Re: [PATCH v4] mm/page_alloc: Consolidate unlikely handling in page_expected_state

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

在 2025/4/3 11:55, Vishal Moola (Oracle) 写道:
> On Wed, Apr 02, 2025 at 10:49:17AM +0800, Ye Liu wrote:
>> 在 2025/3/31 23:59, Matthew Wilcox 写道:
>>> On Mon, Mar 31, 2025 at 08:08:01PM +0800, Ye Liu wrote:
>>>> 在 2025/3/28 22:29, Matthew Wilcox 写道:
>>>>> On Fri, Mar 28, 2025 at 09:47:57AM +0800, Ye Liu wrote:
>>>>>> Consolidate the handling of unlikely conditions in the 
>>>>>> page_expected_state() function to reduce code duplication and improve 
>>>>>> readability.
>>>>> I don't think this is an equivalent transformation.
>>>> Could you explain it in detail?
>>> page_expected_state() is called both at free and alloc.  I think
>>> the correct behaviour on encountering a HWPOISON page should be
>>> different at alloc and free, don't you?
>> In the alloc process, this patch does not modify the code behavior.
>> Regarding the free process, the if (unlikely(PageHWPoison(page)) && !order)
>> code handles the case where order is 0. When order is not 0, it does not
>> matter if __ClearPageBuddy is used to process the last page of the compound
>> page, because page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; will also clear it.
>> Is that right?                                                             
> I believe a page's "Buddy" flag is stored in page->type, so maybe this
> won't do what you expect?


Got it, thanks! I originally thought that page->page_type would also be
cleared in the free process, meaning the free behavior wouldn't change,
but that’s not actually the case. I need to look into what impact it
would have if __ClearPageBuddy is also executed in the free process.   


Thanks,
Ye Liu