2021-03-23 12:44:23

by Liu Shixin

[permalink] [raw]
Subject: [PATCH -next] mm, page_alloc: avoid page_to_pfn() in move_freepages()

From: Kefeng Wang <[email protected]>

The start_pfn and end_pfn are already available in move_freepages_block(),
there is no need to go back and forth between page and pfn in move_freepages
and move_freepages_block, and pfn_valid_within() should validate pfn first
before touching the page.

Signed-off-by: Kefeng Wang <[email protected]>
Signed-off-by: Liu Shixin <[email protected]>
---
mm/page_alloc.c | 28 +++++++++++++---------------
1 file changed, 13 insertions(+), 15 deletions(-)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index c53fe4fa10bf..ccfaa8158862 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -2425,19 +2425,21 @@ static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
* boundary. If alignment is required, use move_freepages_block()
*/
static int move_freepages(struct zone *zone,
- struct page *start_page, struct page *end_page,
+ unsigned long start_pfn, unsigned long end_pfn,
int migratetype, int *num_movable)
{
struct page *page;
+ unsigned long pfn;
unsigned int order;
int pages_moved = 0;

- for (page = start_page; page <= end_page;) {
- if (!pfn_valid_within(page_to_pfn(page))) {
- page++;
+ for (pfn = start_pfn; pfn <= end_pfn;) {
+ if (!pfn_valid_within(pfn)) {
+ pfn++;
continue;
}

+ page = pfn_to_page(pfn);
if (!PageBuddy(page)) {
/*
* We assume that pages that could be isolated for
@@ -2447,8 +2449,7 @@ static int move_freepages(struct zone *zone,
if (num_movable &&
(PageLRU(page) || __PageMovable(page)))
(*num_movable)++;
-
- page++;
+ pfn++;
continue;
}

@@ -2458,7 +2459,7 @@ static int move_freepages(struct zone *zone,

order = buddy_order(page);
move_to_free_list(page, zone, order, migratetype);
- page += 1 << order;
+ pfn += 1 << order;
pages_moved += 1 << order;
}

@@ -2468,25 +2469,22 @@ static int move_freepages(struct zone *zone,
int move_freepages_block(struct zone *zone, struct page *page,
int migratetype, int *num_movable)
{
- unsigned long start_pfn, end_pfn;
- struct page *start_page, *end_page;
+ unsigned long start_pfn, end_pfn, pfn;

if (num_movable)
*num_movable = 0;

- start_pfn = page_to_pfn(page);
- start_pfn = start_pfn & ~(pageblock_nr_pages-1);
- start_page = pfn_to_page(start_pfn);
- end_page = start_page + pageblock_nr_pages - 1;
+ pfn = page_to_pfn(page);
+ start_pfn = pfn & ~(pageblock_nr_pages - 1);
end_pfn = start_pfn + pageblock_nr_pages - 1;

/* Do not cross zone boundaries */
if (!zone_spans_pfn(zone, start_pfn))
- start_page = page;
+ start_pfn = pfn;
if (!zone_spans_pfn(zone, end_pfn))
return 0;

- return move_freepages(zone, start_page, end_page, migratetype,
+ return move_freepages(zone, start_pfn, end_pfn, migratetype,
num_movable);
}

--
2.25.1


2021-03-23 13:39:22

by Matthew Wilcox

[permalink] [raw]
Subject: Re: [PATCH -next] mm, page_alloc: avoid page_to_pfn() in move_freepages()

On Tue, Mar 23, 2021 at 09:12:15PM +0800, Liu Shixin wrote:
> From: Kefeng Wang <[email protected]>
>
> The start_pfn and end_pfn are already available in move_freepages_block(),
> there is no need to go back and forth between page and pfn in move_freepages
> and move_freepages_block, and pfn_valid_within() should validate pfn first
> before touching the page.

This looks good to me:

Reviewed-by: Matthew Wilcox (Oracle) <[email protected]>

> static int move_freepages(struct zone *zone,
> - struct page *start_page, struct page *end_page,
> + unsigned long start_pfn, unsigned long end_pfn,
> int migratetype, int *num_movable)
> {
> struct page *page;
> + unsigned long pfn;
> unsigned int order;
> int pages_moved = 0;
>
> - for (page = start_page; page <= end_page;) {
> - if (!pfn_valid_within(page_to_pfn(page))) {
> - page++;
> + for (pfn = start_pfn; pfn <= end_pfn;) {
> + if (!pfn_valid_within(pfn)) {
> + pfn++;
> continue;
> }
>
> + page = pfn_to_page(pfn);

I wonder if this wouldn't be even better if we did:

struct page *start_page = pfn_to_page(start_pfn);

for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
struct page *page = start_page + pfn - start_pfn;

if (!pfn_valid_within(pfn))
continue;

> -
> - page++;
> + pfn++;
> continue;

... then we can drop the increment of pfn here

> }
>
> @@ -2458,7 +2459,7 @@ static int move_freepages(struct zone *zone,
>
> order = buddy_order(page);
> move_to_free_list(page, zone, order, migratetype);
> - page += 1 << order;
> + pfn += 1 << order;

... and change this to pfn += (1 << order) - 1;

Do you have any numbers to quantify the benefit of this change?

