 8 0 0 0 0 0
Node 0, zone DMA
202 1774 335 12 1 1 2 0 2 0 0
0 303 486 496 178 8 0 0 0 0 0
Node 0, zone DMA
1800 1905 334 12 1 1 2 0 2 0 0
0 319 479 496 178 8 0 0 0 0 0
Node 0, zone DMA
1560 1802 361 13 1 1 2 0 2 0 0
0 319 485 496 178 8 0 0 0 0 0
Node 0, zone DMA
1011 1822 366 15 1 1 2 0 2 0 0
0 319 489 495 178 8 0 0 0 0 0
Node 0, zone DMA
589 1814 364 14 1 1 2 0 2 0 0
0 315 492 495 179 8 0 0 0 0 0
Node 0, zone DMA
63 1789 313 14 1 1 2 0 2 0 0
0 320 499 495 179 8 0 0 0 0 0
Node 0, zone DMA
417 1818 306 14 1 1 2 0 2 0 0
0 314 499 495 179 8 0 0 0 0 0
Node 0, zone DMA
372 1861 292 14 1 1 2 0 2 0 0
0 315 510 495 179 8 0 0 0 0 0
Node 0, zone DMA
1902 1931 295 13 1 1 2 0 2 0 0
0 325 503 496 179 8 0 0 0 0 0
Node 0, zone DMA
354 1934 306 11 1 1 2 0 2 0 0
0 322 510 496 179 8 0 0 0 0 0
Node 0, zone DMA
786 1858 350 12 2 1 2 0 2 0 0
0 326 514 496 179 8 0 0 0 0 0
Node 0, zone DMA
660 1843 345 12 2 1 2 0 2 0 0
0 321 514 496 179 8 0 0 0 0 0
Node 0, zone DMA
461 1893 295 12 2 1 2 0 2 0 0
0 321 516 498 179 8 0 0 0 0 0
Node 0, zone DMA
261 1811 323 12 2 1 2 0 2 0 0
0 323 516 498 179 8 0 0 0 0 0
Node 0, zone DMA
254 1819 318 11 2 1 2 0 2 0 0
0 317 514 498 179 8 0 0 0 0 0
Node 0, zone DMA
1731 1813 352 11 2 1 2 0 2 0 0
0 318 518 498 179 8 0 0 0 0 0
Node 0, zone DMA
3197 1901 364 11 2 1 2 0 2 0 0
0 354 528 498 179 8 0 0 0 0 0
Node 0, zone DMA
2877 1885 374 10 1 1 2 0 2 0 0
0 350 527 494 179 8 0 0 0 0 0
Node 0, zone DMA
2790 1851 378 10 0 1 2 0 2 0 0
0 335 523 494 178 8 0 0 0 0 0
Node 0, zone DMA
1736 2165 419 13 2 0 2 0 2 0 0
0 450 551 494 178 8 0 0 0 0 0
Node 0, zone DMA
134 401 219 7 1 0 1 0 1 0 0
0 165 722 507 179 8 0 0 0 0 0
Node 0, zone DMA
390 425 245 7 1 0 1 0 1 0 0
0 162 739 507 179 8 0 0 0 0 0
Node 0, zone DMA
1052 658 290 23 3 0 1 0 2 0 0
0 214 929 510 179 8 0 0 0 0 0
Node 0, zone DMA
751 640 297 23 3 0 1 0 2 0 0
0 226 990 509 179 8 0 0 0 0 0
Node 0, zone DMA
17 17 360 15 1 1 1 0 1 0 0
0 31 1014 511 178 8 0 0 0 0 0
Node 0, zone DMA
121 436 10 0 1 1 1 1 0 0 0
0 53 973 516 178 8 0 0 0 0 0
Node 0, zone DMA
1388 343 12 1 0 1 1 1 0 0 0
0 87 989 516 178 8 0 0 0 0 0
Node 0, zone DMA
1496 342 5 1 0 1 1 1 0 0 0
0 93 924 519 178 8 0 0 0 0 0
Node 0, zone DMA
844 362 10 1 0 1 1 1 0 0 0
0 87 905 519 178 8 0 0 0 0 0
Node 0, zone DMA
20 435 218 60 0 0 0 0 2 0 0
0 102 419 629 180 8 0 0 0 0 0
Node 0, zone DMA
288 161 80 23 2 0 0 0 0 0 0
0 65 1449 674 216 9 0 0 0 0 0
Node 0, zone DMA
2164 217 77 23 37 22 6 2 0 0 0
0 60 241 23 139 16 10 2 0 0 0
Node 0, zone DMA
402 1001 202 80 48 24 8 2 0 0 0
0 187 653 72 69 16 11 2 0 0 0
Node 0, zone DMA
3832 807 240 80 48 24 8 2 0 0 0
0 256 704 86 69 16 11 2 0 0 0
Node 0, zone DMA
3819 786 246 80 48 24 8 2 0 0 0
0 256 706 86 69 16 11 2 0 0 0
Node 0, zone DMA
987 964 469 193 9 6 3 0 0 0 0
0 197 1378 157 52 10 2 2 0 0 0
Node 0, zone DMA
477 1113 475 193 9 6 3 0 0 0 0
0 218 1363 160 52 10 2 2 0 0 0
Node 0, zone DMA
1087 1208 399 193 9 6 3 0 0 0 0
0 242 1448 161 52 10 2 2 0 0 0
Node 0, zone DMA
1856 1211 416 194 9 6 3 0 0 0 0
0 261 1438 161 52 10 2 2 0 0 0
Node 0, zone DMA
105 1171 410 197 9 6 3 0 0 0 0
0 257 1426 162 52 10 2 2 0 0 0
Node 0, zone DMA
59 39 32 202 16 7 3 0 0 0 0
0 17 958 137 69 10 2 2 0 0 0
Node 0, zone DMA
210 462 29 133 5 3 0 0 0 0 0
0 17 539 327 58 9 2 1 0 0 0
Node 0, zone DMA
2190 1104 420 193 33 6 2 1 0 0 0
0 119 881 469 67 9 2 1 1 0 0
Node 0, zone DMA
2163 1132 434 199 33 6 2 1 0 0 0
0 121 910 474 69 9 2 1 1 0 0
Node 0, zone DMA
3762 2049 452 199 33 6 2 1 0 0 0
0 168 919 475 69 9 2 1 1 0 0
Node 0, zone DMA
3102 1990 459 200 33 6 2 1 0 0 0
0 163 926 475 69 9 2 1 1 0 0
Node 0, zone DMA
1168 116 56 2 1 0 1 0 0 0 0
0 20 945 474 69 9 1 1 1 0 0
Node 0, zone DMA
1732 375 194 23 2 0 0 1 0 0 0
0 50 896 503 69 9 2 1 1 0 0
Node 0, zone DMA
754 399 203 107 14 10 5 1 0 0 0
0 91 827 599 102 10 2 1 1 0 0
Node 0, zone DMA
1842 248 194 109 14 10 5 1 0 0 0
0 33 786 603 102 10 2 1 1 0 0
Node 0, zone DMA
1992 251 209 109 14 10 5 1 0 0 0
0 38 811 604 102 10 2 1 1 0 0
Node 0, zone DMA
2252 2082 544 178 15 2 1 1 0 0 0
0 473 1138 667 101 10 2 1 1 0 0
Node 0, zone DMA
2248 2154 551 182 15 2 1 1 0 0 0
0 490 1103 667 101 10 2 1 1 0 0
Node 0, zone DMA
338 2035 536 183 15 2 1 1 0 0 0
0 492 1107 667 101 10 2 1 1 0 0
Node 0, zone DMA
140 1612 537 183 15 2 1 1 0 0 0
0 490 1120 667 101 10 2 1 1 0 0
Node 0, zone DMA
79 119 86 178 14 2 1 1 0 0 0
0 51 934 663 101 11 2 1 1 0 0
Node 0, zone DMA
278 345 117 178 14 2 1 1 0 0 0
0 44 768 665 101 11 2 1 1 0 0
Node 0, zone DMA
299 529 160 154 14 2 1 1 0 0 0
0 106 762 632 101 11 2 1 1 0 0
Node 0, zone DMA
3364 823 180 154 14 2 1 1 0 0 0
0 199 788 633 101 11 2 1 1 0 0
Node 0, zone DMA
1181 1351 153 155 14 2 1 1 0 0 0
0 235 804 645 101 11 2 1 1 0 0
Node 0, zone DMA
2759 780 157 155 14 2 1 1 0 0 0
0 157 802 646 101 11 2 1 1 0 0
Node 0, zone DMA
641 174 83 20 13 2 0 0 0 0 0
0 68 1057 674 106 12 4 1 0 0 0
Node 0, zone DMA
267 237 142 34 13 2 0 0 0 0 0
0 68 926 647 108 12 4 1 0 0 0
Node 0, zone DMA
2591 569 40 27 13 2 0 0 0 0 0
0 143 227 637 108 12 4 1 0 0 0
Node 0, zone DMA
2211 591 72 39 13 2 0 0 0 0 0
0 152 251 638 108 12 4 1 0 0 0
Node 0, zone DMA
1701 728 329 96 4 3 0 0 0 0 0
0 217 568 631 110 12 3 1 0 0 0
Node 0, zone DMA
3590 1438 498 104 57 15 1 1 1 0 0
0 159 726 191 148 14 5 1 0 0 0
Node 0, zone DMA
3670 1307 501 104 57 15 1 1 1 0 0
0 162 735 192 148 14 5 1 0 0 0
Node 0, zone DMA
3675 1615 523 105 57 15 1 1 1 0 0
0 191 1021 192 148 14 5 1 0 0 0
Node 0, zone DMA
3644 1615 523 105 57 15 1 1 1 0 0
0 191 1021 192 148 14 5 1 0 0 0
Node 0, zone DMA
2210 1592 521 106 36 1 1 1 1 0 0
0 189 1022 192 145 12 4 1 0 0 0
Node 0, zone DMA
120 7 2 0 1 1 1 1 0 0 0
0 58 974 249 148 12 3 1 0 0 0
Node 0, zone DMA
2222 1079 259 31 30 5 1 1 0 0 0
0 182 1182 265 151 12 3 1 0 0 0
Node 0, zone DMA
53 832 249 37 30 5 1 1 0 0 0
0 177 1200 267 151 12 3 1 0 0 0
Node 0, zone DMA
482 313 97 22 21 8 3 0 0 0 0
0 107 377 306 159 12 3 1 0 0 0
Node 0, zone DMA
44 776 265 26 6 1 0 0 0 0 0
0 201 551 308 152 12 3 1 0 0 0
Node 0, zone DMA
2 2 16 27 17 4 1 0 0 0 0
0 9 602 364 162 17 3 1 0 0 0
Node 0, zone DMA
15 144 332 73 2 1 1 0 0 0 0
0 139 776 564 159 17 3 1 0 0 0
Node 0, zone DMA
1620 1255 275 3 2 1 1 0 0 0 0
0 161 883 568 159 17 3 1 0 0 0
Node 0, zone DMA
8 4 0 1 1 1 1 0 0 0 0
0 18 956 570 159 17 3 1 0 0 0
Node 0, zone DMA
54 437 242 2 0 1 1 0 0 0 0
0 258 961 571 159 17 3 1 0 0 0
Node 0, zone DMA
2358 524 65 9 4 0 0 0 0 0 0
0 62 415 564 160 16 3 1 0 0 0
