* [PATCH] mm/page_alloc: Consolidate unlikely handling in page_expected_state
@ 2025-03-21 1:13 Ye Liu
2025-03-21 5:40 ` Anshuman Khandual
0 siblings, 1 reply; 3+ messages in thread
From: Ye Liu @ 2025-03-21 1:13 UTC (permalink / raw)
To: akpm; +Cc: linux-mm, linux-kernel, Ye Liu
From: Ye Liu <liuye@kylinos.cn>
This patch consolidates the handling of unlikely conditions in the
page_expected_state function, reducing code duplication and improving
readability.
Previously, the check_new_page_bad function contained logic to handle
__PG_HWPOISON flags, which was called from check_new_page. This patch
moves the handling of __PG_HWPOISON flags into the page_expected_state
function and removes the check_new_page_bad function. The check_new_page
function now directly calls bad_page if the page has unexpected flags.
This change simplifies the code by reducing the number of functions and
centralizing the unlikely condition handling in one place.
Signed-off-by: Ye Liu <liuye@kylinos.cn>
---
mm/page_alloc.c | 26 ++++++++++----------------
1 file changed, 10 insertions(+), 16 deletions(-)
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 2842da893eea..d8d04ac1d709 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -892,6 +892,13 @@ static inline bool page_expected_state(struct page *page,
if (unlikely(atomic_read(&page->_mapcount) != -1))
return false;
+ if (unlikely(PageHWPoison(page))) {
+ /* Don't complain about hwpoisoned pages */
+ if (PageBuddy(page))
+ __ClearPageBuddy(page);
+ return false;
+ }
+
if (unlikely((unsigned long)page->mapping |
page_ref_count(page) |
#ifdef CONFIG_MEMCG
@@ -1586,29 +1593,16 @@ static __always_inline void page_del_and_expand(struct zone *zone,
account_freepages(zone, -nr_pages, migratetype);
}
-static void check_new_page_bad(struct page *page)
-{
- if (unlikely(PageHWPoison(page))) {
- /* Don't complain about hwpoisoned pages */
- if (PageBuddy(page))
- __ClearPageBuddy(page);
- return;
- }
-
- bad_page(page,
- page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP));
-}
-
/*
* This page is about to be returned from the page allocator
*/
static bool check_new_page(struct page *page)
{
- if (likely(page_expected_state(page,
- PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
+ if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_PREP)))
return false;
- check_new_page_bad(page);
+ bad_page(page,
+ page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP));
return true;
}
--
2.25.1
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] mm/page_alloc: Consolidate unlikely handling in page_expected_state
2025-03-21 1:13 [PATCH] mm/page_alloc: Consolidate unlikely handling in page_expected_state Ye Liu
@ 2025-03-21 5:40 ` Anshuman Khandual
2025-03-21 5:59 ` Ye Liu
0 siblings, 1 reply; 3+ messages in thread
From: Anshuman Khandual @ 2025-03-21 5:40 UTC (permalink / raw)
To: Ye Liu, akpm; +Cc: linux-mm, linux-kernel, Ye Liu
On 3/21/25 06:43, Ye Liu wrote:
> From: Ye Liu <liuye@kylinos.cn>
>
> This patch consolidates the handling of unlikely conditions in the
> page_expected_state function, reducing code duplication and improving
> readability.
>
> Previously, the check_new_page_bad function contained logic to handle
> __PG_HWPOISON flags, which was called from check_new_page. This patch
> moves the handling of __PG_HWPOISON flags into the page_expected_state
> function and removes the check_new_page_bad function. The check_new_page
> function now directly calls bad_page if the page has unexpected flags.
>
> This change simplifies the code by reducing the number of functions and
> centralizing the unlikely condition handling in one place.
>
> Signed-off-by: Ye Liu <liuye@kylinos.cn>
> ---
> mm/page_alloc.c | 26 ++++++++++----------------
> 1 file changed, 10 insertions(+), 16 deletions(-)
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 2842da893eea..d8d04ac1d709 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -892,6 +892,13 @@ static inline bool page_expected_state(struct page *page,
> if (unlikely(atomic_read(&page->_mapcount) != -1))
> return false;
>
> + if (unlikely(PageHWPoison(page))) {
> + /* Don't complain about hwpoisoned pages */
> + if (PageBuddy(page))
> + __ClearPageBuddy(page);
> + return false;
Should this be return 'true' instead ?
Let's consider a scenario where PageHWPoison(page) is true.
Previously bad_page() was not getting called as check_new_page_bad() will
return earlier before reaching bad_page().
But now with the proposed change here page_expected_state() returns false
and hence bad_page() still gets called later on in check_new_page().
There is a change in behaviour - or am I missing something here ?