2021-03-27 03:38:08

by Liu Shixin

[permalink] [raw]
Subject: Re: [PATCH -next] mm, page_alloc: avoid page_to_pfn() in move_freepages()

Sorry to reply to you after a so long time and thanks for your advice. It does seem that your proposed change will make the code cleaner and more efficient.

I repeated move_freepages_block() 2000000 times on the VM and counted jiffies. The average value before and after the change was both about 12,000. I think it's probably because I'm using the Sparse Memory Model, so pfn_to_page() is not time-consuming.


On 2021/3/23 20:54, Matthew Wilcox wrote:
> On Tue, Mar 23, 2021 at 09:12:15PM +0800, Liu Shixin wrote:
>> From: Kefeng Wang <[email protected]>
>>
>> The start_pfn and end_pfn are already available in move_freepages_block(),
>> there is no need to go back and forth between page and pfn in move_freepages
>> and move_freepages_block, and pfn_valid_within() should validate pfn first
>> before touching the page.
> This looks good to me:
>
> Reviewed-by: Matthew Wilcox (Oracle) <[email protected]>
>
>> static int move_freepages(struct zone *zone,
>> - struct page *start_page, struct page *end_page,
>> + unsigned long start_pfn, unsigned long end_pfn,
>> int migratetype, int *num_movable)
>> {
>> struct page *page;
>> + unsigned long pfn;
>> unsigned int order;
>> int pages_moved = 0;
>>
>> - for (page = start_page; page <= end_page;) {
>> - if (!pfn_valid_within(page_to_pfn(page))) {
>> - page++;
>> + for (pfn = start_pfn; pfn <= end_pfn;) {
>> + if (!pfn_valid_within(pfn)) {
>> + pfn++;
>> continue;
>> }
>>
>> + page = pfn_to_page(pfn);
> I wonder if this wouldn't be even better if we did:
>
> struct page *start_page = pfn_to_page(start_pfn);
>
> for (pfn = start_pfn; pfn <= end_pfn; pfn++) {
> struct page *page = start_page + pfn - start_pfn;
>
> if (!pfn_valid_within(pfn))
> continue;
>
>> -
>> - page++;
>> + pfn++;
>> continue;
> ... then we can drop the increment of pfn here
>
>> }
>>
>> @@ -2458,7 +2459,7 @@ static int move_freepages(struct zone *zone,
>>
>> order = buddy_order(page);
>> move_to_free_list(page, zone, order, migratetype);
>> - page += 1 << order;
>> + pfn += 1 << order;
> ... and change this to pfn += (1 << order) - 1;
>
> Do you have any numbers to quantify the benefit of this change?
> .
>

2021-03-29 15:32:58

by Vlastimil Babka

[permalink] [raw]
Subject: Re: [PATCH -next] mm, page_alloc: avoid page_to_pfn() in move_freepages()

On 3/23/21 2:12 PM, Liu Shixin wrote:
> From: Kefeng Wang <[email protected]>
>
> The start_pfn and end_pfn are already available in move_freepages_block(),
> there is no need to go back and forth between page and pfn in move_freepages
> and move_freepages_block, and pfn_valid_within() should validate pfn first
> before touching the page.
>
> Signed-off-by: Kefeng Wang <[email protected]>
> Signed-off-by: Liu Shixin <[email protected]>

Acked-by: Vlastimil Babka <[email protected]>

Agreed with Matthew's suggestion, also:

> @@ -2468,25 +2469,22 @@ static int move_freepages(struct zone *zone,
> int move_freepages_block(struct zone *zone, struct page *page,
> int migratetype, int *num_movable)
> {
> - unsigned long start_pfn, end_pfn;
> - struct page *start_page, *end_page;
> + unsigned long start_pfn, end_pfn, pfn;
>
> if (num_movable)
> *num_movable = 0;
>
> - start_pfn = page_to_pfn(page);
> - start_pfn = start_pfn & ~(pageblock_nr_pages-1);
> - start_page = pfn_to_page(start_pfn);
> - end_page = start_page + pageblock_nr_pages - 1;
> + pfn = page_to_pfn(page);
> + start_pfn = pfn & ~(pageblock_nr_pages - 1);

Since you touch this already, consider pageblock_start_pfn()

> end_pfn = start_pfn + pageblock_nr_pages - 1;

I'd also drop the "- 1" here and instead adjust the for loop's ending condition
from "pfn <= end_pfn" to "pfn < end_pfn" as that's more common way of doing it.

Thanks.

>
> /* Do not cross zone boundaries */
> if (!zone_spans_pfn(zone, start_pfn))
> - start_page = page;
> + start_pfn = pfn;
> if (!zone_spans_pfn(zone, end_pfn))
> return 0;
>
> - return move_freepages(zone, start_page, end_page, migratetype,
> + return move_freepages(zone, start_pfn, end_pfn, migratetype,
> num_movable);
> }
>
>