Node 0, zone DMA
333 1 3 0 0 0 0 0 0 0 0
0 135 464 569 160 16 3 1 0 0 0
Node 0, zone DMA
655 170 0 0 0 0 0 0 0 0 0
0 141 351 569 160 16 3 1 0 0 0
Node 0, zone DMA
1683 674 147 28 8 8 0 0 0 0 0
0 138 396 591 139 15 0 0 0 0 0
Node 0, zone DMA
1908 931 186 19 4 3 0 0 0 0 0
0 142 580 649 135 16 0 0 0 0 0
Node 0, zone DMA
133 134 4 1 0 0 0 0 0 0 0
0 245 621 651 135 16 0 0 0 0 0
Node 0, zone DMA
1355 1200 131 0 0 0 0 0 0 0 0
0 188 438 706 165 16 0 0 0 0 0
Node 0, zone DMA
1603 1236 111 0 0 0 0 0 0 0 0
0 184 492 705 165 16 0 0 0 0 0
Node 0, zone DMA
3420 1831 178 8 0 0 0 0 0 0 0
0 452 683 722 165 16 0 0 0 0 0
Node 0, zone DMA
2824 1752 200 11 0 0 0 0 0 0 0
0 455 688 723 165 16 0 0 0 0 0
Node 0, zone DMA
606 1807 167 0 0 0 0 0 0 0 0
0 467 697 725 165 16 0 0 0 0 0
Node 0, zone DMA
1062 1052 211 17 4 1 0 0 0 0 0
0 151 362 743 170 16 0 0 0 0 0
Node 0, zone DMA
147 174 6 9 0 0 0 0 0 0 0
0 19 620 761 180 16 0 0 0 0 0
Node 0, zone DMA
306 147 595 30 3 0 0 0 0 0 0
0 223 224 531 207 16 0 0 0 0 0
Node 0, zone DMA
3944 1292 552 30 3 0 0 0 0 0 0
0 227 234 531 207 16 0 0 0 0 0
Node 0, zone DMA
4716 1368 561 29 3 0 0 0 0 0 0
0 251 250 538 207 16 0 0 0 0 0
Node 0, zone DMA
359 375 33 5 0 0 0 0 0 0 0
0 10 723 1087 229 20 0 0 0 0 0
Node 0, zone DMA
2 6 9 32 5 2 0 0 0 0 0
0 0 1 626 236 20 1 0 0 0 0
Node 0, zone DMA
1673 744 114 20 18 4 0 0 0 0 0
0 125 280 528 241 21 2 0 0 0 0
Node 0, zone DMA
643 649 119 22 19 3 0 0 0 0 0
0 228 317 528 238 21 2 0 0 0 0
Node 0, zone DMA
167 98 12 6 18 2 0 0 0 0 0
0 15 49 329 238 21 1 0 0 0 0
Node 0, zone DMA
1180 1246 275 42 34 9 0 0 0 0 0
0 346 277 304 241 22 1 0 0 0 0
Node 0, zone DMA
228 1400 400 57 7 4 0 0 0 0 0
0 477 425 328 241 21 1 0 0 0 0
Node 0, zone DMA
294 1391 252 0 3 4 0 0 0 0 0
0 470 485 345 241 21 1 0 0 0 0
Node 0, zone DMA
1993 1741 76 0 3 4 0 0 0 0 0
0 492 1015 354 241 21 1 0 0 0 0
Node 0, zone DMA
236 1638 7 1 2 4 0 0 0 0 0
0 505 1033 353 241 21 1 0 0 0 0
Node 0, zone DMA
424 657 152 15 11 6 2 0 0 0 0
0 85 221 354 242 21 2 0 0 0 0
Node 0, zone DMA
4117 2228 130 7 0 0 0 0 0 0 0
0 251 773 447 258 20 0 0 0 0 0
Node 0, zone DMA
3751 2398 137 8 0 0 0 0 0 0 0
0 243 790 448 258 20 0 0 0 0 0
Node 0, zone DMA
7539 2788 210 13 0 0 0 0 0 0 0
0 332 848 453 258 20 0 0 0 0 0
Node 0, zone DMA
1763 2306 71 0 0 0 0 0 0 0 0
0 295 1122 481 258 20 0 0 0 0 0
Node 0, zone DMA
7409 3990 596 80 2 0 0 0 0 0 0
0 767 1860 541 270 20 0 0 0 0 0
Node 0, zone DMA
779 245 56 30 10 1 1 0 0 0 0
0 52 1052 601 279 20 0 0 0 0 0
Node 0, zone DMA
3568 1160 181 68 33 5 1 0 0 0 0
0 158 1122 633 294 20 0 0 0 0 0
Node 0, zone DMA
4162 1591 246 72 34 5 1 0 0 0 0
0 228 2234 643 295 20 0 0 0 0 0
Node 0, zone DMA
4050 1672 246 73 34 5 1 0 0 0 0
0 230 2235 641 295 20 0 0 0 0 0
Node 0, zone DMA
6457 2781 466 168 22 5 0 0 0 0 0
0 406 2468 646 290 20 0 0 0 0 0
Node 0, zone DMA
4861 2198 15 2 1 0 0 0 0 0 0
0 661 2843 649 290 20 0 0 0 0 0
Node 0, zone DMA
4422 2059 14 0 0 0 0 0 0 0 0
0 665 2840 648 290 20 0 0 0 0 0
Node 0, zone DMA
2223 2098 31 0 0 0 0 0 0 0 0
0 666 2838 648 290 20 0 0 0 0 0
Node 0, zone DMA
4363 2189 31 0 0 0 0 0 0 0 0
0 686 2901 648 290 20 0 0 0 0 0
Node 0, zone DMA
6288 3092 38 3 0 0 0 0 0 0 0
0 836 3738 659 289 20 0 0 0 0 0
Node 0, zone DMA
926 633 381 115 11 0 0 0 0 0 0
0 201 1127 717 293 20 0 0 0 0 0
Node 0, zone DMA
2692 1122 276 48 15 4 0 0 0 0 0
0 301 1404 768 299 20 0 0 0 0 0
Node 0, zone DMA
3074 489 288 48 11 1 0 0 0 0 0
0 318 1427 771 298 20 0 0 0 0 0
Node 0, zone DMA
6803 1700 320 50 11 1 0 0 0 0 0
0 462 1454 770 298 20 0 0 0 0 0
Node 0, zone DMA
25092 8052 125 11 1 0 0 0 0 0 0
0 1994 2786 811 308 20 0 0 0 0 0
Node 0, zone DMA
21794 7145 98 12 1 0 0 0 0 0 0
0 2105 2861 814 308 20 0 0 0 0 0
Node 0, zone DMA
938 1262 89 17 0 0 0 0 0 0 0
0 1752 4445 1048 358 22 0 0 0 0 0
Node 0, zone DMA
981 1293 103 15 0 0 0 0 0 0 0
0 1758 4455 1051 358 22 0 0 0 0 0
Node 0, zone DMA
2970 1700 109 0 0 0 0 0 0 0 0
0 1846 4892 1051 358 22 0 0 0 0 0
Node 0, zone DMA
3175 1647 104 3 0 0 0 0 0 0 0
0 1868 4895 1053 358 22 0 0 0 0 0
Node 0, zone DMA
4980 2340 87 2 0 0 0 0 0 0 0
0 1900 5040 1060 358 22 0 0 0 0 0
Node 0, zone DMA
10308 1612 378 69 1 0 0 0 0 0 0
0 565 5269 1180 373 23 0 0 0 0 0
Node 0, zone DMA
1453 981 222 6 1 0 0 0 0 0 0
0 224 5906 1275 399 23 0 0 0 0 0
Node 0, zone DMA
1687 246 9 0 0 0 0 0 0 0 0
0 200 5977 1264 399 23 0 0 0 0 0
Node 0, zone DMA
1141 579 22 2 0 0 0 0 0 0 0
0 220 5766 1291 399 23 0 0 0 0 0
Node 0, zone DMA
1047 104 0 0 0 0 0 0 0 0 0
0 192 5788 1284 400 23 0 0 0 0 0
Node 0, zone DMA
7550 2234 94 2 0 0 0 0 0 0 0
0 2167 6245 1314 413 23 0 0 0 0 0
Node 0, zone DMA
7592 2252 100 2 0 0 0 0 0 0 0
0 2125 6285 1314 413 23 0 0 0 0 0
Node 0, zone DMA
6614 2308 99 2 0 0 0 0 0 0 0
0 2132 6290 1314 413 23 0 0 0 0 0
Node 0, zone DMA
4722 2650 70 1 0 0 0 0 0 0 0
0 2086 6526 1339 418 23 0 0 0 0 0
Node 0, zone DMA
3132 2283 115 3 0 0 0 0 0 0 0
0 2066 6850 1347 425 24 0 0 0 0 0
Node 0, zone DMA
1076 1485 45 1 0 0 0 0 0 0 0
0 2053 7389 1357 434 24 0 0 0 0 0
Node 0, zone DMA
836 1087 4 0 0 0 0 0 0 0 0
0 1762 7435 1354 434 24 0 0 0 0 0
Node 0, zone DMA
409 1045 1 1 0 0 0 0 0 0 0
0 1760 7469 1349 434 24 0 0 0 0 0
Node 0, zone DMA
337 877 14 0 1 0 0 0 0 0 0
0 1672 7476 1347 434 24 0 0 0 0 0
Node 0, zone DMA
15 0 0 0 0 0 0 0 0 0 0
0 581 7479 1343 440 24 0 0 0 0 0
Node 0, zone DMA
1 0 8 0 0 0 0 0 0 0 0
0 20 5288 1390 457 27 0 0 0 0 0
Node 0, zone DMA
16 38 6 0 0 0 0 0 0 0 0
0 26 4500 1394 457 27 0 0 0 0 0
Node 0, zone DMA
33 27 13 0 0 0 0 0 0 0 0
0 20 3573 1397 457 27 0 0 0 0 0
Node 0, zone DMA
5177 758 27 0 0 0 0 0 0 0 0
0 180 4137 1397 457 27 0 0 0 0 0
Node 0, zone DMA
7670 1207 87 3 1 0 0 0 0 0 0
0 294 4705 1407 457 27 0 0 0 0 0
Node 0, zone DMA
5767 740 32 6 0 0 0 0 0 0 0
0 277 4714 1399 457 27 0 0 0 0 0
Node 0, zone DMA
9843 1966 25 0 0 0 0 0 0 0 0
0 1447 5696 1390 442 27 0 0 0 0 0
Node 0, zone DMA
10032 1829 25 0 0 0 0 0 0 0 0
0 1473 5709 1388 442 27 0 0 0 0 0
Node 0, zone DMA
9687 2626 60 4 0 0 0 0 0 0 0
0 1537 5782 1385 442 25 0 0 0 0 0
Node 0, zone DMA
8866 2495 101 1 0 0 0 0 0 0 0
0 1585 5798 1384 442 25 0 0 0 0 0
Node 0, zone DMA
8830 2450 21 0 0 0 0 0 0 0 0
0 1577 5802 1384 442 25 0 0 0 0 0
Node 0, zone DMA
11145 2800 69 3 0 0 0 0 0 0 0
0 1655 5923 1389 443 25 0 0 0 0 0
Node 0, zone DMA
12822 3004 52 3 0 0 0 0 0 0 0
0 1708 5936 1391 443 25 0 0 0 0 0
Node 0, zone DMA
11282 3108 115 7 0 0 0 0 0 0 0
0 1693 5941 1386 443 25 0 0 0 0 0
Node 0, zone DMA
10508 3040 111 6 2 0 0 0 0 0 0
