在 2025/3/21 13:40, Anshuman Khandual 写道:
>
> On 3/21/25 06:43, Ye Liu wrote:
>> From: Ye Liu <liuye@xxxxxxxxxx>
>>
>> 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@xxxxxxxxxx>
>> ---
>> 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;
>> }
>>
