If buddy page has order n, (2^n) pages are filtered at the same
time. Then, next (2^n) pages are already filterd, so we can skip the
(2^n) pages for free.
By this, on 32GB memory system, there's the following improvement:
- with mmap() on /proc/vmcore from 0.594421 seconds to 0.496517
seconds, and
- with normal read() from 1.533796 seconds to 0.750935 seconds.
Note that the two improvements are measured on the same /proc/vmcore.
Note also that amount of the improvement depends on amount of free
pages, though it's obvious.
Signed-off-by: HATAYAMA Daisuke <d.hatayama at jp.fujitsu.com>
---
makedumpfile.c | 7 +++++--
1 file changed, 5 insertions(+), 2 deletions(-)
diff --git a/makedumpfile.c b/makedumpfile.c
index 7acbf72..9075f38 100644
--- a/makedumpfile.c
+++ b/makedumpfile.c
@@ -4303,9 +4303,9 @@ __exclude_unnecessary_pages(unsigned long mem_map,
&& info->flag_cyclic
&& info->page_is_buddy
&& info->page_is_buddy(flags, _mapcount, private, _count)) {
- int i;
+ int i, nr_pages = 1 << private;
- for (i = 0; i < (1 << private); ++i) {
+ for (i = 0; i < nr_pages; ++i) {
/*
* According to combination of
* MAX_ORDER and size of cyclic
@@ -4318,6 +4318,9 @@ __exclude_unnecessary_pages(unsigned long mem_map,
clear_bit_on_2nd_bitmap_for_kernel(pfn + i);
}
pfn_free += i;
+
+ pfn += nr_pages - 1;
+ mem_map += (nr_pages - 1) * SIZE(page);
}
/*
* Exclude the cache page without the private page.
--
1.8.1.4