0 1722 5940 1388 443 25 0 0 0 0 0
Node 0, zone DMA
9796 3265 147 10 2 0 0 0 0 0 0
0 1721 5942 1387 443 25 0 0 0 0 0
Node 0, zone DMA
11928 3940 118 16 20 1 0 0 0 0 0
0 1913 6228 1413 453 25 0 0 0 0 0
Node 0, zone DMA
13113 4003 92 16 20 1 0 0 0 0 0
0 1969 6257 1412 453 25 0 0 0 0 0
Node 0, zone DMA
11343 4020 61 16 20 1 0 0 0 0 0
0 1959 6279 1414 453 25 0 0 0 0 0
Node 0, zone DMA
10528 3426 99 20 20 1 0 0 0 0 0
0 1949 6299 1412 453 25 0 0 0 0 0
Node 0, zone DMA
113 321 85 7 0 0 0 0 0 0 0
0 27 6326 1540 465 26 0 0 0 0 0
Node 0, zone DMA
2879 305 136 3 0 0 0 0 0 0 0
0 54 6350 1540 465 26 0 0 0 0 0
Node 0, zone DMA
1872 393 95 0 0 0 0 0 0 0 0
0 85 6322 1541 465 26 0 0 0 0 0
Node 0, zone DMA
2674 606 121 4 0 0 0 0 0 0 0
0 132 6258 1542 465 26 0 0 0 0 0
Node 0, zone DMA
986 921 94 4 0 0 0 0 0 0 0
0 153 6326 1546 465 26 0 0 0 0 0
Node 0, zone DMA
2609 2148 27 6 1 0 0 0 0 0 0
0 1235 5574 1607 480 26 0 0 0 0 0
Node 0, zone DMA
68 61 5 0 0 0 0 0 0 0 0
0 508 5825 1591 473 26 0 0 0 0 0
Node 0, zone DMA
213 265 13 0 0 0 0 0 0 0 0
0 208 5306 1589 473 26 0 0 0 0 0
Node 0, zone DMA
2143 238 18 0 0 0 0 0 0 0 0
0 101 5096 1594 473 26 0 0 0 0 0
Node 0, zone DMA
1192 236 34 0 0 0 0 0 0 0 0
0 99 5145 1594 473 26 0 0 0 0 0
Node 0, zone DMA
400 244 17 4 0 0 0 0 0 0 0
0 33 2064 1637 473 26 0 0 0 0 0
Node 0, zone DMA
0 0 0 0 0 0 0 0 0 0 0
0 0 1667 1666 474 26 0 0 0 0 0
Node 0, zone DMA
8 5 14 0 0 0 0 0 0 0 0
0 35 1759 1665 474 26 0 0 0 0 0
Node 0, zone DMA
36 14 9 1 0 0 0 0 0 0 0
0 17 24 905 474 26 0 0 0 0 0
Node 0, zone DMA
375 196 40 4 0 0 0 0 0 0 0
0 203 331 939 478 26 0 0 0 0 0
Node 0, zone DMA
1128 2184 771 0 0 0 0 0 0 0 0
0 667 3345 1265 496 26 0 0 0 0 0
Node 0, zone DMA
17493 3123 94 12 3 0 0 0 0 0 0
0 1258 2740 1057 509 26 0 0 0 0 0
Node 0, zone DMA
17494 2939 81 10 3 0 0 0 0 0 0
0 1276 2758 1056 509 26 0 0 0 0 0
Node 0, zone DMA
15407 3682 415 29 3 0 0 0 0 0 0
0 1384 2849 1075 510 26 0 0 0 0 0
Node 0, zone DMA
13485 3961 498 37 4 0 0 0 0 0 0
0 1429 2913 1135 510 26 0 0 0 0 0
Node 0, zone DMA
15665 4486 599 44 4 0 0 0 0 0 0
0 1562 3038 1152 511 26 0 0 0 0 0
Node 0, zone DMA
410 1364 168 0 0 0 0 0 0 0 0
0 1200 4979 1360 525 26 0 0 0 0 0
Node 0, zone DMA
77 30 164 0 0 0 0 0 0 0 0
0 246 4976 1360 525 26 0 0 0 0 0
Node 0, zone DMA
1492 281 230 7 0 0 0 0 0 0 0
0 152 4621 1377 525 26 0 0 0 0 0
Node 0, zone DMA
2780 727 267 6 2 0 0 0 0 0 0
0 591 4680 1382 525 26 0 0 0 0 0
Node 0, zone DMA
12995 2974 169 2 2 0 0 0 0 0 0
0 1351 4884 1390 528 26 0 0 0 0 0
Node 0, zone DMA
11002 4237 325 34 2 2 0 0 0 0 0
0 1784 5068 1449 530 26 0 0 0 0 0
Node 0, zone DMA
10062 4278 348 31 2 2 0 0 0 0 0
0 1789 5064 1450 530 26 0 0 0 0 0
Node 0, zone DMA
10558 4405 374 32 1 2 0 0 0 0 0
0 1908 5080 1453 530 26 0 0 0 0 0
Node 0, zone DMA
17822 6827 288 24 1 0 0 0 0 0 0
0 2676 5638 1455 528 26 0 0 0 0 0
Node 0, zone DMA
16082 6534 277 24 1 0 0 0 0 0 0
0 2685 5644 1456 528 26 0 0 0 0 0
Node 0, zone DMA
1179 2890 926 201 34 1 0 0 0 0 0
0 2751 7098 1587 567 26 0 0 0 0 0
Node 0, zone DMA
2032 2723 575 21 5 1 0 0 0 0 0
0 2745 7147 1604 567 26 0 0 0 0 0
Node 0, zone DMA
1149 2808 613 18 6 1 0 0 0 0 0
0 2731 7061 1609 567 26 0 0 0 0 0
Node 0, zone DMA
11101 5306 307 24 6 1 0 0 0 0 0
0 2801 7798 1588 567 26 0 0 0 0 0
Node 0, zone DMA
10648 5227 307 27 6 1 0 0 0 0 0
0 2805 7811 1588 567 26 0 0 0 0 0
Node 0, zone DMA
2740 2243 71 109 36 3 0 0 0 0 0
0 2280 8503 1643 589 26 0 0 0 0 0
Node 0, zone DMA
1634 2034 1 105 35 3 0 0 0 0 0
0 2272 8406 1645 589 26 0 0 0 0 0
Node 0, zone DMA
1549 1957 87 112 17 3 0 0 0 0 0
0 2273 8297 1617 575 26 0 0 0 0 0
Node 0, zone DMA
7128 3535 134 96 22 3 0 0 0 0 0
0 2705 8907 1618 577 26 0 0 0 0 0
Node 0, zone DMA
7125 3511 137 96 22 3 0 0 0 0 0
0 2707 8913 1618 577 26 0 0 0 0 0
Node 0, zone DMA
557 1049 10 109 41 5 0 0 0 0 0
0 2434 9464 1637 592 27 0 0 0 0 0
Node 0, zone DMA
379 52 11 8 20 5 0 0 0 0 0
0 20 6219 1545 592 27 0 0 0 0 0
Node 0, zone DMA
112 214 81 27 7 0 0 0 0 0 0
0 20 5716 1569 567 26 0 0 0 0 0
Node 0, zone DMA
5735 719 58 16 0 0 0 0 0 0 0
0 245 5661 1563 552 26 0 0 0 0 0
Node 0, zone DMA
5351 798 60 16 0 0 0 0 0 0 0
0 279 5685 1563 552 26 0 0 0 0 0
Node 0, zone DMA
889 800 76 19 0 0 0 0 0 0 0
0 256 5820 1566 552 26 0 0 0 0 0
Node 0, zone DMA
334 253 56 16 0 0 0 0 0 0 0
0 271 5808 1566 555 26 0 0 0 0 0
Node 0, zone DMA
98 170 53 16 0 0 0 0 0 0 0
0 272 5835 1567 554 26 0 0 0 0 0
Node 0, zone DMA
7129 1753 288 67 4 0 0 0 0 0 0
0 1401 6568 1602 561 26 0 0 0 0 0
Node 0, zone DMA
6498 1832 311 70 5 0 0 0 0 0 0
0 1434 6590 1608 561 26 0 0 0 0 0
Node 0, zone DMA
6582 1801 320 71 5 0 0 0 0 0 0
0 1446 6598 1608 561 26 0 0 0 0 0
Node 0, zone DMA
6849 1466 311 72 5 0 0 0 0 0 0
0 1451 6591 1610 562 26 0 0 0 0 0
Node 0, zone DMA
5618 1252 252 66 6 2 0 0 0 0 0
0 1357 6651 1618 566 26 0 0 0 0 0
Node 0, zone DMA
10321 2108 290 70 7 2 0 0 0 0 0
0 1645 7612 1642 571 26 0 0 0 0 0
Node 0, zone DMA
9897 2177 253 59 7 2 0 0 0 0 0
0 1649 7642 1643 571 26 0 0 0 0 0
Node 0, zone DMA
1013 1234 190 57 7 2 0 0 0 0 0
0 1083 8175 1682 578 26 0 0 0 0 0
Node 0, zone DMA
1546 535 2 54 6 2 0 0 0 0 0
0 1093 8055 1682 578 26 0 0 0 0 0
Node 0, zone DMA
1127 428 2 0 0 0 0 0 0 0 0
0 1095 7313 1613 566 26 0 0 0 0 0
Node 0, zone DMA
5391 1599 27 49 1 0 0 0 0 0 0
0 1153 8400 1667 568 26 0 0 0 0 0
Node 0, zone DMA
11053 2284 44 49 1 0 0 0 0 0 0
0 1389 8534 1663 573 26 0 0 0 0 0
Node 0, zone DMA
2632 858 214 112 25 1 0 0 0 0 0
0 1433 9339 1735 584 27 0 0 0 0 0
Node 0, zone DMA
1104 857 218 116 29 1 0 0 0 0 0
0 1433 9420 1739 585 28 0 0 0 0 0
Node 0, zone DMA
710 770 158 123 27 5 1 0 0 0 0
0 1448 9414 1752 589 29 1 0 0 0 0
Node 0, zone DMA
5530 1253 162 120 28 5 1 0 0 0 0
0 1478 9414 1754 589 29 1 0 0 0 0
Node 0, zone DMA
16582 5138 174 120 28 5 1 0 0 0 0
0 2021 9810 1761 589 29 1 0 0 0 0
Node 0, zone DMA
14834 5126 145 121 28 6 1 0 0 0 0
0 2037 9791 1767 591 29 1 0 0 0 0
Node 0, zone DMA
16874 5157 143 121 28 6 1 0 0 0 0
0 2145 9838 1766 591 29 1 0 0 0 0
Node 0, zone DMA
17034 4974 169 124 34 6 1 0 0 0 0
0 2167 9690 1772 594 29 1 0 0 0 0
Node 0, zone DMA
19112 5082 174 124 34 6 1 0 0 0 0
0 2190 9721 1774 594 30 1 0 0 0 0
Node 0, zone DMA
17870 5110 182 130 34 6 1 0 0 0 0
0 2205 9733 1776 594 30 1 0 0 0 0
Node 0, zone DMA
16621 4997 154 123 35 6 1 0 0 0 0
0 2207 9739 1780 596 30 1 0 0 0 0
Node 0, zone DMA
16130 4641 142 123 35 6 1 0 0 0 0
0 2206 9732 1780 596 30 1 0 0 0 0
Node 0, zone DMA