> + }
> +
> if (unlikely((unsigned long)page->mapping |
> page_ref_count(page) |
> #ifdef CONFIG_MEMCG
> @@ -1586,29 +1593,16 @@ static __always_inline void page_del_and_expand(struct zone *zone,
> account_freepages(zone, -nr_pages, migratetype);
> }
>
> -static void check_new_page_bad(struct page *page)
> -{
> - if (unlikely(PageHWPoison(page))) {
> - /* Don't complain about hwpoisoned pages */
> - if (PageBuddy(page))
> - __ClearPageBuddy(page);
> - return;
> - }
> -
> - bad_page(page,
> - page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP));
> -}
> -
> /*
> * This page is about to be returned from the page allocator
> */
> static bool check_new_page(struct page *page)
> {
> - if (likely(page_expected_state(page,
> - PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
> + if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_PREP)))
> return false;
>
> - check_new_page_bad(page);
> + bad_page(page,
> + page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP));
> return true;
> }
>
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] mm/page_alloc: Consolidate unlikely handling in page_expected_state
2025-03-21 5:40 ` Anshuman Khandual
@ 2025-03-21 5:59 ` Ye Liu
0 siblings, 0 replies; 3+ messages in thread
From: Ye Liu @ 2025-03-21 5:59 UTC (permalink / raw)
To: Anshuman Khandual, akpm; +Cc: linux-mm, linux-kernel, Ye Liu
在 2025/3/21 13:40, Anshuman Khandual 写道:
>
> On 3/21/25 06:43, Ye Liu wrote:
>> From: Ye Liu <liuye@kylinos.cn>
>>
>> This patch consolidates the handling of unlikely conditions in the
>> page_expected_state function, reducing code duplication and improving
>> readability.
>>
>> Previously, the check_new_page_bad function contained logic to handle
>> __PG_HWPOISON flags, which was called from check_new_page. This patch
>> moves the handling of __PG_HWPOISON flags into the page_expected_state
>> function and removes the check_new_page_bad function. The check_new_page
>> function now directly calls bad_page if the page has unexpected flags.
>>
>> This change simplifies the code by reducing the number of functions and
>> centralizing the unlikely condition handling in one place.
>>
>> Signed-off-by: Ye Liu <liuye@kylinos.cn>
>> ---
>> mm/page_alloc.c | 26 ++++++++++----------------
>> 1 file changed, 10 insertions(+), 16 deletions(-)
>>
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index 2842da893eea..d8d04ac1d709 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -892,6 +892,13 @@ static inline bool page_expected_state(struct page *page,
>> if (unlikely(atomic_read(&page->_mapcount) != -1))
>> return false;
>>
>> + if (unlikely(PageHWPoison(page))) {
>> + /* Don't complain about hwpoisoned pages */
>> + if (PageBuddy(page))
>> + __ClearPageBuddy(page);
>> + return false;
> Should this be return 'true' instead ?
>
> Let's consider a scenario where PageHWPoison(page) is true.
>
> Previously bad_page() was not getting called as check_new_page_bad() will
> return earlier before reaching bad_page().
>
> But now with the proposed change here page_expected_state() returns false
> and hence bad_page() still gets called later on in check_new_page().
>
> There is a change in behaviour - or am I missing something here ?
Thank you for the suggestion. You're right, it makes sense to return true in this case.
I'll update the patch accordingly. Appreciate your feedback!
>
>> + }
>> +
>> if (unlikely((unsigned long)page->mapping |
>> page_ref_count(page) |
>> #ifdef CONFIG_MEMCG
>> @@ -1586,29 +1593,16 @@ static __always_inline void page_del_and_expand(struct zone *zone,
>> account_freepages(zone, -nr_pages, migratetype);
>> }
>>
>> -static void check_new_page_bad(struct page *page)
>> -{
>> - if (unlikely(PageHWPoison(page))) {
>> - /* Don't complain about hwpoisoned pages */
>> - if (PageBuddy(page))
>> - __ClearPageBuddy(page);
>> - return;
>> - }
>> -
>> - bad_page(page,
>> - page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP));
>> -}
>> -
>> /*
>> * This page is about to be returned from the page allocator
>> */
>> static bool check_new_page(struct page *page)
>> {
>> - if (likely(page_expected_state(page,
>> - PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
>> + if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_PREP)))
>> return false;
>>
>> - check_new_page_bad(page);
>> + bad_page(page,
>> + page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP));
>> return true;
>> }
>>
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2025-03-21 5:59 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2025-03-21 1:13 [PATCH] mm/page_alloc: Consolidate unlikely handling in page_expected_state Ye Liu
2025-03-21 5:40 ` Anshuman Khandual
2025-03-21 5:59 ` Ye Liu
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox