Thread (6 messages) 6 messages, 5 authors, 2016-06-16

[PATCH] ARM: mm: Speed up page list initialization during boot

flat view

From: Jungseung Lee <hidden>
Date: 2015-12-28 10:15:44
Also in: linux-mm

Hi,
quoted hunk
During boot, we populate the page lists by using the page freeing 
mechanism on every individual page.  Unfortunately, this is very 
inefficient because the memory manager spends a lot of time coalescing 
pairs of adjacent free pages into bigger blocks.

Rather than adding a single order 0 page at a time, we can take 
advantage of the fact that we know that all the pages are available and 
free up big blocks of pages at a time instead.

Signed-off-by: Chirantan Ekbote <redacted>
---
arch/arm/mm/init.c  | 19 +++++++++++++++++--  include/linux/gfp.h |  1 
+
mm/internal.h       |  1 -
3 files changed, 18 insertions(+), 3 deletions(-)
diff --git a/arch/arm/mm/init.c b/arch/arm/mm/init.c index 
97c293e..c7fc2d8 100644
--- a/arch/arm/mm/init.c
+++ b/arch/arm/mm/init.c
@@ -22,6 +22,7 @@
#include <linux/memblock.h>
#include <linux/dma-contiguous.h>
#include <linux/sizes.h>
+#include <linux/bitops.h>

#include <asm/mach-types.h>
#include <asm/memblock.h>
@@ -469,8 +470,22 @@ static void __init free_unused_memmap(struct 
meminfo
*mi)
quoted hunk
#ifdef CONFIG_HIGHMEM
static inline void free_area_high(unsigned long pfn, unsigned long 
end)  {
-	for (; pfn < end; pfn++)
-		free_highmem_page(pfn_to_page(pfn));
+	while (pfn < end) {
+		struct page *page = pfn_to_page(pfn);
+		unsigned long order = min(__ffs(pfn), MAX_ORDER - 1);
+		unsigned long nr_pages = 1 << order;
+		unsigned long rem = end - pfn;
+
+		if (nr_pages > rem) {
+			order = __fls(rem);
+			nr_pages = 1 << order;
+		}
+
+		__free_pages_bootmem(page, order);
+		totalram_pages += nr_pages;
+		totalhigh_pages += nr_pages;
+		pfn += nr_pages;
+	}
}
#endif
diff --git a/include/linux/gfp.h b/include/linux/gfp.h index 
39b81dc..a63d666 100644
--- a/include/linux/gfp.h
+++ b/include/linux/gfp.h
@@ -367,6 +367,7 @@ void *alloc_pages_exact_nid(int nid, size_t size, 
gfp_t
gfp_mask);
quoted hunk
#define __get_dma_pages(gfp_mask, order) \
		__get_free_pages((gfp_mask) | GFP_DMA, (order))

+extern void __free_pages_bootmem(struct page *page, unsigned int 
+order);
extern void __free_pages(struct page *page, unsigned int order);  
extern void free_pages(unsigned long addr, unsigned int order);  extern 
void free_hot_cold_page(struct page *page, int cold); diff --git 
a/mm/internal.h b/mm/internal.h index 29e1e76..d2b8738 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -93,7 +93,6 @@ extern pmd_t *mm_find_pmd(struct mm_struct *mm, 
unsigned
long address);
/*
 * in mm/page_alloc.c
 */
-extern void __free_pages_bootmem(struct page *page, unsigned int 
order);  extern void prep_compound_page(struct page *page, unsigned 
long order);  #ifdef CONFIG_MEMORY_FAILURE  extern bool 
is_free_buddy_page(struct page *page);
--
1.9.1.423.g4596e3a
This patch really could save boot time. 
Is there any reason this patch is not merged to mainline kernel?

Best Regards,
Jungseung Lee
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help