15672 4607 146 122 36 6 1 0 0 0 0
0 2204 9607 1794 598 32 1 0 0 0 0
Node 0, zone DMA
15425 4509 148 122 36 6 1 0 0 0 0
0 2204 9611 1794 598 32 1 0 0 0 0
Node 0, zone DMA
15540 4603 168 124 36 6 1 0 0 0 0
0 2212 9641 1796 598 32 1 0 0 0 0
Node 0, zone DMA
15140 4583 157 126 37 6 1 0 0 0 0
0 2189 9537 1803 598 32 1 0 0 0 0
Node 0, zone DMA
16955 5021 175 126 36 6 1 0 0 0 0
0 2210 9554 1805 601 33 1 0 0 0 0
Node 0, zone DMA
14956 4909 168 126 36 6 1 0 0 0 0
0 2209 9552 1805 601 33 1 0 0 0 0
Node 0, zone DMA
14765 4270 165 128 38 6 1 0 0 0 0
0 2204 9466 1814 602 33 1 0 0 0 0
Node 0, zone DMA
14432 3975 168 128 38 6 1 0 0 0 0
0 2171 9467 1814 602 33 1 0 0 0 0
Node 0, zone DMA
14115 4137 153 135 38 6 1 0 0 0 0
0 2107 9337 1817 603 33 1 0 0 0 0
Node 0, zone DMA
14808 4471 161 136 37 7 1 0 0 0 0
0 2152 9368 1821 603 33 1 0 0 0 0
Node 0, zone DMA
13415 4459 176 144 38 7 1 0 0 0 0
0 2146 9237 1828 607 33 1 0 0 0 0
Node 0, zone DMA
13162 4247 161 144 38 7 1 0 0 0 0
0 2144 9225 1828 608 33 1 0 0 0 0
Node 0, zone DMA
12873 4222 45 145 38 7 1 0 0 0 0
0 1919 9238 1828 608 33 1 0 0 0 0
Node 0, zone DMA
19 215 283 267 75 9 1 0 0 0 0
0 2125 10674 1889 643 35 1 0 0 0 0
Node 0, zone DMA
30 22 4 140 60 9 1 0 0 0 0
0 191 10715 1915 643 35 1 0 0 0 0
Node 0, zone DMA
1325 343 34 139 60 9 1 0 0 0 0
0 61 10194 1914 643 35 1 0 0 0 0
Node 0, zone DMA
3849 353 108 163 61 9 1 0 0 0 0
0 169 9901 1945 645 35 1 0 0 0 0
Node 0, zone DMA
2046 1362 225 139 58 10 1 0 0 0 0
0 305 9988 1960 647 35 1 0 0 0 0
Node 0, zone DMA
388 146 62 15 65 10 1 0 0 0 0
0 17 8387 1890 665 36 1 0 0 0 0
Node 0, zone DMA
8 6 3 0 62 10 1 0 0 0 0
0 12 8376 1771 665 36 1 0 0 0 0
Node 0, zone DMA
1337 426 158 59 28 0 0 0 0 0 0
0 16 8438 1733 607 27 0 0 0 0 0
Node 0, zone DMA
737 359 151 34 0 0 0 0 0 0 0
0 84 8370 1702 581 26 0 0 0 0 0
Node 0, zone DMA
377 333 155 35 0 0 0 0 0 0 0
0 82 8390 1702 581 26 0 0 0 0 0
Node 0, zone DMA
91 21 3 0 0 0 0 0 0 0 0
0 12 5769 1705 581 26 0 0 0 0 0
Node 0, zone DMA
14 58 106 47 20 0 0 0 0 0 0
0 269 5846 1789 608 26 0 0 0 0 0
Node 0, zone DMA
89 19 5 0 0 0 0 0 0 0 0
0 3 2768 1715 582 26 0 0 0 0 0
Node 0, zone DMA
143 51 6 0 0 0 0 0 0 0 0
0 17 833 1045 582 26 0 0 0 0 0
Node 0, zone DMA
1890 449 27 0 0 0 0 0 0 0 0
0 44 694 1058 582 26 0 0 0 0 0
Node 0, zone DMA
141 5 3 1 0 1 0 0 0 0 0
0 178 1192 986 582 26 0 0 0 0 0
Node 0, zone DMA
377 136 30 24 2 0 0 0 0 0 0
0 1538 4183 1208 586 26 0 0 0 0 0
Node 0, zone DMA
935 223 45 9 1 0 0 0 0 0 0
0 1427 4076 1184 586 26 0 0 0 0 0
Node 0, zone DMA
70 33 13 10 0 0 0 0 0 0 0
0 18 3979 1147 586 26 0 0 0 0 0
Node 0, zone DMA
1611 146 18 11 0 0 0 0 0 0 0
0 35 3741 1147 586 26 0 0 0 0 0
Node 0, zone DMA
361 538 209 44 0 0 0 0 0 0 0
0 145 3952 1190 590 26 0 0 0 0 0
Node 0, zone DMA
240 203 45 0 0 0 0 0 0 0 0
0 70 3382 1186 590 26 0 0 0 0 0
Node 0, zone DMA
239 376 101 3 0 0 0 0 0 0 0
0 39 2726 1193 590 26 0 0 0 0 0
Node 0, zone DMA
2554 306 84 9 0 0 0 0 0 0 0
0 85 3010 1199 590 26 0 0 0 0 0
Node 0, zone DMA
9614 746 88 0 0 0 0 0 0 0 0
0 416 2874 1197 590 26 0 0 0 0 0
Node 0, zone DMA
8997 676 92 0 0 0 0 0 0 0 0
0 413 2877 1195 590 26 0 0 0 0 0
Node 0, zone DMA
9492 1566 190 2 0 0 0 0 0 0 0
0 658 3019 1292 590 26 0 0 0 0 0
Node 0, zone DMA
2397 1548 543 58 1 0 0 0 0 0 0
0 738 3935 1344 590 26 0 0 0 0 0
Node 0, zone DMA
2639 1572 493 28 0 0 0 0 0 0 0
0 755 3955 1339 590 26 0 0 0 0 0
Node 0, zone DMA
9199 3716 334 5 1 0 0 0 0 0 0
0 1013 4452 1334 590 26 0 0 0 0 0
Node 0, zone DMA
8788 3731 324 6 1 0 0 0 0 0 0
0 1018 4422 1335 590 26 0 0 0 0 0
Node 0, zone DMA
6503 3490 225 5 1 0 0 0 0 0 0
0 964 4511 1335 590 26 0 0 0 0 0
Node 0, zone DMA
7197 3569 229 6 0 0 0 0 0 0 0
0 978 4522 1327 590 26 0 0 0 0 0
Node 0, zone DMA
10472 3961 212 11 0 0 0 0 0 0 0
0 1052 4626 1332 590 26 0 0 0 0 0
Node 0, zone DMA
10116 4065 238 15 0 0 0 0 0 0 0
0 1103 4677 1334 590 26 0 0 0 0 0
Node 0, zone DMA
9264 4014 241 18 0 0 0 0 0 0 0
0 1118 4685 1335 590 26 0 0 0 0 0
Node 0, zone DMA
9041 4004 248 18 0 0 0 0 0 0 0
0 1121 4685 1335 590 26 0 0 0 0 0
Node 0, zone DMA
1060 2320 288 77 33 1 0 0 0 0 0
0 977 5342 1465 622 27 0 0 0 0 0
Node 0, zone DMA
7086 2573 195 41 23 1 0 0 0 0 0
0 750 4912 1475 623 27 0 0 0 0 0
Node 0, zone DMA
807 1412 174 41 21 2 0 0 0 0 0
0 787 4938 1478 625 27 0 0 0 0 0
Node 0, zone DMA
1935 1590 180 42 21 2 0 0 0 0 0
0 806 4739 1500 625 27 0 0 0 0 0
Node 0, zone DMA
1069 1600 183 41 21 2 0 0 0 0 0
0 804 4739 1499 625 27 0 0 0 0 0
Node 0, zone DMA
985 1646 192 50 21 2 0 0 0 0 0
0 803 4607 1511 625 27 0 0 0 0 0
Node 0, zone DMA
4765 2289 288 57 21 2 0 0 0 0 0
0 902 4793 1520 625 27 0 0 0 0 0
Node 0, zone DMA
4452 2217 279 58 21 2 0 0 0 0 0
0 904 4790 1520 625 27 0 0 0 0 0
Node 0, zone DMA
3921 2207 265 59 21 2 0 0 0 0 0
0 850 4819 1522 625 27 0 0 0 0 0
Node 0, zone DMA
3596 2224 269 59 21 2 0 0 0 0 0
0 850 4820 1522 625 27 0 0 0 0 0
Node 0, zone DMA
3457 2202 229 59 21 2 0 0 0 0 0
0 893 4828 1523 625 27 0 0 0 0 0
Node 0, zone DMA
9412 2771 233 57 20 2 0 0 0 0 0
0 1067 5023 1529 627 27 0 0 0 0 0
Node 0, zone DMA
9357 2752 232 57 20 2 0 0 0 0 0
0 1067 5024 1529 627 27 0 0 0 0 0
Node 0, zone DMA
9114 2729 240 57 20 2 0 0 0 0 0
0 1059 5036 1530 627 27 0 0 0 0 0
Node 0, zone DMA
8924 2731 243 57 20 2 0 0 0 0 0
0 1059 5036 1530 627 27 0 0 0 0 0
Node 0, zone DMA
9110 2780 249 57 20 2 0 0 0 0 0
0 976 5045 1531 627 27 0 0 0 0 0
Node 0, zone DMA
11652 3785 253 58 20 2 0 0 0 0 0
0 1236 5360 1554 630 27 0 0 0 0 0
Node 0, zone DMA
11410 3785 249 58 20 2 0 0 0 0 0
0 1237 5362 1554 630 27 0 0 0 0 0
Node 0, zone DMA
11243 3696 241 59 20 2 0 0 0 0 0
0 1272 5365 1555 630 27 0 0 0 0 0
Node 0, zone DMA
10801 3700 242 59 20 2 0 0 0 0 0
0 1272 5365 1555 630 27 0 0 0 0 0
Node 0, zone DMA
10642 3728 239 59 20 2 0 0 0 0 0
0 1190 5365 1555 630 27 0 0 0 0 0
Node 0, zone DMA
10399 3708 239 59 20 2 0 0 0 0 0
0 1190 5365 1555 630 27 0 0 0 0 0
Node 0, zone DMA
10186 3674 239 59 20 2 0 0 0 0 0
0 1193 5365 1555 630 27 0 0 0 0 0
Node 0, zone DMA
8006 3646 240 59 20 2 0 0 0 0 0
0 1255 5365 1555 630 27 0 0 0 0 0
Node 0, zone DMA
6483 3648 238 59 20 2 0 0 0 0 0
0 1255 5365 1555 630 27 0 0 0 0 0
Node 0, zone DMA
6647 3662 237 59 20 2 0 0 0 0 0
0 1159 5365 1555 630 27 0 0 0 0 0
Node 0, zone DMA
6516 3670 237 59 20 2 0 0 0 0 0
0 1170 5365 1555 630 27 0 0 0 0 0
Node 0, zone DMA
6227 3688 242 59 20 2 0 0 0 0 0
0 1174 5366 1555 630 27 0 0 0 0 0
Node 0, zone DMA
7248 3728 262 59 20 2 0 0 0 0 0
0 1178 5370 1556 630 27 0 0 0 0 0
Node 0, zone DMA
6810 3730 263 59 19 1 0 0 0 0 0
0 1178 5370 1556 629 27 0 0 0 0 0
Node 0, zone DMA
8588 4498 412 77 20 1 0 0 0 0 0
0 1299 5554 1554 629 27 0 0 0 0 0
Node 0, zone DMA
8430 4605 424 78 20 1 0 0 0 0 0
0 1324 5550 1554 629 27 0 0 0 0 0
Node 0, zone DMA
8165 4649 416 78 20 1 0 0 0 0 0
0 1333 5557 1554 629 27 0 0 0 0 0
Node 0, zone DMA
8806 4655 423 79 20 1 0 0 0 0 0
0 1365 5557 1555 629 27 0 0 0 0 0
Node 0, zone DMA
7289 4644 432 80 20 1 0 0 0 0 0
0 1363 5555 1554 629 27 0 0 0 0 0
Node 0, zone DMA
5796 4524 382 80 20 1 0 0 0 0 0
0 1350 5543 1554 629 27 0 0 0 0 0
Node 0, zone DMA
5478 1047 253 64 21 1 0 0 0 0 0
0 1375 5730 1573 631 27 0 0 0 0 0
Node 0, zone DMA
4837 745 253 63 21 1 0 0 0 0 0
0 1327 5730 1572 632 27 0 0 0 0 0
Node 0, zone DMA
7629 898 250 46 21 1 0 0 0 0 0
0 1272 5536 1589 634 27 0 0 0 0 0
Node 0, zone DMA
7813 1100 253 47 21 1 0 0 0 0 0
0 1330 5540 1590 634 27 0 0 0 0 0
Node 0, zone DMA
7892 1099 210 53 21 1 0 0 0 0 0
0 1355 5571 1591 635 27 0 0 0 0 0
Node 0, zone DMA
7746 1104 214 53 21 1 0 0 0 0 0
0 1357 5571 1591 635 27 0 0 0 0 0
Node 0, zone DMA
7800 1138 213 54 21 1 0 0 0 0 0
0 1365 5582 1591 635 27 0 0 0 0 0
Node 0, zone DMA
7488 1112 206 55 21 1 0 0 0 0 0
0 1363 5586 1591 635 27 0 0 0 0 0
Node 0, zone DMA
7382 1127 206 55 21 1 0 0 0 0 0
0 1366 5587 1591 635 27 0 0 0 0 0
Node 0, zone DMA
2044 1234 248 111 41 2 0 0 0 0 0
0 1350 6021 1638 650 27 0 0 0 0 0
Node 0, zone DMA
481 574 261 92 38 2 0 0 0 0 0
0 1130 6033 1639 650 28 0 0 0 0 0
Node 0, zone DMA
913 52 0 0 0 2 0 0 0 0 0
0 19 3946 1365 642 28 0 0 0 0 0
Node 0, zone DMA
1122 478 35 14 1 0 0 0 0 0 0
0 29 3836 1384 611 26 0 0 0 0 0
Node 0, zone DMA
1555 788 68 31 1 0 0 0 0 0 0
0 80 3941 1393 611 26 0 0 0 0 0
Node 0, zone DMA
5886 736 79 26 0 0 0 0 0 0 0
0 396 3993 1385 600 26 0 0 0 0 0
Node 0, zone DMA
5616 642 77 27 0 0 0 0 0 0 0
0 430 3999 1386 600 26 0 0 0 0 0
Node 0, zone DMA
2286 98 1 2 1 0 0 0 0 0 0
0 291 3927 1368 594 26 0 0 0 0 0
Node 0, zone DMA
4572 640 76 26 2 0 0 0 0 0 0
0 447 4007 1380 597 26 0 0 0 0 0
Node 0, zone DMA
4492 653 76 25 3 0 0 0 0 0 0
0 446 4012 1380 598 26 0 0 0 0 0
Node 0, zone DMA
2751 1371 161 28 0 0 0 0 0 0 0
0 522 4051 1376 594 26 0 0 0 0 0
Node 0, zone DMA
3288 791 85 17 0 0 0 0 0 0 0
0 537 4055 1375 594 26 0 0 0 0 0
Node 0, zone DMA
3193 791 85 17 0 0 0 0 0 0 0
0 538 4055 1375 594 26 0 0 0 0 0
Node 0, zone DMA
3082 1143 148 14 0 0 0 0 0 0 0
0 634 4080 1371 594 26 0 0 0 0 0
Node 0, zone DMA
2880 1159 146 14 0 0 0 0 0 0 0
0 634 4081 1370 594 26 0 0 0 0 0
Node 0, zone DMA
8221 2231 393 99 13 0 0 0 0 0 0
0 1173 4888 1492 609 26 0 0 0 0 0
Node 0, zone DMA
7803 2070 391 100 13 0 0 0 0 0 0
0 1172 4895 1497 609 26 0 0 0 0 0
Node 0, zone DMA
4451 1940 393 99 12 0 0 0 0 0 0
0 1174 4897 1497 609 27 0 0 0 0 0
Node 0, zone DMA
6630 1786 396 99 13 0 0 0 0 0 0
0 1181 4904 1498 609 27 0 0 0 0 0
Node 0, zone DMA
6126 1789 397 99 13 0 0 0 0 0 0
0 1180 4904 1498 609 27 0 0 0 0 0
Node 0, zone DMA
5940 1699 401 100 13 0 0 0 0 0 0
0 1191 4906 1497 609 27 0 0 0 0 0
Node 0, zone DMA
6303 2471 501 101 13 0 0 0 0 0 0
0 1519 4913 1498 609 27 0 0 0 0 0
Node 0, zone DMA
6100 2477 504 102 13 0 0 0 0 0 0
0 1524 4915 1498 609 27 0 0 0 0 0
Node 0, zone DMA
6199 2458 508 101 13 0 0 0 0 0 0
0 1530 4916 1498 609 27 0 0 0 0 0
Node 0, zone DMA
4400 2447 507 101 13 0 0 0 0 0 0
0 1530 4916 1498 609 27 0 0 0 0 0
Node 0, zone DMA
4862 2392 508 100 14 0 0 0 0 0 0
0 1535 4917 1498 609 27 0 0 0 0 0
Node 0, zone DMA
4689 2401 508 100 14 0 0 0 0 0 0
0 1535 4917 1498 609 27 0 0 0 0 0
Node 0, zone DMA
4532 2402 508 100 14 0 0 0 0 0 0
0 1535 4917 1498 609 27 0 0 0 0 0
Node 0, zone DMA
4459 2416 508 100 14 0 0 0 0 0 0
0 1536 4920 1498 609 27 0 0 0 0 0
Node 0, zone DMA
4306 2440 509 100 14 0 0 0 0 0 0
0 1538 4921 1498 609 27 0 0 0 0 0
Node 0, zone DMA
4114 2376 509 100 14 0 0 0 0 0 0
0 1541 4921 1498 609 27 0 0 0 0 0
Node 0, zone DMA
3653 2374 509 100 14 0 0 0 0 0 0
0 1541 4921 1498 609 27 0 0 0 0 0
Node 0, zone DMA
150 948 508 100 14 0 0 0 0 0 0
0 1541 4921 1498 609 27 0 0 0 0 0
Node 0, zone DMA
2950 1372 508 100 14 0 0 0 0 0 0
0 1094 4919 1498 609 27 0 0 0 0 0
Node 0, zone DMA
2602 1388 509 100 14 0 0 0 0 0 0
0 1097 4919 1498 609 27 0 0 0 0 0
Node 0, zone DMA
2380 1342 486 101 14 0 0 0 0 0 0
0 769 4920 1498 609 27 0 0 0 0 0
Node 0, zone DMA
2489 1382 486 101 14 0 0 0 0 0 0
0 782 4920 1499 609 27 0 0 0 0 0
Node 0, zone DMA
3438 1453 486 105 16 0 0 0 0 0 0
0 788 4933 1504 609 27 0 0 0 0 0
Node 0, zone DMA
3373 1443 489 106 16 0 0 0 0 0 0
0 797 4936 1506 609 27 0 0 0 0 0
Node 0, zone DMA
3153 1553 423 110 18 0 0 0 0 0 0
0 655 4875 1519 609 27 0 0 0 0 0
Node 0, zone DMA
3009 1500 422 110 18 0 0 0 0 0 0
0 661 4877 1519 610 27 0 0 0 0 0
Node 0, zone DMA
2834 1528 420 110 18 0 0 0 0 0 0
0 665 4878 1519 610 27 0 0 0 0 0
Node 0, zone DMA
3823 1574 244 53 14 0 0 0 0 0 0
0 710 4884 1519 610 27 0 0 0 0 0
Node 0, zone DMA
3746 1628 228 53 14 0 0 0 0 0 0
0 712 4885 1520 610 27 0 0 0 0 0
Node 0, zone DMA
3737 1623 230 53 14 0 0 0 0 0 0
0 715 4884 1520 610 27 0 0 0 0 0
Node 0, zone DMA
3835 1534 234 53 14 0 0 0 0 0 0
0 717 4885 1520 610 27 0 0 0 0 0
Node 0, zone DMA
5601 2503 324 53 14 0 0 0 0 0 0
0 728 4898 1520 610 27 0 0 0 0 0
Node 0, zone DMA
9443 3281 236 53 14 0 0 0 0 0 0
0 767 4911 1520 610 27 0 0 0 0 0
Node 0, zone DMA
19296 6049 296 65 15 0 0 0 0 0 0
0 2782 5776 1540 613 27 0 0 0 0 0
Node 0, zone DMA
19233 6137 302 66 15 0 0 0 0 0 0
0 2795 5790 1540 613 27 0 0 0 0 0
Node 0, zone DMA
19462 6070 310 66 15 0 0 0 0 0 0
0 2801 5801 1540 613 27 0 0 0 0 0
Node 0, zone DMA
19934 6316 322 70 15 0 0 0 0 0 0
0 2833 5803 1539 613 27 0 0 0 0 0
Node 0, zone DMA
19668 6327 309 66 15 0 0 0 0 0 0
0 2831 5803 1539 613 27 0 0 0 0 0
Node 0, zone DMA
19677 6280 319 66 15 0 0 0 0 0 0
0 2834 5804 1539 613 27 0 0 0 0 0
Node 0, zone DMA
19520 6280 319 66 15 0 0 0 0 0 0
0 2835 5804 1539 613 27 0 0 0 0 0
Node 0, zone DMA
20794 6354 330 65 15 0 0 0 0 0 0
0 2840 5800 1539 614 27 0 0 0 0 0
Node 0, zone DMA
20329 6377 342 65 15 0 0 0 0 0 0
0 2841 5799 1539 614 27 0 0 0 0 0
Node 0, zone DMA
20755 6397 344 65 15 0 0 0 0 0 0
0 2850 5800 1539 614 27 0 0 0 0 0
Node 0, zone DMA
20732 6345 352 65 15 0 0 0 0 0 0
0 2853 5799 1539 614 27 0 0 0 0 0
Node 0, zone DMA
20418 6352 356 65 15 0 0 0 0 0 0
0 2853 5798 1539 614 27 0 0 0 0 0
Node 0, zone DMA
19902 6298 355 65 15 0 0 0 0 0 0
0 2855 5798 1539 614 27 0 0 0 0 0
Node 0, zone DMA
