Re: [PATCH] mm/page_alloc: Consolidate unlikely handling in page_expected_state

[Ye Liu <ye.liu@linux.dev>]

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;
>>  }
>>