Re: [PATCH v4] mm/huge_memory: preserve PG_has_hwpoisoned if a folio is split to >0 order

[Zi Yan <ziy@nvidia.com>]

On 24 Oct 2025, at 11:44, Lorenzo Stoakes wrote:

> On Wed, Oct 22, 2025 at 11:05:21PM -0400, Zi Yan wrote:
>> folio split clears PG_has_hwpoisoned, but the flag should be preserved in
>> after-split folios containing pages with PG_hwpoisoned flag if the folio is
>> split to >0 order folios. Scan all pages in a to-be-split folio to
>> determine which after-split folios need the flag.
>>
>> An alternatives is to change PG_has_hwpoisoned to PG_maybe_hwpoisoned to
>> avoid the scan and set it on all after-split folios, but resulting false
>> positive has undesirable negative impact. To remove false positive, caller
>> of folio_test_has_hwpoisoned() and folio_contain_hwpoisoned_page() needs to
>> do the scan. That might be causing a hassle for current and future callers
>> and more costly than doing the scan in the split code. More details are
>> discussed in [1].
>>
>> This issue can be exposed via:
>> 1. splitting a has_hwpoisoned folio to >0 order from debugfs interface;
>> 2. truncating part of a has_hwpoisoned folio in
>>    truncate_inode_partial_folio().
>>
>> And later accesses to a hwpoisoned page could be possible due to the
>> missing has_hwpoisoned folio flag. This will lead to MCE errors.
>>
>> Link: https://lore.kernel.org/all/CAHbLzkoOZm0PXxE9qwtF4gKR=cpRXrSrJ9V9Pm2DJexs985q4g@mail.gmail.com/ [1]
>> Fixes: c010d47f107f ("mm: thp: split huge page to any lower order pages")
>> Cc: stable@vger.kernel.org
>> Signed-off-by: Zi Yan <ziy@nvidia.com>
>
> This seems reasonable to me and is a good spot (thanks!), so:
>
> Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
>
>> ---
>> From V3[1]:
>>
>> 1. Separated from the original series;
>> 2. Added Fixes tag and cc'd stable;
>> 3. Simplified page_range_has_hwpoisoned();
>> 4. Renamed check_poisoned_pages to handle_hwpoison, made it const, and
>>    shorten the statement;
>> 5. Removed poisoned_new_folio variable and checked the condition
>>    directly.
>>
>> [1] https://lore.kernel.org/all/20251022033531.389351-2-ziy@nvidia.com/
>>
>>  mm/huge_memory.c | 23 ++++++++++++++++++++---
>>  1 file changed, 20 insertions(+), 3 deletions(-)
>>
>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>> index fc65ec3393d2..5215bb6aecfc 100644
>> --- a/mm/huge_memory.c
>> +++ b/mm/huge_memory.c
>> @@ -3455,6 +3455,14 @@ bool can_split_folio(struct folio *folio, int caller_pins, int *pextra_pins)
>>  					caller_pins;
>>  }
>>
>> +static bool page_range_has_hwpoisoned(struct page *page, long nr_pages)
>> +{
>> +	for (; nr_pages; page++, nr_pages--)
>> +		if (PageHWPoison(page))
>> +			return true;
>> +	return false;
>> +}
>> +
>>  /*
>>   * It splits @folio into @new_order folios and copies the @folio metadata to
>>   * all the resulting folios.
>> @@ -3462,17 +3470,24 @@ bool can_split_folio(struct folio *folio, int caller_pins, int *pextra_pins)
>>  static void __split_folio_to_order(struct folio *folio, int old_order,
>>  		int new_order)
>>  {
>> +	/* Scan poisoned pages when split a poisoned folio to large folios */
>> +	const bool handle_hwpoison = folio_test_has_hwpoisoned(folio) && new_order;
>
> OK was going to mention has_hwpoisoned is FOLIO_SECOND_PAGE but looks like you
> already deal with that :)

Right. And has_hwpoisoned is only set for large folios.

>
>>  	long new_nr_pages = 1 << new_order;
>>  	long nr_pages = 1 << old_order;
>>  	long i;
>>
>> +	folio_clear_has_hwpoisoned(folio);
>
> OK so we start by clearing the HW poisoned flag for the folio as a whole, which
> amounts to &folio->page[1] (which must be a tail page of course as new_order
> tested above).
>
> No other pages in the range should have this flag set as is a folio thing only.
>
> But this, in practice, sets the has_hwpoisoned flag for the first split folio...

handle_hwpoison is only true when after-split folios are large (new_order not 0).
All folio has_hwpoisoned set code is guarded by handle_hwpoison.

>
>> +
>> +	/* Check first new_nr_pages since the loop below skips them */
>> +	if (handle_hwpoison &&
>> +	    page_range_has_hwpoisoned(folio_page(folio, 0), new_nr_pages))
>> +		folio_set_has_hwpoisoned(folio);
>>  	/*
>>  	 * Skip the first new_nr_pages, since the new folio from them have all
>>  	 * the flags from the original folio.
>>  	 */
>>  	for (i = new_nr_pages; i < nr_pages; i += new_nr_pages) {
>>  		struct page *new_head = &folio->page + i;
>> -
>
> NIT: Why are we removing this newline?

It is a newline between two declarations.

>
>>  		/*
>>  		 * Careful: new_folio is not a "real" folio before we cleared PageTail.
>>  		 * Don't pass it around before clear_compound_head().
>> @@ -3514,6 +3529,10 @@ static void __split_folio_to_order(struct folio *folio, int old_order,
>>  				 (1L << PG_dirty) |
>>  				 LRU_GEN_MASK | LRU_REFS_MASK));
>>
>> +		if (handle_hwpoison &&
>> +		    page_range_has_hwpoisoned(new_head, new_nr_pages))
>> +			folio_set_has_hwpoisoned(new_folio);
>> +
>
> ...We then, for each folio which will be split, we check again and propagate to
> each based on pages in range.

Yes, but this loop only goes [new_nr_pages, nr_pages), so the code above is
needed for [0, new_nr_pages). The loop is done in this way to avoid redundant
work, flag and compound head setting, for [0, new_nr_pages) pages and the
original folio, since there is no change between the original values and
after-split values.

>
>>  		new_folio->mapping = folio->mapping;
>>  		new_folio->index = folio->index + i;
>>
>> @@ -3600,8 +3619,6 @@ static int __split_unmapped_folio(struct folio *folio, int new_order,
>>  	int start_order = uniform_split ? new_order : old_order - 1;
>>  	int split_order;
>>
>> -	folio_clear_has_hwpoisoned(folio);
>> -
>>  	/*
>>  	 * split to new_order one order at a time. For uniform split,
>>  	 * folio is split to new_order directly.
>> --
>> 2.51.0
>>


--
Best Regards,
Yan, Zi