19652 6299 355 65 15 0 0 0 0 0 0
0 2855 5798 1539 614 27 0 0 0 0 0
Node 0, zone DMA
19373 6299 356 65 15 0 0 0 0 0 0
0 2855 5798 1539 614 27 0 0 0 0 0
Node 0, zone DMA
19136 6214 356 65 15 0 0 0 0 0 0
0 2857 5798 1539 614 27 0 0 0 0 0
Node 0, zone DMA
19101 6217 352 65 15 0 0 0 0 0 0
0 2857 5798 1539 614 27 0 0 0 0 0
Node 0, zone DMA
19001 6141 352 65 15 0 0 0 0 0 0
0 2860 5798 1539 614 27 0 0 0 0 0
Node 0, zone DMA
16538 6116 333 66 15 0 0 0 0 0 0
0 2861 5798 1537 614 27 0 0 0 0 0
Node 0, zone DMA
16213 6107 338 65 15 0 0 0 0 0 0
0 2861 5798 1537 614 27 0 0 0 0 0
Node 0, zone DMA
15179 6124 337 65 15 0 0 0 0 0 0
0 2769 5798 1537 614 27 0 0 0 0 0
Node 0, zone DMA
15505 6148 335 65 15 0 0 0 0 0 0
0 2829 5798 1537 614 27 0 0 0 0 0
Node 0, zone DMA
15282 6138 331 65 15 0 0 0 0 0 0
0 2732 5791 1537 614 27 0 0 0 0 0
Node 0, zone DMA
16524 6299 327 65 15 0 0 0 0 0 0
0 2729 5789 1537 614 27 0 0 0 0 0
Node 0, zone DMA
15431 6303 326 65 15 0 0 0 0 0 0
0 2729 5789 1537 614 27 0 0 0 0 0
Node 0, zone DMA
16640 6251 327 65 15 0 0 0 0 0 0
0 2804 5793 1537 614 27 0 0 0 0 0
Node 0, zone DMA
16320 6255 330 65 15 0 0 0 0 0 0
0 2805 5793 1537 614 27 0 0 0 0 0
Node 0, zone DMA
16299 6171 330 65 15 0 0 0 0 0 0
0 2805 5787 1537 614 27 0 0 0 0 0
Node 0, zone DMA
16014 6170 330 65 15 0 0 0 0 0 0
0 2805 5789 1537 614 27 0 0 0 0 0
Node 0, zone DMA
15826 6171 333 65 15 0 0 0 0 0 0
0 2805 5790 1537 614 27 0 0 0 0 0
Node 0, zone DMA
15850 6202 334 65 15 0 0 0 0 0 0
0 2723 5793 1537 614 27 0 0 0 0 0
Node 0, zone DMA
15247 6203 335 65 15 0 0 0 0 0 0
0 2725 5794 1537 614 27 0 0 0 0 0
Node 0, zone DMA
15267 6207 332 65 15 0 0 0 0 0 0
0 2680 5793 1537 614 27 0 0 0 0 0
Node 0, zone DMA
14567 6208 333 65 15 0 0 0 0 0 0
0 2688 5794 1537 614 27 0 0 0 0 0
Node 0, zone DMA
14323 6206 333 65 15 0 0 0 0 0 0
0 2688 5794 1537 614 27 0 0 0 0 0
Node 0, zone DMA
13627 6208 334 65 15 0 0 0 0 0 0
0 2600 5794 1537 614 27 0 0 0 0 0
Node 0, zone DMA
13482 6204 335 65 15 0 0 0 0 0 0
0 2599 5794 1537 614 27 0 0 0 0 0
Node 0, zone DMA
13526 6214 292 65 15 0 0 0 0 0 0
0 2602 5794 1537 614 27 0 0 0 0 0
Node 0, zone DMA
11299 6221 294 65 15 0 0 0 0 0 0
0 2604 5794 1537 614 27 0 0 0 0 0
Node 0, zone DMA
11526 6228 245 65 15 0 0 0 0 0 0
0 2626 5826 1537 614 27 0 0 0 0 0
Node 0, zone DMA
11328 6230 244 65 15 0 0 0 0 0 0
0 2630 5825 1537 614 27 0 0 0 0 0
Node 0, zone DMA
12256 6355 264 65 15 0 0 0 0 0 0
0 2774 5827 1537 614 27 0 0 0 0 0
Node 0, zone DMA
12121 6349 276 66 15 0 0 0 0 0 0
0 2709 5828 1536 614 27 0 0 0 0 0
Node 0, zone DMA
12892 6423 285 65 15 0 0 0 0 0 0
0 2708 5831 1536 614 27 0 0 0 0 0
Node 0, zone DMA
12514 6348 286 66 15 0 0 0 0 0 0
0 2765 5825 1536 614 27 0 0 0 0 0
Node 0, zone DMA
12295 6349 286 66 15 0 0 0 0 0 0
0 2765 5825 1536 614 27 0 0 0 0 0
Node 0, zone DMA
12167 6351 285 66 15 0 0 0 0 0 0
0 2765 5826 1536 614 27 0 0 0 0 0
Node 0, zone DMA
12041 6310 286 66 15 0 0 0 0 0 0
0 2776 5826 1536 614 27 0 0 0 0 0
Node 0, zone DMA
11564 6313 286 66 15 0 0 0 0 0 0
0 2777 5827 1536 614 27 0 0 0 0 0
Node 0, zone DMA
11268 6340 287 66 15 0 0 0 0 0 0
0 2650 5827 1536 614 27 0 0 0 0 0
Node 0, zone DMA
11020 6340 287 66 15 0 0 0 0 0 0
0 2650 5827 1536 614 27 0 0 0 0 0
Node 0, zone DMA
11182 6349 287 65 15 0 0 0 0 0 0
0 2653 5827 1536 614 27 0 0 0 0 0
Node 0, zone DMA
10961 6354 283 65 15 0 0 0 0 0 0
0 2615 5830 1536 614 27 0 0 0 0 0
Node 0, zone DMA
10620 6354 284 65 15 0 0 0 0 0 0
0 2615 5830 1536 614 27 0 0 0 0 0
Node 0, zone DMA
10462 6359 235 65 15 0 0 0 0 0 0
0 2659 5832 1536 614 27 0 0 0 0 0
Node 0, zone DMA
10349 6353 242 65 15 0 0 0 0 0 0
0 2661 5832 1536 614 27 0 0 0 0 0
Node 0, zone DMA
10076 6350 242 65 15 0 0 0 0 0 0
0 2661 5830 1536 614 27 0 0 0 0 0
Node 0, zone DMA
10482 6364 213 65 15 0 0 0 0 0 0
0 2655 5832 1536 614 27 0 0 0 0 0
Node 0, zone DMA
10325 6364 213 65 15 0 0 0 0 0 0
0 2656 5832 1536 614 27 0 0 0 0 0
Node 0, zone DMA
11014 6422 234 65 15 0 0 0 0 0 0
0 2742 5837 1536 614 27 0 0 0 0 0
Node 0, zone DMA
11201 6346 263 65 15 0 0 0 0 0 0
0 2765 5841 1536 614 27 0 0 0 0 0
Node 0, zone DMA
10974 6351 263 65 15 0 0 0 0 0 0
0 2765 5841 1536 614 27 0 0 0 0 0
On 2017/7/4 14:52, Michal Hocko wrote:
> On Tue 04-07-17 09:21:00, zhouxianrong wrote:
>> the test was done as follows:
>>
>> 1. the environment is android 7.0 and kernel is 4.1 and managed memory is 3.5GB
>
> There have been many changes in the compaction proper since than. Do you
> see the same problem with the current upstream kernel?
>
>> 2. every 4s startup one apk, total 100 more apks need to startup
>> 3. after finishing step 2, sample buddyinfo once and get the result
>
> How stable are those results?
>
--
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>
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH mm] introduce reverse buddy concept to reduce buddy fragment
2017-07-04 8:04 ` zhouxianrong
@ 2017-07-04 11:24 ` Michal Hocko
2017-07-04 11:46 ` Mel Gorman
0 siblings, 1 reply; 12+ messages in thread
From: Michal Hocko @ 2017-07-04 11:24 UTC (permalink / raw)
To: zhouxianrong
Cc: linux-mm, linux-kernel, akpm, vbabka, alexander.h.duyck, mgorman,
l.stach, vdavydov.dev, hannes, minchan, npiggin, kirill.shutemov,
gi-oh.kim, luto, keescook, mark.rutland, mingo, heiko.carstens,
iamjoonsoo.kim, rientjes, ming.ling, jack, ebru.akagunduz,
bigeasy, Mi.Sophia.Wang, zhouxiyu, weidu.du, fanghua3,
won.ho.park
On Tue 04-07-17 16:04:52, zhouxianrong wrote:
> every 2s i sample /proc/buddyinfo in the whole test process.
>
> the last about 90 samples were sampled after the test was done.
I've tried to explain to you that numbers without a proper testing
metodology and highlevel metrics you are interested in and comparision
to the base kernel are meaningless. I cannot draw any conclusion from
looking at numbers you have posted. Are high order allocations cheaper
to do with this patch? What about an averge order-0 allocation request?
You are touching memory allocator hot paths and those are really
sensitive to changes. It takes a lot of testing with different workloads
to prove that no new regressions are introduced. That being said, I
completely agree that reducing the memory fragmentation is an important
objective but touching the page allocator and adding new branches there
sounds like a problematic approach which would have to show _huge_
benefits to be mergeable. Is it possible to improve khugepaged to
accomplish the same thing?
--
Michal Hocko
SUSE Labs
--
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>
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH mm] introduce reverse buddy concept to reduce buddy fragment
2017-07-04 11:24 ` Michal Hocko
@ 2017-07-04 11:46 ` Mel Gorman
0 siblings, 0 replies; 12+ messages in thread
From: Mel Gorman @ 2017-07-04 11:46 UTC (permalink / raw)
To: Michal Hocko
Cc: zhouxianrong, linux-mm, linux-kernel, akpm, vbabka,
alexander.h.duyck, l.stach, vdavydov.dev, hannes, minchan,
npiggin, kirill.shutemov, gi-oh.kim, luto, keescook,
mark.rutland, mingo, heiko.carstens, iamjoonsoo.kim, rientjes,
ming.ling, jack, ebru.akagunduz, bigeasy, Mi.Sophia.Wang,
zhouxiyu, weidu.du, fanghua3, won.ho.park
On Tue, Jul 04, 2017 at 01:24:14PM +0200, Michal Hocko wrote:
> On Tue 04-07-17 16:04:52, zhouxianrong wrote:
> > every 2s i sample /proc/buddyinfo in the whole test process.
> >
> > the last about 90 samples were sampled after the test was done.
>
> I've tried to explain to you that numbers without a proper testing
> metodology and highlevel metrics you are interested in and comparision
> to the base kernel are meaningless. I cannot draw any conclusion from
> looking at numbers you have posted. Are high order allocations cheaper
> to do with this patch? What about an averge order-0 allocation request?
>
I have to agree. The patch is extremely complex for what it does which
is working around a limitation of the buddy allocator in general
(buddy's must be naturally aligned). There would have to be *strong*
justification that allocations fail even with compaction or a reclaim
cycle or that the latency is severely reduced -- neither which is
evident from the data presented. It would also have to be proven that
there is no overhead added in the general case to justify this so
without extensive justification for the complexity;
Naked-by: Mel Gorman <mgorman@suse.de>
> You are touching memory allocator hot paths and those are really
> sensitive to changes. It takes a lot of testing with different workloads
> to prove that no new regressions are introduced. That being said, I
> completely agree that reducing the memory fragmentation is an important
> objective but touching the page allocator and adding new branches there
> sounds like a problematic approach which would have to show _huge_
> benefits to be mergeable. Is it possible to improve khugepaged to
> accomplish the same thing?
Or if this is CMA related, a justification why alloc_contig_range cannot do
the same thing with a linear walk when the initial allocation attempt fails.
--
Mel Gorman
SUSE Labs
--
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>
^ permalink raw reply [flat|nested] 12+ messages in thread
* [PATCH mm] introduce reverse buddy concept to reduce buddy fragment
@ 2017-06-30 11:22 zhouxianrong
0 siblings, 0 replies; 12+ messages in thread
From: zhouxianrong @ 2017-06-30 11:22 UTC (permalink / raw)
To: linux-mm
Cc: linux-kernel, akpm, vbabka, alexander.h.duyck, mhocko, mgorman,
l.stach, vdavydov.dev, hannes, minchan, npiggin, kirill.shutemov,
gi-oh.kim, luto, keescook, mark.rutland, mingo, heiko.carstens,
iamjoonsoo.kim, rientjes, ming.ling, jack, ebru.akagunduz,
bigeasy, Mi.Sophia.Wang, zhouxianrong, zhouxiyu, weidu.du,
fanghua3, won.ho.park
From: z00281421 <z00281421@notesmail.huawei.com>
Signed-off-by: z00281421 <z00281421@notesmail.huawei.com>
---
include/linux/gfp.h | 8 +-
include/linux/mmzone.h | 2 +
include/linux/page-flags.h | 9 ++
include/linux/thread_info.h | 5 +-
mm/compaction.c | 17 ++++
mm/internal.h | 7 ++
mm/page_alloc.c | 222 +++++++++++++++++++++++++++++++++++++++----
mm/vmstat.c | 5 +-
8 files changed, 251 insertions(+), 24 deletions(-)
diff --git a/include/linux/gfp.h b/include/linux/gfp.h
index db373b9..f63d4d9 100644
--- a/include/linux/gfp.h
+++ b/include/linux/gfp.h
@@ -40,6 +40,7 @@
#define ___GFP_DIRECT_RECLAIM 0x400000u
#define ___GFP_WRITE 0x800000u
#define ___GFP_KSWAPD_RECLAIM 0x1000000u
+#define ___GFP_NOREVERSEBUDDY 0x2000000u
/* If the above are modified, __GFP_BITS_SHIFT may need updating */
/*
@@ -171,6 +172,10 @@
* __GFP_NOTRACK_FALSE_POSITIVE is an alias of __GFP_NOTRACK. It's a means of
* distinguishing in the source between false positives and allocations that
* cannot be supported (e.g. page tables).
+ *
+ * __GFP_NOREVERSEBUDDY does not allocate pages from reverse buddy list
+ * of current order. It make sure that allocation is alignment to same order
+ * with length order.
*/
#define __GFP_COLD ((__force gfp_t)___GFP_COLD)
#define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
@@ -178,9 +183,10 @@
#define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
#define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK)
#define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
+#define __GFP_NOREVERSEBUDDY ((__force gfp_t)___GFP_NOREVERSEBUDDY)
/* Room for N __GFP_FOO bits */
-#define __GFP_BITS_SHIFT 25
+#define __GFP_BITS_SHIFT 26
#define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
/*
diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 8e02b37..94237fe 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -89,7 +89,9 @@ enum {
struct free_area {
struct list_head free_list[MIGRATE_TYPES];
+ struct list_head free_list_reverse[MIGRATE_TYPES];
unsigned long nr_free;
+ unsigned long nr_free_reverse;
};
struct pglist_data;
diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
index 6b5818d..39d17d7 100644
--- a/include/linux/page-flags.h
+++ b/include/linux/page-flags.h
@@ -675,6 +675,15 @@ static inline int TestClearPageDoubleMap(struct page *page)
#define PAGE_KMEMCG_MAPCOUNT_VALUE (-512)
PAGE_MAPCOUNT_OPS(Kmemcg, KMEMCG)
+/*
+ * ReverseBuddy is enabled for the buddy allocator that allow allocating
+ * two adjacent same free order blocks other than buddy blocks and
+ * composing them as a order + 1 block. It is for reducing buddy
+ * fragment.
+ */
+#define PAGE_REVERSE_BUDDY_MAPCOUNT_VALUE (-1024)
+PAGE_MAPCOUNT_OPS(ReverseBuddy, REVERSE_BUDDY)
+
extern bool is_free_buddy_page(struct page *page);
__PAGEFLAG(Isolated, isolated, PF_ANY);
diff --git a/include/linux/thread_info.h b/include/linux/thread_info.h
index 5837387..b4a1605 100644
--- a/include/linux/thread_info.h
+++ b/include/linux/thread_info.h
@@ -28,9 +28,10 @@
#ifdef CONFIG_DEBUG_STACK_USAGE
# define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK | \
- __GFP_ZERO)
+ __GFP_NOREVERSEBUDDY | __GFP_ZERO)
#else
-# define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK)
+# define THREADINFO_GFP (GFP_KERNEL_ACCOUNT | __GFP_NOTRACK | \
+ __GFP_NOREVERSEBUDDY)
#endif
/*
diff --git a/mm/compaction.c b/mm/compaction.c
index 0fdfde0..a43f169 100644
--- a/mm/compaction.c
+++ b/mm/compaction.c
@@ -768,6 +768,20 @@ static bool too_many_isolated(struct zone *zone)
continue;
}
+ if (PageReverseBuddy(page)) {
+ unsigned long freepage_order = page_order_unsafe(page);
+
+ /*
+ * Without lock, we cannot be sure that what we got is
+ * a valid page order. Consider only values in the
+ * valid order range to prevent low_pfn overflow.
+ */
+ if (freepage_order > 0 &&
+ freepage_order < MAX_ORDER - 1)
+ low_pfn += (1UL << (freepage_order + 1)) - 1;
+ continue;
+ }
+
/*
* Regardless of being on LRU, compound pages such as THP and
* hugetlbfs are not to be compacted. We can potentially save
@@ -1005,6 +1019,9 @@ static bool suitable_migration_target(struct compact_control *cc,
return false;
}
+ if (PageReverseBuddy(page))
+ return false;
+
/* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
if (migrate_async_suitable(get_pageblock_migratetype(page)))
return true;
diff --git a/mm/internal.h b/mm/internal.h
index ccfc2a2..439b0a8 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -143,6 +143,13 @@ struct alloc_context {
return page_pfn ^ (1 << order);
}
+static inline unsigned long
+__find_reverse_buddy_pfn(unsigned long page_pfn, unsigned int order)
+{
+ return (page_pfn & (1 << order)) ? page_pfn + (1 << order) :
+ page_pfn - (1 << order);
+}
+
extern struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
unsigned long end_pfn, struct zone *zone);
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 9f9623d..ee1dc1b 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -715,6 +715,18 @@ static inline void rmv_page_order(struct page *page)
set_page_private(page, 0);
}
+static inline void set_reverse_page_order(struct page *page, unsigned int order)
+{
+ set_page_private(page, order);
+ __SetPageReverseBuddy(page);
+}
+
+static inline void rmv_reverse_page_order(struct page *page)
+{
+ __ClearPageReverseBuddy(page);
+ set_page_private(page, 0);
+}
+
/*
* This function checks whether a page is free && is the buddy
* we can do coalesce a page and its buddy if
@@ -758,6 +770,120 @@ static inline int page_is_buddy(struct page *page, struct page *buddy,
return 0;
}
+static inline int page_is_reverse_buddy(struct page *page, struct page *buddy,
+ unsigned int order)
+{
+ if (PageReverseBuddy(buddy) && page_order(buddy) == order) {
+ /*
+ * zone check is done late to avoid uselessly
+ * calculating zone/node ids for pages that could
+ * never merge.
+ */
+ if (page_zone_id(page) != page_zone_id(buddy))
+ return 0;
+
+ VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
+
+ return 1;
+ }
+ return 0;
+}
+
+static inline
+void set_reverse_free_area(struct zone *zone, struct page *page,
+ unsigned int order, int migratetype)
+{
+ unsigned long buddy_pfn;
+ unsigned long reserve_buddy_pfn;
+ struct page *reverse_buddy;
+ struct free_area *area;
+
+ if (order > MAX_ORDER - 3)
+ return;
+ if (unlikely(is_migrate_isolate(migratetype)))
+ return;
+ buddy_pfn = page_to_pfn(page);
+ reserve_buddy_pfn = __find_reverse_buddy_pfn(buddy_pfn, order);
+ if ((buddy_pfn ^ reserve_buddy_pfn) & ~(pageblock_nr_pages - 1))
+ return;
+ if (!pfn_valid_within(reserve_buddy_pfn))
+ return;
+ reverse_buddy = pfn_to_page(reserve_buddy_pfn);
+ if (page_zone_id(page) != page_zone_id(reverse_buddy))
+ return;
+ if (PageBuddy(reverse_buddy) &&
+ page_order(reverse_buddy) == order) {
+ area = &zone->free_area[order];
+ list_del(&page->lru);
+ rmv_page_order(page);
+ area->nr_free--;
+ set_pcppage_migratetype(page, migratetype);
+ list_del(&reverse_buddy->lru);
+ rmv_page_order(reverse_buddy);
+ area->nr_free--;
+ set_pcppage_migratetype(reverse_buddy, migratetype);
+ area++;
+ if (buddy_pfn < reserve_buddy_pfn) {
+ list_add(&page->lru,
+ &area->free_list_reverse[migratetype]);
+ area->nr_free_reverse++;
+ set_reverse_page_order(page, order);
+ set_reverse_page_order(reverse_buddy, order);
+ } else {
+ list_add(&reverse_buddy->lru,
+ &area->free_list_reverse[migratetype]);
+ area->nr_free_reverse++;
+ set_reverse_page_order(reverse_buddy, order);
+ set_reverse_page_order(page, order);
+ }
+ }
+}
+
+static inline
+void rmv_reverse_free_area(struct zone *zone, struct page *page,
+ unsigned int order, int migratetype)
+{
+ unsigned long pfn, buddy_pfn;
+ unsigned long reserve_buddy_pfn;
+ struct page *buddy, *reverse_buddy;
+ struct free_area *area;
+
+ pfn = page_to_pfn(page);
+ buddy_pfn = __find_buddy_index(pfn, order);
+ buddy = page + (buddy_pfn - pfn);
+ if (!pfn_valid_within(buddy_pfn))
+ return;
+ if (!page_is_reverse_buddy(page, buddy, order))
+ return;
+
+ area = &zone->free_area[order];
+ reserve_buddy_pfn = __find_reverse_buddy_pfn(buddy_pfn, order);
+ reverse_buddy = pfn_to_page(reserve_buddy_pfn);
+ VM_BUG_ON_PAGE(!PageReverseBuddy(reverse_buddy) ||
+ page_order(reverse_buddy) != order, reverse_buddy);
+ if (buddy_pfn < reserve_buddy_pfn) {
+ list_move(&buddy->lru, &area->free_list[migratetype]);
+ rmv_reverse_page_order(buddy);
+ area[1].nr_free_reverse--;
+ area->nr_free++;
+ set_page_order(buddy, order);
+ rmv_reverse_page_order(reverse_buddy);
+ list_add(&reverse_buddy->lru, &area->free_list[migratetype]);
+ area->nr_free++;
+ set_page_order(reverse_buddy, order);
+ } else {
+ list_move(&reverse_buddy->lru, &area->free_list[migratetype]);
+ rmv_reverse_page_order(reverse_buddy);
+ area[1].nr_free_reverse--;
+ area->nr_free++;
+ set_page_order(reverse_buddy, order);
+ rmv_reverse_page_order(buddy);
+ list_add(&buddy->lru, &area->free_list[migratetype]);
+ area->nr_free++;
+ set_page_order(buddy, order);
+ }
+}
+
/*
* Freeing function for a buddy system allocator.
*
@@ -805,6 +931,7 @@ static inline void __free_one_page(struct page *page,
VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
VM_BUG_ON_PAGE(bad_range(zone, page), page);
+ rmv_reverse_free_area(zone, page, order, migratetype);
continue_merging:
while (order < max_order - 1) {
buddy_pfn = __find_buddy_pfn(pfn, order);
@@ -882,6 +1009,7 @@ static inline void __free_one_page(struct page *page,
list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
out:
zone->free_area[order].nr_free++;
+ set_reverse_free_area(zone, page, order, migratetype);
}
/*
@@ -1238,14 +1366,24 @@ void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
static void __free_pages_ok(struct page *page, unsigned int order)
{
+ bool reverse_buddy;
int migratetype;
unsigned long pfn = page_to_pfn(page);
if (!free_pages_prepare(page, order, true))
return;
+ reverse_buddy = order &&
+ ((pfn & ((1 << order) - 1)) == (1 << (order - 1)));
migratetype = get_pfnblock_migratetype(page, pfn);
- free_one_page(page_zone(page), page, pfn, order, migratetype);
+ if (!reverse_buddy)
+ free_one_page(page_zone(page), page, pfn, order, migratetype);
+ else {
+ free_one_page(page_zone(page), page,
+ pfn, order - 1, migratetype);
+ free_one_page(page_zone(page), page + (1 << (order - 1)),
+ pfn + (1 << (order - 1)), order - 1, migratetype);
+ }
}
static void __init __free_pages_boot_core(struct page *page, unsigned int order)
@@ -1651,6 +1789,25 @@ static inline void expand(struct zone *zone, struct page *page,
}
}
+static inline void expand_reverse(struct zone *zone, struct page *page,
+ int low, int high, struct free_area *area,
+ int migratetype)
+{
+ struct page *reverse_buddy;
+
+ reverse_buddy = page + (1 << (high - 1));
+ rmv_reverse_page_order(reverse_buddy);
+ set_pcppage_migratetype(reverse_buddy, migratetype);
+ if (high > low) {
+ area--;
+ high--;
+ expand(zone, page, low, high, area, migratetype);
+ list_add(&reverse_buddy->lru, &area->free_list[migratetype]);
+ area->nr_free++;
+ set_page_order(reverse_buddy, high);
+ }
+}
+
static void check_new_page_bad(struct page *page)
{
const char *bad_reason = NULL;
@@ -1785,25 +1942,45 @@ static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags
*/
static inline
struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
- int migratetype)
+ int migratetype, gfp_t gfp_flags)
{
+ bool reverse_buddy;
unsigned int current_order;
struct free_area *area;
struct page *page;
+ reverse_buddy = !!(gfp_flags & __GFP_NOREVERSEBUDDY);
/* Find a page of the appropriate size in the preferred list */
for (current_order = order; current_order < MAX_ORDER; ++current_order) {
area = &(zone->free_area[current_order]);
page = list_first_entry_or_null(&area->free_list[migratetype],
struct page, lru);
- if (!page)
+ if (page) {
+ list_del(&page->lru);
+ rmv_page_order(page);
+ area->nr_free--;
+ expand(zone, page, order,
+ current_order, area, migratetype);
+ set_pcppage_migratetype(page, migratetype);
+ return page;
+ }
+ if (current_order + reverse_buddy == 0 ||
+ current_order + reverse_buddy > MAX_ORDER - 2)
continue;
- list_del(&page->lru);
- rmv_page_order(page);
- area->nr_free--;
- expand(zone, page, order, current_order, area, migratetype);
- set_pcppage_migratetype(page, migratetype);
- return page;
+ area += reverse_buddy;
+ page = list_first_entry_or_null(
+ &area->free_list_reverse[migratetype],
+ struct page, lru);
+ if (page) {
+ list_del(&page->lru);
+ rmv_reverse_page_order(page);
+ area->nr_free_reverse--;
+ expand_reverse(zone, page, order,
+ current_order + reverse_buddy,
+ area, migratetype);
+ set_pcppage_migratetype(page, migratetype);
+ return page;
+ }
}
return NULL;
@@ -1828,13 +2005,13 @@ struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
#ifdef CONFIG_CMA
static struct page *__rmqueue_cma_fallback(struct zone *zone,
- unsigned int order)
+ unsigned int order, gfp_t gfp_flags)
{
- return __rmqueue_smallest(zone, order, MIGRATE_CMA);
+ return __rmqueue_smallest(zone, order, MIGRATE_CMA, gfp_flags);
}
#else
static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
- unsigned int order) { return NULL; }
+ unsigned int order, gfp_t gfp_flags) { return NULL; }
#endif
/*
@@ -2136,7 +2313,8 @@ static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
/* Remove an element from the buddy allocator from the fallback list */
static inline struct page *
-__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
+__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype,
+ gfp_t gfp_flags)
{
struct free_area *area;
unsigned int current_order;
@@ -2190,17 +2368,18 @@ static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
* Call me with the zone->lock already held.
*/
static struct page *__rmqueue(struct zone *zone, unsigned int order,
- int migratetype)
+ int migratetype, gfp_t gfp_flags)
{
struct page *page;
- page = __rmqueue_smallest(zone, order, migratetype);
+ page = __rmqueue_smallest(zone, order, migratetype, gfp_flags);
if (unlikely(!page)) {
if (migratetype == MIGRATE_MOVABLE)
- page = __rmqueue_cma_fallback(zone, order);
+ page = __rmqueue_cma_fallback(zone, order, gfp_flags);
if (!page)
- page = __rmqueue_fallback(zone, order, migratetype);
+ page = __rmqueue_fallback(zone, order, migratetype,
+ gfp_flags);
}
trace_mm_page_alloc_zone_locked(page, order, migratetype);
@@ -2221,7 +2400,7 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order,
spin_lock_irqsave(&zone->lock, flags);
for (i = 0; i < count; ++i) {
- struct page *page = __rmqueue(zone, order, migratetype);
+ struct page *page = __rmqueue(zone, order, migratetype, 0);
if (unlikely(page == NULL))
break;
@@ -2718,12 +2897,13 @@ struct page *rmqueue(struct zone *preferred_zone,
do {
page = NULL;
if (alloc_flags & ALLOC_HARDER) {
- page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
+ page = __rmqueue_smallest(zone, order,
+ MIGRATE_HIGHATOMIC, gfp_flags);
if (page)
trace_mm_page_alloc_zone_locked(page, order, migratetype);
}
if (!page)
- page = __rmqueue(zone, order, migratetype);
+ page = __rmqueue(zone, order, migratetype, gfp_flags);
} while (page && check_new_pages(page, order));
spin_unlock(&zone->lock);
if (!page)
@@ -5286,7 +5466,9 @@ static void __meminit zone_init_free_lists(struct zone *zone)
unsigned int order, t;
for_each_migratetype_order(order, t) {
INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
+ INIT_LIST_HEAD(&zone->free_area[order].free_list_reverse[t]);
zone->free_area[order].nr_free = 0;
+ zone->free_area[order].nr_free_reverse = 0;
}
}
diff --git a/mm/vmstat.c b/mm/vmstat.c
index 69f9aff..26007df 100644
--- a/mm/vmstat.c
+++ b/mm/vmstat.c
@@ -1146,10 +1146,13 @@ static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,
{
int order;
- seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
+ seq_printf(m, "Node %d, zone %8s\n", pgdat->node_id, zone->name);
for (order = 0; order < MAX_ORDER; ++order)
seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
seq_putc(m, '\n');
+ for (order = 0; order < MAX_ORDER; ++order)
+ seq_printf(m, "%6lu ", zone->free_area[order].nr_free_reverse);
+ seq_putc(m, '\n');
}
/*
--
1.7.9.5
--
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>
^ permalink raw reply related [flat|nested] 12+ messages in thread
end of thread, other threads:[~2017-07-04 11:46 UTC | newest]
Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-06-30 11:25 [PATCH mm] introduce reverse buddy concept to reduce buddy fragment zhouxianrong
2017-07-03 7:48 ` Michal Hocko
2017-07-03 9:01 ` zhouxianrong
2017-07-03 12:02 ` zhouxianrong
2017-07-03 15:33 ` Michal Hocko
2017-07-04 1:21 ` zhouxianrong
2017-07-04 6:52 ` Michal Hocko
2017-07-04 7:00 ` zhouxianrong
2017-07-04 8:04 ` zhouxianrong
2017-07-04 11:24 ` Michal Hocko
2017-07-04 11:46 ` Mel Gorman
-- strict thread matches above, loose matches on Subject: below --
2017-06-30 11:22 zhouxianrong
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).